Excluded causal variants produce false discoveries in genetic fine-mapping

Filtering out variants with MAF below 1% and omitting indels in GWAS/QTL fine-mapping at 499 All of Us loci produced false high-confidence calls, with 20% (and 32% with indels) lacking support unfiltered.

• Why it matters: Fine-mapping is meant to separate causal variants from LD-correlated proxies, but common preprocessing can remove true causal signals before inference, undermining algorithm assumptions and inflating false discoveries.
• What they did: The study compared filtered versus unfiltered fine-mapping across 499 All of Us loci, then tested how errors arose when candidate variants tagged omitted variants; it also introduced SuSiE-OUT and evaluated strategies including infinitesimal-effect modeling versus imputing omitted variants or using SuSiE-OUT.
• The result: Errors were enriched for proxy tagging of omitted variants, and at ABO omitting the frameshift defining blood group O yielded near-certain PIPs for linked SNPs; imputing omitted variants or using SuSiE-OUT prevented most errors.

AI summary of the abstract · may contain errors · not medical advice · read the paper.