DINTD

DINTD detects and infers tandem duplications (TDs) from short sequencing reads generated by next-generation sequencing technologies to characterize structural variations relevant to cancer evolution and tumor genesis.

Key Features:

• Tandem duplication detection: Detects and infers tandem duplications (TDs) from short sequencing reads produced by next-generation sequencing technologies.
• Signal extraction: Extracts read depth and mapping quality signals from sequencing data.
• Uneven read handling: Addresses the challenges of uneven distribution of reads and uncertain amplitude of TD regions.
• Clustering: Uses DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to identify spatially dense candidate TD regions and filter out noise.
• Denoising: Applies a total variation penalized least squares model to denoise read depth and mapping quality signals.
• Neighbor search: Implements a 2D binary search tree structure to efficiently search neighboring points.
• Breakpoint resolution: Integrates split-read signals to pinpoint exact breakpoints within TD regions.
• Validation and performance: Demonstrated on simulated datasets and real samples to improve sensitivity, precision, F1-score, and reduce boundary bias compared to existing methods.

Scientific Applications:

• Cancer evolution and tumor genesis: Analysis of TDs implicated in cancer evolution and tumor genesis.
• Early detection, diagnosis, and treatment studies: Identification of TDs that inform early detection, diagnostic biomarkers, and therapeutic targets.
• Structural variant interpretation: High-precision breakpoint identification for interpretation of structural variations.
• Method benchmarking: Comparative evaluation of TD detection methods using simulated datasets and real samples.

Methodology:

Extract read depth and mapping quality signals from short sequencing reads; denoise signals with a total variation penalized least squares model; cluster candidate points using DBSCAN (Density-Based Spatial Clustering of Applications with Noise) aided by a 2D binary search tree for neighbor queries; integrate split-read signals to refine breakpoint positions.