Ginex 2019
Ginex 2019
Ginex 2019
The pharmacophore concept is a fundamental cornerstone in drug discovery, playing a critical role in de-
termining the success of in silico techniques, such as virtual screening and 3D-QSAR studies. The reliability
of these approaches is influenced by the quality of the physicochemical descriptors used to characterize the
chemical entities. In this context, a pivotal role is exerted by lipophilicity, which is a major contribution to
host–guest interaction and ligand binding affinity. Several approaches have been undertaken to account
for the descriptive and predictive capabilities of lipophilicity in 3D-QSAR modeling. Recent efforts encode
the use of quantum mechanical-based descriptors derived from continuum solvation models, which open
novel avenues for gaining insight into structure–activity relationships studies.
First draft submitted: 30 August 2018; Accepted for publication: 4 February 2019; Published online:
25 February 2019
10.4155/fmc-2018-0435
C 2019 Newlands Press Future Virol. (Epub ahead of print) ISSN 1746-0794
Review Ginex, Vazquez, Gilbert, Herrero & Luque
proach, the nature of the molecular features relevant for similarity assessment, and the definition of the similarity
function [12]. A sensitive and effective estimation of molecular similarity is a fundamental pre-requisite for the
identification of potential leads starting from a chemical reference, which represents the paradigm of VS.
Another successful application of the pharmacophore concept is linked to 3D-quantitative structure–activity
relationships (3D-QSAR) [13], such as CoMFA [14], CoMSIA [15] and GRID/GOLPE [16]. These methods permit
to identify a pharmacophore from the relationships between the biological activities of a set of aligned molecules
and the projection of selected physicochemical descriptors into the surrounding space, leading to the disclosure of
regions favorable or not to the bioactivity of compounds. 3D-QSAR approaches are also used to model ADME(T)
properties in the attempt to predict whether a molecular candidate would be able to achieve its biological target [17].
Optimization of both ligand potency and ADME(T) profile is absolutely required to translate promising molecular
candidates to successful low-dose therapeutics. However, the success of this operation is not trivial, since the final
result depends on factors such as the quality of the input data, as well as the adequacy and level of description of the
physicochemical parameters used in the analysis. In fact, Gleeson and collaborators [18] have observed the existence
of a diametrically opposed relationship between descriptors that efficaciously model drug potency and ADME(T)
properties, making more challenging the drug discovery process.
G o/w
tr
= - 2.303 RT logP (Equation 1)
Lipophilicity reflects the complex interplay between the intermolecular forces that dictate the differential solvation
in the aqueous and organic phases. Accordingly, it can be factorized in terms of selected physicochemical properties
of the compound that may be relevant for the preferential solvation in aqueous and nonaqueous solvents, as shown
in Equation 2 [21], and references therein.
logP = vV - + I + IE (Equation 2)
where v is a constant, V is the molar volume, which encompasses the ability of the solute to elicit nonpolar
interactions, is related to the polarity of the compound, and finally I and IE accounts for the solute capacity
to form ionic interactions, which favor partitioning into the aqueous phase, and for the contribution due to
intramolecular effects, respectively.
Let us note that lipophilicity and hydrophobicity, which are often used as equivalent concepts, are not strictly
synonymous, the latter being in fact one of the contributions to molecular lipophilicity [22]. Thus, while hydropho-
bicity can be defined as the tendency of nonpolar groups of a molecule to aggregate in order to minimize the
unfavorable exposition to the surrounding polar (water) solvent, lipophilicity is a measure of the affinity of the
molecule for the nonpolar solvent in a biphasic system constituted by a polar and a nonpolar solvent.
Lipophilicity affects a number of pharmacokinetic parameters (Figure 1). Low lipophilicity is responsible of high
aqueous solubility, which is a key factor for drug-likeness, but an excessively low lipophilicity could compromise
the ability of the drug to achieve the biological target. On the opposite site, highly soluble compounds possess poor
permeability through biological membranes, limiting absorption along the gastrointestinal tract or the transport
across the blood–brain barrier. Therefore, optimal requirements for efficient solubility and permeability properties
Drug disposition
- + - -
Lipophilicity
Figure 1. Schematic representation of the central role of lipophilicity in drug potency and pharmacokinetics profile.
Direct (+) and inverse (-) correlation of lipophilicity with each of the main steps of ADME process are also highlighted.
are inevitably enclosed in a very narrow range of lipophilicity. Another key aspect for drug-likeness is bioavailability,
which is inversely correlated to low first-pass clearance. Once again, lipophilicity is crucial since high lipophilicity
is associated with high clearance and low metabolic stability. Overall, a careful handling of lipophilicity is required
to optimize compound availability at the biological target.
On the other hand, lipophilicity has rarely been used as the primary descriptor in ligand–receptor recognition.
Indeed, following the IUPAC recommendation for the definition of a pharmacophore, it is defined as “the ensemble
of steric and electronic features that is necessary to ensure the optimal supramolecular interactions with a specific biological
target structure” [8]. This definition hides the key role played by (de)solvation in the recognition and binding of
a drug-like compound to its macromolecular target [23], especially keeping in mind that the maximal achievable
affinity that can be attained for target binding sites is largely influenced by nonpolar desolvation [24]. This is
consistent with the concept that favorable drug binding is largely driven not only by the global lipophilicity of a
compound, but more importantly by the spatial distribution of polar and apolar regions along the chemical skeleton.
Thus, while apolar regions determine the binding affinity with complementary lipophilic regions of the binding
site, polar interactions would provide ‘anchor points’ contributing to ligand specificity and/or directionality in the
binding pocket, as well as to modulate binding kinetics of the ligand [25–30].
Taken together, these data suggest that a concomitant optimization of both pharmacokinetic profile and drug
potency have to be done to obtain successful drug products. This is encoded in the concept of lipophilicity efficiency
(LipE), which provides a metric that normalizes the potency (generally measured as Ki or IC50 ) of the ligand against
a protein target for the lipophilicity of the compound [31–33]. This is achieved by substracting the logP (or the
distribution coefficient for ionizable molecules, logD) from the negative logarithm of the potency (Equation 3).
Lipophilicity efficiency can be useful to provide guidelines to study the simultaneous effects exerted by structural
changes on potency and lipophilicity, which is central for drug design and lead optimization programs, thus giving
support to the formulation of the ‘lipophilic pharmacophore’ concept.
a i f i + j=1 b j Fj
n m
logP = i=1 (Equation 4)
where logP is the sum of the weighted (ai ) contribution of each fragment/atom (fi ) and a correction factor (bj Fj ).
