Resolving single-cell copy number profiling for large datasets

Ruohan Wang, Yuwei Zhang, Mengbo Wang, Xikang Feng, Jianping Wang, Shuai Cheng Li

TL;DR

This paper introduces SeCNV, a novel computational method for accurately and efficiently profiling copy number variations (CNVs) in single cancer cells. SeCNV leverages "structural entropy" to overcome the challenges of high noise and low coverage in single-cell DNA sequencing data, enabling robust analysis of large datasets and revealing intricate tumor heterogeneity.

What Problem Does This Paper Address?

Single-cell DNA sequencing (scDNA-seq) holds immense promise for understanding cancer's genetic diversity. However, analyzing this data to identify copy number variations (CNVs) – duplications or deletions of DNA segments that drive cancer progression – is incredibly challenging. The data is inherently noisy and has low coverage, making it difficult for existing computational tools, often designed for bulk sequencing, to accurately detect CNVs. Furthermore, many current methods are computationally intensive, failing to scale to the large datasets (tens of thousands of cells) now becoming common in single-cell research. This gap limits our ability to resolve complex tumor subclones and understand cancer evolution at a granular level.

Methods

SeCNV tackles these challenges with a multi-step approach:

  • Bin-specific Normalization: Instead of relying solely on common biases like GC content, SeCNV identifies normal cells using a density-based method on coefficient of variation (CV) values. These normal cells then serve as negative controls to infer and correct for more nuanced, bin-specific biases present in the data.
  • Depth Congruent Map (DCM): SeCNV constructs a DCM that captures the similarity of copy number states between different genomic bins across all cells. This is achieved using a local Gaussian kernel, which helps to focus on genuine copy number differences while downplaying noise.
  • Structural Entropy for Segmentation: The core innovation is the use of "structural entropy" to segment the genome. By minimizing this measure on the DCM, SeCNV partitions the genome into segments with consistent copy number states. This approach is inspired by its successful application in analyzing graph structures in other biological contexts.
  • Integer Copy Number Estimation: Finally, SeCNV estimates the absolute copy number for each segment, employing an integer approximation algorithm that considers cell ploidy.

Datasets

The authors evaluated SeCNV using:

  • Simulated Datasets: Nine simulated datasets were generated with varying breakpoint distributions, normal cell percentages, and noise amplitudes to rigorously benchmark SeCNV against existing methods.
  • Single-Nucleus Sequencing (SNS) Data: Data from two breast cancer patients (one genetically heterogeneous, one homogeneous) were analyzed to demonstrate SeCNV's ability to reproduce distinct subclones and infer tumor heterogeneity.
  • Acoustic Cell Tagmentation (ACT) Sequencing Data: Datasets from eight breast cancer patients, each with approximately 1,000 cells, were used to showcase SeCNV's scalability and its capacity to identify superclones and subclones.

Key Insights and Ideas

  • Structural Entropy as a Novel Objective: The application of structural entropy for genomic segmentation is a key novelty, demonstrating its power in capturing complex patterns in noisy biological data.
  • Cross-Sample Segmentation is Crucial: Jointly segmenting across multiple cells (cross-sample segmentation) significantly improves breakpoint detection sensitivity and leverages shared information, which is vital for resolving clonal structures in cancer.
  • Robustness to Noise: SeCNV's structural entropy-based approach proves remarkably resilient to the high noise inherent in scDNA-seq data, outperforming statistical methods in challenging scenarios.
  • Scalability is a Game-Changer: The ability to process massive datasets (50,000+ cells) in minutes, compared to hours or days for other tools, democratizes large-scale single-cell CNV analysis.

Key Results

  • High Accuracy: SeCNV achieved F1-scores higher than 0.95 for breakpoint detection on simulated datasets, significantly outperforming state-of-the-art methods.
  • Exceptional Speed: SeCNV processed datasets of over 50,000 cells in under four minutes, while other tools failed to complete within 120 hours.
  • Biological Validation: SeCNV successfully identified distinct subclones in real breast cancer patient data (SNS and ACT-seq), with results highly concordant with FACS and arrayCGH ground truth data.
  • Reproduced Findings: The method reproduced known tumor heterogeneity patterns and identified clonal structures consistent with previous studies.

Limitations and Future Work

  • Limitations: The paper notes that while SeCNV performs well, the performance of any CNV detection method can be influenced by the quality and depth of the sequencing data. Some cells with extremely low coverage were excluded from analysis.
  • Future Work: The authors suggest that SeCNV's structural entropy framework could be extended to other biological segmentation and clustering problems. The ability to infer tumor evolutionary history from genetic heterogeneity is also highlighted as a promising future direction.

Summary updated

Abstract

ABSTRACT The advances of single-cell DNA sequencing (scDNA-seq) enable us to characterize the genetic heterogeneity of cancer cells. However, the high noise and low coverage of scDNA-seq impede the estimation of copy number variations (CNVs). In addition, existing tools suffer from intensive execution time and often fail on large datasets. Here, we propose SeCNV, a novel method that leverages structural entropy, to profile the copy numbers. SeCNV adopts a local Gaussian kernel to construct a matrix, depth congruent map , capturing the similarities between any two bins along the genome. Then SeCNV partitions the genome into segments by minimizing the structural entropy from the depth congruent map. With the partition, SeCNV estimates the copy numbers within each segment for cells. We simulate nine datasets with various breakpoint distributions and amplitudes of noise to benchmark SeCNV. SeCNV achieves a robust performance, i.e., the F1-scores are higher than 0.95 for breakpoint detections, significantly outperforming state-of-the-art methods. SeCNV successfully processes large datasets (>50,000 cells) within four minutes while other tools failed to finish within the time limit, i.e., 120 hours. We apply SeCNV to single-nucleus sequencing (SNS) datasets from two breast cancer patients and acoustic cell tagmentation (ACT) sequencing datasets from eight breast cancer patients. SeCNV successfully reproduces the distinct subclones and infers tumor heterogeneity. SeCNV is available at https://github.com/deepomicslab/SeCNV .
BibTeX
@article{wang2022resolving,
  author = {Ruohan Wang and Yuwei Zhang and Mengbo Wang and Xikang Feng and Jianping Wang and Shuai Cheng Li},
  title = {{Resolving single-cell copy number profiling for large datasets}},
  journal = {Briefings in Bioinformatics},
  year = {2022},
  doi = {10.1093/bib/bbac264},
}

Loading related content...

Figure