GFLASSO-LR: Logistic Regression with Generalized Fused LASSO for Gene Selection in High-Dimensional Cancer Classification

Round 1

Reviewer 1 Report

In this paper, the authors tackled an important problem of automated selection of important genes in high-dimensional cancer classification which effectively can be cast to the feature selection problem. The topic is certainly worthy of interest, especially given the extreme dimensionality of the data acquired nowadays in the field, and the presented ideas are sound. There are, however, several issues which might be tackled to further improve the quality of the manuscript:

1.       It would be useful to announce the structure of the manuscript in the introductory section of the paper.

2.       The authors tend to use “novel” while discussing their method – if the algorithm or a method is novel then it speaks for itself, whereas if it is not, then adding “novel” will certainly not help.

3.       What is the theoretical (O-notation) time and memory complexity of the proposed pipeline?

4.       We are currently facing the reproducibility crisis in the machine learning field – to tackle this issue, the authors should provide a link to a repository containing their implementation together with scripts showing how to reproduce the results reported in this work.

5.       While looking at the tables it seems a bit weird to see the number of genes being a floating point number, e.g., in Table 3. It would be useful to emphasize (in the caption) that it is an aggregated value across the folds.

6.       Although the manuscript reads well, it would still benefit from careful proofreading – there are open parentheses around (e.g., in the table captions), I also spotted several grammar errors here and there. Also, please see the very end of page 14.

7.       The authors should back up their quantitative claims with appropriate statistical testing – there are some differences (e.g., in the ablation study) which seem to be fairly minor. Are those statistically significant? Please report appropriate p-values.

Overall, the English is fine, but the manuscript still requires proofreading.

Author Response

Please check the attached Responses to Reviewer #1

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper focuses on DNA microarray technology to pinpoint significant genes implicated in tumor development (cancer), to aid the development of sophisticated computerized diagnosis tools.

The abstract can be improved by adding one or two sentences on the general context before presenting what has been done in this paper.

Please avoid the term “we” in the paper to increase its quality.

Line 64: A synthesis on methods presented above has to be introduced for explaining clearly what this paper adds to the research of this topic. This synthesis will increase the quality of the paper.

I think that a specific section on the background will increase the quality of the paper. The question is what has been done in the literature and what is directly relevant to this paper and its research. Maybe sentences to link sub-sections are required. The added value of the paper has to be linked to the previous works in the literature.

Lines 139-140 : a more detailed sentence is required to link the previous section to the following.

I really need after each paragraph to understand why a method has been presented, and what is the problem related to this paragraph and that could be solved by the method presented in the paper.

I find a section 2.1.1 but no section 2.1.2!!! Please check the titles of your sections and subsections.

Does the subsection 2.2.2 is your research or a background. It is really important to be clear.

The approach that is proposed seems good but has to be presented in response to the inconsistencies or problems noticed in the literature.

There is no section 2.3.2!!!!Idem for the section 3.3.2

The section on the result is clear.

A discussion section would increase the quality of the paper, by showing what has been solved and how it improves the literature.

Good language!

Author Response

Please check the attached Responses to Reviewer #2

Author Response File: Author Response.pdf

Reviewer 3 Report

1. Line 91, " M the number of samples" should be " M is the number of samples".

2. Line110, "parameter of the model which render the following maximum value" should be "parameter of the model which renders the following maximum value".

3. Table 5, "Performances" should be "Performances".

4. Table 5, "Acccuracy" should be "Accuracy". The accuracy is misspelled for several times.

5. Line 407, "Prostate T umor, andProstate.Theseresultsstem f romexperimentsemployingarangeo f genesele" you main accidentally put these text in the latex formula and quoted by "$"

6. In Figure 3, you may need to keep the first 2000 iterations since your objective function won't decrease after 2000 iterations.

7. Line 106, since xi' is a vector, you may consider using boldface to denote the vector. The same rule applies to the beta.

8.  Line 110, " the latter has a value in ]0, 1[" should be " the latter has a value in [0, 1]".
9. Line 310 to 320, I am guessing that you are describing an algorithm here, but it is not properly presented due to unknown compilation errors. Please correct it.

10. In Figure 2, I recommend removing the part of the figures after 9704 iterations since no information is presented.

11. In Table 2, I recommend truncating the numbers, keeping only two digits after decimal, and making the table smaller.

12.  In section 3.2.2, for selecting the appropriate lambda parameters, you may use the grid search method, which is more efficient.

13. The subgradient method typically converges slower than other methods, like ADMM and Proximal Gradient Descent, you may consider comparing the converging rate with other methods.

Author Response

Please check the attached Responses to Reviewer #3

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thank you for addressing my concerns.