A Statistical Learning Method for the Prediction of Anti-HIV Activity Using Topological Indices Based on Steiner Eccentricity

The prediction of molecular activity plays an important role in drug discovery. Various approaches have been devised to predict anti-HIV activity based on topological indices. This study proposes Steiner 3-eccentric connectivity index and Steiner 3-eccentric distance sum and applies them to predict anti-HIV activity of a molecule. A support vector machine model is established for the prediction of the anti-HIV activity over a dataset comprising 1795 compounds. Different dimensions of feature vectors over Wiener index, Randić index, graph energy, Steiner 3-eccentric connectivity index, and Steiner 3-eccentric distance sum are considered in our experiment. Cross-validation shows that the Steiner 3-eccentric connectivity index and Steiner 3-eccentric distance sum could be used to predict the anti-HIV activity. A combination of indices would earn a good performance than a single index. However, sometimes, more indices could not provide a better performance in prediction of anti-HIV activity.