Introduction

vcfkit is a fast, single-binary VCF toolkit for bioinformaticians. It does three things every pipeline needs — normalize, liftover, and filter — rewritten in Rust, validated nightly against bcftools on real 1000 Genomes chr22 data.

No htslib. No Python. No C dependencies. One static binary that works on macOS, Linux, and Windows.

What it does

normalize

Left-align indels, split multi-allelic sites, validate against a reference FASTA. A fast path handles biallelic SNPs/MNPs without full parsing (~80% of typical VCFs).

normalize docs →

liftover

Convert between genome builds — hg19, hg38, T2T-CHM13 — using UCSC chain files. Handles strand flips and b37/UCSC contig name mismatches automatically.

liftover docs →

filter

Keep variants matching expressions over INFO, FORMAT, CHROM, POS, QUAL, and FILTER fields. Or describe the filter in plain English with --ask and review before it runs.

filter docs →

Quick start

# Install (pre-built binary or Cargo)
# See https://vcfkit.dev/install
cargo install vcfkit-cli

# Filter: keep rare variants
vcfkit filter -e "INFO/AF < 0.01" input.vcf

# Normalize: split multi-allelic sites
vcfkit normalize -r ref.fa input.vcf

# Liftover: hg19 → hg38
vcfkit liftover --chain hg19ToHg38.chain.gz input.vcf

# Filter with plain English (requires ANTHROPIC_API_KEY)
vcfkit filter --ask "rare variants in Europeans but common in Africans" input.vcf

How it compares to bcftools

vcfkit does not replace bcftools. It does three operations well, faster, with a simpler interface. Key differences:

• 4× faster on normalize (biallelic SNP/MNP fast path) and filter (expression evaluation) — see benchmarks
• Single static binary — no shared libraries, no conda environment, no htslib
• --ask mode — translates plain English to a deterministic filter expression using Claude; you review before it runs
• WASM build — the same core runs in the browser at vcfkit.dev

Known divergences from bcftools are documented.

Performance

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

Operation	vcfkit	bcftools	Speedup
filter -e 'INFO/AF < 0.01'	422 ms	1,695 ms	4.0×
normalize --fast --no-split	682 ms	2,820 ms	4.1×
normalize (standard path)	6,481 ms	2,820 ms	0.43×
liftover	6,713 ms	—	~164K rec/s

The fast path applies to biallelic SNPs and MNPs (~80% of typical VCFs). Standard normalize uses full noodles parsing — slower than bcftools’ C implementation but required for correctness on indels and multi-allelic records. Full methodology →

Not for clinical use

vcfkit is research and pipeline tooling. It is not validated for clinical or diagnostic use.