The abstract is read at the publisher; the summary is JClub's.
Domain-informed effector gene assignment from GWAS shows strong training-data dependence, with AUPRC dropping from 0.91 to 0.46 across domains for the same method.
- Why it matters: Most GWAS lead variants fall outside protein-coding regions, making effector gene nomination a key bottleneck. Without large-scale experimental ground truth, prioritization depends on quasi-true-positive sets that can embed curation assumptions.
- What they did: The study benchmarked established prioritization methods across three independent evaluation sets, trained five domain-informed models, and annotated lead variants for six common diseases. It found performance transfer failure linked to how models weight variant–gene distance (p<5.3x10-123).
- The result: Drug target recovery for 340 approved targets varied by model, and simple score averaging (38.8%) or joint training (36.5%) did not reach the 52.1% recovered when using all five together. The DEFA framework reports curation assumptions to enable examination and testing.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.