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!
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 protocolradEmu
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 pseudocountsradEmu
does not require that you have a "reference taxon" that is not changing in abundance across samples
radEmu
estimates differences in abundance across taxa 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 confoundersradEmu
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
Sadly we do not yet have a logo nice-looking logo. If you would like to design us one, please let Amy know!
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.
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)
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).
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!
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.
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."