Fragmental methods are illustrated by the work of Leo, Hansch and Elkins [43] as well as Nys and Rekker [44]. The
former relies on the concept of substituent constant, which encodes the lipophilicity contribution of a chemical
group or atom when it replaces a hydrogen atom in a reference compound, and the theoretical estimation of logPo/w
follows an additivity scheme, named cLOGP. This method permits to extrapolate the partition coefficients starting
from a list of experimentally fitted fragmental contributions to lipophilicity. An arbitrary set of interfragmental
rules was then used to compile a database library of fragment-weighted lipophilicity contributions. On the other
hand, Nys and Rekker [44] introduced the concept of hydrophobic fragmental constant (f), which represents the
lipophilicity contribution of a constituent part of a structure to the total lipophilicity of a given compound.
Fragments range from atoms to heterocyclic rings, so that functional groups with direct contribution to resonance
interactions were left intact, and are differentitated upon linkage to aliphatic and aromatic structures. The differences
between experimental logP and the additive value estimated from the f approach was accounted for by correction
rules, reflecting factors such as the presence of vicinal electronegative centers in the chemical structure, aromatic
condensation, cross-conjugation or hydrogen-bonding [45].
An example of atom-based partitioning strategy was undertaken by Ghose and Crippen, who developed a
procedure that combines lipophilicity contributions at an atomic level leading to the ALOGP method. This
method encompassed a list of 120 atom types for carbon, hydrogen, oxygen, nitrogen, sulfur and halogens [46–48].
An alternative strategy is the XLOGP method [49], which is based on the summation of atomic contributions
derived from experimental lipophilicity data of 1831 organic molecules, and includes correction factors for some
intramolecular interactions.
In the last decades, the evolution of computer performances enabled the development of whole molecule-based
strategies to predict the lipophilicity by taking into account the 3D-structure of compounds, and thus the effect
of molecular conformation. Among all the available techniques, the molecular lipophilicity potential (MLP) [50]
offers an empirical quantitative 3D-description of the lipophilicity potential from all the molecular fragments on
the surrounding space of a compound. The MLP approach is then intended to model the lipophilic interactions
between ligand and receptor as noted in Equation 5,
where Fi is the lipophilic fragmental contribution and f(dik ) is a distance function which depends on the
separation between a given fragment (i) and any point on the molecular surface or volume (k).
Molecular fields derived from the MLP potential have found a wide range of pharmaceutical applications,
including the prediction of skin permeation and distribution of new chemical entities [51], modeling of peptides
and proteins [52,53], and structure–activity relationships studies [54].
The Hydrophobic INTeraction (HINT) method represents an alternative, promising strategy for the study of
lipophilicity in biomolecular interactions [55,56]. This method exploits a scale of hydrophobic fragments constants
at the atomic level by means of an adaptation of the CLOGP method, which are then used to evaluate a pairwise
interaction energy term (bij ) between atoms i and j in the interacting partners according to Equation 6,
where ai and Si are respectively the hydrophobic constant and the accessible surface area of the atom i, Tij is a
logic function describing the character of interacting pairs (attraction or repulsion), and Rij and rij denote functions
of the distance between atoms i and j, the former following an exponential form and the latter a Lennard–Jones
implementation.
Equation 5 encodes the formalism of the ‘natural’ HINT force-field, which has been used to explore a variety of
applications in ligand–protein and protein–protein interactions [57–61].
Other approaches have relied on molecular properties derived from quantum mechanical treatments of molecules.
An early attempt is the work by Roger and Cammarata [62,63], who related the logP of aromatic compounds with
the charge density of both π and σ electron frameworks and the induced polarization. In a distinct approach, the
BLOGP method relied on semiempirical AM1 calculations to derive geometrical and quantum chemical descriptors
for the prediction of logP [64,65]. In a similar approach, Clark and coworkers performed AM1 and PM3 calculations
to derive a series of descriptors, including electrostatic potentials, total dipole moments, mean polarizabilities,
surfaces, volumes and charges, which were used in the prediction of partition coefficients [66,67].
These efforts can also be exemplified with the concept of heuristic MLP [68,69]. In this approach, the
lipophilic/hydrophilic features of a compound are determined from the analysis of the electrostatic potential
computed at the molecular surface. To this end, a dimensionless distance-dependent screening function is used to
compare the local electron density at the surface of a given atom with the electrostatic potential generated on the
rest of atoms. The screening function, which was derived from statistical mechanical treatment of polar solvent
molecules as dipoles, accounts for the influence exerted by the atomic descriptors of the electrostatic potential from
surrounding atoms. Ultimately, such a comparison leads to the definition of an atomic lipophilicity index, which
can adopt positive or negative values, reflecting the lipophilic and hydrophilic nature, respectively, of such an atom.
Finally, a distinct approximation comes from the usage of solute–solvent correlation functions derived by using the
reference interaction site model (RISM) as descriptors for QSAR studies. By using a classical statistical mechanics-
based solvent model combined with machine learning, 1D solute–solvent correlation functions were used to predict
Caco-2 cell permeabilities [70]. As an extension of this approach, Gussregen et al. proposed the Comparative Analysis
of 3D-RISM Maps (CARMa) methodology [71]. In this computational strategy, the classical electrostatic and steric
fields generally used in CoMFA are replaced by solute–solvent distribution functions determined from 3D-RISM
computations, which are subsequently treated as descriptors to perform QSAR analysis. The method was validated
using a set of serine protease inhibitors as a test system.
Even though CARMa uses a statistical mechanics solvent model, the electrostatic and steric effects implemented
in CoMFA cannot be directly captured. This issue has been recently addressed by solving 3D-RISM equations for a
solvent comprising CoMFA probes in aqueous solution, this extension being referred to as CARMa (electrolyte) [72].
The analysis performed for six protein–ligand systems reveals a small but consistent increase in prediction accuracy
compared with CoMFA.
Fragmental lipophilicity model from the Miertus–Scrocco–Tomasi method: the Hyphar approach
The Miertus–Scrocco–Tomasi (MST) solvation model has been used to develop 3D-distribution patterns of
lipophilicity, which in turn have been exploited in predicting molecular overlays and 3D-QSAR studies [82,83]. The
MST model is a parametrized version of the polarizable continuum model developed by Tomasi and coworkers [84,85]
at both semiempirical, Hartree–Fock and B3LYP levels [86–89] (for a review see [90]). From the solvation free energies
in water and n-octanol, one can derive the n-octanol/water partition coefficient (Equation 1), which is a property
of the whole molecule. Nevertheless, by decomposing the solvation free energy into atomic contributions, one can
obtain the 3D profile of lipophilicity from the corresponding atomic contributions to the logP. For a molecule (M)
containing N atoms, this is achieved by decomposing the logP (or the corresponding transfer free energy, G tr,M )
o/w
into electrostatic (logPele,i ), cavitation (logPcav,i ) and van der Waals (logPvw,i ) components, which can be derived
from the polar ( G ele,i ) and nonpolar ( G cav,i’G vW,i ) contributions to the solvation free energy (Equations 7
o/w o/w o/w
& 8).
