Suggestions for the main README.md #1
Description
Based on my efforts to reproduce Utkarsh's results, I compiled the following suggestions for the main README.md file:
- Content of the Processing directory: mentioning 'Processing' here is confusing, since for someone who enters the project via GitHub there is no Processing directory.
- Content of the Processing directory, Data: mentioning 'Data' here is confusing, since it is not available at GitHub. Of course the current text in README.md should be clarified and add a link should be added to where the data is stored (Zenodo; note: check under which conditions we can do this for the Tabula Sapiens data).
- Indicate which deconvolution methods are benchmarked. Also indicate which methods are R-based and which are Python-based.
- Indicate which Python version to install. I used 3.9.21, which came with Miniconda3-py39_24.11.1-0-Windows-x86_64.exe. Probably any 3.9 version will work (but definitely not >=3.10, since then config.py will throw an error)?
- For those who only want to run the R-based deconvolution methods, indicate which Python packages (and their versions) to install:
- numpy==1.22.0
- anndata==0.8.0
- pandas==1.3.4
- (not sure whether this is really required): numexpr==2.9.0
- Module-2:
- When explaining the five arguments, refer to them as arg1,...,arg5 instead of a,...,e.
- Make more explicit that the 3 ST datasets are the ones used in the manuscript. Now it seems as if these are three arbitrary examples.
- Module-3:
- Add Rscript Init_env.R
- When explaining the three arguments, refer to them as arg1,...,arg3 instead of a,...,c.
- arg2 (b) is defined as a range, but only a single number has to be provided. In fact if a value of n is provided then the first n cell type removal combinations are executed for the scenario indicated as arg3 (c). This should be better explained.
- Make more explicit that the code indicated here will regenerate the results for ST1 shown in the manuscript. Now it seems as if these are slightly arbitrary examples.
- Mention that for CARD and Seurat some runs will fail (and indicate which ones).
- The arguments for the standalone scripts should be better explained.
- Module-4:
- What is meant by "Note: Estimating performance metrics needs all the deconvolution results generated in Module-3."? All scripts have an argument where specific deconvolution methods can be indicated.
- As already mentioned use the same names for the (optional) arguments.
- Should "rm0" also be included as an argument when only executing the scripts for specific removal scenarios? At least that's what I did, but I don't remember if things went wrong if I didn't.
- Module-5:
- Better explain what is meant by "Note: this module expects calculations results from Module-4"
- Remove "(list of methods and removal scenarios are adapted from JSD/RMSE calculations in Module-4 )"
- Known issues, 2. Compilation failed error on Windows OS:
- rtools -> RTools
- Indicate the required version of RTools (4.0) and include link to it: https://cloud.r-project.org/bin/windows/Rtools/rtools40.html.
- I would also mention this much earlier as a basic requirement for those using Windows and not only under 'Known issues'.