Drug Discovery Today  Volume 16, Numbers 17/18  September 2011 REVIEWS

Reviews  POST SCREEN

Structure-based drug design to augment
hit discovery
Subha Kalyaanamoorthy and Yi-Ping Phoebe Chen
Faculty of Science, Technology and Engineering, La Trobe University, Australia

Several technology-based strategies have been developed to address the significance of the two phases of
drug discovery: hit identification and lead identification. Structure-based drug design (SBDD), a method
that depends on possessing the knowledge of 3D structures of biological targets, is growing swiftly with
the development of new technologies for searching potential ways to combat disease. The past decade
has evidenced a threefold increase in the amount of software and tools in the online repositories. Herein,
we review the in silico strategies and modules applied at the level of hit identification and confer the
different challenges with possible solutions in enhancing the success rate of the ‘hit-to-lead’ phase that
could eventually help the progress of SBDD in the drug discovery arena.

Introduction influence in the search for new drugs [7]. Thus, SBDD has been
The approaches and methodologies used in drug design have widely realized and accepted as an essential part of drug research.
changed over time, exploiting and driving new technological The most popularly followed flowchart of SBDD is shown in Fig. 1.
advances to solve the varied, new, difficult setbacks that occur During the past few decades, there has been a steep rise in the
along the path to drug discovery. In addition to the experimental volumes of software packages that can assist in carrying out the
techniques, a variety of computational approaches have been different phases of SBDD effectively. Although these computa-
applied at the various stages of the drug-design process: in the tional resources have much to offer SBDD, it has eventually
early stages these approaches focused on reducing the number of become a challenge to choose successful combinations of strate-
possible ligands, whereas in the later stages, during lead-optimiza- gies and tools for efficient lead discovery [8]. Thus, in the following
tion, the emphasis is on decreasing experimental costs and redu- sections, we review the different protocols and relevant computa-
cing the period of discovery [1]. Structure-based drug design tional programs that are used in in silico hit discovery.
(SBDD) is one such computational approach that can go hand-
in-hand with major phases of drug discovery such as ‘hit identi- Target identification
fication’ and ‘hit-to-lead’; the initial phase involves the identifica- Complete genome sequences have provided a glut of potential
tion of a list of chemical compounds, known as ‘hits’, that ideally biological targets [9]. Practices such as systems biology, clustering,
exhibit some degree of potency and specificity against the target. probabilistic networks and drug affinity response have matured to
Whereas, the latter engages evaluation of the screened hits to help with the identification of biological targets [10]. A typical
identify the promising lead molecules before proceeding toward a SBDD begins with the identification and validation of the target
large-scale lead optimization [2,3]. These approaches have often structure [11]. The structural information for all targets is generally
shown multifaceted outcomes that escorted the way to retrace the obtained by X-ray crystallography or NMR. However, in the case of
underlying principles and nurture novel concepts such as receptor targets with no experimentally determined structures, several
flexibility [4], multiple conformation accountability [5], and phar- computational approaches such as ab initio modeling, threading
macophore-based virtual screening [6]. The amount of time, cost and comparative modeling can be used to predict 3D structures.
and labor spent in SBDD is comparatively low, but it can have great Homology modeling or comparative modeling is the most reliable
method for target structure prediction that builds 3D structures for
Corresponding author:. Chen, Y.-P. ([email protected]) unknown proteins based on the known homologous protein

1359-6446/06/$ - see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.drudis.2011.07.006 www.drugdiscoverytoday.com 831
 REVIEWS Drug Discovery Today  Volume 16, Numbers 17/18  September 2011
[(Figure_1)TD$IG]

Target identification

(b)
3D No
Structure prediction
structure through homology De novo ligand Virtual compound
available modeling / threading design library

Yes
Reviews  POST SCREEN

Structure optimization
Structure optimization

Compound structure library

Target bindingsite

Target structure

(c)

Dock and score compounds

Docked complex optimization

Analyse ranked list and pick toppers

Are hits No New analogs generation through

potential? modification of existing hits

‘Hit’ identification phase

Yes

'Hit to lead' phase

Molecular simulations

QSAR, ADME and toxicity

Synthesis of compound and testing for

binding in biochemical assays

Satisfied results with desired potency

Lead optimization and experimental trials

Drug Discovery Today

FIGURE 1
Process flow for the ‘hit identification’ phase and ‘hit-to-lead’ phase – the structure-based drug design (SBDD) approach. (a) Target structure preparation and
binding-site analyses; (b) compound library preparation; (c) docking of compounds against the target binding pocket and analyses.

