Proteomics

Technology Landscape

1.1 Mass spectrometry remains the discovery standard

Bottom-up proteomics digests proteins into peptides, identifies them by tandem mass spectra, and assembles protein-level evidence. A key limitation is that database-dependent searches can miss novel peptides or sequence variants. Deep-learning methods such as DeepNovo address this by doing de novo sequencing directly from spectra [arXiv:1701.08635].

1.2 Affinity/PEA: high multiplexing with low sample input

Olink’s proximity extension assay, or PEA, uses pairs of antibodies conjugated to oligonucleotide probes. When both probes bind the same protein, the oligos are brought into proximity and can be ligated or extended into an amplifiable DNA barcode, enabling qPCR or NGS readout [Olink PEA technology page].

1.3 Single-molecule proteomics: the next commercialization frontier

Nautilus Biotechnology is developing a single-molecule protein-analysis platform based on immobilizing proteins on arrays and repeatedly probing them with affinity reagents. Its SEC filings indicate that it is still development-stage and pre-revenue [Nautilus 10-K SEC page].

Patent Dissection

Patent: US 6,878,515 B2 — “Ultrasensitive immunoassays and kits using proximity probes”

Assignee: Olink Bioscience

Source: Google Patents US6878515B2

The patent describes proximity probes that each contain:

• a binding moiety for the target analyte, and
• a nucleic acid moiety.

When both probes bind the same analyte, their nucleic acid moieties are brought into proximity, allowing ligation/extension to create a detectable nucleic-acid signal. This is the core enabling mechanism behind Olink PEA.

Why this matters: The requirement for dual binding reduces background and improves specificity for low-abundance proteins. It also converts protein abundance into a nucleic-acid amplification problem, enabling high-multiplex panels.

arXiv Paper

Paper: “De Novo Peptide Sequencing by Deep Learning”

arXiv ID: arXiv:1701.08635

DeepNovo uses a deep neural network to predict peptide sequences directly from tandem mass spectra. The paper reports that the model improves identification of novel peptides and single amino acid variants compared with database-dependent search methods.

Cross-reference implication: For affinity-based proteomics, MS/deep-learning methods remain important for validating antibody targets and discovering novel biomarkers that are not yet part of affinity panels.

Company Filing Cross-Reference

4.1 Olink Holding 20-F

Olink Holding AB filed 20-F reports with the SEC before its acquisition by Thermo Fisher. The filing describes PEA as the core technology platform.

• SEC Olink page: Olink Holding 20-F filings

4.2 Nautilus Biotechnology 10-K

Nautilus Biotechnology’s 10-K describes its single-molecule protein-analysis platform as still in development. The company has not yet generated product revenue, so performance parameters around its commercial instrument remain pre-commercial.

• SEC Nautilus page: Nautilus 10-K filings

Cross-reference finding: The patent-to-filing path is clearer for Olink than for Nautilus. Olink’s PEA patent is materially expressed in a commercial product line and described in its SEC filings. Nautilus’s filings confirm a strong patent/technical narrative but no commercial revenue yet, so single-molecule performance claims should be held to a higher validation bar.

Cross-Reference Summary

What the technology is

Proteomics is large-scale measurement of proteins, their post-translational modifications, interactions, and abundance changes. The two workhorse multiplex technologies are:

• Mass spectrometry (MS) — bottom-up/shotgun and targeted LC-MS/MS proteomics; broad discovery but dynamic range and throughput remain bottlenecks. Source: Aebersold & Mann, Nature 537:347–355 (2016), DOI:10.1038/nature19949, published 2016-09-14, accessed 2026-05-05.
• Affinity-based multiplexing — aptamer-based SomaScan (now Standard BioTools) and proximity-extension assay Olink (now Thermo Fisher). SomaScan uses modified DNA aptamers; the core off-rate-enrichment method is protected by a granted patent. Source: US Patent 7,964,356 B2, granted 2011-06-21, accessed 2026-05-05: patents.google.com .

A key tier distinction: US 7,964,356 B2 is a granted patent, so it supports recorded legal claims and potential moat, but it is not proof of analytical performance or clinical value.

