Use of Support Vector Machine in Pattern Classification: Application to QSAR Studies

Abstract

The Support Vector Machine (SVM) approach for classification and regression problems was originally developed by Vapnik and co-workers [1]. For the last few years it has been gaining acceptance in the machine learning community [2]. The purpose of this paper is to evaluate SVM performance in the quantitative structure-activity relationship (QSAR) domain for classification applications and to compare the performance of one particular implementation of an SVM [3] to one particular implementation of an artificial neural network (ANN) [4]. For this purpose, we used artificial data simulating various response surfaces, and biological data derived from the literature covering various pharmacological domains. The results obtained on biological data are also compared to previous work using other modeling techniques. We also discuss the usage of SVM in building QSAR models for biological activity of drugs.