FEATURE EXTRACTION USING LINEAR AND NONLINEAR QSAR STUDY ON SEVERAL TAXOL DERIVATIVES AS ANTICANCER DRUGS

The activity of the Paclitaxel derivatives was estimated using multiple linear regression (MLR), artificial neural network (ANN) as modelling tools, and genetic algorithm (GA) and simulated annealing algorithm (SA) as optimization techniques. These models were employed to choose the best set of descriptors in a cross-validation procedure for non-linear-log (IC50) (the empirical negative logarithm half maximal inhibitory concentration) prediction. A high predictive ability was observed for the MLRMLR, GA-MLR, GA-ANN models, with root mean sum square errors (RMSE) of 0.421, 0.0712, 0.160, 0.0534 in gas phase and 0.910, 0.965, 0.922, 0.976 in solvent, respectively. The results obtained using the GA-ANN method indicated that the activity of the derivatives of Paclitaxel depends on different parameters such as E2u, BELP1, HATS6p, piPC05, Morl4u, BELv8, RDF120m, RDF025p descriptors in gas phase including BEHe8, Mor07u, H5u, Eeig11r in the solvent phase.