Radial basis function neural networks based on the projection pursuit and principal component analysis approaches: QSAR analysis of fullerene[C60]-based HIV-1 PR inhibitors

Abstract

In this work, projection pursuit (PP) and principal component analysis (PCA) are combined with radial basis function networks (RBFNs) to perform the quantitative structure–activity relationship (QSAR) studies on the binding affinities (pEC50, i.e., minus decimal logarithm of the 50 % effective concentration) of 47 fullerene derivatives as inhibitors of the human immunodeficiency virus type 1 protease. RBFN is applied to construct the nonlinear QSAR models. The input of RBFN is the scores of PP or PCA, and genetic algorithm is used to select the centers of RBFN. The methods are called PP-GA-RBF and PCA-GA-RBF, respectively. The aim is the performance comparison of the proposed methods. To evaluate the performance of the methods, various statistical parameters such as \( Q_{F2}^{2} \) and \( r_{m}^{2} \) are calculated. The results demonstrated that the predictive performance of the proposed PP-GA-RBF is better than PCA-GA-RBF and previous studies. The applicability domain of the models is assessed by leverage and distance approaches.