IIT Madras, Ohio State University develop AI framework to aid drug discovery

Researchers from IIT Madras’ Wadhwani School of Data Science and Artificial Intelligence (WSAI) and The Ohio State University have developed an AI framework that can generate drug-like molecules which are easier to synthesise in real laboratory settings.

The model, called PURE (policy-guided unbiased representations for structure-constrained molecular generation), aims to reduce the early-stage drug discovery process, which typically costs billions of dollars and can take a decade or more. It could be particularly useful in tackling drug resistance in cancer and infectious diseases, said IIT Madras in a press release issued on Monday.

Srinivasan Parthasarathy of The Ohio State University said the model could accelerate the search for alternative drug candidates, especially in cases of resistance or toxicity, and support discovery in new materials research.


Unlike existing AI tools that rely on pre-defined scoring or optimisation metrics, PURE uses reinforcement learning to simulate how molecules transform through real chemical reactions. This allows it to generate novel, diverse and synthetically viable molecules without being explicitly trained on those evaluation parameters. It also identified plausible synthetic routes for its generated molecules, making it a general-purpose molecular discovery engine, the institute said in the press release.

B Ravindran, head of WSAI, said the framework treats “chemical design as a sequence of actions guided by real reaction rules”, enabling AI systems to reason through synthesis, much like a chemist.

Karthik Raman, also from WSAI, added that PURE’s reaction rule-based approach “grounds molecule generation in synthesisability”, addressing a key challenge in computational drug design.

PURE’s approach—blending self-supervised and reinforcement learning—helps overcome a persistent limitation of AI-driven drug design, where many virtual molecules are not lab-synthesisable. By linking digital discovery to real chemical synthesis, the researchers say, the model could help compress development timelines and improve the success rate of early-stage drug candidates.