Skip to content

How Olink Proteomics Works

Oak
A lab machine's two robotic arms trail glowing threads that meet over a droplet and twist into one bright helix, with rows of similar stations fading behind.

Olink measures proteins in blood using the Proximity Extension Assay, or PEA. Each protein target is detected by a pair of antibodies, and each antibody carries a short single-stranded DNA oligonucleotide. When both antibodies bind the same protein molecule, the two oligos are held close enough to hybridize. A DNA polymerase extends the hybridized pair into a double-stranded reporter sequence unique to that target. That reporter is then counted, either by microfluidic qPCR or by DNA sequencing. The output is a relative, log2-scaled value called NPX. It is not an absolute concentration, and understanding that distinction is most of the work in using the data well.

Why dual recognition matters

A conventional sandwich immunoassay also uses two antibodies, but it reports optical signal from any antibody that sticks. Cross-reactive binding by either antibody produces background that is hard to separate from true signal. PEA adds a second requirement: the two oligos must be in proximity and hybridize before any amplifiable product exists. A single antibody bound to the wrong protein produces nothing detectable.

This is what allows Olink to run many assays in one well. In a standard ELISA, multiplexing fails because antibody pairs cross-talk. In PEA, each target has its own oligo sequence, so the readout separates targets by sequence rather than by physical compartment 1. The practical consequence is that thousands of proteins can be measured from one or two microliters of plasma or serum.

The product tiers and what they cost you in sample volume

Olink Target 96 panels each contain 92 protein assays plus four internal controls and consume about 1 µL of sample. Panels are organized by biology: Inflammation, Cardiovascular II and III, Neurology, Oncology, and others. Readout is qPCR on a Fluidigm-style 96×96 integrated fluidic circuit. Most published Olink work from the last decade used these panels, including studies of inflammatory proteins and incident myocardial infarction 2 and serum profiling after ischemic stroke 3.

Olink Explore and Explore HT move the readout to Illumina sequencing. Explore 3072 covers roughly 2,900 unique proteins across four 384-plex blocks. Explore HT covers roughly 5,400 proteins in a single reaction from about 2 µL. Sequencing counts replace Ct values as the raw measurement, which raises dynamic range and throughput but makes turnaround dependent on flow cell scheduling.

Target 48 and Flex are the quantitative options. These run a calibrator curve on each plate and report concentrations in pg/mL for a smaller number of targets. If you need an absolute number to compare against a clinical reference interval, this is the tier that produces one.

What NPX represents

NPX stands for Normalized Protein eXpression. It is a log2 unit. An NPX difference of 1.0 between two samples corresponds to a twofold difference in the measured signal for that assay. A difference of 3.0 corresponds to eightfold.

NPX is normalized per assay. It is not normalized across assays. A value of 6.0 for IL-6 and 6.0 for CXCL1 tell you nothing about their relative abundance in the blood. The two assays have different antibody affinities, different oligo efficiencies, and different background. Comparisons are valid within one assay across samples, and invalid across assays within one sample. This is the single most common error we see in reading Olink output.

Each assay also has a limit of detection, defined as the negative control signal plus three standard deviations. Values below LOD are reported rather than censored. We keep the reported NPX for exploratory work and drop any assay where more than about half of samples fall below LOD before running formal statistics. Imputing at LOD creates an artificial floor that inflates correlations.

Controls and the QC flags in your file

Every sample well contains three internal controls. The incubation control is a non-human antigen spiked in before the antibody step and reports on immunoreaction efficiency. The extension control is an antibody pre-bound to a matched oligo pair that is always in proximity, so it reports on polymerase and amplification performance. The detection control is a double-stranded template that bypasses both earlier steps and reports on the readout alone.

Each plate also carries a sample control, a negative control, and an inter-plate control. Target 96 flags a sample when its incubation or detection control deviates from the plate median by more than 0.3 NPX. Explore applies count-based thresholds instead. These flags land in your delivered data as a QC_Warning column.

Treat flagged samples as suspect rather than automatically discarded. In our experience the most frequent cause is hemolysis or a partial pipetting failure, and both are worth checking against the sample log before you delete a row.

Working with the files

Target 96 deliveries arrive as long-format CSV or Excel from NPX Manager, with one row per sample-assay pair. The columns you will use are the following.

  • SampleID
  • OlinkID
  • UniProt
  • Assay
  • Panel
  • Plate_ID
  • NPX
  • LOD
  • QC_Warning

Explore deliveries are Parquet files containing both raw counts and NPX.

The reference tooling is the OlinkAnalyze R package. Start with read_NPX(), then olink_qc_plot() for a per-sample median NPX versus IQR scatter, then olink_pca_plot() to check whether plate rather than biology explains the first component. For differential analysis, olink_ttest() and olink_lmer() handle the long format directly and return Benjamini-Hochberg adjusted p-values.

If you are merging batches run months apart, you need bridging. Re-run a subset of samples on both batches, then use olink_bridgeselector() and olink_normalization() to compute the per-assay median shift. Eight bridge samples is the documented minimum for Target 96, and we would use at least sixteen for Explore. Bridging with too few samples leaves per-assay offsets that look exactly like a treatment effect.

