Introduction

Docking/scoring approaches are widely used now in virtual screening experiments. Large databases can be searched for suitable lead compounds that are active against a macromolecular target if the 3D structure of the target (usually an enzyme or receptor) is known [1-4]. While the fast sampling of possible binding modes can be accomplished very efficiently by various docking programs [5-12], the identification of the correct binding mode and the ranking of different ligands according to their binding affinities to the protein target remain the Achilles' heel of docking/scoring approaches [13].

The fast ranking of binding modes of putative protein-ligand complexes is accomplished by using so called scoring functions [14,15]. There are three main types ofscoring functions that are implemented in docking/scoring programs today. Force field scoring functions have been used for more than a decade. They rely on calculated non-bonded interaction terms between protein and ligand atoms based on standard force fields such as AMBER [16,17]. Additional solvation or entropy terms are sometimes considered [18] and chemical as well as contact scores are used [8]. Recent improvements include the introduction of a generalized Born treatment for long range electrostatic interactions [19]. Empirical scoring functions, introduced a few years ago [20-23], use multivariate regression methods to fit a set of physically motivated parameters like hydrogen bonding energy, lipophilicity, ion pair interactions, entropic contributions, and solvation contributions for a training set of protein-ligand complexes of known 3D structure to measured binding constants. Several scoring functions have been introduced very recently that are partially or completely knowledge-based [24-29]. They use the sum ofpoten-tials of mean force (PMF) between protein and ligand atoms derived from the Brookhaven Protein Data Bank [30] (PDB) as a measure for protein-ligand binding affinity.

Knowledge-based scoring functions have been successfully used in docking studies of different protein targets and have shown some improvement over empirical and force field-based scoring functions in predicting correct binding modes and in ranking putative protein-ligand complexes [4,24-26, 31,32]. Although there is no direct theoretical rationale of linking knowledge-based potentials, as used in scoring functions, to binding free energies of protein-ligand complexes [33], the results of these docking studies are very encouraging [4,31,32]. One of the most important aspects of knowledge-based scoring functions is the introduction of an appropriate reference state to derive meaningful PMF. Since there is no unique concept of how this reference state should be designed, we attempt in this perspective to compare three possible reference states, as well as the effect of different radii of the reference sphere on the predictive power of a knowledge-based scoring function. Interpretations of these reference states as well as their effect on ranking diverse protein-ligand complexes taken from the PDB, according to their binding affinities, are discussed based on our recently introduced PMF scoring function [27].

0 0

Post a comment