Implement reference mapping components

[VladimirShitov]

• Implement integrate/scarches component to map query data to the reference (currently only supports HLCA)
• Implement integrate/scvi_tools component to build reference and map query data on it
• Add scVI model support for integrate/scvi_tools
• Add scanVI model support for integrate/scvi_tools
• Add totalVI model support for integrate/scvi_tools
• Add scripts for downloading test files for reference mapping

[LuckyMD]

PR concept feedback: User workflow 1: User has new lung data and wants to map to known reference -> HLCA Requirements:

• QC'ed data
• Note: HLCA ref model loaded in code

User workflow 2: Build custom reference model, annotate it, map new data on top Requirements:

• Reference model file from previous integration effort
• Annotated data object (from previous integration & annotation)
• QC'ed query data

User workflow 3: User loads annotated data with same annotation schema, builds a new reference, maps new data on top Requirements:

• Annotated datasets with same annotation schema
• Model specification (model class and parameterization)
• QC'ed query data

[LuckyMD]

Workflows 1 and 2 are more important, workflow 3 has more challenging requirements (only met by CELLxGENE data corpus with cell types mapped to CL)

Note: CL is not a great representation of cell types

Other issue:

• have to trust provided annotations for this, which can be problematic

[LuckyMD]

Workflow 2 is missing:

• adapt integration modules to output ref models that you can ingest here

[LuckyMD]

To do:

• [x] Move workflow 3 to another PR -> unclear whether this workflow is useful
• [ ] Separate PR for workflow 2

[rcannood]

@VladimirShitov Please let us know when this PR is ready for another review! :)

[VladimirShitov]

The repository is ready for review. I removed unnecessary details and fixed bugs. Local tests are running fine, but the test pipeline will fail due to the lack of test files. @rcannood , can you please run the script workflows/resources_test_scripts/hlca_reference_model.sh and put the resulting files on AWS so that test script could use them?

[DriesSchaumont]

LGTM! Very nice job @VladimirShitov, thanks!