Use of eHiTS CrossDocking Approaches to Survey the Ligand/Target Landscape
Monday, September 29th, 2008Virtual high throughput screening requires the use of approximate approaches to distinguish the complimentarity of ligands to target. Often this falls to the domain of 2D approaches such as fingerprint, topological, and shape screens, given the computational efficiency of such approaches. High throughput docking, employing the 3D structures of ligand and targets, however, provides a facile first approach to glean principles for ligand recognition of targets. The endpoint might be screening a database against a single target to determine which ligands are plausible lead compounds to be explored with other computational approaches (e.g. free energy computation and estimation of binding free energies) or recommendation for synthesis or testing in vitro high throughput assays. Alternatively, one might wish to know how such compounds ’score’ against an array of targets that one doesn’t want to “hit” with a novel lead candidate. Such an approach may be important in eliminating unwanted side-effects/toxicity
Before using eHiTS for such a screening enterprise, it is useful to perform a benchmark crossdocking screen using a set of crystal structures of ligand-receptor complexes. Each of the ‘native’ ligands extracted from each of the co-complexes is then docked against the panel of other targets. The matrix below shows an example of such a hypothetical screen to examine how each of the native ligands of 13 targets score against targets in the same or distinct pharmacological families. Each cell entry corresponds to the eHiTS docking score of the ligand extracted from the top row complex against the target extracted from the left column complex. The diagonal elements (bold-font) show the best docking score for (self docking of) each ligand against its own crystal structure target. The blue font entries in each column illustrate that while quite (most) often the ‘native ligand’ for a target crystal structure is amongst the best scoring (within 1-2 eHiTS scoring units) it is not always the best score. Quite often ‘native’ ligands for other targets will have significant affinity for your target of interest (and will plausibly have good eHiTS docking scores) and this is both the challenge of drug design in attempts to control specificity and selectivity for a panorama of target proteins/receptors.
One is looking for several facets in such a preliminary screen. One expects that each of the native ligands found in a co-complex crystal structure of course does dock to the native target. One expects, in most instances, that the docking score for each native ligand for its ‘natural’ target will be near the minimum score compared to other ligands (typically within a few scoring units). The figures below shows in what portion of the cases the native ligand is among the best N scoring ligands. It is of course possible that ‘native’ ligands derived from other target complexes will score well against your target of interest, but that kind of ‘crosstalk’ is precisely the type of information that one wants to glean from such a low cost computer ‘experiment.’ One also wants to know how use of different crystal structures with alternative conformations of binding site residues affects the docking scores. While a docking score is not a binding free energy, eHiTS docking scores often give at least monotonic score correlations with lnKd’s in instances where the scoring function calibration has included interaction types of the nature of your target of interest.
What about the timescale for such screens? For eHiTS 6.2 on Intel platforms versus the new Lighting release? In a recent benchmark I found it to be ~3.5 minutes/ligand (using 4-Xeon 3.2GHz nodes) versus an average of 30 seconds per ligand on Cell processor in a Sony Playstation 3 (PS3). Note that this corresponds to docking at the highest accuracy. For screening of large libraries of molecules, a lower level of accuracy is recommended with a typical docking time of less than 10 seconds per ligand on the PS3.
While the timescale of such 3D approaches has not been competitive with 2D methods, the advances in porting eHiTS to the Cell will make this approach feasible for high throughput analysis with quantitative estimation of both docking-score/IC50 correlations and accurate pose prediction.
Blog post by Dan Harris
