AI-powered clinical decision support for breast cancer
From snapshots to disease modeling.
SignPost Cancer Analytics is developing BreastDefense™, an AI-powered biological intelligence platform designed to read whole-genome cell-free DNA signals from a routine blood draw, and read them again with each draw, to model each patient’s tumor phenotype, trajectory, and velocity over time.
A single draw is a point, and a point has no direction. Cell-free DNA dynamics appear only in the change between draws: whether the disease is stable, changing, responding, or returning.
Patient journey
Care is a sequence of decisions, and each one waits on an answer that arrives late.
| Phase | Decision | Today | Wait |
|---|---|---|---|
| Detection | Is there disease? | mammography, then biopsy | moderate wait |
| Treatment | Is the treatment working? | imaging, weeks to months later | long wait |
| Is any disease left? | pathology and risk scores | moderate wait | |
| Surveillance | Is it coming back? | symptoms, examination, or mammography | long wait |
| Recurrence | Has the tumor’s biology changed? | re-biopsy, when feasible | moderate wait |
Tissue biopsy characterizes the tumor at one moment, usually the one in which it was removed. Imaging confirms a change only once it is large enough to see. Between those two points a patient may spend months on a regimen that is not working, taking on its toxicity without its benefit.
Blood-based testing was meant to close that distance. Most assays read one class of signal and are blind to the others, and the signal itself is scarce: when disease burden is low, tumor-derived DNA is a small fraction of the cell-free DNA in plasma, most of which comes from normal blood cells. Any blood test has to work under that constraint.
Today’s tools tell a clinician what is present right now. BreastDefense is designed to show where a patient’s disease is heading.
The opportunity is not just earlier detection. It is earlier biological understanding.
The BreastDefense engine
One blood draw. Three signals. One model of the disease.
- 01
Draw
A routine blood draw, with no new procedure and no new device.
- 02
Read
Methylation, fragmentomics, and mutations, profiled together from the same cell-free DNA in a single sequencing run.
- 03
Model
BreastDefense’s AI engine integrates the signals into one longitudinal model of the disease, and revises it with every draw.
- 04
Return
A unified biological report, returned to the clinician and shared with the patient, to inform decisions that today wait on imaging.
Methylation, fragmentation, and mutation are complementary signals, and all three can be read from the same fragment of cell-free DNA. Alone, none can say where a fragment came from. Integrated, they let the engine estimate, fragment by fragment, the probability that the DNA is tumor-derived.
Methylation
The tissue the DNA came from, and whether that tissue carried the methylation changes seen in cancer.
Fragmentomics
Fragment length and end pattern, which reflect how the DNA was packaged in the nucleus of the cell it came from.
Mutations
Which of the tumor’s own sequence changes are present in plasma, and how their share of the DNA moves from draw to draw.
The framework is built to incorporate new molecular signals as sequencing advances. BreastDefense is designed to be read alongside mammography and the patient’s clinical record, so the molecular model and the clinical picture inform each other.
BreastDefense adds a biological intelligence layer that complements existing imaging systems, screening programs, and clinical workflows rather than replacing them.
Why now
Two technologies had to converge before BreastDefense could exist.
Whole-genome sequencing that reads three signals from one run
Commercial-scale whole-genome sequencing now reads methylation, fragmentomics, and mutations from the same molecules in a single run, at a cost a clinical lab can carry, and produces more data per patient than any panel. Public datasets of this kind barely exist yet.
AI-powered models that weigh it all at once
Multimodal deep learning can now integrate several classes of signal per patient rather than scoring each on its own, and follow how the disease changes from draw to draw.
BreastDefense is built on both by a team with decades in diagnostics, and will be trained on proprietary data rather than adapted from public models.
Validation
Validation, one stage at a time.
Analytical validation comes first. No result is claimed before its stage is reached.
Academic partner
Specimen access and genomics core capacity, under a signed institutional research agreement.
- 01 ANALYTICAL VALIDATIONCURRENT
Establish that the assay is reproducible and that the model recovers tumor-derived signal in plasma from patients whose tumors have been characterized.
- 02 CLINICAL VALIDATIONNEXT
Follows analytical validation, on a regulatory path as software as a medical device.
Team
Oncology, computation, and operating experience.
Board of Directors
