Journal of Molecular Liquids 309 (2020) 113165

Contents lists available at ScienceDirect

Journal of Molecular Liquids

journal homepage: www.elsevier.com/locate/molliq

Quantitative structure properties relationship for deep eutectic solvents

using Sσ-profile as molecular descriptors
Tarek Lemaoui a, Nour El Houda Hammoudi a, Inas M. Alnashef b, Marco Balsamo c, Alessandro Erto d,
Barbara Ernst e, Yacine Benguerba a,⁎
a
Laboratoire des Matériaux Polymères Multiphasiques, LMPMP, Université Ferhat ABBAS Sétif-1, 19000 Sétif, Algeria
b
Department of Chemical Engineering, Khalifa University, Sas Al Nakhl campus, Abu Dhabi 127788, United Arab Emirates
c
Dipartimento di Scienze Chimiche, Università degli Studi di Napoli Federico II, Complesso Universitario di Monte Sant'Angelo, 80126 Napoli, Italy
d
Dipartimento di Ingegneria Chimica, dei Materiali e della Produzione Industriale, Università di Napoli Federico II, P. le Tecchio, 80, 80125 Napoli, Italy
e
Université de Strasbourg, CNRS, IPHC UMR 7178, Laboratoire de Reconnaissance et Procédés de Séparation Moléculaire (RePSeM), ECPM 25 rue Becquerel, F-67000 Strasbourg, France

a r t i c l e i n f o a b s t r a c t

Article history: Computer assisted Quantitative Structure Property Relationship (QSPRs) has proven to be an accurate, re-
Received 14 February 2020 liable and cost-effective method for predicting the physicochemical properties of DESs, via a set of molec-
Received in revised form 10 April 2020 ular descriptors. In this work, experimental data on the properties of DESs at different temperatures were
Accepted 14 April 2020
taken from different bibliographic sources. The Conductor like Screen Model for Real Solutions (COSMO-
Available online 17 April 2020
RS) was used to predict the thermodynamic properties of DESs. A modeling analysis was conducted in
Keywords:
order to provide a model for the prediction of specific DESs properties, such as viscosity density, etc. The
Deep eutectic solvents used methodology allowed achieving reliable results as all the models showed high regression perfor-
Physiochemical properties mances. The corresponding model parameters were determined and an analysis of variance allowed iden-
Multilinear regression tification of the most significant factors of the retrieved models. Finally, an independent set of experimental
Modeling data relevant to the modelled physical properties of DESs was used to test the obtained models. In most
COSMO-RS cases, there was a good agreement between the experimental and predicted values of the investigated
Quantitative Structure Property Relationship properties.
(QSPR)
© 2020 Elsevier B.V. All rights reserved.

1. Introduction for their use in green technologies because they have an intrinsic toxic-
ity and a high volatility [1].
The use of solvents in many industrial applications is of paramount In recent decades, efforts have been made to replace organic sol-
importance. Large-scale applications include industrial separations in vents with alternative classes of chemical compounds. These ap-
different fields such as pharmaceutical, food, metal refining, biochemi- proaches include the use of easily recyclable systems, such as
cal, and in wastewater treatment. While the extraction methods have fluorinated solvents, the elimination of solvents from productive cycle
now become a routine procedure in separation technologies, the correct (whenever possible) and the use of non-volatile compounds, such as
identification of the optimal solvent with adequate properties for a spe- ionic liquids (ILs) and deep eutectic solvents (DESs).
cific application still represents one of the challenges in this research Ionic Liquids are salts that are usually liquid below 100 °C. Over the
field. last two decades, the number of published articles about ILs has in-
In fact, the choice of an appropriate solvent is essential for both tech- creased exponentially [3]. The great advantage of ILs is that they can
nical and economic reasons, since it represents about 80% of the total be tuned by combining different cations and anions. However, their
volume of chemicals used in a generic process [1]. Solvents present main disadvantages include the difficulty of their processing, mainly
many environmental, health and safety concerns, including human due to their general high viscosity. In addition, the cost of ILs is high
and ecotoxicological problems, process safety hazards and waste man- compared to commercially available solvents. This is due to their rela-
agement issues [2]. Most organic solvents do not fulfil the requirements tively complicated synthesis and purification [4]. To overcome these
disadvantages, DESs have been proposed as a new class of analogues
of the ILs. Although they share many characteristics and properties
⁎ Corresponding author. with ILs, they represent a different type of solvents and have different
E-mail address: [email protected] (Y. Benguerba). chemical nature [5].

