UC Irvine

Advancements in computational methods have significantly impacted the field of drug discovery by enabling the exploration of complex molecular interactions and the design of diverse chemical libraries. This dissertation presents a study into various computational approaches aimed at enhancing the efficiency and efficacy of early stage drug discovery. In Chapter 2, we explore how machine learning methods can be used to more efficiently select compounds within large chemical databases. We demonstrate how active learning approaches identify promising drug candidates with reduced computational cost and how machine learning (ML) models can be used to filter large chemical databases. We then shift our focus to characterizing discrete binding conformations of T4 L99A using Markov state models (MSMs) in Chapter 3. Using MSMs, we characterize the dynamic behavior of protein-ligand interactions and provide insights into the binding mechanisms crucial to rational drug design that need to be addressed in future studies. Chapters 4 and 5 delve into strategies for building block selection in DNA-encoded library (DEL) design. Leveraging building block-centric approaches, we provide guidelines to construct libraries under specific design constraints and develop predictive models to inform prior additional computational and experimental follow-up. Collectively, we discuss a diverse set of computational techniques which we hope will lead to more efficient and effective strategies for drug design and library construction in the future.