Multi-Omics Companies: Who Sells What, and What You Can Get

Oak
multi-omics genomics proteomics bioinformatics vendors
Four different laboratory instruments in a row, each emitting a ribbon of light that merges into one braided line along the floor.

“Multi-omics companies” is a single phrase covering four businesses that barely overlap. Instrument makers (Illumina, Thermo Fisher, Bruker, 10x Genomics, Oxford Nanopore, PacBio) sell the machines and consumables. Assay platforms (Olink, SomaLogic, Alamar, Seer, Metabolon) sell a specific measurement modality with its own units and quirks. Service labs and CROs (Psomagen, Novogene, Azenta, Sapient) run those platforms on your samples and return files. Discovery companies (MultiOmic Health, BostonGene, Regeneron Genetics Center, Tempus) generate or license omics data to build models, and sell the models or the drugs, not the data. If you are an individual who wants your own molecular data in your own hands, only the third layer sells to you directly, and most of it sells to institutions rather than people. That gap is the practical problem.

The four layers, and which one you are buying from

The distinction matters because it determines what you receive. An instrument vendor’s marketing numbers describe what the hardware can do under ideal conditions; the service lab’s actual protocol determines what your sample yields. When a provider says “whole-genome sequencing,” ask whether it is 30x mean coverage short-read on a NovaSeq X or a 4x low-pass array-imputation hybrid. Those differ by roughly an order of magnitude in what they can call, especially for rare variants and indels.

Layer two is where the units get strange. Olink Explore HT measures around 5,400 proteins by proximity extension assay and reports NPX, a normalized log2 relative unit that is not comparable across panels or across plates without bridging samples. SomaScan uses modified aptamers and reports relative fluorescence units for roughly 11,000 reagents, some of which bind targets other than their nominal one. Mass spectrometry platforms (Seer Proteograph into a Bruker timsTOF, or a Thermo Orbitrap Astral) give you spectra and quantities that are absolute in a way the affinity platforms are not, but with fewer proteins per run at depth. None of the three is a drop-in substitute for the others, and correlation between affinity platforms for the same protein is frequently modest. Anyone integrating genomics and proteomics runs into exactly this class of problem: the technologies answer different questions and their outputs resist naive merging.1

Layer four, the discovery companies, is where most of the press lives. These firms are optimizing for cohort-scale inference, and the individual is a data point rather than a customer. That is a reasonable business, but it will not get you a CRAM file.

What good deliverables look like

The single most useful filter is to ask, before paying, for the exact list of files you will receive. A credible genomics deliverable is: paired-end FASTQ (gzip, or better, the original BCL-derived FASTQs), an aligned CRAM against a named reference (GRCh38 with alt contigs, or T2T-CHM13v2.0, and you want to know which), a gVCF plus a filtered VCF, and the QC bundle from whatever caller was used. CRAM 3.1 is reference-dependent, so get the reference FASTA and its .fai too, or your file is an expensive brick in three years. Ask which caller produced the VCF: DRAGEN on Illumina hardware, GATK HaplotypeCaller, or DeepVariant. Each has different indel behavior and different filter fields, and you will want to know before you compare against a public callset.

For RNA sequencing, the questions are library chemistry and depth. Poly-A selection gives you mature mRNA cheaply, ribo-depletion keeps non-coding and degraded transcripts. Twenty to fifty million paired reads per sample is a reasonable working range for differential expression on whole blood or PBMCs, more if you care about low-expressed transcripts or splicing. You want the FASTQs plus a quantification you did not have to trust: salmon quant -l A --validateMappings --seqBias --gcBias against a decoy-aware transcriptome index takes minutes and lets you check the vendor’s numbers against your own. Ask for the transcriptome annotation version (GENCODE v44 and v47 will not give identical gene IDs).

For proteomics, insist on the raw or near-raw layer alongside the normalized table: mzML for mass spectrometry, plate and QC-sample metadata for Olink or SomaScan, and the limit-of-detection flags per assay. Without plate identifiers you cannot correct batch effects later, and batch effects on affinity proteomics are large enough to swamp real longitudinal signal. For continuous glucose monitoring, get the raw interstitial readings at native cadence (Dexcom G7 at five minutes, Abbott Libre 3 at one) rather than a daily summary. Derived metrics like time-in-range and MAGE are trivial to recompute and impossible to un-average.

