BAM, CRAM, FASTQ or VCF: Which Whole-Genome File Do I Use?

For almost every consumer DNA tool, use the VCF (the file ending .vcf.gz). It lists where your genome differs from the reference, and it's what trait, health, medication and ancestry tools read. FASTQ, BAM and CRAM are the raw sequencing reads underneath it: tens of gigabytes, and only for bioinformatics software. You can read a VCF free, in your browser, at healthosx.com/dna.

The files are steps in a pipeline

Sequencing produces your data in stages, and each file is the output of one stage:

FASTQ: the raw reads

The sequencer reads your DNA as hundreds of millions of short fragments, each around 150 letters. The FASTQ file lists every fragment and a quality score for every letter. At 30x coverage that's roughly 100GB uncompressed, often split across several files. Nothing is placed on the genome yet.

When you need it: only to re-run the whole analysis from scratch with different software.

BAM and CRAM: the reads, aligned

Alignment works out where on the human reference genome each fragment belongs. A BAM stores the aligned reads. A CRAM stores the same thing, compressed further against the reference. Both come with an index (.bai or .crai). Expect tens of gigabytes.

When you need them: to call variants again with better software, to look at the evidence behind one particular variant, or for things a VCF can't capture — copy-number changes such as CYP2D6 being the classic consumer example.

VCF: your variants

Variant calling compares the aligned reads with the reference and writes down every position where you differ. That's about 4 to 5 million rows for a whole genome, typically 100MB to 1GB compressed. The .tbi file beside it is an index.

When you need it: almost always. Every consumer report, ancestry tool and variant lookup reads this.

Some providers split it up: Dante Labs, for example, ships a .snp.vcf.gz and an .indel.vcf.gz. Consumer findings are nearly all SNPs, so use the SNP file. A gVCF also records stretches where you match the reference, which makes it much larger; use the ordinary VCF if you have one.

Two things to check in your VCF

1. 1.rsIDs. If the third column is . on every row, tools that search by rsID find nothing. How to check and fix it.
2. 2.Build. The ##reference or ##contig header lines say whether it's GRCh37 or GRCh38. Positions differ between the two.

Need your VCF as a 23andMe-style file for a tool that only takes those? Convert it free in your browser. The same VCF also gives your maternal and paternal haplogroups, and the file checker reads its build and rsIDs for you.

What your VCF gets you on /dna

healthosx.com/dna

Where each finding sits in your genome.

healthosx.com/dna

Medication genes, kept behind a tap rather than sprung on you.

The sample genome, read in the browser. A whole-genome VCF carries far more markers than the sample.

Your browser streams even a multi-gigabyte VCF, locally, into findings on medication genes, inherited risk factors (lipoprotein(a), iron overload, alpha-1 antitrypsin, LRRK2, TTR), food and traits. Ancestry is optional. Ask your DNA can go deeper on a stored whole genome, with haplogroups, polygenic scores and any variant you name. Company notes: Nebula · Dante Labs · Sequencing.com.

What to keep, and what to delete

Storage is the practical question once you have all four files:

• Keep the VCF forever. It's small, and it's what every tool reads.
• Keep the CRAM if you can. It's the only file that lets anyone re-call your variants later with better software, or work out something a VCF can't capture — CYP2D6 copy number being the standard example. A CRAM is smaller than the BAM it came from.
• FASTQ is usually safe to delete once you have a CRAM or BAM, because the reads are in there, aligned. Keep it only if you plan to re-align against a different reference.
• Store what you keep encrypted. A genome identifies you and your relatives permanently.

Checking a file is what it claims to be

You don't need bioinformatics tools for a sanity check:

• A VCF starts with header lines beginning ##, and its data rows have the chromosome and position in the first two columns.
• A FASTQ repeats in blocks of four lines, the first starting with @.
• A BAM or CRAM is binary. Opening it in a text editor shows unreadable characters — that's normal, not corruption.
• Size is the quickest tell. Hundreds of megabytes means variants; tens of gigabytes means reads.

FAQ

My provider only gave me a BAM. Can I get a VCF?

Yes, but it needs variant calling, which is real bioinformatics work (bcftools or GATK against a matching reference). Ask the provider first — most can export a VCF.

What's a .tbi or .csi file?

An index that lets software jump to a position without reading the whole file. Harmless, and not needed for a browser read.

Is a gVCF fine?

It works, but it's much bigger because it also records the stretches where you match the reference. Use the plain VCF if you have one.

Found your VCF? Choose it and read it now: healthosx.com/dna