MASTERS: A General Sequence-based MultiAgent System for Protein TERtiary Structure Prediction

Abstract

As of May 12, 2014, there are approximately 39 million of non-redundant (nr) protein sequences in the NCBI nr database. The Protein Data Bank just now surpassed 100,000 three-dimensional (3-D) protein structures representing 1,393 different SCOP protein folds. Clearly, there is a huge gap between our capacities to generate protein sequences and to determine their experimental 3D structures. Structural bioinformatics, which addresses the problem of how a protein attains its 3-D structure starting only from its amino acid sequence, can reduce this gap. This is the so called the Protein Structure Prediction (PSP) problem. Thermodynamics considerations presented by Christian Anfinsen and co-workers in 1973 stated that a protein native structure is the one that minimizes its global free energy. Hence, we can treat the PSP problem as a minimization one within an NP-complete class of computational complexity. Several techniques have been proposed to predict the 3-D structure of proteins. In this work, we supplement these techniques by adding artificial intelligence concepts still not well explored in this scenario. More specifically, to address the PSP problem, we propose an ab initio-based framework for a cooperative hierarchical multiagent system guided by combined Simulated Annealing/Monte Carlo simulations. The framework was implemented in Netlogo, a widely used multiagent platform. MASTERS' main idea is to provide the user with the freedom to choose both the abstraction level and the energy function/force field model to perform the simulation. To demonstrate a typical MASTERS' application, we present a simple construct to the PSP problem, analyze its behavior and compare the results obtained with state-of-the-art optimization methods for equivalent coarse-grained abstractions.