RDBKE

RDBKE refines breakpoint resolution for structural variant (SV) detection by applying a UNet deep segmentation model to base-wise read-depth (RD) data to produce single-nucleotide breakpoint predictions.

Key Features:

• UNet deep segmentation: Adapted UNet architecture trained on base-wise RD patterns to analyze sequencing data at single-nucleotide resolution and mitigate noise and binning artifacts in RD data.
• Enhanced breakpoint resolution: Integrates UNet predictions with RD-based SV callers to refine breakpoint positions to single-nucleotide resolution.
• Data efficiency: UNet can be effectively trained with a limited dataset, enabling application across samples without extensive retraining.
• Validation on simulated and real datasets: Demonstrated increased numbers of detectable SVs with more precise breakpoints on both simulated and real sequencing datasets.

Scientific Applications:

• High-resolution SV discovery: Improves nucleotide-level detection and mapping of structural variants in genomic studies.
• Disease-associated variant analysis: Enhances characterization of structural variants relevant to genetic disease research.
• Evolutionary genomics: Enables more precise resolution of structural variation for evolutionary biology studies.
• Personalized medicine: Supports accurate SV calls that inform individual genomic analyses and clinical interpretation.

Methodology:

Computational steps include adapting a UNet segmentation model to base-wise RD signals, training the model on RD patterns surrounding known breakpoints, and integrating UNet-derived breakpoint predictions with existing RD-based SV callers to refine breakpoint positions.