Activity prediction of some nontested anticancer compounds using GA-based PLS regression models

This study is an attempt to formulate 3D quantitative structure-activity relationship (3D-QSAR) model for 6-(2,6-dichlorophenyl)-pyrido[2,3-d]pyrimidin-7(8H)-one compounds based on computed molecular descriptors. Molecular field analysis technique has been employed to find out specific contribution of structural features such as steric, electrostatic, and hydrophobic fields of these compounds showing anticancer activities by the inhibition of epidermal growth factor kinase. Three-dimensional QSAR model is developed based on the training set using genetic algorithm feature selection combined with partial least square method. The training model is then used to predict the biological activities of some similar class of compounds which were synthesized only, but the activities were not tested. An accuracy of activity prediction has been cross-checked by introducing a new way of QSAR model validation approach utilizing random normalization correction procedure in the data set.