run susie_rss for MHC region

[PhoebeGuo97]

Hello, I had some issues on running susie_rss for MHC region (CHR6) recently. At beginning, susie didn't converge so I increased the region to 1Mb and removed some SNPs with high logLR value to finally get susie converged. However, I checked the credible sets and found some SNPs with high p-values were there, which doesn't make sense to me. I'm wondering what may cause this happen. Many thanks!

The script and pip of SNPs in the CS are as follows: rss_Jansen_6 <- susie_rss(z=Jansen_6RL_filter$Z, R=as.matrix(ld), n=455258, max_iter = 1000)

rss_Jansen_6$sets $cs $cs$L1 [1] 9064

$cs$L2 [1] 9002

$cs$L3 [1] 6728

$cs$L4 [1] 5760

$cs$L5 [1] 6073

$cs$L8 [1] 14730

$cs$L9 [1] 11918

$cs$L10 [1] 8107

$cs$L6 [1] 12620 12694

$cs$L7 [1] 13868 13957 13959 14269 14491 14778 14826 14828

$purity min.abs.corr mean.abs.corr median.abs.corr L1 1.0000000 1.0000000 1.0000000 L2 1.0000000 1.0000000 1.0000000 L3 1.0000000 1.0000000 1.0000000 L4 1.0000000 1.0000000 1.0000000 L5 1.0000000 1.0000000 1.0000000 L8 1.0000000 1.0000000 1.0000000 L9 1.0000000 1.0000000 1.0000000 L10 1.0000000 1.0000000 1.0000000 L6 0.9997559 0.9997559 0.9997559 L7 0.9997559 0.9998954 0.9999084

$cs_index [1] 1 2 3 4 5 8 9 10 6 7

$coverage [1] 0.9996261 1.0000000 1.0000000 1.0000000 0.9999988 1.0000000 1.0000000 0.9998296 0.9929598 0.9571046

$requested_coverage [1] 0.95

[pcarbo]

@PhoebeGuo97 Did you run with refine = TRUE?

[PhoebeGuo97]

@PhoebeGuo97 Did you run with refine = TRUE?

@pcarbo No, I didn't.

[stephens999]

when you control for SNPs that have an affect, previously non-significant SNPs can become "significant". (remember, non-significant does not imply H0 is true). However, I would be cautious about any results using out-of-sample LD matrix, especially in such a complex area. Finemapping complex areas reliably and confidently probably requires getting the in-sample LD matrix.

[PhoebeGuo97]

I added refine=TRUE then susie_rss didn't converge this time.