Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closeReferee Comments: Referee 1 (Michael Schroeder)
Posted by PLOS_ONE_Group on 08 Apr 2008 at 23:06 GMT
Referee 1's Review (Michael Schroeder):
With the advent of protein interaction interface databases the alignment of interfaces has become an important problem. Comparison of interfaces reveals structural and geometric insights into binding modalities and hot spots, which sequence based method cannot reveal. A particular problem, which can be addressed with interface alignments, is the study of convergently evolved interaction interfaces.
The paper pursues a novel graph-based approach, which represents an interface as a graph of connected atoms, which are characterised by a certain type. With this representation, aligning interfaces reduces to finding the largest common subgraph.
The proposed method is compared to two other approaches. The I2I-SiteEngine is based on the alignment of functional groups at a set of binding sites and DaliLite takes an indirect approach ( based on backbone structures alignments. The proposed method is applied to confirm 3 cases of protein mimicry.
The paper is well written. The method is explaied in a clear way.
Accept the paper with minor revisions:
- the authors should clarify what is the exactly the similarity score between two interfaces. They state that "all resulting alignments of NCIVs are sorted according to their sizes, and the
largest alignments are reported." The authors should discuss how this influences the ranking of interfaces of different sizes. There will be examples, where both interfaces are of the same size and where one is much larger than the other. In the current ranking both cases would not be distinguished.
- The authors should consider different atom types to distinguish unspecific matches of atoms from significant ones.
This helps to distinguish Van der Waals interactions from electrostatic ones (not considered at all in the procedure so far, but often very strong and important at protein-protein interfaces).
- Run-time: The authors report "reasonable run time (within minutes)". Could they elaborate. How does this compare to the other systems. Is time spent mostly on the clique computation? Could the method be use to run large-scale screens to identify novel cases of mimicry or is it limited due to computational demands to small datasets.
- The authors should show (in the last two examples of mimicry, scorpio-toxin mimicry of CD4 in complex with gp120 and expecially SP4206 mimic of IL-2Ralpha) how the I2I-SiteEngine method performs in these cases (do they bring to similar alignments, which are the regions with larger RMSD fluctuations, where do the two methods complement each other?). Expecially in the last example, where the number of overlapping NCIVs is low the alignment should be compared to the I2I-SiteEngine result.
***********
N.B. These are the comments made by the referee when reviewing an earlier version of this paper. Prior to publication the manuscript has been revised in light of these comments and to address other editorial requirements.
RE: Referee Comments: Referee 1 (Michael Schroeder)
macrozhu replied to PLOS_ONE_Group on 15 Apr 2008 at 13:43 GMT
Since this is the comments of the referee to an earlier version of the paper, we post here relevant excerpt from the reply to the referee's report.
Comment 1: “the authors should clarify what exactly the similarity score between two interfaces constitutes. They state that "all resulting alignments of NCIVs are sorted according to their sizes, and the largest alignments are reported." The authors should discuss how this influences the ranking of interfaces of different sizes. There will be examples, where both interfaces are of the same size and where one is much larger than the other. In the current ranking both cases would not be distinguished. ”
Answer: Currently the resulting alignments of Galinter are sorted by the number of interactions aligned at the two interfaces. We have revised the manuscript (the last sentence of section “Extending aligned representatives to NCIVs”) such that this is measure is described clearer:
Comment 2: “The authors should consider different atom types to distinguish unspecific matches of atoms from significant ones. This helps to distinguish Van der Waals interactions from electrostatic ones (not considered at all in the procedure so far, but often very strong and important at protein-protein interfaces). ”
Answer: That is a very good suggestion. We have added a new section “Contribution of different types of non-covalent interactions to the alignment”, where we investigate the effect of weighting short-range electrostatic interaction on the alignment. We have found that these electrostatic interaction occur relatively seldom in the pilot dataset interfaces. Giving a larger weight to the electrostatic interactions, only a few Galinter alignments are changed.
Comment 3: “The authors report "reasonable run time (within minutes)". Could they elaborate. How does this compare to the other systems. Is time spent mostly on the clique computation? Could the method be use to run large-scale screens to identify novel cases of mimicry or is it limited due to computational demands to small datasets. ”
Answer: The run time depends on the size of the interfaces to be compared. For the pilot dataset, the mean run time is 138.5 seconds (median 71.5 seconds) on a normal desktop (3.0GHz CPU, 1GB memory). Only approximately 10% of the run time is spent on clique search. On average, approximately 50% of the run time is spent on the extension step. This indicates that large-scale investigations are feasible with Galinter. The current implementation is in Python and is not optimized for speed.
I2I-SiteEngine is faster at comparing interfaces from the pilot dataset. The average run time for the 240 comparisons of interfaces in the pilot dataset is 20.8 seconds on the same desktop.
Comment 4: “The authors should show (in the last two examples of mimicry, scorpio-toxin mimicry of CD4 in complex with gp120 and especially SP4206 mimic of IL-2Ralpha) how the I2I-SiteEngine method performs in these cases (do they result in similar alignments, which are the regions with larger RMSD fluctuations, where do the two methods complement each other?). Especially in the last example, where the number of overlapping NCIVs is low the alignment should be compared to the I2I-SiteEngine result.”
Answer: This issue has already been addressed in section “Comparison to I2I-SiteEngine results”, where the I2I-SiteEngine alignments of the first two mimicry cases are compared to the corresponding Galinter alignments. As also stated in the same section, I2I-SiteEngine is not applicable to the third mimicry case because one of the two interfaces is formed between a non-peptitic small molecule (SP4206) and a protein. I2I-SiteEngine relies on the identification of functional groups of amino acids at the interfaces. SP4206 contains no amino acids, thus functional groups are not well defined in it. Therefore we could not obtain alignment results for this case. We now make this point more visible in the first paragraph of the discussion.