Maaslin2 N and N.not.0 output appears incorrect


I am doing an analysis with R version 4.0.2 and Maasiln2 version 1.4.0. I noticed that the N and N.not.0 columns of the results output .tsv were incorrect; specifically, they were higher than the actual N with metadata and OTU counts (i.e., the complete cases). When I removed all non-complete cases and re-ran the analysis, the N and N.not.0 were correct.

Is this seemingly incorrect treatment of NA values something that has been fixed in the latest (developer) version of Maaslin2? Or will I need to manually filter out non-complete cases for all future analyses? If the latter, it would be helpful if: 1) Maaslin2 documentation could describe somewhere how the N and N.not.0 columns are generated. I could see some people not realizing NAs had been counted in those Ns when interpreting results, which might lead to reporting errors and 2) This was fixed in future Maaslin2 versions such that only complete cases are included in those numbers, just to make implementation as easy as possible.

Thank you for your help!


Hi @fquerdasi,

In the code, these are currently being called after normalization and transformation steps on just the feature table, before the matching to the metadata and the na.action happens. We are having a discussion on the best way to handle this now and will update the documentation etc. accordingly. Thanks for pointing this out.