Following presentations were prepared for International Symposium on Advances in Synthetic and Medicinal Chemistry, Saint-Petersburg, August 27-31, 2007. Read more about ASMC >>
HOW SOPHISTICATED SHOULD A SCORING FUNCTION BE TO ENSURE SUCCESSFUL DOCKING, SCORING AND VIRTUAL SCREENING?
Tarasov D.N., Tovbin D.G.
DIDIALL INC., 382 Central Park W, #5K, New York, NY, 10025 tarasov@didiall.com
Increasingly more complicated scoring functions requiring increasingly large amounts of experimental data for their training are currently used for docking, scoring, and virtual screening of ligands for protein binding sites. Such are, e.g., the scoring functions of the GLIDE software. However, when tested independently on targets different from those used for training, these complicated scoring functions yield results that are substantially worse than the results of tests performed by the developers.
To estimated how sophisticated a scoring function should be to ensure successful docking, scoring and virtual screening, we have developed and tested a new scoring function, NScore (naive score). NScore is a very simple scoring function with as few parameters as possible; yet it allows all the main effects determining the ligand-protein interaction to be taken into account. We selected all NScore parameters on the basis of the most general considerations, without any adjustments or trainings using experimental data on ligand-protein interaction. The results of ligand docking and scoring with the use of an independent test set of protein-ligand complexes were as good as the results obtained using the ICM, GOLD, and GLIDE software, and even considerably better than those with respect to some parameters. We also used NScore to perform virtual screening for 8 targets. The results of this screening proved to be almost as good as those provided by the GLIDE and Flex software.
SCORING FUNCTIONS FOR SCORING AND SCORING FUNCTIONS FOR DOCKING: THE DIFFERENCE AND HOW THEY CAN WORK TOGETHER.
Tovbin D.G., Tarasov D.N.
DIDIALL INC., 382 Central Park W, #5K, New York, NY, 10025 dtovbin@didiall.com
In the general case, the scoring functions used for protein-ligand docking and those used for protein-small molecule scoring (prediction of ligand binding affinity) are different scoring functions. The scoring functions suitable for scoring may yield unacceptable results when used for docking, and vise versa.
We have developed a new method for obtaining docking functions intended exclusively for docking and demonstrated that these docking functions correctly predict the most probable position of the ligand in the active site of the protein in almost 100% of cases. If the specially developed docking functions are used for docking, and the subsequent scoring with the use of scoring functions specially developed for scoring is performed for the resultant best position of the ligand in the active site, then the binding affinity of the ligand-protein interaction can be predicted considerably more accurately than if both docking and scoring are performed using scoring functions that have not been developed specially for these purposes.