Eight-mask strategies increase association discovery in rare-variant gene burden tests
Background: Variant grouping criteria can change the results
In gene-level burden tests of rare variants, changing only the criteria for including variants can substantially alter which associations are discovered. A study published in Nature Genetics analyzed UK Biobank data and proposed strategies comprising eight masks each for low-frequency and rare variants. The two strategies identified about twice as many significant associations as the average of existing strategies.
A burden test groups multiple variants within a gene to assess their association with a trait. Because it assumes that the included variants have the same effect on gene function, variant selection matters. Masking selects coding variants to aggregate according to functional annotations and allele frequency; it is broader than simply removing common variants.
The research team reviewed 234 papers published from 2012 to 2024 and collected 664 masks. Even after harmonizing their definitions within a common analytical framework, 92.2% of masking strategies had been used in only one paper. Many studies also did not adequately explain the rationale for their selections.
Key findings: Considering discovery yield and multiple-testing burden together
The main comparison used exome data from 189,947 UK Biobank participants and 54 quantitative traits. After quality control and other steps, the researchers evaluated 271 masks and 146 distinct strategies. The number of significant associations after strategy-specific multiple-testing correction ranged from 58 to 2,523. This full range also includes results from masks that included common variants, so it should not be interpreted as the performance of rare variants alone.
The team used a minor allele frequency (MAF) below 1% as the threshold for low-frequency variant analyses and below 0.1% for rare-variant analyses. They combined predicted loss-of-function variants, potentially damaging missense variants and other variants according to frequency and functional criteria. They then built strategies with a greedy algorithm that sequentially selected the mask adding the largest number of significant associations not yet captured. Adding masks broadens the search, but also makes multiple-testing correction more stringent, so more masks are not always advantageous.
The final proposals comprise eight masks each for low-frequency and rare variants. They captured about 95% of the associations identified by larger strategies with 22 and 35 masks, while finding about twice as many significant associations as the average of existing strategies. Rather than providing an optimum for every study, the proposals offer a starting point and a benchmark for follow-up analyses.
Implications and outlook: Transparency in analytical criteria and further validation
In an expanded UK Biobank analysis of 439,829 participants, the researchers identified 114 significant associations not reported by three previous studies, across 46 traits shared with those studies. They also evaluated the strategies in All of Us and AMP T2D GENES. Discovery performance was generally high, but the strategies did not outperform alternatives in every comparison.
It is too early to conclude that all newly identified associations reflect true biological effects. The researchers explicitly stated that replication in independent data would increase confidence. Reliance on predictions of functional impact remains a limitation. Although the evaluation included data from multiple ancestry groups, it lacked sufficient statistical power to adequately compare performance across ancestry groups.
Frequency filters alone cannot eliminate linkage disequilibrium with common variants—the correlation between variants that are inherited together. Conditional analyses are needed to determine whether gene-level signals from rare variants are independent of signals from common variants. These findings provide a basis for improving mask selection and reporting standards; they do not validate individual disease-risk prediction or clinical utility.
The study’s methods and results are available in the publicly available full text of the Nature Genetics paper.
Nature Genetics, Published online: 08 May 2026; doi:10.1038/s41588-026-02597-9 Systematic comparisons of different masking approaches for rare variant association tests across 54 traits in UK Biobank highlight strategies for gene-level burden analyses that increase study power and replicability.
Teams using exome data to identify genes associated with diseases or quantitative traits can use the eight-mask strategy as a starting benchmark for burden testing. Those developing their own masks can compare them with the benchmark using the same data and multiple-testing correction conditions to assess the benefit of additional discoveries. Explicitly reporting variant-selection criteria, which have varied across studies, can also make it easier to determine whether differences in results stem from the data or the analytical design. The identified gene–trait associations may help select candidates for follow-up functional studies, but independent replication and conditional analyses that account for common-variant signals are still needed. The strategy itself does not guarantee therapeutic target validity or predictive accuracy for individual patients.