Which Multi-Omics Companies Are Worth Using, and For What

There is no single best multi-omics company, because “multi-omics” names four different businesses that are usually sold under one word. The market splits into the groups below.

• instrument manufacturers (Illumina, Oxford Nanopore, PacBio, Bruker, and Thermo Fisher)
• assay platform owners (Olink, SomaLogic, and Metabolon, plus Nightingale, Seer, and 10x Genomics)
• service labs that run those assays on your samples (Novogene, Azenta, Psomagen, Sapient, and Discovery Life Sciences)
• analysis software (DNAnexus, Velsera, Terra, Lifebit, and QIAGEN Digital Insights)

If you want to work with your own molecular data, the decision that matters most is whether the vendor returns raw files in standard formats, and whether their assay version will still be comparable to itself in three years. The brand on the instrument does not matter as much.

The four layers of the market

Instrument manufacturers set the ceiling on data quality and almost never sell to individuals. Their choices still reach you: an Illumina NovaSeq X run gives you 2×150 bp paired-end reads with a well-characterized error profile, while an Oxford Nanopore PromethION run gives you 10–100 kb reads with base modification calls included in the same pass.

Assay platform owners define the measurement itself. Olink’s Proximity Extension Assay and SomaLogic’s aptamer array both claim thousands of plasma proteins, and they do not measure the same thing. Service labs are the layer you can usually buy from directly. Software vendors sell workflow orchestration and interpretation on top of everything else.

Rank vendors within a layer, never across layers. A question like “is 10x better than Metabolon” has no answer, because one sells single-cell library chemistry and the other sells untargeted plasma metabolomics.

Sequencing: what to demand back

For whole-genome sequencing, we would buy 30× PCR-free short-read WGS from a service lab running NovaSeq X, and we would insist on the raw files before paying. The deliverable you want is gzipped FASTQ plus an aligned CRAM against a named reference build, ideally GRCh38 with alt contigs handled explicitly, plus a gVCF. A 30× human genome is roughly 90 gigabases of sequence, which lands near 50–60 GB of gzipped FASTQ and about 20 GB as a reference-compressed CRAM.

Check three numbers in the QC report before anything else: mean coverage, the fraction of the genome at ≥20× (should be above 90% for a clean 30× run), and duplicate rate. Run your own check with samtools stats and mosdepth --by 1000. If the lab used Illumina DRAGEN for alignment and variant calling, note the version, because DRAGEN 3.x and 4.x calls differ enough to break naive longitudinal comparison. Re-running nf-core/sarek yourself with BWA-MEM2 and GATK HaplotypeCaller gives you a pipeline you control.

Long-read sequencing is the better buy if you care about structural variants, phasing, or methylation. PacBio Revio HiFi reads support small-variant calling that is competitive with short reads, and both PacBio and Nanopore emit 5-methylcytosine calls as MM and ML tags inside the BAM. You can extract per-CpG methylation with modkit pileup and get a bedMethyl file without a separate bisulfite assay.

For RNA sequencing from blood, the sample tube matters more than the sequencer. PAXgene or Tempus tubes stabilize transcripts at draw. Ask whether the lab performs globin depletion, since hemoglobin transcripts can dominate whole-blood mRNA and waste most of your reads. Request the FASTQs and a transcript-level quantification from Salmon or kallisto, delivered as quant.sf with both TPM and estimated counts.

Proteomics and metabolomics: the choice that constrains everything after

Plasma proteomics is where vendor choice locks you in hardest. Olink Explore HT reports roughly 5,400 protein assays as NPX, a log2 relative unit. SomaScan reports around 11,000 aptamer measurements as relative fluorescence units. Neither gives absolute concentration in ng/mL without separate calibration, and cross-platform agreement for the same named protein is often weak. Many protein pairs correlate below a Spearman of 0.5 across the two technologies.

Our view is to pick one platform and stay on it for every timepoint. Longitudinal signal within a platform is far more interpretable than a single cross-sectional snapshot from the “better” platform. Ask the vendor which panel version and which normalization release produced your data, and keep that string with the file. Mass spectrometry options such as Seer’s Proteograph or a Biognosys DIA service give you mzML files and identification tables, which are more portable but less consistent across labs.

Metabolomics has the same split. Metabolon’s untargeted platform returns roughly a thousand named metabolites as scaled, batch-normalized intensities. Nightingale’s NMR panel returns about 250 measures in absolute units, heavily weighted toward lipoproteins and a handful of amino acids and ketones. Nightingale is cheaper, more reproducible over years, and narrower. Metabolon is broader and harder to compare across runs. For a longitudinal personal baseline we would start with NMR and add untargeted work when a specific question justifies the batch-effect risk.

Public resources help you interpret what comes back. ProteomicsDB aggregates protein expression across human tissues and cell lines alongside other omics layers, which is a reasonable first stop when a protein identifier means nothing to you.¹

Integration software, and why we run it locally

The commercial integration platforms (DNAnexus, Velsera, Lifebit) exist mostly because large institutions need audit trails and access control across many datasets. Collaborative platform building of this kind has been the pattern in translational research for over a decade.² For one person’s data, the same work fits on a workstation with 64 GB of RAM and a few terabytes of disk.

We would use Nextflow with nf-core pipelines under Docker or Singularity, pinning revisions with -r 3.14.0 so that a re-run next year produces the same output. Store everything in open formats. Use CRAM for alignments and VCF or a Hail MatrixTable for variants. Use Parquet or a tidy CSV for proteomics and metabolomics. Keep one metadata table keyed by sample ID and draw timestamp. The recurring failures in multi-omics work are inconsistent identifiers, missing assay versions, and batch effects that nobody recorded. The failures are not algorithmic.³

Harmonization is the other half. Consortium efforts in dementia biomarker development have shown how much of the effort goes into standardizing sample handling and reporting before any modeling starts.⁴ Apply the same discipline to yourself. Use the same draw time of day and the same fasting state, freeze aliquots at −80 °C, and write a protocol you follow each time.

