In Silico Prediction of Human Organ Toxicity via Artificial Intelligence Methods

Unpredicted human organ level toxicity remains one ofthe majorreasons for drug clinical failure. There is a critical need for cost-efficientstrategies in the early stages of drug development for human toxicityassessment. At present, artificial intelligence methods are popularlyregarded as a promising solution in chemical toxicology. Thus, weprovided comprehensive in silico prediction modelsfor eight significant human organ level toxicity end points usingmachine learning, deep learning, and transfer learning algorithms.In this work, our results showed that the graph-based deep learningapproach was generally better than the conventional machine learningmodels, and good performances were observed for most of the humanorgan level toxicity end points in this study. In addition, we foundthat the transfer learning algorithm could improve model performancefor skin sensitization end point using source domain of invivo acute toxicity data and in vitro dataof the Tox21 project. It can be concluded that our models can provideuseful guidance for the rapid identification of the compounds withhuman organ level toxicity for drug discovery.