Local ancestry-informed rare variant burden testing improves gene discovery in admixed populations

Rare-variant association studies enable the discovery of high-impact genetic contributors often missed by conventional genome-wide association studies focused on common variation. However, standard burden tests aggregate variants without accounting for local ancestry in admixed genomes, reducing power when rare variant frequencies or genetic effects differ across ancestral backgrounds. Here, we introduce Tractor-Burden, an ancestry-aware gene-based association method that partitions rare-variant burden by inferred local ancestry and estimates ancestry-specific effects within a unified regression. In simulations, Tractor-Burden is well calibrated and improves power over standard burden tests under effect heterogeneity. Applied to whole-genome sequencing data from 47,152 admixed African-European individuals in the All of Us Research Program, Tractor-Burden recapitulates known associations, including ancestry-enriched effects at LDLR, and identifies additional suggestive genes and pathways for type 2 diabetes. Tractor-Burden extends rare-variant association testing to admixed genomes and provides a scalable framework for detecting and interpreting gene-level effects across local ancestry backgrounds.