Partitioning of the electrostatic term into atomic contributions can be made resorting to a perturbation ap-
proximation of the coupling between the solute charge distribution and the solvent reaction field [91], leading to
Equation 9,
1 qk w q1o
o/w o k L
logPele,i = ¨ =1 - =1 ¤o (Equation 9)
2 k ki w
rk - r I Ii o
r1 - r
where o is the solute wave function in the gas phase, and K and L stand for the total number of reaction field
w o
charges in water ( qk ) and n-octanol ( qi ), located at positions rkw and ri o .
The atomic decomposition of the cavitation and van der Waals terms takes advantage of the linear dependence
with the solvent-exposed surface of the atoms in the molecule (Equations 10 & 11).
Si
= i = 1
N
o/w
logPcav,i G o/w
P,i (Equation 10)
ST
= i = 1 Si
o/w N o
/w
logPvW,i (Equation 11)
where G p,i =G P,i - G P,i , G P,i being the cavitation free energy of atom i, = - , with i
o/w W o o/w w o
being the atomic surface tension and Si denotes the contribution of atom i to the total molecular surface (ST ).
In contrast to the COSMO-RS-based approaches, which rely on the concept of σ-profile (see above), the MST-
derived applications use the atomic contributions to the thermodynamic components of the differential solvation
free energy in water and n-octanol, which are encoded under the partition coefficient between these two solvents.
Accordingly, they take into account the effect of specific chemical features of the molecule, such as the existence of
specific tautomers or conformational species, or the formation of specific intramolecular interactions (i.e., hydrogen
bond), in the computation of the 3D-distribution pattern of molecular lipophilicity.
These patterns have been exploited to predict the chemical similarity between compounds [92]. By using the MST-
o/w o/w
based hydrophobic descriptors logPeles,i and logPcav,i , a computational procedure has been proposed to identify
the molecular overlay that maximizes the lipophilic similarity. To this end, molecular similarity was achieved
by comparing the hydrophobic fields generated by the molecules, which were prealigned following multipole
expansions of the atomic lipophilic contributions. On the other hand, simple descriptors of the hydrogen-bond
(HB) donor/acceptor character of atoms were used to complement the information about the chemical nature
of polar atoms in a molecule (briefly, the current implementation assigns an arbitrary value of +1 to hydrogen
atoms in HB donors, and -1 to N and O atoms that may act as acceptors). This choice obeys to the fact that
the polar nature of hydrophilic groups cannot distinguish the HB donor/acceptor character, as this information is
o/w
not implicitly encoded by the logPele,i term. Hydrophobic and HB properties are then projected into a 3D grid
using the exponential function (Equation 12) implemented in CoMSiA [15], and then compared by means of the
Tanimoto coefficient.
2
p q = i=1 w i e
N -–riq
(Equation 12)
The method was implemented in PharmScreen software [83,93] and was successfully used to evaluate the molecular
overlay for a collection of 121 molecular systems compiled by AstraZeneca, denoted as the AstraZeneca Overlays
Validation Test Set [94]. This set contains molecular overlays experimentally characterized for 119 targets, which were
grouped in four categories according to the expected difficulty in predicting the experimental overlay: easy, moderate,
hard and unfeasible. The results pointed out that correct overlays were predicted for 94% (easy), 79% (moderate) and
54% (hard) of the cases. Moreover, the overall performance obtained from classical electrostatic/steric descriptors
and from Hyphar ones was fairly similar for easy and moderate subsets, but the accuracy obtained with Hyphar
for the subset of hard cases exceeded the performance obtained with electrostatic/steric properties. Finally, it was
found that the similar performance of Hyphar and electrostatic/steric descriptors does not imply that they lead
to identical overlays. Rather, the analysis of the predicted poses revealed that the degree of identity in molecular
overlays was reduced with the increase in the difficulty of the target. Overall, these findings point out that Hyphar
descriptors may be a valuable alternative for molecule superposition and VS of chemical libraries, especially for
targets that may be challenging for predictive molecular similarity techniques.
On the other hand, the atom-centered MST-derived hydrophobic contributions have also been used as physic-
ochemical descriptors to derive 3D-QSAR models using PharmQSAR [82]. MST/IEFPCM calculations were
performed for five sets of compounds, including dopamine D2/D4 receptor antagonists, antifungal chromanones,
glycogen synthase kinase-3 inhibitors, cruzain inhibitors and thermolysin inhibitors. The compounds in these
sets covered a wide range of variance in selected physicochemical properties (molecular weight, hydrogen-bond
donor/acceptor, clogP and number of rotatable bonds). The 3D-QSAR models obtained with the hydrophobic
pharmacophore (HyPhar) were found to have a predictive accuracy comparable to standard CoMFA and CoMSiA
techniques. Moreover, Hyphar descriptors were also valuable to discriminate the selectivity of compounds acting
as inhibitors of thrombin, trypsin and factor Xa [83].
Overall, these findings support the usefulness of the MST-derived lipophilic descriptors as a valuable alternative
to electrostatic/steric properties to carry out VS of chemical libraries for molecular similarity, as well as to derive 3D-
lipophilic pharmacophores, thus providing valuable complementary information to gain insight into the molecular
determinants of bioactivity.
Table 1. Statistical parameters of the 3D-QSAR HyPhar models obtained from Miertus–Scrocco–Tomasi/B3LYP and
Miertus–Scrocco–Tomasi/RM1 calculations for the four sets of compounds.†
System Training set Test set Nc Field (%)
r2 q2 S Spress r2 S Elec Nonelec
D2
MST/B3LYP 0.94 0.77 0.31 0.60 0.78 0.57 3 68.6 31.4
MST/RM1 0.93 0.74 0.28 0.65 0.71 0.63 3 70.9 29.1
Chromanones
MST/B3LYP 0.77 0.51 0.49 0.29 0.81 0.20 3 34.3 65.7
MST/RM1 0.76 0.42 0.51 0.32 0.66 0.82 3 42.1 57.9
GSK3
MST/B3LYP 0.91 0.80 0.12 0.19 0.79 0.21 3 54.5 45.5
MST/RM1 0.91 0.82 0.30 0.18 0.79 0.21 5 64.7 35.3
Cruzain
MST/B3LYP 0.81 0.50 0.31 0.51 0.69 0.47 2 53.0 47.0
MST/RM1 0.91 0.65 0.31 0.44 0.70 0.46 3 58.4 41.6
† See [91] for a proper description of the molecular sets. Nc denotes the number of PLS components in the best 3D-QSAR model, and the terms Elec and Nonelec stand for the fraction
(in percentage) of electrostatic (logPele,i ) and nonelectrostatic (logPcav,i ) hydrophobic contributions to the final model.
MST: Miertus–Scrocco–Tomasi.
for the THER set, where Gasteiger–Marsili charges were used. For the sake of comparison, an additional model,
denoted CoMFA (RM1), which exploits RM1 electrostatic-potential fitted partial charges in conjunction with an
steric field obtained from the Lennard–Jones potential with a positively charged C.3 atom probe, was also examined.
