Multi-agent list-based noising algorithm for protein structure prediction

Abstract

Protein structure prediction (PSP) with ab initio model problem is still a challenge in bioinformatics on account of high computational complexity. To solve this problem within a limited time and resource, a multi-agent list-based nosing (MLBN) algorithm is presented. MLBN contains three main features. First, a flexible noising list is designed to adjust the solution acceptance condition according to the convergence. An adaptive multiple sampling strategy is included to provide a strong exploitation. A parallel framework explores the searching space in a more effective way. Compared to traditional Simulated Annealing (SA) algorithm, MLBN introduces only one extra parameter for the length of noising list and it is insensitive to specific problems. Conducted experiments in a range of protein sequences indicate MLBN performs better than, or at least is comparable with, several state-of-the-art algorithms for PSP.