Particle swarm optimization-based support vector regression and Bayesian networks applied to the toxicity of organic compounds to tadpoles (Rana japonica)

Abstract

Particle swarm optimization (PSO) is a new optimization method with strong global search capability. In present work, PSO-support vector regression (SVR) model was proposed to predict the toxicity of organic compounds to tadpoles (Rana japonica), in which PSO was used to determine free parameters of SVR. These results showed that the prediction accuracy of PSO-SVR model is higher than those mode of MLR and PLS. Moreover, Bayesian networks (BNs) was adopted to describe the relationship between toxicity associated with molecular descriptors in this work. The result of BNs was considered to be reasonable.