This model, therefore, is intended to explore the efficiency of RM1-based partial charges in defining electrostatic
features of molecules at the atomic level.
Table 2 shows the statistical parameters of the 3D-QSAR models. In general, similar performances were obtained
for the different 3D-QSAR models determined for molecules in the training test included in a given system, as
noted in the large resemblance between the statistical values of the regression (r2 ) and cross-validation (q2 ) models.
The same trend can be observed for the test set compounds, although a small improvement was found for CoMFA
(RM1) and Hyphar models in GPB and THERM systems compared with reference CoMFA/CoMSiA models. In
addition, a higher level of accuracy was also achieved by the models derived from RM1 calculations since the number
of outliers in the test set was lower than in classical CoMFA/CoMSIA (Supplementary Material, Supplementary
Table 4). On the other hand, both BZR and COX2 were confirmed to be challenging systems for QSAR modeling,
as already noted by Sutherland and coworkers [95]. For instance, in case of COX2, part of the reason for the poor
predictive behavior may probably be ascribed to the fact that training and test set cover different ranges of in the
property space.
The predictive performance of the models was also examined by analyzing their capacity to discriminate between
active and inactive compounds. To this end, for each molecular system the compounds in the test set were
ranked according to their experimental potency: ‘active/positive’ (P) and ‘inactive/negative’ (N) were categorized
by applying a threshold value of 6.0 (in pIC50 /pKi units). Then, test set compounds with a predicted pIC50 /pKi
value larger than the threshold value were considered ‘actives/positives’ (TP), whereas compounds with a predicted
pIC50 /pKi value lower than the threshold were considered ‘inactives/negatives’ (TN). For each molecular system,
the number of P, N, TP and TN compounds, as well as false positives (FP) and false negatives (FN) are compiled
in Supplementary Material (Supplementary Table 5). In turn, these values were used to identify correctly negative
(specificity or TNR; in green in Figure 2) and positive (sensitivity or TPR; in blue in Figure 2) compounds, and to
reduce the false negative rate (‘fall-out’ or FPR; in red in Figure 2) by applying Equations. 13-15.
TN TN
Specificity(TNR) = = (Equation 13)
N (TN+FP)
Table 2. Statistical parameters obtained for CoMFA and CoMSiA models reported with the results determined by using
COMFA (RM1) and Hyphar models in this study for the eight molecular systems (ACE, AChE, BZR, COX2, DHFR, GPB, THERM
and THR).†
System Training set Test set Nc* Field (%)
r2 q2 S Spress r2 S Ele N-Ele HB
ACE‡
CoMFA 0.80 0.68 1.04 – 0.49/0.55 1.54/1.47 3 – – –
CoMSiA 0.76 0.65 1.15 – 0.52/0.58 1.48/1.41 3 – – –
CoMFA (RM1) 0.82 0.67 0.42 1.37 0.54/0.61 1.45/1.32 3 29.4 70.6 –
Hyphar 0.75 0.64 0.51 1.43 0.42/0.62 1.62/1.35 2 28.8 53.5 17.7
AChE
CoMFA 0.88 0.52 0.41 – 0.47/0.56 0.95/0.87 5 – – –
CoMSiA 0.86 0.48 0.45 – 0.44/0.60 0.98/0.81 6 – – –
CoMFA (RM1) 0.90 0.54 0.32 0.85 0.35/0.52 1.07/0.86 6 20.0 80.0 –
Hyphar 0.76 0.45 0.50 0.92 0.65 0.78 4 64.1 18.7 17.2
BZR
CoMFA 0.61 0.32 0.41 – 0.00/0.18 0.97/0.81 3 – – –
CoMSiA 0.62 0.41 0.41 – 0.08/0.30 0.93/0.75 3 – – –
CoMFA (RM1) 0.60 0.36 0.64 0.53 0.21/0.21 0.81/0.80 3 30.5 69.5 –
Hyphar 0.67 0.37 0.58 0.54 0.00/0.02 0.91/0.86 6 48.8 16.7 34.5
COX2
CoMFA 0.70 0.49 0.56 – 0.29/0.37 1.24/1.09 5 – – –
CoMSIA 0.69 0.43 0.56 – 0.03/0.22 1.44/1.20 6 – – –
CoMFA (RM1) 0.74 0.51 0.52 0.72 0.19/0.34 1.20/1.07 5 28.6 71.4 –
Hyphar 0.60 0.52 0.63 0.71 0.26/0.40 1.15/0.99 3 85.4 4.3 10.3
DHFR
CoMFA 0.79 0.65 0.59 – 0.59/0.70 0.89/0.73 5 – – –
CoMSiA 0.76 0.63 0.62 – 0.52/0.63 0.96/0.81 5 – – –
RM1 CoMFA 0.81 0.67 0.44 0.73 0.42/0.55 1.04/0.91 4 17.7 82.3 –
Hyphar 0.72 0.63 0.53 0.78 0.53/0.56 0.94/0.89 5 36.2 38.8 25.0
GPB
CoMFA 0.84 0.42 0.43 – 0.42/0.37 0.94/0.70 4 – – –
CoMSiA 0.78 0.43 0.50 – 0.46/0.34 0.90/0.82 4 – – –
CoMFA (RM1) 0.88 0.43 0.36 0.85 0.51 0.89 4 24.4 75.6 –
Hyphar 0.83 0.54 0.42 0.75 0.71 0.68 3 52.0 2.7 45.3
THERM
CoMFA 0.94 0.51 0.55 1.54 0.60 1.26 7 – – –
CoMSiA 0.85 0.54 0.73 – 0.36/0.46 1.87/1.60 6 – – –
CoMFA (RM1) 0.90 0.46 0.33 1.57 0.51/0.66 1.39/1.18 5 25.5 74.5 –
Hyphar 0.84 0.49 0.41 1.51 0.67 1.13 4 37.9 25.5 36.6
THR¶
CoMFA 0.86 0.59 0.36 – 0.54/0.73 1.59/0.56 4 – – –
CoMSiA 0.88 0.62 0.34 – 0.55/0.62 0.76/0.66 5 – – –
CoMFA (RM1) 0.89 0.59 0.33 0.64 0.45/0.58 0.86/0.82 5 16.0 84.0 –
Hyphar 0.87 0.64 0.37 0.59 0.53/0.56 0.79/0.74 4 37.5 41.7 20.8
† For 2
test sets compounds, statistical parameters (r and S) with (left) and without (right) outliers (i.e., compounds with residuals higher than 2.5-fold the standard deviation) are indicated.
The number of outliers for each system is reported in Supplementary Material (Supplementary Table 4).
‡ mol0088 (original file name mol 17) was excluded because it contains iodine atom.
¶ mol0088 (original file name 82) was excluded due to problems with the input geometry.