https://doi.org/10.1016/j.molliq.2020.113165
0167-7322/© 2020 Elsevier B.V. All rights reserved.
2 T. Lemaoui et al. / Journal of Molecular Liquids 309 (2020) 113165

The first article on DESs was published only in 2001, so it is clear that It is a continuation of the previous work with more properties and
the subject, compared with ILs, is in its infancy. In contrast to ILs, which more DESs used for model set-up [9].
are composed of discrete anions and cations, DESs are formed by a eu- The experimental data used to validate the obtained QSPR models
tectic mixture of Lewis and Bronsted acids and bases. They can include were taken from different and independent studies reported in the liter-
a variety of anionic or cationic species and, as contradistinctive charac- ature [8]. The quantum chemical approach (COSMO-RS) developed by
ter, present a great drop of the melting point as compared to those of Klamt et al. [10] was used to predict the thermodynamic properties of
the pure substances from which they are composed [5]. There are pure and mixed polarity distribution [11].
many potential applications of DESs, for example they can be used in
the electrochemical treatment of metals, as reaction media for various
industrial processes, biochemical processes, drug delivery, etc. Since 2. DESs experimental data retrieved from the literature
DESs serve as an environmental-friendly alternative to organic solvents,
their field of applications can be potentially further enlarged in the next Quantitative Structure-Property Relationship constitutes a powerful
years. analytical method for understanding how the chemical structure of a
The biodegradability and biocompatibility of a DES depend on DES can be correlated with its physicochemical properties. The starting
the compounds used for their preparation. Naturally-based DESs point for deriving QSPR models is the availability of reliable
can be prepared using primary metabolites, namely amino acids, or-
ganic acids, sugars or choline derivatives [6]. These DESs perfectly
fulfil the principles of green chemistry and engineering. Although
there is a large group of DESs with known properties and many of Table 1
them have been already used in different applications, the mecha- DESs used in the training set.

nism of their formation is not well understood and the prediction HBA HBD Mole Abbrev Ref
of their phase diagrams is difficult since various interactions can si- ratio
multaneously occur. On the other hand, since the possible combina- Benzyl triphenyl-phosphonium Ethylene glycol 1:3 DES 1 [12]
tions of constituents is potentially infinite (in terms of both chloride (BTPC) Glycerine 1:5 DES 2 [12]
substances and their relative concentrations), ability to predict Choline chloride (ChCl) D-Fructose 1:1 DES 3 [13]
the properties of a given DES would be an invaluable tool for the D-Fructose 1.5:1 DES 4 [13]
rapid and inexpensive identification of suitable high- D-Fructose 2:1 DES 5 [13]

performances materials. The best option would be the availability D-Fructose 2.5:1 DES 6 [13]

