search

stephenslab / mashr

An R package for multivariate adaptive shrinkage.
https://stephenslab.github.io/mashr
Other
88 stars 19 forks source link

Bayes Factor calculation at the unit level #96

Open pekjoonwu opened 3 years ago

pekjoonwu commented 3 years ago

Hi,

Thanks for developing this great tool for analyzing the data. I have a question more about comparing the results from mashr to the other methods such as Metasoft.

I am wondering when you compared the power in Figure 2 a, b panel with Metasoft, were you using the Bayes factor to test the global null so that both methods have significance measurement for unit level? If so, is there any function in the package that can output this factor for each unit and what threshold should be used for each unit?

Thank you so much!

Best

stephens999 commented 3 years ago

yes that is what we did.

I notice that our documentation of the mash output is quite incomplete! The log BF can be directly computed from the mash output (res.mash) by res.mash$alt_loglik - res.mash$null_loglik I believe this is log base e.

The threshold is up to you... not that easy to give a general answer to that.

pekjoonwu commented 3 years ago

Thank you Matthew for the clarification!

Since I am not a Bayesian, I still have one question about the procedure. Specifically, did you guys treat the Bayes factor as an alternative option for p-value (ranking the Bayes factor for all the units from the biggest to the smallest) and then calculate the power and FDR like what we usually do with the p-value?

Best

stephens999 commented 3 years ago

yes, to plot a curve like this all you need is a ranking. p values and BFs both provide a ranking, and the curves are just quantifying how effective that ranking is from distinguishing the effects and the non-effects. You could do this for any method that provides a ranking, it doesn't have to be a p value or a BF....

On Mon, Jun 21, 2021 at 9:34 PM pekjoonwu @.***> wrote:

Thank you Matthew for the clarification!

Since I am not a Bayesian, I still have one question about the procedure. Specifically, did you guys treat the Bayes factor as an alternative option for p-value (ranking the Bayes factor for all the units from the biggest to the smallest) and then calculate the power and FDR like what we usually do with the p-value?

Best

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/stephenslab/mashr/issues/96#issuecomment-865477007, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANXRRPVD7OWMQJKUZYV2GTTT7ZDFANCNFSM47CQC36A .