How to evaluate the sample size for a metagenomic experiment

[AndrwOnar]

Hi everyone,
I'm not sure this is exactly the right place for this question but i'm bit in trouble with a metagenomic article and this is the only forum about the argoument that i know.
I've analysed my data using Qiime but a reviewer asked me to estimate sample size for my experiment to know if the number of sample that i have choosen is correct.
I did not found any function of Qiime to do that but i discovered an R package named "micropower" that can perform that analysis..... I've tried to understand how does it works by myself but i found that it bit difficult to understand, does anyone know it or know another method by Qiime to estimate the sample size for a metagenomic experiment?
thank you all for your attention

[Tomasz]

Hi!

I forwarded your question to my colleague. I think he should be able to advise you.

best,

Tomasz

[Justine Debelius]

There are currently two published method for power in microbiome data. Which you want to use depends on your analysis, and what you want to know.

Micropower (Kelly, et al, 2015) would be my current first choice for doing this analysis. It's recent, and more likely to be directly relevant to what you're doing. Micropower estimates sample size based on permanova for beta diversity. I've not run micropower, but they have a GitHub page, and that might be a way to reach out. I assume, based on reading the paper, that you calculate an effect size statistic and that's either compared against a bootstrap of your data, or an existing power curve.

The other option is to use the dirichlet multinomial model from La Rosa et al, 2012. This works based on the assumption you've done taxonomic analysis using their method. Again, I've not used this package, so I don't know how easy or hard it is to use.

Not sure if that helps make anything clearer.

Happy Holidays,

Justine

[AndrwOnar]

thank you for your help, yes I've founde both articles but i wonder if someone have use them to guide me in the analysis cause it is not so much clear to me

[jonsan]

It doesn't sound like anyone around here has run them either... The GitHub page for MicroPower seems to indicate that there was a tutorial/vignette that was going to be posted, but perhaps never was. 

There doesn't currently exist an alternative out-of-the-box power calculator within Qiime, so my thinking is that contacting the authors of MicroPower directly might be your best bet at the moment.

-j

[Colin Brislawn]

out-of-the-box power calculator within Qiime

That would be a excellent feature for Qiime 2. I've worked with a statistician to estimate power for DESeq2 differential abundance tests after-the-fact, so I can probably contribute code. This is salient problem and I'm glad that Andrw brought this to the forums.

Colin

[francesco cerutti]

Hi everybody,
I'm running in the same issue as Andrw, trying to use the "micropower" package to estimate the power of my study.
How did you solve your problem?

[AndrwOnar]

Hi Francesco,

to be true I've never solved that problem... I did not find any help with "micropower"... if you find something please share it if you can.