DDI-SSL: Drug–Drug Interaction Prediction Based on Substructure Signature Learning

Round 1

Reviewer 1 Report

To add space between the first "interactions(DDI)" to become "interactions (DDI)" in the Abstract.

Impressive as the work is only performed by one Author, but the term "we" was used throughout the article.

Author Response

1）To add space between the first "interactions(DDI)" to become "interactions (DDI)" in the Abstract.

 

Answer: I appreciate that the reviewer points out this problem, I add space between the first "interactions(DDI)" to become "interactions (DDI)" in the Abstract in the revised manuscript.

 

 

 

2）Impressive as the work is only performed by one Author, but the term "we" was used throughout the article.

 

Answer: Thank you very much for your constructive suggestion. In this manuscript, although there is only one author, I carefully reviewed the Applied Sciences journal, and there were no writing guidelines or requirements to use 'I.' This manuscript leans towards using the third person 'we' to maintain objectivity and formality in sentences, even with a single author.

Author Response File: Author Response.docx

Reviewer 2 Report

In the paper under consideration, the authors proposed new deep learning-based method including graph embedding, multihad attention, and collaborative attention. DDI-SSL, their method, is interesting and sounds rational architecture. To enhance the article, I recommend the author to improve the presentation of results.   [Major] M1. We cannot see the result of DrugComb DB because of poor representation of Table 3. M2. The authors should claim the DDI-SSL is statistically better than other methods in a metric (I think one of metrics is acceptable) via significance testing. M3. I guess the bolded values of Table 3 mean the best values among models but sometimes it is not. please clarify and please check. M5. "Although DeepSynergy exhibits ... Even with very low true positive rates, multiclass classifiers often have high AUC scores." I know ROC-AUC is sometimes unstable with imbalanced data, but I am not sure the nature of overestimation. It should be explained more. M6. "On the TwoSides dataset, the accuracy is significantly improved, and the convergence is also faster." The comparison in Figure 2 is insufficient (it includes only some methods, not all). Furthermore, "significantly" must be used with significance testing. I strongly recommend not to use "significant", "significantly" and so on without any statistical testings.   [Minor] m1. "lr = 1e-2 times 0.96^t" does it means "0.01*(0.96^t)"? it is confusing because at first I misunderstood as "e minus 2*(0.96^t)". m2. Table 3 has "plus-minus" and bolded values. The author must note their meanings as caption. m3. I wonder why MHCADDI and CASTER were not included in Table 3. Especially, the author mentioned MHCADDI, but it did not appear in any results. m4. "Test Accuracy" is better than "Accuracy" on Figure 2.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

 

The manuscript provides a comprehensive overview of the proposed DDI-SSL method, highlighting its novel approach in identifying drug-drug interactions and predicting related side effects. The introduction effectively establishes the problem with previous methods relying on global structures and the potential inaccuracies they lead to in handling drug information. The description of the self-attention mechanism for generating signal strength scores and the utilization of weighted drug attributes to mark substructures is intriguing and well-explained. The flexibility of adapting to various subgraph shapes and sizes is emphasized convincingly. The article effectively highlights the model's credibility through comprehensive experiments on DDI datasets, reinforcing the improvements introduced by the DDI-SSL model.

Overall, the manuscript adeptly captures the essence of the proposed DDI-SSL method, its innovative mechanisms, and its potential impact on the field of drug-drug interaction analysis. The clarity of explanation, strategic organization, and forward-looking perspective enhance the value of the article.

Improvation is needed in the sentence structure in certain places is quite complex, which might lead to confusion for readers unfamiliar with the subject matter. Simplifying sentence structure could improve clarity

 

 

 

Author Response

Thank you very much for taking the time to review this manuscript. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.

Author Response File: Author Response.docx

Reviewer 4 Report

 The manuscript entitled “DDI-SSL: Drug–Drug Interaction Prediction Based on Substructure Signature Learning” introduces the crucial problem of drug-drug interactions (DDIs) and the limitations of existing methods. It presents the proposed solution, DDI-SSL, outlines its key features, such as substructure signature learning, deep clustering, and a layer-independent collaborative attention mechanism. The manuscript deals with crucial problem on drug-drug interactions and methods to analyze it. Manuscript can be published after incorporating suggested changes.

 

1) Instead of simply stating that DDI-SSL demonstrates improved performance, consider providing specific metrics or percentages to highlight the extent of improvement. This can make the claim more compelling.

2) Please provide abbreviation of GAT layer when used first time.

3) According to journal criteria, add separate section for Materials and Methods, Results, Discussion and Conclusions.

4) Table 1, in manuscript management context and number of labels are not discussed exclusively in the manuscript.

5) Please add limitation statement and explain the real-world impact of accurate predictions, such as improving patient safety, drug development efficiency, or healthcare cost reduction. Also mention it clearly if this method is specifically used for organic molecules or metallo-ligands (ex. Cisplatin) as well.

It will be good if language is simplified for most of the researchers.

Author Response

Thank you very much for taking the time to review this manuscript. Please see the attachment and the corresponding revisions/corrections highlighted/in track changes in the re-submitted files.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I have reviewed the authors' revised version. I believe it is acceptable.