Proteomics

Olink proximity extension assay analysis: bridge-normalized NPX QC through differential testing and pathway enrichment

Olink proximity extension assay (PEA) analysis quantifies targeted proteins in plasma, serum, or CSF from Normalized Protein eXpression (NPX) exports. Pepkio delivers QC, bridging, differential-abundance tables, figures, commented R code, and a Methods draft for academic, biotech, and pharma clients. Custom analyses are scoped at kickoff. The UK Biobank Pharma Proteomics Project measured 2,923 proteins across 54,219 participants on Olink Explore 3072 [1].

Key facts

Key facts about Olink Proximity Extension
FactValue
Supported platforms / instrumentsOlink Target 96; Explore 384 modules (Cardiometabolic, Inflammation, Neurology, Oncology, and II panels); Explore 3072; Explore HT; Focus; Reveal; NPX exports from NPX Manager, NPX Map, and NPX Explore software
Input requirementsNPX or QUANT exports (CSV, Excel, Parquet, or ZIP as exported by Olink software); sample manifest with condition, plate/batch, and replicate IDs; bridge samples for multi-batch or multi-project designs (8-64 overlapping samples depending on Olink product pairing; [3]); multiple biological replicates per condition (often ≥3) for variance estimation [5]. Confirmed at kickoff.
Reference builds supportedNot genome-based. UniProt accessions per Olink assay panel (OlinkID to UniProt mapping validated via `check_npx` in OlinkAnalyze 5.0+). No GRCh38 or other assembly is required.
Primary tools (with versions)OlinkAnalyze 5.0+; OlinkAnalyzeVignettes; limma 3.60+; clusterProfiler via `olink_pathway_enrichment`; lme4/lmerTest via `olink_lmer`, pinned per project
Typical turnaround time2-4 weeks (single NPX project, one contrast, standard DE and enrichment); 4-8 weeks (multi-batch bridging, cross-product normalization, mixed models, or large cohorts). Confirmed at kickoff.
Deliverable formatsLong and wide NPX matrices (`.csv`, `.tsv`, `.rds`); differential protein tables; PDF/SVG QC, PCA, volcano, and heatmap plots; HTML QC report; commented R scripts; Methods draft
Key cited best-practice referenceOlink Proteomics [6], OlinkAnalyzeVignettes bridging tutorials; Assarsson et al. [2], *PLOS ONE* (96-plex PEA validation)
Custom / bespoke analysisNon-standard contrasts, cross-cohort integration, proteogenomic correlation, custom enrichment gene sets, client-specified statistical models, or alternate deliverable formats scoped at kickoff

What is Olink proximity extension assay analysis?

Olink PEA analysis imports vendor NPX matrices, checks assay and sample integrity, optionally bridges batches or Olink products, and tests per-protein abundance differences on log₂-scaled NPX values. That measures which targeted proteins change between conditions without peptide identification or spectral library search [2, 3]. Unlike LC-MS discovery proteomics, PEA uses dual-antibody proximity ligation with qPCR or NGS readout on fixed panels [4]. Unlike single-target ELISA, 96-plex panels consume 1 µL per sample [2]. Explore HT measures about 5,400 proteins from 2 µL with about a 10-log dynamic range [3]. Custom workflows are agreed at kickoff. See the Olink proximity extension assay glossary.

When should you use Olink proximity extension assay analysis?

Olink PEA fits when you need multiplex measurement of known or candidate proteins in biofluids at low sample volume, for example biomarker panels, pathway monitoring, or genetic-proteomic association, without untargeted MS depth [1].

