SeCNV

SeCNV profiles copy number variation (CNV) from single-cell DNA sequencing (scDNA-seq) data to segment the genome and estimate per-cell copy numbers for analysis of genomic heterogeneity.

Key Features:

• Depth Congruent Map (DCM): Constructs a DCM using a local Gaussian kernel to capture genomic similarities between bins.
• Structural entropy segmentation: Partitions the genome by minimizing structural entropy computed from the DCM.
• Per-cell copy number estimation: Estimates copy numbers for individual cells within the derived genomic partitions.
• Robustness to noise and low coverage: Specifically addresses high noise levels and low coverage characteristic of scDNA-seq data.
• Benchmark performance: Achieved F1-scores higher than 0.95 for breakpoint detection across nine simulated datasets with varied breakpoint distributions and noise amplitudes.
• Scalability: Processes datasets comprising over 50,000 cells within four minutes, whereas comparison methods failed to complete within a 120-hour limit.
• Real-data validation: Applied to single-nucleus sequencing datasets from two breast cancer patients and acoustic cell tagmentation sequencing datasets from eight patients to identify subclones and characterize tumor heterogeneity.

Scientific Applications:

• Single-cell CNV profiling: Detection and quantification of CNVs from scDNA-seq data.
• Breakpoint detection benchmarking: Evaluation of breakpoint calling accuracy under varied simulated breakpoint distributions and noise amplitudes.
• Large-scale CNV analysis: Analysis of tens of thousands of single cells to resolve population-level and rare CNV events.
• Cancer subclone identification: Identification of distinct subclones and inference of tumor heterogeneity in breast cancer and other sequencing modalities.

Methodology:

Builds a depth congruent map (DCM) using a local Gaussian kernel, partitions the genome by minimizing structural entropy on the DCM, and estimates copy numbers for individual cells within those partitions.