Annick Panaye

Comparative study of QSAR/QSPR correlations using support vector machines, radial basis function neural networks, and multiple linear regression.

Support vector machines (SVMs) were used to develop QSAR models that correlate molecular structures to their toxicity and bioactivities. The performance and predictive ability of SVM are investigated and compared with other methods such as multiple linear regression and radial basis function neural network methods. In the present study, two different data sets were evaluated. The first one involves an application of SVM to the development of a QSAR model for the prediction of toxicities of 153 phenols, and the second investigation deals with the QSAR model between the structures and the activities of a set of 85 cyclooxygenase 2 (COX-2) inhibitors. For each application, the molecular structures were described using either the physicochemical parameters or molecular descriptors. In both studied cases, the predictive ability of the SVM model is comparable or superior to those obtained by MLR and RBFNN. The results indicate that SVM can be used as an alternative powerful modeling tool for QSAR studies.

Steric effects—I: A critical examination of the taft steric parameter—Es. Definition of a revised, broader and homogeneous scale. Extension to highly congested alkyl groups

Abstract The basic assumptions used in obtaining the Taft Es scale of steric parameters have been examined critically. Certain inconsistencies in the existing scale are pointed out which motivate a rigorous recalculation of the scale based on a single defining reaction: the acid catalysed esterification of carboxylic acids in MeOH at 40°. This revised Taft scale is termed Es. The scale includes 44 of the original groups cited by Taft with 50 additional values obtained from literature data and has been extended to extremely hindered alkyl groups (13 in number) by measurement based on competitive reactivity. The complete scale from its least hindered group (H, Es = 1.12) to its most hindered (t-BuPriMeC-, Es = −7.56) spans more than eight powers of ten. A definite levelling effect has been observed in the region of Es ~−6 and an inversion effect in two cases (i-Pr3C- and t-BuPriEtC-). Some currently used quantitative approaches to steric effects are discussed and compared in the light of the Es scale. The corrected steric parameters, Esc and Eso, have been shown not to be of general applicability and the use of the van der Waals radius relative to hydrogen subject to limitation.

Nonlinear SVM Approaches to QSPR/QSAR Studies and Drug Design

Recently, a new promising nonlinear method, the support vector machine (SVM), was proposed by Vapnik. It rapidly found numerous applications in chemistry, biochemistry and pharmacochemistry. Several attempts using SVM in drug design have been reported. It became an attractive nonlinear approach in this field. In this review, the theoretical basis of SVM in classification and regression is briefly described. Its applications in QSPR/QSAR studies, and particularly in drug design are discussed. Comparative studies with some linear and other nonlinear methods show SVMs high performance both in classification and correlation.

13C NMR CHEMICAL SHIFT PREDICTION OF SP2 CARBON ATOMS IN ACYCLIC ALKENES USING NEURAL NETWORKS

The 13C NMR chemical shift of sp2 carbon atoms in acyclic alkenes was estimated with multilayer feedforward artificial neural networks (ANNs) and multilinear regression (MLR), using as structural descriptors a vector made of 12 components encoding the environment of the resonating carbon atom. The neural network quantitative model provides better results than the MLR model calibrated with the same data. The predictive ability of both the ANN and MLR models was tested by the leave-20%-out (L20%O) cross-validation method, demonstrating the superior performance of the neural model. The number of neurons in the hidden layer was varied between 2 and 7, and three activation functions were tested in the neural model:  the hyperbolic tangent or a bell-shaped function for the hidden layer and a linear or a hyperbolic tangent function for the output layer. All four combinations of activation functions give close results in the calibration of the ANN model, while for the prediction a linear output function performs ...

Quantitative prediction of liquid chromatography retention of N-benzylideneanilines based on quantum chemical parameters and radial basis function neural network.

Based on quantum chemical parameters and a simple numerical coding, the liquid chromatography retention of bifunctionally substituted N-benzylideneaniles (NBA) has been predicted using a radial basis function neural network (RBFNN) model. The quantum chemical parameters involved in the model are dipole moment (m), energies of the highest occupied and lowest unoccupied molecular orbitals (E(homo,) E(lumo)), net charge of the most negative atom (Q(min)), sum of absolute values of the charges of all atoms in two given functional groups (Delta), total energy of the molecule (E(T)), weight of the molecule (W), and numerical coding (N). N was used to indicate the different positions of two substituents. The predictive values are consistent with the experimental results. The mean relative error of the testing set is 1.6%, and the maximum relative error is less than 5.0%. In this work the success of the whole modeling process only depends on the optimization of the spread parameter in network.

13C NMR Chemical Shift Prediction of the sp3 Carbon Atoms in the Position Relative to the Double Bond in Acyclic Alkenes.

