search

stephenslab / susieR

R package for "sum of single effects" regression.
https://stephenslab.github.io/susieR
Other
176 stars 46 forks source link

couldn't get susie converged although LD matrix seems fine #182

Closed PhoebeGuo97 closed 1 year ago

PhoebeGuo97 commented 1 year ago

Hello, I tried fine-mapping some loci of my interest using susie_rss. The LD matrix was calculated using UK biobank as the reference panel. It got converged for some loci easily using default settings, but it never converged for a locus on Chr17q21 . I have tried increasing max_iter to 5000, setting check_prior=FALSE and removing SNPs with logLR>0 but still couldn't fix this issue.

I have attached the condz$plot here and the lambda is ~0.01. I notice that some SNPs look like outliers in the plot but they are not red dots. Thanks for your help in advance!

Bell_17RL_susie

pcarbo commented 1 year ago

Yes, you could try removing these 5 additional SNPs and see if it helps.

pcarbo commented 1 year ago

The plot shows conditional_mean$z vs. conditional_dist$condmean, so you would want to look for rows in the conditional_dist data frame that diff a lot in those two columns.

PhoebeGuo97 commented 1 year ago

I removed 6 SNPs with the largest absolute value of conditional_dist$z_std_diff and susie_rss converged. 5/6 SNPs' logLR are smaller than 0, so I guess maybe that's the reason that they are not shown as red dots in the plot.