Referee 1's review:

The authors described a new method CSIDOP with high accuracy in protein functional assignment (using GO terms) through investigating cross-species protein-protein interaction data and protein domain patterns. The idea behind this method is that protein functions are implemented through protein-protein interactions (PPI), while PPIs are modulated by combinations of protein domains in both PPI partners. Thus identifying these combinations of patterns, together with the PPI info, would provide a very accurate way in functional assignment. The manuscript was well written, and should be understandable for non-specialists. Genome scientists, system biologists, and computational biologists would find this paper of special interests, because of its potential applications in these areas.

There are several issues that authors need to address:
1. Sensitivity issue: Because of the special requirement of both PPI data and domain information, this method has a good accuracy rate of 95%. Similar to many bioinformatics papers published, authors ignored the discussion of the other side of the story: sensitivity, which is an important factor to consider for potential applications. Although the sensitivity is expected to be low, a discussion on this direction will help readers to understand this algorithm better.
2. Orthology issue: This methodology first identifies the domain combination patterns, which should lead to homologs; the additional requirement of PPI information will probably lead to orthologs. As also described by authors, lots of functional assignments made by CSIDOP are actually based on the GO assignments of their orthologs. For example, the case authors discussed on the last paragraph of Page 3: Q9XVV3 interacts with Q18688 in C. elegans, while P03372 interacts with P07900 in human. Actually Q18688 is the ortholog of P07900, thus it's no wonder that they share the same function. Despite of the non-orthologous relationship between Q9XVV3 and P03372, they still share the same function, which is where the extra power of this method CSIDOP comes into play. Thus, it would be really informative to include an orthology detection algorithm (Inparanoid or OrthoMCL) in the benchmarking analysis shown in Table 1, and also to calculate what fraction of CSIDOP's (new) assignments can be also covered by orthology detection.
3. It would be useful to know how CSIDOP performs on multiple species other than human.
4. In Table 5, the "Verification with Evidence" is not clear at all, e.g. on the first line, how does Q99829 relate with Q96A23, and why this can be an evidence that Q96A23 should have the GO:0001786 assignment?
5. It would be useful to have the "Look-up Table of Domain Patterns & Associated Functions" (described in the METHODS and Figure 4) online.

*********
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.