Which Genomic Testing Company Should You Use, and What You Get Back

Oak
genomics sequencing wgs variant-calling testing
Four identical lab instruments in a row, each holding the same glowing DNA strand but illuminating a different amount of it.

There is no single best genomic testing company, and the reason is that the companies appearing in your search results are solving four different problems. Clinical diagnostic labs such as GeneDx, Invitae/Labcorp, Myriad, Blueprint, and Ambry answer a specific medical question and return a PDF report. Direct-to-consumer services such as 23andMe and AncestryDNA genotype a fixed set of common variants on an array and return a tab-delimited file. Sequencing providers such as Broad Clinical Labs, Nebula, Dante, Novogene, and Psomagen sell you coverage and give you FASTQ, BAM/CRAM, and VCF files. Tumor profiling labs such as Foundation Medicine, Tempus, Caris, and Guardant sequence a cancer panel and report somatic findings to an oncologist.

If you are technically fluent and want data you can compute on, the third category is the one you want, and the sensible basis for comparing vendors is the deliverables and the sequencing depth rather than the analysis layer they bolt on top. The sections below walk through what each category provides and what 30x whole genome sequencing gets you in practice. They also cover the pipeline we would run on our own files and the questions to ask a vendor before sending a sample.

The four categories, and what each one will and won’t give you

Knowing which category you are shopping in saves the most time, because the four are not interchangeable and each is genuinely good at its own job.

Clinical diagnostic labs run targeted panels or exomes under CLIA/CAP accreditation, classify variants against ACMG/AMP criteria, and return a signed report. The analytic quality is high, and the interpretation is the product you are buying. For a technical reader the limitation is that raw data is usually not part of the deliverable, and when it is available at all, it arrives only after a records request. These tests also require an ordering clinician in most states, which is a deliberate feature of the market rather than an accident of it: clinicians act as both gatekeepers and translators for commercial genomic results, and the boundary shifts depending on the test and the vendor.1 If your family history points at a specific gene, this is the right category, and your time is better spent with a genetic counselor than with a VCF.

Direct-to-consumer arrays genotype somewhat under a million pre-selected sites. A 23andMe raw file is roughly 600k–700k rows of rsid, chromosome, position, genotype. That serves ancestry work and polygenic scores built on common variants perfectly well, and it serves almost nothing else. An array cannot see a variant it has no probe for, so the absence of a finding tells you nothing at all. Arrays descend directly from the hybridization approach that first made genome-scale measurement cheap, and they inherit that approach’s central constraint, which is that you measure what you designed the probes for.2 The commercial DTC model has also always blended consumer product with research resource in ways that make the consent form worth reading closely.3

Sequencing providers are the category we recommend for a baseline. You pay for depth and you receive files. Before you pay, ask for three things in writing:

  1. FASTQ or unaligned BAM, rather than only the aligned product.
  2. The reference build used.
  3. Per-sample QC metrics.

Tumor profiling sits apart from the other three. These tests sequence tumor tissue or circulating tumor DNA against a panel of a few hundred genes and report somatic alterations, tumor mutational burden, and microsatellite instability. They are ordered by an oncologist and interpreted by an oncologist, and they do not substitute for germline testing. A somatic report is not something to self-interpret.

What 30x whole genome sequencing delivers

Since 30x whole genome sequencing is the usual entry point for people who want computable data, it helps to know exactly what arrives and what it can support. “30x WGS” means mean coverage of about 30 reads per base across the genome, typically 150 bp paired-end reads on an Illumina NovaSeq X or a comparable instrument. The files you should expect are these:

  • FASTQ: ~90–120 GB gzipped for a 30x human genome, split R1/R2 across lanes
  • BAM aligned to GRCh38: ~80–100 GB
  • CRAM, reference-compressed: ~15–20 GB, which is what you should store
  • gVCF/VCF: roughly 4–5 million SNVs and indels per genome versus GRCh38, of which ~3–4 million are common

Thirty-fold coverage is the right default for germline SNV and indel calling. It falls short of what confident somatic calling requires, where you want 100x or more on the tumor with a matched normal. Short reads also remain structurally weak across repeats, segmental duplications, and large structural variants. Long reads from PacBio HiFi or Oxford Nanopore resolve those regions and can be combined with short-read or linked-read data in hybrid assemblies to improve contiguity.4 Every sequencing generation has traded read length, accuracy, and cost against one another. No platform yet wins on all three.56

The reference build matters more than most vendors acknowledge. A VCF called against GRCh37 and a VCF called against GRCh38 use different coordinates, and annotating one with tools configured for the other produces garbage without raising an error. Insist on GRCh38 with alt contigs handled explicitly. If you intend to work against T2T-CHM13, get FASTQ so that you can realign.

