Machine learning-based identification of biomarkers and drugs in immunologically cold and hot pancreatic adenocarcinomas

Abstract

Abstract Background Pancreatic adenocarcinomas (PAADs) often exhibit a “cold” or immunosuppressive tumor milieu, which is associated with resistance to immune checkpoint blockade therapy; however, the underlying mechanisms are incompletely understood. Here, we aimed to improve our understanding of the molecular mechanisms occurring in the tumor microenvironment and to identify biomarkers, therapeutic targets, and potential drugs to improve PAAD treatment. Methods Patients were categorized according to immunologically hot or cold PAAD subtypes with distinct disease outcomes. Cox regression and weighted correlation network analysis were performed to construct a novel gene signature, referred to as ‘Downregulated in hot tumors, Prognostic, and Immune-Related Genes’ (DPIRGs), which was used to develop prognostic models for PAAD via machine learning (ML). The role of DPIRGs in PAAD was comprehensively analyzed, and biomarker genes able to distinguish PAAD immune subtypes and predict prognosis were identified by ML. The expression of biomarkers was verified using public single-cell transcriptomic and proteomic resources. Drug candidates for turning cold tumors hot and corresponding target proteins were identified via molecular docking studies. Results Using the DPIRG signature as input data, a combination of survival random forest and partial least squares regression Cox was selected from 137 ML combinations to construct an optimized PAAD prognostic model. The effects and molecular mechanisms of DPIRGs were investigated by analysis of genetic/epigenetic alterations, immune infiltration, pathway enrichment, and miRNA regulation. Biomarkers and potential therapeutic targets, including PLEC, TRPV1, and ITGB4, among others, were identified, and the cell type-specific expression of the biomarkers was validated. Drug candidates, including thalidomide, SB-431542, and bleomycin A2, were identified based on their ability to modulate DPIRG expression favorably. Conclusions By combining multiple ML algorithms, we developed a novel prognostic model with excellent performance in PAAD cohorts. ML also proved to be powerful for identifying biomarkers and potential targets for improved PAAD patient stratification and immunotherapy.