Blackbird: structural variant detection using synthetic and low-coverage long-reads.

Motivation: Recent benchmarks show that most structural variations, especially within 50-10,000 bp range cannot be resolved with short-read sequencing, but long-read structural variant callers perform better on the same datasets. However, high-coverage long-read sequencing is costly and requires substantial input DNA. Reducing coverage lowers cost but significantly impacts the performance of existing structural variation (SV) callers. Synthetic long-read technologies offer long-range information at lower cost, but leveraging them for SVs under 50 kbp remains challenging.

Results: We propose a novel hybrid alignment- and local-assembly-based algorithm, Blackbird, that uses synthetic long reads and low-coverage long reads to improve structural variant detection. Instead of relying on whole-genome assembly, Blackbird uses a sliding window approach and synthetic long-read barcode information to assemble local segments, integrating long reads to improve structural variant detection accuracy. We evaluated Blackbird on real human genome datasets. On the HG002 Genome in a Bottle (GIAB) benchmark, Blackbird in hybrid mode demonstrated results comparable to state-of-the-art long-read tools, while using less long-read coverage. Blackbird requires only 5 $\times$ coverage to achieve F1-scores (0.835 and 0.808 for deletions and insertions) similar to PBSV and Sniffles2 using 10 $\times$ PacBio Hi-Fi long-read coverage.

Availability and implementation: Blackbird is available at https://github.com/1dayac/Blackbird.