Closed ONeillMB1 closed 4 years ago
The algorithm doesn't care where the data comes from, as long as the references are appropriate.
Now, getting appropriate references is the more difficult challenge. I don't work much with ATAC-seq data, but I used to do a lot of work with ChIP-seq data, and one obvious challenge is the fact that the features (peaks) need to be empirically detected and are not the same across experiments. In this case, everyone would have to agree on what features the references should have. Even the simplest method of creating bins would require a decision on a "reasonable" bin size.
While untested for this purpose, there's no reason SingleR couldn't work already. As @LTLA said, SingleR learns markers empirically from the reference provided, so if you have an annotated ATACseq reference set, with regions that are consistent between the ref
and test
data, theoretically, it should work.
Doubling back, @ONeillMB1, if you can generate your test
& ref
data with compatible bins, I'm quite interested to hear how SingleR works for you. It would be really great if ATAC data can be deemed "officially" supported!
closing as the issue seems resolved, or at least out of our hands.
Thanks for developing this! Any plans to extend to scATAC-seq data?