Alternative methods for sorting input markers

[zrlewis]

Hi, I have enjoyed trying out this package. The results are surprisingly informative and it is really nice to get a summary of cell type characteristics based on marker expression without having to look up all the genes that I am unfamiliar with. It is a great starting point for analysis with unfamiliar cell types.

I am wondering about including a feature where there are options for how to sort the DE marker marker_list input. I see that you currently appear to have thresholds for values like p_val_adj, and pct.1, etc., and then sort on avg_log2FC. However, sometimes I find it most effective to just do a simple sort by p_val_adj and take the top n genes.

[zrlewis]

One other thought to tag on here: I think that the CASSIA approach is best suited to the highest taxonomic level of clustering, where you have key cell-type specific marker genes as well as genes that may distinguish some subtypes on a per-cluster level. However, when you have already subclustered to say only neurons and then run CASSIA, the cell typing may be less accurate because there isn't a set of neuron-specific shared markers to help with the cell type assignment. So thinking of ways to traverse a hierarchy of cell type information would be beneficial such as providing the option to do nested sets of DE markers.

[ElliotXie]

Hi @zrlewis!

Thanks a lot for your interest in CASSIA and for the thoughtful feedback!

It's super helpful to hear that alternative ranking and filtering strategies can work better in some cases. We’ll definitely add support for customizing how marker sets are selected and ranked in the next update.

Regarding your second point, we completely agree, and we’ve actually been working on that recently too! One idea we’re testing is to bring the subcluster labels back to the original dataset and run FindAllMarkers again, or to include higher-level clustering markers directly in the prompt to give more context when annotating subclusters. We're planning to roll these into the next major update as well.

Again, really appreciate you taking the time to share these ideas!

[zrlewis]

You're welcome and it's great to hear about the work in progress. I'll watch this space.
Cheers, Zack