Comparison with Xray data

The effectiveness of an indirect approach to drug design like molecular superpositioning is difficult to evaluate. A quasi-standard metric is to compare

Figure 4. Distance-based merging of Gaussian functions. Dots display positions of Gaussians and edges connect dots representing Gaussians which lie closer to each other than a certain threshold. The encircled Gaussians are to be merged.

the computed alignments with crystallographic data. This is possible if the 3D structures of protein-ligand complexes with different ligands binding to the same protein are available. To this end we extended a dataset initially compiled by Klebe et al. [31] to a total of 76 protein-ligand complexes of 14 proteins with the ligands varying in size from 18 to 158 atoms and up to 35 rotatable bonds. This way, 460 pairs of molecules are found that bind to the same protein. We mutually superimposed the complexes by minimizing the positional differences between the backbone Ca atoms. Then we extracted the ligands from the complexes keeping their obtained relative orientation in space to define our reference alignment.

The approach is now to take the reference compound in the given orientation and conformation, to superimpose the test compound whose conformation is taken to be unknown flexibly onto the reference and to measure the distance of the observed versus the predicted conformation and orientation of the test compound in terms of the traditionally used root-mean-square deviation (RMSD) of all non-hydrogen atoms.

Of course there are a lot ofpros and cons to this kind of evaluation. However, it gives at least an idea as to what extent a method is able to reproduce configurations determined by experiment. Since the superposition can only be meaningful in those regions where reference compound and test compound overlap to a reasonable extent we propose a minimum overlap volume requirement of 60% of the volume of the smaller compound and measure the RMSD of only those atoms of the test compound that overlap with the reference in the experimentally observed configuration. Table 1 shows the respective adjusted values (differentiated by the rank at which the alignment

Table 1. X-ray data comparison

Rank RMSD_

Top 20. 7% 37.7% 45.4% Top 10 28.7% 45.7% 53.7% All 39.4% 55.4% 62.0%

The table displays the percentage of the 284 alignment pairs for which reasonable alignments are achieved. The results are categorized by different RMSD thresholds and different fractions of the rank lists considered.

appears on the solution list) for a reasonable subset of 284 pairs of molecules (definedby the 60%-criterion, cf. above).

With almost 50% of the alignments reproduced with a reasonable RMSD on one of the first ten ranks this is an encouraging result. Similar validation studies recently performed for docking methods achieved reproduction rates of about 60-70% [35,36].

0 0

Post a comment