ludvigla
commented
2 months ago Hi Whitney,
I would try to run NMF on the batch corrected data and the count data. I can't tell you what the most appropriate strategy is because I don't know, but I would expect to get quite similar results. I assume you are using the singlet R package for NMF so maybe you could reach out to the package author and ask him for advice.
If you run it on the raw count data, you might be able to pick up factors which correlate with batches. If this is the case and you have good reason to believe that these are "technical" batch effects, you could simply focus on the other factors that pick up biological signals.
I find it useful to plot the factors on the H&E images to see if they align with morphological features, which can help determine if they are biologically relevant or not.
If you want to share details about the experimental design and batch correction strategy we might be able to help out more!
Hope this helps, Ludvig
Hi! Thank you for the exceptional package.
I am hoping to run NMF on 36 integrated visium reactions. They have been processed with the Seurat pipeline and integrated after SCT transformation. Do you have advice on which assay I should use to run NMF on? Would I run it on the counts slot because those values will be non-negative? How can I account for batch correction/integration?
Thank you!