Application of the Stochastic Tunneling Method for Computational High Trhoughpur Screening
H. Merlitz, B. Burghardt, W. Wenzel
Forschungszentrum Karlsruhe, DE
Keywords: receptor ligand docking, scoring, in silico screeing, stochastic optimization
Virtual screeing of large databases of chemical compounds against known structurally resolved protein receptors emerges as a promising route for lead discovery in drug design. In this approach each ligand of the database is aligned with the receptor to optimize a given scoring function. Then all ligands in the database are ranked according to their optimal score and the best few percent are considered for further, possibly experimental, investigation. The two most important ingredients in this approach are the choice of the scoring function and the reliability and efficiency of the optimization method that is used to find the optimal orientation for each of the ligands.
An initial investigation revealed that a novel optimization method, the stochastic tunneling method (STUN)  was superior in performance and accuracy for rigid ligand docking in comparison to a number of other widely used stochastic optimization techniques (see Fig 1). In the work reported here, we STUN technique to screen a database of chemical compounds to the active site of dihydrofolate reductase for lead candidates in the receptor ligand docking problem. We use an atomistic force field (InsightII/ESFF) and allow internal rotations of the ligands. In our scoring run the natural ligand (methotrexate, MTX) scored best among 10000 tested compounds (see Fig 2). The RMSD between the simulated receptor-ligand complex and the crystal strcuture was less than 1.4 A (see Fig 3). An analysis of the similarities of the the leading five recoptor-ligand complexes permitted us to identify the "docking hot spots" of the receptor that are of paramount importance for its selectivity. In silico site directed mutagenisis of these hotspots resulted in a total loss of specificity of the recpetor for the natural and the next five leading ligands in comparison to the best 1000 ligands of the previous run. This approach to docking is thus able to identify the important characteristics of the receptor even in the absence of the crystal structure with the ligand and able to suggest experimental tests of this hypothesis.
NSTI Nanotech 2003 Conference Technical Program Abstract