structures (i.e. >40% similarity) [12]. There are numerous auto- When the target structure is predicted a few steps of validation
mated programs that can make such high quality protein structure are essential to confirm the stereochemical quality of the predicted
predictions. Because 3D structures are the fundamental require- structures before proceeding further in SBDD. One common
ments to begin SBDD, most of the drug design packages listed in method for such quality checks includes a Ramachandran plot
Table 1 include structure prediction utilities. For example, Dis- analysis that displays the backbone conformational angles for
covery Studio from Accelrys, Prime Module from Schrödinger and all of the amino acids in the protein structures [13]. Structural
Advanced Protein Modeling from Sybyl are a few widely used analysis and verification server (SAVES; http://nihserver.mbi.
modeling programs. ucla.edu/SAVES/) is a structure validation server that offers a

832 www.drugdiscoverytoday.com
Drug Discovery Today  Volume 16, Numbers 17/18  September 2011 REVIEWS

TABLE 1
Drug discovery software packages
Software Features Source
TS BS MB DB MM Dc Pc MD QSAR Sc ADME TP
Discovery studio            http://accelrys.com/
Biograf         http://www.biograf.ch/
BiosolveIT      http://www.biosolveit.de/

Reviews  POST SCREEN

Sybyl          http://www.optive.com/
Forecaster 2.6.1     http://www.fitted.ca/
Aurora drug     http://www.aurorafinechemicals.com/
discovery software
Spartan       http://www.computational-chemistry.co.uk/
ChemTree       http://www.goldenhelix.com/
Quantum   j    http://www.q-pharm.com/
Molecular modeling           https://www.schrodinger.com/
platform
MoE           http://www.chemcomp.com/
Hyperchem       http://www.hyper.com/
Abbreviations: TS, target 3D structure prediction; BS, binding site prediction; MB, molecule builder; DB, database search; MM, molecular mechanics; Pc, pharmacophore; MD, molecular
dynamics simulation; QSAR, quantitative structure activity relationship; Sc, screening; ADME, absorption distribution metabolism excretion; TP, toxicity prediction; [j: selective channels].

combination of structure evaluation tools attributing to different identification approaches that can handle varieties of biological
structural parameters. It is one example of the programs that are targets effectively and identify pharmacologically sound hits
used for structure quality evaluation. A detailed description of the becomes inevitable [21]. This demands the use of SBDD in hit
evaluation procedures can be found in several reviews [12,14,15]. discovery. SBDD incorporates two diverse strategies for the iden-
tification of potential hits such as virtual high-throughput screen-
Binding site recognition ing (vHTS) and de novo design.
The binding site is a small region, a pocket or bumps, where ligand
molecules can best fit or bind to activate the receptor and/or target Virtual screening
and produce the desirable effect. Thus, recognizing the binding Although looking for a ‘fresh recipe’ for novel medicines is a
site or the active site residues in the target structure is of high process of invariable selection, researchers try to identify the most
importance in SBDD. Because the proteins are capable of under- effective assortment from millions of potential compounds by the
going conformational changes, recognizing the accurate binding most efficient method. Virtual screening (VS), also called vHTS,
site residues is difficult [5]; but still there are just a few computa- has been globally attributed as being an alternative approach to
tional programs, such as Ligsitecsc [16], Qsite finder [17] and CASTp HTS. vHTS computationally screens large chemical libraries to
[18], that can capably spot out the binding site residues. Qsite search for compounds that possess complementarities toward
finder [17] locates and clusters the favorable binding sites using the the targets [2,22]. The screening of compounds in vHTS is carried
interaction energy and Van der Waal’s probes, whereas CASTp [18] out using docking calculations where the compounds are filtered
employs functionally annotated residues for mapping the surface based on their binding energies against the target [22,23]. Because
pockets. Thus, more-reliable active site predictions can be carried these types of screening techniques are mainly data driven, data
out with the available software resources for SBDD. accessibility remains highly significant. The past decade has seen a
With the targets and their binding site having been defined, the number of useful chemical databases that serve as knowledgebases
next crucial step in SBDD is hit discovery, which probably results for efficient VS [24]. Table 2 lists a few commonly available small
in a library of compounds that can interact with the target. It is an molecule databases that hold millions of chemical information
accepted fact that proper selection of chemical compounds, with that is freely accessible. Also, several chemical suppliers provide
minimal potency and specificity, during the early phases of drug free access to their compound catalogues which can be used for
discovery plays a vital part in the success of the final lead opti- screening purposes [25]. These days, faster and relatively inexpen-
mization stages [19]. Thus, hunting for novel chemical entities sive clusters of computers have increased the speed at which drug
with therapeutic values remains the central goal of pharmaceutical leads can be identified and evaluated [26].
chemistry. To date, HTS has been considered to be the main VS is generally categorized into two methods: viz. structure-
process used for hit identification. Although this technique is very based VS and ligand-based VS [27], structure-based VS is very useful
efficient and powerful in screening compounds of interest, it when the target structure is available and is more popular among
consumes a lot of time and materials to perform experimental the research community. Equally, ligand-based VS that uses pre-
studies for huge combinatorial space (i.e. cost is high). Further, dictive compound activity models based on previously available
with the increased size of a screening library the efficiency of experimental data is also gaining popularity [28,29]. Any type of
HTS tends to decrease [20]. Hence, employing alternative hit VS results in a library of short-listed compounds with increased

www.drugdiscoverytoday.com 833
 REVIEWS Drug Discovery Today  Volume 16, Numbers 17/18  September 2011

