SPRINGER BRIEFS IN APPLIED SCIENCES AND

TECHNOLOGY FORENSIC AND MEDICAL BIOINFORMATICS

Naresh BabuMuppalaneni
Vinit KumarGunjan Editors

Computational
Intelligence
in Medical
Informatics

SpringerBriefs in Applied Sciences

and Technology
Forensic and Medical Bioinformatics

Series editors
Amit Kumar, Hyderabad, India
Allam Appa Rao, Hyderabad, India

More information about this series at http://www.springer.com/series/11910

Naresh Babu Muppalaneni

Vinit Kumar Gunjan
Editors

Computational Intelligence
in Medical Informatics

123

Editors
Naresh Babu Muppalaneni
C.R. Rao Advanced Institute
of Mathematics, Statistics
and Computer Science
Hyderabad
India

ISSN 2191-530X
ISBN 978-981-287-259-3
DOI 10.1007/978-981-287-260-9

Vinit Kumar Gunjan

Annamacharya Institute of Technology
and Sciences
Kadapa
India

ISSN 2191-5318 (electronic)

ISBN 978-981-287-260-9 (eBook)

Library of Congress Control Number: 2014953255

Springer Singapore Heidelberg New York Dordrecht London
The Author(s) 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publishers location, in its current version, and permission for use must always
be obtained from Springer. Permissions for use may be obtained through RightsLink at the Copyright
Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Foreword

This book is an enthusiastic contribution of the best research work in the eld of
bioinformatics, biotechnology, and allied domains to the International Conference
on Computational Intelligence: Health and Disease (CIHD 2014) to be held at
Visakhapatnam, India during December 2728, 2014. The main objective of this
conference is to create an environment for (1) cross-disseminating state-of-the-art
knowledge to CI researchers, doctors and computational biologists; (2) creating a
common substrate of knowledge that both CI people, doctors and computational
biologists can understand; (3) stimulating the development of specialized CI
techniques, keeping in mind the application to computational biology; (4) fostering
new collaborations among scientists having similar or complementary backgrounds.
Yet another element is provided by many interesting historical data on diabetes
and an abundance of colorful illustrations. On top of that, there are innumerable
historical vignettes that interweave computer science and biology in a very
appealing way.
Although the emphasis of this work is on diabetes and other diseases, it contains
much that will be of interest to those outside this eld and to students of
Biotechnology, Bioinformatics, Chemistry and Computer Scienceindeed to
anyone with a fascination for the world of molecules. The authors have selected a
good number of prominent molecules as the key subjects of their essays. Although
these represent only a small sample of the world of biologically related molecules
and their impact on our health, they amply illustrate the importance of this eld of
science to humankind and the way in which the eld has evolved.
I think that the contributors can be condent that there will be many grateful
readers who will have gained a broader perspective of the disciplines of diabetes
and their remedies as a result of their efforts.
Hyderabad, India

Naresh Babu Muppalaneni

Vinit Kumar Gunjan

Preface

This volume contains a selection of the best contributions delivered at the International Conference on Computational Intelligence: Health and Disease (CIHD
2014) held at Visakhapatnam, India during December 2728, 2014. This conference is organized by Institute of Bioinformatics and Computational Biology
(IBCB), Visakhapatnam, India jointly with Andhra University and JNTU Kakinada.
The IBCB is a research organization established in Visakhapatnam, India. It is a
community of scholars devoted to the understanding of mysteries that remain in the
catalogue of human genes through intellectual inquiry. The Institute encourages and
supports curiosity-driven research in the elds of Bioinformatics and Computational Biology. The institute nurtures speculative thinking that produces advances in
knowledge that change the way we understand the world. It provides for the
mentoring of scholars, and ensures the freedom to undertake research that will make
signicant contributions in any of the broad range of elds in Bioinformatics and
Computational Biology.
CIHD 2014 is aimed to bring together computer professionals, doctors, academicians, and researchers to share their experience and expertise in Computational
Intelligence. The goal of the conference is to provide computer science professionals,
engineers, medical doctors, bioinformatics researchers, and other interdisciplinary
researchers a common platform to explore research opportunities.
A rigorous peer-review selection process was applied to ultimately select the
papers included in the program of the conference. This volume collects the best
contributions presented at the conference.
The success of this conference is to be credited to the contribution of many people.
In the rst place, we would like to thank Prof. Allam Appa Rao, Director, C.R. Rao
AIMSCS, who motivated and guided us in making this conference a grand success.
Our sincere thanks to Dr. Amit Kumar, Editor for Springer Briefs in Applied Sciences
and Technology, who helped us in bringing this series. Moreover, special thanks are
due to the Program Committee members and reviewers for their commitment to the
task of providing high-quality reviews. We thank Prof. B.M. Hegde (Padma Bhushan
Awardee, Cardiologist and Former Vice Chancellor, Manipal University) who
delivered the keynote address. Last but not least, we would thank the speakers Grady
vii

viii

Preface

Hanrahan (California Lutheran University, USA), Jayaram B. (Coordinator, Supercomputing Facility for Bioinformatics and Computational Biology, IIT Delhi),
Jeyakanthan J. (Professor and Head, Structural Biology and Biocomputing Lab,
Alagappa University), Nita Parekh (International Institute of Information Technology
Hyderabad (IIIT-H), Hyderabad, India), Pinnamaneni Bhanu Prasad (Advisor,
Kelenn Technology, France), Rajasekaran E. (Dhanalakshmi-Srinivasan Institute of
Technology, Tiruchirappalli), and Sridhar G.R. (Endocrine and Diabetes Centre,
Krishnanagar Visakhapatnam, India).
December 2014

Naresh Babu Muppalaneni

Vinit Kumar Gunjan

Committees

International Conference on Computational

Intelligence: Health and Disease (CIHD 2014)
2728 December 2014
Visakhapatnam, India
The International Conference on Computational Intelligence: Health and Disease
(CIHD 2014) held at Visakhapatnam, India during December 2728, 2014 is
organized by Institute of Bioinformatics and Computational Biology (IBCB),
Visakhapatnam, India jointly with Andhra University and JNTU Kakinada.

General Chair
Dr. Allam Appa Rao, SDPS Fellow, Director, C.R. Rao AIMSCS, UoH,
Hyderabad, India

Conference Secretary
Dr. P. Sateesh, Associate Professor, MVGR College of Engineering

Organizing Committee
Dr. Ch. Divakar, Secretary, IBCB

ix

Committees

Prof. P.V. Nageswara Rao, Head of the Department, Department of CSE, GITAM
University
Prof. P.V. Lakshmi, Head of the Department, Department of IT, GITAM
University
Prof. P. Krishna Subba Rao, Professor, Department of CSE, GVP College of
Engineering (Autonomous)
Dr. G. Satyavani, Assistant Professor, IIIT Allahabad
Dr. Akula Siva Prasad, Lecturer, Dr VS Krishna College
Shri. Kunjam Nageswara Rao, Assistant Professor, AU College of Engineering
Shri. D. Dharmayya, Associate Professor, Vignan Institute of Information
Technology
Shri. T.M.N. Vamsi, Associate Professor, GVP PG College

Advisory Committee
Prof. P.S. Avadhani, Professor, Department of CS and SE, AU College of
Engineering
Prof. P. Srinivasa Rao, Head of the Department, Department of CS and SE, AU
College of Engineering
Dr. Raghu Korrapati, Professor, Walden University, USA
Prof. Ch. Satyanarayana, Professor, Department of CSE, JNTU Kakinada
Prof. C.P.V.N.J. Mohan Rao, Professor and Principal, Avanthi Institute of
Engineering and Technology
Dr. Anirban Banerjee, Assistant Professor, IISER Kolkata
Dr. Raghunath Reddy Burri, Scientist, GVK Bio Hyderabad
Dr. L. Sumalatha, Professor and Head, Department of CSE, JNTU Kakinada
Dr. D. Suryanarayana, Principal, Vishnu Institute Technology, Bhimavaram
Dr. A. Yesu Babu, Professor and Head, Department of CSE, Sir C.R. Reddy
College of Engineering, Eluru
Dr. T.K. Rama Krishna, Principal, Sri Sai Aditya Institute of Science and
Technology

Finance Committee
Shri. B. Poorna Satyanarayana, Professor, Department of CSE, Chaitanya
Engineering College
Dr. T. Uma Devi, GITAM University
Dr. R. Bramarambha, Associate Professor, Department of IT, GITAM University
Smt. P. Lakshmi Jagadamba, Associate Professor, GVP
Smt. Amita Kasyap, Women Scientist, C.R. Rao AIMSCS

Committees

xi

Publication Committee
Dr. Amit Kumar, Publication Chair, Director, BDRC
Dr. Kudipudi Srinivas, Co-chair, Professor, V.R. Siddhartha Engineering College
Dr. G. Lavanya Devi, Assistant Professor, Department of CS and SE, AU College
of Engineering
Dr. P. Sateesh, Associate Professor, MVGR College of Engineering
Dr. A. Chandra Sekhar, Principal, Sankethika Institute of Technology
Dr. K. Karthika Pavani, Professor, RVR and JC College of Engineering

Website Committee
Dr. N.G.K. Murthy, Professor of CSE, GVIT Bhimavaram
Dr. Suresh Babu Mudunuri, Professor of CSE, GVIT Bhimavaram
Shri. Y. Ramesh Kumar, Head of the Department, CSE, Avanthi Institute of
Engineering and Technology
Financing Institutions
Department of Science and Technology, Government of India

Contents

Analysis of the Structural Details of DsrO Protein from

Allochromatium vinosum to Identify the Role of the Protein
in the Redox Transport Process Through the dsr Operon . . . . . . . . . .
Semanti Ghosh and Angshuman Bagchi
LCMV Interaction Changes with T192M Mutation in
Alpha-Dystroglycan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simanti Bhattacharya, Sanchari Bhattacharjee, Prosun Kumar Biswas,
Amit Das, Rakhi Dasgupta and Angshuman Bagchi
Structural and Functional Characterization of Arabidopsis
thaliana WW Domain Containing Protein F4JC80 . . . . . . . . . . . . . . .
Amit Das, Simanti Bhattacharya, Angshuman Bagchi
and Rakhi Dasgupta

11

25

Structural Insights into IbpAIbpB Interactions to Predict Their

Roles in Heat Shock Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sanchari Bhattacharjee, Rakhi Dasgupta and Angshuman Bagchi

41

Improving the Performance of Multi-parameter Patient

Monitor System Using Additional Features . . . . . . . . . . . . . . . . . . . . .
S. Premanand, C. Santhosh Kumar and A. Anand Kumar

53

Rough Set Rule-Based Technique for the Retrieval of Missing

Data in Malaria Diseases Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . .
B.S. Panda, S.S. Gantayat and Ashok Misra

59

Automatic Image Segmentation for Video Capsule Endoscopy . . . . . . .

V.B. Surya Prasath and Radhakrishnan Delhibabu

73

xiii

xiv

Contents

Effect of Feature Selection on Kinase Classication Models . . . . . . . . .

Priyanka Purkayastha, Akhila Rallapalli, N.L. Bhanu Murthy,
Aruna Malapati, Perumal Yogeeswari and Dharmarajan Sriram
Rheumatoid Arthritis Candidate Genes Identication
by Investigating Core and Periphery Interaction Structures. . . . . . . . .
Sachidanand Singh, V.P. Snijesh and J. Jannet Vennila

81

87

Analysis of the Structural Details of DsrO

Protein from Allochromatium vinosum
to Identify the Role of the Protein
in the Redox Transport Process Through
the dsr Operon
Semanti Ghosh and Angshuman Bagchi

Abstract Sulfur oxidation is one of the oldest known redox processes in our
environment mediated by phylogenetically diverse sets of microorganisms. The
sulfur oxidation process is mediated mainly by dsr operon which is basically
involved in the balancing and utilization of environmental sulfur compounds.
DsrMKJOP complex from the dsr operon is the central player of this operon. DsrO
is a periplasmic protein which binds FeS clusters responsible for electron transfer to
DsrP protein from the dsr operon. DsrP protein is known to be involved in electron
transfer to DsrM protein. DsrM protein would then donate the electrons to DsrK
protein, the catalytic subunit of this complex. In the present work, we tried to
analyze the role of DsrO protein of the dsr operon from the ecologically and
industrially important organism Allochromatium vinosum. There are no previous
reports that deal with the structural details of the DsrO protein. We predicted the
structure of the DsrO protein obtained by homology modeling. The structure of the
modeled protein was then docked with various sulfur anion ligands to understand
the molecular mechanism of the transportation process of sulfur anion ligands by
this DsrMKJOP complex. This study may therefore be considered as a rst report of
its kind that would therefore enlighten the pathway for analysis of the biochemical
mechanism of sulfur oxidation reaction cycle by dsr operon.

Keywords Sulfur oxidation Ecological importance dsr operon Dsro protein

Homology modeling Molecular docking

S. Ghosh
A. Bagchi (&)

Department of Biochemistry and Biophysics, University of Kalyani,
Kalyani 741235, Nadia, India
e-mail: [email protected]
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_1

S. Ghosh and A. Bagchi

1 Introduction
Sulfur oxidation reaction cycle is one of the important biogeochemical cycles in the
world. Sulfur has a wide range of oxidation states viz., +6 to 2. This makes the
element capable of taking part in a number of different biological processes. Sulfurbased chemo or photolithotrophy is one of such processes involving the transfer of
electrons from reduced sulfur compounds like sulte, thiosulfate, elemental sulfur,
etc. The sulfur oxidation process is mediated by a diverse set of microorganisms.
Only little is known about the molecular mechanisms of this sulfur oxidation
process in these microorganisms. One of the sulfur oxidizers is Allochromatium
vinosum (A. vinosum), a dominant member of purple sulfur bacteria. This bacterium
uses reduced sulfur compounds as electron donor for anoxygenic photosynthesis
[1]. Recent studies with A. vinosum revealed that a multiple gene cluster comprising
genes dsrA, dsrB, dsrE, dsrF, dsrH, dsrC, dsrM, dsrK, dsrL, dsrJ, dsrO, dsrP,
dsrN, dsrS, and dsrR is involved in the sulfur oxidation process [2]. The organism
A. vinosum has a wide range of applications in different industrial processes like
waste remediation and removal of toxic compounds, e.g., odorous sulfur
compounds like sulde and explosives and production of industrially relevant
organo-chemicals such as vitamins, bio-polyesters, and biohydrogen [1]. It is well
known that A. vinosum uses the DsrMKJOP protein complex to carry out the sulfur
oxidation process [3]. DsrJ, a periplasmic protein, may be involved in the oxidation
of a putative sulfur substrate in the periplasm and the released electrons would be
transported across the membrane via the other components, viz., DsrO, DsrP,
DsrM, DsrK successively, of the DsrMKJOP complex [3]. DsrO is a periplasmic
iron-sulfur protein and is known to be involved in electron transfer process [24]. It
is not yet known which amino acids of DsrO are involved in the electron transport
or interact with sulfur anions. So, in this work, we have attempted to characterize
DsrO protein at the structural level. We have predicted the putative active site
geometry of the DsrO protein. In order to predict the molecular mechanism of the
electron transport through DsrO protein we have docked the different sulfur anions
present in the environment with DsrO protein. Till date there are no reports that deal
with the analyses of the detailed structural information of DsrO as well as binding
of sulfur anions with this protein to predict the mechanism of electron transport.
This work is therefore the rst of its kind. Since there are no previous reports
regarding the molecular and structural details of DsrO protein, our work would
therefore be important to analyze the biochemical mechanism of sulfur oxidation
process by this ecologically and industrially important microbial species.

Analysis of the Structural Details of DsrO Protein

DsrO

PLMCQHCEHPPCVDVCPTGASFKRADGIVMVDRHLCIGCRYCMMACPYKARSFIHQPTTG 187

2VPW PEQCLHCENPPCVPVCPTGASYQTKDGLVLVDPKKCIACGACIAACPYDAR-YLHP---------- 109

DsrO QLTAVPRGKGCVESCNLCVHRRDNGEESTACVDAC 221
2VPW ------------------AGYVSKCTFCAHRLEKG-KVPACVETC 135

Fig. 1 Sequence alignment of the DsrO functional domain with PsrABC (PDB code: 2VPW) as
template

2 Materials and Methods

2.1 Sequence Analysis and Homology Modeling
of DsrO Protein
The amino acid sequence of DsrO protein from A. vinosum was obtained from
Entrez database (Accession No. YP_003443232). To identify the presence of
domains in DsrO protein, the amino acid sequence of DsrO protein was used to
search Pfam [5]. The Pfam results have shown the presence of the domain 4Fe-4S
dis-cluster spanning amino acid residues 125224. This protein superfamily
includes proteins that bind to iron-sulfur clusters. The amino acid sequence of DsrO
protein has been used to search Brookhaven Protein Data Bank (PBD) [6] using the
software BLAST [7] for nding suitable template for homology modeling. The
BLAST search result of DsrO has revealed it to have 45 % identity with X-ray
crystal structure of Polysulde (PsrABC) from Thermus thermophilus (T. thermophilus) (PDB code: 2VPW) [8, 9]. The sequence alignment is shown in Fig. 1.
The protein has been modeled using the corresponding crystal structure as
template. Homology modeling has been performed using the software suite DS
Modeling [MSI/ and Accelrys, San Diego, CA, USA].
The modeled structure has been superimposed on the crystal template without
altering the coordinate system of atomic positions in the template. The root-meansquare deviation (r.m.s.d.) for the superimposition is 0.906 for DsrO with its
template (Fig. 2).
The model is then subjected to energy minimization using the CHARMm force
eld [10] xing the backbone of the modeled protein to ensure proper interactions
until the structure reached the nal derivative of 0.001 kcal/mol.

2.2 Validation of the Model

The stereo-chemical qualities of the three-dimensional model obtained after energy
minimization are checked by VERIFY3D [11] and PROCHECK [12] using
Structural Analysis and Verication Server (SAVes). Regarding the main chain

S. Ghosh and A. Bagchi

Fig. 2 Superimposition of
the -carbon backbone of
DsrO on 2VPW (B). DsrO is
presented in red and 2VPW in
green

properties of the modeled protein neither considerable bad contacts nor C tetrahedron distortion nor hydrogen bond energy problems have been found. Moreover,
the average G factor, the measure of the normality degree of the protein properties,
has been found to be 0.10, which is inside the permitted values for homology
models. Furthermore, no distortions of the side chain torsion angles are found. The
Ramachandran plot [13] has been drawn. No residues are found to be present in the
disallowed regions of the Ramachandran plot. The residue prole of the model has
been checked by VERIFY3D and it indicates a good model quality.

2.3 Molecular Docking of DsrO with Sulfur Anions

In order to identify the active site residues in DsrO protein, InterPro database was
searched using the amino acid sequence of DsrO protein. The search result revealed
that DsrO has similar domain architecture to InterPro entry IPR017900. It has been
observed that four cysteine residues at amino acid positions 162, 165, 168, and 172
[IPR017900] are predicted to be involved in binding of sulfur anions. These active
site amino acid residues are used for docking purposes. To study the interactions
between DsrO and various sulfur anions that take part in sulfur oxidation process,
the model of DsrO protein has been docked with sulte (SO32), sulfate (SO42),

Analysis of the Structural Details of DsrO Protein

and thiosulfate (S2O32) using the program GOLD [14]. The docked complexes that
have yielded the best GoldScore and ChemScore are selected and analyzed to study
the interactions.

2.4 Molecular Mechanics Simulations of DsrO-Sulfur

Anion Complexes
The docked complexes of DsrO with different sulfur anions have been subjected to
molecular mechanics simulations. During simulation, the backbone of the DsrO
protein has been kept xed. The simulations have been performed in explicit
solvent with a di-electric constant of 80. For the simulations, 2,000 cycles of
steepest descent followed by 1,000 cycles of conjugate gradient methods have been
used with CHARMm [10] force elds. The nal coordinates of the docked complexes have been saved for analyses.

2.5 Calculations for ProteinLigand Interactions

The stereo-chemical qualities of the energy minimized docked complexes are
checked by PROCHECK and VERIFY3D. The interactions between the sulfur
anions and DsrO protein are then analyzed by DISCOVERY STUDIO software
suite.

3 Results and Discussion

3.1 Description of the Three-Dimensional Structure
of DsrO from Allochromatium vinosum
The DsrO protein from A. vinosum has a functional domain consisting of 95 amino
acids. The modeled structure is similar to X-ray crystal structure of Polysulde
Reductase (PsrABC) from T. thermophilus strain HB27 (PDB code: 2VPW, Chain
B). The predicted structure of DsrO protein from A. vinosum consists of a mixture
of helices, coils, and sheets (Fig. 3). The functional domain starts with a coil (amino
acid residues 127136) followed by a helix (amino acid residues 137140). Then,
there are two beta sheets (amino acid residues 147149 and 155157) and three
helices (amino acid residues 167171, 205210, 216219) joined by nine coil
regions (amino acid residues 142146, 150, 158, 162166, 172204, 211, 214215,
220221).

S. Ghosh and A. Bagchi

Fig. 3 Three-dimensional
model of DsrO protein. The
helices are shown in red. The
strands are presented in
yellow. The rest of the part is
coil region shown in green

3.2 Putative Active Site Geometry of DsrO

from Allochromatium vinosum
DsrO belongs to the 4Fe-4S ferredoxin, iron-sulfur binding family of proteins.
Proteins belonging to this class are involved in transport of electrons in a range of
metabolic reactions; they fall into several subgroups according to the nature of their
iron-sulfur cluster(s) [15, 16]. The various Fe-S sites found in many electron
transfer proteins (e.g., ferredoxins) are also found in many enzymes, where these
centers are involved in intra or inter-protein electron transfer processes. For
example, sulte reductase contains a siroheme and Fe4S4 center, which are strongly
coupled and involved in the six electron reduction of SO32 to H2S [17]. In most
ferredoxins, the protein ligands are cysteines, which provide four thiolate donors to
the 1Fe, 2Fe, or 4Fe centers [17]. IPR017900 entry represents at least one conserved domain, including four cysteines residues that bind to a 4Fe-4S center.
Alignment of the amino acid sequence of DsrO from A. vinosum has revealed that
DsrO also has four cysteine residues at the identical positions viz., Cys162, Cys165,
Cys168, Cys172. During molecular docking in GOLD program Cys168 was used
as the binding site of DsrO protein with the sulfur anions as it covers all the
required flexible side chains for docking. The flexible side chains of the amino acid
residues which are mainly involved in the interactions with sulfur anions are listed
in Table 1. This positioning of the amino acid residues is known to play an
important role in ligand binding described in the next section.
Table 1 List of amino acids of DsrO protein involved in proteinligand interactions
DsrO

Cys130, Cys138, Val139, Asp140, Val141, Cys142, Thr144, Ser147, Phe148,

Val155, Met156, Val157, Asp158, Cys162, Ile163, Cys165, Arg166, Tyr167,
Cys168, Met169, Met170, Cys172, Tyr174, Leu175, Ser178, Phe179, Ile180, Cys197,
Val198, Glu199, Ser200

Analysis of the Structural Details of DsrO Protein

Table 2 Comparative values of GOLD program of proteinligand complexes

Proteinligand
complex name

GoldScore tness

ChemScore DG

ChemScore H-bond
weighted

DsrO-thiosulfate
DsrO-sulfate
DsrO-sulte

36.9534
30.7964
26.9615

16.0675
13.9717
17.7079

10.5875
8.4917
12.2279

3.3 GOLD Docking Scores for ProteinLigand Interactions

The energetics of DsrO-sulfur anion interactions were calculated by considering the
parameters of GOLD program as listed in Table 2. We have used thiosulfate, sulte,
and sulfate anions as the potential sulfur substrates since the actual sulfur anion
ligand is not yet known clearly.
From the GoldScore tness values among three complexes it may be safely
concluded that DsrO protein binds strongly with thiosulfate. It was previously
shown that sulde/thiosulfate grown cells show high expression of DsrO protein
[4]. Now, from these docking complexes it is conrmed that thiosulfate is the most
suitable ligand among the different sulfur anions.

3.4 Interactions of DsrO with Sulfur Anions

The binding energy values associated with the docked complexes of DsrO-thiosulfate,
DsrO-sulfate, and DsrO-sulte are presented in Table 3.
It has also been observed that more number of amino acid residues of DsrO
protein is involved in binding with thiosulfate as compared to other sulfur anions
(Fig. 4). The amino acid residues Cys142, Gly164, Cys165, Tyr167, and Cys168
from DsrO protein are involved in binding with thiosulfate. On the other hand, for
DsrO-sulte complex the amino acids Gly164, Cys165, Arg166 are involved in
binding, whereas for DsrO-sulfate complex, the amino acid residues Cys162 and
Gly164 are involved in binding (Figs. 5 and 6). From this analysis it is apparent that
DsrO protein binds strongly with thiosulfate as greater number of amino acid
residues are involved in interactions. Thiosulfate being the largest of the three sulfur

Table 3 Binding energy calculations of proteinligand complexes

Docked complex

Total binding energy of the

proteinligand complex

Energy of the complex

Entropy

DsrO-thiosulfate
DsrO-sulte
DsrO-sulfate

0.0975
0.55491
0.31514

3717.26643
3671.98443
3655.62403

17.03740
17.01440
16.59680

S. Ghosh and A. Bagchi

Fig. 4 Interactions of the DsrO protein with thiosulfate. Cys142, Gly164, Cys165, Tyr167,
Cys168 amino acids from DsrO protein are involved in binding with thiosulfate

Fig. 5 DsrO-sulte complex; the amino acids Gly164, Cys165, Arg166 are involved in binding

anion ligands used has the maximum chance of interactions with the protein. On the
other hand in sulfate, the sulfur atom has the highest oxidation state of +6. Hence,
sulfate has no ability to be oxidized and is not used in the oxidation process.

