Dear MS. Pancheva,

Thank you very much for submitting your manuscript "Using topic modeling to detect cellular crosstalk in scRNA-seq" for consideration at PLOS Computational Biology.

As with all papers reviewed by the journal, your manuscript was reviewed by members of the editorial board and by several independent reviewers. In light of the reviews (below this email), we would like to invite the resubmission of a significantly-revised version that takes into account the reviewers' comments.

We cannot make any decision about publication until we have seen the revised manuscript and your response to the reviewers' comments. Your revised manuscript is also likely to be sent to reviewers for further evaluation.

When you are ready to resubmit, please upload the following:

[1] A letter containing a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

[2] Two versions of the revised manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file).

Important additional instructions are given below your reviewer comments.

Please prepare and submit your revised manuscript within 60 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email. Please note that revised manuscripts received after the 60-day due date may require evaluation and peer review similar to newly submitted manuscripts.

Thank you again for your submission. We hope that our editorial process has been constructive so far, and we welcome your feedback at any time. Please don't hesitate to contact us if you have any questions or comments.

Sincerely,

Qing Nie

Associate Editor

PLOS Computational Biology

Ilya Ioshikhes

Deputy Editor

PLOS Computational Biology

***********************

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: Comments to the Author

In this manuscript, the authors seek to use LDA as part of a procedure to identify genes that change their expression as a result of cell-cell interaction. Code is made available through the first author's Gihub. I found the application to be relevant and current, and the method to be interesting. However, I did think that the exposition of the approach needed tightening, and had a few other minor queries/comments, as described below.

Minor points:

1. The LDA model (Section 2.1). Typo: "\\alpha and \\beta are Dirichlet priors" -- I think these are vectors.

2. "Identifying topics linked to a cell type": for completeness and precision, it would be useful in this section to state the hypothesis that is being tested using Mann-Whitney's U.

3. In Table 1, why is 30 chosen as the lower bound for the number of topics? In general, how is an appropriate candidate set of values chosen?

4. Some of the language needs a little tightening, e.g. "Ranking genes potential candidates of interaction", "Let N be the doublets/interacting population of interest" (is N the sample size?), missing articles (e.g. "We expect the probability of observing counts greater or equal to the actual count for gene changed...", "Metadata file was used to filter ...", "Full list of filtering cut-offs..."), avoiding contractions (e.g. "a gene we haven’t modified...", "It's important to note...")

5. A 2-stage approach is adopted here, in which LDA is first performed on a reference population, and then another LDA is performed to try to identify new topics that capture any changes as result of interaction. I am not suggesting that the authors try this, but I wondered if they had instead considered a hierarchical model that would seek to model the two datasets jointly? If so, what would be the potential advantages/disadvantages of such an approach?

6. "Finding the number of topics" - here, average cosine distance is used, but are other options available? Why was this particular criterion chosen?

Reviewer #2: Pancheva et al. present a method for the analysis of cell-cell interactions through latent Dirichlet allocation (topic modeling). The methods introduced are interesting and I could see their potential impact when applied to current and emerging single-cell sequencing technologies. The paper in its current form is unclear in several places on the methods and the results presented, which limits comprehension of the main points being made. Specific concerns are listed below.

Major points

1. Definition of terms. I find the introduction of cell-cell communication via ligand-receptor interactions confusing, since ligands can in general diffuse, so might define a very different set of interacting cells than PICs. I urge the authors to be more careful / define more precisely what they mean by “an interaction”? In its current form, I don’t understand clearly when this refers to a general interaction between two cell types,. e.g. {T cell—Dendritic cell}, and when it refers to an interaction between/mediated by (the products of) two genes, e.g. cytokines/receptors/adhesion molecules, etc. This is an essential point as it is key to the paper. It also influences statements such as “do not allow for new interactions to be identified” (line 18) - cell/cell or gene/gene interactions?

2. Details on the two-step LDA procedure are not sufficient. Please explain the intuition behind this procedure in more detail (a diagram may help here). Specific questions: is it reasonable to assume that cells are not interacting in reference population? Surely some could have been/were interacting by chance before they were dissociated and sequenced? Use of the phrase “fixing topics” is not clear - as in: fixing the number of topics or fixing the entire distributions? “By fitting LDA on the co-cultures of one cell type in a dataset…” does this really mean one cell type? Or one sample of cells? My understanding was that you are fitting LDA for the _set_ of cell types that are present in the entire (non-interacting) cell population sample? Please explain.

