Direct-to-Consumer Genetic Testing Companies: What You Get and What the Data Can't Tell You

Most direct-to-consumer genetic testing companies sell you a SNP genotyping array. The product is not a sequence. You spit in a tube, a lab runs an Illumina chip that interrogates roughly 500,000 to 700,000 pre-chosen positions, and you get back a tab-separated file of rsIDs and genotype calls covering about 0.02% of your genome. That file is fine for ancestry, fine for common-variant lookups, and unreliable for anything rare. Consumer whole-genome sequencing is sold by a smaller set of companies for a few hundred to a couple thousand dollars. It gives you FASTQ/BAM/CRAM and a VCF, which is a different class of artifact. You can re-call it, re-annotate it, and re-analyze it for the rest of your life. If you are technically fluent and want data you can work with, buy sequence. Do not buy chips.

The three products sold under one name

Array genotyping. 23andMe v5 and AncestryDNA v2 both run variants of the Illumina Global Screening Array, roughly 650k probes chosen for ancestry informativeness, common GWAS hits, and a scattering of clinically famous variants. Output is a text file with rsid, chromosome, position, and genotype, on GRCh37 coordinates for 23andMe. Probes are designed for a specific allele. If your variant is not on the chip, it does not exist in your data, and absence looks identical to reference.

Consumer whole-genome sequencing. Nebula, Dante Labs, Sequencing.com and similar vendors sell 30x short-read WGS. Ask before you buy whether you get FASTQ or CRAM, and whether you get a gVCF or only a filtered VCF. Ask what reference build was used, ideally GRCh38 with alt contigs handled correctly. A vendor that only hands you a web dashboard is selling you a report. A report is not data.

Clinician-ordered testing sold direct. Several labs now use a “DTC 2.0” model: you order online, a contracted physician signs off, and a CLIA lab runs a targeted panel or exome, with results returned through a genetic counselor. This hybrid has grown precisely because it sidesteps the regulatory and interpretive problems of pure consumer testing while keeping the online ordering path.¹ It is the right route when you have a specific clinical question, such as a strong family history of a specific cancer.

Where array data breaks

The dominant failure mode is false positives on rare variants. Array calling relies on clustering signal intensities across a population of samples. For a variant present in 1 in 10,000 people, the “rare allele” cluster is empty or nearly so in the training data, and a mediocre probe can land a heterozygous call on someone who is homozygous reference. An analysis by a clinical lab of raw consumer data brought in for confirmation found roughly 40% of the variants flagged as present were false positives on orthogonal clinical testing. This is why third-party “raw data upload” interpretation sites are dangerous: they parse a BRCA1 or MTHFR row out of a file that was never validated for that purpose.

The second failure mode is silent incompleteness. Arrays do not detect:

• copy-number variants and structural variants (no SMN1 copy counting, no CYP2D6 duplications or hybrids)
• short tandem repeat expansions (HTT, FMR1, C9orf72)
• anything not on the probe list, including nearly all novel or family-private variants
• mitochondrial heteroplasmy at any useful resolution
• phase, unless you impute and statistically phase it yourself

Third, regulatory clearance is variant-specific and narrow. FDA authorization for consumer reports covers particular assays for particular variants, and the review history around consumer genetic health risk reporting is a patchwork rather than blanket approval of a platform.² The reported risk variants in a consumer product are a small, chosen subset. Three founder BRCA1/2 variants common in Ashkenazi populations are not a BRCA test, and a negative result there says close to nothing about the thousands of other pathogenic variants in those genes.

Getting real work out of array data anyway

If you already have a 23andMe or Ancestry file, here is the pipeline we would run.

Convert to VCF and sanity-check:

plink2 --23file genome.txt FID IID M \
       --ref-from-fa --fa hs37d5.fa \
       --recode vcf bgz --out me_grch37
bcftools stats me_grch37.vcf.gz | grep -E "TSTV|number of"

Watch the no-call rate. Anything above ~2% missing suggests a poor sample, and clustered missingness on one chromosome usually means a failed probe region rather than biology. Check the X/Y call pattern against your known sex as a sample-swap test.