Analysis of the Structural Details of DsrO Protein

Fig. 6 DsrO-sulfate complex; the amino acid residues Cys162 and Gly164 are involved in
binding

4 Conclusions
In this study we elucidate the structural basis of the involvement of DsrO protein
from A. vinosum in electron transport during the oxidation of sulfur compounds. We
built the three-dimensional structure DsrO protein using comparative modeling
technique. The dockings of sulfur anions with DsrO allowed us to identify the details
of their mode of interactions. We identied the amino acid residues from the DsrO
protein that are involved in the binding of DsrO protein with sulfur anions. Results
from this study will be important for understanding the pathway of electron transport
via this protein in the global sulfur cycle. This homology model of DsrO provides a
rational framework for designing experiments aimed at determining the contribution
of amino acid residues responsible for electron transport via dsr operon.
Acknowledgments Ms. Semanti Ghosh is thankful to the University of Kalyani, Govt. of West
Bengal, India, and UGC for the nancial support. We would like to thank Bioinformatics Infrastructure Facility and the DST-PURSE program 20122015 going on in the department of Biochemistry and Biophysics, University of Kalyani for the support.

References
1. Weissgerber T, Zigann R, Bruce D, Chang Y, Detter JC, Han C, Hauser L, Jeffrie CD, Land
M, Munk AC, Tapia R, Dahl C (2011) Complete genome sequence of Allochromatium
vinosum DSM 180T. Stand Genomic Sci 5:311330
2. Grein F (2010) Biochemical, biophysical and functional analysis of the DsrMKJOP
transmembrane complex from Allochromatium vinosum. PhD. thesis, Rhenish Friedrich
Wilhelm University, Germany

10

S. Ghosh and A. Bagchi

3. Grein F, Pereira IAC, Dahl C (2010) Biochemical characterization of individual components

of the Allochromatium vinosum DsrMKJOP transmembrane complex aids understanding of
complex function in vivo. J Bacteriol 192(24):63696377
4. Dahl C, Engels S, Pott-Sperling AS, Schulte A, Sander J, Lbbe Y, Deuster O, Brune DC
(2005) Novel genes of the dsr gene cluster and evidence for close interaction of Dsr proteins
during sulfur oxidation in the phototrophic sulfur bacterium Allochromatium vinosum.
J Bacteriol 187(4):13921404
5. Bateman A, Coin L, Durbin R, Finn RD, Hollich V, Grifths-Jones S, Khanna A, Marshall M,
Moxon S, Sonnhammer EL, Studholme DJ, Yeats C, Eddy SR (2004) The Pfam protein
families database. Nucleic Acids Res 32(Database issue):D138-41
6. Berman HM (2008) The protein data bank: a historical perspective. Acta Crystallogr A
64:8895
7. Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ (1990) Basic local alignment search
tool. J Mol Biol 215(3):403410
8. http: www.rcsb.org/pdb/explore/explore.do?structureId=2VPW
9. Jormakka M, Yokoyama K, Yano T, Tamakoshi M, Akimoto S, Shimamura T, Curmi P, Iwata
S (2008) Molecular mechanism of energy conservation in polysulde respiration. Nat Struct
Mol Biol 15(7):730737
10. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan S, Karplus M (1983)
CHARMM: a program for macromolecular energy minimization and dynamics calculations.
J Comp Chem 4:187217
11. Lthy R, Bowie JU, Eisenberg D (1992) Assessment of protein models with three-dimensional
proles. Nature 356(6364):8385
12. Laskowski RA, McArthur MW, Moss DS, Thornton JM (1993) PROCHECK: a program to
check the stereochemical quality of protein structures. J Appl Crystallogr 26:283291
13. Ramachandran GN, Ramakrishnan C, Sasisekharan V (1963) Stereochemistry of polypeptide
chain congurations. J Mol Biol 7:9599
14. Jones G, Willett P, Glen RC, Leach AR, Taylor R (1997) Development and validation of a
genetic algorithm for flexible docking. J Mol Biol 267(3):727748
15. George DG, Hunt LT, Yeh LS, Barker WC (1985) New perspectives on bacterial ferredoxin
evolution. J Mol Evol 22(1):2031
16. Otaka E, Ooi T (1989) Examination of protein sequence homologies: V. New perspectives on
evolution between bacterial and chloroplast-type ferredoxins inferred from sequence evidence.
J Mol Evol 29(3):246254
17. Bertini I, Gray HB, Lippard SJ, Valentine JS (1994) Bioinorganic chemistry. In: Stiefel EI,
George GN (eds) Chapter 7: Ferredoxins, hydrogenases, and nitrogenases: metal-sulde
proteins. University Science Books, Mill Valley. ISBN:0-935702-57-1

LCMV Interaction Changes with T192M

Mutation in Alpha-Dystroglycan
Simanti Bhattacharya, Sanchari Bhattacharjee,
Prosun Kumar Biswas, Amit Das, Rakhi Dasgupta
and Angshuman Bagchi

Abstract Limb girdle muscular dystrophy (OMIM: 613818) is a severe disease in

humans, which broadly affects brain development. The disease is caused by T192M
mutation in the protein alpha-dystroglycan (-DG). -DG is an important component of dystrophindystroglycan complex which links extracellular matrices with
actin cytoskeleton and thereby maintains signalling cascades essential for the
development of tissues and organs. The mutation T192M in -DG hampers proper
glycosylation of -DG thereby developing limb girdle muscular dystrophy. Prototype virus for Old World Arenaviruses (OWV), Lymphocytic Choriomeningitis
virus (LCMV) also uses this -DG as host cell receptor and invades the host cell
causing a disease called Lymphocytic choriomeningitis, an infection to meninges.
Thereby, interaction of -DG and LCMV has become an interesting object of study
to predict the mode of the disease onset. In our current work, we have used
homology modelling, molecular docking and molecular dynamics (MD) with
temperature variation. We have identied signicant structural differences between
wild type (WT) and mutant (MT) -DG in terms of spatiotemporal orientations of
amino acids. This change in the folding patterns of the WT and MT -DG has
brought forth a different interaction pattern of the WT and MT -DG with GP1
protein from LCMV as reflected in our docking simulations. Further MD simulations with the complexes over tropical and temperate environment have revealed
that MT--DG-LCMV GP1 complex is relatively more stable than the wild type
counterpart. It has also been found that LCMV GP1 has interacted strongly with
mutant -DG. Our studies therefore has shed light on the structure and molecular
interaction pattern of LCMV with MT -DG and also indicate a possibility of
T192M mutant in -DG making the receptor to interact strongly with LCMV GP1.
These insights also provide clues to develop possible therapeutic approaches.

Keywords Homology modelling Molecular dynamics

dystrophy Viral infection-drug design

Limb girdle muscular

S. Bhattacharya
S. Bhattacharjee
P.K. Biswas
A. Das
R. Dasgupta
A. Bagchi (&)

Department of Biochemistry and Biophysics, University of Kalyani,
Nadia 741235, West Bengal, India
e-mail: [email protected]
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_2

11

12

S. Bhattacharya et al.

Abbreviations
DG
MDDGC9
OMIM
PDB
RMSD
WT
MT

Dystroglycan
Muscular Dystrophy, Dystroglycanopathy, Type C9
Online Mendelian Inheritance in Man
Protein Data Bank
Root Mean Square Deviation
Wildtype
Mutated

Highlights
1. T192M mutation causes changes in alpha-dystroglycan structure.
2. Interaction between Lymphocytic Choriomeningitis Virus (LCMV) with its
receptor alpha-dystroglycan changes due to the mutation.
3. Molecular dynamics run with temperature variation shows interaction of
LCMV with mutated receptor is stronger than the wild type receptor.

1 Introduction
Dystroglycan (DG) is a cell surface receptor and it belongs to dystrophindystroglycan complex which links extracellular matrices with cellular actin cytoskeleton [1].
Cellular actin cytoskeleton plays crucial role in development of tissues, organs not
only by providing mechanical strengths to cells but also by taking active part in
regulating macromolecular events [2] required for proper development [3]. Matured
DG protein is composed of two subunits: alpha-dystroglycan (-DG), the extracellular
part of DG that interacts and receives signals from extracellular proteins like laminin,
parlecan, argin, etc. and beta dystroglycan (-DG), which remains bound to cell
membrane [1]. This -DG is linked with cellular dystrophin, which is an active acin
binding protein [4]. The protein -DG contains a mucin like region sandwiched
between its N terminal and C terminal globular parts. This mucin rich region is
essential for the proper glycosylation and functioning of -DG [1]. Reportedly, a point
mutation T192M in -DG causes limb girdle muscular dystrophy, MDDGC9 (OMIM:
613818) affecting brain and nervous system growth resulting in severe cognitive
impairments, mental retardation and delayed motor development [5, 6]. This mutation
actually hampers LARGE (a transacetylase) mediated glycosylation leading to the
disease onset. This signies the role of -DG in development and muscle strength.
Interestingly, -DG also serves as a receptor to a large class of arenaviridae (AV),
Old World viruses (OWV) [7]. OWVs invade host system by its preglycoprotein
complex which has three major components: N terminal stable signal peptide (SSP),
followed by glycoprotein G1 (GP1 *42 kDa) and glycoprotein G2 (GP2 *35 kDa)
domains [7]. These viruses interact with the N terminal few amino acids and mucin

LCMV Interaction Changes with T192M Mutation

13

rich region of -DG via their GP1 for attachment and fusion to the host system. Some
of these AVs are also causative agents of diseases like lassa fever, Bolivian hemorrhagic fever, lymphocytic choriomeningitis (LCM), etc. causing signicant mortality
in affected human populations [8]. The prototype for OWV, Lymphocytic Choriomeningitis Virus (LCMV) causing LCM, a chronic asymptomatic, lifelong infection
of meninges which is the membrane to protect central nervous systems [9, 10]. This
has tempted us to study the molecular interaction pattern of -DG and LCMV GP1.
Although detailed studies have been carried out on LCMV regarding its global
distribution, pathogenicity, residues essential for its interaction with -DG [8], the
insight to its molecular interaction with MT -DG is still not deciphered. Therefore,
we have ventured into the prediction of molecular aspects of disease propagation. We
have taken a homology modelling approach to construct the model of LCMV GP1
domain and thereafter have docked this model with wild type (WT -DG) and T192M
mutant (MT -DG) forms of -DG separately. Our initial studies [6] with -DG
structures have also revealed that the mutant structure differs signicantly in terms of
its surface hydrophobicity in vicinity to the mutation site, 192. That is why we have
focused on the structural changes occurring in the vicinity of mutation site and its
effect on the binding orientation of LCMV GP1 with MT -DG to elucidate the effect
of the above-mentioned mutation. Molecular docking and subsequent molecular
mechanistic simulations for the rst time have revealed that MT -DG has increased
numbers of intermolecular hydrogen and hydrophobic bonds with LCMV GP1
establishing a stable interaction. Further we have employed MD with elevated temperature to monitor the changes in this interaction pattern. This study therefore for the
rst time sheds some light into the scenario where MT--DG interacts with LCMV.
And we have observed the pattern indeed changed with mutated receptor protein. In
future, this work will be helpful in designing some effective therapeutic approaches to
defeat the virus borne disease.

2 Materials and Methods

2.1 Template Search
First, the sequence of LCMV GP1 domain, comprising amino acid residues
88241, was extracted from UniprotKB (id: E2D674). This was then used as input
sequence to search for the template in FUGUE [11] and pdb [12] using PSI-BLAST
[13]. In both the cases, the A chain of crystal structure of envelope glycoprotein
Machupo virus GP1 (PDB code: 2WFO) was obtained as the closest match.

2.2 Homology Modelling

This structure was then used as template to construct homology model for LCMV
GP1 using modbase [14] web server with sequence-sequence, prole-sequence

14

S. Bhattacharya et al.

(PSI-BLAST) and sequence-prole (Modeller Built-Prole) methods. Since the

template is a model of New World virus- Machupo virus GP1, the modelling of our
desired protein was not so straightforward. So, alternatively, model for this region
(amino acid residues 88241) has also been constructed using I-TASSER and
RaptorX model building servers independently. Both these servers used 2WFO as
the template. The models so generated were then superimposed on the previously
built model to generate as consensus report (Supplementary File 1). All models
were tested with Verify 3D [15] and PROCHECK [16]. We have also denatured the
structure completely taking it up to 373 K starting from 273 K, then have allowed it
to cool down to 273 K. And the structures were then superimposed in order to
check for the folding accuracy. The Root Mean Square Deviation (RMSD) for these
two superimposed les have been found to be 0.8 (Supplementary File 2) which
implies that they are indeed identical and the folding patterns are conserved.

2.3 Molecular Docking

Next, we used this model to dock with WT -DG and MT -DG [6] separately
using Z dock [17], PyDock [18] and Patchdock [19]. Docked complexes with best
score and lower energy were selected and there were total 25 docked complexes
(5 from Z dock, 10 from Pydock and 10 from Fire Dock) for each type of complex,
i.e. WT--DG with LCMV and MT--DG with LCMV. Resultant docked complexes were then energy minimized keeping their backbones xed to ensure proper
interactions. 800 cycles of conjugate gradient energy minimization steps were
applied with CHARMm [20] force eld until the structure reached a nal derivative
of 0.01 kcal/mole. After each minimization, the structures were veried with
Verify3D and PROCHECK. Further analyses of the docked complexes were done
using Discovery Studio (DS) package version 2.5 and Protein Interaction Calculator
[21]. Binding sites were detected with analyse binding site tool from DS for all the
docked complexes to validate the docking process. Finally, the complexes with the
lowest energy prole and with binding sites consensus with literature [8, 10] have
been selected for wild type and mutated complexes. The complexes were then
further subjected to minimization in explicit solvent system with CHARMm force
eld. Dielectric constant of the molecular surface was kept 80 as of periodic water
bound condition.

2.4 Molecular Dynamics Simulations

Minimized complexes then were brought to elevated tropical temperatures
(i.e. 300320 K) and 10,000 time step of production run was performed in each of
the cases keeping pressure constant. The conformational changes were recorded and
plotted against time.

LCMV Interaction Changes with T192M Mutation

15

3 Results
3.1 The Conformation of MT -DG Contributes
to Altered Interaction with LCMV GP1
Our studies [6] with the models of both WT and MT -DG initially have gured out
that mutant structure has more buried surfaces than that of its wild type form. In
order to investigate further the art of interactions, we have docked LCMV GP1 with
WT and MT -DG, separately. Analyses of docked pattern of the models clearly
have reflected that there are indeed signicant differences both in terms of amino
acid residues involved in interaction as well as the spatiotemporal orientation of
LCMV GP1, docked with either WT or MT -DG (Fig. 1a). As mentioned earlier,
the focus of our study was the immediate vicinity of mutation site at the amino acid
residue position 192. And we have marked those amino acid residues 8896 of
LCMV GP1 have successfully interacted with a groove formed by amino acid
residues 142192 of -DG (Fig. 1b). We have attempted to analyse the total number
of intermolecular hydrogen bonds (H-bond) and hydrophobic interactions (HPI)
formed between LCMV GP1 and -DG in both the cases, in two different systems,
i.e. discovery studio 2.5 and protein interaction calculator. Normalized average
values of the bonds formed have been used to generate a graphical view (Fig. 1c)
which reflected that LCMV-GP1 docked with MT -DG had more numbers of
H-Bonds and HPI interactions than the bonds formed for its docked model with
WT--DG (left panel, Fig. 1c). Hydrogen bonds and hydrophobic interactions are
the non-covalent interactions which are meant to stabilize the overall interaction
occurring between two proteins [22]. Disturbance in these bonds will denitely
affect the binding. Studies with the structures in further details have marked this
dispute to be largely contributed by a huge structural change in mutant form of
-DG in vicinity to the mutation site (Met 192) that has resulted in a new space for
LCMV GP1 interaction and also has created potentially altered kinks among few
residues (arrows, Fig. 1d, e). Superimpositions of models of WT and MT--DG,
minimized with either vacuum [6] or explicit solvent condition (Supplementary
File 3) have revealed that there indeed exist structural differences between the two
structures.

3.2 Identical Amino Acids Bent Differentially to Give

a New Conformation to Mutant -DG
Further assessment of the afore mentioned portion of -DG has identied amino acid
stretch 166171 of -DG to be the key to these disputes. The relative bends and
positioning of these residues have formed a surface which is more closed or buried in
MT -DG than that of its wild type form (Fig. 2a). Surprisingly, dihedral angles
between the adjacent residues in this patch, i.e. Ser166-Val167, Val167-Arg168,

16

S. Bhattacharya et al.

Fig. 1 LCMV GP1 exhibited a different spatiotemporal orientation when interacted with MT -DG.
a Docked structures of LCMV GP1 (red backbone, yellow surface) with WT--DG (blue backbone,
left panel) and with MT -DG (blue backbone, right panel). The common interacting potion has been
shown with red backbone representation. b Amino acids in the vicinity of mutation site 192nd
position and the different orientation of the docked complex. c Differences in intermolecular
hydrogen bond and hydrophobic interactions have been shown. d, e Structural differences between
WT and MT -DG that open up an alternative binding site in case of mutated structure

LCMV Interaction Changes with T192M Mutation

17

Fig. 2 Unusual kinks in amino acid pairs burying the surface in MT -DG. a Different surface
hydrophobicity between WT and MT -DG making the interaction groove more buried. b The
change in dihedral angles contributing to altered kinks in MT -DG

Thr169-Ala170 and Ala170-Ser171, have possessed almost 180 change in two

structures of -DG (Fig. 2b). The corresponding changes of the angles are:
Ser166-Val167 (175.90) (177.41), Val167-Arg168 (178.42) (177.36),
Thr169-Ala170 (178.12) (178.39) and Ala170-Ser171 (178.18) (178.97)
from WT MT -DG. Only amino acid pair Arg168-Thr169 (175.32 176.11)
has similar dihedral angle in both the forms of -DG.

18

S. Bhattacharya et al.

3.3 Tropical Temperature Alters MT -DG

and LCMV GP1 Interactions
Lymphocytic choriomeningitis virus appears everywhere due to the large distribution
of its host house mouse, rodents except for the Antarctica [23]. The serological
prevalence of the virus has been reported to be centralized in Africa, but disease
distribution occurs worldwide. Studies with patients suffering from lymphocytic
choriomeningitis have revealed that this disease is prevalent in temperate zone [23].
Moreover, majority of Old World viral diseases occur at tropical and temperate
regions [24]. So, we have employed MD run with a temperature range 2747 C so
that we can monitor the changes in the interaction occurring between the receptor,
-DG and its ligand, LCMV-GP1. Changes in RMSD of the conformations over time
in three different temperatures, i.e. 27, 37 and 47 C (300, 310 and 320 K, respectively) have clearly demonstrated that the complex of MT--DG and LCMV-GP1
shows less alteration in RMSD with temperature change than the complex of
WT--DG and LCMV-GP1 does. In other words, the interaction between MT--DG
and LCMV-GP1 is much more stable than its wild type counterpart (Fig. 3).

4 Discussions
LCM, an infection of meninges which serve as a protective membrane for central
nervous system, is developed due to the invasion of its causative agent LCMV to
host system via -DG, a host cell surface receptor and this interaction is largely
counted on LARGE mediated proper glycosylation of matured -DG. Additionally,

Fig. 3 Effect of temperature on the interactions. a, b RMSD versus Time plots exhibit that in case
of the complex of LCMV GP1 with wild-type receptor has shown greater deviation in RMSD for
the conformations (a). On the contrary, complex of LCMV GP1 with the mutated receptor has less
conformational changes (b) with change in temperatures

LCMV Interaction Changes with T192M Mutation

19

a mutant form of this -DG (T192M) hampers LARGE mediated glycosylation

resulting in altercation in signalling cascade essential for proper brain development.
Patients with T192M mutation in -DG develop severe mental retardations along
with cognitive impairment. These common features of the two diseases, i.e.
dependence on LARGE, involvement of -DG, proper glycosylation and above all
affecting nervous system, brain, have made the interaction of LCMV and -DG an
invincible study object to us. In our current work, we have found that amino acid
stretches 166171 of -DG in the close vicinity to mutation site of Met 192 have
contributed to structural bends which nally have created relatively more buried
conformation for MT -DG than that of wild type one. Majority of amino acid pairs
have almost 180 orientation difference in their dihedral angles which imply that
replacement of polar Thr 192 with non-polar Met has conferred structural strains to
the positional orientations of its upstream and downstream amino acids. Because of
this, mutant form of -DG has possessed a buried conformation with those amino
acid stretches which otherwise have a widened window in the wild type -DG. And
this has forced LCMV GP1 to occupy a different space for interaction with MT
-DG. But this has been proven to be advantageous for the interaction of the two
participating proteins since interaction of LCMV GP1 with -DG has more main
backboned based H-Bonds formed, compared to its interaction with WT--DG.
And these same results have also been reflected with solvent explicit models.
Interestingly, this result has an indication of likelihood of susceptibility to LCMV
infection in the presence of the T192M mutation. This has urged us to look for the
possible fold pattern of LCMV GP1 that can serve as a potential blocking site to
restrict the disease propagation. In our study, as we already have shown that stretch
of amino acids 8896 of LCMV GP1 has remained a common mediator of interactions for both WT and MT -DG and it also has possessed a particular fold
pattern. In future, analysis of this structure will enlighten a path for developing
novel therapeutic approaches against this viral transmission. Moreover, temperature-based variation on the conformations from our MD run have revealed that the
interaction of LCMV GP1 and -DG with T192M mutation is signicantly stable
than the interaction of LCMV GP1 with the wild type receptor. In conclusion, our
experiments have shed lights on the molecular interactions of LCMV GP1 with MT
-DG, for the rst time of its kind. Experimental results, here, have shown that
structural aberrations in mutant -DG owing to the altered bends in its amino acids
residues have lead to changes in its interaction pattern with LCMV GP1. But this
changed spatiotemporal orientation nally has resulted in strong binding of LCMV
GP1 with MT -DG, compared to WT--DG as is revealed by MD simulations and
hydrogen bond analysis. This indicates a likelihood of LCMV susceptibility when
-DG possesses T192M mutation. T192M mutation is a naturally occurring
mutation in human causing structural changes of the protein and subsequently
diminishing the effective signal transduction. Blocking the activity of this mutant
form with small antagonists may anyway hamper the process. Gene therapies that
deliver functional transcript of the protein may ameliorate the ill effect of the
mutation. But from this study, we have found that LCMV GP1 has stronger
interactions with the mutated form than the wild type one. That implies an effective

20

S. Bhattacharya et al.

therapeutic strategy should be developed against the viral protein to block the
interaction of LCMV GP1 with the host receptor protein. Studying the particular
folding pattern (s) of LCMV GP1 in near future may lead the path to develop new
potential therapeutic approaches to bar the spread of this viral transmission.
The only licensed drug available to certain diseases caused by Arenaviruses is
nucleoside analogue ribavirin [25]. A detailed analysis of this interaction with total
protein model will generate deeper insight to the molecular basis of the viral
attachment to mutant protein. Our future study therefore includes (i) homology
modelling of total -DG with its mutant form as well as that of LCMV GP1, (ii)
docking simulations followed by thereafter analyses of the interactions and (iii)
small molecular library screening to nd effective molecule against the viral GP1.
Acknowledgments Authors are thankful to Dept of Biochemistry and Biophysics, University of
Kalyani for their continuous support and for providing the necessary instruments to carry out the
experiments. The authors would like to thank the ongoing DST-PURSE programme. SB and AD
also are thankful to UGC, India and CSIR, India for their respective fellowships, and the DBT
(project no. BT/PR6869/BID/7/417/2012) for the necessary infrastructural support.
Conflict of Interest
The authors declare no conflict of interest.