TABLE 2
Chemical structure/small molecule database
Database Number of Sub-structure Structure Similarity Molecular Clustering Experimental
records search formats search descriptors data info
Pubchem >37 Million(C) > Yes Smiles, SDF, Yes Yes Yes Yes
70 Million(S) 2D, 3D
Drug Bank Approx.4800 Yes Smiles, SDF, 2D, Yes Yes No Yes
3D, Mol, PDB
Reviews  POST SCREEN

SPRESIweb 7 million Yes SDF, PDF Yes Yes No Yes

KEGG Ligand 17,627 Yes Mol, KCF Yes Yes* No Yes
Chem DB >600,000 Yes Mol, SDF Yes Yes No No
Protein ligand 485 No PDB No No No Yes
database v1.3
Chem Bank 2344 Yes Smiles, Smarts Yes Yes No Yes
ChemPDB 8702 Yes Smiles Yes No No Yes
ChemSpider >20 Million Yes Mol, Smiles, Yes Yes No Yes
SDF, SKC, CDX
ChemIDplus >380,000 Yes SDF, Mol, PDB, Yes Yes No Yes*
Smiles, RDF
Zinc >2.7 Million Yes Smiles, No Yes* No Yes*
Mol2, SDF
Timetec’s bioscreening 600,000 Yes Mol, SDF Yes No No No
database
PDBbind 3214 Yes Smiles Yes BD No Yes
AffinDB 748 No PDB No BD No Yes
BindingMoad 9000 No PDB, SDF, No BD No Yes
Smiles, Mol
BindingDB 11,000 Yes SDF, PDB, Yes Yes No Yes
Mol, Smiles
Yes*: information from other linked database. Abbreviation: BD, binding data.

potency and selectivity. The initial VS library basically has three number of experimentally tested compounds) than the empirical
different classifications: ‘general’, ‘focused’ and ‘targeted’. The screening techniques. This factor makes VS as important as HTS in
‘general libraries’ are the ones that correspond to the broad cate- ‘hit discovery’ [25,33,34].
gory of targets; the ‘focused library’ mainly deals with a family of
related targets only; and the ‘targeted libraries’ are particular to a De novo design
specific target, for instance kinase inhibitor libraries for ‘cyclin- De novo design is a process of creating or building new lead
dependant kinase b’ [30]. The outcome of the screening process is compounds from scratch. This process complements vHTS and
highly dependent on the type of virtual library created. The size of HTS in hit discovery. The main principle of de novo design is to
each library generated through the VS procedure is very high and construct the small-molecule chemical structures that best fit the
libraries often consist of a greater number of compounds than target space [35]. This can be achieved through two different
expected. Every compound in the library is individually docked strategies: namely receptor/target-based design and ligand-based
into the receptor-binding site and is ranked based on the interac- design – the former method being more prevalent than the latter.
tions with the target and the relative scoring functions. Docking In receptor-based de novo design high-quality protein structures
and scoring functions are discussed later as part of this review. and their respective binding sites are essential because the hits are
The potential of computer-based screening has already been designed based on the target structures by placing small fragments
well demonstrated by the identification of several inhibitors and in the key interaction sites of the proteins. The positioning of
antagonists [31]. A successful effort on VS led to the development fragments can be either pre-docked with the structure or placed by
of the patented drug candidate SC12267 for the treatment of the program [36].
rheumatoid arthritis. SC12267 is currently in Phase II clinical Receptor-based design can be carried out by two means: linking
trials. Screening of commercially available compounds with the and growing techniques. In the linking process different small
crystal structure of dihydroorotate dehydrogenase (DHODH) fragments such as amines, single rings and hydrocarbons from the
resulted in the identification of novel cyclic aliphatic carboxylic libraries are added simultaneously to different active site residues
acids that led to the discovery of SC12267 [32]. of the target [37]. Thus, the small fragments positioned at the
VS demonstrated tenfold higher ‘hit rates’ (i.e. number of binding site link to each other and form a final single compound
compounds that bind at a specific concentration and/or total (Fig. 2). This approach is widely preferred by the researchers

834 www.drugdiscoverytoday.com
Drug Discovery Today  Volume 16, Numbers 17/18  September 2011 REVIEWS
[(Figure_2)TD$IG]

