A novel approach to de novo protein structure prediction using knowledge based energy functions and experimental restraints

Abstract

The experimental elucidation of protein structures is one of the central tasks of structural biology. The structure of a protein defines its function within a cell helping us to understand the biology of organisms. Protein structure can be determined with experimental techniques, e.g. x-ray crystallography or nuclear magnetic resonance spectroscopy, to atomic resolution. Despite the success of these techniques, some proteins of interest, including membrane and large proteins, are difficult to work with. Computational structure prediction in combination with sparse experimental data like electron density maps from cryo electron microscopy can help to develop structural models for those difficult protein targets. A new de novo protein structure prediction algorithm is developed that assumes that the majority of stabilizing interaction within a protein structure is dominated by the topology defined by secondary structure elements (SSEs). Knowledge based potentials derived using Bayesian theory discriminate native like protein structures from random models. Data from experiments resulting in sparse data can be added as restraints to the energy potential to evaluate models for their agreement. Models are generated in a novel Monte-Carlo/Metropolis SSE assembly algorithm. It was shown that the energy potentials can enrich for native like protein structures in sets of protein models. Additionally, the assembly algorithm is capable of sampling structures of proteins, recovering the correct topologies and some native interactions.