Dear Prof. Dr. Fischer,

Thank you very much for submitting your manuscript "Recognition and reconstruction of cell differentiation patterns with deep learning" 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,

Jinyan Li

Academic Editor

PLOS Computational Biology

Daniel Beard

Section 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 manuscript submission by Dirk et al describes a series of machine learning algorithms designed to investigate the characteristics associated with inner cell mass (ICM) organoids, which capture an early stage of blastocyst morphogenesis in cultured murine embryonic stem cells. The authors generate a large synthetic dataset of 2D and 3D cell configurations that describe binary state transitions of differentiation according to a “dispersion” parameter that accounts for influences on a cell ranging from only nearest neighbors to influences from cells further away and yields striking differences in multicellular patterning of the system. By generating data from thousands of organoids, a graphical neural network (GNN) model was trained and validated to predict the dispersion parameter value for a given image. The GNN predicts for actual experimental organoid images a range of dispersion values. Another ML model generates predictions of a cell phenotype (one of 4 classes based upon Nanog/Gata6 expression) solely based upon information regarding the closest neighbors’ phenotypes. While the model performs excellent for the synthetic dataset, accuracy of the experimental images is good but with considerable room for improvement. Primary claims in the paper are 1) algorithms can detect patterns that humans can’t; and 2) cell fate reconstruction can be predicted based upon a subset of the nearest neighbors.

The rigor and implementation of the machine learning in this paper is quite good. The use of a computational model yielded insights into the perception of cellular patterning detectable by ML versus human. However in the paper there are several critical shortcomings.

MAJOR:

1) Throughout the paper, the authors make claims (including the title) that the prediction of an individual cell’s phenotype from the neighboring cell information is pattern reconstruction. I find this misleading, one can think of this more as “spatial imputing” in which enough information about the surroundings allows for an estimate of what is happening at the center cell. That is quite different than pattern prediction which occurs on the macroscale.

As a sidenote, the motivation for the entire exercise is a bit weak. Why do we want to predict a single cell’s fate? Pivoting the emphasis toward imputation for spatial transcriptomics data acquisition could be an improvement to the paper.

2) The only description of the model is a reference to a pre-print that is not peer-reviewed. The authors need to provide enough description so that the readers can understand the nature of the model. Is it agent-based? Does the model contain any stochasticity or will the same initialized set yield the same outcome? Why is there variability in the organoid cell numbers for dataset B? How does the reader interpret the meaning of model-defined or experimentally extracted dispersion values?

3) Similarly, the authors never show any examples of the organoid images that they are working with. Statements in the paper such as “the spatial arrangement of epiblast and primitive endoderm precursor cells is non-random but visually unrecognizable” should be backed up by these images. The authors should show an example of where inaccurate predictions occur spatially on the organoid.

4) The estimation of cell fate by both machine and expert is made from cell state and rank order distance from the predicted cell. All spatial information is lost and I don’t understand the choice in discarding this. For example, in Figure 5, if the user knew that all the red cells were clustered to the left side of the cell in question, and the blue cells were clustered to the right, that information has both biological implications (e.g. a larger localized buildup of morphogen) and implications for accuracy.

5) The authors’ choice in binning their cell data does not normalize expression levels by cell area, and subsequent analysis is based upon the centroid coordinates. How are the results impacted if normalization is performed, which will change the distribution of N-G-, N+G-, N+G+, N-G+ cells?

6) Why do bigger, older organoids have higher dispersion values? This result is not sufficiently discussed.

7) I do not understand the conclusion drawn by the sentences lines 444-447: “For the ICM organoid data, we obtained an accuracy of more than 70% for nine neighbors, which is less than the typical 14 cells that are directly in contact with a given cell. This is in agreement with a short range cell-cell communication as predicted by the pattern recognition”.

Why does having less than 14 cells needed make this a short-range process? If all the cells in the vicinity are of similar phenotype (i.e. a cluster), then the immediate cells are reflecting the distant cells, this says nothing about the mode of communication.

MINOR:

8) Throughout the text are references to “2D organoids”. These are 2D colonies.

9) The justification for discarding organoids with > 2/3 of one type is weak. How many organoids fell into this category? The radially organized organoids that appear at this stage may have a central core of less than 1/3 and highly relevant to the morphogenesis under investigation

Reviewer #2: Dirk et al combine machine learning (ML) algorithms and mathematical models to analyze spatial patterns of cellular fate decisions. More specifically, they assume a mathematical model of cellular fate decision where cells can interact with their neighbors and adopt two different fates. In this model, a parameter, q, sets the range of cell-cell interaction. Then, they train ML algorithms on patterns of cell fate decision generated in silico via their model to predict the parameter q, given a specific spatial distribution of cell fates (what they call "pattern recognition" task). Additionally, they train ML algorithms to predict the fate of a cell from the fates of the neighboring cells ("pattern reconstruction" task).

