backed byCombinator

POSTMAN

In silico spatial proteomics from H&E

Request Early Access

From H&E to 180+ spatial protein maps

POSTMAN takes a standard H&E image and predicts virtual multiplex immunofluorescence across 180+ protein biomarkers, without touching the tissue.

Trained on a dataset of 20,000 slides of paired same-section H&E + mIF data, the largest in the field. Spanning immune, structural, and functional markers across pan-cancer, multi-tissue, multi-site cohorts.

Why this matters

$5-10K per slide

Multiplex immunofluorescence is expensive, takes days, and damages the tissue in the process.

Samples are scarce

Most clinical samples are degraded, exhausted, or impossible to re-acquire, especially in retrospective cohorts.

Massive blind spot

Millions of patients. Decades of trials. Spatial proteomics on almost none of them.

180+protein biomarkers
20Kpaired training slides
25+tissue types
28diseases

1B parameter generative vision model. Fine-tunable to your indications and target panels.

Validation

We validate POSTMAN not just on pixel-level concordance with held-out ground truth, but on the metrics that matter: biomarker prediction, patient stratification, and clinical outcome prediction using imputed data.

Applications

Virtual Target Screening

Predict drug target expression across H&E slides where the drug is not yet approved.

Patient Stratification

From one H&E slide, predict all targetable markers simultaneously to guide treatment selection.

Population-Scale Drug Repurposing

Virtually stain entire archived tissue microarray banks to identify new therapeutic opportunities.

Request early access

Get POSTMAN running on your data.

Request Access