Platform
Hundreds of cohorts. Thousands of contrasts.
An AI engine applied to a harmonised multi-omic corpus: every candidate signal is tested against thousands of biological situations. What remains is real, specific and reproducible.
Method
From raw data to a locked panel.
- 01
Harmonise
Each cohort is imported, annotated and normalised. Probes reconciled gene by gene, duplicate patients removed.
- 02
Discover
Our AI models select multi-gene panels, in nested cross-validation to rule out overfitting.
- 03
Map
The signal is projected onto thousands of pathways and contrasts: we know what it measures and what it does not.
- 04
Replicate
The locked panel is tested without retraining on independent cohorts, other laboratories, other technologies.
- 05
Lock
Coefficients, parameters and version are sealed. The score is computed identically at a partner's site.
Mapping a signal
Knowing where a signature lights up.
Three readouts the platform produces for every signal. Illustrations on simulated data.
Rigour
AI proposes, validation decides.
Most signals die along the way. Those that survive have passed these checks. The engine reads transcriptomics and the microbiome, and is extending to proteomics and tissue biopsies.
External validation
No signature is kept without replication on independent cohorts, without retraining.
Specificity maps
Each signal is tested against neighbouring diseases to define its scope.
Confounders first
Age, sex, cell composition, batch effects, generic inflammation.
Against what exists
Systematic comparison with published signatures and marketed tests.
Traceability
Versioned, sealed panels: genes, coefficients, normalisation, cohorts.
Standard technologies
Measurable by RNA-seq or NanoString, with no new instrument.