Can You Get Your Genome Sequenced Without a Doctor's Order?

In most US states you can order sequencing on yourself without a physician referral. Some services sell directly to you under an FDA authorization for a fixed list of reports. Others route the order through a telehealth physician network, so a doctor signs off but you never have to find one. New York has historically been the strict outlier, and a few labs will not ship kits there. The part that matters for you is not the ordering path but the deliverable: a service that returns a PDF of 40 “wellness” traits is a different product from one that returns a CRAM file and a gVCF you can reprocess. Ask that question before you pay.

The three tiers, and why only one of them is a genome

Genotyping arrays (23andMe, AncestryDNA) assay a fixed set of roughly 600,000 to 700,000 probes. They are cheap, reproducible at common variants, and fine for ancestry and for polygenic scores built on common SNPs. They are a poor instrument for rare variants. A probe is designed for a specific allele, the call is a hybridization intensity cluster, and when a variant is present in one person per 50,000 the cluster has almost no data behind it. Raw array data promoted to a “pathogenic BRCA1 variant” is a hypothesis, not a result, and needs orthogonal confirmation in a clinical lab before anyone acts on it. Arrays also miss everything not on the chip: novel variants, most indels, structural variation.

Exome sequencing covers roughly 1 to 2 percent of the genome, the protein-coding portion, typically at 100x mean on-target. It captures most of what current clinical interpretation can use, at the cost of uneven capture (GC-rich first exons drop out) and blindness to deep intronic and regulatory variants.

Whole-genome sequencing at 30x mean coverage, 150 bp paired-end, is what we would buy. Costs fell by orders of magnitude over the past two decades as short-read platforms matured, which is why a personal genome is a consumer purchase at all now¹. Coverage is more uniform than exome capture, and indel and structural variant calling is better. Mitochondrial DNA comes for free at high depth. You can also re-call variants against a new reference or a new caller in five years without re-collecting a sample.

Insist on these files

A personal genome service that will not give you primary data is selling you a report. What to require, in order of how much you should care:

• FASTQ, gzipped, R1 and R2. Roughly 100 GB for a 30x human genome. This is the only artifact that lets you redo alignment from scratch.
• CRAM aligned to a named reference build. GRCh38 with alt contigs, or better, the T2T-CHM13v2.0 assembly as a second alignment. A 30x CRAM is about 15 to 20 GB against roughly 55 GB for the equivalent BAM. Ask which reference, explicitly, including whether ALT contigs and decoys were used.
• gVCF. A filtered VCF on its own is not enough. The gVCF has reference-confidence blocks, so you can distinguish “no variant here” from “no coverage here”. Without it you cannot tell a homozygous-reference site from a dropout.
• The run QC: mean and callable coverage, percent of bases at ≥20x, duplicate rate and insert size distribution. It also includes a contamination estimate (VerifyBamID FREEMIX, want < 0.01).

A 30x genome yields roughly 4 to 5 million variants against GRCh38, of which about 4 to 5 percent are coding and a few tens of thousands are rare protein-altering.

The pipeline we would run

If the lab used DRAGEN, keep their output as the primary call set. It is fast and well-validated. Then build a second, independent call set so you can see where the callers disagree, because disagreement is where the errors live.

# align
bwa-mem2 mem -t 32 -K 100000000 -Y \
  -R '@RG\tID:S1\tSM:S1\tPL:ILLUMINA\tLB:lib1' \
  GRCh38_full_analysis_set_plus_decoy_hla.fa R1.fq.gz R2.fq.gz \
| samtools sort -@ 8 -m 4G -o s1.bam -

samtools markdup -@ 8 --reference GRCh38.fa s1.bam s1.md.cram

# small variants
run_deepvariant --model_type=WGS --num_shards=32 \
  --ref=GRCh38.fa --reads=s1.md.cram \
  --output_vcf=s1.dv.vcf.gz --output_gvcf=s1.dv.g.vcf.gz

# structural variants
manta configManta.py --bam s1.md.cram --referenceFasta GRCh38.fa --runDir mt
smoove call --name S1 --fasta GRCh38.fa --processes 16 --genotype s1.md.cram

DeepVariant is our default small-variant caller for a single sample. GATK HaplotypeCaller is the right choice if you are joint-calling a family, because the joint genotyping step in GenotypeGVCFs is where it earns its keep.

