Why Does My VCF File Have More Than One Sample in It

A VCF file holds more than one sample whenever the variants were called for several people at once: research cohorts are joint-called by the hundred or thousand, and clinical sequencing delivers family trios and tumor-normal pairs in a single file. The format simply adds one column per person, so 2,504 people can sit in one file as easily as one. To get just your genome out, you extract your column with bcftools, and the free DNA file checker at healthosx.com/dna/check tells you in seconds how many samples your file actually contains.

Why does my VCF file have more than one sample in it

A VCF is a table with one row per variant. The first nine columns describe the variant itself: chromosome, position, an identifier, usually an rsID, reference allele, alternate alleles, quality, filter status, per-site annotations, and the FORMAT key. Everything after that is people, one column per sample, holding each person's genotype at every row. Nothing in the format limits a file to one person, because VCF was designed for studies, not consumer kits.

Calling variants jointly, on a whole cohort in a single pass, is statistically stronger than calling each person alone. A site where one person has two reads supporting a change looks like noise on its own; the same change appearing across hundreds of samples is plainly real, so a joint caller keeps it. The output of that pass is one VCF for the whole cohort, often split into one file per chromosome.

Three kinds of multi-sample VCF reach ordinary people:

• Research cohorts. You were in a study, or you downloaded a public one. The 1000 Genomes phase 3 files, for instance, hold 2,504 samples apiece.
• Family trios. Rare-disease diagnostics often sequence the child and both parents, and the lab ships one file with three samples.
• Tumor-normal pairs. Cancer sequencing compares tumor tissue against matched blood, and somatic callers write both into a single file, with columns often literally named TUMOR and NORMAL.

What a multi-sample VCF looks like inside

Open it in a plain text editor, never a spreadsheet: a whole-genome VCF runs to 3 to 5 million rows, and Excel will mangle it long before you see the header. Past the lines starting with ## sits a single line beginning #CHROM, something like #CHROM POS ID REF ALT QUAL FILTER INFO FORMAT HG00096 HG00097 HG00099. Every name after FORMAT is a person. Each data row then carries one cell per person in that same order. Shown as genotypes only, a row in a three-sample file might read 1|0 0|0 0|1; a full cell usually carries more fields, like 1|0:23:99, where the numbers after the genotype are read depth and genotype quality.

So "more than one sample" has a literal meaning: more than one name on the #CHROM line. Count the names and you have counted the people. The command bcftools query -l yourfile.vcf.gz prints just the sample names, without dumping millions of rows onto your screen, and the checker does the same count for you in the browser.

The tool that reads the wrong person quietly

Consumer DNA tools were built around one genome per file, the shape of a 23andMe or AncestryDNA text export. Hand one of them a multi-sample VCF and one of two failure modes is typical.

• It refuses. The parser finds more columns than it expects and rejects the file, sometimes with a clear message, sometimes with a generic invalid-format error.
• It reads the first sample column and never mentions the rest. No error, no warning. It builds the whole report, traits, medication genes, ancestry, from whoever happens to sit in column one.

The second failure is the dangerous one. If column one is your father, the tumor, or a stranger from a public cohort, the report describes them with your name on it. The common mistake is assuming the first column must be you, or that extra samples mean a more complete file. The variant rows are identical either way. The extra columns are other people, nothing more.

A multi-sample VCF is not a richer file. It is several genomes in one container, and until you extract your column, a tool may be reading somebody else's.

Get just your genome out: extract one sample with bcftools

bcftools is the free, standard toolkit for VCF files, and extraction is a one-line job.

1. 1.List the samples: bcftools query -l yourfile.vcf.gz.
2. 2.Extract yours: bcftools view -s NA12878 -Oz -o mygenome.vcf.gz yourfile.vcf.gz, putting your own sample name in place of NA12878.
3. 3.If the delivery arrived as one VCF per chromosome, which cohort files often do, run the same command on each.

No variants are lost. Every row is kept, and only the sample columns are cut down to one. That includes sites where your sample is homozygous reference and only the other people carried a variant: -s keeps those rows too, so the output has the same row count as the original file, with many rows now reading 0/0. The result is an ordinary single-sample VCF that the free report and the VCF to 23andMe converter read like any other whole genome. For a tumor-normal file, extract the NORMAL column for anything about your inherited genetics: the TUMOR column carries those same inherited variants mixed with mutations acquired during the illness, and that mix can produce nonsense in consumer tools.

Which column is actually mine?

The names on the #CHROM line are the only labels that exist. A trio file may use roles (proband, father, mother) or bare lab IDs. A tumor-normal file is the easy case: NORMAL is your constitutional genome, usually sequenced from blood. A research cohort uses coded names like HG00096, and only the study's own sample sheet maps each code to a person.

Two honest limits. If the names are coded and you have no sample sheet, no software can reliably tell you which column is yours: a column dense with Y-chromosome calls is male, which can rule columns in or out, but that narrows the field rather than proving identity. And what you may do with the other columns depends on how the file reached you. Cohort files delivered under controlled access require approval to touch any column but your own; public cohorts such as 1000 Genomes are open data anyone may analyse, so those columns carry no such restriction. Either way, the clean first step is the same: extract your own column and work from that.

Before you convert anything or run a report, spend the few seconds the check takes: drop the file on the free HealthOS DNA file checker at healthosx.com/dna/check and read the sample count it reports; among its fixes it lists extracting one sample with bcftools. It runs entirely in your browser and the file is never uploaded. Once your column is out, the file behaves like any other single-genome VCF.