Open EchoLee0925 opened 2 years ago
for Q1: the ordering of the CSs is not important. I think you are best to ignore the ordering.
for Q2: CSs can be removed because of low purity (or possibly due to estimation of the effect size parameter being zero). In general the number of CSs you get is guaranteed to be <= L, not necessary =L. (And typically it is often much less than L).
I believe you can adjust min_abs_corr
to output greater or fewer CSs.
I got it,but when I change the L ,the result of PIP is changed,too. Is there any method to get the proper number L, or some judgement criterions to stop iterate?
For uni-variate fine-mapping with SuSiE in practice, we start with a large enough (but not crazy large) L based on the context. For example for cis-eQTL a good starting point is L = 10; for a GWAS region identified by using a lead SNP and variants around it, L = 5 may be good enough. We then run the analysis in each region and find regions with detected CS close or equals L, and reanalyze those regions using a larger L.
Thank you! I'll try with your suggestion.
Besides, whether the positive and negative sign of beta for A1/A2 and z-score effect the PIP?
I'm Sorry, I'm confused by the susieR and finemap software. The finemap software need the beta for A1/A2. In your package, the positive and negative sign of z-score whether to effect the PIP?
yes they effect PIP.
Hi, I also have a question about PIP and CS. In the results, I noticed situations when some SNPs have a very high PIP like 0.9, but they are not in any CS (the value of cs column is -1). I'm confused by this. Should I give more weight to CS or PIP results?
@RL-m High without any other SNPs in high LD with it, the high PIP SNP itself with PIP = 0.9 cannot make it to the default 95% CS (intuitively, the sum of PIPs from a "causal" variant and its LD friends should be >= 0.95 to be in a 95% CS).
Should I give more weight to CS or PIP results?
Depends on what you want to do. If mapping the variant is your objective then you should look out for eg 95% CS which is interpreted as there is 95% probability that the "causal" variant is captured in the CS. A PIP of 90% means the probability of that SNP being "causal" is 90%.
By run the susie_rss( ) function of a locus GWAS, I have got the result of PIP and cs for each SNP.