Active site analysis Key interaction sites

De Novo tools
De novo design
LUDI
NEWLEAD
SMoG
CAVEAT
LEGEND

Reviews  POST SCREEN

BUILDER
LIGBUILDER
RECAP
GANDI
Linking process RECORE

Link

First fragment

Growing process Grow

Drug Discovery Today

FIGURE 2
Molecule-building strategies – de novo design. The base fragments grow complementarily to the binding-site cavity, linking to each other. The molecule shown is
mock; [TD$INLE] : receptor sites; [TD$INLE] : new lead.

because the fragment design strategy is insightful in that most Docking and scoring functions
biological targets encompass discrete binding sites for each piece The key advantage with SBDD is the ability of the method to depict
of a ligand. It would be more significant if the ligand key regions the experimental binding mode of a small molecule bound to the
were focused through sectional planning. Whereas, in the growing target structure. Docking, a process of predicting the ligand con-
technique a single small fragment is placed in the active site of the formation and its orientation inside the target structure, plays a
target and this fragment grows well complementarily against the vital part in SBDD. The interaction or fit between the ligand and
receptor-binding site – thereby resulting in a library of chemical the protein structure is best represented as the ‘hand and glove’
compounds that are more specific to the target (Fig. 2). model [43]. Docking is often carried out in two parts. The first part
Fragment placing can be carried out via two different includes the effective search of conformational space through a
approaches: namely, the ‘outside-in’ approach and the ‘inside- ‘posing’ mechanism where the ligand is placed inside the receptor
out’ approach. During the ‘outside-in’ approach the fragment is in different orientations to facilitate the identification of the
initially placed at the edges of the binding site and the fragment is actual binding mode of the ligand molecules. Several algorithms
built inward. Software programs such as Caveat [38] and SPROUT such as genetic algorithms, the Monte Carlo algorithm, Evolu-
[39] are used for this approach. During the ‘inside-out’ approach tionary algorithms, simulated annealing algorithms, empirical
the fragments are randomly attached to the binding site and built approaches, knowledge-based algorithms, the SIS algorithm, the
outward. For this approach, a software program such as LUDI [40] Hammerhead algorithm, and the Fast Fourier Transform approach
can be useful. are used for the effective search of parameter space. An energy-
There are a number of reliable software programs for generating based score, ‘scoring functions’, is provided for each pose in terms
novel scaffolds through a de novo design strategy (Fig. 2). Such of their interactions with the receptor. These scoring functions are
automated de novo design has demonstrated its potential in hit and features that aid in investigating the interactions between the
lead discovery. Recently, Ni et al. have successfully designed a small molecule and the biological target, thereby providing con-
highly potent lead for Cyp A PPIase using the de novo design text about biological activity. The methods such as force field,
strategy and the SAR study revealed that two of the inhibitors empirically derived methods or knowledge-based methods are
showed 31.6  2.0 nm inhibition [41]. Thus, the mixture of a well- generally used to position the ligand into the binding site
characterized target and fragment-based methods favors the dis- (Box 1) [44]. Other possibilities such as rule-based methods,
covery of novel lead molecules with improved affinity, selectivity grid-based methods and multiple copy simultaneous search
and physiochemical properties [42]. (MCSS) are used to obtain the primary target constraints

www.drugdiscoverytoday.com 835
 REVIEWS Drug Discovery Today  Volume 16, Numbers 17/18  September 2011

BOX 1 Following this, there were a number of potential inhibitors

identified by SBDD techniques, for example amprenavir (against
Scoring functions and search algorithms for effective HIV protease) by modeling and molecular dynamics simulations
parameter space search [46,47] [32,49], raltitrexed (against thymidylate synthase) by group mod-
Scoring function – ‘Mathematical methods to predict the strength ification [43,50] and norfloxacin (fluoroquinolone antibiotic) by
of the interaction between docked molecules’.
QSAR modeling [51,52]. Equally, there have also been a lot of
Force fields – Affinities estimated from inter, intra molecular and
desolvation energies between the ligand and the receptor. failures with SBDD, for example the failure of the anti-depressant/
Empirical – Counting of various interaction terms between the anxiolytic PRX00023, which was believed to be a selective agonist
Reviews  POST SCREEN

