Activities Prediction of Drug Molecules by Using the Optimal Ensemble Based on Uniform Design

Abstract

Neural network ensemble is a powerful tool for simulating the quantitative structure activity relationship in drug discovery because of its high generalization ability. However, the architecture of the ensemble and the training parameters of individual neural networks are closely relative to the generalization performance of the ensemble and the convenience of the creation of the ensemble. This paper proposes a novel creation algorithm for neural network ensemble, which employs uniform design to guide users to design the ensemble architecture and adjust the training parameters of individual neural networks. In addition, this algorithm is applied to produce neural network ensemble for predicting activities of drug molecules, which is a convenient way to achieve better results than commonly used bagging methods.