Computation-Driven Development of Next-Generation Viral Vaccines and Antibody Therapeutics

Abstract

Despite successful antiviral countermeasures dating back thousands of years, billions of human infections still occur annually, and pandemic viruses with high mortality rates continue to emerge, often with long wait times for effective interventions. Monoclonal antibody therapeutics and subunit vaccines represent promising technological advances that can offer two important advantages over traditional approaches: enhanced safety profiles and protection against a broader range of viral strains. This dissertation presents research advancing three aspects of next-generation antiviral countermeasures. In Chapter 2, I develop contrastive learning methodologies for antibody epitope representation, establishing both straightforward sequence-feature rules for identifying overlapping-epitope antibody pairs and two antibody language models with epitope-specific embeddings. The more generalizable model, trained across 250 antigen families, demonstrated 50% success in discovering HIV-1 antibodies with epitope overlap to the broadly neutralizing antibody 8ANC195. Chapter 3 details the characterization of one of the most broadly reactive coronavirus spike antibodies identified to date, contextualizing its binding properties and sequence features within the landscape of other cross-reactive coronavirus antibodies. Chapter 4 presents the development of linear epitope-focused influenza vaccines that result in heterosubtypic neutralization, reduced viral load in challenge models, and pan-influenza binding monoclonal antibodies targeting the fusion peptide. Collectively, this work advances our understanding of antibody-antigen interactions and provides novel computational and experimental frameworks that will contribute to the development of broadly protective viral vaccines and antibody therapeutics.