Environment for scArches loading/application

[chartl]

It would be useful (either here or at the Zenodo project) to include the appropriate environment for applying the scArches model to new data. Updates to pytorch, scArches, numba, umap-tool, and scanpy all lead to conflicts for loading, applying, or analyzing new data during or after integration with the core set. After much trial and error, the following environment appears to work:

name: GBmap_env

channels:
  - conda-forge
  - bioconda

dependencies:
  - python==3.8
  - pip
  - pip:
    - torch==1.6.0
    - scvi-tools==0.8.1
    - scArches==0.3.0
    - umap-learn==0.4.6
    - numba==0.52.0
    - leidenalg
    - seaborn
    - jupyter

[ccruizm]

Hello @chartl, you are right. I plan to include more detailed information in the final version of the repository. I know that the new updates in several tools (especially scvi) generate challenges to running the code deposited in Zenodo.

Please find attached the yml file of the environment I used to generate the core GBmap and used for the transfer learning.

scarches_environment.zip

Sorry for the inconvenience. I hope the following updates in the code/repositories make this clearer.