ACE: 113 angiotensin converting enzyme; AChE : 111acetylcholinesterase;BZR : 147 ligands for benzodiazepine receptors; COX-2: 282 cyclooxygenase-2; DHFR : 361 dihydrofolatereductase;
GPB: 66 glycogen phosphorylase b; THER: 74 thermolysin ; THR: 87 thrombine.
1.0
0.7
0.5
0.2
0.0
ACE
AchE
BZR
COX2
DHFR
GPB
THERM
THR
Figure 2. Specificity (in green), sensitivity (in blue) and fall-out (in red) for RM1 CoMFA (left) and H2 (right) models
the test sets of the eight systems.
TP TP
Sensitivity (TPR) = = (Equation 14)
P (TP+FN)
FP FP
Fall - out (FPR) = = = 1 - TNR (Equation 15)
N (FP+TN)
These parameters, which can vary from 0 to 1, can be considered a measure of the predictive performance of
the model. According to this classification, a model can be considered good if it has high specificity/sensitivity
and low fall-out values. Nevertheless, this analysis requires a balanced partition of active and inactive compounds
in the set of compounds, a requirement that is not fulfilled in the case of BZR and GPB systems, since only one
inactive and one active compound are present in these two sets, respectively. Accordingly, the results obtained for
BZR and GPB should be excluded from the analysis. For the rest of molecular systems, both CoMFA (RM1) and
Hyphar models exhibit generally similar trends (Figure 2). The Hyphar model has a slightly better performance
in sensitivity/specificity and fall-out values for AchE, THERM and THR systems, whereas the opposite trend is
found for CoMFA (RM1) in ACE and COX2.
Finally, the ability of CoMFA (RM1) and Hyphar models to rank the compounds according to their potency was
also examined (Figure 3). To this end, the Spearman (Rs) coefficient for the first (Q1; in green), second (Q2; in blue)
and third (Q3; in red) quartiles, which would encompass molecules with highest, medium and low activity/affinity,
were determined for the test set compounds in each system. Although there is a notable resemblance in the general
trends obtained for CoMFA (RM1) and Hyphar models, slightly better performances (higher Rs values) are observed
for Hyphar models, especially for compounds of higher activity/affinity (Q1/Q2), whereas the differences are less
pronounced for compounds in Q3, probably due to the larger noise associated with the biological activity low
active compounds.
Overall, the results obained for the benchmark systems reveal that the Hyphar descriptors yield 3D-QSAR
models with an overall performance that compares with the results obtained using standard CoMFA/CoMSiA.
Hyphar models also seem to be more effective in locating (high sensibility) and ranking (high Rs) true positives,
especially in regions of high and medium activity/affinity.
1.0
0.8
0.5
0.3
0.0
ACE
AchE
BZR
COX2
DHFR
GPB
THERM
THR
-0.3
-0.5
-0.8
-1.0
Figure 3. Spearman (Rs) coefficients for the first (Q1; in green), the second (Q2; in blue) and the third (Q3; in red)
quartiles for RM1 CoMFA (left) and H2 (right) models.
Overall, the analysis of structure–activity relationships in terms of the lipophilic/hydrophilic balance may provide
a useful signature to complement studies performed with electrostatic/steric properties. In this sense, the QM MST-
based hydrophobic descriptors are valuable in predicting molecular overlays and elucidating molecular similarity
patterns. The higher descriptive quality of these descriptors could thus offer interesting clues in searching for novel
bioactive compounds, especially for challenging targets.
Executive summary
• All biological and biochemical processes are driven by the general concept of host–guest complementarity.
Accordingly, an essential but effective description of the ‘guest’ is required for a successful prediction of ‘host’
recognition.
• The pharmacophore concept is a fundamental cornerstone in drug discovery, as it accounts for the common
interaction features of a group of compounds toward their target structure, playing a critical role in determining
the success of in silico techniques.
• Optimized descriptors able to model both pharmacokinetics and pharmacodynamics properties in drug design are
not easily achievable, and the use of suboptimal physicochemical parameters may be a more effective strategy.
• Besides the relevance in predicting ADME(T) properties, lipophilicity exerts a pivotal role in accounting for the
maximal achievable affinity that can be attained between ligand and receptor.
• The usage of lipophilicity descriptors may offer novel opportunities to disclose the underlying relationships
between chemical features and biological activity. In this context, the availability of refined version of QM-based
continuum solvation models may be an effective strategy for deriving novel descriptors well suited for drug
design.
• In 3D-QSAR studies, the Miertus–Scrocco–Tomasi-derived Hyphar descriptors have been shown to provide models
for structure–activity relationships with a predictive accuracy comparable to CoMFA/CoMSiA techniques based on
electrostatic/steric parameters.
• The Hyphar descriptors are also a valuable alternative for molecule superposition and virtual screening of
chemical libraries, especially for targets that may be challenging for predictive molecular similarity techniques.
• The availability of ‘polar’ and ‘non-polar’ fractional descriptors obtained from Miertus–Scrocco–Tomasi-based
continuum solvation models may be valuable to explore the molecular determinants of bioactivity, providing
complementary interpretations to classical descriptors in the rational design of novel compounds.
Acknowledgments
The authors acknowledge J Muñoz-Muriedas (GSK group, Stevenage, UK) for valuable comments and suggestions.
References
Papers of special note have been highlighted as: • of interest; •• of considerable interest
1. Gohlke H, Klebe G. Approaches to the description and prediction of the binding affinity of small-molecule ligands to macromolecular
receptors. Angew. Chem. Int. Ed. Engl. 41, 2644–2676 (2002).
2. Khedkar SA, Malde AK, Coutinho EC, Srivastava S. Pharmacophore modeling in drug discovery and development: an overview. Med.
Chem. 3, 187–197 (2007).
3. Güner OF, Bowen JP. Setting the record straight: the origin of the pharmacophore concept. J. Chem. Inf. Model. 54, 1269–1283 (2014).
4. Schueler FW. Chemobiodynamics and Drug Design. McGrawHill, NY, USA, (1960).
5. Beckett AH, Harper NJ, Clitherow JW. The impact of stereoisomerism in muscarinic activity. J. Pharm. Pharmacol. 15, 362–371 (1963).
6. Kier LB. Receptor mapping using molecular orbital theory. In: Fundamental Concepts in Drug-Receptor Interactions. Academic Press, NY,
USA, 15–46 (1970).
7. Gund P, Wipke WT, Langridge R. Computer searching for molecular structure file for pharmacophoric patterns. In: Computers in
Chemical Research and Education (Volume 3). Hadzi D, Zupan J (Eds)., Elsevier Scientific, Amsterdam, The Netherlands, 5–33 (1973).
8. Wermuth CG, Ganellin CR, Lindberg P, Mitscher LA. Glossary of terms used in medicinal chemistry (IUPAC recommendations 1998).
Pure Appl. Chem. 70, 1129–1143 (1998).