The pipeline we would run on our own data

Our own practice is to start from FASTQ, because re-running alignment and calling yourself is the only way to know what you have. The commands below cover alignment, variant calling, and a first pass at quality control.

# alignment
bwa-mem2 mem -t 32 -R '@RG\tID:s1\tSM:sample\tPL:ILLUMINA' \
  GRCh38_full_analysis_set_plus_decoy_hla.fa R1.fq.gz R2.fq.gz \
| samtools sort -@ 8 -o sample.bam -
samtools index sample.bam

# variant calling
run_deepvariant --model_type=WGS --ref=GRCh38...fa \
  --reads=sample.bam --output_vcf=sample.vcf.gz --num_shards=32

# QC
mosdepth -n --fast-mode -t 4 sample sample.bam
bcftools stats sample.vcf.gz > sample.stats

We prefer DeepVariant to GATK HaplotypeCaller for a single sample, because it gives better indel precision out of the box and requires far less parameter fiddling. GATK is the better choice if you are joint-calling a family. With variants in hand, the next step is annotation:

vep -i sample.vcf.gz --cache --assembly GRCh38 --everything \
  --plugin CADD,whole_genome_SNVs.tsv.gz --vcf -o sample.vep.vcf.gz

Before trusting any of the output, we check a set of QC gates. We want mean coverage of at least 28x and at least 95% of the callable genome at 20x or better. We also want a transition/transversion ratio around 2.0–2.1 genome-wide and a heterozygous/homozygous ratio near 1.5–2.0. Finally, we want a contamination estimate from VerifyBamID FREEMIX under 0.02 and a duplication rate under roughly 15%. A Ti/Tv of 1.7 tells you your filtering is too loose. A high FREEMIX indicates sample contamination, which makes everything downstream suspect.

For the interpretive layer, restrict yourself at first to variants carrying a ClinVar assertion of pathogenic or likely pathogenic with at least two-star review status, intersected with the ACMG SF v3.2 secondary-findings gene list. That yields a list of tens of variants, most of which will be false positives generated by your own pipeline. Treat a pathogenic call from a research-grade pipeline as a hypothesis rather than a result: confirm it in a CLIA lab and take it to a genetic counselor before it changes anything.

How to evaluate a vendor

The following questions are the ones we ask before sending a sample. Vendors that dodge them are selling a report rather than data.

  1. What deliverables, in what formats, and for how long are they hosted? FASTQ plus CRAM plus gVCF, all downloadable, is the answer you want.
  2. What instrument and chemistry, what read length, and what mean coverage? And what is the guaranteed floor if the run underperforms?
  3. What reference build and pipeline version, and do you receive the QC report with mosdepth/Picard metrics and a contamination estimate?
  4. Is the lab CLIA-certified and CAP-accredited? Research-grade sequencing is cheaper and perfectly good for exploration, but it cannot be used clinically without confirmation.
  5. Who else gets the data? Read the secondary-use and de-identification terms.
  6. Saliva or blood? Blood yields higher-molecular-weight DNA and less bacterial contamination. Saliva kits often come back with 10–30% non-human reads, which eats into your effective coverage.

