Recommendation for supplementary methods to validate MaAsLin 3 results on MetaPhlAn 4 data

Downstream analysis and statistics MaAsLin
1

Dear MaAsLin 3 team,

We are using MaAsLin 3 on MetaPhlAn 4 relative abundance data and identified both prevalence-driven (the majority) and abundance-driven biomarkers.
To anticipate potential reviewers’ requests for a “multi-method consensus,” we have three quick questions:

  1. Supplementary Method: Since we only have relative abundances, which supplementary DA method would you recommend to corroborate findings from both models?
  2. ALDEx2/ANCOM-BC2: Your 2026 Nature Methods paper applied these tools to MetaPhlAn 4 profiles. Since they require integer counts, how did you transform the relative abundances to run them?
  3. Read Depth Covariate: The paper recommends including “untransformed read depth” to prevent spurious prevalence associations. Should we use the total clean reads from our upstream QC logs for this covariate?

Thank you very much for your time and guidance!

Best regards,
Chun

2

Hi Chun,

  1. ALDEx2 and ANCOM-BC2 are the most popular other DA tools. I’d use those to corroborate findings, but like in the Figure 3 section of the manuscript, you can get pretty different results on real data depending on the tool since ANCOM-BC2 (and ALDEx2 depending on the mode) are looking for associations on the absolute abundance scale. I’d check some of the individual plots and make sure the the statistical associations match the visual associations regardless of the method.
  2. You can always get psuedo-integer counts by multiplying the read depth by the relative abundance and rounding to the closest integer. We usually did some version of this when necessary.
  3. For untransformed read depth, I’d use however many cleaned reads are going into MetaPhlAn as the count.

Will

3

Hi Will,

Thank you so much for the incredibly clear and helpful explanations!

We will definitely incorporate your practical advice on pseudo-counts, read depth, and visual verification into our workflow.

Thanks again for your time and for developing such an excellent tool!

Best regards,
Chun