Abstract
The development of artificial intelligence (AI) tools and technology has made AI-driven drug discovery a more prominent field. We are firmly in the AI era, with hybrid designs that eventually comprise deep learning (DL) and conventional machine learning (ML). Although traditional models can predict ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties, they remain relatively unsuccessful, and improving the accuracy of predictions remains challenging. Recently, several researchers have developed a hybrid learning model that successfully addresses these problems and improves prediction accuracy. The systematic tendencies facing AI-powered transformation from conventional DL and ML to hybrid learning AI models are examined in this review. Compared with traditional ML and DL, hybrid AI models have increased efficiency by reducing drug development time and costs, and improved success rates. In this context, the ongoing development of new ADMET software based on hybrid AI and multimodeling techniques can enhance the accuracy of pharmacokinetic-pharmacodynamic predictions, improve ADMET endpoint predictions, and expedite the drug discovery of new chemical entities. Moreover, this review covers the future of AI in pharmaceutical sciences and ADMET predictions, including AI-driven prediction models that range from basic ML/DL to newly developed hybrid models, evaluation parameters, and their applications in ADMET property prediction. SIGNIFICANCE STATEMENT: The article covers the compilation of ongoing research in the development of ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) software based on hybrid artificial intelligence and multimodeling techniques, which may increase the accuracy of pharmacokinetic-pharmacodynamic predictions, improve ADMET endpoint predictions, and accelerate drug discovery.