I have a bunch of metagenomic reads that can be grouped into the conditions healthy and disease. I profied them using ShortBRED (for Virulence Factors and Antibiotic Resistance Genes). Moreover, I used PanPhlAn to track strains of certain species (which I before profiled with MetaPhlAn). What I Am wondering about now is: Can I use MaAslin2 to search for differences between my conditions?
ShortBRED reports the frequency of gene-families in RPKM values. Since that is already normalized, does it make sense to switch of normalization and transformation procedures in MaAslin? Could I go with the standard LM model?
PanPhlAn profiles strains based on the presence/absence of genes and reports back a binary matrix. When I cluster the strains, I see patterns of strains of healthy and diseased sites splitting up. I now want to know which genes drive these varation. Can I use Maaslin for this?
Great questions! MaAsLin should work on ShortBred data similar to how it operateson HUMAnN data (see the tutorial for the newly added functional section), for both tools you end with the rel. abundance of functions within the gut microbiome - thus it should be fine to use MaAsLin on these data. The caveat here is to make sure you do not further normalize these data (e.g. TSS etc. - since TSS on RPKM data is hard to interpret).
For PanPhlAn - I don’t think we have ever validated MaAsLin on presence/absence calls like those generated by PanPhlAn. I wouldn’t use MaAsLin for this purpose. That said you could maybe do a similar analysis with the stratified HUMAnN functional results as you are looking to do with the PanPhlAn output. Here you would subset to just the species of interest and use MaAsLin to explore any associations within that species functional profile.
I hope this helps! Let me know if you have any additional questions.