Models built for proprietary biomarkers, with a clear path to the clinic
A proof of concept on archived tissue, validated against pathologist scores and across sites and scanners.
Run retrospectively on clinical trial samples and tested against patient outcomes to build the evidence dossier.
Used in the clinic as a laboratory-developed test or regulated diagnostic, including companion diagnostics.
Example use cases
Some IHC stains are hard to score consistently by eye. We have built over 50 models that score them automatically, including for antibody-drug conjugates (ADCs) and radiopharmaceuticals. Models locate staining by cell compartment, grade intensity, and calculate scores such as TPS and CPS.
F1 > 0.95
Positive vs. negative scoring of proprietary biomarkers
F1 > 0.85
Scoring across intensity levels, from none to strong
F1: agreement with reference scores, 1 is perfect. Source: Aignostics internal analyses
Trained on H&E slides paired with IHC and gene panel results, our models predict biomarker expression and mutation status from H&E alone. Validated models can pre-screen patients for trials or serve as diagnostics.
AUC 0.86
Biomarker expression vs. pathologist scores. 0.94 with clinical data added
AUC 0.71–0.76
EGFR (0.76), STK11 (0.71) and TP53 (0.75) mutations vs. gene panel sequencing. Trained on 2,000+ clinical H&E images, tested on independent data
AUC: 0.5 is chance, 1 is perfect. Source: Aignostics internal analyses [citation to add]
Uncover new biology with multimodal data and AI
We assess novel targets and biomarkers across clinical, spatial and omics data, fusing insights from multiple levels of biology, including:
Fill out the form to discuss custom biomarker models or target and biomarker identification.