Lift to GRCh38 so annotations line up with current resources:

CrossMap.py vcf hg19ToHg38.over.chain.gz me_grch37.vcf.gz \
            GRCh38.fa me_grch38.vcf

Then impute. The TOPMed Imputation Server (Eagle for phasing, Minimac4 for imputation) will take your ~650k typed sites to tens of millions of imputed dosages. Filter on the estimated imputation r² per variant, INFO/R2 > 0.8 for anything you intend to interpret, and understand that r² collapses for low-frequency variants and for ancestries underrepresented in the panel. Imputation gives you coverage for polygenic scoring. The coverage does not support clinical variant calling. Never treat an imputed genotype at a pathogenic site as a result.

With imputed dosages you can run polygenic scores properly: pull a scoring file from the PGS Catalog, apply with plink2 --score using the dosage column, and compare your score against the distribution of a reference cohort of matched ancestry. Scores developed in European-ancestry cohorts lose substantial predictive power in other populations, and a raw score with no reference distribution is a number with no units.

What to check the day your WGS lands

Verify the deliverable before the download link expires.

samtools quickcheck -v sample.cram
mosdepth --by 1000 --fast-mode sample sample.cram   # mean depth, coverage uniformity
verifybamid2 --SVDPrefix 1000g.phase3.100k --Reference GRCh38.fa --BamFile sample.cram
bcftools stats sample.vcf.gz | grep -E "TSTV|number of SNPs|number of indels"

Targets for 30x PCR-free WGS start with mean autosomal depth 28-35x and ≥90% of the genome at ≥20x. Contamination (FREEMIX) should be under 0.01, with Ti/Tv near 2.0-2.1 genome-wide. Expect roughly 4.5-5.0 million SNVs and 800k-1M indels for a single sample against GRCh38. Wildly fewer variants usually means the vendor filtered the VCF hard. A Ti/Tv of 1.7 means noise. If the numbers are off, ask for a re-run before you build anything on top of it.

For annotation, we use Ensembl VEP with the --everything set plus ClinVar, gnomAD v4 allele frequencies, and CADD, then restrict to genes with real evidence rather than scanning the whole exome. Short-read WGS still does not solve CYP2D6 structural variation, HLA typing, or repeat expansions without specialized callers (Cyrius, HLA*LA, ExpansionHunter). Run those explicitly or accept the gaps.

What it costs, and whether it is worth it

Array kits run $99-$200. Consumer WGS runs roughly $300-$1,500 depending on depth and whether you get raw files. Clinician-ordered panels billed out of pocket range from a few hundred dollars to several thousand, and coverage varies enough that out-of-pocket exposure changes what tests get ordered at all, including in oncology where the clinical stakes are highest.³ Sequencing costs have fallen far faster than the cost of interpreting, reporting, and following up on sequences, and total spending on genetic testing is driven increasingly by the sheer number of distinct tests available rather than the price of any one.⁴

The downstream cost is the part people underestimate. In a randomized pilot integrating whole-genome sequencing into primary care and cardiology, the sequencing itself was a minority of the short-term cost once you counted clinician time, confirmatory testing, and follow-up.⁵ Formal economic evaluations of genomic sequencing have struggled to demonstrate cost-effectiveness in unselected healthy adults, largely because the yield of actionable findings per person is low and the cascade of follow-up is not.⁶ The value case is stronger for people with a specific indication, and the ethical argument for universal sequencing rests on considerations beyond near-term cost.⁷

The other downsides are not technical. Consumer genetic data sits outside most clinical privacy protections, and the terms governing research use, law-enforcement access, and what happens in an acquisition are set by a contract you clicked through.⁸ Concerns about analytic validity, over-interpretation, and gaps in regulatory oversight have been documented since the first wave of consumer tests and have not been fully resolved.⁹¹⁰ Read the data-sharing terms and the deletion policy before you spit.

Our recommendation: if you want a file you will still be analyzing in ten years, buy 30x WGS with FASTQ or CRAM delivery and run your own calls. If you have a specific family history or a symptom, go through a clinician and a CLIA-certified clinical lab, because a consumer result should never be the basis for a medical decision and any finding that matters needs orthogonal confirmation and a genetic counselor to interpret it.

