Robust Chromosome Segmentation in Sparse Single-Cell Whole-Genome Sequencing Using a k-Segment Constrained Hidden Markov Model

Abstract

Genome integrity is essential for cellular and organismal fitness. Loss of heterozygosity (LOH), which eliminates the buffering effect of heterozygosity and can expose deleterious recessive alleles, is an important form of somatic genome alteration associated with aging and tumorigenesis. To characterize LOH across cell populations, we use sci-L3, a high-throughput combinatorial-indexing approach for single-cell whole-genome sequencing (WGS). The throughput, however, comes at the cost of sparse coverage of individual genomes, complicating LOH inference from phased heterozygous sites. We therefore developed a k-segment constrained hidden Markov model (HMM) that explicitly limits the maximum number of state transitions and implicitly enforces learning a much lower transition probability than unconstrained HMM models, enabling robust segmentation with learned transition probabilities and optional SNP binning. Benchmarking against orthogonal sci-L3-Strand-seq segmentation, which resolves all seven mitotic crossover outcomes, the method achieved 86% LOH recall and 88% precision. Haplotype-aware analyses support genuine LOH in 68-78% of sampled apparent false-positive calls, suggesting that the reference-based precision of 88% may underestimate the actual precision, potentially approaching 96%. The k-segment HMM thus enables sensitive detection of LOH heterogeneity from sparse single-cell WGS and is applicable to systems with available haplotype phasing.