Prediction of Mutagenicity of Organic Molecules by Ensemble Learning

Abstract

In this paper, the results of construction of classification models for mutagenicity of organic molecules are described. The objective of this study is to construct a model which can predict results of reverse mutation test with high accuracy. For this end, we propose a novel ensemble modeling method in which a lot of support vector machine (SVM) models are constructed as a sub-model and integrated to predict mutagenicity. For constructing sub-model, a part of data matrix which is randomly selected from an original data matrix and randomly determined SVM parameters are used. After the construction of sub-models, a certain number of models which have high accuracy rate are selected and integrated to predict mutagenicity. We constructed an ensemble model using a data set of reverse mutation test which was collected by Hansen et al. [K. Hansen, et al., J. Chem. Inf. Model., 49, 2077-2081] to estimate the proposed method. As a result, an ensemble model with accuracy of 79.6% was successfully obtained. The area under ROC-curve (AUC) is 0.866, which is slightly better than that of Hansen et al. Thus we concluded that the ensemble modeling with SVM sub-models are a promising method for predicting mutagenicity of organic molecules.