Cross-check: marketing vs filed disclosures

Standard BioTools’ SomaScan page claims 7,000 protein measurements per sample. This is a Tier 2 vendor self-report and should be treated as [SINGLE-SOURCE] for the 7,000-plex claim unless an independent technical assessment is found. Source: Standard BioTools SomaScan product page, accessed 2026-05-05: standardbio.com .

Formal SEC filings do not independently validate the 7,000-plex analytical performance. Nautilus Biotechnology is a clear disclosure gap: its corporate materials describe a single-molecule protein analysis platform, while its SEC filing disclosure is that product revenue is not yet material; that gap is expected for an early-stage company, but it underscores that marketing language is ahead of filed performance. Source: Nautilus IR/SEC EDGAR page, accessed 2026-05-05: ir.nautilus.bio .

Quantitative worked calculation

We estimated the sample size needed to detect a 20% protein-abundance difference (FC = 1.20) using a two-group, two-sided test with 5% technical CV, then added biological variation.

Formula:

n_per_group = 2 (Z_alpha/2 + Z_beta)^2 CV_total^2 / [ln(FC)]^2

Inputs:

Z_alpha/2 = 1.96, Z_beta = 0.842, FC = 1.20, technical CV = 0.05, biological CV scenario = 0.20.

CV_total = sqrt(0.05^2 + 0.20^2) = 0.2062

ln(1.20) = 0.1823, squared = 0.03324

2 (1.96 + 0.842)^2 0.2062^2 = 0.6675

n = 0.6675 / 0.03324 = 20.1, round up to 21 per group.

Sensitivity: with only technical CV (0.05), the nominal size is 1.18, round to 2 per group; adding the assumed 20% biological CV raises it 10-fold. The implication is that biological variability, not instrument CV, dominates study design for single-protein discovery.

Parameter table

Unit economics / business model

The value chain monetizes proteomics several ways:

• Instrument + consumable/service razor-blade model: Thermo Fisher (TMO), Danaher/SCIEX (DHR), Waters (WAT), Bruker (BRKR), and Agilent (A) sell mass spectrometers and recurring consumables/service contracts. Exact instrument prices and segment margins need primary-source validation against each company’s 10-K, but the recurring consumables component is disclosed generally in SEC filings and IR materials. Source: company IR pages, accessed 2026-05-05: ir.thermofisher.com ; investors.danaher.com ; ir.waters.com ; ir.bruker.com ; investor.agilent.com .
• Affinity panel/economies of scale model: Standard BioTools SomaScan is sold as high-plex service/kit consumable economics; the value accrues to per-sample assay fees and reagents. Exact list prices need primary-source validation, but the business is structured around per-sample recurring revenue rather than one-time instruments. Source: Standard BioTools IR page, accessed 2026-05-05: investors.standardbio.com .
• Pre-revenue single-molecule future-instrument model: Nautilus (NAUT) and Quantum-Si (QSI) plan an instrument-plus-consumable model, but their filed disclosures show product revenue is not yet material. Source: NAUT IR ir.nautilus.bio ; QSI IR ir.quantum-si.com .

Disconfirming Evidence

The thesis would be weakened or invalidated if:

• an independent, peer-reviewed technical assessment fails to reproduce SomaScan’s 7,000-plex coverage or CV claims;
• patent US 7,964,356 B2 is invalidated, expires, or is easily designed around;
• MS throughput and single-molecule methods improve enough to erode the cost/plex advantages of affinity panels;
• Nautilus or Quantum-Si fail to demonstrate performance against existing MS/immunoassay methods in pre-competitive benchmarking; or
• large biobank proteomic associations do not translate into reproducible drug-target or clinical-test utility.

Original scope and definitions

The large-scale study of proteins — the functional molecules of biology. While genomics tells you what could happen (the blueprint), proteomics tells you what is happening (the machinery). Market ~$40B (2026) → ~$95B (2035), ~10% CAGR. Reagents/kits are ~75% of the market. Unlike DNA sequencing — which converged on a dominant platform (Illumina SBS) — proteomics remains fragmented across five competing physical principles, none of which dominates. This fragmentation is both the bottleneck and the opportunity.

