Goal: predict spatial gene expression from H&E slides What you'll need: a Seurat object for 10x visium data, or similar. We'll need an H&E slide, and spatial transcript counts.
First, we'll extract the necessary data from the Seurat object. The H&E slide is extracted using RDS_to_PNG.R and the transcript counts are extracted using GetCountsForPatches.R
To make predictions from our image, we'll use features calculated from KimiaNet. The code within RunKimiaNet.ipynb splits the image into patches and calculates features using this pipeline.
Finally, we predict individual transcript expression levels, using Spatial Random Forest, using KimiaNet features in RandomForest.Rmd.