Scoring Functions for Predicting Structure and Binding of Proteins

This chapter first discusses the general framework of developing knowledge-based potential functions in terms of molecular descriptors, functional form, and parameter calculations. It also discusses the underlying thermodynamic hypothesis of protein folding. With the assumption that frequently observed protein features in a database of structures correspond to a low energy state, frequency of observed interactions can be converted to energy terms. The chapter then describes in detail the models behind the Miyazawa-Jernigan contact potential, distance-dependent potentials, and geometric potentials. Next, it shows how to weight sample structures of varying degree of sequence similarity in the structural database. The chapter also describes general geometric models for the problem of obtaining optimized knowledge-based potential functions, as well as methods for developing optimal linear and nonlinear potential functions. It presents several applications of the knowledge-based potential functions. Finally, the chapter provides general limitations and possible improvements for the statistical and optimized potential functions.