We present a physics-based method for identifying native configurations of protein-protein interactions amongst a set of nearly native decoys (< 2.0 Å C_a RMSD to the native structure) using a fast new method for performing free energy calculations. The method uses Markov Random Fields to encode the Boltzmann distribution for a given complex, and Generalized Belief Propagation to perform the free energy calculation. Our method is fast, running in a few minutes on a single-processor workstation, making it an attractive alternative to free-energy calculations based molecular dynamics and Monte Carlo simulations, which can require hours or days on multiprocessor machines. The method is also accurate; in an experiment involving 9 targets with an average of 8 nearly native decoys, our method ranks the native structure number one 67% of the time, and in the top three for the remaining cases.

14 pages

*Department of Computer Science and Department of Biological Sciences, Carnegie Mellon University

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu