Joshua LaBaer directs a center for personalized diagnostics, which is a polite way of saying his business is turning a vial of blood into a test. So when the executive director of Arizona State’s Biodesign Institute announced that his lab could predict, before the needle goes in, who will respond to a vaccine and who won’t, the announcement arrived shaped the way these announcements always are: a confident number bolted to the promise, and no number at all bolted to the performance.

What LaBaer’s team actually published, on August 21 in Cell Press Blue, was a retrospective look at 8,687 blood samples drawn from 4,089 people. The researchers measured antibodies against 185 microbial and viral targets, then turned a deep-learning model loose on the patterns to find which pre-existing signatures tracked with a strong response after vaccination. They found some. Higher baseline levels of certain antibodies, including ones aimed at Staphylococcus aureus, RSV, and human respirovirus 3, showed up more often in people who mounted a robust response. The team called these “sentinel” antibodies, a tidy shorthand for baseline immune readiness. The vaccine the model was trained to predict, the one doing the actual work in the study, was the COVID-19 shot. Most of the coverage mentions that part in passing, if at all.

Here is the claim, in LaBaer’s words: certain biomarkers, analyzed with AI, “can predict who is likely to respond well to a vaccine, even before they receive it.” Here is what the paper describes: an association, found in stored blood, after the vaccinations had already happened and the outcomes were already on record. Nowhere in the Arizona State release or the trade write-ups is there an area-under-the-curve, a sensitivity, a specificity, anything that would tell a reader how often the model is right when it stamps a living person a weak responder. The authors’ own caveat, parked where caveats go, is that the findings need confirming in future studies and other vaccines before anyone acts on them.

The press release made a prediction. The paper made a correlation.

The number the coverage led with was the unsettling one: roughly 5 to 6 percent of healthy participants turned out to be weak responders despite nothing being wrong with their immune systems. As immunology, that is a genuine puzzle. Ask what the tool built on top of it is for, and the puzzle turns practical fast. LaBaer said it plainly: profiling these antibodies could help doctors identify patients who need “additional vaccine doses, closer follow-up or alternative protective measures.” Strip the soft clause out of the middle and the use case is a blood test that sorts healthy people into a bin whose recommended fix is another COVID shot.

WEAK RESPONDERS (percent)
of healthy participants56
Roughly 5 to 6 percent of healthy participants turned out to be weak responders. Source: Cell Press Blue, 2026

It matters who is building this and on whose dime. The work was funded by the National Institutes of Health, the National Cancer Institute, and the Autoimmunity Centers of Excellence, public money at a public university. The lab is the Virginia G. Piper Center for Personalized Diagnostics, which LaBaer also runs. “Sentinel antibody profiling,” in that setting, is not a lyrical figure for the immune system. It is a diagnostic product, and diagnostic products get sold. None of the news coverage carried a conflict-of-interest statement, which tells you something about how the story was built to be received.

Set it against the record. For most of a decade the public-health apparatus oversold what the COVID vaccines could do, for how long, and against what, and burned the credibility it took to say so. An apparatus in that hole has two options. Earn the trust back, or build the instrument that identifies, in advance, the people it can tell to line up again, and describe that instrument in the soft vocabulary of personalized medicine. A tool that cannot yet say how often it is right, but can already say who needs another dose, is not aimed at patients. It is aimed at the system that hands out the doses.

LaBaer’s center exists to convert a blood draw into a test. Measured against that, a study that never reports its own accuracy but ships with a clinical use case already attached is not an unfinished piece of science. It is a finished piece of salesmanship.

Sources

  1. Cell Press Blue (2026) – LaBaer et al., “Pre-existing antibody profiles and vaccine response,” DOI 10.1016/j.cpblue.2026.100088
  2. EurekAlert / Arizona State University – “Why immune responses to vaccines vary from person to person”
  3. ScienceDaily – “AI may know how you’ll respond to a vaccine before you get it” (Aug 21, 2026)
  4. Genetic Engineering & Biotechnology News – AI-identified pre-existing antimicrobial antibody profile and vaccine response
  5. News-Medical – “AI predicts vaccine response based on pre-existing antibody patterns”