PhD Oral Proposal: Shruthi Viswanath, Nov. 2, 2012, 4 p.m. CST, ACES 6.336

Contact Name: 
Lydia Griffith
Nov 2, 2012 4:00pm - 5:00pm

PhD Oral Proposal: Shruthi Viswanath

Date: Nov 2nd, 2012
Time: 4 p.m.
Place: ACES 6.336

Title of Dissertation: Scoring Functions for Protein Docking and Drug Design
Advisor: Chandrajit Bajaj and Ron Elber

Proteins perform their function by binding to other proteins or small molecules (ligands). But protein complexes are difficult to obtain through experimental methods, and hence computational methods play an important role in predicting structures of complexes, understanding protein interactions involved in biochemical pathways and potentially designing drugs. In this dissertation, we develop new approaches for improving computational structure prediction of protein complexes.

The first part deals with methods to improve prediction of protein-protein complexes. Also known as protein-protein docking, this problem involves predicting the structure of the complex formed by two proteins, given the three-dimensional structure of the individual protein chains. Docking algorithms consist of two stages: search and scoring. We develop a novel algorithm for the second stage of scoring. A new scoring function that includes atomic and amino acid pair-wise interaction terms is developed and used in conjunction with a side-chain refinement procedure to improve docking results. The parameters of the scoring function are learnt using linear programming. The docking program that includes the novel refinement and scoring algorithm compares favorably to other leading docking packages.

The second part involves the development of a novel atomic scoring function for scoring models of protein-ligand complexes. The novelty of this scoring function lies in the fact that discriminative learning is used on a large scale with exhaustive sampling of negative binding poses.  The parameters of the scoring function are learnt using quadratic programming and the function is scored on a grid with fast calculation of the scoring function using Fast Fourier Transforms. It is hoped that by exhaustive discriminative learning, the new scoring function will be able to rank near-native poses accurately.