Detecting spatially-resolved cell-cell communication using deep learning

Emerging evidence suggests tumor microenvironments promote cancer-cell growth through cooperation among cancer cells and stroma recruitment, suggesting involvement from cell-cell communication. Spatial transcriptomics measures spatially-resolved transcriptomic profiles in tissue, providing tumor microenvironment insight. We developed a new computational method to detect communication at spot and single-cell resolution in spatial transcriptomics data that uses graph attention networks to learn spatially-resolved signals. We applied our tool to pancreatic cancer and identified differential enrichment of signals within and between the tumor and stroma across tissue regions, demonstrating that our tool will help advance the field of communication detection.