Providing pre-defined clusters to feature selection from annotated cluster cell types via paired scRNA-seq data

[newtonharry]

Hi,

I was wondering about the validity of this approach in selecting features. You've mentioned @kaizhang that you're skeptical of the current approach because it could propagate clustering errors and noise for further feature selection. If the data I'm working with is paired with scRNA_seq, and I've annotated clusters of cells from that modality, would it be reasonable to provide the annotated cluster for the scATAC-seq data? That way, potentially valid clusters have been identified for the same cells, ideally mitigating the effects of clustering errors and noise from using arbitrary Leiden resolution parameters.

I think this is more of a temporary solution before a cluster merging algorithm (inspired by Cytocipher/CHOIR) is implemented in order to resolve potential clustering errors. Since scRNA-seq features are more clearly defined, I think it makes sense to take advantage of that when possible, with the caveat that the biology of scATAC-seq doesn't directly reflect that of scRNA-seq and vice versa.

Keen to hear what you think :)