Machine Learning and Artificial intelligence in drug discovery

Machine Learning and Artificial intelligence in drug discovery

Alexander Kötter*, Frankfurt/Germany

*BioAIM, Digital R&D Large Molecule Research, Sanofi-Aventis Deutschland GmbH, Industriepark Höchst, D-65926 Frankfurt am Main, Germany

Preclinical drug discovery from initial target identification to final preclinical toxicology studies is a years-long process and a multi-million-dollar investment. Designing a good clinical candidate molecule is a highly complex task where many parameters need to be optimized simultaneously (e.g., target activity, toxicological parameters, and physico-chemical properties of the molecule). The difficulty of this endeavor manifests in the 90 % failure rate in the clinic, e.g., because of poor clinical efficacy or unmanageable toxicity. Artificial Intelligence and Machine Learning (AI/ML) are poised to increase the efficiency of this process across modalities (e.g., small molecules, biologics like antibodies) by leveraging the wealth of data generated by public research organizations and the pharmaceutical industry: large language models (LLMs) trained on hundreds of millions of sequence data generate numerical representations of proteins that can then be used to train downstream models; graph neural networks directly operate on the molecular graph of small molecules and eliminate the need of handcrafted molecular features; generative AI designs molecules by itself and active learning autonomously selects highly informative molecules for experimental testing. In this talk I will share insights on how these technologies are used at Sanofi and the impact they are having on drug discovery projects.

*A.K. is an employee of Sanofi and may hold shares and/or stock options in the company.