Peer review at Journal of Chemometrics (CEM-22-0030)

Reviewer 1

[x] Pg 5 Ln 33-35 It would be recommended to reveal the total mass of the tablets as the API content could be better interpreted, and it helps to understand the Figures. ACTION: Natalie to find tablets and weight
[x] Pg 7 Ln 24 The msbackadj function is used to correct the baseline of mass spectra. However, the spline approximation method is a very hard method to apply for Raman spectra, where broader peaks appear (mainly in the case of polymers). This approach can lead to information loss. So, it is not surprising that only the sharp peaks are visible in Fig. 1F. Using a proper baseline correction method may be crucial in this study. ACTION: Dimitrios to draft text saying we had tried
[x] Pg 7 Ln 50 There is no Table 1 in Supplementary information ACTION: Adam to make table
[x] Pg 9 Ln 20-29 Why did tablet A show very low signals for paracetamol while the content is as high as in the case of tablet D. Both tablets A and D have film coatings. ACTIONS: Natalie to describe differences betwenn paracatemol levels in A and D between MS and Raman
[x] Pg 10 Ln 33-34 The cosine similarity between types B and C looks too high, almost identical. However, previous spectral analysis (see Fig. 1G) revealed differences between the Raman spectra of types B and C. The caffeine content should have a much more significant effect on spectra. ACTION: Adam to add comment re why cosine sim between Raman B and C is v high
[x] Pg 11 Ln 27-36 The high rate of misclassification (17 and 33%) at the best LDA method is surprising. Types A and D had lower cosine similarity, yet the method distinguishes more similar tablet spectra (types B and C) better. ACTION: With rebutal from above. Couple of sentances to explain why this is the case?
[x] Pg 15 Ln 38-39 The two analytical techniques provide complementary information. It would be worthwhile to think about data fusion, allowing for more efficient classification. ACTION: Agree. Add sentenct Typos:
[x] Pg 4 Ln 60 lineiar discriminant → linear discriminant
[x] Pg 5 Ln 51 tarch → Starch
[x] Pg 11 Ln 42 SFM → SVM

Reviwer 2:

[ ] What is the significance of support vector machines performing being the method of choice for the electrospray data whereas linear discriminant analysis is the method of choice for Raman? Clearly, the classification problem posed by Raman is easier than the classification problem in desorption electrospray. The author should elaborate upon this point in their manuscript.
[x] How was the data preprocessed for the different pattern recognition methods used in this study? Was the same method used?

adamjtaylor / tablet-classification

Peer review at Journal of Chemometrics (CEM-22-0030) #24