search

stephenslab / susieR

R package for "sum of single effects" regression.
https://stephenslab.github.io/susieR
Other
174 stars 44 forks source link

Mixture models/GLM #115

Closed mariellabogoni closed 2 years ago

mariellabogoni commented 3 years ago

is possible to extend to mixture models? or generalized linear models such as logistic model? thank you!

gaow commented 3 years ago

@mariellabogoni what do you mean by mixture model -- would you elaborate?

katherinef commented 3 years ago

I would also be interested in this functionality, particularly logistic regression for binary outcomes. I assume "mixture models" means linear mixed models like those implemented in BOLT-LMM that can account for genetic relatedness between individuals.

Thanks!

Kath

stephens999 commented 3 years ago

the most general approach we have is susie_rss (susie regression with summary statistics) which applies susie to z scores and an LD matrix. So, for example, you can compute z scores under a logistic regression, or mixed model, and apply susie_rss to those z scores. We have attempted to make susie_rss somewhat robust to misspecification of the LD matrix and continue to work on further improving this. You should be able to get more help with ?susie_rss

For the specific case of logistic regression we have also implemented a version of susie specific to the logistic model. It is not really clear whether this is better than using susie_rss on z scores, or even just applying susie with a binary outcome (if effect sizes are small, as they are in most genetic studies, logistic regression and linear regression usually produce very similar results). But we have implemented it. @andrewg3311 is responsible for that work, and I'll let him point you to anything he can share as it is very much "work in progress".