Weak homologies down to about 30% sequence identity can be easily handled and correct fold motifs can be predicted. With higher homology levels, threading methods allow the prediction of the structure in atomic detail to about a 2Å resolution. Of course this will work only if the experimentally determined structure of a protein with a homologous sequence is available. With lower homology levels predictions are still possible, however the accuracy and especially the reliability of such models degrade quickly.

If the 3D structure of the protein is available, docking methods attempt to predict the aggregation of a small molecule with the protein. Our docking method FlexX allows the prediction of the geometric arrangement of a protein-ligand complex to around about 1.5Å resolution in over 80% of the cases (if multiple docking solution candidates are acceptable). Such a calculation can be carried out in about a minute and distributed computing allows for the speed-up required to process entire in-house collections.

In the case of missing 3D structural data, the similarity of compounds becomes a crucial parameter. On the basis of the paradigm 'similar molecules behave similarly', similarities can be assessed at various property levels. Our small molecule alignment tool FlexS allows a detailed comparison of surfaces, property distributions, and other characteristics of molecules and has been successfully used to pre-select molecules for biological testing.

The docking program FlexX has recently been enhanced in three major ways. Firstly, the combinatorial nature of a dataset of molecules can be directly taken into account. The docking procedure has been enhanced such that in an incremental construction procedure, substituents can be enumerated systematically without re-docking the fixed part of the molecule. Therefore the combinatorial docking in FlexX^C is much more efficient than the traditional sequential docking of a fully enumerated library. Secondly, the constraint of a rigid pocket has been overcome by another enhancement of the docking tool (called FlexX-Ensemble) which can take into account a whole ensemble of protein structures at once. Movements of side-chains, the backbone, or even whole domains, captured by a series of snapshots are considered during the docking procedure. Note that this procedure is more than a mere sequential docking into the individual ensemble members. A model of the site, developed on-the-fly, can be any combination of structural elements in the ensemble structures. Third and finally, a frequently requested enhancement, namely consideration of pharmacophore points while docking, has been developed. Again, the procedure is more than a mere filtering of docking solutions. A look-ahead strategy has been implemented that helps to prune a partial docking solution early on that will not be able to fulfill the constraints when completed. The same enhancements are easily applicable to the small-molecule alignment tool FlexS, which has the same software basis as the docking program, and will be implemented in the near future.

In cases where higher throughput is mandatory, the descriptor technology FTrees allows the user to perform similarity calculations on thousands of molecules in a matter of seconds - which makes it applicable to virtual uHTS. FTrees is based on a topology-based descriptor (a Feature Tree) that extends greatly beyond traditional 2D approaches in that quite distant topologies can be mapped onto each other if sufficient agreement of correctly positioned physico-chemical features can be found. In a fragment spaces module FTrees-FS, FTrees has also been extended to enable a fragment based design approach via similarity searching in combinatorial libraries and chemistry spaces.