The performance of genetic-constraint metrics varies significantly across the human noncoding genome

Gnocchi’s constraint performance in the human noncoding genome declines as GC content increases, with the most extreme distortion linked to model bias from mismatched de novo mutation training windows.

• Why it matters: Genome-wide precision-recall evaluations can mislead when neutral-model assumptions are biased or when genetic data have low signal-to-noise, undermining prioritization of noncoding loci for Mendelian traits.
• What they did: The study pinpoints bias in Gnocchi by comparing the de novo mutation (DNM) training set used to adjust expected SNP counts against the genomic windows where those expectations are computed, then recalibrates the mismatch.
• The result: Correcting the training–window mismatch yields a more calibrated Gnocchi, significantly improving interpretability, and motivates adding robust bias measures to constraint scores for user confidence.

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