Operating constraints

Physics: why it's hard — the five-platform fragmentation

Proteins are harder than DNA for three fundamental reasons:

1. No amplification: DNA can be copied (PCR) — a single molecule becomes billions. Proteins cannot be amplified. Every molecule must be detected directly, requiring attomole-to-zeptomole sensitivity.
2. Dynamic range: The plasma proteome spans >10 orders of magnitude in concentration (albumin at ~40 mg/mL vs. cytokines at ~1 pg/mL). No single detector can cover this range without depletion or fractionation.
3. Proteoforms: A single gene can produce dozens of protein variants via alternative splicing, post-translational modifications (PTMs — phosphorylation, glycosylation, acetylation, ubiquitination, cleavage). Measuring "amount of protein X" misses which proteoform is present. Cancer signaling depends on which phosphorylated form, not total abundance.

Five platforms attack these problems with different physics:

1. Mass spectrometry (DIA / TMT — Thermo, Bruker)

The unbiased workhorse. Proteins are digested into peptides, separated by LC, ionized, and fragmented in the mass spectrometer. Data-independent acquisition (DIA) fragments everything in a mass window systematically, enabling retrospective re-analysis. TMT (tandem mass tags) multiplexes up to 18 samples with isobaric labels. The bottleneck: dynamic range limited to ~4–5 orders of magnitude without extensive fractionation; low-abundance proteins in plasma (pg/mL) are below detection without depletion. One run = 6,000–10,000 proteins in cell lysate, ~1,000–2,000 in undepleted plasma.

2. Olink PEA (Proximity Extension Assay — Thermo Fisher, acquired 2024)

Pairs of antibodies bind adjacent epitopes on the same protein; their conjugated DNA oligos hybridize and extend, creating a unique PCR-amplifiable barcode. Dual-binding requirement gives high specificity (reduces cross-reactivity vs. single-antibody methods). The bottleneck: antibody-based — limited to proteins with good antibody pairs (~5,400 in Explore HT panel). PTMs or conformational changes that block epitope binding are invisible. Strengths: minimal sample volume (1 μL plasma), excellent reproducibility, used in UK Biobank and pharma trials.

3. SomaScan (SOMAmer aptamers — Illumina/Standard BioTools, acquired from SomaLogic)

Modified DNA aptamers (SOMAmers) with slow off-rates bind proteins with high affinity. Released aptamers are quantified on microarrays or via NGS. Current plex: ~7,000–11,000 proteins. The bottleneck: concordance with Olink for overlapping proteins is modest (median correlation ~0.26–0.30) — the two platforms often give different biological conclusions for the same protein. Platform-specific effects mean you cannot directly compare Olink and SomaScan data without cross-validation.

4. Single-molecule protein sequencing — Nautilus Voyager (NAUT)

Proteins are immobilized on a single-molecule array and exposed iteratively to a library of affinity reagents against short linear epitopes. The pattern of binding events across cycles creates a unique "fingerprint" that identifies the protein and its proteoform. The bottleneck: still in Early Access Program (2026); throughput and proteome coverage are developing. Strength: proteoform-level resolution — can distinguish tau splice variants and phosphorylation states in Alzheimer's research that bulk methods blur together.

5. Single-molecule sequencing — Quantum-Si Platinum/Proteus (QSI)

Semiconductor-chip-based protein sequencer. Immobilized proteins/peptides undergo sequential N-terminal amino acid cleavage by exopeptidases; fluorescently labeled amino acids are detected via CMOS imaging. Produces actual sequence reads (tens of amino acids) — the closest analog to DNA sequencing. The bottleneck: short read lengths mean identification relies on reference databases; single-pass error rates are higher than NGS for DNA. Strengths: protein-level variant validation, PTM mapping, drug target characterization. Proteus instrument launch H2 2026.

Economic constraints

• No single dominant platform: Unlike Illumina's ~80–85% share in sequencing, no proteomics platform exceeds ~30–40% market share. This fragments instrument purchasing decisions and slows standardized clinical adoption.
• Switching costs are real but lower than sequencing: Affinity-based methods (Olink, SomaScan) have lower instrument switching costs than mass spectrometry (where methods, columns, and trained staff create lock-in). But data incompatibility between platforms (Olink vs. SomaScan) means you cannot easily merge datasets — creating de facto lock-in at the data layer.
• Single-molecule platforms are pre-revenue: Nautilus and Quantum-Si are burning cash building adoption; commercial inflection is 2027–2028 at earliest. This means the near-term investable proteomics exposure is through TMO (Olink + mass spec) and BRKR (timsTOF).