SECEDO

SECEDO identifies single-nucleotide variant (SNV)-defined tumor subclones from ultra-low coverage single-cell DNA sequencing data to characterize intra-tumor heterogeneity.

Key Features:

• Ultra-Low Coverage Capability: Operates effectively with sequencing coverage at or below 0.05x per cell.
• Bayesian Filtering and Clustering: Employs Bayesian filtering to identify relevant loci and clusters cells based on inferred SNVs.
• Read Overlap and Phasing Exploitation: Leverages read overlap and phasing information to enhance subclone detection accuracy at minimal depth.
• High Sensitivity at Low Coverage: Detected 92.11% of somatic SNVs on a synthetic dataset, including SNVs in small clusters representing 6.9% of the population.
• Real-Data Performance: Achieved an Adjusted Rand Index (ARI) of approximately 0.6 at 0.03x coverage on a breast cancer single-cell sequencing dataset.
• Improved Variant Calling on Subclusters: Calling variants on derived subclusters more than doubled the number of detected SNVs and improved allelic ratios versus non-clustered calling.
• Outperforms Existing SNV-Based Methods: Outperformed state-of-the-art SNV-based clustering methods, which required substantially higher coverage to match its performance.

Scientific Applications:

• SNV-driven intra-tumor heterogeneity analysis: Enables detection of SNV-defined subclonal architecture in tumors not primarily driven by copy number alterations.
• Tumor evolution and treatment resistance studies: Facilitates resolution of subclonal SNV composition to inform analyses of tumor evolution and mechanisms of treatment resistance.

Methodology:

Uses Bayesian filtering to identify relevant loci, exploits read overlap and phasing information, clusters genetically similar cells based on inferred SNVs, performs variant calling on identified subclusters, and was validated on a simulated dataset (7,250 cells from eight subclones) and on a breast cancer single-cell sequencing dataset.