3. For the number of topics: the example of table 1 is not clear. The cosine distances at 30 topics and 100 topics are very similar, implying potentially high sensitivity of # topics determined to the input data. Given the four distances calculated, it really begs the question: what are the cosine distances for topics < 30 and > 100?

4. “in the case of our synthetic experiments, we plot how choosing between 2 and 20 genes per topic affect the true positive and false positive rates” - where are these plots?

5. Why is adding 10 counts to the expression of a gene a good model for evidence that its expression is influenced by interaction — What is justification for this? Does the absolute change in counts used affect the results? More generally, more discussion of the results using synthetic data is needed. Figs 1 and 2 contain little information - what other plots would provide more detailed summaries of the inference results? e.g. (at least) plot more single genes than 2 in Fig 1; in fig 2 would be more helpful to provide total # mis-classified genes since the absolute numbers are small (ROC curves not v informative). Also: what is the effect of the abs count increase (if less than or greater than 10) on the results obtained? What is the detection limit? i.e. Can the method still pick up these interactions if they occur in 1 or 2 or 3 pairs of cells?

6. PIC-seq data: in Fig 3, it looks like there are some topics with very little variation across cells (e.g. topic 3), how to explain these? Also, I am curious what topics that pick up specific subsets of cells represent? e.g. topics 25 or 26. What are the top genes driving these? It would be helpful to include (supplementary) figures with genes contributing to topics for at least a good subset of the topics identified in Fig 3.

7. COVID-19 data: are the inconclusive results here due to: use of DoubletFinder + LDA not able to resolve interactions well (as the authors suggest), or due to the complexity/particularities of this dataset itself? Was DoubletFinder + LDA tried on any other dataset? In particular, there are a large number of clusters in the COVID dataset: perhaps trying DoubletFinder on a simpler (by # of cell types) 10X dataset would yield better results?

Minor points

- line 399: While this is informative, the current setup of the 10x Chromium protocol is fully suitable for studying cellular crosstalk of physically interacting cells” - is this missing “not”?

**********

Have the authors made all data and (if applicable) computational code underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data and code underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data and code should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data or code —e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Figure Files:

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

Data Requirements:

Please note that, as a condition of publication, PLOS' data policy requires that you make available all data used to draw the conclusions outlined in your manuscript. Data must be deposited in an appropriate repository, included within the body of the manuscript, or uploaded as supporting information. This includes all numerical values that were used to generate graphs, histograms etc.. For an example in PLOS Biology see here: http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1001908#s5.

Reproducibility:

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Dear MS. Pancheva,

Thank you very much for submitting your manuscript "Using topic modeling to detect cellular crosstalk in scRNA-seq" for consideration at PLOS Computational Biology.

As with all papers reviewed by the journal, your manuscript was reviewed by members of the editorial board and by several independent reviewers. In light of the reviews (below this email), we would like to invite the resubmission of a significantly-revised version that takes into account the reviewers' comments.

The reviewer #2 still has major concerns. It's important in the re-submission that those concerns will be fully addressed. 

We cannot make any decision about publication until we have seen the revised manuscript and your response to the reviewers' comments. Your revised manuscript is also likely to be sent to reviewers for further evaluation.

When you are ready to resubmit, please upload the following:

[1] A letter containing a detailed list of your responses to the review comments and a description of the changes you have made in the manuscript. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

[2] Two versions of the revised manuscript: one with either highlights or tracked changes denoting where the text has been changed; the other a clean version (uploaded as the manuscript file).

Important additional instructions are given below your reviewer comments.

Please prepare and submit your revised manuscript within 60 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email. Please note that revised manuscripts received after the 60-day due date may require evaluation and peer review similar to newly submitted manuscripts.

Thank you again for your submission. We hope that our editorial process has been constructive so far, and we welcome your feedback at any time. Please don't hesitate to contact us if you have any questions or comments.

Sincerely,

Qing Nie

Associate Editor

PLOS Computational Biology

Ilya Ioshikhes

Deputy Editor

PLOS Computational Biology

***********************

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: Thanks to the authors for largely addressing my points.

Two of the minor points I mentioned previously seem to have been missed:

1. The authors still assert that "α and β are Dirichlet priors that control..." -- but I think these are parameters of Dirichlet priors, not priors themselves.

2. Section title: "Ranking genes potential candidates of interaction" - I am not sure this currently makes sense. Is there a missing word?

Reviewer #2: In this revision the authors have addressed many of the concerns and the manuscript is improved for it. However, important questions remain, these are listed below.

