SNP weights in LD score regression

241 views
Skip to first unread message

nora.m...@gmail.com

unread,
Mar 22, 2017, 11:05:48 PM3/22/17
to ldsc_users

Hi,


I identified a set of variants known to have come from a particular part of Africa. I used that data to build a polygenic score, and gave all the relevant SNPs the same weight. I was wondering if I could run a bivariate LD score regression and use this as one of the traits, or does the algorithm require there to be variation in the weight value?


Thank you, 

Nora

Raymond Walters

unread,
Mar 23, 2017, 12:49:12 PM3/23/17
to nora.m...@gmail.com, ldsc_users
Hi Nora,
The bivariate LD score regression model is fairly directly linked to assuming standard GWAS results (i.e. from linear or logistic regression) as input, so I imagine it’s not a great fit for your analysis.

On the other hand, your identified set of variants could be a good candidate for use as a custom annotation for partitioned LD score regression of the other trait (depending on the details of your variant set and what you’re hoping to test with the analysis). See Finucane et al 2015 for details of the partitioned heritability method, and then the tutorials on github for running partitioned heritability analysis and creating your own annotations for the analysis.

Cheers,
Raymond


-- 
You received this message because you are subscribed to the Google Groups "ldsc_users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ldsc_users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ldsc_users/fe35a8e6-b323-4378-bfd3-7f1690921267%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

nora.m...@gmail.com

unread,
Apr 29, 2017, 11:04:29 AM4/29/17
to ldsc_users, nora.m...@gmail.com

Hi Raymond,


Thank you so much for your previous response. One more question - is there a minimum threshold for the number of SNPs used in an annotation?


Best,

Nora

Raymond Walters

unread,
May 1, 2017, 3:11:54 PM5/1/17
to nora.m...@gmail.com, ldsc_users
Hi Nora,
There’s not a strict threshold, but the current rule of thumb is roughly that the annotation should be at least 1% of the genome. The idea is that you want an annotation that’s large enough to be safe treating the effects of individual SNPs as random effects (with some average heritability per SNP) and spread out enough across the genome for the block jackknife SEs to work as intended.
Cheers,
Raymond


Reply all
Reply to author
Forward
0 new messages