SGER: Hidden Markov models for three-dimensional packing arrangements in proteins

Rensselaer Polytechnic Institute, Troy NY

Investigators

Abstract

EIA-0229454 Bystroff, Christopher Rensselaer Polytechnic Institute SGER: Hidden Markov Models for Three-Dimensional Packing Arrangements in Proteins Our goal is to design a hidden Markov model (HMM) that captures the sequence dependence of non-local interactions in proteins. The model will focus on recurrent substructures that are known to exist in proteins but whose sequence similarity is undetectable by current methods. These recurrent packing arrangements, or cores, are encoded in the spatial relationships between the HMM states, which represent individual residues. New algorithms will allow the prediction of the protein chain's self-avoiding path through the 3D states. The approach tests the topology-driven hypothesis for protein folding, and if successful it will represent an advance in our ability to predict protein structure from sequence. The approach is broadly applicable to systems that can be described in two spaces, sequential/temporal and spatial, and for which (1) a large database of known examples exists, and (2) a common underlying principle led to all of the examples. Applications include weather prediction, brain function, ecosystems and natural languages.