Pearson residuals work well to normalize UMI data, as originally suggested by @satijalab: https://t.co/7A21KSdP8n Our prior work on Analytic Pearson residuals comes to a similar conclusion: https://t.co/bdkc68AMY8 The key assumption is that UMI counts ~
704 followers
686 followers
On genes selection, good to add that even randomly reducing no. dimensions from the total ~20k genes is beneficial when working with typical current sc dataset sizes of ~10k cells, so Brennecke or other selections also help (hopefully better than random!)
686 followers
@lpachter @arjunrajlab This paper is good for seeing how a working variance stabilisation helps for selection of „interesting“ genes (and also data visualisation) https://t.co/Ntqr5s7CTc
704 followers
With all that being said, keep in mind I might be a bit biased when looking at these results because my last paper with @hippopedoid & @CellTypist is on Pearson residuals ;-) See here: https://t.co/gsm1IZkEe8