Appendix

Supplementary
File 1 Superimposed models
built from Modbase, RaptorX
and I-TASSER

LCMV Interaction Changes with T192M Mutation

Supplementary
File 2 Superimposed
Reformed structure with its
original structure after heat
denaturation

Supplementary
File 3 Superimposed WT
and MT alpha-dystroglycan
minimized with explicit
solvent system

21

22

S. Bhattacharya et al.

References
1. Henry MD, Campbell KP (1999) Dystroglycan inside and out. Curr Opin Cell Biol
11:602607. http://dx.doi.org/10.1016/S0955-0674(99)00024-1
2. Pollard TD (1986) Mechanism of actin lament self-assembly and regulation of the process by
actin-binding proteins. Biophys J 49:149151. http://dx.doi.org/10.1016/S0006-3495(86)
83630-X
3. Cheever TR, Ervasti JM (2013) Actin isoforms in neuronal development and function. Int Rev
Cell Mol Biol 301:157213. http://dx.doi.org/10.1016/B978-0-12-407704-1.00004-X
4. Henry MD, Campbell KP (1996) Dystroglycan: an extracellular matrix receptor linked to the
cytoskeleton. Curr Opin Cell Biol 8:625631. http://dx.doi.org/10.1016/S0955-0674(96)
80103-7
5. Diner P, Balci B, Yuva Y, Talim B et al (2003) A novel form of recessive limb girdle
muscular dystrophy with mental retardation and abnormal expression of alpha-dystroglycan.
Neuromuscul Disord 13:771778. http://dx.doi.org/10.1016/S0960-8966(03)00161-5
6. Bhattacharya S, Das A, Ghosh S, Dasgupta R, Bagchi A (2014) Hypoglycosylation of
dystroglycan due to T192M mutation: a molecular insight behind the fact. Gene 537:108114.
http://dx.doi.org/10.1016/j.gene.2013.11.071
7. Spiropoulou CF, Kunz S, Rollin PE et al (2002) New world arenavirus clade C, but not clade
A and B viruses, utilizes alpha-dystroglycan as its major receptor. J Virol 76:51405146
8. Oldstone MB, Campbell KP (2011) Decoding arenavirus pathogenesis: essential roles for
alpha-dystroglycan-virus interactions and the immune response. Virology 411:170179. http://
dx.doi.org/10.1016/j.virol.2010.11.023
9. Lapoov K, Pastorekov S, Tomkov J, Lymphocytic choriomeningitis virus: invisible but
not innocent. Acta Virol 57:160170. http://dx.doi.org/10.4149/av_2013_02_160
10. Kunz S, Sevilla N, McGavern DB et al (2001) Molecular analysis of the interaction of LCMV
with its cellular receptor [alpha]-dystroglycan. J Cell Biol 155:301310. http://dx.doi.org/10.
1083/jcb.200104103
11. Shi J, Blundell TL, Mizuguchi K (2001) FUGUE: sequence-structure homology recognition
using environment-specic substitution tables and structure- dependent gap penalties. J Mol
Biol 310:243257. http://dx.doi.org/10.1006/jmbi.2001.4762
12. Berman HM (2008) The protein data bank: a historical perspective. Acta Crystallogr A
64:8895. http://dx.doi.org/10.1107/S0108767307035623
13. Altschul SF, Gish W, Miller W et al (1990) Basic local alignment search tool. J Mol Biol
215:403410. http://dx.doi.org/10.1006/jmbi.1990.9999
14. Pieper U, Webb BM, Barkan DT et al (2011) ModBase, a database of annotated comparative
protein structure models, and associated resources, Nucleic Acids Res 39(Database issue):
D465D474. http://dx.doi.org/10.1093/nar/gkq1091
15. Eisenberg D, Lthy R, Bowie JU (1997) VERIFY3D: assessment of protein models with
three-dimensional proles. Methods Enzymol 277:396404. http://dx.doi.org/10.1016/S00766879(97)77022-8
16. Laskowski RA, MacArthur MW, Moss DS et al (1993) PROCHECKa program to check the
stereochemical quality of protein structures. Appl Crystallogr 26:283291. http://dx.doi.org/
10.1107/S0021889892009944
17. Chen R, Li L, Weng Z, ZDOCK: an initial-stage protein-docking algorithm. Proteins
52:8087. http://dx.doi.org/10.1002/prot.10389
18. Jimnez-Garca B, Pons C, Fernndez-Recio J (2013) pyDockWEB: a web server for rigidbody protein-protein docking using electrostatics and desolvation scoring. Bioinformatics
29:16981699. http://dx.doi.org/10.1093/bioinformatics/btt262
19. Schneidman-Duhovny D, Inbar Y, Nussinov R, Wolfson HJ (2005) PatchDock and
SymmDock: servers for rigid and symmetric docking. Nucl Acids Res 33:W363W367

LCMV Interaction Changes with T192M Mutation

23

20. Brooks BR, Bruccoleri RE, Olafson BD et al (1983) CHARMM: A program for
macromolecular energy, minimization, and dynamics calculations. J Comput Chem
4:87217. http://dx.doi.org/10.1002/jcc.540040211
21. Tina KG, Bhadra R, Srinivasan N (2007) PIC: Protein interactions calculator. Nucleic Acids
Res 35(Web Server issue):W473W476. http://dx.doi.org/10.1093/nar/gkm423
22. Jaenicke R (1991) Protein stability and molecular adaptation to extreme conditions. Eur J
Biochem 202:715728
23. The Centre for Food Security and Public Health (2010) Institute for International Co-operation
and in Animal Biologics. Iowa State University, Ames. http://www.cfsph.iastate.edu/
Factsheets/pdfs/lymphocytic_choriomeningitis.pdf
24. Pasqual G, Rojek JM, Masin M, Chatton JY, Kunz S (2011) Old world arenaviruses enter the
host cell via the multivesicular body and depend on the endosomal sorting complex required
for transport. PLoS Pathog 7:e1002232. doi:10.1371/journal.ppat.1002232
25. Lee AM, Pasquato A, Kunz S (2011) Novel approaches in anti-arenaviral drug development.
Virology 411:163169

Structural and Functional

Characterization of Arabidopsis thaliana
WW Domain Containing Protein F4JC80
Amit Das, Simanti Bhattacharya, Angshuman Bagchi
and Rakhi Dasgupta

Abstract WW domains are the smallest known independently foldable protein

structural motifs that are involved in cellular events like protein turnover, splicing,
development, and tumor growth control. These motifs bind the polyproline rich
ligands. While the WW domains of animal origin are well characterized, the same
from plant origin are not well documented yet. Despite the small repertoire of WW
proteome of plants (in comparison to animal WW proteome) functional diversity is
reported to be equally vivid for plants also. Here, for the rst time, we report the
structural and functional properties of an Arabidopsis thaliana (At) WW domain
containing protein F4JC80 by using homology modeling and docking techniques.
Our ndings report that the At F4JC80 protein contains two WW domains which
bear the standard triple sheet structure and structurally and functionally resemble
Class I WW domains of E3 ubiquitin ligase family but their structural differences
impact their polypeptide binding abilities differently.

Keywords WW domain
Arabidopsis thaliana
Homology modeling Molecular docking

Transcription

Splicing

1 Introduction
Different cellular phenomena like protein ubiquitination, splicing, organ development, tumor progression, and suppression are associated with large multi-protein
interactions. Conserved patches of proteins, known as domains, often assign the
structural and functional properties of a protein. Due to their functional importance,
A. Das
S. Bhattacharya
A. Bagchi (&)
R. Dasgupta (&)
Department of Biochemistry and Biophysics, University of Kalyani,
Kalyani, Nadia 741235, WB, India
e-mail: [email protected]
R. Dasgupta
e-mail: [email protected]
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_3

25

26

A. Das et al.

evolutionary forces often enable protein domains to accommodate more sequence

variations keeping the overall structural folding pattern intact. One such great
example is the WW domain which is known to be the smallest protein module (only
3040 amino acids long) that is able to fold independently in solution [1]. This
domain harbors its name from the presence of two conserved tryptophan residues
residing toward the two termini of this 30 amino acids stretch [2]. Other than the
two conserved residues, there are 23 aromatic amino acids, mostly tyrosine, are
found in the middle of this domain and a conserved proline residue is located at the
end of this domain [1]. Other than these 45 conserved residues, the remaining parts
of these WW domains are found to be highly variable in nature. Despite this huge
sequence variation, this small 3040 amino acids stretch of a WW domain is always
found to fold into a triple sheet structure where the individual sheets are
generally connected by small turns [1, 2]. However, the lengths of the individual
sheets vary greatly among different WW domains. Generally, proteins or peptides
containing polyproline rich regions act as ligands of this small protein module.
Different WW domains interact with polyproline rich ligands or polypeptides
(PRL/PRP) of different compositions and lengths. Based on their PRL/PRP choices,
WW domains are classied into ve different classes [1]. Among these ve classes,
class I WW domain comprises of E3 ubiquitin ligases like NEDD4, ITCH as well as
transcription regulator like YAP1 and oxidoreductase WWOX (NEDD4: E3
ubiquitin-protein ligase NEDD4/Neural precursor cell expressed developmentally
down-regulated protein 4; ITCH: E3 ubiquitin-protein ligase Itchy; YAP1: Yorkie
homolog/65 kDa Yes-associated protein; WWOX: WW domain containing
oxidoreductase; APBB1: Amyloid beta A4 precursor protein-binding family
B member 1; FBP/FNBP: Formin binding protein; WBP: WW domain binding
protein). Proteins like FBP11 and Prp40 which are part of the eukaryotic splicing
machinery form the Class II and Class V, respectively, while prolyl isomerases like
Pin1 is the sole member of the Class IV. Other WW domain containing proteins like
FE65 (also known as APBB1) and FBP21 (also known as WBP4) form the class III
WW domains. Salah et al. [1] in their review about WW domains have mentioned
in details about the specic ligand choices of each class of WW domains [38].
Irrespective of this small length, the functional impact of WW domain is
enormous and it is found to be associated with all the above-mentioned cellular
phenomena. It is often associated with large cellular protein networks. Diseases like
Alzheimer and Huntington disease are also associated with different WW domain
containing proteins (WW-CP) whos altered or impaired interactions are associated
with these disease prognosis [1]. Like the functional diversity the distribution and
architectural diversity of WW domains are also signicantly diversied. The pfam
WW domain family (http://www.pfam.xfam.org) consists of 8,927 WW domain
sequences form 357 species with 403 different architecture.
Over the last decade, the WW domain has been structurally and functionally well
characterized. Unfortunately most of these studies whether cellular, biochemical,
structural, or at organism level, have been associated with WW domains of either
animals or fungi. Studies about plant WW domain is restricted to a handful of
articles despite the ndings that WW domains of plants are also associated with

Structural and Functional Characterization of Arabidopsis thaliana

27

diverse cellular events like inhibition of plant virus replication, cuticle development, plant flowering time control, morphological development, RNA processing,
and prolyl isomerization [914]. While the last two functions are quite common to
the WW domain family, others are specic to plants. Considering the conserveness
of different pathways involving RNA processing and prolyl isomerization, it would
not be surprising to nd out the involvement of WW domains in similar events of
plants. In this context, it is also surprising that the functional diversity of the plant
WW domains have not been well addressed yet. While plants are much less
diversied in terms of their WW proteome (Table 1) when compared to that of
animals, the WW domain-related studies of model plants like Arabidopsis thaliana
(A. thaliana) and Oryza sativa (O. sativa) are signicantly limited. Without better
understanding of the plant WW domain functions and their involvement in different
cellular events, it would not be possible to specically categorize the similarities
and differences of plant and animal WW domains, their PRL/PRP preferences as
well as their respective protein interactions. Especially characterization of those
WW domains and WW-CPs that are involved in evolutionary conserved processes
like RNA processing, prolyl isomerization, protein turnover would provide
signicant insight into better understanding of these essential cellular processes that
encompass all kinds of eukaryotes. In this context, we have undertaken homology
modeling- and docking-based studies to characterize and functionally annotate the
uncharacterized WW-CPs of model plants like A. thaliana.
As described in Table 1, the A. thaliana WW proteome consists of only 28
proteins (22 unique sequences and six redundant sequences) which has a total of 41
WW domain sequences. Among these 22 unique sequences, 8 are yet to be
functionally annotated while others were reported to be involved in most of the
above-mentioned plant WW domain functions. Among these eight uncharacterized
proteins, this study reports about the structural and functional characterization of

Table 1 Comparison of WW-CP proteome of plants and animals

Serial no.

Species name

Number of proteins with

WW domains

Number of total
WW domains

1
Homo sapiens
204
314
2
Mus musculus
139
243
3
Danio rerio
108
174
4
Drosophila melanogaster
76
139
5
Arabidopsis thaliana
28
41
6
Oryza sativa
16
25
According to the pfam database, WW domain containing proteins are more in number in animals
than to the same of plants. While vertebrates (like human, mouse, or zebra sh) contain more than
100 WW domain containing proteins and even larger amounts of WW domain sequences (on
average 1.5 WW domains per protein), variation of plant (A. thaliana and O. sativa) WW domains
is much less, even in comparison to Arthropods (fruit fly). However, the functional diversity of
plants and animals are comparable (details in text)

28

A. Das et al.

one such A. thaliana WW-CP, annotated as WW domain containing protein

(Uniprot ID: F4JC80) which harbors two WW domains. (During the rest of the
article, this protein will be mentioned as At F4JC80).

2 Materials and Methods

The amino acid sequence of At F4JC80 (Uniprot KB ID: F4JC80, NCBI Accession
number: NP_001154613.1) was retrieved from Uniprot KB (http://www.uniprot.org)
and its corresponding conserved domain regions were identied via pfam [15]. The
pfam output showed the presence of two WW domains in At F4JC80. The pfam
identied WW domain regions were further used for homology modeling and subsequent docking and functional annotation studies.
For structural and functional characterization of a protein, it is essential to
generate stereo-chemically t structures that can be further used in different studies.
Although there are couple of models of WW domains available at the Swiss Model
repository for the protein of our interest, none of the models were found to be stereochemically t. Thus, fresh models for the WW domains of At F4JC80 were required
to be built. To identify potential templates for homology modeling of the At F4JC80
WW domains, NCBI PSI-BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE=
Proteins) was performed against the protein databank (PDB) (http://www.rcsb.org)
using the At F4JC80 WW domain region sequences as query sequences. For the
BLAST search the default parameters were used (BLOSUM62 matrix, Word size: 3,
Expect threshold: 100, Gap costs: Existence: 11, Extension: 1). However, lowsequence similarity and poor e-values of the PDB PSI-BLAST outputs rendered no
direct choice of structures that can be used for homology modeling. The small length
and conserved folding patterns of WW domains often enables us to produce stereochemically t structure even with templates of low similarity. So we have used the
modbase [16] server (slow mode) for homology modeling of At F4JC80 WW
domains. Modbase while generating the homology models identies the best
template for homology modeling. Modbase generates models using ModPipe, which
is an automated modeling pipeline. ModPipe uses Modeler for fold assignment,
sequencestructure alignment, model building, and model assessment. The validations of the structures obtained as outputs from Modbase were performed using the
SAVES (v3) server with Procheck, Verify_3D, and Errat analysis [1719] as well as
through the ProSA [20] tool. Model visualization, inspection, analysis, and gure
preparation were performed using Discovery Studio Client 3.5 [Accelrys, United
States, presently BIOVIA, Dassault systems, France]. Built models were further
subjected to energy minimizations. The energy minimization processes were run
with conjugate gradient algorithm for 4,000 steps by applying the CHARMm [22]
force eld until the structures reached the nal derivative of 0.01 kcal/mole using the
Discovery Studio 2.5 package [Biovia].
Functional annotations of the At F4JC80 WW domains were performed by
structurally comparing the models with structures of known function through the

Structural and Functional Characterization of Arabidopsis thaliana

29

DALI server [21]. Only the outputs with a RMSD of <1.5 were used to assign
functions to our domains of interest.
The central groove created by the three sheets form the PRL/PRP binding
region of any known WW domain [1] and till date all the known PRLs/PRPs of
WW domains have been reported to be binding to this groove. In case of our
docking studies, only those docked complexes where the PRL/PRP was found to be
docked into this groove were considered for further analysis. Any other docking
poses of ligand-receptor complex were rejected.
We have extracted the 17 different PRL/PRP structures from different known
complexes of proteins with WW domains as receptors bound to different ligand
molecules or polypeptides which are available at the Protein Data Bank (http://
www.rcsb.org). These ligand structures were further heated to 373 K and then
cooled to 273 K to release the structural constraints, if any, that might be present in
them due to crystallization conditions or other structural limitations from the
reported structural studies. These heating and cooling procedures were performed
using the Accelrys Discovery Studio 2.5 package (Accelrys, USA; presently
BIOVIA, Dassault Systems, France). Required energy minimization for these
ligand structures were performed in Discovery Studio 2.5 using the Conjugate
Gradient algorithm. All the docking studies were performed through the Z-DOCK
server [23] (for rigid docking). Only those docked complexes that were found to
have least energy as well as stereo-chemically t, were selected for further
renement. These reliable best outputs of the Z-DOCK server were subjected to
further renement via the FlexPepDock server [24]. Since 10 different output
models are generated by FlexPepDock, similar criteria as of selecting the Z-DOCK
output were also applied to choose the output of the FlexPepDock server.
Validations of the docked complexes were performed through SAVES server using
previously mentioned analytical tools. Energy minimizations of the docked
complexes were performed as mentioned above with backbone xation to ensure
proper interactions of the docked complexes. After each minimization, the structures of the complexes were checked through Verify_3D and Procheck to determine
their stereo-chemical tness.

3 Results and Discussions

3.1 Sequence Analysis
The sequence of At F4JC80 is 892 amino acids long. It harbors two WW domains
(Fig. 1a). The N terminal WW domain is located between amino acids 186 and 213
whereas the other WW domain is located at extreme C termini of the protein
(860890 amino acids) (Fig. 1a). Both of these pfam identied WW domains were
found to contain the two conserved Trp residues as well as two aromatic Tyr
residues in the middle of each region and the conserved terminal proline residue

30
Fig. 1 Domain organization
of At F4JC80 WW domains
and structural properties of
these WW domains (left panel
for 1st WW domain and right
panel for 2nd WW domain).
a 892 amino acids long At
F4JC80 harbors two WW
domains (marked with green
boxes with their amino acid
positions marked on each
side). The pfam match
qualities of these two WW
domains are shown in details.
b The triple sheet structure
of the generated models of
these two WW domains with
each sheet numbering starting
from N terminal. c Side view
of the models shows the
difference in the upper
concave ligand binding
groove. d ProSA plot of the
energy functions of these two
WW domains validates their
structural properties. e and
f Structural alignment of these
two models with their
templates shows the areas of
differences (template shown
in yellow, major differences
are marked with red arrow
head). g Sequence view of
these model and template
structures alignment

A. Das et al.

Structural and Functional Characterization of Arabidopsis thaliana

31

(Fig. 1a). The level of condence to identify these two regions as WW domains
were very high (2.4e07 for the 1st WW domain and 8.9e06 for the 2nd WW
domain of At F4JC80). Since protein domains are the most important region of any
protein in terms of mediating its interaction with ligands, substrates, or other
proteins; presence of only WW domain in At F4JC80 rendered these two WW
domains as the sole functional motif of At F4JC80. So for further understanding of
the functional properties of At F4JC80, these two WW domains were subjected to
homology modeling.
Another important feature of At F4JC80 is the presence of polyproline rich
stretches in this protein. PRLs are known as the ligands for WW domains and based
on the composition of PRLs/PRPs, WW domains are classied into ve classes [1].
The At F4JC80 contains four polyproline stretches among which three are four
proline residues long (537540aa, 546549aa, and 568571aa) while one is ve
proline residues long (556560aa). Presence of both ligand and receptor domain
(i.e., polyproline stretches and WW domains, respectively) could have signicant
impact in determining the state of activation (through the intra-molecular interactions of these ligand and domain), oligomerization (through mediating intermolecular interactions), or binding partners (through interaction with other WW
domains or polyproline stretches of other proteins) of At F4JC80.

3.2 Homology Modeling: Modeled Structure Description

3.2.1 1st WW Domain
The 1st WW domain of At F4JC80 was build using the structure having PDB ID
2L5Fchain A as template. The stereo-chemically t model consists of three
sheets which are connected by turns (Fig. 1b, left panel). The length of each of the
three sheets is different from each other. The side view of the At F4JC80 1st WW
domain model shows a deep concave groove (Fig. 1c, left panel) which represents
the primary PRL/PRP binding region of any WW domain. Along with Ramachandran plot (via Procheck), Verify_3D, and Errat analysis (See Appendix for details of
about model quality), ProSA analysis of the model also validated its structural and
energy-related properties (Fig. 1d, left panel). Superimposition of the model with its
template identied the regions of deviations (arrow heads in Fig. 1e, f, left panel)
which resulted in a RMSD of 1.79 between the model and its template. This
difference in RMSD may indicate slight difference from its template but it is mainly
due to low sequence similarity with the template (Fig. 1g, left panel). However, this
structurally validated model showed 80 % structural overlap with its template as well
an overall Z-score of 2.27 which conrmed the good overall quality of the model
which was also represented as a point in the plot (Fig. 1d, left panel) that is in a range
solved by X-ray and NMR studies.

32

A. Das et al.

3.2.2 2nd WW Domain

The 2nd WW domain of At F4JC80 (modeled with PDB ID: 2YSB chain A as
template) is also made of three variable length sheets with a turn connecting sheet 2
and 3 (Fig. 1b, c, right panel). Unlike the model of the 1st WW domain, sheet 1 and 2 of
the 2nd WW domain model is connected by a longer loop structure and compared to
the 1st WW domain model, the PRL/PRP binding groove of the 2nd WW domain is
shallower (Fig. 1c, right panel). However, this model was also found to be stereochemically t and the ProSA energy function analysis of the model rendered an
overall Z-score of 0.54 which also falls in a range solved by X-ray and NMR
techniques. Superimposition of this model structure (Fig. 1eg, right panel) with its
template was found to be more accurate (96.67 % structural overlap to template) than
the 1st WW domain model and this also reflected in the lower RMSD of 0.74 .
Both the models of the 1st and 2nd WW domain of At F4JC80 were found to be
structurally suitable and they represented the overall common architecture of WW
domains. Even with similar architectures, WW domains are quite different in terms
of their functions and PRL/PRP choices. We have used the two above-mentioned
models to perform functional annotations to each of them.

3.3 Ligand Binding Properties and Functional Properties

of Arabidopsis thaliana F4JC80 WW Domains
WW domains bind to polyproline rich stretches and their classication is also based
on their PRL/PRP preferences. To understand the functional properties of the At
F4JC80 WW domains, we performed docking of the generated models with the
different PRLs/PRPs that are available at PDB. Renement of the docked structure
removed the structural constrains of the ligands while increasing its docking
parameters (Fig. 2). The deep ligand binding groove of the 1st WW domain model
was found to negatively impact the docking of the At F4JC80 1st WW domain
(Fig. 2a, b, Table 2) with different PRLs/PRPs. Especially it affected the formation of
Hydrogen bonds (H-bonds) in the deepest region of the groove by imposing distance
constrains and thus no H-bond was found in the middle region of the ligand that spans
over the triple sheet structure. However, the 2nd WW domain was found to be more
general in terms of its PRL/PRP choices (Table 2). Especially the shallow ligand
binding groove (Fig. 2c, d) also contributed signicantly toward interacting with
PRLs/PRPs in the form of forming more hydrogen bonds in this region. Making the
PRL/PRP structure flexible via Flexpepdock [24], this server have put the PRL/PRP
in proper pose where it can better interact with the receptor WW domain (Fig. 2e, f).
Docking studies showed that both the WW domains mainly bound to the PRLs/PRPs
of class I WW domains (comprising of WW domains from E3 ubiquitin ligases like
NEDD4, ITCH, and YAP as mentioned earlier) [1]. Beyond this, the 2nd WW domain
also showed preference toward PRLs/PRPs that primarily interacts with Class V WW
domains (which are formed by WW domains of transcription factor like Prp40) [1].

Structural and Functional Characterization of Arabidopsis thaliana

Fig. 2 Docking of At F4JC80
WW domains with their
ligands. a, b 1st WW domain
with ligand from PDB ID:
2JO9 structure, (cf) 2nd WW
domain with ligand from PDB
ID: 2V8F structure.
ad Receptor WW domain
and ligand regions is shown
as ribbon and Hydrogen
bonds are shown in green
dashed lines. c, d The
interacting surface (according
to hydrophobicity) is shown
for the 2nd WW domain.
(Due to interference of the
clear visibility of H-bonds,
similar surface is not shown
for the 1st WW domain).
e, f Front and side view,
(respectively), of the
flexpepdock output of the 2nd
WW and its ligand interaction
which highlights the variation
in the ligand backbone to
maximize the docking
qualities. (For simplicity of
representation, only one such
flexible docking output is
shown but similar approaches
to maximize docking qualities
were taken for all
combinations of successful
Z-DOCK output with both At
F4JC80 WW domains and
their ligands)

33

Ligand PDB ID

Ligand structure receptor protein

Ligand sequence

At 1st WW domain

At 2nd WW domain

1
1I5H
NEDD4-WW1
GSTILPIPGTPPPNYDSL
N/A
N
2
1K5R
YAP1
GTPPPPYTVG
N/A
Y
3
2DJY
SMURF2-WW3
GPLGSELESPPPPYSRYPMD
N/A
Y
4
2DYF
FBP11-WW1
GSTAPPLPR
N/A
N
5
2EZ5
NEDD4-WW3
TGLPSYDEALH
N/A
Y
6
2HO2
FE65
PPPPPPPPPL
N/A
Y
7
2JMF
Su(dx)-WW4
GPLGSPNTGAKQPPSYEDCIK
N/A
Y
8
2JO9
ITCH-WW3
EEPPPPYED
Y
Y
9
2JUP
FBP28-WW2
PPLIPPPP
Y
Y
10
2KQ0
NEDD4-WW3
ILPTAPPEYMEA
Y
Y
11
2KXQ
SMURF2-WW2
GPLGSELESPPPPYSRYPMD
N/A
Y
12
2LAW
YAP1
TPPPAYLPPEDP
N/A
Y
13
2LB1
SMURF2-WW2
DTPPPAYLPPEDP
Y
Y
14
2LB2
NEDD4L-WW2
ETPPPGYLSEDG
N/A
Y
15
2RLY
FBP28-WW2
PTPPPLPP
N/A
Y
16
2RM0
FBP28-WW2
PPPLIPPPP
N/A
Y
17
2V8F
PROFILIN
IPPPPPLPGV
N/A
Y
Although WW domains form a triple sheet structure, connected by turns, variation of sequences result in their variable ligand choices. We have used online
docking server Z-DOCK to nd out the ligand binding ability of both the WW domains of At F4JC80 with 17 different ligand sequences extracted from already
reported structures of different WW domainligand complexes (taken from protein database: http://www.rcsb.org). The result clearly shows that the 1st WW
domain of At F4JC80 can bind to only a handful of ligands while the 2nd WW domain binds to almost all the types of proline rich ligands that were taken into
consideration in this study. Since WW domains bind to their ligands only through their well known ligand binding groove (detailed in text), the deep groove
that we have found in case of the 1st WW domain of F4JC80 could well be the reason for the inability of this domain to bind to different ligands as such deep
grooves imposes a distance constrain to form successful atomic interactions

Serial no.

