From Computational Algorithm Development to GPCR Drug Discovery: Structure-based Ultra-large Library Screening and Beyond

Abstract

This dissertation advances structure-based computer-aided drug discovery (SB-CADD) with a particular focus on G protein-coupled receptor (GPCR)–targeted therapeutics. The work centers on the development and application of Rosetta Automated Monte Carlo Reaction-based Ligand Design (RosettaAMRLD), a novel de novo drug design algorithm that integrates Monte Carlo Metropolis optimization with reaction-driven and similarity-guided ligand generation. By leveraging combinatorial ultra-large libraries, RosettaAMRLD enables efficient exploration of chemical space while ensuring synthetic accessibility, overcoming a major limitation of previous approaches. Benchmarking across diverse protein targets demonstrates the method’s ability to generate novel, synthetically feasible molecules with active-like binding modes, improved docking scores, and enhanced scaffold diversity. Building upon this methodological foundation, the dissertation applies RosettaAMRLD to the discovery of protease-activated receptor 4 (PAR4) antagonists, an important target in thrombosis and hemostasis. A complete virtual screening pipeline is established, combining RosettaAMRLD with complementary approaches such as REvoLd and DOCK ultra-large library screening, to identify and optimize structurally diverse antagonists. Computational modeling further illuminates key binding interactions and provides insights for prioritizing compounds for experimental validation. Beyond ligand discovery, this work extends computational methods to study GPCR signaling mechanisms. Using AlphaFold2-predicted active conformations, structural models of PAR4 bound to Gαq and Gα13 proteins reveal divergent interaction networks that underlie selective G protein coupling. In parallel, comparative docking of pharmacochaperone candidates in wild-type and mutant rhodopsin variants provides mechanistic explanations for mutation-specific rescue effects, informing therapeutic strategies for inherited retinal diseases. Together, these studies contribute new algorithms, workflows, and biological insights that enhance the utility of SB-CADD in drug discovery. By integrating de novo design, ultra-large library screening, and structural modeling of protein-ligand complexes, this dissertation demonstrates how computational tools can accelerate the discovery of novel therapeutics and deepen mechanistic understanding of GPCR function.