of a calculation tool for testing many possible mixtures and deter- Diethanolamine 1:6 DES 7 [14]
Diethanolamine 1:8 DES 8 [14]
mining their properties before preparing them, based on the knowl- Diethanolamine 1:10 DES 9 [14]
edge of single constituents' properties. In addition, important Ethylene glycol 1:1.8 DES 10 [15]
information about construction principles and intermolecular in- Ethylene glycol 1:2 DES 11 [15]
teractions could be retrieved, which could help in the prediction Ethylene glycol 1:2.5 DES 12 [15]
Glycerine 1:1 DES 13 [15]
of their behavior when used in industrial processes.
Glycerine 1:2 DES 14 [15]
One of the most widely adopted methods to augment experi- Glycerine 1:3 DES 15 [15]
mental analytical techniques is the use of Computer-assisted Quan- Malonic acid 1:1 DES 16 [16]
titative Structure Property Relations (QSPR) [7], which have proven Methyl 1:6 DES 17 [17]
to be an accurate, reliable and cost-effective method for predicting diethanolamine
Methyl 1:8 DES 18 [17]
DESs properties [8]. The general idea is to derive the physicochem- diethanolamine
ical properties of DESs from a set of molecular descriptors, which Methyl 1:10 DES 19 [17]
can be represented by physiochemical properties or theoretical diethanolamine
molecular properties of the used chemicals. The use of QSPR Monoethanolamine 1:4 DES 20 [18]
Monoethanolamine 1:5 DES 21 [18]
models can also significantly help in understanding the relation-
Monoethanolamine 1:6 DES 22 [18]
ships between the microscopic properties of molecular compo- Monoethanolamine 1:7 DES 23 [18]
nents and the macroscopic properties. Many experimental data of Monoethanolamine 1:8 DES 24 [18]
DESs properties formed by a specific hydrogen bond acceptor Monoethanolamine 1:10 DES 25 [18]
(HBA) and a hydrogen bond donor (HBD) are currently available Triethanolamine 1:2 DES 26 [15]
Oxalic acid 1:1 DES 27 [16]
in the literature. It is well known that the type of the HBD and Methyltriphenyl Ethylene glycol 1:3 DES 28 [15]
HBA and their molar ratio in a DES can drastically affect the proper- phosphonium-bromide (MTPB) Ethylene glycol 1:4 DES 29 [12]
ties of the DES. However, since very few methods are currently Ethylene glycol 1:5 DES 29 [15]
available for the prediction of DESs properties without the use of Glycerine 1:2 DES 30 [15]
Glycerine 1:3 DES 31 [15]
experimental data, the set-up of a predictive modeling tool is
Glycerine 1:4 DES 32 [15]
gaining great interest in the scientific community. In order to Glycerine 1:1.8 DES 33 [12]
achieve a reliable model for DESs properties prediction, the avail- Monoethanolamine 1:6 DES 34 [18]
ability of an extensive set of experimental data is necessary, for Monoethanolamine 1:7 DES 35 [18]
model set-up, tuning, and validation. Monoethanolamine 1:8 DES 36 [18]
Monoethanolamine 1:9 DES 37 [18]
The σ-profiles distribution area (Sσ-profile) was previously used as a N, N-diethylethanol-ammonium Malonic Acid 1:1 DES 38 [16]
quantitative representation of the polar surface screen charge of the chloride (DEAC) Ethylene glycol 1:2 DES 39 [15]
molecule on the polarity scale, calculated using the σ-profile given by Ethylene glycol 1:3 DES 41 [15]
COSMO-RS. In recent years, the Sσ-profile has been used by different au- Ethylene glycol 1:4 DES 42 [15]
Glycerine 1:2 DES 43 [15]
thors as a molecular descriptor in QSPR models for the prediction of dif-
Glycerine 1:3 DES 44 [15]
ferent physicochemical properties, such as viscosity and density of ILs Glycerine 1:4 DES 45 [15]
[8]. Tetrabutylammonium bromide Monoethanolamine 1:4 DES 46 [18]
The main objective of this work is to develop reliable tools to extend (TBAB) Monoethanolamine 1:5 DES 47 [18]
the field of application of QSPR models and allow the prediction of fur- Monoethanolamine 1:6 DES 48 [18]
Monoethanolamine 1:7 DES 49 [18]
ther physicochemical properties of DESs, such as density, viscosity, etc.
 T. Lemaoui et al. / Journal of Molecular Liquids 309 (2020) 113165 3

experimental data. In this study, a total of 49 DESs reported in the liter- 3.2. Model set-up
ature (Table 1) were considered. Each DES was individuated by HBA,
HBD, and HBA: HBD molar ratio. The model set-up requires the definition of the independent vari-
The DESs listed in Table 1 contain a total number of five HBAs: Ben- ables, i.e. the Sσ-profile (1−10) for each stated surface and the tempera-
zyl triphenyl phosphonium chloride (BTPC), Choline chloride (ChCl), ture. The dependent variables represent the properties that will be
Methyl triphenyl phosphonium bromide (MTPB), N,N-diethylethanol estimated. In this work, the attention was focused on viscosity (μ) and
ammonium chloride (DEAC) and Tetra Butyl Ammonium Bromide density (ρ). The generic model expression for each property can be
(TBAB) and nine HBDs: D-Fructose, Diethanolamine (DEA), Ethylene expressed as:
glycol (EG), Malonic acid (MA), Glycerine (Gly), Methyl diethanolamine
(MDEA), Monoethanolamine (MEA), Oxalic acid (OA), Triethanolamine
(TEA).

These HBAs and HBDs are representative of a sufficiently wide range Property ¼ ƒ S1mix ; S2mix ; :…; S10
mix ; T ð1Þ
of molecules that allow a robust approach to the modeling prediction of
DESs properties. Figs. 1 and 2 show the chemical structures of the HBAs
and HBDs used to form the 49 DESs. The green coloured area represents In which Simix (i = 1 : 10) are the σ-profile surface area descriptors of
the nonpolar part of the DES, the blue coloured area is the HBD part and the DES.
the red one is the HBA part of the DES. As an example, S1mix can be expressed as:

3. Calculation methods
X
NC
Simix ¼ X j Siσ −profile; j ; i ¼ 1; NC ð2Þ
The COSMO-RS approach was used to predict the important thermo- j¼1
dynamic properties of the DESs. A full and detailed description of the
COSMO-RS theory is given by Klamt and Eckert [10] and Lin and Sandler
where Xj is the mole fraction of the HBA, HBD, or any other component
[19]. The calculation procedure adopted in the COSMO-RS frame work
that may be present in the DES. Siσ−profile, j is the profile-specific surface
was detailed in previous works [20,21]. The geometry of the HBD and
descriptor for the components; NC is the number of components used to
HBA molecules was optimized at the density functional theory (DFT)
form the DES.
level using B3LYP functional with the def-TZVP (triple-ζ valence polar-
The mathematical expression of the model, the commonly used way
ized) basis set [20,21]. The generated COSMO files contain all the infor-
to quantify a simple relation between the dependent variable y and the
mation needed to calculate the σ-profile function and, consequently, to
independent variables (descriptors) xi, i = 1: n, is the multi-linear re-
calculate the descriptor Sσ-profile [22]. Subsequently, ten surfaces of
gression model (MLR), given as:
0.005 e/Å2 wide in the range of −0.025 to 0.025 e/Å2 (x axis) were
used [22].

X
n X
n X
n

3.1. Interpretation of the σ-profiles y ¼ a0 þ ai xi þ aij xi x j ð3Þ

i¼1 i¼1 j¼1
Figs. 3 and 4 show the probability distribution, P(σ), of a molecular
surface segment having a specific charge density for the investigated j≠i
HBAs and HBDs, respectively.
Based on the σ values, the σ-profile curve can be divided into three where ai is the regression coefficient of the descriptor, aij is the interac-
regions corresponding to HBD, nonpolar and HBA, on the. In the HBD re- tion coefficient, and a0 is the intercept.
gion, [−0.025, −0.008], the characteristic peaks of the hydrogen's
atoms (e.g. in N\\H or O\\H bonds) are present. The nonpolar region 4. Results and discussion
[−0.008, 0.008] is characteristic of the CH3 and CH2 alkyl groups. Finally,
the region [0.008, 0.025] is characteristic of hydrogen acceptor atoms The DES properties were analyzed and the corresponding model
(e.g. N, O). equations were derived using the experimental data listed in Table 1.

BTPC ChCl DEAC MTPB TBAB

Fig. 1. COSMO surfaces and chemical structures of HBAs. (For interpretation of the references to colour in this figure, the reader is referred to the web version of this article.)
4 T. Lemaoui et al. / Journal of Molecular Liquids 309 (2020) 113165

Fructose DEA EG

Gly MA MDEA

MEA OA TEA
Fig. 2. COSMO surfaces and chemical structures of HBDs. (For interpretation of the references to colour in this figure, the reader is referred to the web version of this article.)

4.1. Density model density experimental values are reported in Table 4, while the corre-
sponding coefficients of the significant descriptors are listed in Table 5:
For the determination of the model expression for density, 310 ex- In particular, it was found that S1 and S10 have no significant ef-
perimental data points for 17 DESs were used (Table 2). fect on the dependent variable with a p-value N5%. Therefore, a 1
The stepwise regression algorithm was used for the analysis of the and a10 were set equal to zero. However, all the other descriptors
experimental density data expressed by the MLR model descriptors of were significant with a p-value lower than 5% and accordingly
the σ-profile's surfaces, temperature, and the interaction between were retained in the model (Table 5). It was found that there was
them. The summary of MLR performance for the entire data set is a combined effect on the density of S 2 with S 4 (+), S8 (−), and T
given in Table 3. (−); of S3 with S5 (+), S6 (−), and T(−); of S5 with S9 (−); of S6
From the obtained values of the coefficient of determination (R2) with T(−); and finally S7 and S9 with T(+). Where the sign, given
and the root mean squared error (RMSE), 0.9839, 0.0097, respectively, between parenthesis, indicates positive (+) or negative (−) ef-
it can be concluded that the MLR model fits well the experimental fects. It should be noted that the temperature has a double effect
data set. The relationship between the density and the descriptors is sat- on density: a simple negative effect (a11 b 0) and a combined effect
isfactorily multi-linear. with sigma surface segments, which could be positive or negative,
The parameters of the ANOVA analysis carried out for the individua- e.g., S 7 and S 9 (a 711 and a 911 N 0). The temperature has an effect
tion of the descriptors that have a significant influence in modeling the on polarity which in turn influences density. The positive sign of
 T. Lemaoui et al. / Journal of Molecular Liquids 309 (2020) 113165 5