Table 2 Ligand binding choices of At F4JC80 WW domains

34
A. Das et al.

Structural and Functional Characterization of Arabidopsis thaliana

35

For better insight into the functional properties of the At F4JC80 WW domains,
we have used the DALI server for functional assessment by structural comparison.
The DALI output (only those outputs with a RMSD of <1.5 were considered)
showed that both the At F4JC80 WW domains are structurally similar to the WW
domains of NEDD4 family which also synergized with our ndings from the
docking studies.
Overall this study on the in silico characterization of At F4JC80 has identied
the presence of two WW domains in this protein as well as assigned functional
annotations two these domains by molecular docking techniques and structurebased functional assessment studies. Such computational studies which is a rst of
its kind on plant WW domains would provide a future direction on further characterization of this or similar types of proteins. Future studies addressing these
proteins from experimental approach will prove to be signicant in understanding
the overall interactome of these proteins.

4 Conclusions
This study on structural and functional properties of At F4JC80 identied presence
of two WW domains and four polyproline rich stretches in the protein. Both the
models of the WW domain were found to contain the standard triple sheet structure
but there were signicant difference in the PRL/PRP binding groove of the WW
domains. This difference caused better ability of the 2nd WW domain to bind to
PRL/PRPs as the deep groove was found to have a negative impact on the 1st WW
domains PRL/PRP binding capabilities. However, docking and structural comparison-based function annotation studies found both the WW domains belong to the
Class I WW domains (specically related to NEDD4 and SMURF family). Based on
the well-characterized roles of NEDD4 and SMURF E3 ubiquitin ligases [4], it can
be hypothesized that the At F4JC80 WW domains are most probably involved with
gene expression control by regulating protein turnover during TGF signaling. Also
the presence of nuclear localization signal (regions 746-KRTKKK-751 795-KRKR798 826-WREKVKRKRERAEKSQKKDPE846, identied through NLStradamus
server [25]), makes this protein an important player in nuclear context. Moreover,
the ability of the 2nd WW domain to bind to ligands of WW domains belonging to
Prp40 (PDB ID: 2DYF) and ca150 (PDB ID: 2RM0) as well as with formin
polyproline region (PDB ID: 2VDF) shows a wider arena of functional interactions
of At F4JC80 in cellular context where it has the possibility to directly regulate
cellular actin cytoskeleton [26] and gene transcription.
Acknowledgments The authors would like to thank the members of RD and AB laboratory for
their continuous support and critical assessments. AD and SB would like to thank CSIR (India)
and UGC (India), respectively, for their Ph.D. fellowships.
Author Contributions AD and SB have performed the modeling and docking studies and drafted
the manuscript. RD and AB have analyzed the results and prepared the nal version of the
manuscript. All the authors have read and approved the nal version of the manuscript.

36

A. Das et al.

Appendix
Details of At F4JC80 1st WW domain model template, secondary structure
and model quality

fbp21ww2
f4jc80ww1

LLSKCPWKEYKSDSGKPYYY-NSQTKESRWAKP 32
-----QWKMILHEESNQYYYWNTETGETSWELP 28
**
:.. *** *::* *: * *

(35.71 % sequence identity with template)

PSIPRED secondary structure prediction

Yellow marked residues form sheets

Structural and Functional Characterization of Arabidopsis thaliana

37

Verify_3D score: 96.77 %

Errat analysis overall quality factor: 72.77
Details of At F4JC80 2nd WW domain model template, secondary structure
and model quality

sav1ww1
f4jc80ww2

LPPGWSVDWTMRGRK-YYIDHNTNTTHWSHP 30
LPSEWQAYWDESTKKVYYGNTSTSQTSWTRP 31
**. *.. *
:* ** : .*. * *::*

(35.48 % sequence identity with template)

PSIPRED secondary structure prediction
Yellow marked residues form sheets

Verify_3D score: 100 %

Errat analysis overall quality factor: 59.091 (Supplementary Table 1)

38

A. Das et al.

Supplementary Table 1 Detailed list of ligands used for docking studies from reported WW
domainligand complex structures
Serial no.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

PDB ID
1I5H
1JMQ
1K5R
2DJY
2EZ5
2HO2
2JMF
2JO9
2JUP
2KQ0
2KXQ
2LAW
2LB1
2LB2
2RLY
2RM0
2V8F
2V8F

PDB receptor molecule

NEDD4-WW1
YAP1
YAP1
SMURF2-WW3
NEDD4-WW3
FE65
Su(dx)-WW4
ITCH-WW3
FBP28-WW2
NEDD4-WW3
SMURF2-WW2
YAP1
SMURF2-WW2
NEDD4L-WW2
FBP28-WW2
FBP28-WW2
Prolin
Prolin

Ligand sequence
GSTILPIPGTPPPNYDSL
GTPPPPYTVG
GTPPPPYTVG
GPLGSELESPPPPYSRYPMD
TGLPSYDEALH
PPPPPPPPPL
GPLGSPNTGAKQPPSYEDCIK
EEPPPPYED
PPLIPPPP
ILPTAPPEYMEA
GPLGSELESPPPPYSRYPMD
TPPPAYLPPEDP
DTPPPAYLPPEDP
ETPPPGYLSEDG
PTPPPLPP
PPPLIPPPP
IPPPPPLPGV
IPPPPPLP

References
1. Salah Z, Alian A, Aqeilan RI (2012) WW domain-containing proteins: retrospectives and the
future. Front Biosci (Landmark Ed) 17:331348. doi:http://dx.doi.org/10.2741/3930
2. Sudol M (1996) Structure and function of the WW domain. Prog Biophys Mol Biol
65:11332. doi:http://dx.doi.org/10.1016/S0079-6107(96)00008-9
3. Mouchantaf R, Azakir BA, McPherson PS, Millard SM, Wood SA, Angers A (2006) The
ubiquitin ligase itch is auto-ubiquitylated in vivo and in vitro but is protected from degradation
by interacting with the deubiquitylating enzyme FAM/USP9X. J Biol Chem 281:3873838747.
doi:http://dx.doi.org/10.1074/jbc.M605959200
4. Yang B, Kumar S (2010) Nedd4 and Nedd4-2: closely related ubiquitin-protein ligases with
distinct physiological functions. Cell Death Differ 17:6877. doi:10.1038/cdd.2009.84
5. DelMare S, Salah Z, Aqeilan RI (2009) WWOX: its genomics, partners, and functions. J Cell
Biochem 108:737745. doi:10.1002/jcb.22298
6. Sudol M, Shields DC, Farooq A (2012) Structures of YAP protein domains reveal promising
targets for development of new cancer drugs. Semin Cell Dev Biol 23:827833. doi:10.1016/j.
semcdb.2012.05.002
7. Kato Y, Miyakawa T, Kurita J, Tanokura M (2006) Structure of FBP11 WW1-PL ligand
complex reveals the mechanism of proline-rich ligand recognition by group II/III WW
domains. J Biol Chem 281:4032140329. doi:http://dx.doi.org/10.1074/jbc.M609321200
8. Lippens G, Landrieu I, Smet C (2007) Molecular mechanisms of the phospho-dependent
prolyl cis/trans isomerase Pin1. FEBS J 274:52115222. doi:http://dx.doi.org/10.1111/j.17424658.2007.06057.x

Structural and Functional Characterization of Arabidopsis thaliana

39

9. Qin J, Barajas D, Nagy PD (2012) An inhibitory function of WW domain-containing host

proteins in RNA virus replication. Virology 426:106119. doi:http://dx.doi.org/10.1016/j.
virol.2012.01.020
10. Wu R, Li S, He S, Wassmann F, Yu C, Qin G et al. (2011) CFL1, a WW domain protein,
regulates cuticle development by modulating the function of HDG1, a class IV homeodomain
transcription factor, in rice and Arabidopsis. Plant Cell 23:33923411. doi:http://dx.doi.org/
10.1105/tpc.111.088625
11. Noh YS, Bizzell CM, Noh B, Schomburg FM, Amasino RM (2004) EARLY FLOWERING 5
acts as a floral repressor in Arabidopsis. Plant J 38:664672
12. Hong F, Attia K, Wei C, Li K, He G, Su W et al (2007) Overexpression of the rFCA RNA
recognition motif affects morphologies modications in rice (Oryza sativa L.). Biosci Rep
27:225234. doi:http://dx.doi.org/10.1007/s10540-007-9047-y
13. Kang CH, Feng Y, Vikram M, Jeong IS, Lee JR, Bahk JD et al (2009) Arabidopsis thaliana
PRP40s are RNA polymerase II C-terminal domain-associating proteins. Arch Biochem Biophys
484:3038. doi:10.1016/j.abb.2009.01.004, http://dx.doi.org/10.1016/j.abb.2009.01.004
14. Landrieu I, Wieruszeski JM, Wintjens R, Inz D, Lippens G (2002) Solution structure of the
single-domain prolyl cis/trans isomerase PIN1At from Arabidopsis thaliana. J Mol Biol
320:321332. doi:http://dx.doi.org/10.1016/S0022-2836(02)00429-1
15. Punta M, Coggill PC, Eberhardt RY, Mistry J, Tate J, Boursnell C et al (2012) The Pfam
protein families database. Nucleic Acids Res (database issue) 40:D290D301. doi:http://dx.
doi.org/10.1093/nar/gkr1065
16. Pieper U, Webb BM, Barkan DT, Schneidman-Duhovny D, Schlessinger A, Braberg H et al.
(2011) ModBase, a database of annotated comparative protein structure models, and
associated resources. Nucleic Acids Res (database issue) 39:D465474. doi:http://dx.doi.org/
10.1093/nar/gkq1091
17. Lthy R, Bowie JU, Eisenberg D (1992) Assessment of protein models with three-dimensional
proles. Nature 356:8385. doi:http://dx.doi.org/10.1038/356083a0
18. Bowie JU, Lthy R, Eisenberg D (1991) A method to identify protein sequences that fold into
a known three-dimensional structure. Science 253:164170. doi:http://dx.doi.org/10.1126/
science.1853201
19. Colovos C, Yeates TO (1993) Verication of protein structures: patterns of nonbonded atomic
interactions. Protein Sci 2:15111519. doi:http://dx.doi.org/10.1002/pro.5560020916
20. Wiederstein M, Sippl MJ (2007) ProSA-web: interactive web service for the recognition of
errors in three-dimensional structures of proteins. Nucleic Acids Res 35:W407410. doi:http://
dx.doi.org/10.1093/nar/gkm290
21. Holm L, Rosenstrm P (2010) Dali server: conservation mapping in 3D. Nucleic Acids Res
38:W545549. doi:http://dx.doi.org/10.1093/nar/gkq366
22. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan S, Karplus M (1983)
CHARMM: a program for macromolecular energy, minimization, and dynamics calculations.
J Comp Chem 4:187217. http://dx.doi.org/10.1002/jcc.540040211
23. Pierce BG, Hourai Y, Weng Z (2011) Accelerating protein docking in ZDOCK using an
advanced 3D convolution library. PLoS One 6(9):e24657. doi:http://dx.doi.org/10.1371/
journal.pone.0024657
24. London N, Raveh B, Cohen E, Fathi G, Schueler-Furman O (2011) Rosetta FlexPepDock web
serverhigh resolution modeling of peptide-protein interactions. Nucleic Acids Res (web
server issue) 39:W249253. doi:10.1093/nar/gkr431.doi, http://dx.doi.org/10.1093/nar/gkr431
25. Nguyen, BAN, Pogoutse A, Provart N, Moses AM (2009) NLStradamus: a simple Hidden
Markov model for nuclear localization signal prediction. BMC Bioinformatics 10(1):202. doi:
http://dx.doi.org/10.1186/1471-2105-10-202
26. Blanchoin,L, Staiger CJ (2010) Plant formins: diverse isoforms and unique molecular
mechanism. Biochem Biophys Acta 1803:201206. doi:10.1016/j.bbamcr.2008.09.015, http://
dx.doi.org/10.1016/j.bbamcr.2008.09.015

Structural Insights into IbpAIbpB

Interactions to Predict Their Roles in Heat
Shock Response
Sanchari Bhattacharjee, Rakhi Dasgupta and Angshuman Bagchi

Abstract Cells respond to stress conditions. As a result of stress, most genes are
deactivated, while a few are activated with antistress response. The latter involves a
variety of molecules including molecular chaperones or heat shock proteins (Shps)
whose levels get increased in stressed conditions, particularly at elevated temperatures. Heat shock proteins help the other cellular proteins to achieve their native
states, i.e. correct folding or functional conformations. Thus, heat shock proteins
play a major role in protein homeostasis network of the cell. Small heat shock
proteins (sHsps) are one of the families of molecular chaperones that prevent the
irreversible aggregation and assist in the refolding of denatured proteins. Two
members of the sHsp family, IbpA and IbpB, are present in Escherichia coli. The
IbpA and IbpB proteins are 48 % identical at the amino acid sequence level and
have the characteristic -crystalline domain. It is known that the cooperation
between IbpA and IbpB is crucial for their chaperone activity in heat stressed
condition. So far, the molecular mechanisms of the stress response of the IbpA/
IbpB protein system have not been well understood. In the present work, an attempt
has been made to identify the amino acid residues of the IbpA and IbpB proteins,
which are found to be involved in proteinprotein interactions. The interactions
between IbpA and IbpB are studied with and without the presence of substrate
Lactate Dehydrogenase (LDH) at cold shock, physiological and heat shock temperatures to observe the changes in the pattern of interaction. This study is the rst
report to elucidate the mechanism of interactions between the proteins.
Keywords Small heat shock proteins
IbpaIbpB interaction

Heat stress
Heat shock temperature

S. Bhattacharjee
R. Dasgupta (&)
A. Bagchi (&)

Department of Biochemistry and Biophysics, University of Kalyani, Nadia 741235,
West Bengal, India
e-mail: [email protected]
A. Bagchi
e-mail: [email protected]
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_4

41

42

S. Bhattacharjee et al.

1 Introduction
Cells react to various types of physical (e.g. heat) or chemical (anoxia, low pH)
stresses. Cell stresses are frequent and recurring which challenge the cells leading
ultimately to stress response. Most of the genes are deactivated as a result of cellular
stress response, but only a few become activated to combat it. The antistress
response mechanism involves a variety of molecules including molecular chaperones and proteases or heat shock proteins (Shps) whose levels increase particularly
at elevated temperatures [1]. Heat Shps help the other cellular proteins to achieve
the native folding conformations, to localize them at their destined sub-cellular
organelles, to prevent their denaturation by heat stresses and retrieve native folding
states after partial denaturations. Thus, heat Shps play a vital role in cellular protein
homeostasis network [2].
Small heat shock proteins (sHsps) belong to the families of molecular chaperones that assist in the refolding of denatured proteins by resisting the irreversible
aggregation [3]. Small heat Shps are wide in distribution. They are characterized by
their low molecular weight (1230 kDa) and the presence of a conserved -crystalline domain [4]. Their robust upregulation under temperature-stressed conditions
make them among the most abundant cellular proteins [5].
Two members of sHsp family, IbpA and IbpB are found in Escherichia coli.
These two proteins are 48 % identical at amino acid sequence level [6] and have the
characteristic -crystalline domain flanked by N- and C-terminal ends [7]. Initially,
both are identied as inclusion body proteins but later they are found in protein
aggregates in temperature-stressed cells [8]. The efciency of IbpA/IbpB depends
on increased temperature with protein reactivation and protection from degradation
found at elevated temperatures [9]. It is reported that the presence of IbpA during
heat denaturation of substrate is sufcient to assist to change the macroscopic
properties of aggregates, yet this alone does not increase the efciency of the
subsequent reactivation of such aggregated polypeptides. The presence of IbpB is
required to assist to increase the efciency of disaggregation and refolding [10].
This observation depicts the cooperative mode of interaction of IbpA and IbpB, in
which IbpA associates rst with the aggregating protein or substrate and then
attracts IbpB to the complex [11]. So the cooperation between IbpA and IbpB is
crucial for their chaperone activity in temperature-stressed condition. So far, the
molecular mechanisms of the stress response of the IbpA/IbpB protein system have
not been well understood. In the present work, we tried to elucidate the interaction
proles between the proteins at the molecular level. We identied the amino acid
residues of the IbpA and IbpB proteins which are found to be involved in
proteinprotein interactions. The interactions between IbpA and IbpB are studied
with and without the presence of substrate Lactate Dehydrogenase (LDH) [12] at
cold shock, physiological and heat shock temperatures to observe the changes in the
pattern of interaction following heat and cold stress. This study is the rst report to
elucidate the mechanism of interactions between the proteins.

Structural Insights into IbpAIbpB Interactions

43

2 Materials and Methods

2.1 Sequence Homology Search
First, the amino acid sequences of IbpA and IbpB, comprising of 137 and 142
amino acid residues, respectively, were extracted from UniProt [13] (id:P0C054 and
id:P0C058, respectively). This was then used as input sequence to search for the
template in PDB [14] using PSI-BLAST [15]. In both the cases, the A chain of
crystal structure of an eukaryotic small heat shock protein (PDB code:1GME) [16]
was obtained as the closest match.

2.2 Homology Modelling of the Proteins

This structure (1GME, chain A) was then used as a template to build the homology
models for IbpA and IbpB separately in Accelrys Discovery Studio Modeller
(version 2.5) [17]. Alternatively, the models of IbpA and IbpB were also built
independently using RaptorX [18] and Phyre2 [19] homology modelling servers.
The models so generated were energy minimized to ensure proper interactions
applying CHARMM force eld [20] using conjugate gradient algorithm [21] until
the structures reached the nal derivative of 0.001 kcal/mol. The stereochemical
qualities of the resultant structures were then checked with PROCHECK [22] and
Verify3D [23] and Ramachandran plots [24] were drawn. No residues were found
to be present in the disallowed regions of the Ramachandran plot.

2.3 Molecular Docking Simulations with IbpA and IbpB

The model of IbpA was docked with the model of IbpB using Zdock [25], PyDock
[26], Patchdock [27], GrammX [28] and Cluspro [29] servers in order to get a
comprehensive result. Docked complexes with best score were selected and the
stereochemical qualities of the models were checked with PROCHECK and Verify3D. The docked complex with the best PROCHECK and Verify3D score was
selected as the nal working model. The nal docked complex was then energy
minimized to ensure proper interactions. 918 cycles of steepest descent (SD) energy
minimization [30] were performed in explicit solvent system [31] applying
CHARMM force eld in GROMACS version 4.6.5 [32] until the structure reached
the nal derivative of 0.001 kcal/mol. The stereochemical qualities of minimized
docked complex were again tested with PROCHECK and Verify3D.

44

S. Bhattacharjee et al.

Next, the crystal structure of the substrate, i.e. LDH was retrieved from PDB
(PDB code-3H3F) [33] and the stereochemical qualities were tested with PROCHECK and Verify3D. Then, this structure was again docked with docked complex
of IbpAB using Zdock, PyDock, Patchdock, GrammX and Cluspro. The stereochemical qualities of the docked complexes so generated were tested with PROCHECK and Verify3D. Docked complex with the best score was selected as the
nal working model and was then energy minimized to ensure proper interactions
using SD algorithm in explicit solvent system applying CHARMM force eld in
GROMACS version 4.6.5 until the structure reached the nal derivative of
0.001 kcal/mol. The docked complex was minimized in 875 steps using SD
algorithm.

2.4 Molecular Dynamics Simulation of the Complexes

In order to study the stress response both the minimized docked complexes (i.e.
IbpAIbpB and IbpAB-LDH) were then elevated from cold shock temperature
(300 K) to physiological (310 K) and heat shock temperature (318 K) [34] separately and 20 ns of production runs were performed in GROMACS version 4.6.5 in
each of the cases, keeping pressure constant. The conformational changes were
recorded and plotted against time. Further analyses were done using Discovery
Studio Package version 2.5.

3 Results
Analyses of docked complexes have revealed that, irrespective of the presence or
absence of substrate, IbpA and IbpB have followed a denite pattern of interactions
following cold shock and heat shock temperature in contrast to physiological
temperature. It has been observed that, in case of IbpAIbpB docked complex, 12
amino acid residues are involved in forming H-bond interactions in cold shock
temperature (300 K) (Fig. 1). The interactions go down at the physiological temperature of (310 K) which involve only 10 amino acid residues (Fig. 2). However,
with rise in temperature from 300 to 310 K the previous 12 amino acid residues
which were involved in H-bond interactions in cold shock temperature are restored
when the docked complex is elevated to heat shock temperature (318 K) (Fig. 3)
(Table 1).
In presence of the substrate LDH, it has been observed that the IbpAB-LDH
docked complex has 18 amino acid residues forming H-bond interactions in cold
shock temperature, among which 8 amino acid residues from IbpA/B interact with
the substrate LDH and the remaining 10 residues show inter-protein H-bond

Structural Insights into IbpAIbpB Interactions

45

Fig. 1 Amino acid residues from IbpA and IbpB proteins involved in intermolecular H-bond
interaction in IbpAB protein complex in absence of substrate at cold shock temperature (300 K)
[backbone of IbpA is marked in orange and backbone of IbpB is marked in red. H-bonds are
marked in green dashed lines] (the relevant interacting residues that are unique to this
conformation are shown)

Fig. 2 Amino acid residues from IbpA and IbpB proteins involved in intermolecular H-bond
interaction in IbpAB protein complex in absence of substrate at physiological temperature (310 K)
[backbone of IbpA is marked in orange and backbone of IbpB is marked in red. H-bonds are
marked in green dashed lines] (the relevant interacting residues that are unique to this
conformation are shown)

Fig. 3 Amino acid residues from IbpA and IbpB proteins involved in intermolecular H-bond
interaction in IbpAB protein complex in absence of substrate at heat shock temperature (318 K)
[backbone of IbpA is marked in orange and backbone of IbpB is marked in red. H-bonds are
marked in green dashed lines] (the relevant interacting residues that are unique to this
conformation are shown)

interactions (Fig. 4). When this docked complex (IbpAB with substrate LDH) is
brought to physiological temperature, only 15 amino acid residues are involved in
H-bond interactions. Of these 15 amino acid residues, 6 residues interact with LDH
and the rest of them show inter-protein interactions (Fig. 5). When the complex
again is elevated to heat shock temperature, the docked complex restores the
interaction pattern as it shows 18 amino acid residues participating in H-bond
interactions, among which 8 amino acid residues interact with the substrate LDH
and the remaining 10 amino acid residues have inter-protein interactions (Fig. 6)
(Table 2).

