Predicting spatial gene expression from H&E images offers a scalable alternative to experimental spatial transcriptomics, but modeling spatial context remains a challenge. The arXiv preprint introduces HistoLoop, a looped transformer that performs iterative contextual inference to improve these predictions.
The abstract notes that while progress has been made in spatial context modeling, the question of how to construct subsequent context from spot states is still open. HistoLoop directly targets this issue by looping over context construction, allowing the model to refine its predictions iteratively.
As a new method, HistoLoop aims to push forward the accuracy and practicality of gene expression prediction from routine histology slides. The paper is available on arXiv, though the abstract is truncated; full details of the architecture and results are in the preprint.