150 Table 2
DESs used for developing the density model.

BTPC Number of experiments DESs

ChCl
MTPB 310 DES 10 DES 14 DES 31 DES 41 DES 45
DEFA DES 11 DES 15 DES 32 DES 42
100 TBAB DES 12 DES 28 DES 33 DES 43
DES 13 DES 30 DES 40 DES 44
P(

50
acceptable difference between observed and predicted values. The
good prediction of the data is confirmed by the residue analysis
(Fig. 6), which in all cases was b0.02 (absolute value).
0
-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03
4.2. Viscosity model
e/ Ų)

The viscosity of DESs is of paramount importance for the industrial

Fig. 3. σ-profiles of the HBAs used in this work.
applications of these solvents. In order to develop an appropriate
model for the prediction of DES viscosity, 193 experimental points re-
trieved from the literature, Table 6, comprising 32 of the DESs listed in
a711 and a911 is due to the fact that the negative effect of S7 and S9 Table 1, were used.
(a 7 b 0, a 9 b 0) combined with the negative effect of temperature The stepwise regression algorithm was used for the regression anal-
(a11 b 0) gives a positive effect on density which means a relaxation ysis of the experimental viscosity data expressed by the MLR descrip-
of the DES system at the molecular level, i.e. reduction of the HBA- tors: σ-profile's surfaces and temperature. A summary of MLR
HBD interactions strength when increasing temperature. performances for the entire data set is given in Table 7.
In conclusion, the best expression of the MLR model for density is as In general, the value of R2 = 0.9874, and RMSE = 0.1047 indicate
follows: that the MLR model for viscosity shows very good performance as it
provides a precise fit of the experimental data set.
%ρ ¼ −4:37 þ 6152:85 S2 þ 380:74 S3 −591 S4 þ 169:87 S5 The relationship between the viscosity and the descriptors Sσ−profiles
−4
þ 1196:74 S
6 −3881:25

S7 þ204:35 S8 −289:62

S
9 −7 10 T
and temperature is satisfactorily multilinear.
þ 3023:52 S2 −S2 S4 −S4 −264:91 S2 −S2 S8 −S8 ð4Þ A detailed statistical analysis to identify the descriptors that have

 

 a significant influence on the description of the experimental viscos-
−0:62 S2 −S2 T−T þ 3297 S3 −S3 S5 −S5 ity values was performed. The purpose behind this analysis is to find



 
−6274:55 S3 −S3 S6 −S6 −0:55 S3 −S3 T−T a reduced expression of the MLR model equation similar to the ge-

 

 neric Eq. (3). To this end, ANOVA analysis was used to determine
−0:06 S4 −S4 T−T −1832:41 S5 −S5 S9 −S9 the model expression coefficients and the results are listed in

 
 
−0:09 S6 −S6 T−T þ 0:14 S7 −S7 T−T Tables 8 and 9.

  The ANOVA analysis showed that the sigma surface S1, …, S10 de-
þ0:24 S9 −S9 T−T
scriptors have no significant effect on the viscosity with a p-value of ac-
ceptance stated at 0.05. Therefore, the constants a1, …, a10 were set to 0
The experimental and calculated values are shown as a parity dia- in the model expression. On the other hand, the description of the vis-
gram in Fig. 5. cosity is significantly affected by the temperature and the interaction
It is remarkable that all density points lie on the diagonal line, with a between temperature and sigma-surfaces (except S3 and S8) at a level
very narrow dispersion. There was only one point that have a fairly of confidence N95% (Table 9).
As expected, the temperature was found to have a negative effect on
viscosity. The interaction's terms with a p-value lower than 5%, de-
scribes the combined effect of sigma surfaces (polarity) with tempera-
20
ture on viscosity. It was found that the negative effect of temperature

Fructose
15 DEA
EG
Gly Table 3
MA
MDEA Statistical parameters of the MLR model for density.
MEA
10 OA R2 0.9839
TEA
R2 adjusted 0.9827
P(

RMSE 0.0097

Table 4
ANOVA parameters of the MLR model for density.
0

-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03

Source Degrees of freedom Sum of squares Mean squares F-ratio

e/ Ų) Model 21 1.6610 0.0790 838.9552

Residual 288 0.0271 0.0001 Prob. N F
Total 309 1.6881 b0.0001
Fig. 4. σ-profiles of the HBDs used in this work.
6 T. Lemaoui et al. / Journal of Molecular Liquids 309 (2020) 113165

Table 5
Values of the coefficients of the density model.

