Deep learning-based enhancing spatial biological context parsing.

CroSP, a multi-view graph autoencoder, learns a shared latent space from gene expression, spatial coordinates, and tissue morphology, improving spatial domain identification across ST datasets.

• Why it matters: Spatial transcriptomics data are noisy and heterogeneous, making it difficult to reliably parse tissue structure and spatial relationships from multimodal measurements.
• What they did: CroSP uses cross-view contrastive learning and metric learning to integrate gene expression profiles, spatial locations, and histological information into low-dimensional embeddings.
• The result: Across multiple platforms and spatial resolutions, CroSP improves spatial domain identification and delineates brain and olfactory bulb structures, and supports downstream analyses in human breast cancer tissue.

AI summary of the abstract · may contain errors · not medical advice · read the paper.