Structural variants

A short read of about a hundred bases cannot span a large rearrangement. Callers therefore infer indirectly from signals: pairs that map too far apart or in the wrong orientation, reads split at a breakpoint, or regions with conspicuously high or low coverage. Repetitive regions further complicate unambiguous assignment because a read can map to several places. From these traces the most plausible rearrangement is reconstructed, not read off directly.

Genome uses Manta (configManta.py configures the run, runWorkflow.py executes it locally). Manta first scans the entire alignment for SV evidence: discordant read pairs with atypical spacing or wrong orientation, and reads split across a breakpoint. From these cues Manta builds a graph of possible connections between distant genome regions and re-assembles the breakpoints locally to reconstruct the junction sequence at base resolution where possible. Each candidate receives a probability-based quality score and a filter status (PASS or a filter tag); Genome takes the file diploidSV.vcf.gz from the run as the technical result. The findings stay candidates with evidence; assessment and confirmation are a qualified step.

Short-read data detect variants with varying reliability, depending on type and location. Deletions, which become visible through pairing distance and coverage, are comparatively reliable; balanced inversions and translocations are harder because they leave traces only at the breakpoints. Insertions longer than the read length often cannot be reconstructed in full, and variants within segmental duplications or repetitive regions frequently stay invisible. Manta limits local assembly to compact breakpoints, so very large insertions are usually reported only as a breakend. Long reads complement this picture because they can span entire rearrangements.