SVNN

SVNN detects structural variations in PacBio long-read sequencing data to increase sensitivity and speed of SV calling for variants larger than 50 base pairs, particularly in low-coverage datasets.

Key Features:

• Input Flexibility: Accepts raw long reads, including PacBio reads, as input without requiring prior preprocessing.
• Variant Size Threshold: Targets structural variants larger than 50 base pairs.
• Aligners: Employs Minimap2 for initial rapid mapping and feature extraction and NGMLR to realign a selected subset of reads for precise downstream analysis.
• Callers: Integrates Sniffles and SVIM for structural variation calling.
• Neural Network Integration: Uses a neural network to analyze Minimap2 output and identify the subset of reads most informative for SV detection, guiding NGMLR realignment.
• Performance Efficiency: Reports up to a 20 percentage point increase in sensitivity compared to existing state-of-the-art methods and approximately threefold faster runtime than traditional tool combinations while maintaining comparable accuracy.
• Low-Coverage Optimization: Optimized to improve detection accuracy in low-coverage long-read datasets with high per-read error rates.

Scientific Applications:

• Complex genome SV profiling: Detection and characterization of structural variants in complex genomes using long-read data.
• Low-coverage sequencing studies: SV discovery in projects constrained to low-coverage PacBio sequencing.
• PACBIO-based variant analysis: Structural variation analysis workflows that rely on PacBio long-read sequencing technologies.

Methodology:

Minimap2 performs an initial rapid alignment and feature extraction, a neural network analyzes Minimap2 output to select informative reads, the selected reads are realigned with NGMLR, and Sniffles and SVIM perform structural variation calling on the NGMLR alignments.