DGAT: A Dual-Graph Attention Network for Inferring Spatial Protein Landscapes from Transcriptomics

To address the lack of spatially resolved proteomic measurements in standard spatial transcriptomics (ST) technologies, we generated this in-house MPM spatial transcriptomics and proteomics dataset and used it to train and evaluate DGAT (Dual-Graph Attention Network), a deep learning framework for imputing protein expression from ST data.

DGAT integrates transcriptomic, proteomic, and spatial information using graph attention networks and jointly reconstructs mRNA and protein profiles through multi-task learning. This dataset serves both as a biological resource for understanding the tumor immune microenvironment in MPM and as a benchmark for spatial protein inference. DGAT’s predictions on this dataset enabled improved identification of immune phenotypes and tumor–stroma spatial organization, with implications for biomarker discovery and therapeutic targeting in mesothelioma.