SDTC-CPMed / scDrugPrio

Network analyses of single cell-based digital twins for personalized treatment of inflammatory disease
10 stars 1 forks source link

DCA on count matrix: incompatibility with recent tensorflow and python versions #1

Closed wenjiezeng08 closed 6 months ago

wenjiezeng08 commented 8 months ago

Thanks for the package! I have a question about the DCA on the count matrix. I checked both the reference and the original DCA package tutorial, and it seems that the package has not been updated for a long time to keep up with the latest tensorflow version.

It requires a tensorflow version >=2.0 but <2.5 to run DCA on the count matrix. However, the current tensorflow version compatible with the latest python version 3.12 is tensorflow 2.16, which is incompatible with DCA.

I just wonder how did you solve this issue. Is there any alternatives?

Thanks!

MartinSmel commented 7 months ago

DCA is not part of scDrugPrio, see our publication: https://rdcu.be/dBWTN.

We are aware that DCA has not been updated which might lead to problems. Though not tested, other denoising methods might work similarly well to DCA and one could therefore consider replacing DCA with another denoising method. One option could be to use scvi (https://docs.scvi-tools.org/en/stable/index.html), where function 'get_normalized_data', which can be used after training the model, should return analogical results to DCA.

When we ran DCA we applied it to each data set through calling the python application from the terminal. In this setting we used python version 3.6 and tensorflow version 2.0.