Advances in artificial intelligence for predictive toxicology: From QSAR and omics integration to clinical safety translation

The benefit of predictive toxicology strategies utilized in drug discovery to predict possible early-stage side effects. Common methods can be performed in vitro or in vivo and are lengthy, expensive, and often poorly correlated with human responses. High attrition rates in clinical trials caused by unexpected off-target toxicities have complicated drug discovery efforts and necessitate the improvement of models. Machine learning (ML), deep learning, and artificial intelligence (AI) are being used to redefine predictive toxicology. With the use of AI technologies, it is more efficient to mine big data and uncover complex associations, which is more feasible to address safety concerns and perform toxicity studies on a broader array of biological platforms. In this review, we have focused on AI, drug safety, efficacy prediction, and clinical translation. This explains how AI models are applicable to early toxicity detection, adverse drug reaction (ADR) prediction, and safer drugs. It is anticipated that the inclusion of explainable AI (XAI) may improve the transparency and credibility of the models. In addition, the rise of digital twins and in silico human models can transform regulation and reduce reliance on traditional animal studies, while still enabling personalization in drug discovery.