2009-11-08 15:40 | 科学网博客 | 龚新奇
The authors described an idea and method to use the interaction fingerprint indexes, IFP, to cluster the docked protein-peptide complex decoys and to select the near-native decoys. This is interesting for protein docking. The similar factor, interface interaction, along with the usually used RMSD, has been used for evaluation of docking methods in CAPRI, the critical assessment of predicted interactions, since 2001. In there, three indexes are used, ligand RMSD, interface RMSD and the fraction of native interface interactions.
I think the idea and the result are illuminating. So I recommend publication after the following revisions and questions were satisfied.
1. The authors used ZDOCK for side-chain trimmed rigid docking and then optimized the side-chains. “volume of each side-chain of the receptor protein was modified to that of alanine before docking. After restoring the volumes of the modified residues to their original values, side-chains of the protein-ligand complex structures were optimized by SCWRL3”. But the description was not clear. How did you restore the side-chain? What was the coordinates of the trimmed side-chain atoms? Why not use RosettaDock for this flexible docking?
2. In docking, the decoys are usually selected and ordered with energy scores, so the authors should compare their results of Tc-IFP with energy scores.
3. Fig.1 tried to point out the weakness of RMSD, but this is not strict. In practice, we often use RMSD for some specified regions, for example, interaction interface, not only for the whole molecules. It is natural to obtain different RMSDs for different regions, although the regions are same in size. RMSD is not only dependent on the superposition step, but also dependent on the superposition region.
4. In page 4, the authors claimed to use the IFP as a scale for measuring unique similarities between complex structures. But, just like RMSD, IFP is not sufficient for that. That’s why we use both in CAPRI.
5. In page 5, the authors claimed “Some complex structures were rejected in the process of side-chain rearrangement because of their unavailable arrangements.” This is not clear. How did you know which side-chains were unavailable, how to define and select them?
6. In page 6, the authors claimed “From a desired viewpoint of docking prediction, a set of b-decoys, generated from the cognate structure, was expected to have more near-native decoys than a-decoys, generated from various single state CaM structures.” In the actual docking practice, this is not true. In many bound docking cases in CAPRI, the failure were very high. And then the analysis thereafter based on this idea should be questioned.
7. In page 7, the authors claimed “This might have resulted from the smaller interacting surface of the CaM-PMCA complex than the respective CNG and CaMKK complexes, as shown in Figure 4.” I think the results of the case CaM-PMCA justify the importance of RMSD, but not only IFP.
8. Generally, the cluster threshold will influence the clustering result, so what’s the different effect of the different thresholds given in page 8?
9. In page 14, the authors claimed the IFP were defined by the software LIGPLOT, but LIGPLOT only gives out hydrogen bonds and hydrophobic interactions, why not consider some other interactions?