Combination of scoring schemes for protein docking
Cologne University Bioinformatics Center (CUBIC), University of Cologne, Zuelpicher Str. 47, 50674 Koeln, Germany
BMC Bioinformatics 2007, 8:279 doi:10.1186/1471-2105-8-279Published: 1 August 2007
Docking algorithms are developed to predict in which orientation two proteins are likely to bind under natural conditions. The currently used methods usually consist of a sampling step followed by a scoring step. We developed a weighted geometric correlation based on optimised atom specific weighting factors and combined them with our previously published amino acid specific scoring and with a comprehensive SVM-based scoring function.
The scoring with the atom specific weighting factors yields better results than the amino acid specific scoring. In combination with SVM-based scoring functions the percentage of complexes for which a near native structure can be predicted within the top 100 ranks increased from 14% with the geometric scoring to 54% with the combination of all scoring functions. Especially for the enzyme-inhibitor complexes the results of the ranking are excellent. For half of these complexes a near-native structure can be predicted within the first 10 proposed structures and for more than 86% of all enzyme-inhibitor complexes within the first 50 predicted structures.
We were able to develop a combination of different scoring schemes which considers a series of previously described and some new scoring criteria yielding a remarkable improvement of prediction quality.