Failure modes worth knowing before you interpret anything

Sequence variants under an antibody epitope can abolish binding without changing protein abundance. The result is a very low NPX that reflects genotype rather than biology. This is why many strong cis-pQTL signals in large Olink cohorts sit in or near coding regions of the assayed gene. If you have your own genome, check for coding variants in any gene whose protein reads unusually low.

Cross-platform agreement is limited. Olink and aptamer-based platforms measure overlapping target lists, and for a large fraction of shared targets the correlation between them is weak. Neither is a straightforward substitute for mass spectrometry or for a clinical immunoassay, and none of these platforms is interchangeable for the purpose of tracking a value over time 4. Pick one platform and stay on it.

Finally, the largest Olink datasets have shown that multi-protein signatures can improve risk prediction over clinical variables for a range of conditions 5. That is a population-level result about discrimination across thousands of people. It does not translate into an individual-level statement about your own risk, and interpreting your numbers in a health context requires a clinician.

Questions people also ask

What does Olink measure? It measures relative abundance of specific, pre-selected proteins in plasma, serum or other fluids. Values are reported as NPX. Panels are curated around biology, with inflammation and cardiovascular targets the best characterized 1. Discovery studies have used the same underlying platform with different panel selections. Examples span glaucoma, autism, acute kidney injury and heart failure 6 7 8 9.

Is Olink part of Thermo Fisher? Yes. Thermo Fisher Scientific completed its acquisition of Olink Holding AB in 2024, and Olink now operates as part of Thermo Fisher.

How many proteins can Olink detect? Explore HT covers roughly 5,400 proteins in one run. Explore 3072 covers roughly 2,900. Individual Target 96 panels cover 92 each.

How much does Olink cost? Pricing depends on volume and on whether you go through a core facility or a commercial service provider. Single Target 96 panels are commonly quoted in the low hundreds of dollars per sample, and Explore HT in the high hundreds to low thousands. Ask for the per-sample price at your actual sample count, since the tiers move sharply.

Is NPX comparable between studies? Not directly. NPX is normalized within a run and within an assay. Comparing across studies requires shared bridge samples or a common reference set, and without them the absolute values should not be pooled 10.

Footnotes

  1. Han Wang, Tian Zhao, Jingjing Zeng, et al. Methods and clinical biomarker discovery for targeted proteomics using Olink technology. PROTEOMICS – Clinical Applications, 2024. https://doi.org/10.1002/prca.202300233 ↩ ↩2

  2. Elsa Valdes-Marquez, Robert Clarke, Michael Hill, et al. Proteomic profiling identifies novel independent relationships between inflammatory proteins and myocardial infarction. European Journal of Preventive Cardiology, 2023. https://doi.org/10.1093/eurjpc/zwad020 ↩

  3. Tian Zhao, Jingjing Zeng, Ruijie Zhang, et al. Serum Olink Proteomics-Based Identification of Protein Biomarkers Associated with the Immune Response in Ischemic Stroke. Journal of Proteome Research, 2024. https://doi.org/10.1021/acs.jproteome.3c00885 ↩

  4. Bhawana Singh, Oleg A. Karpov, Manuel Mayr. Clinical proteomics in cardiovascular medicine: Current capabilities, limitations, and future directions. Atherosclerosis, 2026. https://doi.org/10.1016/j.atherosclerosis.2026.120637 ↩

  5. Julia Carrasco-Zanini, Maik Pietzner, Jonathan Davitte, et al. Proteomic signatures improve risk prediction for common and rare diseases. Nature Medicine, 2024. https://doi.org/10.1038/s41591-024-03142-z ↩

  6. Hui-Ying Zhang, Qing Liu, Feng-Sheng Wang, et al. Targeted Proteomics Profiling for Biomarker Discovery in Glaucoma Using the Olink Proteomics Platform. Journal of Proteome Research, 2024. https://doi.org/10.1021/acs.jproteome.4c00593 ↩

  7. Xiao-Hong Bao, Bao-Fu Chen, Jun Liu, et al. Olink proteomics profiling platform reveals non-invasive inflammatory related protein biomarkers in autism spectrum disorder. Frontiers in Molecular Neuroscience, 2023. https://doi.org/10.3389/fnmol.2023.1185021 ↩

  8. Xiaoyang Li, Xiangyang Zhou, Xinbo Ping, et al. Combined Plasma Olink Proteomics and Transcriptomics Identifies CXCL1 and TNFRSF12A as Potential Predictive and Diagnostic Inflammatory Markers for Acute Kidney Injury. Inflammation, 2024. https://doi.org/10.1007/s10753-024-01993-9 ↩

  9. Nicolas Girerd, Daniel Levy, Kevin Duarte, et al. Protein Biomarkers of New-Onset Heart Failure: Insights From the Heart Omics and Ageing Cohort, the Atherosclerosis Risk in Communities Study, and the Framingham Heart Study. Circulation: Heart Failure, 2023. https://doi.org/10.1161/circheartfailure.122.009694 ↩

  10. Xinyun Chen, Yimei Huang, Liping Liu, et al. Olink proteomics profiling platform reveals myocardial metabolism-associated protein biomarkers in heart failure with preserved ejection fraction (HFpEF). Clinica Chimica Acta, 2026. https://doi.org/10.1016/j.cca.2025.120717 ↩

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.