ligand and the receptor. of the 5-HT1A receptor but turned out to be an antagonist with less-
Knowledge-based – Statistical observations of intermolecular pronounced efficacy [53]. These failures were caused by several
close contacts in large 3D databases are used to derive potentials challenges and limitations found at the various stages of drug
of mean force. design. Although the operational workflow of SBDD involves
Consensus scoring – Use of multiple scores to balance errors in
several effective strategies, each of these components has their
single scores and enhance the probability of identifying true
positives. own restrictions that really defy the accuracy and success of the
Genetic algorithm – Start with random population, random SBDD approach. The process of target identification is smooth
crossovers and mutations and select results with higher scores for only when the experimental 3D structures of the targets are
next iteration. available. If the target 3D structure is not available then the
Fast Fourier transform – A fast method for computing the scores structure can be predicted with the available computational
of many different alignments. resources; but the predictions are effective only if there is a higher
Monte Carlo – Searches a series of samples from the joint
level of sequence identity between the target and template. When
probability distribution of two or more variables.
Evolutionary algorithm – Use of fitness functions on the initial the sequence identity is low, that is <40% similarity, then several
randomly generated solutions for next iteration. levels of refinement have to be carried out to predict a quality
Simulated annealing – Use of uphill moves on a discrete search target structure, which is really complex.
space to provide a random ‘nearby’ solution. In the case of lead generation through de novo design, this
SIS (sequential importance sampling) – Based on recursive process seems to be agreeable yet it does hold drawbacks in the
propagation of weights and support points for each sequential linking process, which involves connecting the key ligand frag-
score.
ments that are placed in accordance with the target cavity and
property. Further, although this process of fragment building is
(i.e. the binding affinity energy). The second part of the docking said to be fully automated, final selection of the potential chemical
process is the ordering of poses based on their computed scores entities is manually carried out, which is extremely tedious. Above
[45]. all, the compounds that are designed de novo are often not the
Although the concept of docking seems to be simple, it appends favorite choice of a medicinal chemist, because the synthesis of
higher complexity at each level and this is usually initiated by the such new fragment combinations in the laboratory is not always
algorithm beneath the docking process. For instance, the insertion feasible. Also, the software fails to address the binding behavior
of a flexibility factor to some active regions of the protein can help (primary constraint) and pharmacokinetic profiles (secondary
in identifying new hits with better interaction very quickly, when constraints) accurately. Thus, the software options that offer de
compared with traditional rigid docking. But, this requires more- novo design need to consider the synthesis factors of the com-
complex calculations because the protein of high conformational pounds, because the ones that are not synthesizable are of no use.
energy score would result in idealistic positioning of the ligand. In the case of the docking process, there are a variety of
Comprehensive reviews featuring comparisons of the different advanced docking software packages that are available and every
algorithms and scoring functions are frequently available in the well known docking program has been developed with a number
literature [47,48]. A list of common software resources available for of scoring functions. Hence, the process of choosing the appro-
protein–ligand and protein–protein docking and their scoring priate scoring function is tedious and complicated. No single
functions are presented in Table 3. software pertains to prove its efficiency to all types of target and
inhibitor combinations. Further, calculation of accurate binding
SBDD challenges energies and absolute binding affinity for the docked complex is
The ‘birth’ of a new drug in the pharmaceutical market usually still in need of solution. A number of factors determine this lack of
costs millions of dollars, very recently estimated at US$800 mil- ability to cope with all types of complex structures and this
lion. Hence, pharmaceutical companies and researchers are ready demands a software system that can integrate multiple strategies
to make use of every chance that can help in reducing the mone- and scoring functions to filter out the true positives, thereby
tary burden of drug design. SBDD is one such computational resulting in a clean rank-list of potential compounds.
methodology that has been quickly recognized and globally VS, a valuable drug design strategy, is recognized to reduce the
accepted by drug researchers and medicinal scientists. Also, this time and cost involved in regular HTS techniques. Although VS is
method has demonstrated a considerable amount of success over successful in generating libraries of reliable hits by screening
the past decades. The foremost success of SBDD includes the millions of compounds quickly with minimal computational
identification of inhibitors targeting HIV-1 that have been power, the main drawback is that it neglects the effect of tauto-
approved by the FDA and have therefore reached the market merism, protonation and ionization states of the chemical struc-
effectively [49]. tures. Most of the docking studies today are oriented with

836 www.drugdiscoverytoday.com
Drug Discovery Today  Volume 16, Numbers 17/18  September 2011 REVIEWS

TABLE 3
Molecular docking tools for virtual screening
Software Method Source Application
AutoDock Lamarckian genetic algorithm http://autodock.scripps.edu/ PL
Gold Genetic algorithm http://www.ccdc.cam.ac.uk/ PL
a
FRED Directed docking with SMARTS patterns http://www.eyesopen.com/fred PL
FlexX SIS-algorithmb http://www.biosolveit.de/FlexX/ PL
Dock Vision Hybrid evolutionary algorithm http://dockvision.com/ PL

