176. Generative Models for Drug Discovery: From Pharmacogenomics to Targeted Molecular Design
Joel P. Arrais, Professor at University of Coimbra, Department of Informatics Engineering, Portugal
Abstract:
The availability of large-scale multi-omics data, together with advances in artificial intelligence and computing hardware, is transforming biomedical research toward hypothesis-driven drug discovery. This talk presents an integrated perspective on the use of deep learning and generative models across the drug discovery pipeline, combining pharmacogenomics, temporal regulatory modelling, and targeted molecular design.
We discuss time-dependent transcriptomic approaches for identifying disease-specific regulatory pathways and therapeutic targets, followed by deep learning models for drug–target interaction prediction using raw protein sequences and molecular representations. We then focus on generative models based on reinforcement learning and transformer architectures, enabling de novo molecule generation guided by multi-objective reward functions that encode biological affinity, pharmacokinetic constraints, and drug-likeness.
Through representative case studies, we illustrate how generative AI models act as in silico hypothesis engines, accelerating candidate identification and optimisation. The talk concludes with perspectives on future challenges and the role of integrative generative models in advancing more efficient and precise drug development.
About the author:
Joel P. Arrais, Professor at University of Coimbra, Department of Informatics Engineering, Portugal