For pharma, biotech and academic groups
Bespoke computational pathology research. For questions that fall outside Precedent and Motif, our scientific team runs scoped computational pathology studies on your data, using the same tissue-language model.
The problem
Some questions do not fit a product
Existing cohorts are underused
Cohorts from completed studies hold more signal than the original read-out extracted, but re-reading them manually is rarely a justifiable use of pathologist time.
Off-the-shelf tools address predefined tasks
Commercial image-analysis products are trained for defined tasks on defined tissues. If your question is not one of them, no product addresses it.
Building in-house is a multi-year programme
A pathology foundation model requires data engineering, self-supervised training capacity and pathology review. Most teams need the answer rather than the infrastructure.
How it works
A scoped study, start to finish
- 01
Scope and ingest
We agree the question, the cohort and the endpoints up front. Your whole-slide images are processed as they are, with no annotation step and no need to decide in advance what matters.
- 02
Map the cohort
The foundation model clusters your tissue into histomorphological phenotypes learned from the images themselves, giving every slide and every patient a phenotype composition.
- 03
Correlate and report
Phenotype composition is tested against your outcomes, omics or treatment response. You get the findings, the phenotypes behind them, and a clear account of what the data does and does not support.
Why this is different
The same engine as our products
Studio studies run on the tissue-language model behind Precedent and Motif, so every finding rests on phenotypes a pathologist can open and inspect. Because the model learns tissue phenotypes without labels, a study can identify morphology that neither party set out to look for.
The deliverable
What you get
You get: a question and scope agreed in advance, analysis by our scientific team, and a written report of the findings with the phenotypes behind them.
Questions
Frequently asked
- What is TileBio Studio?
- Studio is TileBio’s bespoke computational pathology research service. Our scientific team applies the label-free tissue-language model to your own whole-slide images to answer a question agreed in advance, and delivers a written report of the findings.
- What data do you need from us?
- Whole-slide images (H&E is the usual starting point), plus whatever outcome, treatment or molecular data you want the phenotypes correlated against. We can work with de-identified data under a data processing agreement, and we do not need any manual annotation.
- How large does the cohort need to be?
- Self-supervised learning benefits from scale, but a study does not need a population-scale cohort to be informative. At scoping we will advise whether your cohort can support the question you are asking.
- Who owns the results?
- Ownership of study outputs and any downstream rights are agreed in writing before work starts. We work on both a fee-for-service and a co-development basis, and will structure the study accordingly.
- How long does a study take?
- It depends on cohort size, data readiness and the number of endpoints. We give an indicative timeline at scoping, before any commitment.