Reviews  POST SCREEN

Surflex-Dock Hammerhead docking system http://www.tripos.com/index.php?family=modules, PL
SimplePage,,,&page=surflex_dock&s=0
Situs Correlation-based rigid body docking and http://situs.biomachina.org/index.html ARD
density filtering
DOCK Geometric matching algorithm http://dock.compbio.ucsf.edu/ PPL
GRAMM Empirical approach http://vakser.bioinformatics.ku.edu/main/resources_gramm.php PPL
ICM-Docking Flexible docking http://www.molsoft.com/docking.html PPL
3D-Dock Suite Fourier correlation algorithm, self consistent http://www.sbg.bio.ic.ac.uk/docking/index.html PP
mean field optimization procedure, single
distance constraint empirically derived pair potential
BiGGER Soft-docking http://www.cqfb.fct.unl.pt/ PP
DOT Fast Fourier transforms http://www.sdsc.edu/CCMS/DOT/ PP
ArgusLab Genetic algorithm and ArgusDock http://www.arguslab.com PP
HADDOCK Uses ambiguous interaction restraints of NMR http://www.nmr.chem.uu.nl/haddock/ PP
Hex Similar to conventional fast Fourier transform (FFT) http://www.loria.fr/ritchied/hex/ PP
rDock Weighted sum of intermolecular, ligand intramolecular, http://www.ysbl.york.ac.uk/rDock PL and
site intramolecular and external restraint terms RNA-ligand
Abbreviations: PL, protein–ligand docking software; PPL, protein–protein and protein–ligand docking software; PP, protein–protein docking software; ARD, atomic resolution structure
docking software.
a
SMiles ARbitrary target specification.
b
Sequential importance sampling.

compounds in the chemical database that stores their canonical tautomerization easy to ignore but irresistible to consider [54]. For
form. This restricts many other forms of the compound structure example, Tempirini et al. [55] showed that the crystal structure of
such as tautomers, and they do not get the chance of being carbonic anhydrase II (3F4X) is bound with the lactim tautomer
considered as a hit during the screening process. Lack of experi- form of chlorthalidone, a carbonic anhydrase II inhibitor, instead
mental data and authenticated computational predictions make of the usual amide form. This shows that the protein prefers

TABLE 4
Merits and demerits of SBDD ‘hit’ identification strategies
Technique Merits Demerits
Virtual screening Enable filtering of millions of Key factors like tautomerism, ionization and protonation are
compounds from database usually ignored resulting in filtering out some of the significant hits.
to identify potential ‘hits’. Ability to include multiple conformations of protein and ligand in
screening is still hard to reach owing to the demand of
higher computational power.
Redundancy of data in chemical databases.
De novo design An Insightful technique, because Process of linking key ligand fragments and the chemical synthesis
most biological targets of new fragment combinations are difficult.
encompass discrete binding sites Prediction of crucial properties like primary constraints and
for each piece of a ligand. It secondary constraints remain challenging.
offers a broader exploration of the
chemical space to identify novel scaffolds. Inclusion of receptor flexibility by the de novo design softwares
has not been pronounced effectively.
Docking Includes different potential scoring The process of choosing an appropriate scoring function and
functions and algorithms, incorporating algorithm for a specific target and lead compound is intricate
protein and ligand flexibility and tricky. Accurate calculations of ligand receptor binding
to mimic the experimental binding mode. energy are still out of our reach.

www.drugdiscoverytoday.com 837
 REVIEWS Drug Discovery Today  Volume 16, Numbers 17/18  September 2011

specific chemical state of inhibitors for the desired therapeutic Concluding remarks and future outlook
activity. The importance of such factors has also been elucidated Although SBDD holds a few drawbacks, it serves the medicinal
by other studies [56,57]. Although a few protocols, such as quan- researchers in making a drug discovery effort with minimum time
tum-mechanics- and molecular-mechanics-based docking, help in and an added level of confidence. With the vast numbers of
the assignment of proper states to the receptor and ligand molecules available resources it is necessary to choose an appropriate
while docking, they are time-consuming and require high-level resource that could envisage the genuine state of the biological
computational powers to perform large-scale calculations. Once system to fabricate better therapeutic achievements. Synchroniza-
these challenges are overcome by technological and knowledge- tion and enrichment of the experimental data availability and
Reviews  POST SCREEN

based advancements, SBDD would be more beneficial not just for hit addressing the challenges and limitations countenanced can ele-
discovery but also the entire process of drug design and discovery. A vate SBDD to hold the primary role for identifying a compound as
summary of the merits and demerits are listed in Table 4. a ‘lead’ in future drug discovery research.

