Prediction of Low Energy Protein Side Chain Configurations Using Markov Random Fields
The task of predicting energetically favorable amino acid side chain configurations, given the three-dimensional structure of a protein main chain, is a fundamental subproblem in computational structural biology. Specifically, it is a key component in many protocols for de novo protein folding, homology modeling, and protein-protein docking. In addition, fixed main chain protein design can be cast as a generalized version of the side chain placement problem. For all these purposes, the objective of pursuing low energy side chain configurations is equivalent to finding the most probable assignments of a corresponding Markov random field. Consequently, this problem can be addressed using message-passing probabilistic inference algorithms, such as max-product belief propagation (BP) and its variants. In this chapter, we review the inference techniques that have been successfully applied to side chain placement, discuss their current limitations, and outline promising directions for future improvements.
KeywordsProtein Design Belief Propagation Markov Random Field Linear Programming Relaxation Junction Tree
We would like to thank Amir Globerson and Talya Meltzer for their discussions on message-passing algorithms with certificates of optimality.