Development of computational approaches for the design of antibodies and other binding proteins for target epitopes

Abstract

Antibodies are the most studied and successful therapeutic binding protein. Alternative, smaller-sized binding proteins are being modified for therapeutic applications. Computational energy models i.e. forcefields, protein analysis and design tools have been developed to aid in the discovery and engineering of antibodies and other binding proteins. Current binding protein design tools employ iterative cycles of energy calculations and optimizations, which are computationally expensive. In this work, I have gained a deeper understanding of features of binding interface residues and make an attempt towards developing novel computational design approaches that use sequential forcefield-independent steps. The dissertation begins with the development and energetic analysis of a database of non-redundant antibody-antigen complexes. Building upon this analysis, we developed AUBIE, a novel computational tool for the de novo design of binding proteins like antibodies for any target epitope. AUBIE identifies groups of compatible binding loops that can simultaneously make strong interactions with the target epitope. AUBIE-generated antibodies for a HER2 epitope were tested experimentally. In order to get a better understanding of flexible nature of binding surfaces, features related to the conformational stability of binding epitopes and paratopes were studied. Learning from these features, improvements were made to the AUBIE approach to better replicate epitope stabilities in AUBIE designed antibodies. We then developed a novel MutDock, docking approach that can simultaneously perform mutations for fixed backbone binding protein scaffolds for any target epitope. MutDock was benchmarked against commonly used fixed-sequence docking tools. The dissertation will provide in-depth details about the methods and results regarding database generation and analysis, AUBIE and MutDock.