Researchers at the Indian Institute of Technology-Madras (IIT-M) and The Ohio State University have created an innovative artificial intelligence framework designed to facilitate the generation of drug-like molecules that can be synthesized more efficiently in laboratory settings. Named Policy-guided Unbiased Representations for Structure Constrained Molecular Generation (PURE), this framework is poised to significantly shorten the early stages of drug development, which can often extend over decades and incur substantial costs.
According to a press release, the PURE framework could also provide solutions to the pressing issue of drug resistance in cancer treatments and infectious diseases. The team noted that one major challenge in AI-driven drug discovery is that many molecules appear promising in computational models but are nearly impossible to synthesize in practice. PURE addresses this by grounding its molecular generation in realistic synthesis pathways. It autonomously learns chemical similarities without relying on potentially biased metrics and proposes feasible synthetic routes along with corresponding molecular structures.
The implications of this development could lead to faster drug pipelines and alternative solutions for treatments that are currently failing. B. Ravindran, head of the Wadhwani School of Data Science and AI at IIT Madras, emphasized that the PURE framework brings researchers closer to achieving AI systems capable of reasoning through synthesis steps in a manner reminiscent of human chemists.
Karthik Raman from WSAI highlighted that PURE”s approach involves mapping chemical space in a way that avoids biases associated with specific metrics. Srinivasan Parthasarathy from the Department of Computer Science and Engineering at Ohio State noted that this framework provides significant advantages for early-stage pharmaceutical research, particularly in identifying alternative drug candidates when faced with resistance and hepatotoxicity issues.
The findings from this collaborative research were published in the Journal for Cheminformatics, an open-access journal dedicated to original peer-reviewed studies in the field.
