SingleR for methylation

[LeaLe88]

Dear all,

Just wondering whether SingleR can also be used for annotation of methylation cell types?
Or if not, if you have any recommendation for any package.

Thank you!

[j-andrews7]

Snagging the beta values and using that should be easy enough in theory. Seems like the sparsity might be a killer though.

Then again, imputation in scWGBS seems to not completely suck, so may be able to overcome it to a degree. Dunno, think this would take some experimenting.

Label transfer/projection mapping would probably work fine. And random forest classifiers are pretty robust with methylation data and used for array classifications all the time, e.g. the MNP classifier.

I'm guessing Aaron will have deeper insights.

[LTLA]

I don't work much with methylation data, but from what I remember, there is the question of what to do about CpG sites that don't have any coverage, i.e., the beta value is missing. I don't think SingleR knows what to do with missing values in the input data; technically, the correct approach is to just ignore such values, but that would entail some significant changes to SingleR's internals. I suppose you could just put some "neutral" placeholder value, e.g., 0.5.

Putting that aside, it's probably best to use the M-value as input instead of the beta value. This is a bit more like expression data (not bounded to [0, 1]) and should make the marker detection steps a bit happier. In this case, the placeholder would be zero.