Prediction of Superoxide Quenching Activity of Fullerene (C₆₀) Derivatives by Genetic Algorithm-Support Vector Machine

Abstract

In this study, the quantitative structure–activity relationship (QSAR) models for the superoxide quenching activity of fullerene derivatives were developed. The dataset was divided into a training set and test set, based on hierarchical clustering technique. Three variables were selected by the genetic algorithm (GA) variable subset selection procedure. Multiple linear regressions (MLR) and support vector machine (SVM) were used as linear and nonlinear methods. Both the linear and nonlinear models could give very satisfactory prediction results: The square correlation coefficient of R² for the training and test sets were 0.826 and 0.981, by MLR and 0.957 and 0.965, by SVM methods, respectively. The prediction result of the SVM model was better than that obtained by MLR method, which proved SVM was a useful tool in the prediction of the superoxide quenching activity of fullerene derivatives. The results suggested that the minimum atomic partial charge, minimum valence of hydrogen atom, and minimum atomic state energy of oxygen atom, are the main independent factors contributing to the superoxide quenching activity of fullerene derivatives.

Keywords:

• QSAR
• support vector machine
• fullerene; genetic algorithm
• superoxide quenching