In doing so, they use simulated and imaging data generated from mouse ICM organoids. When applying their model to mouse ICM organoids data, they conclude that the analysis suggests the existence of a short-range cell-cell communication controlling cellular fate decision, in agreement with some experimental evidence.

The approach used by the authors (the combination of ML with mathematical modelling) is very interesting, and very much needed in biology, where there's an increasing availability of large-scale datasets from which we can extract potentially interesting patterns. However, we're still largely unable to obtain valuable insights into the molecular interactions that generate them. The combination of approaches from ML and mathematical modelling can help address this issue.

Overall the implementation of the models is sound, and the manuscript is well-written.

However, I have some points that should be addressed before publication.

1. Can one predict the value of q (in both 2D and 3D) by computing some spatial statistics used in marked point processes, such as the Stoyan's mark correlation function, the mean or variance-mark function, etc.? Before using more complex approaches based on deep learning, The authors should show that more straightforward strategies like this do not work or are less accurate. This would be more compelling than comparing the accuracy of the deep learning algorithms with the manual annotation of human experts.

2. For some values of q, the authors observed a decrease in the model performance in the task of pattern reconstruction. For example, in 2D using Model 2, the testing accuracy was below 80% for higher values of q (line 355), where engulfing patterns are obtained. Does the performance improve if, in addition to the cell fate and distances of the nearest neighbors, the cell's position is also given as input (e.g., using some sort of distance from the "boundary")? It seems like, especially in some situations (like in the engulfed patterns), the relative position of the cell might also be predictive of its fate. It would be interesting to see whether the addition of the cell position can improve the models' accuracy with the in vitro data.

3. Given that 4 categories are used with the in vitro data, the authors should generate in silico data with 4 categories as well (instead of 2) and then compare the accuracy they obtain with the in silico vs the in vitro data. This would also be useful to support their claim that the lower accuracy observed with the in vitro data is (partly) due to the greater number of categories.

**********

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: None

**********

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 Prof. Dr. Fischer,

Thank you very much for submitting your manuscript "Recognition and reconstruction of cell differentiation patterns with deep learning" 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. The reviewers appreciated the attention to an important topic. Based on the reviews, we are likely to accept this manuscript for publication, providing that you modify the manuscript according to the review recommendations.

Please prepare and submit your revised manuscript within 30 days. If you anticipate any delay, please let us know the expected resubmission date by replying to this email.

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

[1] A letter containing a detailed list of your responses to all 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.

Thank you again for your submission to our journal. 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,

Jinyan Li

Academic Editor

PLOS Computational Biology

Daniel Beard

Section Editor

PLOS Computational Biology

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

A link appears below if there are any accompanying review attachments. If you believe any reviews to be missing, please contact ploscompbiol@plos.org immediately:

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 done an excellent job revising the manuscript to address concerns. The addition of spatial autocorrelation metrics greatly enhances the paper. I found the interactive web visualization very useful for exploring the data. I do have lingering comments that were not fully addressed by the edits:

1) The referencing of 2D systems as organoids still appears throughout the document, e.g. Figure 2, lines 229, 368

2) Lines 323-325: “We find that 24 h after organoid formation, the Moran's index is consistently larger than 0, indicating a non-random distribution of the cells (Fig. 7). For 48 h organoids, the average Moran's index increases further, indicating stronger clustering of the two cell types.”

It is unclear which dataset is being referred to here. It would be helpful to explicitly label the dataset used for this observation. Is it experimental images or simulations?

3) Unfortunately, despite the addition of Figures 8, 9, 10, S1, S2, there is an unclear degree of overlap in results from other publication. Whatever has been accepted for publication must be quite different from the arxiv link. How much of what is being presented is new analyses on data presented in the other paper versus reproductions of the same figures from that paper? Typically, with a large degree of overlap, a journal will request the accompanying pre-print to make this determination.

4) I find figure 10 very hard to interpret. I understand that the attempt is to show “good” versus “bad” organoids modeled at different q values, however the use of opacity in the image does not make sense to me. Am I to interpret this as only organoid 36 is well-captured by the Moran’s index statistic? What are the implications of this result?

Reviewer #2: The authors have addressed all the points that I've raised, and I have no further comments.

**********

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: None

**********

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

References:

Review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript.

If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Dear Prof. Dr. Fischer,

We are pleased to inform you that your manuscript 'Recognition and reconstruction of cell differentiation patterns with deep learning' 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,

Jinyan Li

Academic Editor

PLOS Computational Biology

Daniel Beard

Section 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 done a great job of addressing my last concerns.

**********

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

**********

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