Dissertation Title: 

AI-Driven Protein Design: Predictive Modeling, Generative Design, and Agentic Workflows

Abstract: 

Foundational models for protein sequence and structure have substantially enhanced our ability to engineer proteins with novel function, enabling applications in therapeutics, industrial enzyme engineering, and synthetic biology. This thesis addresses a central challenge in the development of novel protein therapeutics through the application of machine learning methods to model determinants of immunogenicity in silico. Immune response to de novo proteins arises from a cascade of molecular interactions along the antigen-processing pathway. Depending on biological context and mode of delivery, these responses may be mediated through diverse biological processes, including MHC Class I, MHC Class II, and alternative antigen-processing pathways. However, MHC polymorphism and substantial variation in data availability across antigen-processing stages make accurate prediction particularly challenging. This thesis addresses these limitations through a unified computational toolkit that pools information across stages while incorporating uncertainty estimation to account for variation in data availability.

Beyond immunogenicity modeling, this thesis further explores generative machine learning methods for protein design across a range of biological contexts, including workflows that directly incorporate immunogenicity models as design objectives. Additionally, with LLMs emerging as powerful tools across all stages of scientific inquiry, this thesis develops and evaluates agentic systems for protein modeling and design, illustrating how LLM-based agents can support multimodal workflows across the protein engineering pipeline. Taken together, this thesis reflects a broader shift in biological research toward AI-driven workflows, where foundation models, generative models, and LLM-based agents are increasingly positioned to accelerate the design and experimental realization of proteins with novel function.

Joseph M. Jacobson
Associate Professor, Media Arts and Sciences, MIT

Edward S. Boyden
Y. Eva Tan Professor in Neurotechnology, McGovern Institute and HHMI

Andrew D. White
Chief Technology Officer, FutureHouse & Edison Scientific

Pranam Chatterjee
Assistant Professor of Bioengineering, University of Pennsylvania