A:GLU82:OE2

A:TYR85:O

A:GLN88:OE1

A:GLY33:O

A:GLY32:O

A:GLY32:O

B:ASN25:HD21

B:TYR35:HN

B:ASN36:HD22

B:ARG131:HE

B:ARG131:HH12

B:ARG131:HH22

A:ALA54:O

B:ALA22:HN

A:GLU92:OE2

A:ALA91:O

B:ALA20:HN

A:GLU82:O

B:TYR35:O

A:GLN88:HN

B:GLN24:HE22

B:GLN24:OE1

A:LYS81:HN

B:ALA22:HN

Donor

Donor

B:ARG131:HH21

B:ARG131:HE

B:TYR35:HN
A:GLY32:O

A:GLY32:O

A:LEU86:O

A:GLU82:OE2

A:THR84:OG1

B:ASN25:HD21

A:GLU82:O

B:ASN25:HN

A:GLU92:OE2

B:ASN36:OD1

B:TYR35:O

B:ASN21:OD1

Acceptor

B:GLN24:HE22

B:ALA22:HN

A:GLN88:HE21

A:GLN88:HN

A:TYR34:HH

310 K

Acceptor

300 K

B:ARG131:HH22

B:ARG131:HH12

B:ARG131:HE

B:ASN36:HD22

B:TYR35:HN

B:ASN25:HD21

B:GLN24:HE22

B:ALA22:HN

B:ALA22:HN

B:ALA20:HN

A:GLN88:HN

A:LYS81:HN

Donor

318 K

A:GLY32:O

A:GLY32:O

A:GLY33:O

A:GLN88:OE1

A:TYR85:O

A:GLU82:OE2

A:GLU82:O

A:GLU92:OE2

A:ALA54:O

A:ALA91:O

B:TYR35:O

B:GLN24:OE1

Acceptor

Table 1 Amino acid residues from IbpA and IbpB proteins involved in interactions in the IbpAB protein complex in the absence of substrate at cold shock,
physiological and heat shock temperatures [interacting residues that remain unchanged at these three temperatures are marked in bold pattern. IbpA is denoted
as A, IbpB is denoted as B]

46
S. Bhattacharjee et al.

Structural Insights into IbpAIbpB Interactions

47

Fig. 4 Amino acid residues from IbpA and IbpB proteins involved in intermolecular H-bond
interaction in IbpAB protein complex in presence of substrate LDH at cold shock temperature
(300 K) [backbone of IbpA is marked in orange, backbone of IbpB is marked in red and backbone
of substrate LDH is marked in blue. H-bonds are marked in green dashed lines] (the relevant
interacting residues that are unique to this conformation are shown)

Fig. 5 Amino acid residues from IbpA and IbpB proteins involved in intermolecular H-bond
interaction in IbpAB protein complex in the presence of substrate at physiological temperature
(310 K) [backbone of IbpA is marked in orange, backbone of IbpB is marked in red and backbone
of substrate LDH is marked in blue. H-bonds are marked in green dashed lines] (the relevant
interacting residues that are unique to this conformation are shown)

4 Discussions
Escherichia coli small heat Shps, IbpA and IbpB, function as molecular chaperones
and protect misfolded proteins against irreversible aggregation as well as help the
unfolded proteins to restore their native conformations following heat stress.
In our study, it is observed that irrespective of presence or absence of substrate
LDH, the number of interactions increases both in heat shock and cold shock temperatures in comparison to physiological temperature. In presence of substrate LDH, at
cold shock temperature (300 K) 8 amino acid residues of IbpA and IbpB interact with
substrate LDH and 10 residues show inter-protein interactions. Among these 8

48

S. Bhattacharjee et al.

Fig. 6 Amino acid residues from IbpA and IbpB proteins involved in intermolecular H-bond
interaction in IbpAB protein complex in the presence of substrate at heat shock temperature
(318 K) [backbone of IbpA is marked in orange, backbone of IbpB is marked in red and backbone
of substrate LDH is marked in blue. H-bonds are marked in green dashed lines] (the relevant
interacting residues that are unique to this conformation are shown)

residues, IbpA is involved in seven interactions, whereas the remaining one is

contributed by IbpB. At physiological temperature (310 K), only two amino acid
residues of IbpB interact with the substrate, whereas four amino acid residues of IbpA
interact with the same and only 9 residues show inter-protein interactions. When the
complex is brought to heat shock temperature (318 K), 8 amino acid residues of IbpAB
interact with substrate LDH, of which ve residues are contributed by IbpA and
remaining three residues by IbpB, whereas at heat shock temperature, another 10
residues are involved in inter-protein interactions. In each of the cases, the number of
interacting residues is more in IbpA than IbpB, which suggests that IbpA interacts with
the substrate rst, thus allowing it to be more exposed to interact with IbpB to make
efcient refolding of heat stressed substrate. It is also noted that inter-protein interaction increases at cold shock and heat shock temperature in comparison to physiological temperature, which indicates that in heat stressed condition, the cooperative
interaction between IbpA and IbpB is also important for their chaperone activity. So
far, this is the rst study to elucidate the molecular mechanism of the binding interactions of IbpA and IbpB at heat stress conditions. This work would therefore be
essential to predict the hitherto unknown molecular mechanisms of IbpAB
interactions.

A:GLN88:OE1

A:GLY32:O

A:GLY32:O

B:ASN36:HD22

B:ARG131:HE

B:ARG131:HH21

B:ARG131:HH21

B:ARG131:HE
A:GLY32:O

A:GLY32:O

A:THR84:OG1
A:LEU86:O

B:ASN25:HN

A:GLU82:O

A:GLU92:OE2

A:TYR34:OH

B:TYR35:O

B:GLN24:OE1

C:PHE70:O

C:ASP63:O

A:GLN100:OE1

B:TYR35:HN

B:GLN24:HE22

B:ALA22:HN

B:ASN21:HD21

A:GLN88:HN

A:GLU82:HN

A:ARG48:HH22

A:ASN38:HD22

C:TYR246:HH

A:LEU101:O

B:ALA100:O

B:SER135:O

Acceptor

B:ARG131:HH21

B:ARG131:HE

B:TYR35:HN

B:ASN25:HN

B:GLN24:HE22

B:ALA22:HN

B:ASN21:HD22

B:MET1:H2

A:ARG123:HH12

A:GLU82:HN

A:LYS81:HN

A:ARG48:HH22

A:ASP43:HN

C:TYR246:HH

C:LYS244:HZ2

C:HIS66:HE2

C:LYS58:HZ3

A:GLY32:O

A:GLY32:O

A:LEU86:O

A:THR84:OG1

A:GLU82:O

A:GLU92:OE2

A:TYR34:OH

A:ASN111:OD1

C:TRP249:O

B:GLN24:OE1

B:GLN24:OE1

C:PHE70:O

C:MET40:O

A:GLN100:OE1

A:LEU101:O

B:ALA100:O

B:SER135:O

Acceptor
B:SER135:O

Donor
C:LYS58:HZ1

318 K

Interacting residues that remain unchanged at these three temperatures are marked in bold pattern. Amino acid residues from IbpA and IbpB of IbpAB protein complex that interact
with the substrate are marked in blue. IbpA is denoted as A, IbpB is denoted as B and substrate is denoted as C

A:GLU82:O

A:THR84:OG1

B:ASN25:HN

A:GLU92:OE2

B:GLN24:HE22

A:TYR34:OH

B:ALA22:HN

C:TRP249:O

A:ARG123:HH12

B:ASN21:HD21

C:LYS244:O

A:GLN100:HE22

A:ASN111:OD1

B:GLN24:OE1

A:LYS81:HN

A:ASN111:OD1

C:PHE70:O

A:ARG48:HH22

B:MET1:H1

A:GLN100:O

C:TRP249:HE1

B:MET1:H3

A:GLN100:OE1

C:TYR246:HH

C:LYS244:HZ2

C:HIS66:HE2

A:PRO126:O

A:LEU101:O

C:LYS58:HZ1

B:ALA100:O

C:HIS66:HE2

C:HIS180:HE2

Donor

Donor

C:LYS244:HZ2

310 K

Acceptor

300 K

Table 2 Amino acid residues from IbpA and IbpB proteins involved in interactions in the IbpAB protein complex in the presence of substrate at cold shock,
physiological and heat shock temperature

Structural Insights into IbpAIbpB Interactions

49

50

S. Bhattacharjee et al.

Acknowledgments The authors are thankful to Department of Biochemistry and Biophysics and
BIF centre, University of Kalyani for their continuous support and for providing the necessary
instruments to carry out the experiments. The authors would like to acknowledge the ongoing DSTPURSE programme (20122015) and DBT (project no. BT/PR6869/BID/7/417/2012) for support.
Conflict of Interest
The authors declare no conflict of interests.

References
1. Gill RT, Valdes JJ, Bentley WE (2000) A comparative study on global stress gene regulation
in response to over-expression of recombinant proteins in Escherichia coli. Metab Eng
2:178189
2. Thomas JG, Baneyx F (1998) Roles of the Escherichia coli small heat shock proteins IbpA
and IbpB in thermal stress management: comparison with ClpA, ClpB, and HtpG in vivo.
J Bacteriol 180:51655172
3. Kitagawa M, Matsumura Y, Tsuchido T (2000) Small heat shock proteins, IbpA and IbpB, are
involved in resistances to heat and superoxide stresses in Escherichia coli. FEMS Microbiol
Lett 184:165171
4. Kuczyska-Winik D, Kdzierska S, Matuszewska E, Lund P, Taylor A, Lipinska B, Laskowska
E (2002) The Escherichia coli small heat shock proteins IbpA and IbpB prevent the
aggregation of endogenous proteins denatured in vivo during extreme heat shock.
Microbiology 148:17571765
5. Carri MM, Villaverde A (2003) Role of molecular chaperones in inclusion body formation.
FEBS Lett 537:215221
6. Strzecka J, Chrusciel E, Grna E, Szymanska A, Zietkiewicz S, Liberek K (2012) Importance
of N- and C-terminal regions of IbpA, Escherichia coli small heat shock protein, for chaperone
function and oligomerization. J Biol Chem 287:28432853. doi:10.1074/jbc.M111.273847
7. Van Montfort R, Slingsby C, Vierling E (2001) Structure and function of the small heat shock
protein -crystallin family of molecular chaperones. Adv Protein Chem 59:105156
8. Allen SP, Polazzi JO, Gierse JK, Easton AM (1992) Two novel heat shock genes encoding
proteins produced in response to heterologous protein expression in Escherichia coli.
J Bacteriol 174:69386947
9. Lee GJ, Roseman AM, Saibil HR, Vierling E (1997) A small heat shock protein stably binds
heat-denatured model substrates and can maintain a substrate in a folding-competent state.
EMBO J 16:659671
10. Jiao W, Qian M, Li P, Zhao L, Chang Z (2005) The essential role of the flexible termini in the
temperature-responsiveness of the oligomeric state and chaperone-like activity for the
polydisperse small heat shock protein IbpB from Escherichia coli. J Mol Biol 347:871884
11. Kuczyska-Winik D, Kedzierska S, Matuszewska E, Lund P, Taylor A, Lipiska B,
Laskowska E (2002) The Escherichia coli small heat shock proteins IbpA and IbpB prevent
the aggregation of endogenous proteins denatured in vivo during extreme heat shock.
Microbiology 148:17571765
12. Motohashi K, Watanabe Y, Yohda M, Yoshida M (1999) Heat-inactivated proteins are rescued
by the DnaKJ-GrpE set and ClpB chaperones. Pnas 13:71847189. doi:10.1073/pnas.96.13.
7184
13. Leinonen R, Diez FG, Binns D, Fleischmann W, Lopez R, Apweiler R (2004) UniProt archive.
Bioinformatics 20(17):32363237. doi:10.1093/bioinformatics/bth191 PMID15044231
14. Berman HM (2008) The Protein Data Bank: a historical perspective. Acta Crystallogr Sect A
Found Crystallogr A64(1):8895. doi:10.1107/S0108767307035623, PMID 18156675

Structural Insights into IbpAIbpB Interactions

51

15. Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K, Madden TL (2009)

BLAST+: architecture and applications. BMC Bioinformatics 10:421. doi:10.1186/14712105-10-421, PMID 20003500
16. Van Montfort RLM, Basha E, Friedrich KL, Slingsby C, Vierling E (2001) Crystal structure
and assembly of a eukaryotic small heat shock protein. Nat Struct Biol 8:1025. doi:10.1038/
nsb722, PubMed: 11702068
17. Eswar N, Marti-Renom MA,Webb B, Madhusudhan MS, Eramian D, Shen M, Pieper U, Sali
A (2006) Comparative protein structure modeling with MODELLER. In: Current protocols in
bioinformatics, vol 15, Wiley, New York, pp 5.6.15.6.30
18. Peng J, Jinbo X, Raptor X (2011) Exploiting structure information for protein alignment by
statistical inference. Proteins 79:16171. doi:10.1002/prot.23175, PMC 3226909, PMID
21987485
19. Bennett-Lovsey RM, Herbert AD, Sternberg MJE, Kelley LA (2007) Exploring the extremes
of sequence/structure space with ensemble fold recognition in the program Phyre. Proteins
Struct Funct Bioinf 70(3):611. doi:10.1002/prot.21688
20. Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan S, Karplus M (1983)
CHARMM: a program for macromolecular energy, minimization, and dynamics calculations.
J Comp Chem 4(2):187217. doi:10.1002/jcc.540040211
21. Fletcher R, Powell MJD (1963) A rapidly convergent descent method for minimization.
Comput J 6:163168
22. Laskowski RA, MacArthur MW, Moss DS, Thornton JM (1993) PROCHECK: a program to
check the stereochemical quality of protein structures. J Appl Cryst 26:283291
23. Lthy R, Eisenberg JU, Bowie D (1992) Assessment of protein models with three-dimensional
proles. Nature 356(6364):8385
24. Ramachandran GN, Ramachandran C, Sasisekharan V (1963) Stereochemistry of polypeptide
chain congurations. J Mol Biol 7:9599. doi:10.1016/S0022-2836(63)80023-6 PMID
13990617
25. Chen R, Weng Z (2003) ZDOCK: an initial-stage protein-docking algorithm. Proteins
52:8087
26. Jimenez-Garcia B, Pons C, Fernandez-Recio J (2013) pyDockWEB: a web server for rigid
body protein-protein docking using electrostatics and desolvation scoring. Bioinformatics 29
(13):16981699
27. Schneidman-Duhovny D, Inbar Y, Nussinov R, Wolfson HJ (2005) PatchDock and
SymmDock: servers for rigid and symmetric docking. Nucleic Acids Res 33:W363W367
28. Tovchigrechko A, Vakser IA (2006) GRAMM-X public web server for proteinprotein
docking. Nucleic Acids Res 34:W310W314
29. Comeau SR, Gatchell DW, Vajda S, Camacho CJ (2004) ClusPro: a fully automated algorithm
for protein-protein docking. Nucleic Acids Res 32(Web Server issue):W96-9. PMID
15215358
30. Deift P, Zhou X (1993) A steepest descent method for oscillatory Riemann-Hilbert problems.
Asymptotics for the MKdV equation. Ann Math 137(2):295368. doi:10.2307/2946540
31. Lee MS, Salsbury FR, Olson MA (2004) An efcient hybrid explicit/implicit solvent method
for biomolecular simulations. J Comput Chem 25(16):19671978. doi:10.1002/jcc.20119
PMID 15470756
32. Hess B, Kutzner C, Van Der Spoel D, Lindahl E (2008) GROMACS 4: algorithms for highly
efcient, load-balanced, and scalable molecular simulation. J Chem Theory Comput 4
(2):435447. doi:10.1021/ct700301
33. Swiderek K, Panczakiewicz A, Bujacz A, Bujacz G, Paneth P (2009) Modelling of isotope
effects on binding oxamate to lactic dehydrogenase. J Phys Chem B 113(38):1278212789.
doi:10.1021/jp903579
34. Spiess C, Beil A, Ehrmann M (1999) A temperature-dependent switch from chaperone to
protease in a widely conserved heat shock protein. Cell 97(3):339347

Improving the Performance

of Multi-parameter Patient Monitor
System Using Additional Features
S. Premanand, C. Santhosh Kumar and A. Anand Kumar

Abstract Multi-parameter patient monitor (MPM) keep track of the condition of a

patient in intensive care units (ICU) or general wards using the human vital
parameters, heart rate, blood pressure, respiration rate and oxygen saturation
(SpO2). A high accuracy for the overall classication, specicity and sensitivity is
extremely important in providing quality health care to the patients. Support vector
machine (SVM) is a powerful supervised algorithm that is effectively used in
MPMs for classication. A careful study of the vital parameters in a healthy person
reveals that there exists an intrinsic relationship between the four vital parameters,
for example when heart rate is on the higher side, blood pressure is expected to be
on the lower side and vice versa. Hence, it would be highly required to understand
the correlation between the vital parameters and to integrate it into the MPM
system. In this work, we present the results of the MPM using the SVM as back-end
classier. Further, we use correlation features (feature expansion) along with base
parameters in an effort to improve the performance of MPM and note that the
performance of the MPM enhanced signicantly.
Keywords Multi-parameter patient monitor
expansion Intersection kernel

SVM
Vital parameters
Feature

1 Introduction
Multi-parameter patient monitoring (MPM) plays an important role in ensuring
quality health care in the intensive care units (ICU) and general wards to continuously monitor patients vital parameters, heart rate, blood pressure, respiratory rate
S. Premanand
C. Santhosh Kumar
Machine Intelligence Research Laboratory, Department of Electronics
and Communication Engineering, Amrita Vishwa Vidyapeetham, Ettimadai 641112, India
A. Anand Kumar (&)
Department of Neurology, Amrita Institute of Medical Sciences, Cochin 682041, India
e-mail: [email protected]
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_5

53

54

S. Premanand et al.

and oxygen saturation (SpO2) and alert as and when the condition of the patient
deteriorates. Studies of vital parameters to show evidence of physiological deterioration even before the patients abnormal condition, which leads to improvement
in mortality rates [1]. The current early warning score (EWS) [2] systems assign
scores to each vital sign based on clinical experience, depending on its diversion
away from some assumed normal range. If the scores for a vital parameters, exceeds
threshold, then a clinical analysis of the patient is prompted. There is a substantial
error rate associated with manual scoring and they do not consider the intrinsic
relationship between the vital parameters that exist in healthy people [36].
Machine learning techniques can be applied to detect physiological deterioration
in patients health, with high accuracy as compared to the EWS system [7]. MPM
uses support vector machine (SVM) classication, found to be effective in
providing state-of-the-art performance [8, 9]. The vital parameter data gathered
from the bedside monitors can be utilized to train an MPM system that could be
used to effectively predict the condition of a patient under observation. One class
SVMs [10] using the prior intelligence gathered from some assumed normal
behaviour was found to be effective in the development of MPMs. However, when
sufcient samples from the abnormal condition is available, it was seen that the
two class SVMs [11] outperform one class SVMs.
Studies on the vital parameters in a healthy person suggests that there exists a
well-established intrinsic relationship [35] between the four vital parameters, for
example, when heart rate tend to increase, blood pressure is expected to get lower
and vice versa.
In this work, we capture the correlation between human vital parameters, to
capture the intrinsic relationship between the parameters for diagnosis more
accurately and also achieve higher sensitivity, specicity and overall classication
accuracy. We experimented with SVM algorithm [8] using linear, non-linear and
homogeneous kernels for constructing a model to verify the effectiveness of the
proposed approach and to enhance the operation of the MPMs. The system
developed in the present work uses MIMIC-II database to determine the developments in the vital parameters.

2 Database Used for the Experiments

Experiments were performed on the vital parameters, obtained from MIMIC-II
database [12, 13]. MIMIC-II consists of two major parts, namely clinical data and
physiological waveforms. From bedside patient monitors (Component Monitoring
System Intellivue MP-70; Philips Healthcare), the physiological waveforms are
collected, used in medical, surgical, coronal care and neonatal ICUs in a tertiary
hospital. The waveform database includes high resolution (125 Hz) waveforms
(e.g., ECG), derived numeric time series such as heart rate, blood pressure,
respiration rate and oxygen saturation [12]. The database also consists of monitor
generated alarms along with the physician annotated les.

Improving the Performance of Multi-parameter Patient Monitor

55

3 Support Vector Machine

Support Vector Machine [8] is a set of supervised learning methods used for
classication and regression [14], given by a set of data with associated labels.
SVM constructs a hyperplane that separates two classes, and the separating hyperplane is chosen to maximize the margin between the examples of two classes.
Breaking up the classes with a big margin, minimizes a bound on expected
generalization error (chance of predicting error during classication). On the
bounding planes, the points in the data set which are falling are called support
vectors, sv. If the given data points are non-linear kernel, k(x, y), where x and y are
the points in the dataset with labels naming, lx and ly, with lx, ly 1. Classication
of the test vector x is dened by,
lx sign

M
X

!
ai kx; svi

i1

where sv(i) means ith support vector, M is the number of support vectors and a be
the dual representation of separating hyperplanes normal vector.
In this work, we experiment with basic kernels, linear, polynomial and RBF
kernels and the additive kernels [15] like intersection [16, 17], Chi-square [15] and
JS kernels [18], in the SVM backend used in the MPM system.
1. Linear kernel: KLinear x; y xT y
2. Polynomial kernel: KPolynomial x; y 1 xT yd for any d [ 0


Z
1jjxyjj
3. RBF kernel: KRBF x; y exp
2r2
4. Intersection kernel: KIntersection x; y RNi1 minxi ; yi
:yi
5. Chi-square kernel: KChisquare x; y RNi1 xxi iy
i

i
i
6. JS kernel: KJS x; y RNi1 x2i log2 xi y
y2i log2 xi y
xi
yi

The kernel, K(x, y), interpreted as a measure of similarity [19] between the two
examples, x and y. N is the dimension of the feature vector.

4 Improving the Performance of Multi-parameter Patient

Monitor
The rejection of additional algorithmic complexity, towards the improving the
performance of the MPM, we use geometric means of the vital parameters taken in
pairs of two, in addition to four vital parameters used in the baseline system.
Additional information about the balance/imbalance of the patients health status
are there in the expanded feature set, for example, the relationship between the

56

S. Premanand et al.

parameters exposes the risk factor for problems like cardiovascular abnormalities,
hyperoxia (high oxygen content in the blood), hypoxia (low oxygen content in the
blood), etc. The baseline system features [6] set is given by,
F x1; x2; x3; x4;

pppppp
x1x2; x1x3; x1x4; x2x3; x2x4; x3x4

In this work, we gathered more additional information by increasing the feature

set to G, to make the performance of MPM improved, and the feature set G is given
by,
h
pppppp
G x1; x2; x3; x4; x1x2; x1x3; x1x4; x2x3; x2x4; x3x4;

x1x2
x1x3
x1x4
x2x3
x2x4
x3x4
;
;
;
;
;
x1 x2 x1 x3 x1 x4 x2 x3 x2 x4 x3 x4

where x1, x2, x3 and x4 are the human vital parameters, heart rate, blood pressure,
respiration rate and oxygen saturation (SpO2) values, respectively, and the
remaining correlation features helps to capture the intrinsic relationship between the
vital parameters along with the help of explicit feature maps for addictive kernels
such as intersection, Chi square and JS kernels.

5 Experimental Results
The MIMIC-II database consists of four vital parameters collected from 413
patients. Among them, data from 12 patients were not suitable, and the remaining
401 patients data are used for the experiment. From the 401 patients, 1454,010
samples are separated as, 1100,510 (300 patients) samples as training data and
311,423 (101 patients) samples as testing data. Afterwards, the integral training and
testing data are shuffled randomly as 50,000 and 20,000 samples with the corresponding labels. The process is repeated to generate samples for seven independent
trials, and the results obtained from these trails are averaged to obtain the nal
result. We used LIBSVM for all our experiments. Table 1 illustrates the performance of SVM MPM system by using four vital parameters, Table 2 illustrates the
performance of SVM MPM system by using four vital parameters with additional
six correlation features by using geometric mean calculation thereby, increasing the
efciency and Table 3 explains the expansion of six features with the previous
process by the proportion of geometric mean to the arithmetic mean calculation,
hence we achieve better efciency than by 10-feature parameter. Results conclude
that the use of correlation features (16 features) along with the four vital parameters
helped to enhance the MPM system performance when compared to the baseline
system with ten features.

Improving the Performance of Multi-parameter Patient Monitor

57

Table 1 Result for 4 feature by using baseline kernels along with homogeneous kernels
Kernels

Overall accuracy

Sensitivity

Specicity

Linear
Polynomial
RBF
Intersection
Chi-squared
JS

77.14
92.62
97.38
99.61
93.56
92.76

1.55
79.54
95.54
98.36
79.65
77.59

100
96.58
97.93
99.99
97.77
97.35

Table 2 Result for 10 feature using baseline kernels along with homogeneous kernels
Kernels

Overall accuracy

Sensitivity

Specicity

Linear
Polynomial
RBF
Intersection
Chi-squared
JS

91.94
95.06
96.82
99.02
93.08
91.92

77.09
86.48
95.86
97.84
80.09
77.48

96.43
97.72
97.67
99.56
97.16
96.37

Table 3 Result for 16 feature using baseline kernels along with homogeneous kernels
Kernels

Overall accuracy

Sensitivity

Specicity

Linear
Polynomial
RBF
Intersection
Chi-squared
JS

92.61
95.19
97.91
99.87
93.84
92.95

79.44
86.57
96.25
98.39
80.84
78.61

96.60
97.76
98.41
99.97
97.82
97.29

6 Conclusion
We notice that the values of the four vital parameters heart rate, blood pressure,
respiration rate and oxygen saturation are always positive quantities, and thus the
correlation features derived from these four vital parameters. In this report, we
proposed a novel approach to improve the process of the MPMs taking advantage
of the intrinsic relationship between the vital parameters, using additional six
features by a proportion of geometric mean to the arithmetic mean calculation of the
vital parameters required in a pair of two, making the total number of features in the
proposed system as 16. In our experiments with the feature expansion for MPM,
we are able to make an improvement in the proposed system. We measured the
baseline system and the proposed system using sensitivity, specicity and overall
classication accuracy.