Comparison of Olink PEA (NPX), LC-MS DDA/DIA, and single-target ELISA approaches
ApproachBest forLimitationsApproximate cost range
Olink PEA (NPX)Targeted multiplex biomarkers in plasma, serum, or CSF; 96-5,400 proteins; 1-2 µL sample inputRelative NPX quantification; bridging required across batches or products; fixed antibody panelsModerate per-sample assay cost; moderate bioinformatics
LC-MS DDA/DIAUntargeted proteome discovery, PTMs, novel protein identificationHigher sample volume and MS time; complex missing-value handlingHigher MS acquisition and bioinformatics cost
Single-target ELISAOne analyte with an established clinical-grade assayNo multiplex discovery; poor scalability for panelsLower per analyte; limited for multi-protein signatures
  • Population-scale pQTL mapping: Sun et al. [1] profiled 2,923 plasma proteins in 54,219 UK Biobank participants on Olink Explore 3072 and identified 14,287 primary protein quantitative trait loci, 81% of them previously undescribed [1].
  • AD pathology staging in CSF: Pichet Binette et al. [7] measured CSF proteins on Olink Explore 3072 in 877 BioFINDER-2 participants and identified 127 differentially abundant proteins across Aβ and tau pathology stages, including glial markers such as SMOC1 linked to Aβ plaques [7].
  • ALS plasma biomarker panel: Chia et al. [6] used Olink Explore 3072 on plasma from 183 ALS patients and 309 controls, identifying 33 differentially abundant proteins and a machine-learning panel with area under the curve 98.3% [6].

How the analysis works — step by step

  1. 1. Validate NPX exports and experimental design

    Pepkio confirms NPX format, panel version, contrast definitions, and bridge-sample plans before import. Confounded batch-by-condition designs, or too few bridge samples for multi-project merging, are flagged at kickoff [3].

    Tools and outputs

    Tools used: Custom validation scripts; OlinkAnalyze 5.0+ design checklist

    Output: sample_manifest.csv with sample IDs, conditions, plate/batch IDs, bridge flags, and design notes

  2. 2. Import NPX data

    NPX files load from Olink software exports without manual editing. QUANT exports can be log-transformed when scoped (OlinkAnalyze 5.0+).

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`read_npx` / `read_NPX`)

    Output: npx_long_raw.rds; import log with sample and assay counts

  3. 3. Check data integrity

    OlinkIDs, duplicate sample IDs, assay QC warnings, and UniProt mapping are validated via `check_npx`. The check log feeds downstream functions (OlinkAnalyze 5.0+).

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`check_npx`)

    Output: check_log.rds; data_integrity_report.csv

  4. 4. Clean unsuitable records

    Invalid assays, failed QC flags, and external controls are removed per agreed thresholds (OlinkAnalyze 5.0+).

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`clean_npx`)

    Output: npx_cleaned.rds; cleaning_log.csv

  5. 5. Run plate and sample QC

    Plate distribution and sample NPX boxplots are reviewed. Outlier samples flagged by IQR or median rules are documented before exclusion or retention [3].

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`olink_dist_plot`, `olink_qc_plot`)

    Output: sample_qc_summary.csv; plate and sample QC plots (PDF/SVG)

  6. 6. Profile LOD and missingness

    LOD is calculated from negative controls or fixed Olink LOD values. High-missingness assays are filtered before statistics [3].

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`olink_lod`); OlinkAnalyzeVignettes LOD tutorial

    Output: lod_summary.csv; missingness_profile.csv; assay_filter_log.csv

  7. 7. Bridge and normalize across batches or projects

    Bridge samples are selected with `olink_bridge_selector`. Normalization uses median-of-paired-differences bridging, or quantile smoothing for cross-product merges (Explore 3072 to HT). PCA before and after bridging is used to check that batch separation is reduced [3].

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`olink_bridge_selector`, `olink_normalization_bridge`); OlinkAnalyzeVignettes bridging tutorials

    Output: npx_normalized.csv; bridging_adjustment_factors.csv; pre/post bridging PCA plots

  8. 8. Explore data structure

    PCA on normalized NPX values is used to separate condition from plate or batch when that is possible. UMAP is added when scoped [3].

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+; custom R (`umap` when scoped)

    Output: PCA/UMAP plots (PDF/SVG); unsupervised_qc_summary.csv

  9. 9. Test differential protein abundance

    Per-protein models match the design: `olink_ttest` or `olink_wilcox` for two groups, `olink_anova` for multi-group, or `olink_lmer` with plate or subject random effects. Benjamini-Hochberg FDR controls multiplicity (OlinkAnalyze 5.0+; [8] for limma contrasts).

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`olink_ttest`, `olink_anova`, `olink_lmer`); limma 3.60+ when design fits matrix contrasts

    Output: dep_results_<contrast>.csv (OlinkID, UniProt, Assay, Estimate, SE, DF, pvalue, adj.pvalue, N)

  10. 10. Run pathway enrichment and package deliverables

    Significant proteins feed GSEA or ORA via `olink_pathway_enrichment` against MSigDB, Reactome, KEGG, or GO. Volcano plots, heatmaps, and enrichment dot plots are bundled with scripts, README, and a Methods draft (OlinkAnalyze 5.0+; [9]).

    Tools and outputs

    Tools used: OlinkAnalyze 5.0+ (`olink_volcano_plot`, `olink_pathway_enrichment`)

    Output: pathway_enrichment.csv; figure bundle (PDF/SVG); README; Methods draft

What Pepkio delivers

Processed data

  • Raw and cleaned long-format NPX tables (`.csv`, `.tsv`, `.rds`)
  • Wide-format matrices when requested
  • Normalized NPX after bridging with per-assay adjustment factors
  • Parameter log documenting LOD thresholds, bridging method, statistical model, and FDR cutoffs

Figures (PDF/SVG)

  • Plate and sample QC plots
  • Missingness heatmap
  • PCA or UMAP pre- and post-bridging
  • Volcano plot per contrast; top differential protein heatmap
  • Pathway enrichment dot plots

Tables

  • sample_manifest.csv, data_integrity_report.csv, cleaning_log.csv
  • sample_qc_summary.csv, lod_summary.csv, missingness_profile.csv
  • assay_filter_log.csv, bridging_adjustment_factors.csv
  • dep_results_<contrast>.csv, pathway_enrichment.csv

Code

  • Commented R scripts per analysis stage
  • `renv` lock file or `sessionInfo()` export
  • Delivery via private Git repository or agreed file transfer

Documentation

  • HTML QC report with thresholds, bridge-sample rationale, and outlier flags
  • README with reproduction instructions
  • Methods draft citing OlinkAnalyze version, normalization, LOD handling, and statistical tests
  • Post-delivery reviewer support for clarification and minor revisions within agreed scope (typically ≤20% of deliverables)

Technical decisions we make — and why

Normalization: bridge normalization default; quantile smoothing when cross-product diagnostics require it
NPX is relative. Overlapping bridge samples align projects via the median of paired differences [3]. Explore 3072 to HT bridging may need quantile smoothing when distribution shapes diverge beyond a median shift [3].
LOD handling: filter high-missingness assays; half-LOD imputation only when pre-specified
Assays with many undetectable values can inflate false discovery if they are ignored. Pepkio documents below-LOD rates and excludes assays above agreed thresholds before DE testing [3].
Statistical model: `olink_lmer` with plate random effect for multi-plate designs; simpler tests when design permits
Mixed models with `(1|Plate)` or `(1|Subject)` handle repeated measures without treating technical replicates as independent (OlinkAnalyze 5.0+). Two-group designs without nesting may use `olink_ttest`, `olink_wilcox`, or limma contrasts [8].
Multiple testing: Benjamini-Hochberg FDR across all tested proteins
OlinkAnalyze applies BH adjustment by default in `olink_lmer` and related functions (OlinkAnalyze 5.0+).
Bridge sample selection: QC-passing biological samples, not external controls
`olink_bridge_selector` prioritizes samples with high detectability and QC pass status. External controls are excluded because they do not represent the study sample dynamic range [3].

Common questions

What is the minimum sample size and replicate count for Olink PEA analysis?

Multiple biological replicates per condition, often three or more, help estimate per-protein variance on NPX [5]. For a two-group design with mean ΔNPX = 0.5 and pooled SD = 0.8 (Cohen's d ≈ 0.63), roughly 41 samples per group yields 80% power at α = 0.05 for a single test. Testing hundreds of proteins requires larger cohorts and FDR control. Multi-project designs also need bridge samples (8-64 depending on Olink product pairing; [3]). Pepkio reviews power and bridge plans at kickoff.

Can you analyze samples with Olink QC warnings or high below-LOD rates?

Yes, with documented caveats. Warning-flagged samples and high-missingness assays are flagged in the QC report before differential testing. Outlier samples in PCA are discussed with you. A re-run on the Olink platform is recommended when QC failure threatens the study question. Retained warning samples are noted in the parameter log.

Do you support Olink Target 96, Explore 3072, Explore HT, Focus, and Reveal NPX exports?

Yes, for NPX files exported from Olink software without manual alteration, when panel metadata and export format are confirmed at kickoff. OlinkAnalyze 5.0+ reads CSV, Excel, Parquet, and ZIP archives from NPX Manager, NPX Map, and NPX Explore pipelines. QUANT data can be analyzed when log-transformed per Olink guidance.

How long does Olink PEA analysis take at Pepkio?

A standard project (single NPX project, roughly 20-200 samples, one primary contrast, DE testing, and pathway enrichment) typically completes in 2-4 weeks from data receipt. Multi-batch bridging, cross-product normalization (Explore 3072 to HT), mixed-model designs, or cohorts that need extensive QC review may take 4-8 weeks. Exact timelines are confirmed at kickoff.

How do you handle batch effects across Olink plates or project runs?

When batch or project is known and not fully confounded with condition, Pepkio applies Olink bridge normalization using overlapping bridge samples and documents adjustment factors per assay [3]. For within-project multi-plate designs, plate may enter as a random effect in `olink_lmer`. Generic post-hoc correction without bridge samples is not applied to cross-project NPX comparisons.

Do I own the code, and in what format is it delivered?

Yes. You retain full ownership of all code, scripts, and results delivered under the project agreement. Pepkio provides commented R scripts with `renv` lock files or `sessionInfo()` exports. NPX matrices use `.csv`, `.tsv`, and `.rds` formats. R Markdown delivery is available on request.

Can I be involved during analysis?

Yes. Checkpoint reviews occur after data integrity QC, after bridging normalization, and before final delivery. You can review contrast definitions, bridge-sample lists, LOD filtering thresholds, and FDR cutoffs within agreed scope. A dedicated scientific contact leads the project and documents decisions in the QC report and Methods draft.

What does post-delivery reviewer support include?

Clarification of normalization, bridging, LOD handling, statistical models, and minor figure or table revisions within agreed scope (typically ≤20% of deliverables). It does not include open-ended reanalysis or new biological contrasts. Substantial new work is scoped as separate milestones.

Is co-authorship required?

No. Pepkio operates as a fee-for-service provider unless co-authorship is discussed and agreed in advance. You retain ownership of results and code delivered under the project agreement.

Can I compare NPX values across two Olink projects without bridge samples?

Not for valid joint inference. NPX is a relative log₂-scaled unit normalized within each project, so separate projects need bridge normalization using overlapping biological samples [3]. Pepkio analyzes single-project datasets without bridging when that is appropriate, or sets bridge-sample requirements before merging multi-batch data.

How do you handle proteins with NPX below the limit of detection?

Per-assay LOD is calculated using `olink_lod` [3]. Assays where a large fraction of samples fall below LOD are typically excluded from DE testing. For retained assays with sporadic below-LOD values, Pepkio documents the handling rule (complete-case analysis, censoring, or pre-specified half-LOD imputation) in the parameter log before modeling.

