Refinement of Protein Structure Predicted Models Using Minimum Spanning Tree.

Abstract

The protein structure prediction is of three categories: homology modeling, fold recognition and ab initio modeling, and this division into categories depends on whether similar protein structures were previously determined using X-ray crystallography or NMR or not. Protein structure models predicted by the free modeling (ab initio modeling) are considered as low-resolution models. Progress has recently been made in refining low-resolution models (ab initio modeling) closer to the native ones; this can be done by minimizing the energy funnel of physics-based force fields. In this paper, we present a new refinement method based on applying minimum spanning tree to produce a connected graph from the atoms forming the protein. This connected graph represents the minimum van der Waals energy path. The new refinement method causes supplementary execution time (about 55.83486 sec. in the average for a protein sequence of length 166 amino acids) but enhance the predicted model of low resolution. We used a small set of 18 different targets got from CASP10. The results show the improvement in about (83.333%) of the cases. We compare our results with state-of-art algorithms results.