FFT-Correlation is one of the fastest and most accurate techniques for protein docking. Based on the original work of [Katchalski-Katzir, 1992] the program KORDO had been developed in our group for predicting the structure of protein complexes [Meyer et al., 1996].
While FFT-Correlation is invariant to translation, the complete rotational space has to be scanned if no prior knowledge is used. Using bound structures near-native orientations with an RMSD around one Angstroem are obtained in less than an hour on a normal workstation. Although this is fast compared to other methods, weeks of computation would be required if 1:N dockings to a whole database of hundreds of structures were performed. In order to speed up the docking process at least one order of magnitude we have applied a prefilter based on hydrogen bonding patterns. This filter effectively reduces the number of orientations to be calculated.
While most of the cocrystallized complexes have been successfully docked from their respective monomers, there are some complexes which show either rather bad geometric complementarity or have very bumpy interfaces, both leading to low correlation values in the rotational scanning. Attempts to deal with these cases are discussed. Furthermore, in addition to the native interface there are other orientations which show high geometric complementarity. To distinguish the native orientation from other high ranking orientations a large survey of interface features from several hundred native and non-native interfaces was performed. The performance of different postfilters based on the results of this survey are compared.
As an application some results obtained by docking bound structures of transmembrane regions and unbound structures, respectively, are shown.