search

statdivlab / corncob

Count Regression for Correlated Observations with the Beta-binomial
103 stars 22 forks source link

Appropriate use of corncob? #127

Closed msmcfarlin closed 3 years ago

msmcfarlin commented 3 years ago

Hello!

First, I'd like to echo #118, you put together a fantastic, easy to follow, vignette that makes understanding the package quite simple.

I am very interested in using this package in my research but noticed in your manuscript the statement on the limitations of this method, "A limitation of this approach is that it does not enforce the compositionality constraint..."

I am rather new to differential abundance analysis, could you give examples of the types of questions/hypotheses that would be appropriate to test with corncob analysis and what would inappropriate with corncob?

Thank you so much for your time.

bryandmartin commented 3 years ago

Hi @msmcfarlin ,

Sure, happy to! Essentially, any setting in which you are testing for differential abundance or differential variability (as defined in the paper/vignette), corncob will be an appropriate method. It is designed specifically to test for differential abundance in a statistically valid way.

However, what I wouldn't use corncob for is estimating the composition of an entire microbiome. For example, you wouldn't want to take all of your relative abundance estimates by taxon from corncob, then put them together, then assume that's the composition. One of the main reasons for this is that corncob does not impose that the sum of all the estimated abundances sum to 1.

Does this clarify things? Closing for now but happy to re-open and further clarify.

Bryan

msmcfarlin commented 3 years ago

Hi @bryandmartin,

Yes, this clears it up! Thank you!