radEmu

radEmu is an R package for estimating changes in the abundance of microbial taxa using amplicon or shotgun sequencing technologies.

If you are a microbial ecologist or bioinformatician, some of the things that you may like about radEmu include

radEmu uses your amplicon or shotgun sequencing to estimate changes in the "absolute abundance" of microbial taxa. Here, "absolute abundance" could be interpreted on the cell count, cell concentration or DNA concentration scale. Yes! It's true!
- We know this sounds magical! You can check out Section 2 of the manuscript for details.
- In brief, we can't recover the absolute abundance of taxa in any individual sample from amplicon or shotgun sequencing. However, we can estimate fold-differences in abundances across samples.
radEmu formalizes some of the nice things about log-ratio-type methods for differential abundance, including
- radEmu is robust to differential detection of taxa, so you don't have to worry about (e.g.) the different extraction/PCR efficiency of your protocol
- radEmu is robust to unequal sampling effort. No need to rarefy! (Actually, please don't.)
- radEmu deals with zeroes natively, without any need for arbitrary parameters like pseudocounts
- radEmu does not require that you have a "reference taxon" that is not changing in abundance across samples
  - Instead, radEmu estimates differences in abundance across taxa
Amongst existing methods, radEmu is most similar in flavor to ALDEx2 and ANCOM (and ANCOM relatives), but doesn't require priors, log-ratio transformations (and thus pseudocounts), nor a reference taxon!
radEmu can adjust for relevant covariates, including precision variables and confounders
radEmu achieves all of the above by jointly modeling all taxa (i.e., it's not a taxon-by-taxon model like corncob). This makes it harder to parallelize, but fortunately testing can be parallelized easily. (There's is an example in the preprint's supplementary material, but let us know if you want a tutorial on how!) On a standard desktop, radEmu can handle 1000 taxa, 800 samples and 12 covariates. You may want to get a 35-minute coffee break while it runs, though.
radEmu is publicly available in open-source software... right here!

If you are a statistician, some of the things that you may like about radEmu include

A clearly defined and interpretable target estimand
Fast algorithms for estimation under alternative and null hypotheses
Type 1 error rate control even under pathological distribution misspecification and small sample sizes... at the same time!
- Check out the preprint for details! Our model isn't built to cater to zero-inflated Negative Binominally distributed data, but it still did awesome!
- Note that the robust score tests have better error rate control than the robust Wald tests (they are a bit slower)

Sadly we do not yet have a ~~logo~~ nice-looking logo. If you would like to design us one, please let Amy know!

Installation

To download the radEmu package, use the code below.

# install.packages("devtools")
devtools::install_github("statdivlab/radEmu")
library(radEmu)

We are currently only releasing radEmu via GitHub. If you'd like us to consider submitting to CRAN, please let us know by opening an issue.

Use

The vignettes demonstrate example usage of the main functions. Please file an issue if you have a request for a tutorial that is not currently included. The following code shows the easy-to-use syntax if your data is in a phyloseq object:

ch_fit <- emuFit(formula = ~ Group + Study + Gender + Sampling, 
                 Y = my_phyloseq_object)

and if your abundances and covariates are in a dataframe, you can use the following:

all_fit <- emuFit(formula = ~ Group + Study + Gender + Sampling,
                  data = my_covariates_df, 
                  Y = my_abundances_df)

Documentation

We additionally have a pkgdown website that contains pre-built versions of our function documentation and our vignettes (an introductory vignette, an introductory vignette that uses phyloseq data, a vignette for running radEmu tests in parallel for more efficient computation, and a vignette for running radEmu with clustered data).

Citation

If you use radEmu for your analysis, please cite our open-access preprint, available on arXiv.

David S Clausen and Amy D Willis. 2024+. "Estimating Fold Changes from Partially Observed Outcomes with Applications in Microbial Metagenomics." arxiv.org/abs/2402.05231

Huge thanks to the NIGMS for funding this work through Amy's R35!

Bug Reports / Change Requests

If you encounter a bug or would like make a change request, please file it as an issue here.

If you're a developer, we would love to review your pull requests.

Nomenclature

When we are not developing fast, robust and interpretable estimation methods, we enjoy making up silly names for our fast, robust and interpretable estimation methods. radEmu abbreviates radEmuAbPill, which denotes "using relative abundance data to estimate multiplicative differences in absolute abundances with partially identified log-linear models."

statdivlab / radEmu