A detailed outline of the scoring function, including the derivation of the ligand volume correction factor, is provided elsewhere [27].

The PMF are derived from a set of 697 protein-ligand complexes taken from Reference 27. A bin size of 0.2 A is applied. Sixteen protein atom types and 34 ligand atom types are defined and lead to 544 PMF of which 282 are statistically significant [27]. Protein-ligand contacts that have less than 1000 occurrences in the database are considered statistically insignificant and are ignored. PMF derived for hydrogen atom types are also ignored throughout the work motivated by the fact that many ligands and proteins used to derive the potentials do not have all hydrogen atoms present in the database. Test runs with smoothed potentials, which are generated by using Savitzky-Golay filters, do not improve the scoring. Therefore, the derived potentials are taken at their face values. We show below, that the predictive power ofthe potentials are, indeed, very sensitive to small changes.

0 0

Post a comment