Scoring Functions

High throughput screening identifies inhibitors of the target enzyme by measuring the change in activity of the protein kinase in the presence of the chemical compound. For an ATP-competitive inhibitor, the stronger the affinity of the molecule for the enzyme-binding site the greater the inhibition. The ultimate goal of high throughput docking is to select inhibitors from the compound dataset through prediction of both binding mode and affinity. As discussed below, the prediction of binding modes has outpaced the prediction of binding affinities. Some success has been described in the use of more computationally intensive algorithms to predict binding affinities based on known experimental data [79-81]. At present, though, there is no fast and reliable method for predicting binding affinities.

Considerable resource has therefore been devoted to the development of functions for scoring and ranking the large numbers of poses generated in virtual screening. There are three main types of scoring function; physical, empirical and knowledge-based. Physical, energy-based functions are cut-down versions of atomic forcefields, focusing on the calculation of the intermolecular component of the potential energy of the protein-ligand complex. In empirical functions, a model is generated from experimental observations of binding modes and affinities. Most models can be divided into terms that describe hydrogen bonding interactions, hydrophobic interactions, a ligand intramolecular strain energy and a desolvation term. Knowledge-based functions originate from the prediction of the tertiary structure of proteins. In docking, the functions are derived from a statistical analysis of known protein-ligand complexes. A pairwise potential of mean force is computed from the radial distribution function for each pair of atom types as observed in the protein-ligand complexes. Calculation of knowledge-based functions is fast and their development is facilitated by an ever increasing number of protein-ligand complexes, as compared to the availability of the binding affinities required to derive empirical models.

0 0

Post a comment