Annotate with VEP, pinned to a cache version you record:

vep -i s1.dv.vcf.gz --cache --offline --assembly GRCh38 --everything \
  --plugin CADD,whole_genome_SNVs.tsv.gz \
  --custom clinvar.vcf.gz,ClinVar,vcf,exact,0,CLNSIG,CLNREVSTAT \
  --custom gnomad.genomes.v4.1.sites.vcf.bgz,gnomAD,vcf,exact,0,AF_grpmax \
  --vcf -o s1.annot.vcf.gz --compress_output bgzip

Then filter. Here is a first pass we would run on ourselves. Restrict to the ACMG SF v3.2 secondary-findings gene list. Keep ClinVar pathogenic or likely pathogenic with a review status of at least two stars. Drop anything with gnomAD popmax AF above 0.001, and read every remaining line manually in IGV. Expect somewhere between zero and a handful of candidates. Every one of them needs orthogonal confirmation in a CLIA lab.

For pharmacogenomics, run PharmCAT on the VCF, and call CYP2D6 separately with Cyrius or Aldy, because CYP2D6 has a pseudogene neighbor and common hybrid alleles that a standard short-read VCF will not resolve. Treat the output as a research annotation and take it to a clinician or pharmacist before it touches any prescribing decision.

What a genome in a healthy person tells you

This is where most personal genome services oversell. Penetrance estimates for most pathogenic variants come from families ascertained because they were sick. Applied to an unselected person, the same variant carries a substantially lower probability of disease, and for many genes we do not know the number. Sequencing healthy people reliably produces findings whose clinical meaning is genuinely unresolved, plus a long tail of variants of uncertain significance that no amount of reprocessing will resolve². European professional societies have argued for exactly this reason that genome-wide sequencing outside a targeted clinical question should be approached with a defined analysis plan rather than an open-ended look at everything³.

The interpretation burden is the hard part. The sequencing is not where the difficulty lies. The first efforts to turn an individual whole genome into clinically usable statements found that the bottleneck was curating variant-disease evidence, integrating family history, and handling pharmacogenomic haplotypes, all of which required expert human time per case⁴. That has improved, but not to the point where a pipeline output is a conclusion.

Where you need a clinician

Three situations, plainly. If a candidate pathogenic variant survives your filtering in a gene tied to cancer risk, cardiomyopathy, arrhythmia, or aortopathy, you need a clinical-grade confirmation and a genetic counselor before anything else. If you are considering a reproductive decision, carrier status from a research-grade VCF is not adequate on its own. If a pharmacogenomic call would change a medication, that conversation belongs with your prescriber, and we will not make a recommendation here.

Questions people also ask

Can you get genetic testing without a doctor’s order? In most US states, yes. Some labs sell direct to consumers under FDA authorization for specified reports, others attach a telehealth physician who signs the requisition. State rules vary and New York is the most restrictive, so check before ordering.

Can I get genetic testing on my own? You can order it, receive the raw data, and analyze it yourself. What you cannot do on your own is convert a variant call into a clinical conclusion. Research-grade output is not a diagnostic result, and confirmation happens in a CLIA-certified lab.

Is 30x enough, or should I pay for 60x? Thirty-x mean is the standard for germline SNV and indel calling and is sufficient for almost everything you will do. Go higher if you care about mosaicism or somatic signals. Long reads are the better second purchase, since they resolve structural variants and repeat expansions that short reads cannot.

Do I need to re-sequence as science improves? No. Keep the CRAM and FASTQ. Re-annotation against a current ClinVar and gnomAD release, once a year, captures most of the gain at near-zero cost.

Footnotes

1. Sang Tae Park, Jayoung Kim. Trends in Next-Generation Sequencing and a New Era for Whole Genome Sequencing. International Neurourology Journal, 2016. https://doi.org/10.5213/inj.1632742.371 ↩

2. Noralane M. Lindor, Stephen N. Thibodeau, Wylie Burke. Whole-Genome Sequencing in Healthy People. Mayo Clinic Proceedings, 2017. https://doi.org/10.1016/j.mayocp.2016.10.019 ↩

3. Carla van El, Martina C. Cornel, Pascal Borry, et al. Whole-genome sequencing in health care. European Journal of Human Genetics, 2013. https://doi.org/10.1038/ejhg.2013.46 ↩

4. Kelly E. Ormond, Matthew T. Wheeler, Louanne Hudgins, et al. Challenges in the clinical application of whole-genome sequencing. The Lancet, 2010. https://doi.org/10.1016/s0140-6736(10)60599-5 ↩