9. Bender A, Glen RC. Molecular similarity: a key technique in molecular informatics. Org. Biomol. Chem. 2, 3204–3218 (2004).
10. Wolber G, Seidel T, Bendix F, Langer T. Molecule-pharmacophore superpositioning and pattern matching in computational drug
design. Drug Discov. Today 13, 23–29 (2008).
11. Kaserer T, Beck KR, Akram M, Odermatt A, Schuster D. Pharmacophore models and pharmacophore-based virtual screening: concepts
and applications exemplified on hydroxysteroid dehydrogensases. Molecules 20, 22799–22832 (2015).
12. Maggiora G, Vogt M, Stumpfe D, Bajorath J. Molecular similarity in medicinal chemistry. J. Med. Chem. 57, 3186–3204 (2013).
13. Verma J, Khedkar VM, Coutinho EC. 3D-QSAR in drug design - a review. Curr. Top. Med. Chem. 10, 95–115 (2010).
14. Cramer RD III, Patterson DE, Bunce JD. Comparative molecular field analysis (CoMFA). 1. Effect of shape on binding of steroids to
carrier proteins. J. Am. Chem. Soc. 110, 5959–5967 (1988).
15. Klebe G, Abraham U, Mietzner T. Molecular similarity indices in a comparative analysis (CoMSIA) of drug molecules to correlate and
predict their biological activity. J. Med. Chem. 37, 4130–4146 (1994).
16. Nilsson J, Wikström H, Smilde A, Glase S, Pugsley T, Cruciani G, Pastor M, Clementi S. GRI D/GOLPE 3D quantitative
structure-activity relationship study on a set of benzamides and naphthamides, with affinity for the dopamine D3 receptor subtype. J.
Med. Chem. 40, 833–40 (1997).
17. Winiwarter S, Ridderström M, Ungell A-L, Andersson T, Zamora I. Use of molecular descriptors for absorption, distribution,
metabolism, and excretion predictions. In: Comprehensive Medicinal Chemistry II. (Volume 5). Testa B, van de Waterbeemd H (Eds).,
Elsevier, Amsterdam, The Netherlands, 531–554 (2006).
18. Gleeson MP, Hersey A, Montanari D, Overington J. Probing the links between in vitro potency, ADMET and physicochemical
parameters. Nat. Rev. Drug Discov. 10, 197 (2011).
19. Testa B, Carrupt PA, Gaillard P, Tsai RS. Intramolecular interactions encoded in lipophilicity: Their nature and significance. In:
Lipophilicity in Drug Action and Toxicology. Pliska V, Testa B, van de Waterbeemd H (Eds)., VCH, Weinheim, Germany, 49–71 (1996).
20. Drug Bioavailability: Estimation of Solubility, Permeability, Absorption and Bioavailability. van de Waterbeemd H, Lennernäs H, Artursson
P (Eds)., Wiley-VCH, Weinheim, Germany, (2003).
21. Caron G, Ermondi G, Scherrer RA. Lipophilicity, polarity, and hydrophobicity. In: Comprehensive Medicinal Chemistry II. Taylor JB,
Triggle DJ (Eds), Elsevier Science, Oxford, 5, 425–452 (2007).
22. Van de Waterbeemd H, Carter RE, Grassy G. et al. Glossary of terms used in computational drug design (IUPAC Recommendations
1997). Pure Appl. Chem. 69, 1137–1152 (1997).
23. Spyrakis F, Ahmed MH, Bayden AS, Cozzini P, Mozzarelli A, Kellog GE. The roles of water in the protein matrix: a largely untapped
resource for drug discovery. J. Med. Chem. 60, 6781–6827 (2017).
• This contribution provides an updated perspective on the roles of water molecules in protein structure, function and dynamics,
with a particular focus on the applications in drug discovery and design.
24. Cheng AC, Coleman RG, Smyth KT. et al. Structure-based maximal affinity model predicts small-molecule druggability. Nat.
Biotechnol. 25, 71–75 (2007).
•• Reports a model-based approach to predict druggable binding sites and estimate the maximal affinity acievable by a small
compound that relies on the hydrophobic desolvation, and the nonpolar surface and curvatuve of the target binding site.
25. Davis AM, Teague SJ. Hydrogen bonding, hydrophobic interactions, and failure of the rigid receptor hypothesis. Angew. Chem. Int. Ed.
Engl. 38, 736–749 (1999).
26. Hajduk PJ, Huth JR, Fesik SW. Druggability indices for protein targets derived from NMR-based screening data. J. Med. Chem. 48,
2518–2525 (2005).
27. Egner U, Hillig RC. A structural biology view of target drugability. Expert Opin. Drug Discov. 3, 391–401 (2008).
28. Schmidtke P, Barril X. Understanding and predicting druggability. A high-throughput method for detection of drug binding sites. J.
Med. Chem. 53, 5858–5867 (2010).
29. Schmidtke P, Luque FJ, Murray JB, Barril X. Shielded hydrogen bonds as structural determinants of binding kinetics: application in
drug design. J. Am. Chem. Soc. 133, 18903–18910 (2011).
30. Tsopelas F, Giaginis C, Tsantili-Kakoulidou A. Lipophilicity and biomimetic properties to support drug discovery. Expert Opin. Drug
Discov. 12, 885–896 (2017).
31. Freeman-Cook KD, Hoffman RL, Johnson TW. Lipophilic efficiency: the most important efficiency metric in medicinal chemistry.
Future Med. Chem. 5, 113–115 (2013).
32. Jopkins AL, Keserü GM, Leeson PD, Ress DC, Reynolds CH. The role of ligand efficiency metrics in drug discovery. Nat. Rev. Drug
Discov. 13, 105–121 (2014).
33. Johnson TW, Gallego RA, Edwards MP. Lipophilic efficiency as an important metric in drug design. J. Med. Chem. 61, 6401–6420
(2018).
• An updated overview of the role of lipophilic efficiency as a metric with increasing impact in guiding drug discovery.
34. Chen Z, Weber SG. A high-throughput method for lipophilicity measurement. Anal. Chem. 79, 1043–1049 (2007).
35. Giaginis C, Tsantili-Kakoulidou A. Alternative measures of lipophilicity: from octanol-water partitioning to IAM retention. J. Pharm.
Sci. 97, 2984–3004 (2008).
36. Andrés A, Rosés M, Ràfols C, Bosch E, Espinosa S, Segarra V, Huerta JM. Setup and validation of shake-flask procedures for the
determination of partition coefficients (logD) from low drug amounts. Eur. J. Phar. Sci. 76, 181–191 (2015).
37. Mannhold R, Dross K. Calculation procedures for molecular lipophilicity: a comparative study. Quant. Struct. Act. Relat. 15, 403–409
(1996).
38. Ghose AK, Viswanadhan VN, Wendoloski JI. Prediction of hydrophobic (lipophilic) properties of small organic molecules using
fragmental methods: an analysis of ALOGP and CLOGP methods. J. Phys. Chem. A 19, 172–178 (1998).
