Machines

A Setting Is Not a Promise: Inspecting the Hinge

A pancreatic-cancer patent arrives after years of bold blood-test bets, where the gap between a promising signal and safe screening remains vast.

Clay-diorama editorial illustration related to Diagnosis Of Pancreatic Cancer Using Targeted Quantification Of Site-Specific Protein Glycosylation.

What changes in this design

A reader might expect a pancreatic cancer diagnosis to come directly from one test result or one kind of information.

Here, information about at least three chosen small protein-related pieces is used to create a disease indicator, an intermediate result used to generate the diagnosis output.

That is worth knowing because the diagnosis output comes after information from at least three chosen protein-related pieces is used to create an intermediate indicator. What the indicator measures is not stated.

The first doctor visit

The number I started with makes every pancreatic-cancer test feel urgent: roughly half of patients have advanced disease by the time they first see a doctor. Five-year survival is reported at 13% overall, but about 83% for stage IA disease. A diagnosis from a simple biological sample has such pull that I can understand why. That pull is mine too. But early detection is not automatically good screening. In 2019, a national preventive-services panel concluded that screening asymptomatic adults was likely to bring no more than small benefit and at least moderate harm. Even high-risk programs found only 15.6 cases per 1,000 people screened. Those numbers, to me, are the threshold every new test has to clear.

A $201 million turn

By August 2021, the business behind this application had raised $201 million for a glycoproteomics platform pairing mass spectrometry with machine learning and being discussed across 16 oncology uses, including pancreatic cancer detection. The application is assigned to Venn Biosciences, the patent-holding entity for InterVenn. At the start of 2023, it had already launched laboratory-developed tests for melanoma and ovarian cancer, and planned a colon-cancer test later that year. In January, InterVenn replaced its founder-CEO with its chief operating officer to push commercialization. A second chief executive arrived that September, hired to move work from research settings into clinical pathways. Leadership changes don't tell us whether a test works; I won't pretend otherwise. Those changes do, I think, place this filing in a moment when laboratory promise was under pressure to become patient care.

November 30, 2022

That is why I see US20260253677A1, filed on November 30, 2022 and claiming priority to a November 30, 2021 provisional application, as a record of a particular bet rather than a finished diagnostic. It takes peptide structure data from a person's biological sample and uses at least three selected peptide structures, targeted measures of site-specific protein glycosylation, and a supervised machine-learning system to produce a pancreatic-cancer indicator and then a diagnosis output. The filing reflects, to me, a serious effort to move beyond one familiar marker. This filing doesn't let me infer how its results would hold up in a real-world screening population. Evidence that it can safely screen people without symptoms, sort every confusing look-alike, or improve outcomes on its own is not what I see here.

The test that stopped selling

That caution is not theoretical. In July 2023, Immunovia stopped US commercialization of IMMray PanCan-d, a pancreatic-cancer laboratory-developed blood test launched nearly two years earlier. It cited reliance on CA19-9 and inaccurate results for people who do not produce that marker. A 2025 review says 5% to 10% of patients are Lewis antigen-negative and cannot synthesize CA19-9. The race has not ended: Avantect reported an independent 1,445-person validation cohort with 82.6% overall sensitivity, 76.8% sensitivity for stage I-II disease, and 97.5% specificity. Those figures draw attention, and I can see why. Specificity matters because a false positive does not remain a number once it reaches a clinic. But the review's warning that retrospective, single-center panels can overstate accuracy through spectrum bias and often weaken in realistic populations carries more weight with me. It says screening demands specificity near 99%.

The part worth watching

The useful question, I think, is not whether the filing found an interesting signal. It may have found one worth studying. The question is whether a locked test, tried in the people who would actually receive it, can show that it finds disease earlier without setting off more harm than benefit. Prospective, multi-site evidence in a representative intended-use population, with false positives accounted for alongside sensitivity, is what I would want. Knowing who was missed and what followed a positive result is also something I would want. The route to patients is something I will also watch. A federal final rule issued in 2024 treated in vitro diagnostics as devices even when a laboratory manufactures them; its announced schedule put moderate- and low-risk laboratory-developed tests on a path to premarket review by May 6, 2028. That does not validate this test. It sharpens the line for me: a patent can define a bet, while clinical evidence must earn the right to act on it.

Sources

Reporting for this piece drew on the published filing and the references below.

The record, briefly. US20260253677A1, “Diagnosis Of Pancreatic Cancer Using Targeted Quantification Of Site-Specific Protein Glycosylation” Receives peptide structure data from a subject's biological sample and uses a supervised machine learning model to generate a pancreatic cancer disease indicator from at least three selected peptide structures, then generates a diagnosis output from that indicator.

Identity details. The saved title-page transcription needs review, so names are not repeated here. Check the official published title page.

More angles on this

A Handoff Worth Watching: Fragment-Based Drug Design With an Assistant at the BenchAnother reading from the same G16B classification family, bioinformatics.
Measurement And Analysis Method For Detecting And Quantifying Makes a Small Case for Better TimingAnother reading from the same G16B classification family, bioinformatics.
CPC G16B: BioinformaticsThe official classification family this record was filed into, published in the USPTO Cooperative Patent Classification scheme.
CPC G01N: Investigating or analysing materials by determining their chemical or physical propertiesThe official classification family this record was filed into, published in the USPTO Cooperative Patent Classification scheme.
Methods For Treatment Of Thyroid Cancer Treats the Relationship as the PointAnother reading from the neighbouring G16Z classification family, information and communication technology [ict] specially adapted for specific application fields.

Read around this one:Public patent record USPTO published figure