For clinical trial and translational teams
Tissue patterns that separate responders from non-responders. Motif maps every trial slide into our tissue vocabulary, ranks the phenotypes linked to response, and reports at cohort level. It is built for exploratory biomarker discovery in clinical trials.
Applications
Two ways to use Motif
Trial enrichment
Find who responds, to design the next Phase Ib/II trial.
- Ranks the tissue phenotypes linked to response, with example tiles for each
- Continuous histology scores for exploratory endpoints
Trial recovery
Explain why a failed programme failed, and who it worked for.
- Maps every slide from the programme into the tissue vocabulary
- Cohort-level reports, never per-patient calls
The problem
Trial slides hold more information than the original read-out extracted
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 fixed plan, then the analysis
- 01
Fix the plan
We agree the question, the cohort and the endpoints up front. The analysis plan is fixed and the model frozen before results are seen.
- 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
A study that is not limited by a label set
Supervised tools can only find what has already been annotated, so a study using them is limited by its label set from the outset. Because our 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: ranked response-linked phenotypes with example tiles, continuous histology scores for exploratory endpoints, and a cohort-level report.
Questions
Frequently asked
- What is TileBio Motif?
- Motif applies TileBio’s label-free tissue-language model to clinical trial slides. It maps each slide into a vocabulary of tissue phenotypes, ranks the phenotypes linked to response, and delivers a cohort-level report for exploratory biomarker discovery.
- Does Motif make per-patient calls?
- No. Motif reports at cohort level and never makes per-patient calls. It is research and decision-support software, not a medical device.
- 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.