Targeting metal ion transport in pancreatic cancer: A prognostic signature and GraphBAN-predicted therapeutics framework.

Abstract

e16000 Background: Pancreatic ductal adenocarcinoma (PDAC) remains a lethal malignancy with limited therapeutic options. Dysregulation of metal ion homeostasis is implicated in tumor progression and therapy resistance. This study aims to systematically identify metal ion transport-related prognostic genes, construct a reliable risk model, and discover novel therapeutic agents using artificial intelligence. Methods: Transcriptomic data from GEO databases (GSE183795 as training set, GSE28735 as validation set) and single-cell RNA-seq data (GSE155698) were analyzed. Differential expression analysis, functional enrichment, and protein-protein interaction networks were constructed. A prognostic risk signature was developed using LASSO-Cox regression. The tumor immune microenvironment was characterized using CIBERSORT. An innovative GraphBAN model integrated with CNN, ESM, GCN, and ChemBERTa was employed for drug prediction, followed by molecular docking. Single-cell analyses including trajectory inference and cell-cell communication were performed. Results: We identified and validated a five-gene prognostic signature (SLC20A1, SLC39A10, SLC5A3, SLC11A1, SLC4A4) significantly associated with overall survival in PDAC. The risk model demonstrated robust predictive power in both training (3-year AUC = 0.73) and external validation cohorts (5-year AUC = 0.97). High-risk patients exhibited an immunosuppressive microenvironment characterized by M2 macrophage infiltration. GraphBAN prediction and molecular docking identified Elephantin and Sinularin as high-affinity binders to SLC39A10 and SLC4A4, respectively. Single-cell analysis revealed the specific expression dynamics of these genes in macrophage and neutrophil subpopulations and delineated their evolving roles along pseudotemporal trajectories. Conclusions: We established a novel metal ion transport-related gene signature as an independent prognostic indicator for PDAC. Our integrative AI-driven framework successfully predicted candidate drugs targeting this pathway, providing a promising strategy for personalized therapy and highlighting the therapeutic potential of modulating metal ion homeostasis in PDAC.