DATE: Thursday, November 30, 2006
LOCATION: E40-298
TIME: 4:15pm
Reception immediately following in the Philip M. Morse Reading Room, E40-106
ABSTRACT
I shall present our recent work motivated by a fundamental and
challenging problem in computational structural biology. Protein-protein
interactions play a central role in metabolic control, signal
transduction, and gene regulation. Determining the 3-dimensional
(3D) structure of a complex from the atomic coordinates of two
interacting proteins (the receptor and the ligand) is known as
the protein-protein docking problem. Experimental techniques
can provide such 3D structures but are time-consuming, expensive,
and not universally applicable. As a result, solving these problems
computationally is critical and has attracted a lot of attention.
Nature being efficient, protein-protein docking can be formulated
as a the problem of minimizing the Gibbs free energy of the complex.
Optimization is performed over translations (R^3) and rotations
(the rotation group SO(3)) of the ligand with respect to the
receptor as well as over conformational changes (especially side-chains
at the interface). However, the free-energy functional is very
complex having multiple deep funnels and a huge number of local
minima of less depth that are spread over the domain of the function.
We present a systematic way of performing this optimization.
A key tool to that end is a new stochastic global optimization
method we have developed, the so called Semi-Definite programming
based Underestimation (SDU) method. The method is based on finding
general convex underestimators to the binding energy function
that is funnel-like. The underestimator is used to bias sampling
in the search region. We provide probabilistic convergence guarantees,
comparisons with related work, and an array of computational
results illustrating our approach.
