What Is in a Genome Report, and What It Costs

A genome report is a human-readable document, usually 5 to 40 pages, that summarizes a handful of variants selected from the millions found in your genome. It is a filter, not the data. A 30x whole-genome run produces roughly 3.5 million single-nucleotide variants, 500,000 to 900,000 short indels, and a few thousand structural variants relative to GRCh38. A clinical report narrows that to somewhere between zero and a dozen lines. Everything interesting about the exercise happens in the filter, which is why the file you should care about is the VCF and the aligned reads behind it, not the PDF. Current pricing: a research-grade 30x WGS with raw data runs roughly $300 to $1,000, a clinical diagnostic WGS with interpretation from a CLIA/CAP lab runs $1,500 to $5,000 and is often billed to insurance, and consumer array products (“DNA tests” that genotype 600,000 sites rather than sequencing the genome) are $50 to $200 and are a different measurement entirely.

The files, ranked by how long they stay useful

Ask for all of these. Labs that will not release them are selling you a subscription to their interpretation.

• CRAM (or BAM) plus index. Aligned reads. A 30x human CRAM 3.0 against GRCh38 is about 15–25 GB; the equivalent BAM is 60–100 GB. This is the only file that lets you re-call variants with a better caller in five years, inspect a specific locus in IGV, or check whether a “variant” is three reads of noise at the edge of a homopolymer.
• gVCF. Per-position genotype likelihoods, including reference calls. Distinguishes “reference at this site” from “no coverage here,” which the plain VCF does not. Needed for joint calling and for honest negative results.
• VCF 4.2/4.3. The variant calls. ~1–2 GB uncompressed for a genome, ~150 MB bgzipped.
• FASTQ. Only if you distrust the alignment. 100+ GB, and you will almost never go back to it.
• The report PDF. Useful as a record of what the lab asserted on a given date with a given ClinVar snapshot. Variant classifications change. The PDF does not.

Store the CRAM and the reference FASTA it was aligned to. CRAM is reference-compressed, so a CRAM without the exact matching reference (including the specific GRCh38 build, with or without alt contigs) is partly unreadable. samtools view -T GRCh38_full_analysis_set_plus_decoy_hla.fa and keep that FASTA next to the data.

What a clinical report chooses to say

Most clinical genome reports are built from four buckets, and each has its own rules:

1. Primary findings. Variants in genes related to the indication that prompted the test. If there is no indication, this section is empty.
2. Secondary findings. The ACMG SF v3.2 list, 81 genes where a pathogenic or likely pathogenic variant is considered medically actionable regardless of why you were sequenced: BRCA1/2, the Lynch syndrome mismatch repair genes, cardiomyopathy and arrhythmia genes, familial hypercholesterolemia genes, MEN1/2, RET, TTN in a restricted way, and so on. Only a low single-digit percent of people carry a reportable variant here.
3. Carrier status. Heterozygous variants in recessive genes. Relevant for reproductive planning, not for you.
4. Pharmacogenomics. Star-allele diplotypes for CYP2C19, CYP2D6, DPYD, TPMT, SLCO1B1, and about a dozen others, mapped to CPIC guideline levels.

Everything else — the 3.5 million other variants — is dropped, because nobody can say anything defensible about them. Note what this excludes. The ENCODE pilot project, annotating 1% of the genome in depth, found pervasive transcription and far more functional elements outside protein-coding exons than the gene-centric view predicted.¹ Roughly 98% of your variants sit in that territory, and interpretation methods for them remain weak. A report that says nothing about your non-coding variation is being accurate about the state of the field.

Building your own annotated report from a VCF

This is the part worth doing yourself. The pipeline we would run:

# 1. Normalize. Non-negotiable before any join against ClinVar or gnomAD.
bcftools norm -m -any -f GRCh38.fa -Oz -o norm.vcf.gz input.vcf.gz
bcftools index -t norm.vcf.gz

# 2. Annotate with VEP, offline, pinned cache version.
vep -i norm.vcf.gz --cache --offline --assembly GRCh38 \
    --vcf --everything --pick_allele_gene \
    --plugin AlphaMissense,file=AlphaMissense_hg38.tsv.gz \
    --plugin SpliceAI,snv=spliceai_snv.hg38.vcf.gz,indel=spliceai_indel.hg38.vcf.gz \
    --custom gnomad.v4.genomes.sites.vcf.gz,gnomADg,vcf,exact,0,AF,AF_nfe,AF_afr,AF_eas \
    --custom clinvar_20260601.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT \
    -o annotated.vcf.gz --compress_output bgzip --fork 8

Then filter. The single most productive filter is population frequency, and the trap is which population. Frequency panels are ancestry-specific: deep sequencing of 1,070 Japanese individuals found large numbers of variants absent from existing reference panels, most of them rare and population-specific.² If your ancestry is poorly represented in gnomAD, a variant that looks “rare and therefore suspicious” at AF < 0.0001 in non-Finnish Europeans may be common in your own background. Always read the ancestry-stratified AF fields, never the global one alone.

