Near-perfect discrimination of half-sibling and avuncular pairs using genomic data without pedigrees

Large-scale genomic biobanks contain thousands of second-degree relatives without pedigree data. Accurately distinguishing half-sibling and avuncular pairs-both of which share approximately 25% of the genome-remains a significant challenge. Current SNP-based methods rely on aggregate Identical-By-Descent (IBD) segment counts and age differences, but substantial overlap in these distributions leads to high misclassification rates. There is a need for a scalable, genotype-only method that can resolve these second-degree ambiguities without requiring observed pedigrees or phenotypic information.

We present a novel computational approach that achieves near complete separation of half-siblings and avuncular pairs using haplotype-level sharing features based on data that has been computationally phased across-chromosomes. These features also make it possible to determine which member of an avuncular pair is the niece/nephew and which is the aunt/uncle. By modeling these features with a multivariate Gaussian mixture model, we achieve exceptional classification performance in biobank-scale data. Ground-truth labels for validation were established through a multi-step inference process within a family graph constructed from high-confidence first, second and third-degree relationships. All 444 ground-truth avuncular pairs and 97 of the 99 half-sibling pairs were correctly classified. Treating avuncular status as the positive class, this produces a specificity of 97.98% and sensitivity of 1. Our results suggest that the framework could perform comparably on other datasets of similar size. This method provides a robust, scalable means of distinguishing half-sibling and avuncular pairs, improving relationship annotation for pedigree inference, genomic analyses that utilize relatives, and forensic genetic applications.

emmanuel.sapin@colorado.edu , kristen.kelly@colorado.edu , and matthew.c.keller@colorado.edu.