Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript is a valuable contribution to the field of distributed deep learning in Raman spectral classification, especially in the area of pathogen detection in Litopenaeus Vannamei. The focus on data enhancement within deep learning networks is particularly commendable and holds promise for applications in spectral analysis. However, I recommend several improvements:

The manuscript should address the role and optimization of feature selection in the algorithm more comprehensively. Including references such as PMID: 33547786 and 33532820 would be beneficial, as they offer insights into effective feature selection strategies, like L0L1 regularization, which could enhance your work.

Improving the English language usage would significantly enhance the manuscript's readability and professional presentation.

Expanding the literature review to cover recent studies would provide a more robust background, helping to justify the study's relevance and situate it within current research.

Include more detailed information about the data set used, such as size, source, and sample characteristics. This would enhance the transparency and replicability of the study.

The legends, such as in Figure 1, need more detail. For example, the description of the proposed network's framework, including the RSEM, RSDM, and DLCM modules, could be expanded for clarity.

Comments on the Quality of English Language

Improving the English language usage would significantly enhance the manuscript's readability and professional presentation.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Conclusively, my suggestion for this article is to accept it after a minor revision. The manuscript contains only some minor methodological deficiencies. Correcting these deficiencies can make this article better. They are as follows:

1. I suggest the authors calculate the accuracy for each class, not just one averaged value.

2. The authors may compute the AUC value for classification accuracy.

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript cannot be published.

- The authors of the paper trained a 1D signal classifier using three components: a GAN for data augmentation, an autoencoder for denoising, and a neural network for classification. In each of them, the authors introduced either new methods or improvements to existing methods, but there is no baseline to compare against. Each improvement has to be independently evaluated against existing methods. A single ablation study is not sufficient.

- The UNet encoder-decoder design has a critical flaw - the shortcut connections may resolve the vanishing gradient problem, but it also leaks the information from the early encoding layers to the decoding layers, therefore the assumption that the network will compress a signal to key features is no longer valid. The effective denoising layers may only be the layers on the shortest path.

- The classification result does not have significance. The task is trivial: there are only four classes to be identified; there are distinct features on the 1D data for each class; and the test set does not have negative samples. The accuracy should approach 100% under such a setup. The augmentation made by previous steps is only effective in preventing overfitting. When it turns to real-world applications, the network will always give a false positive on any bacteria other than these four types, i.e., the accuracy is infinitely close to 0.

In summary, the paper introduces a stack of fancy things but the significance is not established for any of them. Despite the interesting background and research motivation, the designed task is trivial and the proposed solution has no real-world value.

Comments on the Quality of English Language

The writing includes grammar and formatting errors. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The authors used deep learning approach to classify litopenaeus vannamei pathogens from few Raman spectra. The authors used generative adversarial network (GAN) to accomplish the task. Further, they used learning classification model (DLCM) to train the network. Although interesting, there are few questions that need answers.

1. How many number of spectra (at a minimum) are generally required for this?

2. How many independent spectra were used for the prediction? Were there any overlap of trained spectra and the prediction spectra?

3. Is the model generalized - if it works on "litopenaeus vannamei pathogens", can it work on other virus or even other cells (e.g. cancer cells)? What needs to be modified to make it work for other cell types?

4. The ROC curve should be plotted. 

5. Since, the accuracy is dependent on the number of epoch, is there a criterion on how many epoch should be employed?

Comments on the Quality of English Language

It's adequate.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

Overall, the manuscript describes an interesting topic of how to improve on the quality of classification of bacterial specimen, measured by Raman spectroscopy. The manuscript covers all aspects of the work reasonably well although on the one hand there are a few sections which contain repetitive statement / descriptions, while in other parts a bit more in depth discussion would likely help potential readers. However, these are only very minor niggles.

     One major grievance I have is associated with the appearance of the figures. The content is displayed very well, providing all required scientific details. However, in most figures the annotation is next to illegible: the font size is much to small, even when zooming the page on a large computer screen. This is an annoying feature I increasingly encounter during reviewing, and when reading online articles.

     In the attached list for figures I make requests of enlaging font sizes to levels being readable in print (normally 1.2-1.5 mm).

Comments for author File: Comments.pdf

Comments on the Quality of English Language

In general, the English is OK. A few typos have crept in which should be eradicated. A list is given in the accompanying pdf-file.

    

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed all my concerns.

Author Response

Thank you again for your suggestion and kindly advice! 

Reviewer 3 Report

Comments and Suggestions for Authors

The authors did not addressed the fundimental flaw of the paper in the revision and the changes are cosmetic. To reiterate, the paper presented a overly complex system to solve a problem that is trivial from the perspective of a classifier, and the method has close-to-zero accuracy in real-world applications.

Comments on the Quality of English Language

The writing can be improved.

Author Response

Please see the attachment

Author Response File: Author Response.pdf