Conclusions

When docking calculations are used as a means of database prioritization, a substantial number of approximations have to be accepted: The conforma-tional and orientational space of the ligand is not searched exhaustively, and, above all, the scoring functions employed can provide only crude approximations to binding free energies. Further restrictions are the neglect of structural water and protein flexibility. Within FlexX, water molecules can be placed together with the incremental construction of the ligand [80], but contributions to the score by each water molecule are difficult to estimate. Several methods have been proposed to account for movement of protein side chains [81-84], but these methods are either ineffective or too time consuming to be applied in virtual screening. A promising approach is docking into ensembles of protein structures, representing important conformational states of the binding site [85].

In spite of a long list of approximations, satisfactory enrichment rates can be achieved by means of database docking. This has been shown by retrospective analyses of several Roche in-house compound libraries. The excellent results obtained for the DHFR library point to the importance of accurate ligand poses. Enrichment rates achieved in this study are somewhat higher than those obtained in a retrospective analysis of cathepsin D inhibitors [13]. It has also been demonstrated that it is of great importance to test virtual screening tools under realistic conditions, i.e. not by trying to retrieve a set of selected high-affinity ligands within otherwise randomly assembled test libraries. Since the ideal general-purpose scoring function has yet to be found, in order to yield acceptable results, scoring functions must often be tailored to solve specific problems. A 'tailored' version of the FlexX scoring function has been presented and could be shown to give good enrichment factors in library screening with thrombin as a target. On the other hand, the Vertex group has reported that 'consensus scoring' has proven to be successful in many applications, i.e. the selection of those potential inhibitors that score well with several general-purpose scoring functions [86]. This strategy can help to reduce the number of false positives in virtual screening. As more alternative scoring schemes become available, this strategy might become even more interesting. Potentials of mean force like the Muegge scoring function and other recently published alternatives [48,50] should be given special attention.

0 0

Post a comment