Term Model coefficient Estimation Standard error t-Ratio Prob. N |t|

Constant a0 −4.37 1.9273 −2.27 0.0240

S2 a2 6152.85 2269.4830 2.71 0.0071
S3 a3 380.74 69.0285 5.52 b0.0001
S4 a4 −591 229.3685 −2.58 0.0105
S5 a5 169.87 64.8950 2.62 0.0093
S6 a6 1196.74 433.6576 2.76 0.0062
S7 a7 −3881.25 1416.3480 −2.74 0.0065
S8 a8 204.35 75.1621 2.72 0.0069
S9 a9 −289.62 62.2160 −4.66 b0.0001
T a11 −7 10−4 2.53 10−5 −28.73 b0.0001
(S2-S2 )*(S4-S4 ) a24 3023.52 1048.0160 2.88 0.0042
(S2-S2 )*(S8-S8 ) a28 −264.91 73.7267 −3.59 0.0004
(S2-S2 )*(T-T) a211 −0.62 0.0403 −15.41 b0.0001
(S3-S3 )*(S5-S5 ) a35 3297 264.8593 12.45 b0.0001
(S3-S3 )*(S6-S6 ) a36 −6274.55 1043.7130 −6.01 b0.0001
(S3-S3 )*(T-T) a311 −0.55 0.0658 −8.30 b0.0001
(S4-S4 )*(T-T) a411 −0.06 0.0107 −5.48 b0.0001
(S5-S5 )*(S9-S9 ) a59 −1823.41 168.1669 −10.84 b0.0001
(S6-S6 )*(T-T) a611 −0.09 0.0090 −10.09 b0.0001
(S7-S7 )*(T-T) a711 0.14 0.0166 8.41 b0.0001
(S9-S9 )*(T-T) a911 0.24 0.0356 6.60 b0.0001

T: Average temperature values; Si : mean values of the descriptors Si; S2 = 0.0017; S3 = 0.0060; S4 = 0.0177; S5 = 0.0760; S6 = 0.0303; S7 = 0.0138; S9 = 0.0088; T = 333.15.

Table 6
DESs used in the development of the viscosity model.

Number of experimental data points DESs

193 DES1 DES 8 DES 21 DES 29 DES 46

DES 2 DES 9 DES 22 DES 34 DES47
DES 3 DES 16 DES 23 DES 35 DES 48
DES 4 DES 17 DES 24 DES 36 DES 49
DES 5 DES 18 DES 25 DES 37
DES 6 DES 19 DES 26 DES 38
DES 7 DES 20 DES 27 DES 39

(HBD region: high polarity), and S4, S5 (nonpolar region: positive

charges).
The MLR model for viscosity is given as follows:

 
LogðηÞ ¼ −0:02  T−34:55 S1 −S1 T−T
Fig. 5. Experimental values of density versus predicted values using MLR model.
 
 
þ 3:86 S2 −S2 T−T −2:51 S4 −S4 T−T ð5Þ

 
 
−0:62 S5 −S5 T−T þ 0:60 S6 −S6 T−T

 
  
when combined with the positive effect of S2, S6, S7, S9 and S10 (the di- þ2:69 S7 −S7 T−T þ 3:09 S9 −S9 T−T

 
rect effect of these descriptors is found statistically insignificant) gives a þ18:64 S10 −S10 T−T
negative effect on the viscosity. This means that S2 (HBD region: me-
dium polarity), S6, S7 (nonpolar region: negative charges) and S9, S10
(HBA region), are the important regions when searching for reducing The parameters that are preceded by a negative sign cause a de-
viscosity with increasing temperature. On the contrary, if we want to in- crease in the DES's viscosity, while a positive sign gives a positive effect
crease viscosity with temperature, we have to increase the surface S1 (increases the viscosity-value).
Eq. (5) was determined by the regression analysis performed on 193
total experimental data points. In Fig. 7, the model results were repre-
sented as a parity diagram:
As can be observed, most of the viscosity values were correctly pre-
dicted by the model as the points are very close to the diagonal line with

Table 7
Statistical parameters of the MLR for viscosity.

R2 0.9874
R2 adjusted 0.9845
RMSE 0.1047
Fig. 6. Residuals vs. predicted values of density.
 T. Lemaoui et al. / Journal of Molecular Liquids 309 (2020) 113165 7

Table 8
Analysis of variance of the MLR for viscosity.

