ALDEx2 new version (1.32.0) output effect size difference from (1.28.1)

[davis-rudolfs]

Description

I have noticed a significant difference in the number of ZOTUs with effect sizes greater than 1 between ALDEx2 version 1.28 and version 1.32. Unfortunately I have not found information what has changed between these two versions. However, in version 1.28.1, I observed 8 ZOTUs with an effect size greater than 1, but in version 1.32.0, this number has increased to 20 (see pictures).

Expected Behavior and main question

I would like to understand what changes were made in version 1.32 that could account for this difference in results. Can you please provide some insights into the modifications made to the algorithm or any other factors that might explain this variation? How should I better approach further analysis?

Steps to Reproduce the Issue

1. Load sample data (in this case phyloseq object, same individuals, 2 separate visits).
2. Perform the ALDEx2 analysis (version 1.28.1), Bioconductor version 3.15, R version 4.2.1
3. Load R version 4.3.1, update Bioconductor to 3.17, do not change any of the previous code and perform ALDEx2 analysis (now the version should be 1.32.0)
4. Observe the difference in the number of ZOTUs with effect sizes greater than |1|.

Additional Information

• Operating System test one:
• R Version: R-4.2.1 with Bioconductor version 3.15
• ALDEx2 Version: 1.28.2

• Operating System test two:
• R Version: R-4.3.1 with Bioconductor version 3.17
• ALDEx2 Version: 1.32.0

[ggloor]

Without seeing your code, I suspect that it is the calculation of effect size that has changed. The newest version that you are using is calculating paired effect size properly (i.e., the mean effect size of differences in effect on a paired-sample basis) rather than in bulk (i.e., the mean effect size with samples unpaired). This usually causes effect sizes to increase a little bit.

I see that roughly the same number of features have a significant adjusted p-value.

You could test this by running aldex.clr(), then aldex.effect( paired=T or F) and seeing if this addresses your issue

[davis-rudolfs]

Thank you for your time and the quick response!

The number of significant features did not change, that is correct. I have checked that the difference between both ALDEx2 versions and only when aldex.effect paired= T there is a difference. Thank you for clarifying.

I was considering mentioning all features with the effect size > |1| in my manuscript, but I think I will use the p-value as a cut off for the main findings.