The motivation for this post was also recent question on twitter asking how one might approach the analysis of an AB/BA crossover study when the goal is to estimate the difference in the relative abundance of microbial taxa under treatments A and B.
The motivation for this post was a recent twitter exchange discussing difficulties faced when analyzing microbiome data with missing covariate values (i.e., incomplete sample metadata fields). I have run into this issue as well, and given the increasing size of microbiome studies, expect so to have many others.
Inherent limitations with one-at-a-time (OaaT) feature testing (i.e., single feature differential abundance analysis) have contributed to the increasing popularity of mixture models for correlating microbial features with factors of interest (i.
I was recently asked how many samples we might need to formally power a differential abundance analysis of prevalent species in the human stool microbiome between “otherwise healthy” control samples and patients with a specific disease condition.
I recently analyzed some data from an experiment with a pre-post study design where our estimands of interest were the baseline adjusted mean difference in Shannon diversity and species differential abundance between arms at post-treatment.
In a previous post, I highlighted some difficulties one faces with differential abundance (DA) testing when trying to identify those features (i.e., OTUs/ASVs, species, pathways, etc.) that “differ the most” according to some condition of interest.
Microbiome studies often seek to identify individual features (i.e., OTUs/ASVs, species, pathways, etc.) associated some condition (i.e., exposure, experimental treatment, etc.) of interest. This problem can be approached in many different ways, but most commonly, one-at-a-time (OaaT) feature screening is undertaken.
I recently attended a seminar where Wrench normalization was shown to have good performance in the presented simulation studies. So I thought I would give it another look.
A common goal in many microbiome studies is to identify features (i.e., species, OTUs, gene families, etc.) that differ according to some study condition of interest. While often done, this is a difficult task, and in the Introduction to the Statistical Analysis of Microbiome Data in R post I touch on some of the reasons for this.