Seminario Datalab
Large Language Models and Reinforcement Learning for de Novo Molecular Design
Ponente: Pablo Varas (ICMAT y Aitenea Biotech)Fecha: lunes 17 de marzo de 2025 - 12:30Lugar: Sala 310, Módulo 0, Facultad de Ciencias, UAM
Resumen:
This talk focuses on using Large Language Models (LLMs) combined with Reinforcement Learning (RL) to generate novel molecules with specific desired properties. Traditional pocket-based generative models often produce molecules that cannot be synthesized and do not effectively optimize for multiple pockets. Our approach seeks to address these limitations by utilizing LLMs trained on extensive molecular datasets and refining their outputs through RL with a customized reward function. Additionally, this method allows for the accurate prediction of molecular properties, thereby enhancing the overall design and evaluation of the generated compounds.
