Report for 1^st quarter of second year (August - October 2005)

TArget Specific Scoring FUNctions (TASSFUN)

In the TASSFUN project, the structure-based COMparative BINding Energy (COMBINE) method should be improved for developing target-specific tailor-made scoring functions. These scoring functions will be used as 3D QSAR tools in virtual screening to select specific ligands for different blood coagulation cascade related targets (e.g. thrombin, factor Xa, urokinase, trypsin). The project is supervised by Dr. Niklas Blomberg (GSI CompChem AZ Mlndal) and Dr. Rebecca Wade (EML Research, Heidelberg).

In the first year of the project, we focused on acquisition of published data (e.g., structure and kinetic information), the issues specific to modelling structures of the complexes of the urokinase, trypsin and thrombin proteases, as well as on programming and implementing necessary tools for handling a large number of protein-ligand-complexes. Following the work in the present report period, the programs have reached a stage, which makes it possible to build COMBINE models based on a training set of structures of complexes and apply these COMBINE models to a large test set of ligands with unknown X-ray crystal structures.

For testing the procedure, 30 ligands with known binding conformation and inhibitor constants were minimized by molecular mechanics together with a urokinase receptor model. The van der Waals and electrostatic interaction energies were calculated and decomposed for each amino acid-ligand-pair by the ANAL program of the AMBER 8 software suite. In addition, electrostatic desolvation energy terms were computed by solving the Poisson-Boltzmann equation for the receptor and the ligand using the program UHBD. The decomposed interaction energies and the electrostatic desolvation energy terms formed X variables in a matrix and were correlated by Partial Least Squares to binding free energy values given as Y variables. After variable selection and removing five potential outlier complexes from the training set, correlation coefficients of r²=0.84, q²= 0.43 and SDEP=1.05 kcal/mol were obtained.

In a second step, to create an initial test set, the same ligands were docked using the program GOLD 2.2 into the active site of a urokinase receptor model. Afterwards, the docked ligands were minimized and their energy terms were calculated in the same way as for the training set. Because of problems in selecting the best binding mode from the docking solutions, the correct binding affinity could not be predicted by the COMBINE model for all ligands. In a further test, a virtual screen was performed. 180 ligands with experimentally known inhibitor constants for urokinase, but unknown binding conformations, were docked into the receptor model. Again the results were of variable quality, with good and poor predictions of binding affinity obtained.

The prediction of inhibitor constants by the COMBINE model in conjunction with GOLD docking procedures is still in progress. In the next quarter, we will work on improving the docking procedure and the scoring and selection of docked poses. We will try use of the new version of GOLD, version 3.0, which permits treatment of explicit water molecules. We will also consider more GOLD poses (only the pose ranked highest with ChemScore has been used for the COMBINE models to date) and alternative scoring procedures. The COMBINE models themselves will be analyzed to identify the most important energetic terms and to detect outliers. In parallel, we will prepare structures of complexes of trypsin and of thrombin so as to apply the complete procedure to other interesting targets.