Thread by @mikelove

Amy Willis @AmyDWillis
about to present on model misspecification in microbiome studies at #bioc2022

An April '22 preprint from David Clausen and Dr. Willis:

arxiv.org/abs/2204.12733

@AmyDWillis Great and simple motivation from looking at mock communities (over-detection) as to why to move away from:

E(W_ij) = c_j mu_ij (just size factor scaling)

instead use:

@AmyDWillis Oops that should have been c_i, the scaling factor for sample i.

i for sample here, j for taxa here (wide count tables)

Now introducing taxon specific efficiencies e_j to account for over-detection:

Why develop new models specifically for microbiome:

Just realizing that David Clausen = @davidandacat author of the relevant tweet:

@davidandacat Also this one which sounds like it is about to be relevant with respect to the Poisson modeling assumption:

@davidandacat Estimation involves "a sequence of unconstrained optimizations, permitting solutions progressively
closer to the boundary"

Then, to obtain limiting distributions, when the true parameter may be on the boundary of the parameter space, weighted bootstrap where w ~ Dirichlet

@davidandacat Dr. Willis on her thoughts on interesting directions for future work in modeling microbiome data

All of the methods shown here are available at this github link:

github.com/statdivlab/tin…

Collaborators and contributors: