Digital decoding tissue microenvironment heterogeneity from spatial proteomics through graph-enhanced transfer learning.

Spatial proteomics studies the protein localization patterns within tissues, providing new insights into cellular ecosystems and disease mechanisms. One critical challenge is its restricted spatial resolution, where each measured spot contains mixtures of cells, obscuring cell-type-specific proteomic signatures. Here, we propose spatial digital cytometry (Spatial-DC), a graph-enhanced transfer learning framework that computationally profiles single-cell-type-resolved signatures from spatial proteomics. Comprehensive benchmarking demonstrates that Spatial-DC outperforms eight state-of-the-art transcriptomics-based methods in estimating the cell-type composition accurately. Applied to diverse data from antibody-based and mass spectrometry (MS)-based technologies, Spatial-DC generates more refined cell-type distribution maps than marker-based distributions and successfully reconstructs proteomic profiles resolved by both spatial and cell types. In a self-collected MS-based pancreatic cancer dataset, Spatial-DC identifies cell-type-specific spatial interactions linked to tumor outcomes. Collectively, Spatial-DC serves as a versatile framework for spatial proteomics, enabling multiscale decoding of tissue microenvironments in a single-cell-type- and spatial-context-resolved manner for downstream analysis.