Prediction of single cell expression from low-plex immunofluorescence images for guiding precision oncology.
Machine learning can now predict detailed cell-level molecular patterns from cheaper, simpler cancer images, potentially making advanced tumor profiling more accessible.
The authors developed Image2Count, a deep learning model using Graph Neural Networks trained contrastively to predict single-cell molecular expression profiles from low-cost 4-marker immunofluorescence images. Validated across ovarian, colorectal, and non-small cell lung cancer datasets, the model captures biologically meaningful tumor immune microenvironment patterns and pathway enrichment concordant with true high-plex measurements, suggesting a practical path to clinical TIME stratification without expensive spatial omics assays.
What the study was
- Study design
- Computational/AI model development with multicohort validation
- Population
- Ovarian cancer (GeoMx 72-plex), colorectal cancer (t-CyCIF 25-plex), non-small cell lung cancer (CosMx 960-plex) tissue datasets
- Category
- Diagnostics
- Maturity
- Validated
- Journal
- NPJ Precision Oncology
Why it surfaced
Technically novel AI framework enabling high-plex molecular profiling from 4-marker standard pathology images; validated in three cancer types; potentially democratizes spatial omics for clinical decision support in immuno-oncology.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.