Multi-agent simulated annealing algorithm with parallel adaptive multiple sampling for protein structure prediction in AB off-lattice model

Abstract

Protein structure prediction (PSP) with ab initio model keeps a challenge in bioinformatics on account of high computational complexity. To solve the problem within a limited time and resource, the parallel capacity and search efficiency are of significance for a successful algorithm. Traditional simulated annealing (SA) algorithm is extremely slow in convergence, and the implementation and efficiency of parallel SA algorithms are typically problem-dependent. To overcome such intrinsic limitation, in this paper a multi-agent simulated annealing (MASA) algorithm with parallel adaptive multiple sampling (MASA-PAMS) that features better search ability is proposed. The MASA-PAMS contains two main issues. First, a parallel elitist sampling strategy overcomes the inherent serialization of the original SA, provides benefit information for the iteration which is helpful for the convergence. Then an adaptive neighborhood search and a parallel multiple move mechanism displace the random sampling scheme which balance the intensification and diversification iteratively. Conducted experiments with 2D and 3D AB off-lattice models indicate that the MASA-PAMS performs better than, or at least comparable to other MASAs with different sampling schemes and several state-of-the-art algorithms for PSP.