scATACseq + scRNAseq integration scores

[sky0m1n]

Hey everyone,

I'm having some trouble integrating scATACseq and scRNA-seq data. The samples are from the same organ/stage but from two independent experiments. I plotted the gene integration scores in the UMAP and I have low scores for some cell populations (~0.4), especially in the border between clusters. I was wondering how I could get around with this issue (or even if this is really a big deal).

• Is there a general threshold to consider - e.g., integration scores over 0.7? I couldn't find many papers mentioning this.
• Does it make sense to subset the cells based on integration score and call the atacseq marker peaks only from them? This will decrease the cell numbers in some clusters by a lot.
• Can I infer that if the integration worked for the other populations, that this result is a matter of the biology of those cell types? I.e., the cell populations on that cluster are not fully differentiated and the gene expression patterns are not completely overlapping with chromatin accessibility. They are from mesenchyme origin, so pretty pluripotent.

Anyway, I can ofc recover which cluster from rnaseq has a higher contribution to a individual cluster in atac. That should be sufficient to back up my annotation, right?

Thanks for the help!