MarioniLab/miloR

Choice of fdr.weighting in testNhoods

bazelep opened this issue · 2 comments

Hello,
Is there some guidance on when to use the non-default ("k-distance") option for the fdr.weighting parameter in the testNhoods() function? In the "Making comparisons for differential abundance using contrasts" vignette, the fdr.weighting parameter for the testNhoods() function is set to "graph-overlap", but it's not clear why this is chosen.
Thank you,
Peter Bazeley

Reading the issues, it seems like the k-distance is an older and much slower approach.

#99 and #249 have some details on this, but that's my read at least. I've used both on some datasets, and haven't really observed a difference in results, except the fact that k-distance takes ~15 mins on my dataset while graph overlap was seconds.

Hi @bazelep - the recommendation is to use graph-overlap from now on. This is described in the Milo2.0 pre-print: https://doi.org/10.1101/2023.11.08.566176