Large number in lbf_variable matrix

[hbliu]

Thank you very much for this amazing tool SusieR. I exported the lbf_variable matrix. However, I found very large values (>10,000) in this matrix.

Would you please help fix this issue? Data and code are available at https://www.dropbox.com/s/204sg2ryo17u38t/ExampleData.zip?dl=0.

Thank you.

Related to #109

[pcarbo]

@hbliu Are the association p-values for these SNPs (e.g., calculated using PLINK) also very small? If so, this would suggest that these are just very strong associations.

In any case for fine-mapping we typically focus on the PIPs and the credible sets (CSs), not the Bayes factors.

[hbliu]

Thank you for the reply.

Panels 2-4 showed three credible sets ordered based on their significance of index SNP. The index SNP in each panel is as follows: variable variable_prob cs variant RSID Zscore PVAL N 1 1589 0.9998976 1 1:214744893:T:C rs7535253 -8.983 2.634e-19 2087620 (Index SNP in second Panel) 2 580 0.6838601 3 1:214575122:T:C rs17022807 -6.309 2.801e-10 2181101 (Index SNP in third Panel) 3 781 1.0000000 5 1:214618185:C:T rs4472734 6.189 6.050e-10 2089393 (Index SNP in fourth Panel)

Thus, I am confused that the SNPs in panel 4 get larger lbf than those in panels 2 and 3.

Thank you.

[stephens999]

Are you using in sample ld? This kind of thing could be caused by use of an out of sample ld matrix that does not match in sample well enough

[hbliu]

I am using an out of sample ld matrix, which does not match GWAS samples. I did not observe such large numbers when I tried fine mapping on the same locus using the matched GWAS and ld matrix.

Thank you.

[pcarbo]

Excellent. You might also find the "kriging_rss" function to be useful.

[hbliu]

Got it. Thank you very much! Merry Christmas and a happy holiday!