Thoughts on cluster/celltype filtering

[LinearParadox]

Hi all,

Loving the tool and have been playing around with it. I haven't delved too far into the gsVA/ssGSEA code, so that might be addressed there. Feel free to close if you do. But my thought is, could it be useful to implement a function to filter by cluster somehow? This mainly comes from how the background is computed. If we're running GSEA on a subset of immune cells within a tumor, for example, wouldn't the immune cells possibly have a higher background due to rows that are all zero for that cell being retained within the matrix? If this isn't the case, my apologies.

One way to accompany this would be to remove features that have < some user-defined parameter within an annotation grouping. So for example if one looks at a cell the function would remove rows that are all zero counts within its cluster/grouping before computing background

[ncborcherding]

Hey Alexei,

Interesting idea - I don't know if the zero values affect the rank based enrichment score. I would have to do testing to show possivle inflation. UCell has a upper cap on the number of genes ranked (n=1500), which might prevent that from happening, but I do not believe GSEA or GSVA have that. This might actually speed up computation time as well, as rank list generation would be smaller.

If you are interesting it trying things out or have code, let me know. I will add this to enhancement category, but will need to test things out before I start.

Nick