58

S. Premanand et al.

References
1. National Patient Safety Association and others (2007) Safer care for acutely ill patients:
learning from serious accidents. Technical report, NPSA
2. Tarassenko L, Clifton DA, Pinsky MR, Hravnak MT, Woods JR, Watkinson PJ (2011)
Centile-based early warning scores derived from statistical distributions of vital signs.
Resuscitation 82(8):10131018
3. Dornhost AC, Howard P (1952) Respiratory variations in blood pressure. Circulation
6:553558
4. Yasuma F, Hayano J (2004) Respiratory sinus arrhythmia: why does the heart beat
synchronize with respiratory rhythm? Chest 2:683690
5. Beata G, Anna S et al (2013) Relationship between heart rate variability, blood pressure and
arterial wall properties during air and oxygen breathing in healthy subjects. Auton Neurosci
178(12):6066
6. Vishnuprasad K, Santhosh Kumar C, Ramachandran KI, Vaijeyanthi V, Anand Kumar A
(2014) Towards building low cost multi-parameter patient monitors, ICC, 10 Apr 2014
7. Chan AB, Vasconcelos N, Moreno PJ (2004) A family of probabilistic kernels based on
information divergence, Statistical Visual Computing Laboratory, SVCL-TR 2004/01, June
2004
8. Vapnik VN (1999) The nature of statistical learning theory, 2nd edn. Springer, Berlin,
pp 237240, 263265, 291299
9. Chang C-C, Lin C-J (2013) LIBSVM: a library for support vector machines. Initial version:
2001 last updated: March 2013, pp 553558
10. Clifton L, Clifton DA, Watkinson PJ, Tarassenko (2011) Identication of patient deterioration
in vital-sign data using one-class support vector machines. In: IEEE federated conference
computer science and information systems (FedCSIS), pp 125131
11. Khalid S, Clifton DA, Clifton L, Tarassenko L (2012) A two-class approach to the detection of
physiological deterioration in patient vital signs, with clinical label renement. IEEE Trans Inf
Technol Biomed 16(6):1231
12. Lee J, Scott DJ et al (2011) Open-access MIMIC-II database for intensive care research. In:
33rd annual international conference of the IEEE EMBS
13. MIMIC II Database. www.physionet.org/physiobank/database/mimic2db/
14. Ruiz-Llata M, Guarnizo G, Yebenes-Calvino M (2010) FPGA implementation of a support
vector machine for classication and regression. In: The 2010 international joint conference on
IEEE neural networks (IJCNN), pp 15
15. Vedaldi A, Zisserman A (2011) Efcient additive kernels via explicit features maps. IEEE
Trans Pattern Anal Mach Intell, 114
16. Barla A, Odone F, Verri A (2003) Histogram intersection kernel for image classication. In:
Proceeding of ICIP
17. Sharma G, Jurie F (2013) A novel approach for efcient SVM classication with histogram
intersection kernel. Oral presentation at the British Machine Vision Conference (BMVC),
pp. 111
18. Callut J, Dupont P, Saerens M (2011) Sequence classication in the Jensen-Shannon
embedding. In: International conference on machine learning
19. Scholkopf B, Smola AJ (2001) Learning with kernels-support vector machines, regularization,
optimization and beyond. MIT Press, Cambridge

Rough Set Rule-Based Technique

for the Retrieval of Missing Data
in Malaria Diseases Diagnosis
B.S. Panda, S.S. Gantayat and Ashok Misra

Abstract Malaria disease is a major tropical public health problem in the world.
The diagnosis of this type of tropical diseases involves several levels of uncertainty
and imprecision. It causes severe infection to the brain and prevents brain from its
proper functioning. Hence prior detection of the malaria is much essential. Soft
Computing Techniques provide excellent methodologies to process the medical
data and help medical experts in nding out the nature of illness and to take
decision. True data set collection, feature squeezing, and classication are the basic
steps followed in designing an expert system. The designed expert system acts with
intelligence, prevents erroneous decisions, and produces sharp results in time. This
paper discusses on malaria investigation with missing data using rough set rulebased soft computing technique.
Keywords Accuracy
Rule set

Malaria
Missing data
Medical diagnosis
Rough set

1 Introduction
In India, malaria constitutes a great threat to the health of many communities. The
harmful effects of malaria parasites to the human body cannot be under estimated.
Malaria is a parasitic disease caused mainly by species of Anopheles mosquitoes. The
B.S. Panda (&)
MITS, Rayagada, Odisha, India
e-mail: [email protected]
S.S. Gantayat
GMRIT, Rajam, Andhra Pradesh, India
e-mail: [email protected]
A. Misra
CUTM, Paralakhemundi, Odisha, India
e-mail: [email protected]
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_6

59

60

B.S. Panda et al.

emergence of information technology (IT) has opened unprecedented opportunities

in health care delivery system as the demand for intelligent and knowledge-based
systems has increased as modern medical practices become more knowledgeintensive. The diagnosis of tropical diseases involves several levels of uncertainty
and imprecision [1].
The task of disease diagnosis and management is complex because of the
numerous variables involved in data sets, which may contain a lot of imprecision
and uncertainties. Patients cannot describe exactly how they feel, doctors and
nurses cannot tell exactly what they observe, and laboratories results are dotted with
some errors caused either by the carelessness of technicians or by malfunctioning of
the instrument. Medical researchers cannot precisely characterize how diseases alter
the normal functioning of the body [2, 3]. All these complexities in medical practice
make traditional quantitative approaches of analysis inappropriate. Computer tools
help to organize, store, and retrieve missing appropriate medical knowledge needed
by the practitioner in dealing with each difcult case and suggesting appropriate
diagnosis and decision-making technique.
In this article, we provided a rule-based soft computing technique for the
diagnosis of malaria to nd the appropriate (missing) symptoms to diagnosis
the malaria. This technique is based on clinical observations, medical diagnosis, and
the experts knowledge. The objective of the system is to provide a decision support
platform to medical researchers, physicians, and other healthcare practitioners in
malaria endemic regions. In addition, the system will assist medical personnel in the
tedious and complication task of diagnosing and further provide a scheme that will
assist medical personnel especially in rural areas, where there are shortages of
doctors, thereby, and offering primary health care to the people.

2 Literature Review
Missing data are questions without answers or variables without observation. Even
a small percentage of missing data can cause serious problems with the analysis
leading to draw wrong conclusions and imperfect knowledge. There are many
techniques developed in the literature to manipulate the knowledge with uncertainty
and manage data with incomplete items, but no results, and the results are not of the
same type and absolutely better than the others [46].
To handle such problems, researchers are trying to solve it in different approaches
and then proposed to handle the information system in their way. We know that the
attribute values are important for information processing in a data set or information
table. In the eld of databases, various efforts have been made for the improvement
and enhance of database or information table query process to retrieve the data. The
methodology followed by different approaches such as fuzzy sets [7, 8], rough sets
[9, 10], Boolean logic, possibility theory, statistically similarity [11], etc.

Rough Set Rule-Based Technique

61

3 Rough Sets
3.1 Denition and Notations
The concept of rough set is another approach to deal with imperfect knowledge. It
was introduced by Pawlak in 1982 [10]. From a philosophical point of view, rough
set theory is a new approach to deal vagueness and uncertainty, and from a practical
point of view, it is a new method of data analysis.
This method has the following important advantages:

It
It
It
It
It
It
It
It

provides efcient algorithms for nding hidden patterns in data;

nds reduced set of data (data reduction);
evaluates signicance of data;
generates minimal set of decision rules from data;
is easy to understand;
offers straightforward interpretation of results;
can be used in both qualitative and quantitative data analyses; and
identies relationship that would not be found by using statistical methods.

Let R UxU denote an equivalence relation on U, that is, R is a reflexive,

symmetric, and transitive relation. The equivalence class of an element x U with
respect to R is the set of elements y U such that xRy. If two elements x, y in
U belong to the same equivalence class then we say that x and y are indistinguishable with respect to relation R.
Given an arbitrary set A U, it may not be possible to describe A precisely in the
approximation space apr(R) = (U, R) instead one may only characterize A by a pair
of lower and upper approximations. This leads to the concept of rough sets (Fig. 1).
Fig. 1 Rough set approach

62

B.S. Panda et al.

We dene,
RA [ fY 2 U=R : Y  Ag;
and
RA [ fY 2 U=R : Y \A 6 ug:
RA and RA are respectively called the R-lower and R-upper approximation of
A with respect to R.
It can be noted that
RA fx 2 U : xR   Ag
and



RA x 2 U : xR \X 6 u :
The set BNR ARARA is called the R-boundary of A. The set RA consists of
all those elements of U which can with certainty be classied as elements of A,
employing the knowledge R. The set RA consists of all those elements of U which
can possibly be classied as elements of A, employing the knowledge R. Set
BNR(A) is the set of elements which cannot be classied as either belonging to A or
belonging to A having the knowledge R. We say that a set A is R-denable if and
only if RARA. Otherwise A is said to be R-rough.
The borderline region is the undecidable area of the universe. We say that X is
 equivalently BNR X 6 /. X is
rough with respect to R if and only if RX 6 RX,
 or BNR X /.
said to be R-denable if and only if RX RX,

3.2 Application of Rough Set

We briefly highlight few applications of Rough set theory [12]. We again feel
discussing on various applications will increase the length of paper.

Representation of uncertain or imprecise knowledge.

Empirical learning and knowledge acquisition from experience.
Knowledge analysis.
Analysis of conflicts.
Evaluation of the quality of the available information with respect to its
consistency and the presence or absence of repetitive data patterns.
Identication and evaluation of data dependencies.
Approximate pattern classication.
Reasoning with uncertainty.
Information-preserving data reduction.

Rough Set Rule-Based Technique

63

4 Malaria Data Set with Missing Attribute

Table 1 shows the attributes Temperature, Headache, Nausea, Vomiting, Joint
Pain, and Body Weakness and with the decision Malaria. However, many real-life
data sets are incomplete. The missing attribute value is denoted by ?.
From the above, the different classes can be generated using rough set concept as
follows.
Temperature = {C01, C06, C07, C10, C13, C15}, {C02, C05, C08, C09, C12},
{C03, C14}, {C04}, {C11}}
Headache = {{C01, C02, C04, C05, C07, C09, C10, C15}, {C03, C06, C13},
{C08}, {C11, C12, C14}}
Nausea = {{C01, C02, C03, C05, C15}, {C04, C08}, {C06, C07, C09, C10,
C12, C13, C14}, {C11}}
Vomiting = {{C01, C03, C04, C06}, {C02}, {C05, C07, C08, C09, C10, C13,
C15}, {C11, C12, C14}}
Joint_Pain = {{C01, C03, C15}, {C02, C05, C08, C09, 12}, {C04, C07, C11,
C14},{C06, C10,C13}}
Body_Weakness = {{C01, C05}, {C02, C08, C09, C13}, {C03, C04, C07, C12,
C14, C15}, {C06, C10},{C11}}
Malaria = {{C01, C05, C06, C10, C13, C15}, {C02, C03, C04, C07, C08, C09,
C11, C12, C14}}
In the above classication, the bold classes are the missing data, which will be
lled in the currently proposed technique.
Here, the rules are in the following two classes for Malaria types Yes or No.
If the symptoms lie in any of the temperature, head ache, nausea, vomiting, joint
pain, body weakness, then the Malaria type is either Yes or No.

5 Proposed System
The proposed system for malaria diagnosis uses rough set rule-based system. The
steps included in the proposed system are given below in Fig. 2.
Algorithm:
Step 1: Malaria Data Set Collect the Malaria data set which includes relevant
objects and attributes.
Setp 2: Divide Data Set Divide the malaria set into training and testing sets
using split factor of 60 and 40 %.
Step 3: Cut Set Form cut set for the training set which represents the
nonoverlapping subsets thus divides the attributes into set of intervals.
Cuts act as a boundary values representing those intervals.

Temperature

Mild
Moderate
Severe
Very severe
Moderate
Mild
Mild
Moderate
Moderate
Mild
?
Moderate
Mild
Severe
Mild

Case#

C01
C02
C03
C04
C05
C06
C07
C08
C09
C10
C11
C12
C13
C14
C15

Mild
Mild
Moderate
Mild
Mild
Moderate
Mild
?
Mild
Mild
Severe
Severe
moderate
Severe
Mild

Headache

Table 1 Malaria data set with missing attributes

Mild
Mild
Mild
?
Mild
Moderate
Moderate
?
Moderate
Moderate
Severe
Moderate
Moderate
Moderate
Mild

Nausea
Mild
?
Mild
Mild
Moderate
Mild
Moderate
Moderate
Moderate
Moderate
Severe
Severe
Moderate
Severe
Moderate

Vomiting
Mild
Moderate
Mild
Severe
Moderate
?
Severe
Moderate
Moderate
?
Severe
Moderate
?
Severe
Mild

Joint_Pain
?
Moderate
Severe
Severe
?
Mild
Severe
Moderate
Moderate
Mild
Very severe
Severe
Moderate
Severe
Severe

Body_Weakness

No
Yes
Yes
Yes
No
No
Yes
Yes
Yes
No
Yes
Yes
No
Yes
No

Malaria

64
B.S. Panda et al.

Rough Set Rule-Based Technique

65

Fig. 2 Flow chart of the rulebased system

Step 4:

Step 5:

Rule Set Rules are formed from the attribute by common attribute values
through logical AND. It also states the result class to which a particular
rule point to.
Substitute Common Attribute One of the simplest methods to handle
missing attribute values, such values are replaced by the most common value
of the attribute. In different words, a missing attribute value is replaced by the
most probable known attribute value, where such probabilities are
represented by relative frequencies of corresponding attribute values.

Note: The rules generated here can be extended with the logical operators OR and
NOT for a large dataset.

6 Cut Set for Overlapping Data

Cut set denes the mechanism of decomposing attribute value set in a dataset or
information table. The attributes are discretized to generate a set of intervals. These
are edge points to highlight the intervals. In case of representative attributes, cuts
stand for the nonoverlapping subsets of actual values. Table 2 represents cut set
consists of attributes for which the cuts are created. Size represents the quantity of
cuts created. In the following table, the missing attributes are represented by ?.

Temperature

Mild
Moderate
Very severe
Moderate
Mild
Moderate
Mild
?
Mild

Case#

C01
C02
C04
C05
C06
C08
C10
C11
C13

Table 2 After applying cut set

Mild
Mild
Mild
Mild
Moderate
?
Mild
Severe
Moderate

Headache

Nausea
Mild
Mild
?
Mild
Moderate
?
Moderate
Severe
Moderate

Vomiting
Mild
?
Mild
Moderate
Mild
Moderate
Moderate
Severe
Moderate

Joint_Pain
Mild
Moderate
Severe
Moderate
?
Moderate
?
Severe
?

Body_Weakness
?
Moderate
Severe
?
Mild
Moderate
Mild
Very severe
Moderate

Malaria
No
Yes
Yes
No
No
Yes
No
Yes
No

66
B.S. Panda et al.

Rough Set Rule-Based Technique

67

Cut set states the result attribute in terms of other attributes, which is a feature
reduction based on Boolean logics of algebra.

7 Rough Set Rule-Based Technique

The rough set rule-based technique usually have several antecedents that are combined
using logical operators, such as AND, OR, and NOT, though their denitions tend to
vary. In the rules, AND simply uses minimum weight of all the antecedents, while OR
uses the maximum value, the IF part is called the antecedent, and the THEN part is
called the consequent [13, 14]. The conditions are the set of attributes such as
temperature, head ache, nausea, vomiting, joint pain, body weakness, etc.
There are different approaches to generate rules, direct and indirect methods.
Direct methods generate rule from training data like sequential covering algorithms.
Indirect methods build the classication model from which rule are extracted, e.g.,
decision tree, neural network, genetic algorithms, etc.
The rule can be assessed by its coverage and accuracy. Coverage, if all the
antecedents in a rule R hold true for a given D. Accuracy, is the number of tables
that are correctly classied in the coverage. The coverage and accuracy are determined by the following formula.
CoverageR

ncovers
jD j

AccuracyR

ncorrect
ncovers

Information gain is a supervised feature selection method. It is a method that

measures the decrease in entropy when the feature is given rather than not given. It
can generalize to any number of classes. The entropy of a discrete random variable
X with a probability distribution p(x) is dened as
X
H p
p x log p x
x2X

These formulas are determined by aid of both the expert doctors in the eld of
tropical medicine and literature. Some of the rules (Rules 1, Rules 2, Rules 5, Rule
8, Rule 15) can be interpreted as follows:
Rule 1:
IF temperature = mild and headache = mild and nausea = mild and
vomiting = mild and joint pain = mild and body weakness = ? THEN
malaria = No.
Rule 2:
IF temperature = moderate and headache = mild and nausea = mild
and vomiting = ? and joint pain = moderate and body weakness = moderate THEN malaria = Yes.
Rule 3:
IF temperature = severe and headache = moderate and nausea = mild
and vomiting = mild and joint pain = mild and body weakness = severe
THEN malaria = Yes.

68

Rule 5:

Rule 8:

Rule 15:

B.S. Panda et al.

IF temperature = moderate and headache = mild and nausea = mild

and vomiting = moderate and joint pain = moderate and body
weakness = ? THEN malaria = No.
IF temperature = moderate and headache = ? and nausea = ? and
vomiting = moderate and joint pain = moderate and body weakness = moderate THEN malaria = Yes.
IF temperature = mild and headache = mild and nausea = mild and
vomiting = moderate and joint pain = mild and body weakness = severe
THEN malaria = Yes.

The notion of rough set [1517] is used to classify the patients with malaria or
not and also the other attributes which gives a group of similar information in the
data set. Here the other rules are not mentioned to avoid the complexity and
confusion with the given rules.

8 Results and Discussion

To implement the concept of rough set rule-based technique to simplifying the
diagnosis of malaria and impute the missing attributes, we have considered 15 data
sets with six attributes out of 20 real data sets with ten attributes collected from
different doctors. The technique used on knowledge of domain experts (ve
medical doctors), applied with rough set theory. Rough set classication is utilized
to remove uncertainty, ambiguity, and vagueness inherent in medical diagnosis.
The bold attributes in the above table show the unmatched attribute with the
actual attributes, to measure the accuracy using the formula we found 70 % of
accuracy, i.e., out of 10 cases 7 cases are matched. After imputation of common
attribute using rough set rule-based technique, the following tables 3 and 4 shows
the results.

Table 3 Imputation result for

missing data with actual data

Case#

Common attribute

Actual attribute

Diagnosis

C01
C02
C04
C05
C06
C08
C10
C11
C13

Mild
Moderate
Severe
Moderate
Mild
Moderate
Mild
Severe
Moderate

Mild
Mild
Mild
Moderate
Mild
Moderate
Mild
Severe
Mild

No
Yes
Yes
No
No
Yes
No
Yes
No

Temperature

Mild
Moderate
Severe
Very severe
Moderate
Mild
Mild
Moderate
Moderate
Mild
Severe
Moderate
Mild
Severe
Mild

Case#

C01
C02
C03
C04
C05
C06
C07
C08
C09
C10
C11
C12
C13
C14
C15

Mild
Mild
Moderate
Mild
Mild
Moderate
Mild
Moderate
Mild
Mild
Severe
Severe
moderate
Severe
Mild

Headache

Table 4 The missing data lled with observed data

Mild
Mild
Mild
Severe
Mild
Moderate
Moderate
Moderate
Moderate
Moderate
Severe
Moderate
Moderate
Moderate
Mild

Nausea
Mild
Moderate
Mild
Mild
Moderate
Mild
Moderate
Moderate
Moderate
Moderate
Severe
Severe
Moderate
Severe
Moderate

Vomiting
Mild
Moderate
Mild
Severe
Moderate
Mild
Severe
Moderate
Moderate
Mild
Severe
Moderate
Moderate
Severe
Mild

Joint_Pain
Mild
Moderate
Severe
Severe
Moderate
Mild
Severe
Moderate
Moderate
Mild
Very severe
Severe
Moderate
Severe
Severe

Body_Weakness

No
Yes
Yes
Yes
No
No
Yes
Yes
Yes
No
Yes
Yes
No
Yes
No

Diagnosis

Rough Set Rule-Based Technique

69

70

B.S. Panda et al.

The bold word in the above table shows the replacement of approximated
information with the missing data in Table 4.
After replacing proper values to the missing data, we get the different classes
from the above table as follows.
Temperature = {C01, C06, C07, C10, C13, C15}, {C02, C05, C08, C09, C12},
{C03, C11, C14}, {C04}}
Headache = {{C01, C02, C04, C05, C07, C09, C10, C15}, {C03, C06, C08,
C13}, {C11, C12, C14}}
Nausea = {{C01, C02, C03, C05, C15}, {C06, C07, C08, C09, C10, C12, C13,
C14}, {C04, C11}}
Vomiting = {{C01, C03, C04, C06}, {C02, C05, C07, C08, C09, C10, C13,
C15}, {C11, C12, C14}}
Joint_Pain = {{C01, C03, C06, C10, C15}, {C02, C05, C08, C09, C12, C13},
{C04, C07, C11, C14}}
Body_Weakness = {{C02, C05, C08, C09, C13}, {C03, C04, C07, C12, C14,
C15}, {C01, C06, C10}, {C11}}
Malaria = {{C01, C05, C06, C10, C13, C15}, {C02, C03, C04, C07, C08, C09,
C11, C12, C14}}
After the imputation, the different classes for individual attributes are reassigned
with the currently added new data.
From the above table, it is clear that the classes generated for the decision
making are reducing after imputing the missing values by using the rule generation
method (Table 5).
The analysis of the classication shows the following result.

Table 5 Comparison of before and after imputing missing values

Symptom types

No. of classes before

imputing missing values
(A)

No. of classes after

imputing missing values (B)

No. of changes in
classes |(A-B)|

Temperature
Headache
Nausea
Vomiting
Joint_Pain
Body_Weakness

5
4
4
4
4
5

4
3
3
3
3
4

1
1
1
1
1
1

Rough Set Rule-Based Technique

71

9 Conclusion
Malaria investigation using rough set rule-based soft computing technique shows
good imputation and coverage. It has handled 15 cases and 6 attributes. The above
method handles attributes even in the presence of absent values. In future, the same
method can be used to investigate other diseases by replacing the absent values with
common attribute value. This approach can be combined with other classiers to
enhance further to retrieve the missing data in a data set or information table. This
technique can be enhanced by using rough logic, which is a future direction.

