Modeling the flexibility of alpha helices in protein interfaces : structure based design and prediction of helix-mediated protein-protein interactions

Abstract

Protein-protein interactions play an essential role in many biological functions. Prediction and design of these interactions using computational methods requires models that can be used to efficiently sample structural variation. This thesis identifies methods that can be used to sample an important sub-space of protein structure: alpha helices that participate in protein interfaces. Helices, the global structural properties of which can be described with only a few variables, are particularly well suited for efficient sampling. Two methods for sampling helical backbones are presented: Crick parameterization for coiled coils and normal-mode analysis for all helices. These are shown to capture most of the variation seen in the PDB. In addition, these methods are applied to problems in protein structure prediction and design. Normal-mode analysis is used to design novel nanomolar peptide inhibitors of the apoptosis-related Bcl-2 family member, Bcl-xL, and a modification of Crick Parameterization is used to predict the binding orientation of dimeric coiled coils with greater than 80% accuracy. Finally, this study addresses the increase in computational time required by flexible-backbone methods and the use of cluster expansion to quickly map structural energies to sequence-based functions for increased efficiency.