Assess whether there are better ways to generate credible sets from susie outputs.
Finngen credible sets have problems. Their pipeline orders credible sets by purity instead of log(BF). This is a legacy decision on their part which has knock on consequences, especially for the inclusion of variants into 95% or 99% credible sets:
Some variants appear in different credible sets. For example the top variant in 95% credible set 1 will also be assigned to 99% credible set 2 instead of one.
Daniel-Considine
commented
1 hour ago
Assess whether there are better ways to generate credible sets from susie outputs.
Finngen credible sets have problems. Their pipeline orders credible sets by purity instead of log(BF). This is a legacy decision on their part which has knock on consequences, especially for the inclusion of variants into 95% or 99% credible sets:
| trait| rsid| cs|cs_99| alpha1| alpha2| |ANTIDEPRESSANTS|chr10_118081733_G...| 1| 2| 0.812342679128655|4.04201788934286e-05|Not yet investigated eQTL-catalogue sets for issues.