On the Physical Relevance and Statistical Interpretation of Knowledge-Based Potentials

Abstract

Knowledge based potentials (KBP) are energy functions that are obtained from databases of known protein structures, rather than by physical or chemical analysis. While KBPs have been successfully used in protein structure prediction, their physical interpretation remains unclear. Often KBPs are loosely justified as a type of inverse Boltzmann construction or by analogies to the construction of the potential of mean force (PMF) in statistical mechanics. The chapter provides an overview of various types of KBPs and discusses conceptual differences with the construction and use of PMFs in statistical mechanics, with particular emphasis on the key role played by the reference state. Although essential to the construction of both types of potentials, its precise definition has been elusive in the context of KBPs. We discuss the construction of the reference state of KBPs both in the traditional ‘energy’-oriented approach as well as in a probabilistic framework. In particular, we show by simple Bayesian reasoning that the reference state in fact is uniquely defined by the choice of statistics and mode of application. This view provides a statistical rigorous basis for various possible extensions and generalizations of KBPs.