Questions people also ask

How accurate are DTC genetic tests? For common variants with well-separated genotype clusters, array calls are typically above 99% concordant with sequencing. For rare pathogenic variants, raw-data flags are frequently wrong, with around 40% false positives in one clinical lab’s review of consumer files submitted for confirmation. Accuracy is per-variant. A per-company figure does not capture it.

How much does direct-to-consumer genetic testing typically cost? $99-$200 for an array kit, often less during sales, with subscriptions layered on top for ongoing reports. Consumer 30x whole-genome sequencing runs roughly $300-$1,500. Clinician-ordered panels billed out of pocket generally land between a few hundred and several thousand dollars.⁴

Is it worth getting genetic testing done? For a defined clinical question with a family history behind it, yes, through a clinician. For general curiosity plus a desire for a reusable dataset, whole-genome sequencing is worth more than an array at two to five times the price. Population-wide sequencing of healthy adults has not shown clear cost-effectiveness in formal analyses.⁶

How much does genetic testing for autism typically cost? Diagnostic workups typically use chromosomal microarray and/or exome sequencing ordered by a clinical geneticist, commonly billed in the $1,000-$5,000 range before insurance. Consumer arrays are not a substitute, and this is a clinician-ordered evaluation. The workup is not a self-service one.

What are the downsides of direct-to-consumer genetic testing? The first downsides are incomplete variant coverage that reads as reassurance and false positives on rare variants. The others are no clinician in the loop to contextualize results, and data-governance terms that give the company broad latitude over your genome.⁸

Footnotes

1. Megan A. Allyse, David H. Robinson, Matthew J. Ferber, et al. Direct-to-Consumer Testing 2.0: Emerging Models of Direct-to-Consumer Genetic Testing. Mayo Clinic Proceedings, 2018. https://doi.org/10.1016/j.mayocp.2017.11.001 ↩

2. Sharon Jiang, Lawrence Liberti, David Lebo. Direct-to-Consumer Genetic Testing: A Comprehensive Review. Therapeutic Innovation & Regulatory Science, 2023. https://doi.org/10.1007/s43441-023-00567-5 ↩

3. K. Robin Yabroff, Kewei Sylvia Shi, Jingxuan Zhao, et al. Importance of Patient Health Insurance Coverage and Out-of-Pocket Costs for Genomic Testing in Oncologists’ Treatment Decisions. JCO Oncology Practice, 2024. https://doi.org/10.1200/op.23.00153 ↩

4. Kathryn A. Phillips, Patricia A. Deverka, Gillian W. Hooker, et al. Genetic Test Availability And Spending: Where Are We Now? Where Are We Going?. Health Affairs, 2018. https://doi.org/10.1377/hlthaff.2017.1427 ↩ ↩²

5. Kurt D. Christensen, Jason L. Vassy, Kathryn A. Phillips, et al. Short-term costs of integrating whole-genome sequencing into primary care and cardiology settings: a pilot randomized trial. Genetics in Medicine, 2018. https://doi.org/10.1038/gim.2018.35 ↩

6. Kurt Christensen, Dmitry Dukhovny, Uwe Siebert, et al. Assessing the Costs and Cost-Effectiveness of Genomic Sequencing. Journal of Personalized Medicine, 2015. https://doi.org/10.3390/jpm5040470 ↩ ↩²

7. Leonard M. Fleck, Leslie Francis. Should Whole Genome Sequencing be Publicly Funded for Everyone as a Matter of Healthcare Justice?. Cambridge Quarterly of Healthcare Ethics, 2022. https://doi.org/10.1017/s0963180121000426 ↩

8. Mary A. Majumder, Christi J. Guerrini, Amy L. McGuire. Direct-to-Consumer Genetic Testing: Value and Risk. Annual Review of Medicine, 2021. https://doi.org/10.1146/annurev-med-070119-114727 ↩ ↩²

9. Timothy Caulfield, Amy L. McGuire. Direct-to-Consumer Genetic Testing: Perceptions, Problems, and Policy Responses. Annual Review of Medicine, 2012. https://doi.org/10.1146/annurev-med-062110-123753 ↩

10. J. Kaye. The regulation of direct-to-consumer genetic tests. Human Molecular Genetics, 2008. https://doi.org/10.1093/hmg/ddn253 ↩