Skip to content

Active Region Detection

Before assembling the de Bruijn graph for a window, Lancet2 runs a fast pre-scan to check whether the region contains any evidence of variation. Windows with no mutation signal are skipped entirely, reducing WGS runtime by ~80%.

How It Works

The active region detector iterates all reads overlapping the window using zero-copy alignment proxies (no sequence or quality deep-copy). For each read, it inspects three evidence sources:

  1. MD tag mismatches — If the BAM contains MD tags, each mismatch position (with base quality ≥ Q20) is tracked. If ≥2 reads show a mismatch at the same genomic position, the window is active.

  2. CIGAR-based events — Insertions, deletions, and explicit sequence mismatches (X ops) are counted at each genomic position. Again, if ≥2 independent reads report the same event type at the same position, the window is active.

  3. Soft clips — If a read contains soft-clipped bases, the boundary positions are tracked. ≥2 soft clips at the same position trigger an active call.

The detector terminates immediately on the first evidence hit (≥2 at any position), so the cost is O(R) in the best case and O(R × C) in the worst case, where R = number of overlapping reads and C = mean CIGAR length.

The MD Tag Requirement

BAMs without MD tags cause ~5–10× slower runtime

If the input BAM/CRAM files lack the MD auxiliary tag, Lancet2 automatically disables active region detection and assembles every window. This is because the MD tag is the only source of per-base mismatch information when the CIGAR uses generic M operations (which do not distinguish matches from mismatches).

You will see a runtime log message:

MD tag not found in input BAM. Disabling active region detection.

To restore fast-skip behavior, generate MD tags before running Lancet2:

samtools calmd -b input.bam reference.fasta > input_with_md.bam
samtools index input_with_md.bam

--no-active-region

This flag forces Lancet2 to assemble every window regardless of mutation evidence. Use cases:

  • Completeness auditing — ensure no variant is missed by the heuristic at the cost of runtime.
  • Known variant regions — if you know a window contains a variant that the heuristic might miss (e.g., a mosaic variant with only 1 supporting read per position), forcing assembly guarantees it will be evaluated.

Enabling this flag has the same runtime impact as BAMs lacking MD tags — expect 5–10× longer wall-clock time on WGS runs.