Tissue reassembly with generative AI

The spatial arrangement of cells plays a critical role in determining their functions and interactions within tissues. However, single-cell RNA sequencing (scRNA-seq) dissociates cells from their native tissue context, resulting in a loss of spatial information. Here, we show that complex tissue structures can be reassembled from the gene expression profiles of dissociated cells. To achieve this, we developed LUNA, a generative AI model that reconstructs tissues conditioned solely on gene expressions of cells by learning spatial priors over existing spatially resolved datasets. We show that LUNA effectively reconstructs slices from the MERFISH whole mouse brain atlas with over 1.2 million cells, including cells from cell types never seen during model training. Applying LUNA to the mouse central nervous system scRNA-seq atlas, we show that LUNA is applicable for de novo generation of tissue structures. Additionally, LUNA can infer locations of nuclei lost during cell profiling with Slide-tags technology and correctly places cells belonging to spatially distinct compartments in a human metastatic melanoma sample. We envision that AI-driven tissue reassembly can help to overcome current technological limitations and advance our understanding of tissue organization and function, paving the way towards virtual tissue models.