AI can create super-resolution images from coarse transcriptomic data

Spatial transcriptomics allows researchers to visualize what genes are being expressed where in a piece of tissue. But unfortunately the resolution of spatial omics is not always as fine as we would like. Using advanced generative frameworks, AI can now improve the resolution of these techniques, allowing researchers to see a more detailed view of tumors.

For example, Owkin’s MISO achieves near single-cell resolution predictions from H&E slides alone. AI-trained models can even extend spatial omics to millions of routine pathology slides by reconstructing spatial omics data from images of patient tissue, widely available in the clinic. This innovation transforms existing clinical samples into a virtual treasure trove for cancer research. This could prove valuable for clinical applications, including molecular phenotyping and biomarker identification, offering a practical solution to bridge the gap between routine histopathology and complex molecular profiling techniques that are not yet suitable for widespread clinical or research implementation.