Authors: Matkar Pallavi Rajendra, Prof. R. V. Varhade, Prof. M. C. Kolse, Prof. S. K. Said
Abstract: Drug discovery is a complex, slow and expensive process that involves identifying biological targets, finding and optimizing compounds that act on them, and ensuring that the resulting candidates are effective and safe. Traditional approaches rely heavily on laboratory experiments and screening campaigns that consume years and large budgets, with high failure rates. Artificial Intelligence (AI) and Machine Learning (ML) are increasingly used to accelerate and improve this pipeline. This paper reviews AI methods across the stages of drug discovery: molecular representations such as SMILES strings, fingerprints and molecular graphs; graph neural networks and message passing for property prediction; deep models for drug–target interaction and binding affinity; prediction of toxicity and bioactivity; generative models and reinforcement learning for designing new molecules; protein structure prediction; drug repurposing; and explainable AI for interpreting predictions. Representative successes, including deep-learning-driven antibiotic discovery and rapid design of kinase inhibitors, are discussed alongside benchmark datasets. A layered AI-driven drug discovery architecture is proposed, with research gaps, objectives, a comparison of model families and a discussion of challenges such as data quality, bias, limited datasets and the need for experimental validation. The review concludes that AI can make drug discovery faster and more cost-effective when tightly coupled with laboratory testing.