References
1. Uzoka FME, Osuji J, Obot O (2010) Clinical decision support system (DSS) in the diagnosis
of malaria: a case comparison of two soft computing methodologies. Expert Syst Appl
38:15371553
2. Szolovits P, Patil RS, Schwartz WB (1988) Articial intelligent in medical diagnosis. J Intern
Med 108:8087
3. Szolovits P (1995) Uncertainty and decision in medical informatics. Methods Inf Med
34:111121
4. Little RJ, Rubin DB (2002) Statistical analysis with missing data, 2nd edn. Wiley, New York
5. Kantadzic M (2003) Data mining: concepts, models, methods and algorithms. Wiley, New
York
6. Gantayat SS, Misra A, Panda BS (2013) A study of incomplete dataa review. In: LNCS
Springer FICTA-2013, pp 401408. ISBN: 978-3-319-02930-6
7. Zadeh LA (1965) Fuzzy sets. Inf Control 8:338353
8. Zadeh LA (1973) Outline of a new approach to the analysis of complex system and decision
processes. IEEE Trans Syst Man Cybern 3:2844
9. Grzymala-Busse J (1988) LERS-a system for learning from examples based on rough sets.
J Intell Rob Syst 1:316
10. Pawlak Z (1982) Rough sets. J Inf Comp Sci II:341356
11. Devlin H, Devlin JK (2007) Decision support system in patient diagnosis and treatment.
Future Rheumatol 2:261263
12. Panda BS, Abhishek R, Gantayat SS (2012) Uncertainty classication of expert systemsa
rough set approach. In: ISCON proceedings with IJCA. ISBN: 973-93-80867-87-0
13. Grzymala-Busse J (1988) Knowledge acquisition under uncertaintya rough set approach.
J Intell Rob Syst 1:316
14. Panda BS, Gantayat SS, Misra A (2013) Rough set approach to development of a knowledgebased expert system. Int J Adv Res Sci Technol (IJARST) 2(2):7478. ISSN: 2319-1783
15. Pawlak Z (1991) Rough sets-theoretical aspects of reasoning about data. Kluwer Academic
Publishing, Boston
16. Pawlak Z, Skowron A (2007) Rough sets- some extensions. Inf Sci 177(1):2840
17. Pawlak Z (1996) Why rough sets, fuzzy systems. In: Proceedings of the fth ieee international
conference, vol 2

Automatic Image Segmentation for Video

Capsule Endoscopy
V.B. Surya Prasath and Radhakrishnan Delhibabu

Abstract Video capsule endoscopy (VCE) has proven to be a pain-free imaging

technique of gastrointestinal (GI) tract and provides continuous stream of color
imagery. Due to the amount of images captured automatic computer-aided
diagnostic (CAD) methods are required to reduce the burden of gastroenterologists.
In this work, we propose a fast and efcient method for obtaining segmentations of
VCE images automatically without manual supervision. We utilize an efcient
active contour without edges model which accounts for topological changes of the
mucosal surface when the capsule moves through the GT tract. Comparison with
related image segmentation methods indicate we obtain better results in terms of
agreement with expert ground-truth boundary markings.
Keywords Capsule endoscopy

Image segmentation
Active contours
CAD

1 Introduction
Oesophagel cancer rate in India is one of the highest in the world. Although low
and stable incidence and mortality rates from colorectal cancers were observed in
India [1], these rates were associated with a low 5-year relative survival rate [2].
This low survival rate suggests severe deciencies in early diagnosis and effective
treatment in India. Since as noted in [2] population-based screening of colorectal
V.B. Surya Prasath (&)
University of Missouri-Columbia, Columbia, MO 65211, USA
e-mail: [email protected]
R. Delhibabu
Cognitive Modeling Lab, IT University Innopolis, Kazan, Russia
e-mail: [email protected]
R. Delhibabu
Department of CSE, SSN Engineering College, Chennai, India
R. Delhibabu
Machine cognition lab, Kazan Federal University, Kazan, Russia
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_7

73

74

V.B. Surya Prasath and R. Delhibabu

Fig. 1 Wireless capsule endoscopy: a The patient swallows the capsule and capsule travels
through the tubular intestinal path. The imaging is done through a circular lens and at periodic
intervals. b Pillcam Capsule parts (used with permission from). 1 Optical dome. 2 Lens holder. 3
Lens. 4 Illuminating LEDs. 5 CMOS imager. 6 Battery. 7 ASIC transmitter. 8 Antenna

cancer is not cost-effective, given the low burden of colorectal cancer, early
diagnosis and adequate treatment using imaging techniques are important.
VCE introduced at the turn of this millennium [3] paved the way for painless
non-invasive way to image the gastrointestinal tract, see Fig. 1. A wireless capsule
consist of a tiny imaging device provides continuous video stream of the inner
mucosa-lumen tubular structure. Typically, these VCE exam for each patient
consist of 8 h of video which is around 55,000 frames. This puts an enormous
burden on the gastroenterologists since examining and reviewing the video from
capsule endoscopy requires concentration for a long durations. Thus, automatic
CAD methods are required which can help in analysis and diagnosis [46], see [7]
for a recent review. Computerized video analysis algorithms can reduce the time
required to review VCE exams and can augment the decision making processes.
Lumen detection in other imaging modalities have been studied by some
researchers. In [8, 9] colonoscopy images are considered and adaptive thresholding
type techniques were utilized. The wireless capsule optical system usually consist
of flashing LEDs, as opposed to xed lighting in colonoscopy, and an imaging
sensor which captures the images at predened time intervals. Thus, thresholding
type methods may not capture strong intensity variations between frames. In this
work, we study an automatic image segmentation method for VCE imagery to
segment lumen boundaries without any manual supervision. For this purpose we
start with the well-known image segmentation method based on active contours
without edges [10]. This method involves level sets which are known to handle
topological changes and thus is suitable for mucosa deformations which occur in
VCE data [11]. Though efcient, the traditional implementation was not fast for
real-time imaging and is restricted to takes 0.32 s/frame (of size 512 512). To
speed up the active contour segmentation further, in this paper we propose a fast
and efcient implementation of the active contour model which obtains good
segmentations. Compared with some of the classical and contemporary image

Automatic Image Segmentation for Video Capsule Endoscopy

75

segmentation techniques we obtain better segmentations when benchmarked with

marked ground-truth boundaries.
The rest of the paper is organized as follows. Section 2 gives the proposed fast
active contour based segmentation method. Section 3 provides experimental results
on various VCE images supporting our proposed approach. Finally, Sect. 4
concludes the paper.

2 Active Contours for Lumen-Mucosa Separation

2.1 Active Contour Without Edges
Chan and Vese [10] studied the active contour without edges (ACWE) method
which gives better segmentation results than other active contour based schemes.
ACWE is a modied version of the Mumford-Shah functional [12] and is an
alternating minimization scheme. Implementation of the nonlinear minimization
scheme is undertaken using the level sets methodology [13]. Suppose that the input
image u : X  R2 ! R where represents the image domain, usually a rectangle.
Then the piecewise constant ACWE is given by
Z

Z
d/jr/jdx k ju  c1 j2 H/dx
X
X
Z
2
k ju  c2 j 1  H/dx

min e/; c1 ; c2 l

/;c1 ;c2

where > 0 and 0 are given xed parameters, c1, c2 represent regions inside and
outside the level set function respectively. The function H(z) := 1 if z 0, H(z) :
= 0 if z < 0 is the Heaviside function, and dz : ddz Hz is the Dirac delta function
in the sense distributions. In [10], the Euler-Lagrange equation of the functional (1),
which is a necessary condition for a minimizer triplet (, c1, c2) to satisfy, is
implemented. That is, the following nonlinear PDE is solved for and c1, c2:




o/
r/
de / ldiv
 k1 u  c1 k2 u  c2
ot
jr/j

with
R
uHe /dx
c1 R X
;
X uHe /dx

R
u1  He /dx
c2 R X
X u1  He /dx

Here, de ; He represent the regularized versions of the dirac delta and Heaviside
functions respectively.

76

V.B. Surya Prasath and R. Delhibabu

2.2 Fast Implementation

The discretized version of the ACWE scheme (2) can be obtained using nite
differences and the method proposed in [10] follows a standard explicit central
differences for the spatial variable and forward difference for time variable [11]. In
this work, we modify the standard scheme by using a non-standard discretization


and our method is written as follows: Let /ki;j / nMt; xi ; yj be an approximation
of /t; x; y; k  0; /0 /0 and Mt is the time step size. Consider the rst order
standard differences
Mx /i;j /i;j  /i1;j ;

Mx /i;j /i1;j  /i;j

My /i;j /i;j  /i;j1 ;

My /i;j /i;j1  /i;j :

Then the algorithm for solving the ACWE PDE (2) is an alternating iterative
scheme:
1. Fix l; k; e knowing k, compute c1(k), c2(k) from (3).
2. Compute k+1 by the following discretization and linearization of (2) in :
2
1
0
k
k1
 6 l
C
B
/k1
Mx /i;j
i;j  /i;j
x B
C
r

dh /ki;j 6
M

4
A
@
2




2
2
h
Mt
2
Mx /ki;j =h2 /ki;j1  /ki;j1 =2h
1
0
C
My /i;j
l yB
C
Br
M
@ 
A
2



2
2
h
2
y k
k
k
2
/i1;j  /i1;j =2h M /i;j =h

 2

 2 i
 k1 ui;j  c1 /k
k2 ui;j  c2 /k

4
The modied nite difference scheme (4) involve only rst order discrete
derivatives and theoretical convergence of the above discretization will be reported
elsewhere.

Automatic Image Segmentation for Video Capsule Endoscopy

77

3 Experimental Results
3.1 Setup and Parameters
The core operations of the schemes are implemented in C with mex interface and
MATLAB R2012a is used for visualization purposes. Our fast scheme takes about
0.1 s (40 iterations) for a 512 512 image compared to 0.32 s in a previous work
[11]. This means on average the segmentation can be run for 10 frames per second
(fps), making it very attractive for endoscopic image analysis and diagnosis.
Enabling further code optimization via GPU may signicantly speed up the main
loop of the program. Moreover, instead of having a xed number of iterations for
segmentation, we devised a tolerance check to stop the scheme at convergence. For


example, /k1  /k \106 was set, and we observed convergence of the iterative
scheme (4) in less than 40 iterations for most of the images. Thus, our scheme takes
average runtime of 0.1 s/frame using non-optimized MATLAB code on 2.3 GHz
Intel Core i7, 8 GB RAM laptop to obtain a segmentation of lumen and mucosa.
The ACWE parameters 1 = 2 = 1, and e 106 are xed for all the experiments reported here. This gives equal weight to both inside and outside regions of
the zero level set. The parameter in Eq. (4) controls how regular the nal active
contour is, and hence, affects the nal segmentation in the experiments. After
conducting extensive experiments with respect to this parameter, we xed it at
= 0.2, and this seems to work well for most of the images in VCE data. The time
step parameter t = 0.5, step size for the nite difference grid h = 1 are xed
throughout all the experiments.

3.2 Comparison Results

The rst example in Fig. 2a shows the segmentation based on the proposed scheme.
The original image is a typical normal mucosa of the human colon captured in vivo
by the wireless capsule endoscopy device. To obtain good segmentation we always
start with the initialization shown in Fig. 2a, this type of ne mesh kind of

(a)

(b)

(c)

(d)

(e)

Fig. 2 Endoscopic image mucosa segmentation. a Input image and initialization of the modied
ACWE scheme, see Sect. 2.2. bd Intermediate segmentation results at time t = 20, 40 and 60,
curve laid on top of the original image for visualization. e Final segmentation at T = 80. The
segmentation algorithm usually converges within t = 40

78

V.B. Surya Prasath and R. Delhibabu

initialization provided better segmentations. Moreover, we observed experimentally

that, with a single or seed of coarse circles based initializations the proposed
scheme took more iterations to converge. Figure 2bd shows the intermediate active
contours of the segmentation scheme. The nal segmentation result in Fig. 2e
shows the mucosal folds and the existence of a clear boundary between the lumen
section of the image.
Related schemes from literature are also compared in terms segmentation
accuracy in mucosa-lumen differentiation. Figure 3 we show a comparison result of
other segmentation schemes from the literature against our scheme. The following
methods are compared with the proposed scheme (Table 1).
As can be seen by comparing, white curves superimposed, in Fig. 3g of the
proposed approachs segmentation result, we see that it captures mucosa regions
efciently even if the image is of low contrast. Texture based active contour (TAC)
fails completely and mean-shift obtains spurious lumen regions. On the other hand,

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

Fig. 3 Comparison with other segmentation schemes: a Otsu thresholding based scheme [8].
b Adaptive progressive thresholding scheme [9]. c Mean shift [17]. d ACWE [10]. e TextureACWE [18]. f Global minimization model [19] with Chan-Vese energy. g Our fast active contour.
h Expert (gastroenterologist) segmentation. Dice values () is given for each result and our fast
implementation obtained better agreement with ground-truth (GT) boundaries
Table 1 .
(a) Otsu thresholding based scheme (OT)
(b) Adaptive progressive thresholding (APT)
(c) Mean shift (MS)
(d) Chan-Vese scheme (ACWE)
(e) Texture based active contour (TAC)
(f) Fast global minimization (FGM)

[8]
[9]
[17]
[10]
[18]
[19]

Fig.
Fig.
Fig.
Fig.
Fig.
Fig.

3a
3b
3c
3d
3e
3f

Automatic Image Segmentation for Video Capsule Endoscopy

79

Table 2 Dice similarity coefcient (, Eq. (5)) values for automatic segmentations when
compared with manual ground truth for different schemes in Fig. 3
Sub-gure
Ref.

(a)
[8]

(b)
[9]

(c)
[17]

(d)
[10]

(e)
[18]

(f)
[19]

Our
Result

0.6592

0.4987

0.8202

0.8182

0.3017

0.7444

0.8793

results of other schemes such as APT, ACWE give inaccurate segmentations of the
mucosa folds. We use the Dice Similarity Coefcient,
jA; B

2jA \ Bj
;
jAj jBj

where A and B are automatic and manual segmentations respectively, and |X|
denotes the total number of pixels in a set X. Dice values are in [0,1.0] with values
closer to one indicate good overlap between manual and automatic segmentations.
Table 2 provides Dice values for different schemes corresponding to Fig. 3. As can
be seen our method obtains the highest Dice value and conrms visual comparison
seen with the GT image. Similar active contour methods utilized in other imaging
modalities [1416] can also be adapted for VCE imagery and denes our future
work in this direction.

4 Conclusion
In this paper, we consider automatic segmentation of video capsule endoscopy
imagery. By utilizing an active contour without edges with a fast implementation
we obtain meaningful segmentations of the gastrointestinal tract imaged by the
capsule endoscopy. Numerical implementation of the proposed scheme is carried
out with nite differences and provides efcient segmentation results. Compared
with other related image segmentation methods our fast implementation obtains
better results when compared to ground-truth marked by gastroenterologists. We
believe automatic computer aided diagnostic methods can provide relief to the
big-data handling associated with reading video capsule endoscopy imagery and
further methods for augmenting the diagnostic capabilities are required currently.

References
1. Mohandas KM (2011) Colorectal cancer in india: controversies, enigmas and primary
prevention. Indian J Gastroenterol 30(1):36
2. Pathy S, Lambert R, Sauvaget C, Sankaranarayanan R (2012) The incidence and survival rates
of colorectal cancer in India remain low compared with rising rates in east asia. Dis Colon
Rectum 55(8):900906

80

V.B. Surya Prasath and R. Delhibabu

3. Iddan G, Meron G, Glukhovsky A, Swain F (2000) Wireless capsule endoscopy. Nature 405
(6785):417
4. Figueiredo PN, Figueiredo IN, Prasath S, Tsai R (2011) Automatic polyp detection in pillcam
colon 2 capsule images and videos: Preliminary feasibility report. Diagn Ther Endosc
2011:7 pp Article ID 182435
5. Figueiredo IN, Moreno JC, Prasath VBS, Figueiredo PN (2012) A segmentation model and
application to endoscopic images. In: Campilho A, Kamel M (eds) International conference on
image analysis and recognition (ICIAR 2012). Springer LNCS, vol 7325. Aveiro, Portugal,
pp 164171 (June 2012)
6. Prasath VBS, Pelapur R, Palaniappan K (2014) Multi-scale directional vesselness stamping
based segmentation for polyps from wireless capsule endoscopy. Figshare (June 2014)
7. Karargyris A, Bourbakis N (2010) A survey on wireless capsule endoscopy and endoscopic
imaging. a survey on various methodologies presented. IEEE Eng Med Biol Mag 29(1):7283
8. Asari KV (2000) A fast and accurate segmentation technique for the extraction of
gastrointestinal lumen from endoscopic images. Med Eng Phys 22(2):8996
9. Asari KV, Srikanthan T (2002) Segmenting endoscopic images using adaptive progressive
thresholding: a hardware perspective. J Syst Architect 47(9):759761
10. Chan TF, Vese LA (2001) Active contours without edges. IEEE Trans Image Process 10
(2):266277
11. Prasath VBS, Figueiredo IN, Figueiredo PN, Palaniappan K (2012) Mucosal region detection
and 3D reconstruction in wireless capsule endoscopy videos using active contours. In: 34th
IEEE/EMBS international conference, San Diego, USA, pp 40144017 (September 2012)
12. Mumford D, Shah J (1989) Optimal approximations by piecewise smooth functions and
associated variational problems. Commun Pure Appl Math 42(5):577685
13. Osher S, Sethian JA (1988) Fronts propagating with curvature-dependent speed: algorithms
based on Hamilton-Jacobi formulations. J Comput Phys 79(1):1249
14. Prasath VBS (2009) Color image segmentation based on vectorial multiscale diffusion with
inter-scale linking. In: Chaudhury S, Mitra S, Murthy CA, Sastry PS, Sankar K.P (eds) Third
international conference on pattern recognition and machine intelligence (PReMI-09).
Springer LNCS, vol 5909. Delhi, India, pp 339344 (December 2009)
15. Moreno JC, Prasath VBS, Proenca H, Palaniappan K (2014) Brain MRI segmentation with fast
and globally convex multiphase active contours. Comput Vis Image Underst 125:237250
16. Prasath VBS, Pelapur R, Palaniappan K, Seetharaman G (2014) Feature fusion and label
propagation for textured object video segmentation. In: SPIE Defense + Security (DSS).
Baltimore, MD, USA (May 2014) In Geospatial Info Fusion and Video Analytics, IV
17. Comaniciu D, Meer P (2002) Mean shift: A robust approach toward feature space analysis.
IEEE Trans Pattern Anal Mach Intell 24(5):603619
18. Sandberg B, Chan TF, Vese L (2002) A level-set and Gabor-based active contour algorithm
for segmenting textured images. Technical Report, pp 0239, UCLA CAM (2002)
19. Bresson X, Esedoglu S, Vandergheynst P, Thiran J, Osher S (2007) Fast global minimization
of the active contour/snake model. J Math Imaging Vis 28(2):151167

Effect of Feature Selection on Kinase

Classication Models
Priyanka Purkayastha, Akhila Rallapalli, N.L. Bhanu Murthy,
Aruna Malapati, Perumal Yogeeswari and Dharmarajan Sriram

Abstract Classication of kinases will provide comparison of related human

kinases and insights into kinases functions and evolution. Several algorithms exist
for classication and most of them failed to classify when the dimension of feature
set large. Selecting the relevant features for classication is signicant for variety of
reasons like simplication of performance, computational efciency, and feature
interpretability. Generally, feature selection techniques are employed in such cases.
However, there has been a limited study on feature selection techniques for
classication of biological data. This work tries to determine the impact of feature
selection algorithms on classication of kinases. We have used forward greedy
feature selection algorithm along with random forest classication algorithm. The
performance was evaluated by selecting the feature subset which maximizes Area
Under the ROC Curve (AUC). The method identies the feature subset from the
datasets which contains the physiochemical properties of kinases like amino acid,
dipeptide, and pseudo amino acid composition. An improvised performance of
classication is noted for feature subset than with all the features. Thus, our method
indicates that groups of kinases are classiable with maximum AUC, if good
subsets of features are used.

P. Purkayastha (&)
A. Rallapalli
N.L. Bhanu Murthy
A. Malapati
P. Yogeeswari

D. Sriram
BITS Pilani Hyderabad Campus, Shameerpet, RR District, Hyderabad 500078, AP, India
e-mail: [email protected]
A. Rallapalli
e-mail: [email protected]
N.L. Bhanu Murthy
e-mail: [email protected]
A. Malapati
e-mail: [email protected]
P. Yogeeswari
e-mail: [email protected]
D. Sriram
e-mail: [email protected]
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_8

81

82

P. Purkayastha et al.

Keywords Forward greedy feature selection

ROC curve Kinases classication

Random forest
Area under the

1 Introduction
Kinases have become the most studied class for drug target and play important role
in metabolism, cell signaling, cellular transport, protein regulation, secretory
processes, and many other cellular pathways [1]. Hence, kinases annotations are
necessary for understanding the pathway related to signal transduction and for
understanding the disease pathway associated with impairment of kinases activity
[2]. So, classication of kinases will provide comparison of related human kinases
and insights into kinases functions and evolution. Currently the PROT R package in
R was to extract the features for kinases. With a wide range of features, choosing an
accurate features subset for classication is not an easy task. Feature selection
techniques are employed in such cases. Feature selection method identies subset of
features, based on which a classier will be trained. Feature selection is an
important step in training a classier with a subset of features, instead of training the
classier with the entire features of a dataset [3, 4].
Feature selection is benecial in various ways. First, a packed subset of features
can alleviate the curse of dimensionality and alienates the overtting problem which
is usually encountered during training a classier. Secondly, a model performs with
a highly accurate extent by removing the noisy features and preprocessing the
dataset. Thirdly, the accurate and essential subset of the dataset can be used with the
signicantly reduced computational cost. Finally, an illustrative subset of feature
can make the model output more understandable and reasonable. Computational
cost can be reduced by randomly eliminating a number of features. The major
challenge lies in search of feature subset that leads to enhanced performance of a
classier by removing the redundant and unnecessary features. Consequently, the
efcacy of a feature selection method is commonly assessed by the performance of
the nal model trained with the feature subset [4].
Researchers working on this problem have explored building classication
model using influential features, in which classication accuracy has thought to be
the best measure for assessing the performance of a classier. However, it has been
pointed out that accuracy is not always a suitable assessment metric and the Area
Under the ROC Curve (AUC) has been proven as a better performance metric in
evaluation with classication accuracy [5]. A classier with minimum cost is more
required than a classier with high accuracy [6]. In this work, we have used forward
greedy feature selection algorithm along with random forest classication algorithm
and evaluated the performance based on the feature subset which maximizes AUC.

Effect of Feature Selection on Kinase Classication Models

Table 1 The number of
sequences belongs to each of
the kinases family

83

Kinases families

Number of sequences

AGC
Atypical
CAMK
CK1
CMGC
OTHERS
RGC
STE
TK
TKL
Total

63
43
79
11
63
80
4
47
89
42
530

2 Materials and Methods

The kinases groups are majorly classied into 10 groups, as shown in Table 1. The
sequences are collected from Manning et al. classication and grouped accordingly
[7]. The dataset consist of 530 sequences, which includes physiochemical properties like amino acid composition, dipeptide composition, and pseudo amino acid
composition of human kinases obtained from Prot R package.

3 Results and Discussion

The performance of the random forest classier was evaluated for all the 10 group
of kinases namely, (i) Tyrosine Kinases (TK); (ii) Tyrosine Kinase-Like (TKL); (iii)
Protein kinases with the families with A, G, and C (AGC); (iv) Calmodulin/Calcium regulated kinases (CAMK); (v) Casein kinase 1 (CK1); (vi) Cyclin-dependent
kinases (CDKs), mitogen-activated protein kinases (MAP kinases), glycogen
synthase kinases (GSK), and CDK-like kinases (CMGC); (vii) Other kinases; (viii)
STE kinases; (ix) RGC kinases; and (x) Atypical kinases.
The classication of kinases to all the 10 groups was done using the amino acid,
dipeptide, and pseudo amino acid composition. The amino acid composition is of
xed pattern and is of 20 amino acids. The dipeptide composition consists of 400
amino acid dipeptide combinations. The pseudo amino acid composition also
contains detail information about three properties, hydrophobicity, hydrophilicity,
and side chain mass. The pseudo amino acid composition contains features of length
80. Amino acid and dipeptide composition was used by Bhasin and Raghava for
nuclear receptor classication [8]. Earlier pseudo amino acid composition was used
Krajewski and Tkacz for structural classication of proteins based on pseudo amino
acid composition with a measure of accuracy [9]. But recently, both theoretical and

84

P. Purkayastha et al.

empirical studies revealed that a classier with the highest accuracy extent might not
be idyllic in real-world problems. Instead, the AUC has been demonstrated as the
alternative approach and measure to evaluate the performance of any classier.
Therefore, we attempt to develop classication models using Random Forest
classier [10]. We have developed an algorithm by building the model using 2/3rd of
the training dataset and remaining 1/3rd of the test dataset. The test datasets were
partitioned randomly.
The performance of the classier was evaluated using AUC. Suppose we need to
select k feature subset from a feature set of F = ff 1 ; f 2 ; :::; f m g. Forward greedy
search builds model by considering one feature at a time and by calculating AUC
for each of them [11]. Then, the combination feature subsets are ranked based on
the descending order. The combination of features with maximum AUC value is
selected for classication of kinases.
The performance of the subset of features from the amino acid, dipeptide, and
pseudo amino acid composition are shown in Figs. 1, 2 and 3 respectively. As
shown in Figs. 1, 2 and 3, an improvised performance of the classier was obtained
after applying forward greedy feature selection algorithm than with all the features.
The feature selection maximizes AUC measures for all the 10 classes of kinases.
AUC measured for amino acid, dipeptide and pseudo amino acid composition
for all 10 classes shows a major difference in AUC measure with all 20 features
than compared with feature selected subsets. The AUC measure was seen for amino
acid composition in Fig. 1. The major difference in AUC measure was found for
RGC class than compared with other kinase classes. The feature subset with respect
to RGC class was found to be reduced to three features instead of all 20 features,
using forward greedy algorithm. A negligible difference was identied in case of
CK1 and TK classes, which could further be studied for marking a signicant
difference in all the kinase classes.

1.2
Forward greedy

All features

AUC measure

1
0.8
0.6
0.4
0.2
0
AGC

Atypical

CAMK

CK1

CMGC OTHERS RGC

STE

TK

TKL

Fig. 1 The performance of the classier for all 10 kinase classes using amino acid composition
using feature subset and all features

Effect of Feature Selection on Kinase Classication Models

85

1.2
Forward greedy

All features

AUC measure

1
0.8
0.6
0.4
0.2
0
AGC

Atypical

CAMK

CK1

CMGC OTHERS RGC

STE

TK

TKL

Fig. 2 The performance of the classier for all 10 kinase classes using dipeptide composition
using feature subset and all features

1.2
Forward greedy

All features

AUC measure

1
0.8
0.6
0.4
0.2
0
AGC

Atypical

CAMK

CK1

CMGC OTHERS RGC

STE

TK

TKL

Fig. 3 The performance of the classier for all 10 kinase classes using pseudo amino acid
composition using feature subset and all features

Similarly, the performance in AUC measure was compared using all 400
features and subset of features using dipeptide amino acid composition as shown
in Fig. 2. The major difference was found in case of atypical class of kinases
using all 400 features and subset of 4 features (using forward greedy) and the
negligible difference was found in case of TK class. Similarly, for pseudo amino
acid composition the difference measured was found to be more in case of
atypical than compared with all other classes and very less in case of RGC as
shown in Fig. 3. The number of features generated using forward greedy was
found to contain six features with highest AUC measure. This brings us to the
hypothesis that kinases can be classied with maximum AUC extent, if good
subsets of features are used.