On integration methods, the useful result is that combining layers predicts biological state better than any one layer alone. Multi-omics integration in E. coli accurately predicted cellular state in conditions the model had never seen.⁵ In cancer, network-based integration across expression, methylation, and proteomics has surfaced candidate drivers that single-layer analysis missed.⁶ Machine learning methods built for these data are now well reviewed, and their main constraint is sample size rather than model capacity.⁷⁸ Foundation models such as scGPT extend this to single-cell multi-omics, though they are trained for cell-level tasks rather than personal interpretation.⁹

A ten-minute vendor evaluation

Ask these before you send a sample:

• What raw formats do I receive, and for how long are they hosted? FASTQ or CRAM, mzML, and per-plate NPX or RFU tables with QC flags.
• What reference build and pipeline version produced the processed files?
• What is the assay version and normalization release, in a string I can store?
• Is the lab CLIA-certified and CAP-accredited, and is this a clinical or research-use-only report?
• Can I get the unnormalized data alongside the normalized data?

If a vendor returns only a PDF summary and a web dashboard, treat the purchase as a report rather than data. You cannot reanalyze a dashboard.

Everything here is measurement and interpretation. Molecular data does not diagnose anything on its own. A variant call, a protein shift, or an abnormal blood marker needs a clinician to place it in the context of your history and examination. The clinical genetics literature has been direct that interpreting these datasets is a shared task between analyst and physician.¹⁰

Questions people also ask

Which companies are involved in multi-omics research? Illumina, Oxford Nanopore, PacBio, Thermo Fisher, and Bruker build instruments. Olink, SomaLogic, and Metabolon own assay platforms, as do Nightingale, Seer, and 10x Genomics. Novogene, Azenta, Psomagen, Sapient, and Discovery Life Sciences run those assays as a service. DNAnexus, Velsera, Lifebit, and QIAGEN Digital Insights sell the analysis layer.

Why is Illumina sequencing preferred over Sanger sequencing? Sanger reads one amplicon at a time at roughly 500–1,000 bases per reaction, which is excellent for confirming a single known variant. Illumina sequencing-by-synthesis runs billions of short reads in parallel, which is the only practical way to cover three gigabases at 30×. Sanger remains the standard for orthogonal confirmation of a single finding.

What are examples of multi-omics platforms? At the assay level, there is Olink Explore HT for proteins. For metabolites there is Metabolon HD4 or Nightingale NMR, and for genomes and transcriptomes there is Illumina NovaSeq X. At the software level, Terra, DNAnexus, and self-hosted Nextflow with nf-core pipelines.

Should I pick a vendor based on breadth of analytes? No. Pick on data return policy and version stability first, then breadth. A 5,000-protein panel you can re-run identically for five years beats an 11,000-protein panel that changes normalization between orders.

Footnotes

1. Patroklos Samaras, Tobias Schmidt, Martin Frejno, et al. ProteomicsDB: a multi-omics and multi-organism resource for life science research. Nucleic Acids Research, 2019. https://doi.org/10.1093/nar/gkz974 ↩

2. Axel Schumacher, Tamas Rujan, Jens Hoefkens. A collaborative approach to develop a multi-omics data analytics platform for translational research. Applied & Translational Genomics, 2014. https://doi.org/10.1016/j.atg.2014.09.010 ↩

3. Christopher L. Hemme, Janet Atoyan, Ang Cai, et al. Challenges and Opportunities in Multi-Omics Data Acquisition and Analysis: Toward Integrative Solutions. Biomolecules, 2026. https://doi.org/10.3390/biom16020271 ↩

4. Ekaterina Mavrina, Leighann Kimble, Katharina Waury, et al. Multi-Omics Interdisciplinary Research Integration to Accelerate Dementia Biomarker Development (MIRIADE). Frontiers in Neurology, 2022. https://doi.org/10.3389/fneur.2022.890638 ↩

5. Minseung Kim, Navneet Rai, Violeta Zorraquino, et al. Multi-omics integration accurately predicts cellular state in unexplored conditions for Escherichia coli. Nature Communications, 2016. https://doi.org/10.1038/ncomms13090 ↩

6. Kubra Karagoz, Raghu Sinha, Kazim Yalcin Arga. Triple Negative Breast Cancer: A Multi-Omics Network Discovery Strategy for Candidate Targets and Driving Pathways. OMICS: A Journal of Integrative Biology, 2015. https://doi.org/10.1089/omi.2014.0135 ↩

7. Somayah Albaradei. AI-Driven Approaches to Utilization of Multi-Omics Data for Personalized Diagnosis and Treatment of Cancer: A Comprehensive Review. Computer Modeling in Engineering & Sciences, 2025. https://doi.org/10.32604/cmes.2025.072584 ↩

8. Nupur Biswas, Saikat Chakrabarti. Artificial Intelligence (AI)-Based Systems Biology Approaches in Multi-Omics Data Analysis of Cancer. Frontiers in Oncology, 2020. https://doi.org/10.3389/fonc.2020.588221 ↩

9. 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 ↩

10. Clara D. M. van Karnebeek, Saskia B. Wortmann, Maja Tarailo‐Graovac, et al. The role of the clinician in the multi‐omics era: are you ready?. Journal of Inherited Metabolic Disease, 2018. https://doi.org/10.1007/s10545-017-0128-1 ↩