QSAR classification models for Bcl-2 and Bcl-xL inhibitors using Supervised Kohonen maps and Linear Discriminant Analysis methods

Bcl-2 and Bcl-XL are key apoptosis regulators whose their inhibition is a very attractive strategy for cancer therapy [1-3]. The aim of this study was to apply linear and nonlinear QSAR classification approaches to classify of Bcl-2 and Bcl-XL targets from Binding database. To this end, we created a large database of 1374 molecules which was divided into the active and inactive classes on the basis of their activity values. To classify the data set, molecules were categorized into four classes: Class 1: active selective inhibitors of Bcl-2; Class 2: active selective inhibitors of Bcl-xL; Class 3: inactive compounds for Bcl-2, and Class 4: inactive compounds for Bcl-xL. The classification models of Bcl-2 and Bcl-xL inhibitors were proposed in two different ways: 1) Development of an active-active classifier for separating active inhibitors of Bcl-2 from active inhibitors of Bcl-xL. This model promoted the design of selective inhibitors and the extraction of important pharmacophores which induced selectivity for Bcl-2 or Bcl-xL targets. 2) Development of an active-inactive classifier, that identified the key structural for differentiating active inhibitors of Bcl-2 and Bcl-xL from inactive ones. The general aim of the classification models discovered a discriminatory hyperplane in the feature space to facilitate separation active from inactive inhibitors of Bcl-2 and Bcl-xL and helped to design selective and potent inhibitors. The genetic algorithm (GA) was used to select most efficient subsets of the molecular descriptors. The model containing eight descriptors based on the GA- supervised kohonen maps (SKM) showed a better predictive ability than GA- linear discriminant analysis (LDA). The prediction accuracy of the training set was 83.9%, 91.8%, 82.5% and 79.4% for SKM, and 82.4%, 80.7%, 80.3% and 69.9% for LDA, respectively. The comparison study show that GA-SKM method can be used as a powerful modeling tool For classification molecules according to their activity values and therapeutic targets.