← Remote ARC

Predict gene expression from H&E slides

Implementation of Integrating spatial gene expression and breast tumour morphology via deep learning (DOI). ST-Net tiles a histology image into 224 × 224 px patches, runs each through a pretrained DenseNet-121, and returns a spatial map of predicted expression for up to 250 genes across the tissue.

Checking…
Tools
★ Inference
Infer ST Map
Upload any breast cancer H&E TIFF or JPEG. Get back a predicted spatial gene expression map — select any of the 250 genes and see where it’s high or low.
Validation
Training Predictions
Predicted vs. actual gene expression for the held-out test patients. Shows how well the model generalised, with Pearson r and MSE per gene.
Data
Training Slides
Browse all 68 H&E slides from the 23-patient dataset. Colour spots by tumor/normal or per-gene expression count.
Technical detail