Accurate and scalable genome-wide ancestry estimation using haplotype clusters

Unsupervised genome-wide ancestry estimation in unrelated individuals has been a staple in population and medical genetics for decades, and its importance continues to grow with the increasing number of large genetic cohorts of mixed ancestries. We propose an extension to the hapla framework that scales model-based ancestry estimation to unprecedented sample sizes by leveraging inferred haplotype clusters from phased genotype data. Our haplotype cluster-based approach is approximately 5× and 20× faster than the fastest model-free and model-based SNP-based approaches, respectively, for unsupervised genome-wide ancestry estimation on the harmonized Human Genome Diversity Project and 1000 Genomes Project dataset. Furthermore, we demonstrate that it is the most accurate method to date in an extensive simulation study, across a range of sample sizes from thousands to hundreds of thousands of individuals. Our accurate ancestry estimates can help reduce health disparities and accelerate precision medicine efforts in the growing number of biobanks globally.