kcleal / svbench

Benchmark structural variant calls against a reference set

SVBench

SVBench is a python tool/library for benchmarking variant calls vs a reference dataset.

Installation

Install using:

pip install svbench

Requires Python>=3.6, packages needed are listed in requirements.txt.

Usage

Usage: svbench [OPTIONS] REFERENCE_VCF QUERY_VCFS...

Options:
  --include PATH            Include regions .bed file
  --pass-only               Assess only PASS variants in both reference and
                            query
  --pass-ref                Assess only PASS variants in reference
  --pass-query              Assess only PASS variants in query
  --slop INTEGER            Add intervals +/- slop around breakpoints
                            [default: 250]
  --min-size-ref INTEGER    Min SV length  [default: 0]
  --min-size-query INTEGER  Min SV length  [default: 30]
  --no-duplicates           Don't quantify duplicate true positives
  --version                 Show the version and exit.
  --help                    Show this message and exit.

Benchmark the number of query SVs in a reference/truth set. Results are printed to stderr.:

svbench truthset.vcf query1.vcf

Multiple query vcfs can also be analysed:

svbench truthset.vcf query1.vcf query2.vcf ...

Python API

For a general tutorial on using the API see the ipython notebook; svbench_tutorial.ipynb