Srinivasan Parthasarathy from OSU emphasizes the model’s potential to speed up exploration of resistance or toxicity issues, aiding discovery in challenging therapeutic areas.

IIT Madras researchers have developed a new AI framework called PURE to generate drug-like molecules that are easier to synthesize in real laboratories.

PURE, which stands for Policy-guided Unbiased Representations for Structure-Constrained Molecular Generation, uses reinforcement learning to model chemical synthesis steps rather than relying on rigid metric optimization, aiming for diverse, novel candidates with viable synthetic routes.

The findings were published in the open-access Journal of Cheminformatics, highlighting its contribution to computational chemistry and drug discovery.

Researchers from IIT Madras and The Ohio State University note the approach could reduce discovery bias and improve synthesis feasibility, with the study published under DOI 10.1186/s13321-025-01090-5.

The framework was evaluated on standard molecule-generation benchmarks (QED, DRD2, solubility) and demonstrated higher diversity and novelty while proposing viable synthesis routes without training on those specific metrics.

The work highlights the integration of self-supervised learning with policy-based reinforcement learning and template-driven molecular simulations within PURE.

Beyond accelerating drug development, PURE could help identify alternative and potentially more effective candidates, with potential applications in materials discovery as well.

The research team comprises IIT Madras researchers Abhor Gupta, Barathi Lenin, Rohit Batra, B. Ravindran, Karthik Raman, and OSU’s Srinivasan Parthasarathy and Sean Current, with the study detailing PURE’s capabilities.

PURE treats chemical design as a sequence of actions guided by real reaction rules, enabling AI to reason through synthesis similarly to a chemist.

Overall, PURE promises to compress development timelines and improve early-stage success by enabling AI to reason through synthesis steps and consider resistance or hepatotoxicity early on.

The research suggests broader implications for future research, extending the approach beyond drug discovery to material innovation and other domains.