High-resolution detection of copy number alterations in single cells with HiScanner

Yifan Zhao, Lovelace J. Luquette, Alexander Veit, Xiaochen Wang, Ruibin Xi, Vinay Viswanadham, Yuwei Zhang, Diane D. Shao, Christopher A. Walsh, Hong Yang, Mark D. Johnson, Peter J. Park

TL;DR

HiScanner is a new computational tool that achieves unprecedented resolution in detecting somatic copy number alterations (CNAs) at the single-cell level. By integrating read depth, B-allele frequency, and haplotype phasing, it can reliably identify small genomic changes (down to 200 kb) that previous methods missed, revealing new insights into how mutations accumulate in both healthy brain cells and evolving tumors.

What Problem Does This Paper Address?

Every cell in your body is supposed to have the same DNA, but "somatic mutations"—changes that occur after conception—create a mosaic of genetic diversity. Copy number alterations (CNAs), where segments of the genome are deleted or duplicated, are a major form of this diversity.

While single-cell whole-genome sequencing (scWGS) has improved, detecting these CNAs remains a "noisy" nightmare. The process of amplifying a single cell's tiny amount of DNA (scWGA) introduces artifacts like allelic dropout (ADO)—where one version of a gene simply fails to amplify—and uneven coverage. Existing tools were mostly built for cancer samples with massive, chromosome-scale changes and low sequencing coverage. They often struggle to distinguish real, small-scale CNAs from amplification noise, leading to either missed detections or a flood of false positives.

Methods

HiScanner (High-resolution Single-Cell Allelic copy Number callER) introduces three primary technical innovations to clean up the signal:

  1. Adaptive Binning via HMM: Instead of letting the user guess a "bin size" (the genomic window used for analysis), HiScanner uses a two-state Hidden Markov Model (HMM) to measure the actual size of amplification artifacts (ADO events) in a specific dataset. It then sets the smallest possible bin size that stays above this noise floor.
  2. Joint Segmentation: Rather than analyzing each cell in a vacuum, HiScanner performs joint segmentation across multiple cells using the Bayesian Information Criterion (BIC). This allows the algorithm to "see" shared breakpoints across a population of cells, significantly boosting the sensitivity for clonal mutations.
  3. Robust BAF Estimation: It uses reference-based phasing and a clever two-step procedure to estimate the B-allele frequency (BAF). By looking at the ratio of maternal vs. paternal alleles and correcting for "phase-switch" errors, it adds a second layer of evidence to the standard read-depth signals.

The authors also released ViScanner, a web-based tool built on the HiGlass architecture, allowing researchers to interactively zoom into these high-resolution genomic calls.

Datasets

The authors validated HiScanner using three distinct approaches:

  • Simulated Data (SDX): They created "synthetic diploid X" chromosomes by pooling reads from different male donors. This allowed them to know the "ground truth" of every deletion and gain to test precision and sensitivity.
  • Neurotypical Brain Cells: High-coverage data from 65 neurons and oligodendrocytes across 11 healthy human brains.
  • Meningioma Tumor Serial Dataset: Low-coverage data from 179 cells sampled from a single patient at two time points (initial diagnosis and recurrence).

Key Insights and Ideas

  • The "ADO as a Ruler" Idea: One of the most elegant parts of the paper is using the distribution of allelic dropout events to determine resolution. By treating the noise as a measurable physical property of the sequencing run, they move away from arbitrary parameter tuning.
  • Neurons vs. Glia: The study revealed a striking biological dichotomy in healthy brains. Neurons mostly harbor "private" losses (deletions unique to one cell), likely occurring post-mitotically. In contrast, oligodendrocytes (a type of glial cell) showed "clonal" gains, suggesting these mutations occurred early in development and were passed down through cell division.
  • Evolutionary Tracking: In the meningioma dataset, HiScanner allowed the team to integrate CNAs with point mutations to map out a "family tree" of the tumor. They identified a specific "ancestor" clone in the initial Grade II tumor that eventually gave rise to the more aggressive Grade III recurrence.

Key Results

  • Superior Resolution: In simulations, HiScanner consistently outperformed state-of-the-art methods (Ginkgo, SCOPE, SCYN, and CHISEL). It maintained high precision (~0.96) and sensitivity (~0.90) for CNAs as small as 1 Mb, where other tools began to fail.
  • Sensitivity at Small Scales: HiScanner could detect events as small as 200 kb. In real brain data, it identified 614 small CNAs that were previously undetected; 99% of these were smaller than the 3 Mb limit typically assumed for single-cell studies.
  • Noise Reduction: When compared to Ginkgo on brain data, Ginkgo called 10 times more private losses, but the majority (up to 97% in some cell types) were "BAF-discordant," meaning they were likely false positives caused by coverage fluctuations rather than real genetic losses.

Limitations and Future Work

While HiScanner is a leap forward, the authors note that their current model assumes independence between single-cell genomes for the sake of computational efficiency. While the joint segmentation helps, a more complex model (like a deep learning architecture or conditional random fields) could potentially capture even more subtle dependencies between adjacent genomic regions.

Additionally, while the tool works on low-coverage data (~0.5x), its full "high-resolution" power is best realized with high-coverage (>20x) data, which remains expensive for large-scale studies. Future iterations may focus on further squeezing signal out of ultra-low-pass sequencing.

Summary updated

Abstract

Improvements in single-cell whole-genome sequencing (scWGS) assays have enabled detailed characterization of somatic copy number alterations (CNAs) at the single-cell level. Yet, current computational methods are mostly designed for detecting chromosome-scale changes in cancer samples with low sequencing coverage. Here, we introduce HiScanner (High-resolution Single-Cell Allelic copy Number callER), which combines read depth, B-allele frequency, and haplotype phasing to identify CNAs with high resolution. In simulated data, HiScanner consistently outperforms state-of-the-art methods across various CNA types and sizes. When applied to high-coverage scWGS data from 65 cells across 11 neurotypical human brains, HiScanner shows a superior ability to detect smaller CNAs, uncovering distinct CNA patterns between neurons and oligodendrocytes. We also generated low-coverage scWGS data from 179 cells sampled from the same meningioma patient at two time points. For this serial dataset, integration of CNAs with point mutations revealed evolutionary trajectories of tumor cells. These findings show that HiScanner enables accurate characterization of frequency, clonality, and distribution of CNAs at the single-cell level in both non-neoplastic and neoplastic cells.
BibTeX
@inproceedings{zhao2025high,
  author = {Yifan Zhao and Lovelace J. Luquette and Alexander Veit and Xiaochen Wang and Ruibin Xi and Vinay Viswanadham and Yuwei Zhang and Diane D. Shao and Christopher A. Walsh and Hong Yang and Mark D. Johnson and Peter J. Park},
  title = {{High-resolution detection of copy number alterations in single cells with HiScanner}},
  booktitle = {Nature Communications},
  year = {2025},
  doi = {10.1038/s41467-025-60446-5},
}
Figure