Evidence-based data only
Models learn only from real, documented human samples: origin, consent and licence on record. No synthetic data.
For medical laboratories
The methylation signal of circulating DNA, analysed by a consensus of models.
You draw the sample and run the assay. Tafahi analyses the methylation signal of circulating tumour DNA and returns, for each sample, the verdict of several independent models and their consensus.
Three rules, no exceptions
A result is only useful if it can be checked. Tafahi is built on three rules.
Several models must agree
Independent model families analyse every sample. A signal is reported only when they converge; otherwise the result is declared inconclusive.
Everything is traceable
Every result traces back to the model, its training data and their original source, with versions and checksums.
How it works
- 1
You upload the data
The methylation files produced by your platform, through the web console or the API.
- 2
We analyse
Quality control, harmonisation, a vote from each model and a calibrated consensus, with control samples in every batch.
- 3
You receive the report
Signal, likely tissue of origin, calibrated probability and full trace, ready to include in your own report.
Part of your offer
You keep the relationship with the physician and the patient, and you set your price. Tafahi provides only the signal analysis, billed per sample.
- Per-analysis pricing, decreasing with volume
- Upload through the web console or by API from your laboratory system
- Signed report with the full trace of models and data
- Quality control and control samples in every batch
Research programme: your data in exchange for analyses
Researchers and laboratories can join the programme. You share documented samples with their clinical follow-up; we analyse patients from your cohorts free of charge.
- Protocol approved by an ethics committee, with consent covering data sharing
- Free analyses of patients from your cohorts, for research purposes
- Co-authorship of publications based on the shared data
- Independent, multicentre validation of the signal