Prediction of antifungal activity with support vector machine

Abstract

A set of 67 molecular descriptors, including electronic, topological, geometric descriptors and molecular shape indices, were calculated and used to predict the antifungal activity for 94 organic compounds by means of support vector machine method. The model was validated in two ways: leave-one-out and 5-fold cross-validation. In the 5-fold cross-validation, the compounds were divided into several clusters based on their similarities. The training sets were sorted by selecting molecules randomly from each cluster, the rest of the molecules being the test set. It was shown that two validation methods give similar results and our model has a good prediction ability, and about 84% of the compounds can be correctly classified.