with the new genetics-portal focus on retaining credible set variants only, some discrepancies in variant coverage to the former open targets genetics portal would be expected. Has this been quantified in any way? One example we found is following variant being present in open target genetics but not the new open targets platform:
As @ochoa’s response mentions - we now have some differences in our data model. In the previous genetics portal, 9_89042048_T_C was included as an association for PheWAS plots despite it not being linked as a lead or tag variant for a credible set.
Our susie fine-mapping identifies four credible sets at that locus for the study 9_89042048_T_C is most strongly associated with, and they are much stronger signals.
I had a manual look at this region, and found that susie gives a non-convergence warning after 100 iterations (and takes a relatively long time). This can occur for multiple reasons, such as complex LD at the locus, more real effects than susie has been instructed to model, or effects with too small a size for susie to build a credible set for that coverage. We use a credible set coverage of 0.99, and L = 10 to model up to ten independent effects.
It is possible that 9_89042048_T_C is part of another less significant credible set, and susie just isn’t able to resolve it with those parameters. It’s also possible that the variant is genuinely not part of a credible set.
I have one follow-up question: Would there be a resource where genetic regions that did not converge in finemapping step are summarised?
We don’t have any immediate plans for this, but we could include labels/flags for credible sets which are derived from fine-mapped regions where susie could not reach convergence. This is something we will look into in the future.
