Structure based stabilization of native-like antigens with deep learning

Abstract

Effective vaccines prevent illness and death by stimulating protective immune responses against infectious pathogens. Protective immune responses can recognize and neutralize antigens that are important proteins used by pathogens to infect and replicate. However, these key proteins often evolve multiple conformations or instability to infect and evade the host immune system. Structure-guided design has advanced vaccine development by introducing mutations that stabilize these proteins to elicit stronger immune responses. Deep learning based methods have transformed our ability to predict and design protein structures. In this work, I investigated how to best apply these novel protein design methods to the stabilization of native-like antigens. I focused on four pathogens that cause significant morbidity and mortality: Human Rotavirus, Group A Streptococcus, Mycobacterium tuberculosis, and Rabies virus. For each pathogen, I chose key proteins that are compelling vaccine targets and present distinct challenges for antigen stabilization. By pairing existing methods with novel deep learning based tools, I identified mutations that improve antigen stability and immunogenicity. In the process, I have identified principles which may be applicable for structure guided design across a wide range of antigens.