In Silico ADMET Profiling: Evolution from Traditional Models to Deep Learning Techniques

A key element in the early stages of drug discovery and development is the accurate prediction of Absorption, Distribution, Metabolism, Excretion, and Toxicity properties. Traditional experimental methods for ADMET profiling are costly, time-consuming, and often limited in scalability, leading to high rates of clinical trial failures due to poor toxicological or pharmacokinetic profiles. Machine learning (ML) has emerged as a powerful tool for modeling complex nonlinear relationships between molecular structure and ADMET behavior, due to the increasing availability of chemical and biological data. Looking ahead, integrating explainable AI, transfer learning, and active learning methods offers the potential to address data scarcity and improve model interpretability and regulatory acceptance. As the field progresses, machine learning-based ADMET prediction is expected to become increasingly reliable, helping to streamline drug development pipelines and facilitate the creation of safer, more effective therapies. The purpose of this study is to examine and assess the state-of-the-art machine learning methods for ADMET prediction. Machine learning algorithms, such as support vector machines, gradient boosting algorithms, random forests, graph neural networks, and other software used to forecast all properties, will be examined. To create trustworthy predictive models, we stress the significance of molecular representations, dataset quality, and model evaluation techniques. Recent developments in multi-task learning and data-driven feature building, which improve prediction accuracy across several ADMET endpoints, are given particular emphasis. The web-based tools that we offer in this chapter, together with the hyperlinks that go with them, make it easier to perform In-silico ADMET research on a variety of drug candidates.