References
1 Alonso, H. et al. (2006) Combining docking and molecular dynamic simulations in 27 Krüger, D.M. and Evers, A. (2010) Comparison of structure- and ligand-based virtual
drug design. Med. Res. Rev. 26, 531–568 screening protocols considering hit list complementarity and enrichment factors.
2 Keseru, G.M. and Makara, G.M. (2009) The influence of lead discovery strategies on ChemMedChem 5, 148–158
the properties of drug candidates. Nat. Rev. Drug Discov. 8, 203–212 28 Muegge, I. and Oloff, S. (2006) Advances in virtual screening. Drug Discov. Today
3 Goodnow, J.R.A. (2006) Hit and lead identification: integrated technology-based Technol. 3, 405–411
approaches. Drug Discov. Today: Technol. 3, 367–375 29 Sperandio, O. et al. (2009) wwLigCSRre: a 3D ligand-based server for hit
4 Fischer, B. et al. (2008) Receptor Flexibility for Large-Scale In Silico Ligand identification and optimization. Nucleic Acids Res. 37 (Suppl. 2), 504–509
Screens. In Molecular Modeling of Proteins, Vol. 443 (Kukol, A., ed.), Humana Press pp. 30 Tang, Y. et al. (2006) New technologies in computer-aided drug design: toward
353–364 target identification and new chemical entity discovery. Drug Discov. Today Technol.
5 Liao, J.J. and Andrews, R.C. (2007) Targeting protein multiple conformations: a 3, 307–313
structure-based strategy for kinase drug design. Curr. Top. Med. Chem. 7, 1394–1407 31 Coumar, M.S. et al. (2009) Structure-based drug design of novel aurora kinase A
6 Sun, H. (2008) Pharmacophore-based virtual screening. Curr. Med. Chem. 15, inhibitors: structural basis for potency and specificity. J. Med. Chem. 52, 1050–1062
1018–1024 32 Clark, D.E. (2006) What has computer-aided molecular design ever done for drug
7 Chen, Y.P.P. and Chen, F. (2008) Using bioinformatics techniques for gene discovery? Exp. Opin. Drug Discov. 1, 103–110
identification in drug discovery and development. Curr. Drug Metab. 9, 567–573 33 Doman, T.N. et al. (2002) Molecular docking and high-throughput screening for
8 Searls, D.B. (2005) Data integration: challenges for drug discovery. Nat. Rev. Drug novel inhibitors of protein tyrosine phosphatase-1B. J. Med. Chem. 45, 2213–2221
Discov. 4, 45–58 34 Schneider, G. (2010) Virtual screening: an endless staircase? Nat. Rev. Drug Discov. 9,
9 Smith, C. (2004) Drug target identification: a question of biology. Nature 428, 273–276
225–231 35 Thomas, G. (2007) Computer-aided drug design. In Medicinal Chemistry: An
10 Chan, J.N.Y. et al. (2010) Recent advances and method development for drug target Introduction (2nd Edn.) (Thomas, G., ed.), Wiley pp. 113–143
identification. Trends Pharmacol. Sci. 31, 82–88 36 Schneider, G. and Fechner, U. (2005) Computer-based de novo design of drug-like
11 Grant, M.A. (2009) Protein structure prediction in structure-based ligand design and molecules. Nat. Rev. Drug Discov. 4, 649–663
virtual screening. Combinatorial Chem. High Throughput Screen. 12, 940–960 37 Keseru, G.M. and Makara, G.M. (2006) Hit discovery and hit-to-lead approaches.
12 Song, C.M. et al. (2009) Recent advances in computer-aided drug design. Brief Drug Discov. Today 11, 741–748
Bioinform. 10, 579–591 38 Lauri, G. and Bartlett, P.A. (1994) CAVEAT: a program to facilitate the design of
13 Krieger, E. et al. (2009) Improving physical realism, stereochemistry, and side-chain organic molecules. J. Computer-Aided Mol. Design 8, 51–66
accuracy in homology modeling: four approaches that performed well in CASP8. 39 Gillet, V.J. et al. (1994) SPROUT: recent developments in the de novo design of
Proteins Struct. Funct. Bioinform. 77, 114–122 molecules. J. Chem. Inform. Computer Sci. 34, 207–217
14 Bordoli, L. et al. (2008) Protein structure homology modeling using SWISS-MODEL 40 Böhm, H.-J. (1992) The computer program LUDI: a new method for the de novo
workspace. Nat. Protocols 4, 1–13 design of enzyme inhibitors. J. Computer-Aided Mol. Design 6, 61–78
15 Potapov, V. et al. (2010) Protein structure modelling and evaluation based on a 4- 41 Ni, S. et al. (2009) Discovering potent small molecule inhibitors of cyclophilin A
distance description of side-chain interactions. BMC Bioinform. 11, 374 using de novo drug design approach. J. Med. Chem. 52, 5295–5298
16 Huang, B. and Schroeder, M. (2006) LIGSITEcsc: predicting ligand binding 42 Bembenek, S. et al. (2009) Ligand efficiency and fragment-based drug discovery.
sites using the Connolly surface and degree of conservation. BMC Struct. Biol. 6, 19 Drug Discov. Today 14, 278–283
17 Laurie, A. and Jackson, R. (2005) Q-SiteFinder: an energy-based method for the 43 Anderson, A.C. (2003) The process of structure-based drug design. Chem. Biol. 10,
prediction of protein-ligand binding sites. Bioinformatics 21, 1908–1916 787–797
18 Binkowski, T. et al. (2003) CASTp: computed atlas of surface topography of proteins. 44 Huang, S.-Y. et al. (2010) Scoring functions and their evaluation methods for
Nucleic Acids Res. 31, 3352–3355 protein–ligand docking: recent advances and future directions. Phys. Chem. Chem.
19 Wunberg, T. et al. (2006) Improving the hit-to-lead process: data-driven assessment Phys. 12, 12899–12908
of drug-like and lead-like screening hits. Drug Discov. Today 11, 175–180 45 Fenu, L.A. (2007) Scoring functions. In Structure-Based Drug Discovery (Jhoti, H.,
20 Kato, R. et al. (2005) Novel strategy for protein exploration: high-throughput Leach, A.R., eds), Springer pp. 223–245
screening assisted with fuzzy neural network. J. Mol. Biol. 351, 683–692 46 Sousa, S.F. et al. (2006) Protein-ligand docking: current status and future challenges.
21 Bleicher, K.H. et al. (2003) Hit and lead generation: beyond high-throughput Proteins Struct. Funct. Bioinform. 65, 15–26
screening. Nat. Rev. Drug Discov. 2, 369–378 47 Kitchen, D.B. et al. (2004) Docking and scoring in virtual screening for drug
22 Shoichet, B.K. (2004) Virtual screening of chemical libraries. Nature 432, discovery: methods and applications. Nat. Rev. Drug Discov. 3, 935–949
862–865 48 Kastritis, P.L. and Bonvin, A.M.J.J. (2010) Are scoring functions in protein–protein
23 Phatak, S.S. et al. (2009) High-throughput and in silico screenings in drug discovery. docking ready to predict interactomes? Clues from a novel binding affinity
Exp. Opin. Drug Discov. 4, 947–959 benchmark. J. Proteome Res. 9, 2216–2225
24 Williams, A.J. (2008) A perspective of publicly accessible/open-access chemistry 49 Wlodawer, A. and Vondrasek, J. (1998) Inhibitors of HIV-1 protease: a major
databases. Drug Discov. Today 13, 495–501 success of structure-assisted drug design1. Annu. Rev. Biophys. Biomol. Struct. 27,
25 Hein, M. et al. (2010) Docking compared to 3D-pharmacophores: the scoring 249–284
function challenge. Drug Discov. Today Technol. 7, 229–236 50 Boyd, D.B. (2007) How computational chemistry became important in the
26 Gershell, L.J. and Atkins, J.H. (2003) A brief history of novel drug discovery pharmaceutical industry. In Reviews in Computational Chemistry, Vol. 23 (Lipkowitz,
technologies. Nat. Rev. Drug Discov. 2, 321–327 K.B. et al. eds), Wiley pp. 401–443

