HOROSCOPE: Decoding human centromere architecture from short reads using k -mer signatures

Centromeres safeguard genome integrity, yet their roles in human disease remain understudied because of their repetitive sequence. We present HOROSCOPE, a k -mer-based framework that infers centromere architecture and generates centromere length estimates directly from short-read data. Using a reference atlas of 11,836 centromeres from telomere-to-telomere haplotypes, we derive diagnostic k -mer signatures that classify chromosome-specific architectures with 98.2% precision and 99.1% recall. Applied to 4,029 genomes from 80 populations, HOROSCOPE uncovers continental centromeric structure and African-enriched rare architectures. Across 1,359 cancer genomes, HOROSCOPE reveals a dependency of chromosomal rearrangement locations on the kinetochore attachment site position.