Source Degrees of freedom Sum of squares Mean squares F-ratio

Model 36 134.8731 37,464 341.3182

Residuals 156 1.7123 0,0109 Prob. N F
Total 192 136.5854 b0.0001*

Fig. 8. Residues vs predicted values of viscosity.

Table 9
Values of estimated viscosity model coefficients.

Term Model Estimation Standard t-Ratio Prob. N | depth statistical analysis in which the main descriptors exerting a signif-
coefficient error t| icant influence on the studied properties were considered.
T a11 −0.02 0.0003 −51.95 b0.0001 The modeling results showed that the proposed models for the in-
(S1-S1 )*(T-T) a111 −34.55 11.9135 −2.90 0.0043 vestigated DESs properties were able to predict the properties of the
(S2-S2 )*(T-T) a211 3.86 0.7787 4.96 b0.0001 DESs with an acceptable accuracy with R2 value ∈[0.9839, 0.9874].
(S4-S4 )*(T-T) a411 −2.51 0.6754 −3.72 0.0003 In conclusion, the developed MLR models can be considered as a re-
(S5-S5 )*(T-T) a511 −0.62 0.2041 −3.03 0.0028 liable tool for predicting important DESs properties and can be used for
(S6-S6 )*(T-T) a611 0.60 0.0869 6.95 b0.0001 their determination in the absence of experimental measurements,
(S7-S7 )*(T-T) a711 2.69 1.2743 2.11 0.0361
allowing a significant economy and time saving, and are useful for a
(S9-S9 )*(T-T) a911 3.09 1.1845 2.61 0.0100
thorough and optimal process design.
(S10-S10 )*(T-T a1011 18.64 2.1363 8.73 b0.0001
)

T: Average temperature values; Si: mean values of the descriptors Si; S1 = 1.42 10−5; S2 = CRediT authorship contribution statement
0.0011; S4 = 0.0450; S5 = 0.0987; S6 = 0.0831; S7 = 0.0193; S9 = 0.0194; S10 = 0.0007; T
= 326.98.
Tarek Lemaoui: Formal analysis, Investigation, Validation, Writing -
original draft. Nour El Houda Hammoudi: Formal analysis, Investiga-
no significant dispersion. The same result can be shown by the residues tion, Validation, Writing - original draft. Inas M. Alnashef: Resources,
analysis (Fig. 8), which are in most cases b0.2 (absolute value) and con- Supervision, Methodology, Writing - review & editing. Marco Balsamo:
firm the optimal prediction of the data. Methodology; Validation; Visualization; Writing - review & editing.
Alessandro Erto: Methodology; Validation; Visualization; Writing - re-
view & editing.
5. Conclusion

In this work, empirical models for the prediction of viscosity and Declaration of competing interest
density of DESs were developed. The models were derived by a
multilinear regression analysis, after defining the independent (i.e. the The authors declare that they have no known competing financial
Sσ-profile, Temperature and interaction's terms) and the dependent interests or personal relationships that could have appeared to influ-
(properties) variables. ence the work reported in this paper.
A set of data, including 310 experimental measurements of density
and 193 of viscosity for 49 different DESs were used for the develop- Acknowledgment
ment and validation of the model performance. The DESs were selected
so that their constituents have a wide range of chemical structure. The This work was supported by Université Ferhat ABBAS Sétif-1,
definition of the expression of the models was supported by an in- Algeria, and partially funded by Khalifa University, UAE, through grant
CIRA-2018-69.

