Pathology foundation models
New biology, found at population scale.
TileBio builds foundation models that learn tissue morphology directly from unlabelled whole-slide images, for preclinical toxicology and exploratory biomarker discovery.
- Whole-slide images
- 2M+Available for model training and validation
- Data partners
- NHSNHS Greater Glasgow & Clyde and the NHS West of Scotland Innovation Hub
- Method
- PatentedAnd published in Nature Communications
- Seed round
- £2.261MLed by Twin Path Ventures
Products
For toxicology, clinical trials and bespoke research
For toxicology and safety assessment teams
Morphology search, slide triage and treatment-effect analysis for toxicology archives and studies.
ExploreFor clinical trial and translational teams
Exploratory biomarker discovery for clinical trials: the tissue phenotypes linked to response, with example tiles for each.
ExploreFor pharma, biotech and academic groups
Bespoke computational pathology research: scoped studies on your own cohort, run by our scientific team.
Explore
Approach
A model that learns tissue structure without labels
Most pathology AI can only find the categories it was trained on. Our model learns a vocabulary of tissue phenotypes from unlabelled slides and reads each slide as a spatial composition of them.
Partners
NHS data partnerships and a closed seed round
TileBio holds a data partnership with NHS Greater Glasgow & Clyde and the NHS West of Scotland Innovation Hub, and has closed a £2.261M seed round led by Twin Path Ventures.
- NHS Greater Glasgow & Clyde
- NHS West of Scotland Innovation Hub
We have an announced collaboration with Terrain Life Science and are engaging with global pharmaceutical companies on toxicology and discovery pilots.
Discuss a study
Tell us about your cohort and the question you want to answer, and we will advise whether our model is a good fit.