A new method for protein structure reconstruction from NOESY distances

TL;DR

This paper introduces an efficient protein structure reconstruction method from NMR NOESY distances using the ScaledASD low-rank matrix completion algorithm, enhanced with post-processing steps, showing competitive accuracy with existing methods.

Contribution

The paper applies the ScaledASD low-rank matrix completion algorithm to protein structure reconstruction from NMR data, incorporating multiple refinement procedures for improved accuracy.

Findings

Method achieves results consistent with PDB structures.

Higher validity in Procheck dihedral angles G-factor.

Comparable or improved structural similarity to X-ray structures.

Abstract

Protein structure reconstruction from Nuclear Magnetic Resonance (NMR) experiments largely relies on computational algorithms. Recently, some effective low-rank matrix completion (MC) methods, such as ASD and ScaledASD, have been successfully applied to image processing, which inspires us to apply the methods to reconstruct protein structures. In this paper, we present an efficient method to determine protein structures based on experimental NMR NOESY distances. ScaledASD algorithm is used in the method with several post-procedures including chirality refinement, distance lower (upper) bound refinement, force field-based energy minimization (EM) and water refinement. By comparing several metrics in the conformation evaluation on our results with Protein Data Bank (PDB) structures, we conclude that our method is consistent with the popularly used methods. In particular, our results show higher validities in Procheck dihedral angles G-factor. Furthermore, we compare our calculation results with PDB structures by examining the structural similarity to X-ray crystallographic structures in a special dataset. The software and its MATLAB source codes are available in https://github.com/xubiaopeng/PRASD