Table of Contents:
This thesis summarizes the theory, development, validation, and application of new computer-aided methods to score and optimize protein-ligand complexes. These methods make use of atom-type and distance-dependent pair potentials that were derived from small molecule crystal data.
The new scoring function DrugScoreCSD improves the established DrugScore and identifies native and near-native ligand geometries at the best rate reported so far. Furthermore, the scoring of more deviating poses, correlation with affinities, and computation time were subject to improvement.
The new visualization of per atom interactions allows to easily and intuitively evaluate both attractive and repulsive contributions in the protein or ligand.
With the new online application DrugScoreONLINE, the scoring functions DrugScoreCSD and DrugScore as well as the visualization are made available to the public.
Optimizations of scores and geometries can be achieved by the new minimizer. The pair potentials are modified by introducing artificial terms which model short range repulsive and long-range attractive non-bonding interactions. During the course of one minimization run, the PSS algorithm used here (PSS = potential surface smoothing) generates several local minima which can be rescored externally. This new minimizer generates more relevant ligand geometries, improves scoring by DrugScoreCSD, improves affinity correlation, and predictivity of 3D-QSAR models.
Finally, this work shows that DrugScoreCSD successfully assigns atom types and protonation states to protein ligand complexes.