Abstract
BackgroundAccurate evaluation and modelling of residue-residue interactions within and between proteins is a key aspect of computational structure prediction including homology modelling, protein-protein docking, refinement of low-resolution structures, and computational protein design.ResultsHere we introduce a method for accurate protein structure modelling and evaluation based on a novel 4-distance description of residue-residue interaction geometry. Statistical 4-distance preferences were extracted from high-resolution protein structures and were used as a basis for a knowledge-based potential, called Hunter. We demonstrate that 4-distance description of side chain interactions can be used reliably to discriminate the native structure from a set of decoys. Hunter ranked the native structure as the top one in 217 out of 220 high-resolution decoy sets, in 25 out of 28 "Decoys 'R' Us" decoy sets and in 24 out of 27 high-resolution CASP7/8 decoy sets. The same concept was applied to side chain modelling in protein structures. On a set of very high-resolution protein structures the average RMSD was 1.47 Å for all residues and 0.73 Å for buried residues, which is in the range of attainable accuracy for a model. Finally, we show that Hunter performs as good or better than other top methods in homology modelling based on results from the CASP7 experiment. The supporting web site http://bioinfo.weizmann.ac.il/hunter/ was developed to enable the use of Hunter and for visualization and interactive exploration of 4-distance distributions.ConclusionsOur results suggest that Hunter can be used as a tool for evaluation and for accurate modelling of residue-residue interactions in protein structures. The same methodology is applicable to other areas involving high-resolution modelling of biomolecules.
Highlights
Accurate evaluation and modelling of residue-residue interactions within and between proteins is a key aspect of computational structure prediction including homology modelling, protein-protein docking, refinement of low-resolution structures, and computational protein design
We began by defining pairs of atoms, among which four distances were calculated, for each of 190 side chain-side chain (ScSc) and 18 side chainmain chain (ScMc) residue-residue interactions
We found that the EScMc term by itself did not perform well though combining it with EScSc term gave improvement in side chain packing (Table 2)
Summary
Accurate evaluation and modelling of residue-residue interactions within and between proteins is a key aspect of computational structure prediction including homology modelling, protein-protein docking, refinement of low-resolution structures, and computational protein design. All proteins share the same backbone, with their structure and function determined solely by the side chains of the 20 different amino acids. Computational methods rely on a potential function to evaluate the interactions within proteins [1]. Two types of the potential functions currently exist: physics-based [2,3,4] and knowledge-based [5]. The former rely on the basic physical principles to describe the forces that drive structure and function of proteins. Knowledge-based potentials (KBP) derive statistical preferences on different features from structural and sequence databases and implicitly capture the many factors affecting the protein in its natural environment. Knowledge-based potentials have numerous applications and were successfully used in threading [12], validation of experimentally determined protein structures [13,14], ab initio structure prediction [15,16], decoy discrimination [17,18,19] and more
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
