search

stephenslab / susieR

R package for "sum of single effects" regression.
https://stephenslab.github.io/susieR
Other
172 stars 43 forks source link

No credible sets #163

Open katherinef opened 2 years ago

katherinef commented 2 years ago

Hello,

I am using susie_rss to fine-map a region that was suggestively significant (P < 5 x 10-6) in our discovery GWAS, but I am getting no credible sets returned. I assume this is because, according to Susie, there are no signals in this region. I have tried adjusting the coverage parameter to 0.9, but still no sets. Is there a minimum acceptable threshold for coverage, and is there anything else I should try?

Best wishes,

Kath

gaow commented 2 years ago

Hi @katherinef how does your PIP look like -- would you mind posting here output from susie_plot?

katherinef commented 2 years ago

Hi, thanks for the quick response. Here is the plot. susie

gaow commented 2 years ago

Thanks @katherinef . Assuming the top 3 variants are represneting the same single effect, it looks like a coverage of 55% (for sum of PIP 0.24+0.16+0.18) can capture them -- if you try that CS coverage, although you probably don't want to report such a low coverage in practice. The signal is weak and that's why we dont have a 95% CS .

katherinef commented 2 years ago

That worked, thank you! When you say we shouldn't report such a low coverage in practice, I assume you mean there is insufficient evidence for fine-mapping (ie. for the signal being due to one of these three variants)? Under these circumstances, would you report the top PIPs for the region (with caveats) but disregard the credible set?

gaow commented 2 years ago

@katherinef i would still report the entire CS compared to top PIP. It's just that this CS, by definition, has only 55% chance of capturing a non-zero effect. I should say that whether or not to take this "discovery" seriously depends on the context of your research project. Earlier, I was just assuming that in some GWAS scan where there are dozens other 95% CS and many others in between 55% and 95%, then by definition your 55% CS may be of low priority to study.

katherinef commented 2 years ago

Thanks, that makes sense.