1. Regarding the choice of number of topics, I understand that this is in general a challenge that perhaps no single metric can adequately quantify, but I still find the discussion in light of the metrics used (cosine distance and Jensen-Shannon divergence) to be inconclusive/not informative enough practically for potential users. For both these metrics the differences between a small and large number of topics is minimal. I am not sure what the observation of Fig. S5 adds? Is it not entirely expected that some of the topics from a many-topic model be contained (to some extent) in the topics of a few-topic model? Another question is: why did the average cosine differences in Table 1 change between the initial submission to the revision?

2. Practically, my concern is that there are no clear guidelines on choosing the number of topics for a user. Judging on cosine dist or JS, one would choose low (10) or low/high (10/150) numbers, respectively. Yet in examples, e.g. PIC-seq dataset, 30 topics are used. This may be due to the taking into account of “complexity” (which is not to my understanding well defined). For this method to be useful and adopted, this must be clarified. I think it will also be important to see some comparison of results with different numbers of topics, e.g. how do the heatmaps look for the PIC-seq dataset with 10 or 50 or 100 topics? These could be included as SI figs.

3. For new Fig 2, I appreciate the plots of additional genes in SI. However, the specific genes used are concerning, since the unmodified genes are almost all unrelated to immune cell phenotypes, i.e. characterizing general (housekeeping) cell processes (ribosome, actin). I think it is important to assess this in the analysis: switch the gene sets, and modify the expression of those in Fig S2, leaving the genes from Fig S3 unmodified, how would this affect the predictions of LDA?

4. Fig 3 please add total # genes predicted to the legend so the number of false positives can be understood in context.

5. For the PIC-seq data, thanks for the additional plots, they help with the analysis. A couple questions do arise. Why don’t the cells cluster by time point within type (T cell or DC) in Figs 4, S6? More importantly, I am still left wondering about the number of topics used here. In total, about 1/3 of the PIC-seq topics have been described by the authors, which is indeed helpful although some of them are not informative about immune processes as they contain mostly housekeeping genes. What are the remaining 2/3 of the topics characterizing? I would not hammer on this point so much were it not (as I understand it) the essential core of the methods. Thus I would like to understand if the other topics represent signal (if so what processes?) or background? If the latter, the method may still be useful but there needs to be some way to understand/characterize/rank(?) The topics.

**********

Have the authors made all data and (if applicable) computational code underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data and code underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data and code should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data or code —e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Figure Files:

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org.

Data Requirements:

Please note that, as a condition of publication, PLOS' data policy requires that you make available all data used to draw the conclusions outlined in your manuscript. Data must be deposited in an appropriate repository, included within the body of the manuscript, or uploaded as supporting information. This includes all numerical values that were used to generate graphs, histograms etc.. For an example in PLOS Biology see here: http://www.plosbiology.org/article/info%3Adoi%2F10.1371%2Fjournal.pbio.1001908#s5.

Reproducibility:

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. Additionally, PLOS ONE offers an option to publish peer-reviewed clinical study protocols. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols

Dear MS. Pancheva,

We are pleased to inform you that your manuscript 'Using topic modeling to detect cellular crosstalk in scRNA-seq' has been provisionally accepted for publication in PLOS Computational Biology.

Before your manuscript can be formally accepted you will need to complete some formatting changes, which you will receive in a follow up email. A member of our team will be in touch with a set of requests.

Please note that your manuscript will not be scheduled for publication until you have made the required changes, so a swift response is appreciated.

IMPORTANT: The editorial review process is now complete. PLOS will only permit corrections to spelling, formatting or significant scientific errors from this point onwards. Requests for major changes, or any which affect the scientific understanding of your work, will cause delays to the publication date of your manuscript.

Should you, your institution's press office or the journal office choose to press release your paper, you will automatically be opted out of early publication. We ask that you notify us now if you or your institution is planning to press release the article. All press must be co-ordinated with PLOS.

Thank you again for supporting Open Access publishing; we are looking forward to publishing your work in PLOS Computational Biology. 

Best regards,

Qing Nie

Associate Editor

PLOS Computational Biology

Ilya Ioshikhes

Deputy Editor

PLOS Computational Biology

***********************************************************

Reviewer's Responses to Questions

Comments to the Authors:

Please note here if the review is uploaded as an attachment.

Reviewer #1: The authors have addressed my previous points.

Reviewer #2: I appreciate the authors diligence in addressing my questions and providing addition plots & support to justify the choice of the number of topics. In this revision all of my concerns have been addressed.

**********

Have the authors made all data and (if applicable) computational code underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data and code underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data and code should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data or code —e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No