BioSolveIT at the 229th ACS National Meeting
Mining docking space
Christian Lemmen, Marcus Gastreich, and Holger Claußen.
Modern experimental technology produces an ever increasing mass of data. However, with more and more cheap computer power, also virtual experiments are carried out in high throughput mode, which further adds significantly to this flood of data. One aspect in this regard is virtual docking. Typically hundreds of possible docking poses are generated for every single compound and hundred thousands of compounds are processed overnight on a cluster in a routine fashion. The mere storage of tens of millions of protein-ligand complexes is a challenge, however, analysis down to the level of interaction profiles so far impractical to say the least.
We implemented a system based on Oracle which facilitates the analysis of large volume docking data. It utilizes Flex*-Technology and is equipped with a graphical front-end which provides a spreadsheet-type view on the data. Additionally to interactive filtering, sorting, searching and visualization, this system called Docking Database (DDB) has an interface to a Machine Learning Toolbox (MLT) facilitating the generation of target tailored scoring functions. DDB was taken to the test in an application study which is discussed in this presentation. Significant docking-parameters for the particular target could be detected, filters have been optimized and a new target-specific scoring function was derived, all of which led to an improved performance of the docking method when used on novel compounds.