Where multi-omics integration breaks

Integration failures are rarely about algorithms. They are about identifiers, timing, and missingness. Genes, transcripts, proteins, and metabolites live in different namespaces, and mapping UniProt accessions to Ensembl gene IDs to KEGG compound identifiers is lossy in both directions, with one-to-many relationships everywhere. Pathway databases such as KEGG are the standard bridge for connecting gene, protein, and metabolite layers, and they carry their own curation lag.2 These are long-standing, well-documented problems in the field rather than anything you are doing wrong.3

Timing is the second failure. If your blood draw for proteomics happened at 8am fasted and your RNA sample came from a different draw two weeks later after a night of poor sleep, the strongest axis of variation in your joint model is scheduling. Multi-omic studies that produce interpretable results tend to sample the layers from the same draw, or close to it, and record the covariates. A study of glycemic responders after weight loss found that combining clinical, transcriptomic, and metabolomic features separated phenotypes that any single layer missed, and that work depended on tightly controlled sampling design.4

The third is that most of this remains research-grade. Clinical translation of multi-omics is constrained by standardization, reference ranges, and reimbursement more than by measurement capability.5 A CLIA/CAP-certified lab report and a research-use-only panel are different objects with different legal and interpretive standing. If something in your data looks clinically consequential, the next step is a clinician and a confirmatory test in a certified lab, not a deeper analysis of the same files.

Evaluating a vendor in six questions

Ask these before the sample ships. What is the exact file manifest, including reference build and annotation version? Who holds the sample after the run, and for how long? What are the data-use terms, specifically whether your de-identified data enters a research database by default? What is the turnaround, and what happens on assay failure (free re-run, or re-pay)? Is the assay CLIA-validated or research-use-only? And can you get a raw-tier deliverable, not only a PDF report?

The last one separates providers you can work with from providers you cannot. A PDF is an interpretation you did not make, generated by a pipeline you cannot inspect, against a reference panel you cannot see. If you can run samtools flagstat and bcftools stats on your own files, you can audit everything downstream.

Questions people also ask

What does multi-omics mean?

An “ome” is the complete set of one class of molecule in a sample: the genome (DNA), transcriptome (RNA), proteome (proteins), metabolome (small molecules). Multi-omics means measuring several of these on the same person or sample and analyzing them jointly, on the reasoning that the genome tells you what could happen and the other layers tell you what is happening now.

What are some examples of multi-omics platforms?

At the assay level: Illumina NovaSeq X for DNA and RNA sequencing, Oxford Nanopore PromethION for long reads and native methylation, Olink Explore HT and SomaScan for affinity proteomics, Seer Proteograph with a timsTOF for mass-spec proteomics, and Metabolon or comparable LC-MS untargeted metabolomics. At the analysis level, scGPT and similar transformer models are being trained across single-cell modalities to produce shared embeddings.6

Is AI replacing bioinformatics?

No. It is changing which parts are tedious. Foundation models are useful for embedding and transfer across modalities, and large language models are genuinely good at writing pipeline glue and reading literature.7 The parts that determine whether an analysis is correct (experimental design, batch structure, identifier mapping, knowing which QC metric failed) remain human judgment, and a confidently wrong model output is harder to catch than a crashed job.

Which is the biggest genomics company in the world?

Illumina, by sequencing instrument installed base and revenue, though Thermo Fisher is larger overall as a life-sciences company and Chinese manufacturer MGI/Complete Genomics has taken meaningful share in several markets.

What is omics in simple words?

Measuring everything of one kind at once instead of one thing at a time: all your genes rather than one gene, all your detectable proteins rather than a single blood test.

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. Dr. Seyi Enitan, Esther Adejumo, Dr. Osareniro Osakue, et al. Integrating Genomics and Proteomics Technologies in Biological Research: Advantages, Challenges, and Prospects. Global South Health Horizons, 2025. https://doi.org/10.63950/gshh.2025.1.1.2 ↩

  2. Masaaki Kotera, Mika Hirakawa, Toshiaki Tokimatsu, et al. The KEGG Databases and Tools Facilitating Omics Analysis: Latest Developments Involving Human Diseases and Pharmaceuticals. Methods in Molecular Biology, 2011. https://doi.org/10.1007/978-1-61779-400-1_2 ↩

  3. David Gomez-Cabrero, Imad Abugessaisa, Dieter Maier, et al. Data integration in the era of omics: current and future challenges. BMC Systems Biology, 2014. https://doi.org/10.1186/1752-0509-8-s2-i1 ↩

  4. Armand Valsesia, Anirikh Chakrabarti, Jörg Hager, et al. Integrative phenotyping of glycemic responders upon clinical weight loss using multi-omics. Scientific Reports, 2020. https://doi.org/10.1038/s41598-020-65936-8 ↩

  5. Said I. Ismail, Chadi Saad, Mohamed A. Elrayess, et al. Mapping the path to clinical implementation of multi-omics. Nature Genetics, 2026. https://doi.org/10.1038/s41588-026-02663-2 ↩

  6. Haotian Cui, Chloe Wang, Hassaan Maan, et al. scGPT: toward building a foundation model for single-cell multi-omics using generative AI. Nature Methods, 2024. https://doi.org/10.1038/s41592-024-02201-0 ↩

  7. Yuan Liu, Yamei Chen, Leng Han. Bioinformatics: Advancing biomedical discovery and innovation in the era of big data and artificial intelligence. The Innovation Medicine, 2023. https://doi.org/10.59717/j.xinn-med.2023.100012 ↩