Fast VCF toolkit for bioinformaticians — normalize, liftover, filter — as a single static binary with zero dependencies.

vcfkit.dev · Docs · Install · Changelog

v0.3.0 — suitable for research pipelines and evaluation. Not validated for clinical use. Known differences from bcftools →

Three operations every VCF pipeline needs, rewritten in Rust:

vcfkit normalize -f ref.fa input.vcf > normalized.vcf
vcfkit liftover -s hg19.fa -t hg38.fa -c hg19ToHg38.over.chain.gz input.vcf > lifted.vcf
vcfkit filter -e "INFO/AF < 0.01 && FILTER == 'PASS'" input.vcf > rare_variants.vcf

All three read from stdin and write to stdout by default — pipe them freely.

Measured on 1000 Genomes chr22 (1,103,547 variants), macOS aarch64, bcftools 1.23.1:

The fast path applies to biallelic SNPs/MNPs (~80% of typical VCFs). Standard normalize uses a pure-Rust parser — correct on all inputs but slower than bcftools' C implementation. See BENCHMARKS.md for full methodology.

# Cargo
cargo install vcfkit-cli

# Pre-built binaries (macOS, Linux, Windows) — no Rust required
# See https://vcfkit.dev/install or GitHub Releases

Homebrew tap is planned. Follow the repo or watch Releases for when it lands.

Full install instructions (including Windows, ARM, shell completions): vcfkit.dev/install

# Expression filter
vcfkit filter -e "INFO/AF < 0.01" input.vcf
vcfkit filter -e "QUAL > 30 && FILTER == 'PASS'" input.vcf
vcfkit filter -e "CHROM == 'chr17' && POS >= 43044295" input.vcf
vcfkit filter -e "INFO/CSQ ~ 'missense'" input.vcf
vcfkit filter -e "FILTER == 'PASS'" --invert input.vcf   # keep non-PASS

# Natural-language filter via Claude (requires ANTHROPIC_API_KEY)
vcfkit filter --ask "rare PASS variants on chromosome 17" input.vcf
vcfkit filter --ask "rare PASS variants" --yes input.vcf   # skip confirmation

Fields: INFO/*, FORMAT/*, CHROM, POS, QUAL, FILTER
Operators: < <= > >= == != && || ! ~ (contains) !~

With --ask: the LLM sees only your query and the VCF header schema. Variant data never leaves your machine. The translated expression is shown for review before filtering runs.

vcfkit normalize -f hg38.fa input.vcf > normalized.vcf
vcfkit normalize --fast -f hg38.fa input.vcf > normalized.vcf   # 4× faster for SNPs/MNPs
vcfkit normalize -f hg38.fa --no-split input.vcf                 # keep multi-allelic
vcfkit normalize -f hg38.fa --check-ref error input.vcf          # strict REF validation

vcfkit liftover -s hg19.fa -t hg38.fa -c hg19ToHg38.over.chain.gz input.vcf
vcfkit liftover ... -r rejects.vcf input.vcf   # keep unmapped records

Download chain files: vcfkit liftover --list-chains

vcfkit.dev — all three operations run in WebAssembly in your browser. Nothing uploads anywhere.

git clone https://github.com/robertlangdonn/vcfkit
cd vcfkit
cargo build --release
./target/release/vcfkit --version

Requires Rust 1.75+. No C dependencies, no htslib.

Validated against bcftools in differential tests on 1000 Genomes chr22 (1.1M real variants). Tests run nightly in CI. Known divergences are documented in docs/known_differences.md.

vcfkit builds on the work of:

• htslib / bcftools — Wellcome Sanger Institute, Heng Li, Petr Danecek, and contributors. The reference implementation vcfkit validates against.
• noodles by Michael Macias — the pure-Rust VCF/BCF/FASTA I/O library vcfkit builds on.
• Tan, Abecasis, Kang 2015 — the normalization algorithm implemented in vcfkit normalize.
• UCSC Genome Browser — chain files for liftover.

Portions of this codebase were written with assistance from Claude (Anthropic). See CREDITS.md for full attribution.

MIT