At Prenosis, we use the term “AI-Biomarker” because it is the most accurate word for what we develop, and it’s grounded directly in how the FDA itself defines a biomarker.
The FDA and NIH, through the BEST (Biomarkers, EndpointS, and other Tools) Glossary, define a biomarker as “a defined characteristic that is measured as an indicator of normal biological processes, pathogenic processes, or responses to an exposure or intervention.” Under that definition, a biomarker can take many forms: a protein concentration in the blood, a tissue staining pattern, a finding on an MRI, a heart rate. And it can serve several distinct functions:
- diagnostic (identifying a disease),
- prognostic (predicting how it will progress), or
- predictive (forecasting how a patient will respond to treatment).
Looking across that landscape, we saw a category that was missing. Every example above is measured by a single instrument, reading a single signal. But biology rarely announces itself through one number. It shows up as a pattern, scattered across labs, vitals, imaging, and clinical notes, that no single test can see on its own.
That is the gap an AI-Biomarker fills: a defined, algorithm-derived characteristic, generated by AI or machine learning, that integrates multiple data types into one reproducible measure of biology.
Put simply, it’s a computed biological signal.
Instead of reading one protein or one image in isolation, an algorithm finds the pattern across all of them, and that pattern does the same job a traditional biomarker does: detecting disease, predicting outcomes, or gauging response to treatment.
We hold AI-Biomarkers to the same three tests the FDA applies to any biomarker.
- They must be defined: precisely described, not a vague output.
- They must be measured: derived through a reproducible algorithm, not a one-off judgment call.
- They must be an indicator: reflecting real underlying biology, not just a correlation a black box happened to find.
Our Sepsis ImmunoScore® is the clearest proof this works. Authorized by the FDA through the De Novo pathway, Sepsis ImmunoScore is a locked, validated algorithm that analyzes routinely collected clinical and lab data to quantify a patient’s sepsis risk. Its inputs and outputs are reproducible across hospitals and patient populations, and by integrating signals of immune dysregulation, physiologic instability, and evidence of infection, it reflects the actual biology of sepsis progression rather than a statistical shortcut. In FDA terms, it functions as both a diagnostic biomarker, helping clinicians recognize sepsis earlier, and a prognostic one, stratifying patients by their risk of deterioration.
That’s why we consider “AI-Biomarkers” as a rigorous extension of a regulatory concept that already exists, applied to the reality that modern biology is too complex for any single measurement to capture alone.
It’s also the foundation everything else at Prenosis builds on: the more precisely we can measure a patient’s biology, the more precisely we can match them to the therapy that will actually help.