Martin, Oliver;Wohlfahrt, Gerd;Schomburg, Dietmar - Development and application of functional surface representations of proteins using GRID: A simple but effective post-filter for geometric rigid body protein-protein docking

ECCB 2002 Poster sorted by: Author | Number

Next | Previous poster (in order of the view you have selected)

Title: Development and application of functional surface representations of proteins using GRID: A simple but effective post-filter for geometric rigid body protein-protein docking	P101
Martin, Oliver; Wohlfahrt, Gerd; Schomburg, Dietmar oliver.martin@uni-koeln.de Institute of Biochemistry, University of Cologne, Zuelpicher Strasse 47, D-50674 Cologne, Germany;

GRID is a well established program which is widely used in drug design to calculate energy potentials of chemical probes near the surface of a target protein [1]. In order to analyse a protein surface for potentials that could be of significance for the interaction with other proteins, twenty "protein-like" probes with different properties were chosen out of a set of 64 predefined probes in GRID [1]. The properties of those probes resemble those of atoms or functional groups of amino acids in proteins. After initial analysis ten single atom probes from the subset of "protein-like" probes were finally used for energy calculations at protein surfaces. GRID spans a grid with a spacing of 1Å around the protein expanding 5Å beyond the surface and calculates the energy for every grid point for the chosen probes.
In order to determine functional properties of protein surfaces which allow molecules to form specific interfaces with other proteins, the surfaces of a non-redundant set of 36 co-crystallised non-immunoglobulin protein-protein complexes were analysed using the method described above. The calculated data was stored in a relational SQL-database and a statistical analysis was performed on this data. Comparing the frequency of occurrence of favourable grid points at and outside of protein-protein interfaces, the analysis shows different distributions for some of the probes at certain energy levels. The aim of this analysis was to find significant correlations between calculated probe-specific energies of one partner of a protein-protein complex and six distinct atom types in the other protein. Different radii account for fuzziness caused by methodical errors and structural changes upon complex formation.
Correlations that seem characteristic for interfaces of the analysed complexes have been detected. Using these, a simple post-filter for protein-protein docking results was developed. Based on these observed correlations, a simple normalised count of their occurrence was used to improve the ranking of proposed complex structures obtained by geometric FFT docking with the KORDO program [2]. First tests of the filter with bound complexes show that already this very simple method of post-filtering performs very well. For these tests we used the 20 complex structures ranked highest by KORDO [2]. More systematic testing showed that the filter is able to recognise the correct, respectively 2near-native" structure in about 60-70% of the cases, while it is able to significantly improve the rank of the correct complex in an additional 10-15% of the studied cases.

[1] Goodford, P.J. (1985). A computational procedure for determining energetically favourable binding sites on biologically important macromolecules. J Med Chem, 28(7):849-57.
[2] Meyer M., Wilson P., and Schomburg, D. (1996). Hydrogen bonding and molecular surface shape complementarity as a basis for protein docking. J Mol Biol, 264(1):199-210.