Histology AI
The Hetairos programme: companion models that predict the methylation class from the slide a laboratory has already cut.
The Hetairos programme
Bringing precision classification within reach of every laboratory
Every tumour that reaches a pathology department is already on an H&E slide.
Our histology models read that slide and predict which methylation class the tumour most likely belongs to. A laboratory with a scanner and no molecular platform can still reach a molecularly defined answer; a laboratory that has one can see which cases genuinely need it, and reach the answer days sooner in both cases.
- Works from routine H&E: no arrays, no sequencers, no additional tissue
- Extends molecularly defined classification to laboratories without molecular infrastructure
- Where molecular testing is available, narrows the differential and guides which tests to run
- Predictions come with a confidence score and a heatmap showing which regions drove them
- Methylation profiling remains the reference standard: the model points towards it rather than replacing it
Hetairos Ἑταῖρος, Greek for companion. These models are built to sit alongside the pathologist in the first diagnostic phase, triaging cases, narrowing the differential and pointing to which molecular tests are worth running.
The first model · central nervous system
CNS methylation subtypes predicted from H&E
Hetairos CNS predicts central nervous system tumour methylation subtypes from standard H&E whole-slide images. It was trained and validated on a multi-centre cohort, published in Nature Cancer in 2026, and is now in early access.
The model distinguishes 102 methylation-based subtypes and was trained on 9,606 patients across eleven centres on four continents. On high-confidence predictions it reached 87% accuracy, outperformed five board-certified neuropathologists reading the same slides, and reduced time to a subtype-level answer from roughly twelve days to roughly twelve minutes.
The programme
One companion model per entity we already classify
Hetairos models are developed entity by entity, following the methylation classifiers rather than running ahead of them: a histology model is only meaningful where a validated methylation reference exists to predict against.
Programme models
Early access
Pipeline
More Hetairos models are in development and will join the programme as they are validated.
Entity roadmap
Priority
Central nervous system
Methylation classifier available and in routine research use. Hetairos, our histology model for CNS tumours, is available in early access.
Next
Sarcoma
Methylation classifier available. Histology model coming soon.
Later
Skin
Methylation classifier available in preliminary form. Histology model coming soon.
Shared foundations
One pipeline behind every Hetairos model
The models differ; the machinery around them does not. Building it once means each new entity starts from a validated pipeline rather than from scratch.
Slide ingest
Whole-slide images from the scanner formats laboratories already use, uploaded through the platform.
Preprocessing
Tissue segmentation, tiling, and filtering of artefacts and out-of-focus regions, versioned, so a result can be reproduced.
Inference
GPU-backed inference behind a job queue, so a slide can be submitted and collected rather than waited on.
Reporting
Predictions with confidence and heatmaps showing the regions that drove the call, for review by a pathologist.
Getting access. Hetairos CNS is in early access. If you would like to try it on your own slides, or take part in the programme, write to us at support@epignostix.com.
Hetairos models are for research use only. They are not medical devices, are not intended for diagnostic procedures, and do not replace methylation profiling or histopathological assessment by a qualified pathologist.
Collaborate with us
We are looking for slide cohorts and clinical partners
Histology models need scanned slides paired with methylation ground truth, across scanners, stains and centres. If your department has such a cohort, or works on an entity you would like to see covered, we would like to talk.
