Skip to content

Step 3d: Octopus Variant Caller (Alternative)

Haplotype-aware Bayesian variant caller. Optional 5th caller for benchmarking or potential FreeBayes replacement.

What It Does

Octopus jointly calls SNPs and indels using a haplotype-based model that considers multiple candidates simultaneously. Unlike GATK's fixed-ploidy model, Octopus uses a variable-ploidy approach that handles complex regions more accurately.

Why

  • Better than FreeBayes in benchmarks — higher precision with comparable sensitivity, and significantly faster
  • Lower memory than FreeBayes — typically < 8 GB vs FreeBayes' 13+ GB peaks
  • Multithreaded — unlike FreeBayes which is single-threaded
  • Haplotype-aware — considers multiple variant candidates in a window, reducing false calls from alignment artifacts

Tool

Octopus v0.7.4 — Bayesian haplotype-based mutation calling.

Docker Image

  • OCTOPUS_IMAGE

Pinned in versions.env; Image versions lists the current tag.

Command

export GENOME_DIR=/path/to/data
./scripts/03d-octopus.sh <sample_name>

# Test on one chromosome first:
INTERVALS=chr22 ./scripts/03d-octopus.sh <sample_name>

Output

File Location Description
VCF vcf_octopus/<sample>.vcf.gz Germline variant calls

Runtime

Dataset Threads Time Memory
30X WGS 8 2-4 hours ~8-12 GB
chr22 only 4 ~5-10 min < 4 GB

Notes

  • Outputs to vcf_octopus/ to maintain isolation from the default DeepVariant calls
  • Supports ALIGN_DIR variable for BWA-MEM2 alignments: ALIGN_DIR=aligned_bwamem2 ./scripts/03d-octopus.sh sample
  • Supports INTERVALS for region-restricted testing: INTERVALS=chr22 ./scripts/03d-octopus.sh sample
  • The benchmark script (benchmark-variants.sh) auto-discovers vcf_octopus/ for pairwise concordance analysis
  • Octopus is a good candidate to replace FreeBayes (which is slow, single-threaded, and memory-heavy) as the default alternative caller