Artificial Intelligence for Drug-Induced Liver Injury Prediction: A Systematic Review

Abstract

Drug toxicity and side effects are the cause of clinical trials failure, resulting in wasted investment and time. Liver injury is one of the leading causes of drug rejection in the clinical phase. Machine learning models have been developed to predict induced liver damage with significant results. We carried out a systematic review to identify and describe the AI strategies reported in the scientific literature for drug induced liver injury prediction. Using combinations of keywords, we retrieved articles from Google Scholar and PubMed database and analyzed them according to defined criteria. PRISMA guidelines were used in the articles selection process. The main algorithms reported were Random Forest, Support Vector Machine, Decision Tree, Neural network, K-Nearest Neighbors, XGBoosting and Logistic Regression. Random Forest seemed to perform better than others in same study. The US FDA DILI rank database was the main source of drugs and the AI models integrated diverse features as physicochemical properties, molecular fingerprints and genes expression. The diversity of training data ensures that the training data can provide more discriminating information for the model. Voting combinations-based ensemble is one of innovative strategy used for improving accuracy. Artificial intelligence remains a tool of choice for risk assessment and a definite ally in drug research and development.