Model infers spatial neighborhoods in dissociated single cells

In brain tissue transcriptomic datasets, researchers tested ProxiNet, a computational framework designed to infer cellular proximity from dissociated single-cell RNA sequencing. While spatial transcriptomics captures intact tissue architecture, dissociated single-cell methods lose spatial context. ProxiNet trains on spatial reference data to learn expression signatures of pairwise cell proximity and transfers those relationships to dissociated cells.

The framework predicted proximity across brain regions and across different spatial technologies. In datasets with known coordinates, it recovered anatomical structures and reproducible tissue organization. When applied to dissociated single-cell data, ProxiNet revealed cellular neighborhoods and candidate cell communication programs. In an Alzheimer dataset, the model detected age-associated neighborhood remodeling, including an altered neighborhood at 8 months marked by distinct astrocyte and neuronal transcriptional states.