AI-assisted multi-omics and virtual screening identify the IGF2BP2-SGMS2 axis for PDAC precision immunotherapy

Pancreatic ductal adenocarcinoma (PDAC) is characterized by an immunosuppressive tumor microenvironment (TME), resulting in primary resistance to immune checkpoint therapy and highlighting the need for precision therapeutic targets and effective drug design strategies. Here, we developed an AI-assisted workflow integrating single-cell multi-omics analysis and machine learning (ML) prognostic modeling to identify actionable vulnerabilities in PDAC. Analysis of 24 clinical PDAC single-cell RNA sequencing (scRNA-seq) samples revealed an IGF2BP2-high malignant cell subpopulation as a key driver of TME immunosuppression, with IGF2BP2 and SGMS2 synchronously upregulated during malignant progression. A robust ML signature was constructed and validated for patient stratification, and multidimensional network analysis prioritized the IGF2BP2-SGMS2 axis as a druggable therapeutic target. Virtual screening of FDA-approved compounds identified montelukast as a selective functional inhibitor of IGF2BP2, suppressing SGMS2-mediated lipid raft remodeling and membrane PD-L1 enrichment without altering IGF2BP2 expression. In vitro validation showed that montelukast restored CD8 + T cell cytotoxicity, increased pro-inflammatory cytokine secretion, and synergized with PD-L1 monoclonal antibody to induce tumor cell apoptosis. This study establishes an AI-assisted framework for target discovery and identifies the IGF2BP2-SGMS2 axis as a promising vulnerability for precision immunotherapy in PDAC.