lra

lra aligns long reads from single-molecule sequencing platforms (PacBio and Oxford Nanopore Technologies) and megabase-scale contigs to reference genomes to enable detection of structural variation using sparse dynamic programming.

Key Features:

• Sparse Dynamic Programming (SDP): Identifies exact matches between query and reference and forms optimal chains that represent preliminary alignments.
• Convex Gap Penalty Scoring: Uses a convex-cost gap penalty function for alignment scoring to better model sequence variation and inversions compared to linear-cost gap functions.
• Read and Contig Support: Aligns long reads from PacBio and Oxford Nanopore Technologies and megabase-scale contigs derived from SMS assemblies.
• Structural Variant Detection Improvements: Provides additional evidence for SV calls on PacBio and on ONT improves sensitivity/specificity with a nominal 0.2–0.4% F1 increase over minimap2 (Truvari) and, with Sniffles, yields 3% more calls than minimap2 and 30% more than ngmlr with a 4.6–5.5% F1 increase.
• Contig-based SV Gains: For de novo assembly contigs, increases SV calls by 5.8% relative to minimap2+paftools and improves Truvari F1 by 4.3%.
• Performance: Demonstrates runtime between 52–168% of minimap2 when generating SAM alignments and 9–15% relative to ngmlr for alternative methods.
• SAM Output: Produces alignments in SAM format for downstream analysis and benchmarking.

Scientific Applications:

• Long-read mapping for variant discovery: Mapping PacBio and ONT reads to detect structural variants and support variant calls.
• Structural variant calling and benchmarking: Generating alignments used by SV callers such as Sniffles and evaluated with Truvari to improve call sets and F1 metrics.
• Assembly-based SV discovery: Aligning megabase-scale contigs from de novo assemblies to detect additional structural variants compared with minimap2+paftools.
• Modeling complex genomic structures: Representing inversions and complex gaps through convex gap-penalty alignment scoring.

Methodology:

Uses sparse dynamic programming to identify exact matches and chain them into preliminary alignments, scoring those alignments with a convex gap-penalty function.