Linear and non-linear QSAR studies have been performed in present investigation with multiple linear regressions (MLR) analysis and Support vector machine (SVM) using different kernels. Three relevant descriptors out of fifteen descriptors calculated are identified as LOGP values, G3e and Rte+. Their relationship with biological activity IC50 have provided structural insights in interpretation and serializing the results into a pragmatic approachable technique. QSAR models obtained show statistical fitness and good predictability. SVM using Gaussian kernel function was found more efficient in prediction of IC50 of training set of thirty small molecules HIV-1 capsid inhibitors. Y-scrambling, PRESS and test set were used as validation parameters. SVM was found superior to training set prediction and internal validations and found inferior to external test set (11 molecules) predictions. Wherein MLR analysis it was vice-versa. Mechanistic interpretation of selected descriptors from both the models actuates further research.
Keywords: HIV-1 capsid, AIDS, QSAR, MLR, SVM. linear QSAR, non-linear QSAR, multiple linear regressions (MLR) analysis, Support vector machine (SVM), PRESS