search

omerwe / polyfun

PolyFun (POLYgenic FUNctionally-informed fine-mapping)
MIT License
96 stars 22 forks source link

Interpreting SNPs in credible set = 0 #73

Closed teresa-sansan closed 2 years ago

teresa-sansan commented 2 years ago

Hi Omer,

I used the LD score you provided to calculate the PIP in European-based summary statistics data. However, I recently discovered that there are some SNP with high PIP while being categorized as credible set = 0, which is a little bit confusing to me. (I specified the method to SuSiE and set the max_SNP_per_locus to 10. ) As shown in the attachment, there are still a lot of SNP that have PIP > 0.75 in credible set = 0.

github1

And this is part of the result I got: github2

Please correct me if I am wrong. My understanding is that credible set 0 will not contain any causal/high PIP SNP. That's why SuSiE is not able to put them into any other credible sets.

Thanks, Teresa

omerwe commented 2 years ago

Hi @teresa-sansan,

Usually high-PIP SNPs will belong to at least one credible set, but there's no guarantee. PolyFun simply uses the results of SuSiE as they are. Hence, if you'd like to look more into this I suggest that you contact the authors of SuSiE with a minimal use example. Sorry I can't be more help...

Best,

Omer

jdblischak commented 2 years ago

A similar question was asked in https://github.com/stephenslab/susieR/issues/144

I found the answer informative:

In order for variables to be in a rho credible set, their PIP have to sum to greater than or equal to rho, and their "purity", defined by minimum absolute pairwise correlation, should be greater than or equal to r. Currently, rho is set to 0.95 and r is set to 0.5. You can lower either these numbers, or both, to be more lenient with credible sets, depending on your analysis context (current default is what we believe reasonable for genetic fine-mapping applications)

teresa-sansan commented 2 years ago

Hi @omerwe, no worries. Yeah, I probably should reach out to SuSiE. Still, I appreciate your reply.

@jdblischak Thanks for pointing that out! My group and I checked that one before and still found it can't explain the problem we got. (They wonder why the other SNPs shouldn't be in a third credible set, where the max PIP is less than 0.1; while we have SNPs whose PIP is larger than 0.75 and still weren't assigned for a credible set.)

Thanks, Teresa