H&E-stained IMPRESS triple-negative breast cancer biopsy 939_HE

Holon-S1 Decoding every patient’s hidden biology.

From routine clinical data.

Discuss a collaboration Read the technical report

Live decoding. Drag to explore. You can also focus the biopsy and use the arrow keys.

Our AI helps reveal what makes a tumor unique.

A scan line illustrates feature extraction from a breast cancer H&E slide.
Proliferation Hypoxia Cytotoxic DNA repair Apoptosis Inflammation Interferon Angiogenesis Metabolism Cell stress

Tissue from a biopsy or surgery is prepared on a slide—a routine part of cancer diagnosis.

The animation starts with a routine pathology slide prepared from tissue collected during biopsy or surgery. A scan line indicates image-feature extraction on the tissue. Registered gene maps settle onto the slide, which becomes a colored biological-program map. Program labels illustrate the breadth of coordinated processes that can be studied. The same slide stays in place as six applications appear: expand existing gene panels, discover response biomarkers, explain treatment response, stratify patients for trials, prioritize therapeutic targets and develop clinical tests. These are potential research uses, not validated clinical decisions for this sample. The glass layers illustrate relationships between two-dimensional data.

Making precision medicine more accessible.

Same cancer type. Different treatment needs.

Standard of care

Treatment based on diagnosis
Treatment based on diagnosis Three people have the same cancer type but different tumors, shown by a circle, diamond, and triangle in their chest. Each receives the same blue capsule symbol, shown to their left with an arrow pointing toward them. This is a simplified example of treatment based on cancer type; standard care can also include precision medicine.

Precision medicine

Treatment matched to each tumor
Treatment matched to each tumor The same three people, with the same different tumors, now have different treatment symbols matched to their tumor colors: a capsule, a round tablet, and a vial. Each treatment sits to the left of its patient with an arrow pointing toward them. This illustrates using tumor information to inform treatment choice, not guaranteed treatment matches or outcomes.

Two paths to the same biological insights

Traditional testing for a well-resourced few

  1. High costs
  2. Tissue specimen(if enough available)
  3. Sequencing
  4. Research team
  5. Time
  6. Biological insights

With Holon-S1

  1. Routine slides
  2. Holon-S1
  3. Biological insights
No new gene tests or tissue samples needed for everyone
Biological insights
then Tailored treatment

Case study: Retrospective analysis of osteosarcoma in a real patient.

In 2022, Sid Sijbrandij was diagnosed with osteosarcoma (bone cancer). He spent years exploring his cancer’s biology with a dedicated team and extraordinary resources, and as of 2025 reports no evidence of disease. See what Holon-S1 was able to find from retrospectively analyzing Sid’s original 2022 H&E tissue. (Learn more about Sid’s story)

Each dot will represent a gene’s activity predicted by Holon-S1 from Sid’s 2022 H&E tissue.

Loading predictions for thousands of genes…

Put Holon-S1 to work on your data.

Bring existing slides, a cohort, or a pathology model you want to interpret. Tell us what you want to investigate.

Get in touch

Map gene activity from existing slides.

Predict spatial gene activity from routine H&E, whether molecular profiling was never performed or an earlier panel did not include genes of interest. Investigate new questions without running a new molecular assay.

Discuss your molecular workflow
Validated on
FFPE & fresh frozen slides

Works with archived slides

16 organ types

Pan-cancer coverage

15+ scanners

Robust across your workflow

Discover biomarkers of treatment response.

Most foundation model workflows pool tissue-image features without explicitly modeling their spatial arrangement. Holon-S1 preserves that arrangement and predicts where genes are active. Here we show why this matters in a triple-negative breast cancer cohort: coherent spatial organization of predicted EREG was associated with pathologic complete response to neoadjuvant chemotherapy.

Response to neoadjuvant chemotherapy

In TNBC, stromal tumor-infiltrating lymphocytes (sTIL) are a widely used measure of antitumor immune activity associated with response to neoadjuvant chemotherapy. Here, computational estimates of the percentage of stromal tissue occupied by lymphocytes and plasma cells provide a benchmark for the EREG biomarker.

Precision–recall performance of Holon-S1–predicted EREG spatial gradient and sTIL in discovery and held-out test cohorts Average precision for Holon-S1–predicted EREG spatial gradient was 0.690 in the IMPRESS discovery cohort and 0.754 in the TransNEO held-out test cohort. Average precision for computationally estimated stromal tumor-infiltrating lymphocyte percentage (sTIL) was 0.567 and 0.505 respectively. Higher values indicate better precision–recall performance. IMPRESS discovery cohort Holon-S1–predicted EREG spatial gradient: average precision 0.690, 95% CI 0.513–0.853 IMPRESS discovery cohort sTIL (%): average precision 0.567, 95% CI 0.407–0.778 TransNEO held-out test cohort Holon-S1–predicted EREG spatial gradient: average precision 0.754, 95% CI 0.539–0.908 TransNEO held-out test cohort sTIL (%): average precision 0.505, 95% CI 0.325–0.785
Interactive test

Drag to rearrange predicted EREG activity in a responder’s tissue map while keeping total expression unchanged.

Bulk EREG expressionunchanged
Spatial gradient0.39 · coherent
Marker predictionresponder

The same gene can tell a different story.

Holon reads not only how much EREG is present, but how its expression is arranged across the tissue.

Coherent
Tile-level EREG predictions forming broad, coherent regions across a triple-negative breast cancer biopsy
Magnified tile-level EREG predictions showing a sharply bounded coherent region Sharp boundary and larger concentrations
Fragmented
Tile-level EREG predictions alternating sharply across a triple-negative breast cancer biopsy
Magnified tile-level EREG predictions showing interspersed high and low expression Speckled field, smaller concentrations

EREG has been linked to chemotherapy sensitivity in other cancers. Holon-S1 identified its spatial pattern as a candidate response marker in breast cancer for the first time. Read about our methods and explore more results in the technical report

Discuss a biomarker study

Connect your model’s predictions to biology.

Holon-S1 can combine tissue morphology with predicted gene activity and biological pathways to investigate risk groups in an interpretable manner. Here we use Holon-S1 to analyze biological differences between patient groups stratified by a colorectal cancer prognostic model.

  • Loading biological programs…
Predicted tissue share (%) Each bar is one patient
Loading patient compositions…
Interpret your pathology model

Get in touch.

Discuss your study

Are you a physician-scientist or academic researcher with an idea for using Holon-S1 in your work? Get in touch, we’re seeking collaborators to show off what Holon-S1 can do.

A few places to start