Cost is the last piece of the picture, and it has fallen by orders of magnitude since the Human Genome Project, driven by sustained instrument and chemistry development rather than any single breakthrough.7 A 30x consumer WGS now runs a few hundred dollars from research-grade providers. Clinical exomes and panels run from roughly one to several thousand dollars, and tumor profiling panels are typically billed in the low thousands and usually go through insurance. The sober framing from the late 1990s still holds, in that sequence is a starting point and its predictive value for most common conditions is modest.8 What has changed is who can obtain the data and what they can do with it afterward.9

Questions people also ask

Who are 10x Genomics’ competitors? 10x Genomics sells single-cell and spatial instruments rather than consumer genomic testing. Its direct competitors in droplet-based single-cell RNA-seq include BD Rhapsody, Fluent BioSciences, and Singleron. Parse Biosciences uses combinatorial barcoding and needs no instrument, and Scale Biosciences is also in this group. Head-to-head comparisons show platform-specific differences in cell recovery, gene detection per cell, and ambient RNA. All of these matter for study design.10

What is the best company for genetic testing? It depends on the question you are asking. For a diagnostic question with a family history behind it, the right choice is a clinical lab such as GeneDx or Invitae, ordered through a clinician. For ancestry or common-variant curiosity, a DTC array will do. For a reusable molecular baseline you intend to compute on, choose a sequencing provider that returns FASTQ from 30x WGS.

How much does genomic testing cost? Consumer arrays are typically under $200. Research-grade 30x WGS with raw data included sits in the high hundreds. Clinical panels and exomes range from about $1,000 to several thousand, and are often covered by insurance when medically indicated. Tumor profiling panels are billed in the low thousands.

Is it worth getting genetic testing for cancer? That is a question for a genetic counselor, and the answer turns on your personal and family history rather than on any general rule. Germline hereditary cancer testing and somatic tumor profiling answer different questions and are ordered in different situations. We will not give you an answer here, because the right one depends on clinical facts we do not have.

What does “raw data” mean and should I insist on it? Yes, insist on it. Raw means FASTQ or unaligned BAM, which is to say the reads as they came off the instrument, before anyone’s pipeline touched them. With FASTQ you can realign to any reference, re-call with any caller, and re-annotate as ClinVar updates. With only a PDF you own nothing, and you have to pay again every time you want to ask a new question.

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. Michelle L. McGowan, Jennifer R. Fishman, Richard A. Settersten, et al. Gatekeepers or Intermediaries? The Role of Clinicians in Commercial Genomic Testing. PLoS ONE, 2014. https://doi.org/10.1371/journal.pone.0108484 ↩

  2. Patrick O. Brown, David Botstein. Exploring the new world of the genome with DNA microarrays. Nature Genetics, 1999. https://doi.org/10.1038/4462 ↩

  3. Margaret Curnutte, Giuseppe Testa. Consuming genomes: scientific and social innovation in direct-to-consumer genetic testing. New Genetics and Society, 2012. https://doi.org/10.1080/14636778.2012.662032 ↩

  4. Zhanshan (Sam) Ma, Lianwei Li, Chengxi Ye, et al. Hybrid assembly of ultra-long Nanopore reads augmented with 10x-Genomics contigs: Demonstrated with a human genome. Genomics, 2019. https://doi.org/10.1016/j.ygeno.2018.12.013 ↩

  5. Erik Pettersson, Joakim Lundeberg, Afshin Ahmadian. Generations of sequencing technologies. Genomics, 2009. https://doi.org/10.1016/j.ygeno.2008.10.003 ↩

  6. Jay Shendure, Shankar Balasubramanian, George M. Church, et al. DNA sequencing at 40: past, present and future. Nature, 2017. https://doi.org/10.1038/nature24286 ↩

  7. Jeffery A. Schloss, Richard A. Gibbs, Vinod B. Makhijani, et al. Cultivating DNA Sequencing Technology After the Human Genome Project. Annual Review of Genomics and Human Genetics, 2020. https://doi.org/10.1146/annurev-genom-111919-082433 ↩

  8. R L Zimmern. The human genome project: a false dawn?. BMJ, 1999. https://doi.org/10.1136/bmj.319.7220.1282 ↩

  9. Diana Crow. A New Wave of Genomics for All. Cell, 2019. https://doi.org/10.1016/j.cell.2019.02.041 ↩

  10. A. V. Maluchenko, E. S. Avsievich, Ia. S. Zvorygina, et al. Leveraging two novel droplet−based single−cell RNA Sequencing platforms: a comparative study with 10x genomics. BMC Genomics, 2025. https://doi.org/10.1186/s12864-025-12355-6 ↩