The Miyazawa-Jernigan Contact Energies Revisited

Abstract

The Miyazawa-Jernigan (MJ) contact potential for globular proteins is a widely used knowledge-based potential. It is well known that MJ's contact energies mainly come from one-body terms. Directly in the framework of the MJ energy for a protein, we derive the one-body term based on a probabilistic model, and compare the term with several hydrophobicity scales of amino acids. This derivation is based on a set of native structures, and the only information of structures manipulated in the analysis is the contact numbers of each residue. Contact numbers strongly correlate with layers of a protein when it is viewed as an ellipsoid. Using an entropic clustering approach, we obtain two coarse-grained states by maximizing the mutual information between coordination numbers and residue types, and find their differences in the two-body correction. A contact definition using sidechain centers roughly estimated from C atoms results in no significant changes.