Near perfect identification of half sibling versus niece/nephew avuncular pairs without pedigree information or genotyped relatives

MotivationLarge-scale genomic biobanks contain thousands of second-degree relatives with missing pedigree metadata. Accurately distinguishing half-sibling (H-S) from niece/nephew-avuncular (N/N-A) pairs--both sharing 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 distributional overlap leads to high misclassification rates. There is a critical need for a scalable, genotype-only method that can resolve these "half-degree" ambiguities without requiring observed pedigrees or extensive relative information. ResultsWe present a novel computational framework that achieves near-complete separation of H-S and N/N-A pairs using haplotype-level sharing features derived from across-chromosome phasing. By modeling these features with a multivariate Gaussian mixture model (GMM), we demonstrate exceptional classification performance in biobank-scale data. Ground-truth validation labels were established through a multi-step inference process within a family graph constructed from high-confidence first-degree relationships. To overcome the scarcity of genotyped half-sibling parents, we structurally validated H-S pairs by identifying specific pedigree configurations involving first cousins that are logically incompatible with an avuncular relationship. Our method achieves a sensitivity of 96.9% and a specificity of 99.7% on these structurally-validated pairs. Furthermore, we show that these identified relationships serve as superior phase anchors, that measurably improve the accuracy of across-chromosome homologue assignment. This method provides a robust, scalable solution for pedigree reconstruction and the control of cryptic relatedness in large-scale genomic studies. Contactemmanuel.sapin@colorado.edu, kristen.kelly@colorado.edu, and matthew.c.keller@colorado.edu