39. Mannhold R, van de, Waterbeemd H. Substructure and whole molecule approaches for calculating logP. J. Comput. Aided Mol. Des. 15,
337–354 (2001).
40. Wang J-B, Cao D-S, Zhu M-F, Yun Y-H, Xiao N, Liang Y-Z. In silico evaluation of logD7.4 abd comparison with other prediction
methods. J. Chemometrics. 29, 389–398 (2015).
41. Chen HF. In silico logP prediction for a large data set with support vector machines, radial basis neutral networks and multiple linear
regression. Chem. Biol. Drug Des 74, 142–147 (2009).
42. Mannhold R, Poda GI, Ostermann C, Tetko IV. Calculation of molecular lipophilicity: state-of-the-art and comparison of logP methods
on more than 96,000 compounds. J. Pharm. Sci. 98, 861–893 (2009).
43. Leo A, Hansch C, Elkins D. Partition coefficients and their uses. Chem. Rev. 71, 525–616 (1971).
44. Nys GG, Rekker RF. The concept of hydrophobic fragmental constants (f values). II. Extension of its applicability to the calculation of
lipophilicities of aromatic and heteroaromatic structures. Eur. J. Med. Chem. 9, 361–375 (1974).
45. Mannhold R, Rekker RF. The hydrophobic fragmental constant approach for calculating logP in octanol/water and aliphatic
hydrocarbon/water systems. Perspect. Drug Discovery Des. 18, 1–18 (2000).
46. Ghose AK, Crippen GM. Atomic physicochemical parameters for three-dimensional-structure-directed quantitative structure-activity
relationships. 2. Modeling dispersive and hydrophobic interactions. J. Chem. Inf. Comput. Sci. 27, 21–35 (1987).
47. Viswanadhan VN, Ghose AK, Revankar GR, Robins RK. An estimation of the atomic contribution to octanol-water partition coefficient
and molar refractivity from fundamental atomic and structural properties: its uses in computer-aided drug design. Math. Comput.
Model. 14, 505–510 (1990).
48. Wildman SA, Crippen GM. Prediction of physicochemical properties by atomic contributions. J. Chem. Inf. Comput. Sci. 39, 868–873
(1999)
49. Wang R, Fu Y, Lai L. A new atom-additive method for calculating partition coefficients. J. Chem. Inf. Model. 37, 615–621 (1997).
50. Ottaviani G, Martel S, Carrut P-A, In silico and in vitro filters for the fast estimation of skin permeation and distribution of new
chemical entities. J. Med. Chem. 50, 742–748 (2007).
51. Gaillard P, Carrupt PA, Testa B, Boudon A. Molecular lipophilicity potential, a Tool in 3D QSAR: method and applications. J. Comput.
Aided Mol. Des. 8, 83–96 (1994).
52. Laguerre M, Saux M, Dubost J, Carpy A. MLPP: a program for the calculation of molecular lipophilicity potential in proteins. Pharm.
Pharmacol. Commun. 3, 217–222 (1997).
53. Efremov RG, Chugunov AO, Pyrkov TV, Priestle JP, Arseniev AS, Jacoby E. Molecular lipophilicity in protein modeling and drug
design. Curr. Med. Chem. 14, 393–415 (2016).
54. Bitam S, Hamadache M, Hanini S. QSAR model for prediction of the therapeutic potency of N-benzylpiperidine derivatives as AChE
inhibitors. SAR QSAR Environ. Res. 28, 471–489 (2017).
55. Kellogg GE, Semus SF, Abraham DJ. HINT: a new method of empirical hydrophobic field calculation for CoMFA. J. Comput. Aided
Mol. Des. 5, 454–552 (1991).
56. Kellogg GE, Abraham DJ. Hydrophobicity: is Log P(o/w) more than the sum of its parts? Eur. J. Med. Chem. 35, 651–661 (2000).
57. Fornabaio M, Spyrakis F, Mozzarelli A, Cozzini P, Abraham DJ, Kellogg GE. Simple, intuitive calculations of free energy of binding for
protein-ligand complexes. 3. The free energy contribution of structural water molecules in HIV-1 protease complexes. J. Med. Chem. 47,
4507–4516 (2004).
58. Amadasi A, Spyrakis F, Cozzini P, Abraham DJ, Kellogg GE, Mozzarelli A. Mapping the energetics of water-protein and water-ligand
interactions with the ‘natural’ HINT forcefield: predictive tools for characterizing the roles of water in biomolecules. J. Mol. Biol. 358,
289–309 (2006).
59. Marabotti A, Spyrakis F, Facchiano A. et al. Energy-based prediction of amino acid-nucleotide base recognition. J. Comput. Chem. 29,
1955–1969 (2008).
60. Amadasi A, Surface JA, Spyrakis F, Cozzini P, Mozzarelli A, Kellogg GE. Robust classification of ‘relevant’ water molecules in putative
protein binding sites. J. Med. Chem. 51, 1063–1067 (2008).
61. Ahmed MH, Spyrakis F, Cozzini P. et al. Bound water at protein-protein interfaces: partners, roles and hydrophobic bubbles as a
conserved motif. PLoS ONE 6, e24712 (2011).
62. Rogers KS, Cammarata A. A molecular orbital description of the partitioning of aromatic compounds between polar and non-polar
phases. Biochim. Biophys. Acta 193, 22–29 (1969).
63. Rogers KS, Cammarata A. Superdelocalizability and charge density. A correlation with partition coefficients. J. Med. Chem. 12(4),
692–693 (1969).
64. Bodor N, Gabanyi Z, Wong C. A new method for the estimation of partition coefficient. J. Am. Chem. Soc. 111, 3783–3786 (1989).
65. Bodor N, Huang MJ. An extended version of a novel method for the estimation of partition coefficients. J. Pharm. Sci. 81, 272–281
(1992).
66. Breindl A, Beck B, Clark T, Glen RC. Prediction of the n-octanol/water partition coefficient, logP, using a combination of semiempirical
MO-calculations and a neural network. J. Mol. Model. 3, 142–155 (1997).
67. Beck B, Breindl A, Clark T. QM/NN QSPR models with error estimation: vapor pressure and Log P. J. Chem. Inf. Comput. Sci. 40,
1046–1051 (2000).
68. Du Q, Liu PJ, Mezey PG. Theoretical derivation of heuristic molecular lipophilicity potential: a quantum chemical description for
molecular solvation. J. Chem. Inf. Model. 45, 347–353 (2005).
69. Du Q-S, Li D-P, He W-Z, Chou K-C. Heuristic molecular lipophilicity (HMLP): lipophilicity and hydrophilicity of amino acid side
chains. J. Comput. Chem. 27, 685–692 (2006).
