Fix layer handling in Mixscape.mixscape to resolve errors with adata.raw

[Lilly-May]

PR Checklist

• Referenced issue is linked (Error while using ms.mixscape - ValueError: Input X contains NaN. #617)

Description of changes

• Added use_raw=False to the sc.tl.rank_genes_groups method call
• Currently, Mixscape.mixscape fails for adata objects where adata.raw is present. This is due to the internal call of sc.tl.rank_genes_groups, which uses adata.raw by default for computation if use_raw is not explicitly set to False. If a layer is specified when calling Mixscape.mixscape, this will cause the rank_genes_groups method to fail with the error: ValueError: Cannot specify "layer" and have "use_raw=True". If no layer is specified, rank_genes_groups will use adata.raw for computation, potentially leading to errors like the one reported in Error while using ms.mixscape - ValueError: Input X contains NaN. #617.

Discussion point

• Currently, the layer is set to X_pert by default only after sc.tl.rank_genes_groups is called, meaning it will run on X unless the user provides a different layer. Should we change this? On what basis should the perturbation markers be calculated? Rather based on the X_pert layer?

[Zethson]

Thank you!

Currently, the layer is set to X_pert by default only after sc.tl.rank_genes_groups is called, meaning it will run on X unless the user provides a different layer. Should we change this? On what basis should the perturbation markers be calculated? Rather based on the X_pert layer?

The answer should be somewhere here: https://satijalab.org/seurat/articles/mixscape_vignette.html

Yeah, I think it should be called on X_pert but I thought there was even some code that set X to X_pert?

[Zethson]

Merging this now but please see my comment. We can discuss this further (on slack).