search

qlu-lab / DONUTS

DONUTS: Decomposing nature and nurture using GWAS summary statistics
6 stars 0 forks source link

DONUTS for binary traits #1

Closed tom-a-bond closed 1 year ago

tom-a-bond commented 2 years ago

Hi, could you advise whether DONUTS is valid for binary outcomes analysed using logistic regression please?

yuchang-wu commented 2 years ago

Hi,

Thanks for the question!

We did not extensively examine DONUTS' performance on binary traits, but we believe in general the beta (i.e., logOR) from the logistic regressions can be used as DONUTS inputs since

  1. people usually assume an underlying continuous liability for a binary trait (i.e., liability threshold model) and the effect sizes in the liability scale can certainly be used by DONUTS
  2. when the variance in liability "explained" by a SNP is not too large (empirically < 6%), there's a linear approximation between the effect size in logit scale (i.e., logOR) and the effect size in liability scale; and this linear relationship depends only on the disease prevalence. See Wu and Sham (2021) for the transformation in effect sizes between liability and logit scales.

Feel free to let me know if anything's not clear.

Thank you! Yuchang Wu

tom-a-bond commented 2 years ago

That is helpful, much appreciated!