Comment on “Residual Allelic Activity Likely Underlies the Low Rates of Disease Expression for Predicted Loss-of-Function Variants in Population-Scale Biobanks”

Exome and genome sequencing are increasingly used for rare disease diagnosis even among asymptomatic individuals, though their utility for predicting future disease risk remains uncertain. This is largely due to an incomplete understanding of penetrance, or the probability that a pathogenic genotype results in symptoms, which varies widely across many variables and is difficult to estimate. Population-scale biobanks that link genomic and electronic health record data have enabled “genotype-first” studies, which suggest that many presumed pathogenic variants show low rates of actual disease expression. However, these estimates may be influenced by incomplete phenotyping and differences in diagnostic criteria. In this study, the authors evaluated disease expression among individuals with predicted loss-of-function (pLOF) variants in genes associated with haploinsufficient disorders using large biobank data sets, and investigated factors contributing to low penetrance. This study used exome and genome sequencing data linked to electronic health records from the UK Biobank and All of Us cohorts to assess disease expression among individuals with pLOF variants in genes associated with haploinsufficient disorders. The authors identified 91 autosomal or pseudoautosomal-dominant diseases linked to 117 genes and identified heterozygous carriers of rare variants after quality control and annotation. Disease expression was evaluated using a complex aggregate of diagnostic codes, phenotype risk scores, and a symptom-based probabilistic model based on structured health record data, with noncarrier cohorts for comparison. Additional analyses assessed the impact of incomplete clinical data as a possible explanation for seemingly low penetrance and evaluated for associations between specific types of variants and disease expression using machine-learning models. The study identified over 6000 unique pLOF variants between the two biobanks. Analyses found that 3% to 4.5% of individuals carried at least one variant, which strongly challenges the presumption that each of these variants results in phenotypically expressed disease. While pLOF carriers had a greater risk of symptoms and disease than the reference cohorts, overall disease expression remained low at <10%. After accounting for annotation artifacts, missed diagnoses, and incomplete electronic health record coverage, estimates for disease expression modestly increased, but the discrepancy was not resolved. Machine-learning models that incorporated the genomic features present in each type of pLOF successfully stratified variants by expression risk and outperformed standard pathogenicity filters, suggesting that some variants retain partial function and have intrinsically reduced penetrance. Predicted loss-of-function variants in haploinsufficient disease genes showed consistently low disease expression rates, even after accounting for likely confounders such as annotation errors and incomplete medical records. Standard pathogenicity annotations, including ClinVar, were poorly predictive of disease expression, whereas models incorporating variant-specific genomic features more effectively identified higher-risk variants. These findings suggest that many pLOF variants retain partial function or are rescued through other mechanisms, which results in reduced penetrance and limited predictive value of genomic screening. Although data limitations may lead to underestimation, the results highlight fundamental challenges in prognostic genetic testing and emphasize the importance of cautious interpretation of genomic data in asymptomatic individuals. (Summarized from Blair DR, Risch N. Residual allelic activity likely underlies the low rates of disease expression for predicted loss-of-function variants in population-scale biobanks. Am J Hum Genet . 2025;112:2922–2942. doi:10.1016/j.ajhg.2025.11.002)