issues when running bulkmodel and scmodel

[weijie25]

Hi there, I have encountered several errors and tried to solve them by modifying the code when I was using the app. I was using my own bulk expression and drug response data.

1. the input bulk expression data are expected to have the same rows (same CCLs) as the drug response data. there's plenty of NAs in a typical drug response database such as GDSC, CTRP, and PRISM. I wonder if/how the drug response missing values were imputed?
2. pp.read_sc_file often reports error because it could not interpret the separators in the file, even with the pre-built single-cell data. A quick fix I discovered is to force sep=',' in this method.
3. only uppercase drug names (columns of drug response data) are supported.
4. the --missing_value did not seem to work when I had missing values in my data, either as 1 (default) or NA. the model training step with pytorch will run into error due to more than 2 classes when there's missing (resistant, sensitive, and the missing indicator).
5. i am using a non gpu device and line 200 in scmodel.py (loss_s = torch.tensor(loss_s).cuda()) seems to cause error. loss_s = torch.tensor(loss_s).cpu() worked fine.

[juychen]

Hi, thanks for your suggestions and questions. We are fixing the bugs and seeking help from my colleague about the results issue.

[weijie25]

Hi, thanks for your suggestions and questions. We are fixing the bugs and seeking help from my colleague about the results issue.

Thank you. Does scDEAL support users' own single-cell data? Or currently you only have the pre-built data with the GEO accession numbers?

[juychen]

Hi, thanks for your suggestions and questions. We are fixing the bugs and seeking help from my colleague about the results issue.

Thank you. Does scDEAL support users' own single-cell data? Or currently you only have the pre-built data with the GEO accession numbers?

Hi, we have only tested the functionalities on the pre-built data at the moment. However, we keep the option of setting python scmodel.py --sc_data [Your own data path] that receives alternative input.

Now we have updated the code to include new functions. You may try to start from the very beginning by following the new documentation.