Multiplex-GCN Predicts Clinically Actionable Tumor Microenvironment Subtypes in Pancreatic Cancer

A multi-layered Graph Convolutional Network identifies seven PDAC tumor microenvironment–guided subtypes, grouped into two ecological tumor families, with distinct immune and stromal features.

• Why it matters: PDAC has ~13% five-year survival and limited options, while TME-driven progression and drug resistance lack well-characterized molecular subtype structure.
• What they did: The approach builds a TME reference atlas from single-cell RNA-seq, estimates immune/non-immune proportions in TCGA bulk RNA-seq via in silico cytometry, then integrates multi-omics and clinical data for unsupervised clustering on GCN embeddings.
• The result: Subtypes show subtype-specific molecular programs and significant overall survival differences across ecological families and subtypes, supporting biologically informed patient stratification.