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

By directing kinetochore formation and chromosome segregation, centromeres safeguard genome integrity. Yet, the roles of centromeres in human disease are understudied, as their repetitive architecture comprising -satellite higher-order repeats (HORs) renders them largely inaccessible to short-read sequencing approaches. Here we develop HORoSCOPE (Higher-Order Repeat organization and Size of Centromeres using Oligonucleotide Profiles for Estimation), a computational framework for k-mer-based inference of centromere structure and length from short-read data. Based on a reference atlas of 11,836 human centromeres extracted from completely assembled (telomere-to-telomere) haplotypes, we systematically interrogate chromosome-specific centromere architectures, deriving architecture-specific k-mer signatures as well as centromeric length-informative k-mers from common to rare centromere architectures. Leveraging these diagnostic k-mers, HORoSCOPE achieves a precision of 99.3% and a recall of 99.5% in classifying chromosome-specific centromere architectures from short-read de Bruijn graphs. We perform a population-scale analysis of global centromeric architectures in 4,029 human samples with ancestry from 80 human populations sequenced with short reads, uncovering continental haplotype structure for centromeric regions and highlighting African-enriched rare centromeric architectures. Furthermore, by analyzing 1,359 cancer genomes, we link graded HOR truncation events to arm-level copy-number alterations, and uncover a general dependency of chromosomal rearrangement locations on the position of the centromere dip region (CDR) defining the kinetochore attachment site. HORoSCOPE enables large-scale centromere genomics directly from short-read sample cohorts.