From the perspective of computational de novo screening, the ability to accurately predict the potential bound configurations of small low-molecular weight fragments to a target is essential. The challenge is significant given that the binding site may be large compared to the overall size of the fragment and the affinity of the fragments for particular regions of the binding site may be on the order of mM or less. Finally, it important to not only predict the poses with highest binding affinity (with small root mean square deviation to known structural biology solutions) but to predict a diverse ensemble of bound configurations of each fragment with a large distribution of binding free energies to the target. All of these requirements are crucial given that de novo/fragment design aims to obtain a robust nM affinity binding ligand by tethering weakly bound fragments derived mM to uM range.
The requirements for a fragment docking and scoring approach employed in a denovo design paradigm from our perspective are:
1) a procedure which provides exhaustive sampling of the potential bound fragment configurations in the binding site,
2) a sampling regime in which modest affinity fragment locations are sampled more effectively that low affinity sites,
3) for cases where structural biological solutions of the bound fragment configuration is known, one wants the pose prediction approach to have a good correlation of predicted docking score with docking poses with low RMS deviations from the structural biology solution.
A simple example is shown below that exemplifies these characteristics for the particular case of fragment docking of a fragment found in the pdb code 1OV7 where a small phenolic fragment (2-allyl-6-methyl phenol) is bound to T4 Phage Lysozyme.
Panel A in Figure 1 shows the overlap of the best scoring pose of the phenolic fragment with the crystal structure pose. The top scoring pose has an RMSD of 0.5 Å with respect to the crystallographic configuration. Panel B shows the correlation of the pose root mean square deviation with the eHiTS docking score. The plot illustrates that there is correlation of pose score with low RMSD particularly in the low RMSD/good score region. Panel C and D show the histograms of the RMS deviations and scores of the docked fragment poses. Note that the deviations peak in the good score/low RMSD regime, analogous to what is seen in the distribution in the RMSD vs. Score plot.
I’ll back in coming days to talk about fragment affinity scoring!
Posted by Dan Harris