References
[1] P.T. Anastas, M.M. Kirchhoff, Origins, current status, and future challenges of green
chemistry, Acc. Chem. Res. 35 (2002) 686–694.
[2] R. Gani, C. Jiménez-González, D.J.C. Constable, Method for selection of solvents for
promotion of organic reactions, Comput. Chem. Eng. 29 (2005) 1661–1676.
[3] M. Balsamo, A. Erto, A. Lancia, F. Montagnaro, G. Totarella, R. Turco, Post-combustion
CO2 capture: on the potentiality of amino acid ionic liquid as modifying agent of
mesoporous solids, Fuel 218 (2018) 155–161.
[4] M. Petkovic, K.R. Seddon, L.P.N. Rebelo, C. Silva Pereira, Ionic liquids: a pathway to
environmental acceptability, Chem. Soc. Rev. 40 (2011) 1383–1403.
[5] E.L. Smith, A.P. Abbott, K.S. Ryder, Deep Eutectic Solvents (DESs) and their applica-
tions, Chem. Rev. 114 (21) (2014) 11060–11082.
[6] A. Paiva, R. Craveiro, I. Aroso, M. Martins, R.L. Reis, A.R.C. Duarte, Natural deep eutec-
tic solvents – solvents for the 21st century, ACS Sustain. Chem. Eng. 2 (2014)
1063–1071.
[7] J.A.P. Coutinho, P.J. Carvalho, N.M.C. Oliveira, Predictive methods for the estimation
of thermophysical properties of ionic liquids, RSC Adv. 2 (2012) 7322–7346.
[8] Y. Zhao, Y. Huang, X. Zhang, S. Zhang, Quantitative prediction of viscosity of ionic liq-
uids using Sσ-profile molecular descriptors, Phys. Chem. Chem. Phys. 17 (2015)
3761–3767.
[9] Y. Benguerba, I.M. AlNashef, A. Erto, M. Balsamo, B. Ernst, A quantitative prediction
of the viscosity of amine based DESs using Ss-profile molecular descriptors, J. Mol.
Struct. 1184 (2019) 357–363.
[10] A. Klamt, F. Eckert, Fast solvent screening via quantum chemistry: COSMO-RS ap-
Fig. 7. Observed values of viscosity as a function of the predicted values. proach, AICHE J. 48 (2002) 369–385.
8 T. Lemaoui et al. / Journal of Molecular Liquids 309 (2020) 113165

[11] O. Alioui, Y. BENGUERBA, I.M. AlNashef, Investigation of the CO2-solubility in deep contribution and artificial intelligence prediction techniques, J. Mol. Liq. 256
eutectic solvents using COSMO-RS and molecular dynamics methods, J. Mol. Liq. (2018) 581–590.
307 (2020), 113005. https://doi.org/10.1016/j.molliq.2020.113005 In press. [18] F.S. Mjalli, G. Murshid, S. Al-Zakwani, A. Hayyan, Monoethanolamine-based deep
[12] M.A. Kareem, F.S. Mjalli, M.A. Hashim, I.M. AlNashef, Phosphonium-based ionic liq- eutectic solvents, their synthesis and characterization, Fluid Phase Equilib. 448
uids analogues and their physical properties, J. Chem. Eng. Data 55 (11) (2010) (2017) 30–40.
4632–4637. [19] S.T. Lin, S.I. Sandler, A priori phase equilibrium prediction from a segment contribu-
[13] A. Hayyan, F.S. Mjalli, I.M. AlNashef, T. Al-Wahaibi, Y.M. Al-Wahaibi, M.A. Hashim, tion solvation model, Ind. Eng. Chem. Res. 41 (2002) 899–913.
Fruit sugar-based deep eutectic solvents and their physical properties, Thermochim. [20] T.T. Aissaoui, I.M. AlNashef, Y. Benguerba, Dehydration of natural gas using choline
Acta 541 (2012) 70–75. chloride based deep eutectic solvents: COSMO-RS prediction, J Nat Gas Sci Eng 30
[14] F.S. Mjalli, J. Naser, Viscosity model for choline chloride based deep eutectic solvents, (2016) 571–577.
Asia Pac. J. Chem. Eng. 10 (2015) 273–281. [21] T. Aissaoui, Y. Benguerba, M.K. AlOmar, I.M. AlNashef, Computational investigation
[15] K. Shahbaz, F.S. Mjalli, M.A. Hashim, I.M. AlNashef, Prediction of deep eutectic sol- of the microstructural characteristics and physical properties of glycerol-based
vents densities at different temperatures, Thermochim.Acta. 515 (2011) 67–72. deep eutectic solvents, J. Mol. Model. 23 (10) (2017) 277–289.
[16] L. Bahadori, N.S. Abdul Manan, M.H. Chakrabarti, M.A. Hashim, F.S. Mjalli, I.M. [22] J. Palomar, J.S. Torrecilla, J. Lemus, V.R. Ferro, F. Rodriguez, A COSMO-RS based guide
AlNashef, M.A. Hussain, C.T.J. Low, The electrochemical behaviour of ferrocenein to analyze/quantify the polarity of ionic liquids and their mixtures with organic co
deep eutectic solvents based on quaternary ammonium and phosphoniumsalts, solvents, Phys. Chem. Chem. Phys. 12 (2010) 3473–3482.
Phys. Chem. Chem. Phys. 15 (2013) 1707–1714.
[17] I. Adeyemi, M.R.M. Abu-Zahra, I.M. AlNashef, Physicochemical properties of
alkanolamine-choline chloride deep eutectic solvents: measurements, group