search

nhanhocu / metamicrobiomeR

An R package for analysis of microbiome relative abundance data using zero inflated beta GAMLSS and meta-analysis across microbiome studies using random effects models
https://github.com/nhanhocu/metamicrobiomeR
33 stars 11 forks source link

GAMLSS family BEZI after zero replacement and CLR transformation #10

Closed Nick243 closed 4 years ago

Nick243 commented 4 years ago

Hi Nhan,

Thanks for developing the approaches and user-friendly functions in the metamicrobiomeR package. This is quite impressive!

I have a question on the appropriate probability distribution to use in GAMLSS for relative abundance estimation and testing after zero replacement and CLR transformation. After the replacement and transformation, would I expect to have few/no zeros and negative values? If so, would you still recommend using the BEZI family? Or would you then perhaps recommend also using propmed.rel = "lm" in the taxa.compare function (or fitting the model in GAMLSS assuming a different distribution for the response)?

Thanks in advance for any thoughts on the matter!

Nick

nhanhocu commented 4 years ago

Dear Nick, Thanks for your interest in the package. You are right that if you do CLR transformation, you should use "lm". Bests, Nhan