Machine learning model reveals the risk, prognosis, and drug response of histamine-related signatures in pancreatic cancer.

Background: Histamine, a critical inflammatory mediator, is generated by both mast cells and specific tumor cells, and it plays a fundamental role in inflammatory and immune responses. In the current scientific landscape, histamine-related genes (HRGs) and their associated pathways have been validated to be implicated in the development and advancement of cancer. However, the precise role of HRGs in gauging the risk and predicting the prognosis of pancreatic adenocarcinoma (PAAD) remains nebulous.

Methods: We carried out an elaborate data collection endeavor. Transcriptome data along with pertinent clinical information were obtained from the GSE28735, GSE62452, and TCGA-PAAD cohorts. GWAS data were retrieved from the FinnGen Release 11 and eQTLGen databases. For the drug-target Mendelian randomization (MR) analysis, the "TwoSampleMR" (version 0.5.6) R package was employed. The random survival forest (RSF) model was analyzed using the "randomForestSRC (rfsrc)" R package and further elucidated with the help of the "mlr3" package. Somatic mutation analysis and immune infiltration investigations were conducted by means of the "maftools" (v. 2.12.0) R package and "pRRophetic" R software package, respectively. Targeted drug sensitivity analysis was executed using the "oncopredict" and "parallel" packages.

Results: Through a meticulous drug-targeted MR analysis and an exhaustive exploration of transcriptome databases (including 2 GSE combat and TCGA cohort), 20 upregulated differentially expressed genes (DEGs) were identified. The RSF model emerged as the optimal choice, and a 9-HRGs signature was selected to construct a prognostic model that boasted an average C-index of 0.777. In the training and validation cohorts, the model exhibited remarkable predictive prowess, with 1-, 2-, and 3-year prediction accuracies of 0.898, 0.932, and 0.922 in the training set, and 0.909, 0.974, and 0.962 in the validation set, respectively. A higher HRG score was found to correlate with adverse events and the N1 stage. Additionally, it was associated with an increase in M0 macrophages and a decline in CD8 + T cell function. For patients with a low HRG score, several commonly used chemotherapeutic agents, namely Gemcitabine, Carboplatin, Sorafenib, and Oxaliplatin, were more efficacious.

Conclusion: The HRG signature holds the potential to serve as effective biomarkers for diagnosing, predicting the prognosis, and assessing the sensitivity to chemotherapy in PAAD.