Characterizing Protein Surfaces for Binding Associations

Abstract

Proteins interact with one another and other biomolecules to carry out their function. These interactions are mediated via residues on surfaces of proteins and other biomolecules. In this thesis, we studied both protein-protein and protein-small molecule interfaces physico-chemically and structurally. We characterized the different physico-chemical environments in proteins using depth. We calculated the effect of depth on residue substitutions and created depth dependent amino acid substitution matrices. These matrices were then successfully used to predict deleterious mutations in proteins. We further studied residue environments at protein interfaces to examine how different interface residues transition from one depth level in a monomer to another in a complex. Utilizing this knowledge, we developed a depth based scoring potential for protein-protein interfaces which was used to distinguish near-native interfaces from non-native interfaces.

We created a database of protein-protein/domain-domain interfaces in the PDB and clustered them by geometric similarity. While this library can be used to model protein complex of varying geometry, in this thesis we explored coiled-coil interfaces. We built a machine learning based algorithm to predict coiled-coils that will interact with each other. The algorithm was used to predict the interactions between the coiled-coiled domains of JC virus Agno protein with Rab11B and p53. Along with studying the interfaces, we also wanted to identify the hotspot residues which were important for mediating the interactions. We built a decision tree based classifier to predict these residues and our method was shown to be comparable if not better to other state of the art methods.

We designed a general structure based methodology to identify binding pockets of small molecules on proteins. We tested this methodology in identifying the alternate binding partners of a small molecule drug Nutlin, which is known to bind Mdm2. Our predictions were validated computationally and experimentally. We further extended this to identify alternate drug binding targets for several other known drugs. We also designed/predicted inhibitors against the different proteins of Nipah virus and computationally analyzed their stability. All these inhibitors will plausibly be effective against the different strains of Nipah virus.

This thesis captures various structural and physico-chemical aspects of protein-protein and protein-small molecule interfaces which can be used for scoring protein-protein interfaces, modeling protein complexes, identification of coiled-coiled interfaces, prediction of hotspot residues, predicting off target effects of drugs and other applications.