Identification of Browning in Human Adipocytes by Partial Least Squares Regression (PLSR), Infrared Spectral Biomarkers, and Partial Least Squares Discriminant Analysis (PLS-DA) Using FTIR Spectroscopy

Round 1

Reviewer 1 Report

This work reports the identification of browning in human adipocytes by partial least squares regression (PLSR), infrared spectral biomarkers, and partial least squares discriminant analysis (PLS-DA) using FTIR spectroscopy.

As mentioned by the author the investigation of adipocytes with PLS-DA using FTIR spectroscopy has been reported in the literature (e.g., reference 20 of the manuscript, note that I am not in any way related to these authors or journals in reference 20). Hence, please emphasize the significant differences (not just minor dissimilarity) between this current work and those publications reported in the literature, which could help the general readers to understand clearly on your methodology (preferably in the introduction about what is significantly novel, new and advantage point in this work). Moreover, please also include the substantial advantage point(s) in adopting your methodology/strategy. For example, justification for using PLS over other statistical methods.

Moreover, please provide quantitative comparison for your technique with others. For example, the author stated that “… complex molecular biological methodologies that require a lot of time…” (line 320-322). The author could estimate the hours needed for their strategy, including samples preparation time, such as incubation, or other steps (if any of those steps needed to be included), and that needed for others in the literature. Also please be more specific in providing an example in the name for the complex molecular biological methodologies and the time needed. This inclusion and specific description will give more solid evidence for convincing the general reader (please also apply this style for the other aspects when performing comparison as well, not only for time needed).

Additionally, please also quantitatively compare the results as well. For example, how the PLS-DA results, or the value of RMSECV compare with others (if these methods were reported in literature, please compare, otherwise please compare with RMSECV values for evaluating other types of spectral biomarkers). This comparison (not only on RMSECV) would give a more comprehensive view to the general reader on the effectiveness of the strategy.

Please be more succinct in the abstract, since the word limit has exceeded that suggested by the journal.

Please provide the spell-out version of the acronym at the first occurrence, for example root mean square error of cross-validation (RMSECV), ratio performance deviation (RPD), and so on (please check the entire script).

 

Please also comment on the possibility of overfitting in using PLS, since this technique has a tendency in the issue.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The main questions addressed in the research are identifying white adipocytes' browning using FTIR spectroscopy combined with chemometric methods. Based on IR spectra, the authors developed a method for differentiating between human beige adipocytes and human white adipocytes. The article is generally well-written, and the research is interesting and extensive.

However, I have some questions and minor comments:

1. It is not entirely clear why PLS-DA is used in one case and PLSR in the other if data in both data sets are treated as binary. For a binary response, PLS Discriminant Analysis (PLS-DA) that fits a PLS-R to a dummy variable is used. So what is the difference between the applied PLSR (section 3) and PLS-DA?

2.  The models were built for selected spectral regions. On what basis were specific spectrum regions selected?

3. You should give the number of samples in each group in Table 1 or within the main text (not only in supplementary material).

4. We should not test the model on the same data used for training (for building the model). Of course, one of the validation methods is cross-validation using a training set, but it is internal validation. Why do authors not exclude a few samples from the calibration set to test set for external validation? You can use it to evaluate the prediction power of the model.

5. What cross-validation method was used?

6. Please justify more the use of RPD, ratio performance deviation (which is just inversely related to R2). What do these values of RPd mean for this study? 

7. The RMSE value for calibration is missing. Only RMSE for cross-validation is given. Why?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thank all the authors for checking into my comments and revising the manuscript accordingly.