70. Palmer DS, Mišin M, Fedorov MV, Llinas A. Fast and general method to predict the physicochemical properties of druglike molecules
using the integral equation theory of molecular liquids. Mol. Pharm. 12(9), 3420–3432 (2015).
71. Güssregen S, Matter H, Hessler G, Lionta E, Heil J, Kast SM. Thermodynamic characterization of hydration sites from integral
equation-derived free energy densities: application to protein binding sites and ligand series. J. Chem. Inf. Model. 57, 1652–1666 (2017).
72. Ansari SM, Palmer DS. Comparative molecular field analysis using molecular integral equation theory. J. Chem. Inf. Model. 58(6),
1253–1265 (2018).
73. Orozco M, Luque FJ. Theoretical methods for the description of the solvent effect in biomolecular systems. Chem. Rev. 100, 4187–4226
(2000).
74. Tomasi J, Mennucci B, Cammi R. Quantum mechanical continuum solvation models. Chem. Rev. 105, 2999–3094 (2005).
75. Cramer CJ, Truhlar DG. A universal approach to solvation modeling. Acc. Chem. Res. 41, 760–768 (2008).
76. Klamt A, Mennucci B, Tomasi J. et al. On the performance of continuum solvation methods. A comment on ‘Universal approaches to
solvation modeling’. Acc. Chem. Res. 42, 489–492 (2009).
77. Klamt A. The COSMO and COSMO-RS solvation models. WIRES Comput. Mol. Sci. 8, e1338 (2018).
78. Thormann M, Klamt A, Hornig M, Almstetter M. COSMOsim: bioisosteric similarity based on COSMO-RS σ-profiles. J. Chem. Inf.
Model. 64, 1040–1053 (2006).
•• Reports the application of the σ-profiles derived from Conductor-like Screening Model for Realistic Solvation (COSMO-RS)
calculations in drug similarity measurements.
79. Hornig M, Klamt A. COSM Ofrag: a novel tool for high-throughput ADME property prediction and similarity screening based on
quantum chemistry. J. Chem. Inf. Model. 45, 1169–1177 (2005).
80. Thormann M, Klamt A, Wichmann K. COSMOsim3D: 3D-Similarity and alignment based on COSMO polarization charge densities.
J. Chem. Inf. Model. 52, 2149–2156 (2012).
81. Klamt A, Thormann M, Wichmann K, Tosco P. COSMOsar3D: molecular field analysis based on local COSMO σ-profiles. J. Chem.
Inf. Model. 52, 2157–2164 (2012).
•• The contribution examines the usage of local σ-profiles in molecular field analysis, providing an interpretation about the features
of the virtual free energy field generated from the target binding pocket.
82. Ginex T, Muñoz-Muriedas J, Herrero E, Gibert E, Cozzini P, Luque FJ. Development and validation of hydrophobic molecular fields
derived from the quantum mechanical IEF/PCM-MST solvation models in 3D-QSAR. J. Comput. Chem. 37, 1147–1162 (2016).
•• Examines the performance of QM-based Miertus–Scrocco–Tomasi lipophilic (Hyphar) descriptors for calculation of molecular
fields in the derivation of structure-activity relationships models.
83. Ginex T, Muñoz-Muriedas J, Herrero E, Gibert E, Cozzini P, Luque FJ. Application of the quantum mechanical IEF/PCM-MST
hydrophobic descriptors to selectivity in ligand binding. J. Mol. Model. 22, 136 (2016).
84. Miertus S, Scrocco E, Tomasi J. Electrostatic interaction of a solute with a continuum. A direct utilization of ab initio molecular
potentials for the prevision of solvent effects. Chem. Phys. 55, 117–129 (1981).
85. Cancès E, Mennucci B, Tomasi J. A new integral equation formalism for the polarizable continuum model: theoretical background and
applications to isotropic and anisotrpic dielectrics. J. Chem. Phys. 107 , 3032 (1997).
86. Bachs M, Luque FJ, Orozco M. Optimization of solute cavities and van der Waals parameters in ab initio MST-SCRF calculations of
neutral molecules. J. Comput. Chem. 15, 446–454 (1994).
87. Luque FJ, Bachs M, Orozco M. An optimized AM1/MST method for the MST-SCRF representation of solvated systems. J. Comput.
Chem. 15, 847–857 (1994).
88. Curutchet C, Orozco M, Luque FJ. Solvation in octanol: parametrization of the continuum MST model. J. Comput. Chem. 22,
1180–1193 (2001).
89. Soteras I, Curutchet C, Bidon-Chanal A, Orozco M, Luque FJ. Extension of the MST model to the IEF formalism: HF and B3LYP
parametrizations. J. Mol. Struct. 727, 29–40 (2005).
90. Luque FJ, Curutchet C, Muñoz-Muriedas J. et al. Continuum solvation models: dissecting the free energy of solvation. Phys. Chem.
Chem. Phys. 5, 3827–3836 (2003).
91. Luque FJ, Bofill JM, Orozco M. New strategies to incorporate the solvent polarization in self-consistent reaction field and free-energy
perturbation simulations. J. Chem. Phys. 103, 10183–10191 (1995).
92. Vázquez J, Deplano A, Herrero A. et al. Development and validation of molecular overlays derived from 3D Hydrophobic Similarity
with PharmScreen. J. Chem. Inf. Model. 58, 1596–1609 (2018).
•• A comparative analysis of electrostatic/steric and QM-based lipophilicity (Hyphar) descriptors for predicting molecular overlays
from 3D similarity measurements.
93. PharmScreen - PharmQSAR, Pharmacelera, Barcelona, Spain. www.pharmacelera.com
94. Giangreco I, Cosgrove DA, Packer MJ. An extensive and diverse set of molecular overlays for the validation of pharmacophore programs.
J. Chem. Inf. Model. 53, 852–866 (2013).
95. Sutherland JJ, O’Brien LA, Weaver DF. A comparison of methods for modeling quantitative structure-activity relationships. J. Med.
Chem. 47, 5541–5554 (2004).
96. Forti F, Barril X, Luque FJ, Orozco M. Extension of the MST continuum solvation model to the RM1 semiempirical Hamiltonian. J.
Comput. Chem. 29, 578–587 (2008).
97. Muñoz J, Barril X, Hernandez B, Orozco M, Luque FJ. Hydrophobic similarity between molecules: A MST-based hydrophobic
similarity index. J. Comput. Chem. 23, 554–563 (2002).
98. Muñoz-Muriedas J, Perspicace S, Bech N, Guccione S, Orozco M, Luque FJ. Hydrophobic molecular similarity from MST fractional
contributions to the octanol/water partition coefficient. J. Comput. Aided Mol. Des. 19, 401–419 (2005).
99. Muñoz-Muriedas J, Barril X, Lopez JM, Orozco M, Luque FJ. A hydrophobic similarity analysis of solvation effects on nucleic acid
bases. J. Mol. Model. 13, 357–365 (2007).