Best method for extracting 95% CS if agnostic to no of causal variants

[Marijn-Schipper]

Hi,

I have a very applied answer that I so far could not find an answer to. I want to finemap multiple GWASes from the UKB. Specifically I'm looking to extract a 95% CS for each causal variant in pre-specified loci. Beforehand I do not know how many causal variants are in each loci. Given that SUSIE --max-num-causal uses the exact amount of num causal specified would it be better to use FINEMAP in this instance? Or is there a way to use SUSIE to find the most likely number of causal snps in a locus?

Thanks for your help and the amazing documentation! Marijn Schipper

[omerwe]

Hi @Marijn-Schipper,

The approach we took in our paper is that even if we overestimate L (the estimated number of causal SNPs) we still get robust results. This is based on the SuSiE paper, which shows that if we overestimate L, the "excess probability" gets distributed roughly evenly across all SNPs and thus the effect is negligible. You can find more details in the SuSiE paper. So my short paper is that if you assume a relatively large number for L (e.g. L=10) you should get relatively robust results.

I think that FINEMAP is susceptible to the estimated max. number of causal SNP, but I'm not sure about the exact details there...

Hope this helps, please let me know if not!