Preprint exposes training-data bias in methods used to link GWAS signals to target genes
A benchmarking preprint finds that widely used GWAS effector-gene prioritisation methods are critically dependent on the evidence type used to train them, with implications for how post-GWAS analyses inform drug target selection.
A preprint posted to bioRxiv on 18 November 2026 reports a systematic benchmarking study of methods used to assign effector genes at genome-wide association study (GWAS) loci — the step that translates a statistical association signal into a candidate biological mechanism.
Approximately 90 per cent of trait-associated lead variants from GWAS lie outside protein-coding regions, making effector gene nomination a central bottleneck in translational genetics. The authors benchmarked established prioritisation methods across three independent evaluation sets and trained five domain-informed models, each anchored to a different biological line of evidence. They applied these across six common traits and found that method performance was strongly dependent on which quasi-true-positive set was used for training — a circularity risk that the authors argue is widely underappreciated in the field.
The study does not propose a single preferred method but instead argues for greater transparency about training-data assumptions in published prioritisation pipelines. The findings are particularly relevant to researchers designing post-GWAS functional follow-up studies and to those evaluating the genetic support for drug targets — a step that pharmaceutical and translational research groups routinely perform before committing resources to target validation.
As a preprint, the work has not yet been peer-reviewed. The methodological critique it raises has implications for how the field interprets the strength of genetic evidence for any given gene target.
Sources
Read the original reporting — these are the public sources this summary draws from.
-
Primary sourcePreprint bioRxiv · 2026-11-18Domain-informed effector gene assignment exposes training-data dependence of drug target prioritisation from GWAS