Accucopy

Accucopy infers total copy numbers (TCNs) and allele-specific copy numbers (ASCNs) from low-purity, low-coverage tumor sequencing data to enable accurate detection of copy number alterations.

Key Features:

• Tiered Gaussian Mixture Model: Models the distribution of sequencing coverage to distinguish different copy number states across the genome.
• Autocorrelation-Guided EM Algorithm: Employs an autocorrelation-guided Expectation-Maximization algorithm to accelerate convergence in copy number inference.
• Kernel Smoothing and Signal Processing: Applies kernel smoothing to coverage-differentiation signals and uses time-series/signal-processing concepts to estimate periods in histograms of coverage differentiation.
• Sparse Bayesian Learning (SBL): Integrates sparse Bayesian learning to enhance robustness when handling complex genomic datasets.
• Implementation and Non-human Support: Implemented in C++ and Rust and supports non-human samples for cross-species copy number analysis.

Scientific Applications:

• Cancer genomics: Detection and analysis of copy number alterations (CNAs) in tumor genomes to study oncogenesis and progression.
• Low-purity/low-coverage sequencing analysis: Inference of TCN and ASCN from low-purity, low-coverage tumor sequencing data to enable CNA analysis when sample quality or depth is limited.

Methodology:

Accucopy integrates a tiered Gaussian mixture model, an autocorrelation-guided EM algorithm, kernel smoothing of coverage differentiation, time-series/signal-processing methods for period estimation in coverage-differentiation histograms, and sparse Bayesian learning.