Neural networks and support vector machines based bio-activity classification

Abstract

Classification of various compounds into their respective biological activity classes is important in drug discovery applications from an early phase virtual compound filtering and screening point of view. In this work two types of neural networks, multi layer perceptron (MLP) and radial basis functions (RBF), and support vector machines (SVM) were employed for the classification of three types of biologically active enzyme inhibitors. Both of the networks were trained with back propagation learning method with chemical compounds whose active inhibition properties were previously known. A group of topological indices, selected with the help of principle component analysis (PCA) were used as descriptors. The results of all the three classification methods show that the performance of both the neural networks is better than the SVM.