QSAR of adenosine A3 receptor antagonist 1,2,4-triazolo[4,3-a]quinoxalin-1-one derivatives using chemometric tools

Considering the potential of selective adenosine A3 receptor subtype ligands in the development of prospective therapeutic agents, an attempt has been made to explore physicochemical requirements of 1,2,4-triazolo[4,3-a]quinoxalin-1-one derivatives for A3 receptor binding. In this study, lipophilicity (logP), physicochemical substituent constants (pi, MR, sigma p) of phenyl ring substituents, and Wang-Ford charges of common atoms of the quinoxaline nucleus (calculated from molecular electrostatic potential surface of energy-minimized geometry using AM1 technique) were used as independent variables along with suitable dummy parameters. The best multiple linear regression (MLR) equation obtained from factor analysis (FA-MLR) as the preprocessing step could explain and predict 72.6% and 65.3%, respectively, of the variance of the binding affinity. The same equation also emerged as the best equation in the population of 100 equations obtained from genetic function approximation (GFA-MLR). The results suggested that presence of an electron-withdrawing group at the para position of the phenyl ring would be favorable for the binding affinity. Again, the presence of a nitro group at position R1 increases the binding affinity. When factor scores were used as predictor variables in the principal component regression analysis, the resultant model showed 78.6% explained variance and 63.1% predicted variance. The best equation derived from G/PLS could explain and predict 74.4% and 64.8%, respectively. The results have suggested the importance of Wang-Ford charges of atoms C15 and C19, apart from positive contributions of electron-withdrawing para substituents of the variance of the phenyl ring and nitro group at the R1 position.     

Comparative molecular field analysis of colchicine inhibition and tubulin polymerization for combretastatins binding to the colchicine binding site on beta-tubulin

A molecular modeling study using Comparative Molecular Field Analysis (CoMFA) was undertaken to develop a predictive model for combretastatin binding to the colchicine binding site of tubulin. Furthermore, we examined the potential contribution of lipophilicity (log P) and molecular dipole moment and were unable to correlate these properties to the observed biological data. In this study we first confirmed that tubulin polymerization inhibition (IC50) correlated (R2 = 0.92) with [3H]colchicine displacement. Although these data correlated quite well, we developed two independent models for each set of data to quantify structural features that may contribute to each biological property independently. To develop our predictive model we first examined a series of molecular alignments for the training set and ultimately found that overlaying the respective trimethoxyphenyl rings (A ring) of the analogues generated the best correlated model. The CoMFA yielded a cross-validated R2 = 0.41 (optimum number of components equal to 5) for the tubulin polymerization model and an R2 = 0.38 (optimum number of components equal to 5) for [3H]colchicine inhibition. Final non-cross-validation generated models for tubulin polymerization (R2 of 0.93) and colchicine inhibition (R2 of 0.91). These models were validated by predicting both biological properties for compounds not used in the training set. These models accurately predicted the IC50 for tubulin polymerization with an R2 of 0.88 (n = 6) and those of [3H]colchicine displacement with an R2 of 0.80 (n = 7). This study represents the first predictive model for the colchicine binding site over a wide range of combretastatin analogues.     

Development of CYP3A4 inhibition models: comparisons of machine-learning techniques and molecular descriptors

Computational models of cytochrome P450 3A4 inhibition were developed based on high-throughput screening data for 4470 proprietary compounds. Multiple models differentiating inhibitors (IC(50) <3 microM) and noninhibitors were generated using various machine-learning algorithms (recursive partitioning [RP], Bayesian classifier, logistic regression, k-nearest-neighbor, and support vector machine [SVM]) with structural fingerprints and topological indices. Nineteen models were evaluated by internal 10-fold cross-validation and also by an independent test set. Three most predictive models, Barnard Chemical Information (BCI)-fingerprint/SVM, MDL-keyset/SVM, and topological indices/RP, correctly classified 249, 248, and 236 compounds of 291 noninhibitors and 135, 137, and 147 compounds of 179 inhibitors in the validation set. Their overall accuracies were 82%, 82%, and 81%, respectively. Investigating applicability of the BCI/SVM model found a strong correlation between the predictive performance and the structural similarity to the training set. Using Tanimoto similarity index as a confidence measurement for the predictions, the limitation of the extrapolation was 0.7 in the case of the BCI/SVM model. Taking consensus of the 3 best models yielded a further improvement in predictive capability, kappa = 0.65 and accuracy = 83%. The consensus model could also be tuned to minimize either false positives or false negatives depending on the emphasis of the screening.