86

P. Purkayastha et al.

4 Conclusion
In this paper, we have shown the pros of feature selection method for identifying the
feature subset for classication of kinases. The performance of the classication
model is shown using the feature subset and using all the features. The evaluation of
the performance was done by measuring AUC. The random forest classier is able
to classify kinase groups with a better AUC measure for feature subsets than
compared with all the features. But the difference in AUC measure was found to be
negligible for a few classes of kinase like RGC class using amino acid composition,
atypical class using dipeptide, and pseudo amino acid composition which indicates
that group of kinases are classiable with maximum AUC extent, if a good subset
of features are used. Further, feature selection method could useful to classify large
set of biological data and for dimensionality reduction.

References
1. Cohen P (2002) Protein kinasesthe major drug targets of the twenty-rst century? Nat Rev
Drug Discov 1(4):309315
2. Zhang J, Yang PL, Gray NS (2009) Targeting cancer with small molecule kinase inhibitors.
Nat Rev Cancer 1(9):2839
3. Ding C, Peng H (2003) Minimum redundancy feature selection from microarray gene
expression data. In: Proceedings of the IEEE computer society conference on bioinformatics,
pp 523528. Washington, DC
4. Tang K, Suganthan P, Yao X (2006) Gene selection algorithms for microarray data based on
least squares support vector machine. BMC Bioinform 7:95
5. Huang J, Ling CX (2005) Using AUC and accuracy in evaluating learning algorithms. IEEE
Trans Knowl Data Eng 17(3):299310
6. Rui W, Tang K (2009) Feature selection for maximizing the area under the ROC curve. In:
Data mining workshops, 2009. ICDMW09. IEEE international conference on. IEEE
7. Manning G et al (2002) The protein kinase complement of the human genome. Science 298
(5600):19121934
8. Bhasin M, Raghava GP (2004) Classication of nuclear receptors based on amino acid
composition and dipeptide composition. J Biol Chem 279(22):2326223266
9. Krajewski Z, Tkacz E (2013) Protein structural classication based on pseudo amino acid
composition using SVM classier. Biocybern Biomed Eng 33(2):7787
10. Breiman Leo (2001) Random forests. Mach Learn 45(1):532
11. Bradley Andrew P (1997) The use of the area under the ROC curve in the evaluation of
machine learning algorithms. Pattern Recogn 30(7):11451159

Rheumatoid Arthritis Candidate Genes

Identication by Investigating Core
and Periphery Interaction Structures
Sachidanand Singh, V.P. Snijesh and J. Jannet Vennila

Abstract Rheumatoid arthritis (RA) is a long-term systemic inflammatory disease

that primarily attacks synovial joints and ultimately leads to their destruction. The
disease is characterized by series of processes such as inflammation in the joints,
synovial hyperplasia, and cartilage destruction leading to bone erosion. Since RA
being a chronic inflammatory complex disease, there is a constant need to develop
novel and dynamic treatment to cure the disease. In the present research, network
biology and gene expression proling technology are integrated to predict novel key
regulatory molecules, biological pathways, and functional network associated with
RA. The microarray datasets of synovial broblast (SF) (GSE7669) and macrophages (GSE10500 and GSE8286), which are the primary cells in the synovium and
reported as the key players in the pathophysiology of RA, were considered for
identication of signature molecules related to RA. The statistical analysis was
performed using false discovery rate (FDR), t-test, one-way anova, and Pearson
correlation with favorable p-value. The K-core analysis depicted the change in
network topology which consisted of up- and downregulated genes network, resulted
in six novel meaningful networks with seed genes OAS2, VCAN, CPB1, ZNF516,
ACP2, and OLFML2B. Hence, we propose that, differential gene expression network
studies will be a standard step to elucidate novel expressed gene(s) globally.

Keywords Rheumatoid arthritis Gene expression proling Synovial broblast

Macrophages K-core analysis

S. Singh (&)
V.P. Snijesh

Department of Bioinformatics, School of Biotechnology and Health Sciences,
Karunya University, Coimbatore, India
e-mail: [email protected]
J. Jannet Vennila
Department of Biotechnology, School of Biotechnology and Health Sciences,
Karunya University, Coimbatore, India
The Author(s) 2015
N.B. Muppalaneni and V.K. Gunjan (eds.), Computational Intelligence
in Medical Informatics, Forensic and Medical Bioinformatics,
DOI 10.1007/978-981-287-260-9_9

87

88

S. Singh et al.

1 Introduction
Network biology and gene expression proling with microarrays have turned into a
typical approach for nding genes and biological pathways which are correlated
with diverse complex diseases [3, 5, 12]. Biological network analysis combines
coexpressed gene networks to predict novel associations and key regulatory
molecules. It integrates independent data as well as biologically relevant information which emerge as a reliable method to nd out meaningful novel biological
network(s) [4, 24, 39].
Rheumatoid arthritis (RA) is a systemic autoimmune disease characterized by
inflammation, synovial hyperplasia, cartilage destruction, and bone erosion [31].
The involvement of immune cells and inflammatory molecules are basic hallmark
of RA [13]. The cells of synovium are basically divided into synovial broblast
(SF) and macrophages. The primary function of SF and macrophages are secretion
of hyaluronic acid and phagocytosis, respectively [1, 23]. They are involved in
production of various inflammatory cytokines and chemokines, which in turn attract
more inflammatory molecules to the synovium [16, 35]. As together they also
initiate in production of Matrix Metalloproteinase (MMP), Vascular Endothelial
Growth Factor (VEGF), formation of ectopic germinal layers, and over expression
of Major Histocompatibility Complex Class II (MHC II) [6, 15, 18]. Recent studies
have reported that both SF and macrophages play crucial role in progressive joint
inflammation and destruction as they are found larger in pannus and inflamed
synovial membrane than in normal joints [2]. Current research work focuses on
identication of key regulatory molecules from synovial cell layers.
Microarray technology has been implemented to identify gene expression level
which can be further used to detect transcriptionally altered key signature molecules
involved in the pathophysiology of RA. At current times, various studies have made
to predict differentially expressed genes (DEGs) in RA using multiple gene
expressions [17]. In the current study, molecular expression proles of human
macrophages and SF from Gene Expression Omnibus (GEO) were analyzed to
identify promising genes for RA. Our investigation provided a valuable methodology to analyze these novel genes which are involved in the pathophysiology of RA.

2 Materials and Methods

2.1 Data Collection
The research work focused on the key regulatory molecules involved in the
pathophysiology of RA. Synovial lining consists of two types of cells SF and
macrophages which are reported as the key players in the pathogenesis to RA [25,
38]. RA related microarray datasets of SF and macrophages were retrieved from GEO
(http://www.ncbi.nlm.nih.gov/geo/). Overall 3 datasets, 2 from synovial macrophage

Rheumatoid Arthritis Candidate Genes Identication

89

(GSE10500 and GSE8286) and 1 from SF (GSE7669) under Affymetrix platform

were selected. The series GSE10500 consisted of 8 samples with 5 from RA patients
and 3 from normal donors [44]. The dataset GSE8286 was associated with differentiation of monocyte to macrophage which consisted of total of 9 samples. Three
each samples from monocyte at 0, 16, and 168 h were collected [22]. GSE7669
comprised of 6 samples each from RA patients and osteoarthritis patients [10].

2.2 Data Preprocessing

Each dataset was incorporated into GeneSpring GX 12.6 and raw signal values of
each probe sets were normalized by baseline to median of control samples using
Robust Multiarray Average (RMA) algorithm. It normalizes the raw signals by
creating expression matrix from Affymetrix data by involving background correction of raw signals, log2 transformation, and quantile normalization (Fig. 1) [14].
All samples in the datasets were experimentally grouped based on the content of the
sample. 8 samples of GSE10500 were grouped as 3 control (normal donors) and 5
diseased (RA patients). Similar grouping was made in the case of GSE7669. But for
the dataset GSE8286 grouping was made based on time and condition. Differentiation of monocyte to macrophage under 016 h and 0168 h conditions were
grouped (Fig. 2).

Fig. 1 Box plot of normalized values for series GSE7669, GSE8286, and GSE10500

90

S. Singh et al.

Fig. 2 Prole plot of group for series GSE7669, GSE8286, and GSE10500

2.3 Differentially Expressed Genes

The collected datasets differ in terms of number of samples, experimental conditions, and type. To identify the DEGs, statistical analysis was carried out based on
one-way annova, t-test, and BenjaminiHochberg False Discovery Rate (FDR).
Paired t-test and unpaired t-test were performed for datasets based on the number of
sample on grouping. Datasets having equal number of controlled and treated
samples were performed with paired t-test. Datasets GSE10500, GSE8286, and
GSE7669 followed t-test unpaired, one-way anova, and t-test paired, respectively.
Direct BenjaminiHochberg FDR with p-value cutoff 0.05 and Fold Change (FC)
value 1.5 and 1.5 were carried out for entire datasets to identify the signicant
expressed genes in the dataset.

2.4 Protein Network Analysis

Genegene correlation map, based on Pearson correlation with centroid linkage
clustering method, was constructed for both upregulated and downregulated genes.
The relationship between entire datasets was developed using Pearsons correlation
coefcient (R-value) [27]. Symmetrical n n matrix was generated at R-value cutoff
0.9 and above. From the matrix, genes with high correlation value were imported to
Cytoscape [34] for the network construction of coexpressed genes.
In any biological network, node represents gene or protein and the interactions
between these nodes are termed as degree which represents various pathways

Rheumatoid Arthritis Candidate Genes Identication

91

through which they are associated [36]. Basically, biological networks are scale free
networks with statistically and functionally signicant interacting molecules [33].
Cytoscape plugins-Network analyzer and M-Code were implemented for identifying
the topology of the network keeping into consideration of K-core for identifying
promising hubs [30]. It was further validated by biological analysis by Biointerpreter
(http://www.biointerpreter.com/biointerpreterv3/) [42].

3 Result and Discussion

3.1 Expression Characteristics for RA Datasets
Differentially expressed genes were studied using statistical measures for each series
of macrophages and broblast of synovial membrane. RMA algorithm normalized
the raw signals of all the series which were under Affymetrix platform [9].
The multiple raw signals were compared using multiple hypothesis testing, p-value
(0.05), and FC = 1.5 (1.5 and 1.5). The series GSE7669 of broblast resulted in
79 DEGs with 35 up- and 54 downregulated genes. There were 691 DEGs in the
dataset of GSE10500 with 424 and 267 up- and downregulated genes, respectively.
The series GSE8286 in multiple groups were reported a total of 145 DEGs in which
group 016 h was in 3 and 1 and group 0168 h in 51 and 90 up- and downregulated
genes, respectively (Table 1).
Higher correlation coefcient (r 0.9) with Pearson algorithm implied in
genegene correlation of DEGs resulted in 403 genes interactome (edges = 81,140)
for downregulated and 510 genes interactome (edges = 183,064) for upregulated
genes. The whole interactome gave a good landscape of highly correlated
coexpression matrix.
Table 1 List of up- and downregulated genes in different series of SF and Macrophages
Fibroblast_GSE7669
Upregulated
35
Macrophage_GSE10500
Upregulated
424
Macrophage_GSE8286
016 h
Upregulated
Downregulated
3
1

Downregulated
54
Downregulated
267
0168 h
Upregulated
51

Downregulated
90

92

S. Singh et al.

3.2 K-Core Hubs Identication for DEGs

Both the up- and downregulated networks ended up with three clusters (Rank 1,
Rank 2, and Rank 3) with K-core of 15 (Table 2). The seed nodes obtained through
the analysis were ZNF516, ACP2, and OLFML2B for downregulated (Fig. 3)
whereas OAS2, VCAN, and CPB1 for upregulated networks (Fig. 4). Seed nodes
are the highest scoring nodes in the cluster which connect all other nodes and
represent the central node of connectivity [21]. The M-CODE score value ranged
from 67 to 846 for Rank 13 (Table 2).

Table 2 K-core analysis with highly clustered seed genes for down- and upregulated networks
Clusters

Node

Score

K-core for downregulated genes (1520)

Rank 1
265
528
Rank 2
86
177
Rank 3
49
103
K-core for upregulated genes (1520)
Rank 1
424
846
Rank 2
86
96
Rank 3
49
67

Edges
69,984
7,309
2,351
179,358
2,440
991

Fig. 3 Seed genes on K-core analysis for downregulated network

Seed
ZNF516
ACP2
OLFML2B
OAS2
VCAN
CPB1

Rheumatoid Arthritis Candidate Genes Identication

93

Fig. 4 Seed genes on K-core analysis for upregulated network

Common biological properties like angiogenensis, antigen processing and presentation, immune response, vasculogenesis, chemotaxis, and inflammation
conrmed that these nodes may play role in pathophysiology of RA from the
cytoplasmic and nuclear level. 40 % of genes identied under different categories of
synovial membrane (Rank 13 for up and Rank 13 for down) pointed to have
correlation with the disease hyperlipidemia which is a genetic disorder of increased
blood fats causing cardiovascular risk which is further connected with RA as it
doubles the risk in RA patient [11, 26]. All the ranks from the both networks
showed their influence in cytokinecytokine interaction, TGF beta signaling, ECM
receptor interaction, Osteoclast differentiation, and Wnt receptor signaling pathways. Under upregulated networks, the seed gene Chondroitin Sulfate Proteoglycan
(VCAN) showed its involvement in glycosaminoglycan and hyaluronic acid binding
[29, 43]. Being a major component of cartilage, it is the main point of attraction for
CD44 for initiating the inflammation in synovial cells [8]. Oligoadenylate Synthetase
2 (OAS2) and Carboxypeptidase B1 (CPB1) showed crossregulation and downregulation with autoimmune diseases, respectively [37]. OAS2 was involved in IFN
beta signaling pathway whereas CPB1which is present in plasma helps in brin clot
by acting as a procoagulant [20, 32, 40]. CPB1 also initiates osteopontin, C5a, and
bradykinin-like proinflammatory molecules [7]. Under downregulated networks,
Zinc Finger Protein 516 (ZNF516), Acid phosphatase 2, lysosomal (ACP2), and
Olfactomedin-like 2B (OLFML2B) showed their specic response in stress activity

94

S. Singh et al.

specically in immune response [19, 28]. These genes have already indicated
upregulation in cardiovascular and hormone regulation. ACP2 and OLFML2B also
showed their response to hyperlipidemia disease which was connected to RA [41].
All the proteins molecules reported through network analysis showed their effect in
pathophysiology of RA. Still there is wide scope to further investigate these networks
which can explore more genes/proteins related to RA.

4 Conclusion
The proposed research work is an expression network study frame work for RA
synovial cells which may play signicant role in its pathophysiology. The outcome
resulted in six candidate seed genes which gave a better understanding of the
progression of RA by different pathways involvement. This part of information may
lead to better understand and manage RA in future by considering these potential
targets in therapeutics.
Acknowledgments This research work was supported by Science Engineering and Research
Board-Department of Science and Technology, New Delhi and Karunya University, Coimbatore,
Tamil Nadu.

References
1. Anitua E, Snchez M, Nurden AT, Zalduendo MM, De La Fuente M, Azofra J, Anda I (2007)
Platelet-released growth factors enhance the secretion of hyaluronic acid and induce
hepatocyte growth factor production by synovial broblasts from arthritic patients.
Rheumatology 46(12):17691772
2. Athanasou NA (1995) Synovial macrophages. Ann Rheum Dis 54(5):392
3. Bauer JW, Bilgic H, Baechler EC (2009) Gene-expression proling in rheumatic disease: tools
and therapeutic potential. Nat Rev Rheumatol 5(5):257265
4. Begley TJ, Rosenbach AS, Ideker T, Samson LD (2002) Damage recovery pathways in
Saccharomyces cerevisiae revealed by genomic phenotyping and interactome mapping1 1 NIH
grants RO1-CA-55042 and P30-ES02109; NIH training grant ES07155 and National Research
Service Award F32-ES11733 (to TJB). Mol Cancer Res 1(2):103112
5. Chand Y, Alam MA (2012) Network biology approach for identifying key regulatory genes by
expression based study of breast cancer. Bioinformation 8(23):1132
6. Cho CH, Koh YJ, Han J, Sung HK, Lee HJ, Morisada T, Koh GY (2007). Angiogenic role of
LYVE-1-positive macrophages in adipose tissue. Circ Res 100(4):e47e57
7. Du XY, Zabel BA, Myles T, Allen SJ, Handel TM, Lee PP, Leung LL (2009) Regulation of
chemerin bioactivity by plasma carboxypeptidase N, carboxypeptidase B (activated thrombinactivable brinolysis inhibitor), and platelets. J Biol Chem 284(2):751758
8. Fujimoto T, Kawashima H, Tanaka T, Hirose M, Toyama-Sorimachi N, Matsuzawa Y,
Miyasaka M (2001) CD44 binds a chondroitin sulfate proteoglycan, aggrecan. Int Immunol 13
(3):359366

Rheumatoid Arthritis Candidate Genes Identication

95

9. Gautier L, Cope L, Bolstad BM, Irizarry RA (2004) Affyanalysis of Affymetrix GeneChip

data at the probe level. Bioinformatics 20(3):307315
10. Geurts J, Vermeij EA, Pohlers D, Arntz OJ, Kinne RW, van den Berg WB, van de Loo FA
(2011) A novel Saa3-promoter reporter distinguishes inflammatory subtypes in experimental
arthritis and human synovial broblasts. Ann Rheum Dis 70(7):13111319
11. Han C, Robinson DW, Hackett MV, Paramore LC, Fraeman KH, Bala MV (2006)
Cardiovascular disease and risk factors in patients with rheumatoid arthritis, psoriatic arthritis,
and ankylosing spondylitis. J Rheumatol 33(11):21672172
12. Hupl T, Stuhlmller B, Grtzkau A, Radbruch A, Burmester GR (2010) Does gene
expression analysis inform us in rheumatoid arthritis? Ann Rheum Dis 69(suppl 1):i37i42
13. Huber LC, Distler O, Tarner I, Gay RE, Gay S, Pap T (2006) Synovial broblasts: key players
in rheumatoid arthritis. Rheumatology 45(6):669675
14. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, Speed TP
(2003) Exploration, normalization, and summaries of high density oligonucleotide array probe
level data. Biostatistics 4(2):249264
15. Ito TK, Ishii G, Chiba H, Ochiai A (2007) The VEGF angiogenic switch of broblasts is
regulated by MMP-7 from cancer cells. Oncogene 26(51):71947203
16. Jones JA, Chang DT, Meyerson H, Colton E, Kwon IK, Matsuda T, Anderson JM (2007)
Proteomic analysis and quantication of cytokines and chemokines from biomaterial surfaceadherent macrophages and foreign body giant cells. J Biomed Mater Res Part A 83
(3):585596
17. Kim TH, Choi SJ, Lee YH, Song GG, Ji JD (2014) Gene expression prole predicting the
response to anti-TNF treatment in patients with rheumatoid arthritis; analysis of GEO datasets.
Jt Bone Spine 81(4):325330
18. Kobayashi H, Puri P, OBriain DS, Surana R, Miyano T (1997) Hepatic overexpression of
MHC class II antigens and macrophage-associated antigens (CD68) in patients with biliary
atresia of poor prognosis. J Pediatr Surg 32(4):590593
19. Kovar DR, Wu JQ, Pollard TD (2005) Prolin-mediated competition between capping protein
and formin Cdc12p during cytokinesis in ssion yeast. Mol Biol Cell 16(5):23132324
20. Lepus CM, Song JJ, Wang Q, Wagner CA, Lindstrom TM, Chu CR, Robinson WH (2014)
Brief report: carboxypeptidase B serves as a protective mediator in osteoarthritis. Arthritis
Rheumatol 66(1):101106
21. Li X, Wu M, Kwoh CK, Ng SK (2010) Computational approaches for detecting protein
complexes from protein interaction networks: a survey. BMC Genom 11(suppl 1):S3
22. Liu H, Shi B, Huang CC, Eksarko P, Pope RM (2008) Transcriptional diversity during
monocyte to macrophage differentiation. Immunol Lett 117(1):7080
23. Mantovani A, Sozzani S, Locati M, Allavena P, Sica A (2002) Macrophage polarization:
tumor-associated macrophages as a paradigm for polarized M2 mononuclear phagocytes.
Trends Immunol 23(11):549555
24. Marcotte EM, Pellegrini M, Thompson MJ, Yeates TO, Eisenberg D (1999) A combined
algorithm for genome-wide prediction of protein function. Nature 402(6757):8386
25. Mor A, Abramson SB, Pillinger MH (2005) The broblast-like synovial cell in rheumatoid
arthritis: a key player in inflammation and joint destruction. Clin Immunol 115(2):118128
26. Nelson RH (2013) Hyperlipidemia as a risk factor for cardiovascular disease. Prim Care Clin
Off Pract 40(1):195211
27. Ng A, Bursteinas B, Gao Q, Mollison E, Zvelebil M (2006) pSTIING: a systems approach
towards integrating signalling pathways, interaction and transcriptional regulatory networks in
inflammation and cancer. Nucleic Acids Res 34(suppl 1):D527D534
28. Pantoja-Uceda D, Arolas JL, Garca P, Lpez-Hernndez E, Padr D, Aviles FX, Blanco FJ
(2008) The NMR structure and dynamics of the two-domain tick carboxypeptidase inhibitor
reveal flexibility in its free form and stiffness upon binding to human carboxypeptidase b.
Biochemistry 47(27):70667078
29. Perides G, Biviano F, Bignami A (1991) Interaction of a brain extracellular matrix protein with
hyaluronic acid. Biochim Biophys Acta (BBA) (General Subjects) 1075(3):248258

96

S. Singh et al.

30. Raman MP, Singh S, Devi PR, Velmurugan D (2012) Uncovering potential drug targets for
tuberculosis using protein networks. Bioinformation 8(9):403
31. Rossol M, Schubert K, Meusch U, Schulz A, Biedermann B, Grosche J, Wagner U (2013)
Tumor necrosis factor receptor type I expression of CD4+ T cells in rheumatoid arthritis
enables them to follow tumor necrosis factor gradients into the rheumatoid synovium. Arthritis
Rheum 65(6):14681476
32. Sadler AJ, Williams BR (2008) Interferon-inducible antiviral effectors. Nat Rev Immunol 8
(7):559568
33. Sengupta U, Ukil S, Dimitrova N, Agrawal S (2009) Expression-based network biology
identies alteration in key regulatory pathways of type 2 diabetes and associated risk/
complications. PLoS ONE 4(12):e8100
34. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Ideker T (2003) Cytoscape:
a software environment for integrated models of biomolecular interaction networks. Genome
Res 13(11):24982504
35. Smith RS, Smith TJ, Blieden TM, Phipps RP (1997) Fibroblasts as sentinel cells. Synthesis of
chemokines and regulation of inflammation. Am J Pathol 151(2):317
36. Snijesh VP, Singh S (2014) Molecular modeling and network based approach in explaining
the medicinal properties of nyctanthes arbortristis, lippia nodiflora for rheumatoid arthritis.
J Bioinform Intell Control 3(1):3138
37. Song JJ, Hwang I, Cho KH, Garcia MA, Kim AJ, Wang TH, Robinson WH (2011) Plasma
carboxypeptidase B downregulates inflammatory responses in autoimmune arthritis. J Clin
Invest 121(9):35173527
38. Szekanecz Z, Koch AE (2007) Macrophages and their products in rheumatoid arthritis. Curr
Opin Rheumatol 19(3):289295
39. Tornow S, Mewes HW (2003) Functional modules by relating protein interaction networks
and gene expression. Nucleic Acids Res 31(21):62836289
40. van Baarsen LG, Wijbrandts CA, Rustenburg F, Cantaert T, van der Pouw TC (2010)
Regulation of IFN response gene activity during infliximab treatment in rheumatoid arthritis is
associated with clinical response to treatment. Arthritis Res Ther 12(1):11
41. van Oorschot RA, Birmingham V, Porter PA, Kammerer CM, VandeBerg JL (1993) Linkage
between complement components 6 and 7 and glutamic pyruvate transaminase in the
marsupialMonodelphis domestica. Biochem Genet 31(56):215222
42. Varatharajan S, Karathedath S, Velayudhan SR, Srivastava A, Mathews V, Balasubramanian P
(2013) Harnessing gene expression proling in search of new candidate genes for Ara-C
resistance in acute myeloid leukemia. Blood 122(21):1299
43. Westling J, Gottschall P, Thompson V, Cockburn A, Perides G, Zimmermann D, Sandy J
(2004) ADAMTS4 (aggrecanase-1) cleaves human brain versican V2 at Glu405-Gln406 to
generate glial hyaluronate binding protein. Biochem J 377:787795
44. Yarilina A, Park-Min KH, Antoniv T, Hu X, Ivashkiv LB (2008) TNF activates an
IRF1-dependent autocrine loop leading to sustained expression of chemokines and
STAT1-dependent type I interferonresponse genes. Nat Immunol 9(4):378387