PhD Thesis
Developing Computational Methods in Proximity Pharmacology for Enzyme Discovery, PROTAC Screening, and Conformational Space Exploration
Traditional drug discovery focuses on small molecules that target single proteins. Molecular glues, discovered in the 1990s while studying natural products like cyclosporin, induce de novo protein-protein interaction between proteins. Later, heterobifunctional compounds consisting of two ligands connected by a linker demonstrated that E3 ligases could be chemically recruited to target proteins, inducing de novo protein-protein interactions and leading to target degradation. These compounds, called Proteolysis Targeting Chimeras (PROTACs), have shown clinical promise, with two currently in Phase 3 trials. Beyond degradation, the field of proximity pharmacology explores the recruitment of other enzymes, such as kinases, to induce post- translational modifications. While computational methods have accelerated small molecule discovery and optimization, heterobifunctional compounds pose unique challenges due to their larger size and multi-protein interactions. Therefore, this thesis focuses on assessing existing computational approaches and developing new methods and resources for heterobifunctional molecule design and optimization. Firstly, I analyze a public protein structure repository to identify novel non-catalytic pockets in enzymes that perform post-translational modifications to uncover novel druggable sites for proximity pharmacology. Next, I evaluate computational tools for predicting PROTAC-ligase-target ternary complex configurations and screening PROTAC libraries. I show that incorporating structural dynamics can be crucial for accurate PROTAC complex modeling, which current computational tools do not implement. Therefore, I seek to explore the conformational dynamics of these systems. However, current adaptive sampling methods designed for rapid exploration of conformational landscapes are poorly adapted to high- dimensional and complex landscapes. To address these limitations, in the two final pieces of work, I develop new computational strategies and methods: (1) a new adaptive sampling strategy to accelerate the exploration of complex landscapes; and (2) a feature reduction strategy using convolutional neural networks to reduce conformational space to a dimensionality that could be explored effectively. This thesis advances the field by identifying new opportunities in proximity pharmacology, critically evaluating existing computational tools, and developing innovative algorithms to potentially enhance the modeling of ternary complexes. These contributions may aid in the rational design of heterobifunctional molecules and facilitate a deeper understanding of their structural dynamics, ultimately aiding future drug discovery efforts.