Second filter: ClinVar review status. CLNREVSTAT values of criteria_provided,_multiple_submitters,_no_conflicts (two stars) or reviewed_by_expert_panel (three stars) are worth reading. Single-submitter, no-assertion entries are a large fraction of ClinVar and a large fraction of the false alarms in DIY analysis.

Third: check coverage before you believe any negative. mosdepth --by targets.bed sample cram and look at the fraction of your genes of interest with depth ≥ 20. Short-read WGS systematically under-covers GC-rich first exons, segmental duplications, and paralog-rich genes. PMS2, SMN1, CYP2D6, and the HLA region are the classic offenders and need dedicated tools: PharmCAT or Aldy for star alleles, ExpansionHunter for repeat expansions (HTT, FMR1, C9orf72), GATK gCNV or Manta/Delly for structural variants, HLA-LA for HLA typing. A standard VCF-based report silently says nothing about all of these.

If a DIY pipeline flags a pathogenic variant in an actionable gene, do not act on it. Have it confirmed by orthogonal method in a CLIA/CAP lab and interpreted with a genetic counselor or medical geneticist. Research pipelines produce false positives at rates that matter when the consequence is a surgical decision.

What it costs, and why the sticker price stopped falling

The $1,000 genome was named as a target in 2005, when the cost was six figures and the argument was about sequencing chemistry and instrument throughput.³ Chemistry won. Reagent cost per gigabase is no longer the binding constraint. The binding constraint is interpretation labor: a board-certified variant scientist reviewing candidate variants, curating literature, and writing the report. A systematic review of whole-exome and whole-genome cost-effectiveness studies found wide variation in reported costs and, more to the point, that most studies did not fully account for downstream interpretation and confirmatory testing.⁴ That is why a $400 sequencing run turns into a $3,000 clinical test.

The practical consequence for you: buy sequencing and interpretation separately when you can. The sequencing is a commodity with a stable output format. The interpretation is a snapshot of a database that will be better next year, and you can re-run it yourself against a fresh ClinVar release for the cost of an afternoon.

Questions people also ask

How much does a genome test cost? Research-grade 30x WGS with raw data (FASTQ/CRAM/VCF) delivered: roughly $300 to $1,000. Clinical diagnostic WGS with a CLIA lab report and confirmatory Sanger where needed: $1,500 to $5,000, frequently insurance-billed when there is a clinical indication. Genotyping arrays sold as consumer DNA tests: $50 to $200, and they read fewer than a million pre-chosen sites rather than sequencing your genome.

What does a genome test tell you? Your full sequence of SNVs, indels, and most structural variants, against which a report highlights: pathogenic variants in actionable disease genes, recessive carrier status, pharmacogene diplotypes, and (if you compute them) polygenic scores. It does not tell you anything reliable about most non-coding variation, and it does not measure whether a gene is currently being expressed. That requires RNA or protein measurement.

Is a genome report the same as 23andMe results? No. Array genotyping interrogates a fixed panel of common sites and imputes the rest. It cannot find the novel or private variants that matter most in rare disease, and imputation quality degrades exactly where population reference panels are thin.² WGS reads every base it can align to.

How often should I re-run interpretation? Annually is reasonable. ClinVar classifications change, gnomAD adds samples and refines ancestry-specific frequencies, and new predictors appear. Re-running VEP against your stored VCF costs an hour of compute.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

1. Ewan Birney, Paul Flicek, Damian Keefe, et al. Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature, 2007. https://doi.org/10.1038/nature05874 ↩

2. Masao Nagasaki, Jun Yasuda, Fumiki Katsuoka, et al. Rare variant discovery by deep whole-genome sequencing of 1,070 Japanese individuals. Nature Communications, 2015. https://doi.org/10.1038/ncomms9018 ↩ ↩²

3. Simon T. Bennett, C.L. Barnes, Anthony J. Cox, et al. Toward the $1000 Human Genome. Pharmacogenomics, 2005. https://doi.org/10.1517/14622416.6.4.373 ↩

4. Katharina Schwarze, James Buchanan, Jenny C. Taylor, et al. Are whole-exome and whole-genome sequencing approaches cost-effective? A systematic review of the literature. Genetics in Medicine, 2018. https://doi.org/10.1038/gim.2017.247 ↩