A deep learning framework, CroSP, based on a multi-view graph autoencoder that integrates gene expression profiles, spatial locations, and tissue morphology information into a low-dimensional latent representation that supports the use of CroSP for spatial biological analysis in healthy and diseased tissues.