838 www.drugdiscoverytoday.com
Drug Discovery Today  Volume 16, Numbers 17/18  September 2011 REVIEWS

51 Rutenber, E.E. and Stroud, R.M. (1996) Binding of the anticancer drug ZD1694 to 55 Temperini, C. et al. (2009) Carbonic anhydrase inhibitors. Comparison of
E. coli thymidylate synthase: assessing specificity and affinity. Structure 4, chlorthalidone, indapamide, trichloromethiazide, and furosemide X-ray crystal
1317–1324 structures in adducts with isozyme II, when several water molecules make the
52 Boyd, D.B. and Marsh, M.M. (2006) The history of computers in pharmaceutical difference. Bioorg. Med. Chem. 17, 1214–1221
research and development. In Computer Applications in Pharmaceutical Research and 56 Cho, A.E. et al. (2005) Importance of accurate charges in molecular docking:
Development (Ekins, S., Wang, B., eds), Wiley pp. 1–50 quantum mechanical/molecular mechanical (QM/MM) approach. J. Comput. Chem.
53 De Paulis, T. (2007) Drug evaluation: PRX-00023, a selective 5-HT1A receptor 26, 915–931
agonist for depression. Curr. Opin. Investig. Drugs 8, 78–86 57 ten Brink, T. and Exner, T.E. (2009) Influence of protonation, tautomeric, and
54 Martin, Y. (2009) Let’s not forget tautomers. J. Computer-Aided Mol. Design 23, stereoisomeric states on protein–ligand docking results. J. Chem. Inform. Modeling
693–704 49, 1535–1546

Reviews  POST SCREEN

www.drugdiscoverytoday.com 839