Should I use OlinkAnalyze or limma for differential testing on NPX data?

OlinkAnalyze is the default because the functions are panel-aware, apply BH-FDR across OlinkIDs, and support mixed models with plate random effects (`olink_lmer`; OlinkAnalyze 5.0+). limma 3.60+ is used when the design fits a standard matrix contrast and the client requests limma-specific outputs [8]. The choice is documented at kickoff.

Related services

  • DDA/DIA proteomics — Untargeted LC-MS discovery upstream of Olink panel validation.
  • Single-cell proteomics — Cell-state proteomics when bulk or plasma averages mask heterogeneity.
  • Bulk RNA-seq — Proteogenomic correlation when matched RNA-seq and Olink data share sample IDs.
  • Biomarker discovery — Supervised signature construction and cross-validation on Olink protein features.
  • Custom analysis — Non-standard cross-cohort integration, custom statistical models, or client-specified deliverables beyond the standard Olink pipeline.
References
  1. Sun BB, Chiou J, Traylor M, et al. Plasma proteomic associations with genetics and health in the UK Biobank. Nature. 2023;622(7982):329–338. https://doi.org/10.1038/s41586-023-06592-6 (PMID: 37794186)
  2. Assarsson E, Lundberg M, Holmquist G, et al. Homogenous 96-plex PEA immunoassay exhibiting high sensitivity, specificity, and excellent scalability. PLOS ONE. 2014;9(4):e95192. https://doi.org/10.1371/journal.pone.0095192 (PMID: 24755770)
  3. Nevola K, Sandin M, Guess J, et al. OlinkAnalyze: facilitate analysis of proteomic data from Olink. CRAN. Version 5.0.0. 2026. https://doi.org/10.32614/CRAN.package.OlinkAnalyze
  4. Lundberg M, Eriksson A, Tran B, Assarsson E, Fredriksson S. Homogeneous antibody-based proximity extension assays provide sensitive and specific detection of low-abundant proteins in human blood. Nucleic Acids Research. 2011;39(15):e102. https://doi.org/10.1093/nar/gkr424 (PMID: 21646338)
  5. Cairns DA, Barrett JH, Billingham LJ, et al. Sample size determination in clinical proteomic profiling experiments using mass spectrometry for class comparison. Proteomics. 2009;9(1):74–86. https://doi.org/10.1002/pmic.200800417 (PMID: 19053145)
  6. Chia R, Moaddel R, Kwan JY, et al. A plasma proteomics-based candidate biomarker panel predictive of amyotrophic lateral sclerosis. Nature Medicine. 2025;31(10):3440–3450. https://doi.org/10.1038/s41591-025-03890-6 (PMID: 40830661)
  7. Pichet Binette A, Gaiteri C, Wennström M, et al. Proteomic changes in Alzheimer\'s disease associated with progressive Aβ plaque and tau tangle pathologies. Nature Neuroscience. 2024;27(10):1880–1891. https://doi.org/10.1038/s41593-024-01737-w (PMID: 39187705)
  8. Ritchie ME, Phipson B, Wu D, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Research. 2015;43(7):e47. https://doi.org/10.1093/nar/gkv007 (PMID: 25605792)
  9. Yu G, Wang L-G, Han Y, He Q-Y. clusterProfiler: an R package for comparing biological themes among gene clusters. OMICS: A Journal of Integrative Biology. 2012;16(5):284–287. https://doi.org/10.1089/omi.2011.0118 (PMID: 22455463)

Let's Talk About Your Science

Tell us:

  • • Your biological question
  • • Data type and size
  • • Timeline constraints

We'll tell you:

  • • What's feasible
  • • How long it will take
  • • Exactly what it will cost
Discuss Your Project

Contact us to start with a free consultation. Need everyday bench calculators? Try our free lab tools.