Skip to content

Genome Mapping Companies in the USA, and What Each One Gives You

Oak
A feathered reptilian creature posed on a display stand, its wing feathers marked with glowing blue points, lit against black.

If you are searching for genome mapping companies in the USA, you are probably looking at three different industries that share a word. Optical genome mapping is a specific technology sold in the US mainly by Bionano Genomics, which images long DNA molecules labeled at a sequence motif to find large structural changes. Whole-genome sequencing services are sold by Illumina and its certified service providers. Clinical labs such as Ambry, Invitae, GeneDx, Baylor Genetics and the Broad Clinical Labs sell them too. So do consumer-facing resellers such as Nebula Genomics and Dante Labs. Genotyping arrays, sold by 23andMe and AncestryDNA, read a few hundred thousand preselected sites and are not a genome at all. Which one you want depends entirely on the variant classes you care about and the files you expect to receive.

Why “mapping” and “sequencing” are different words

The distinction is historical and still technically meaningful. In the pre-1995 framing of the Human Genome Project, mapping meant placing markers along chromosomes to build a scaffold, and sequencing meant reading the bases between them.1 Maps came first because sequencing was too slow to be applied genome-wide without a coordinate system to assemble against.

That division shaped the first commercial genomics sector. Companies formed around marker maps, clone libraries, positional cloning and database subscriptions. The subsector grew sharply through the 1990s before contracting after 2000.2 The SNP Consortium is the clearest example of mapping as an industrial product. A group of pharmaceutical companies and the Wellcome Trust funded the discovery and public release of a dense single-nucleotide polymorphism map, explicitly to give everyone a shared coordinate framework rather than a proprietary one.3

Sequencing then became cheap enough to absorb most of mapping’s function. Sustained instrument development after the reference genome drove per-base costs down by orders of magnitude and made short-read resequencing the default assay.4 So when a modern company advertises genome mapping, it is usually selling one of two things: short-read sequencing aligned to a reference, or optical mapping to find the structural variation that short reads miss.

Optical genome mapping

Optical genome mapping (OGM) is what Bionano sells, and it is the only company with meaningful US commercial presence in this niche. Ultra-high-molecular-weight DNA is extracted, labeled enzymatically at a recurring six-base motif, then linearized in nanochannels and imaged. The output is a set of label distance patterns along single molecules hundreds of kilobases long, compared against an in-silico reference map.

What you get is a structural picture: large deletions, duplications, inversions, translocations and repeat expansions, typically down to a few hundred base pairs to a few kilobases in size. What you do not get is base-level sequence. There is no variant call for a single-nucleotide change, no pharmacogenomic star allele, no coding-variant annotation. OGM is ordered in clinical contexts by a physician, usually in hematologic malignancy workups or constitutional cases where a karyotype and a microarray were uninformative. If that describes your situation, this is a conversation with a clinical geneticist and not a direct purchase.

For an individual building a personal baseline, long-read sequencing is the better use of the same budget. PacBio HiFi reads run roughly 15 to 25 kb at high per-base accuracy, and Oxford Nanopore reads can run far longer with the benefit of direct methylation calling. Both give you structural variants and base-level sequence in one dataset. The tradeoff is cost per genome and, for nanopore, a higher raw error rate that matters for small indels in homopolymers.

Short-read whole-genome sequencing providers

This is the category most readers should be in. A standard germline order is 30x mean coverage with 2x150 bp paired-end reads on an Illumina NovaSeq X or similar. In practice, ask providers for the deliverable list before you pay, because the price differences across US vendors are mostly differences in what leaves the lab.

The files you want, in order of importance:

  • Raw reads as gzipped FASTQ, or a CRAM that retains original base qualities. Roughly 80 to 100 GB of gzipped FASTQ at 30x, or 15 to 20 GB as CRAM against GRCh38.
  • An index alongside every alignment and variant file (.crai, .bai, .tbi). Vendors omit these more often than you would expect.
  • A single-sample VCF plus a gVCF. The gVCF records reference confidence at non-variant positions, which is what lets you distinguish “reference” from “not sequenced well enough to say”.
  • The exact reference build and analysis version, stated as a string you can reproduce. GRCh38 with alt contigs behaves differently from GRCh38 no-alt, and hg19-era pipelines are still in circulation.
  • Per-base or per-region coverage, ideally a mosdepth output, so you can see which genes fell below callable depth.

