Multi-Body Knowledge-Based Potentials for Protein Structure Prediction Evaluation

Abstract

Knowledge-based potentials have been widely used in the last 20 years for fold recognition, evaluation of protein structure predictions from amino acid sequences, ligand binding, protein design, and many other purposes. The most commonly known are two-body residue-level contact potentials, especially those first introduced by Miyazawa and Jernigan in 1985, and then rederived using an updated, larger protein dataset in 1996. Dense packing of residues in globular proteins is one of their characteristic features. Because of such dense packing cooperative multi-body interactions, especially in protein cores are important. The four-body contact potentials and short-range interaction potentials have been derived by considering different aspects of protein structures than those used to derive pair-contact potentials. The four-body contact potentials are appropriate for representing the cooperative parts of the protein folding process, and we have shown that they are quite successful for recognizing the native structures among hundreds or even thousands of decoys from the Decoys‘R’Us database. Short-range interaction energies allow us to estimate free energies from the statistical distribution of local conformational descriptors. We developed two types of four-body potentials: sequential and non-sequential ones. We have found that combining the former ones with short-range interactions yields excellent results for threadings, that significantly outperforms all other methods for coarse-grained models of proteins. We have developed also our knowledge-based potential server http://gor.bb.iastate.edu/potential for coarse-grained protein energy estimations that uses two types of four-body potentials, short-range potentials, and 23 different two-body potentials.