Tissue Reconstruction with Generative AI

LUNA, a generative AI model, accurately reconstructs tissue architecture from dissociated gene expression profiles, demonstrated by inferring spatial locations in over 1.2 million cells from the mouse brain atlas.

• Why it matters: Understanding tissue organization is crucial for insights into cell function and disease, but current methods often dissociate cells from their native spatial context, limiting biological interpretation.
• What they did: LUNA learns spatial priors from spatially resolved transcriptomics datasets and applies this knowledge to reconstruct tissue architecture, including unseen cell types and pathological samples, validated through multiple datasets.
• The result: LUNA's reconstructions reveal disease-associated tissue alterations in a Parkinson's disease mouse model, establishing generative tissue reconstruction as a tool to overcome technological limitations and deepen tissue biology understanding.