Spatial Gene Expression Prediction
Overview
The field predicts gene activity at tissue spots or individual cells from H&E images, with the goal of extending costly spatial assays to more samples. Approaches combine pathology foundation-model features with local and global context from transformers or graphs, and add cell-type, gene-program, or ontology priors; some also use retrieval or generative models and sparse measured expression. Methods are commonly evaluated on HEST-1k and spatial platforms such as Visium and Xenium, using expression correlations and measures of spatial-pattern recovery, including held-out-slide or held-out-organ tests. A key challenge is generalizing across patients, tissues, and platforms while preserving spatial heterogeneity and biologically meaningful gene relationships, rather than optimizing pointwise accuracy alone.
Latest papers
39 papers6 papers in the last four weeks, up 20% on the four weeks before. 0.0% of all new papers.