A 30x human genome yields on the order of 4 to 5 million small variants against GRCh38, most of them common and previously catalogued. Clinical labs return an interpreted report on a few dozen curated genes and often withhold the raw data. Consumer resellers return the raw data and a thin interpretation layer. We prefer the second, because the interpretation layer ages badly and the FASTQ does not.

The pipeline we would run on your own files

Start by verifying the data rather than trusting the report. Run samtools quickcheck and samtools flagstat on the alignment, then mosdepth --by 500 to get coverage distribution. Check the fraction of the callable genome above 10x. Anything below about 95% at 10x on a 30x order suggests a library or a coverage-uniformity problem worth raising with the vendor.

If you are realigning, use bwa-mem2 mem -K 100000000 -Y against GRCh38 no-alt, then mark duplicates with samblaster or gatk MarkDuplicates. For calling, DeepVariant is our default for germline short reads because it needs no recalibration step and handles indels well. GATK HaplotypeCaller in gVCF mode remains the right choice if you need joint genotyping across family members later. Normalize with bcftools norm -m -both -f GRCh38.fa before you annotate anything, because comparing unnormalized indel representations will waste an afternoon.

Structural variants from short reads need a separate pass. Run Manta and Delly, then intersect. Expect low sensitivity for insertions and for anything inside segmental duplications. Repeat expansions need ExpansionHunter with a locus catalog. Annotate with Ensembl VEP, adding gnomAD allele frequencies and ClinVar, and treat any single-submitter ClinVar assertion as unresolved. Interpretation of a specific variant in a disease gene belongs with a genetic counselor or a medical geneticist, ideally with confirmation on an orthogonal assay in a CLIA-certified lab.

Questions people also ask

Is Bionano the only optical genome mapping company in the US? It is the only one with a commercial platform at scale for human genomes. Competing approaches to structural variation are now mostly long-read sequencing rather than imaging-based mapping.

Do 23andMe or AncestryDNA map my genome? No. Both use genotyping arrays that measure a few hundred thousand preselected sites, which is well under a tenth of a percent of your 3.1 billion bases. Arrays are a direct descendant of the public SNP maps built in the early 2000s.3 They cannot find a variant that was not designed onto the chip.

Should I buy from a clinical lab or a consumer sequencing company? If you have a specific clinical question or a family history, order through a clinician at a CLIA-certified lab, because you need a report that other clinicians will accept. If you want a reusable dataset to analyze yourself, buy from a provider that contractually delivers FASTQ or CRAM plus a gVCF.

How much coverage do I need? Thirty-fold mean is the workable floor for germline small variants. Go to 60x if you care about mosaicism or somatic signals, and add long reads if structural variation is the question.

Why does my VCF differ between two providers on the same DNA? Reference build, aligner, caller and filtering all shift the call set. Concordance on high-confidence single-nucleotide variants is typically above 99%, while indel and structural variant concordance is much lower.

Footnotes

  1. P.S Harper. Mapping and sequencing the human genome. Endeavour, 1989. https://doi.org/10.1016/0160-9327(89)90035-5 ↩

  2. Ilse R Wiechers, Noah C Perin, Robert Cook-Deegan. The emergence of commercial genomics: analysis of the rise of a biotechnology subsector during the Human Genome Project, 1990 to 2004. Genome Medicine, 2013. https://doi.org/10.1186/gm487 ↩

  3. Arthur L. Holden. The SNP Consortium: Summary of a Private Consortium Effort to Develop an Applied Map of the Human Genome. BioTechniques, 2002. https://doi.org/10.2144/jun0203 ↩ ↩2

  4. 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 ↩

Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.