Aligning Machine Learning Methods with Drug Discovery Objectives

Machine learning is increasingly shaping drug discovery, with computational methods addressing challenges in molecular design, synthesis planning, and optimization. In this talk, I will present recent work on how computational choices influence discovery outcomes, from retrosynthesis to generative chemistry. I will first examine how different combinations of single-step and multi-step retrosynthesis models guide exploration of retrosynthetic space and can enable convergent synthesis planning. I will then discuss de-novo molecular design, highlighting how multi objective optimization strategies shape molecular exploration in many objective design settings. This talk underscores the importance of aligning algorithmic decisions with practical discovery objectives and explores opportunities for integrating data driven methods into medicinal chemistry workflows.

About the speaker
Paula Torren Peraire is a Data Science Innovation Fellow at Novartis in Basel, where she works on machine learning driven methods for drug discovery, with a focus on generative chemistry. She studied Pharmacy at the University of Barcelona, followed by a Master’s in Bioinformatics for Health Sciences at Universitat Pompeu Fabra. Paula received her PhD from the Technical University of Munich with a joint project between Helmholtz Munich and Johnson & Johnson, focusing on computer-aided synthesis planning. Overall, her work aims to develop robust computational frameworks that support molecular design and decision-making in drug discovery, bridging methodological innovation with real-world pharmaceutical applications.