Prediction of Drug-likeness of Central Nervous System Drug Candidates Using a Feed-Forward Neural Network Based on Chemical Structure

The modern medical science has been greatly advanced by the development of new drugs, despite the fact that the process of developing new drugs is costly and time-consuming. An accurate prediction method for the drug-likeness at the early stage of drug discovery is highly desirable, as it will facilitate the discovery process and reduce the overall cost, and eventually contribute to the well-being of human beings. Based on a central nervous system (CNS) drug dataset, we constructed an artificial neural network (NN) to predict the CNS drug-likeness of a given compound. Based on the published results, we first constructed a simple feed-forward neural network to learn and predict the possible correlations between twelve physiochemical properties and the CNS drug-likeness. The accuracy of prediction has reached 80%, which is higher than previous reports. The successful implementation of NN to predict the CNS drug-likeness indicated that NN could be a powerful tool for the prediction. Moreover, we further constructed a neural network based on the chemical structure, and the accuracy has reached 86%. We hope that these methods can serve as an applicable set of protocols for virtual drug screening.