The 13 C NMR chemical shift of sp 3 carbon atoms situated in the R position relative to the double bond in acyclic alkenes was estimated with multilayer feedforward artificial neural networks (ANNs) and multilinear regression (MLR), using as structural descriptors a topo-stereochemical code which characterizes the environment of the resonating carbon atom. The predictive ability of the two models was tested by the leave-20%-out cross-validation method. The neural model provides better results than the MLR model both in calibration and in cross-validation, demonstrating that there exists a nonlinear relationship between the structural descriptors and the investigated 13 C NMR chemical shift and that the neural model is capable to capture such a relationship in a simple and effective way. A comparison between a general model for the estimation of the 13 C NMR chemical shift and the ANN model indicates that general models are outperformed by more specific models, and in order to improve the predictions a possible way is to develop environment-specific models. The approach proposed in this paper can be used in automated spectra interpretation or computer-assisted structure elucidation to constrain the number of possible candidates generated from the experimental spectra.

Steric effects-II: Relationship between topology and the steric parameter. E's—topology as a tool for the correlation and prediction of steric effects

The steric effect of alkyl groups as characterized by the revised Taft Es parameter is analysed using an approach based on the DARC topological system and its PELCO correlation method. This approach involves an analysis of the systemativ variation of Es in a topological dequencing of alkyl groups and shows the existence of three regions of distinct behaviour: R I, a “normal” behaviour region (ca 6 Es units) in which the contribution of the introduction of successive Me groups to the overall steric effect increases monotonically (groups with 1 to 7 carbons); R II, a region, in which a “levelling” effect is observed, i.e. the contribution diminishes and becomes nil (groups with 8 and 9 carbons); and R III, where this contribution changes sign, “inversion” effect (groups with 10 carbons). Using a series of successive approximations, topological models are developed and tested. The conditions under which the topology may be used to represent the topography (i.e. the real 3-dimensional structure) are considered. The correlation of existing Es values and the reliable prediction of experimentally unavailable steric effects are direct consequences of this treatment.

CoMFA/CoMSIA/HQSAR and Docking Study of the Binding Mode of Selective Cyclooxygenase (COX‐2) Inhibitors

Abstract The intermolecular interaction between four types of anti‐inflammatory inhibitors (oxazoles, pyrazoles, pyrroles and imidazoles) and COX‐2 receptor was studied. The results of docking suggest that they have similar interaction mechanism. The most active compounds of these four types of inhibitors could both form several hydrogen bonds with residues His90, Arg513, Leu352 and Arg120, and develop hydrophobic interaction with residues Phe518, Leu352 and Leu359. This is consistent with the investigation reported by R. G. Kurumbail et al. (Nature. 1996, 384, 644‐648). A common 3D‐QSAR model could be constructed with these four categories of COX‐2 inhibitors using the method of docking‐ guided conformer selection. The cross‐validated q2 values are found as 0.741 and 0.632 for CoMFA and CoMSIA respectively. And the non‐cross‐validated r2 values are 0.887 and 0.885. 54 inhibitors constitute the test set used to validate the model. The results show that this model possesses good predictive ability for diverse COX‐2 inhibitors. Furthermore, a HQSAR model was used to evaluate the influence of substituents on anti‐inflammatory activity. Compared with the results of previous works, our model possesses significantly better prediction ability. It could help us to well understand the interaction mechanism between inhibitors and COX‐2 receptor, and to make quantitative prediction of their inhibitory activities.

CISOC-PSCT: a predictive system for carcinogenic toxicity

A SAR based carcinogenic toxicity prediction system, CISOC-PSCT, was developed. It consisted of two principal phases: the construction of relationships between structural descriptors and carcinogenic toxicity indices, and prediction of the toxicity from the SAR model. The training set included 2738 carcinogenic and 4130 non-carcinogenic compounds. Three predefined topological types of substructures termed Star, Path and Ring were used to generate the descriptors for each structure in the training set. In this system, the defined carcinogenic toxicity index (CTI) was obtained from the probability of a structural descriptor to either belong to the carcinogenic or non-carcinogenic compounds. Based on these structural descriptors and their CTI, a SAR model was derived. Then the carcinogenic possibility (CP) and the carcinogenic impossibility (CIP) of compounds were predicted. The model was tested from a testing set of 304 carcinogenic compounds (MDL toxicity database), 460 non-carcinogenic compounds (CMC database) and 94 compounds extracted from two traditional Chinese medicine herbs.

Comparative study of non nucleoside inhibitors with hiv-1 reverse transcriptase based on 3D-QSAR and docking

The intermolecular interaction between two types of non nucleoside reverse transcriptase inhibitors (NNRTIs), HEPT and TIBO, and HIV reverse transcriptase receptor (HIVRT) was investigated. The result of docking study showed that two types of NNRTIs presented similar interaction mechanism with HIVRT. The most active compound of every type of inhibitors could form one hydrogen bond with the residue Lys101 and has hydrophobic interaction with residues Tyr181, Tyr188 and Tyr318, etc. Three 3D-QSAR models including two partial correlation models (one for each family of HEPT and TIBO) and a mixed model gathering two families were constructed. Comparative study of these models indicated that the mixed model offered the strongest prediction ability. For this model, the cross-validated q 2 values were 0.720 and 0.675, non-cross-validated r 2 values were 0.940 and 0.920 for CoMFA and CoMSIA, respectively. It has been validated by using a test set of 27 inhibitors. Compared with previously reported works, our model showed better prediction ability. It could help us to insight the interaction between NNRTIs and HIVRT, and to design new anti-HIV NNRTIs inhibitors.