基于机器学习构建胰腺癌淋巴管生成诊断模型OA

Objective:Pancreatic ductal adenocarcinoma(PDAC)is characterized by an insidious onset and a high propensity for early metastasis.Lymphangiogenesis(LYM)not only provides a pathway for tumor dissemination but also plays a critical role in tumor microenvironment(TME)remodeling and the establishment of immune tolerance.Given the limitations of single biomarkers,this study aims to identify key LYM-related genes in PDAC using machine learning algorithms and construct a multi-gene diagnostic model.Furthermore,the biological roles of these genes in prognostic evaluation,TME remodeling,and prediction of targeted drug sensitivity were investigated. Methods:Gene expression datasets from the Gene Expression Omnibus(GEO)database,including GSE28735,GSE62452,GSE71729,and GSE183795,were integrated as the training cohort,while data from The Cancer Genome Atlas(TCGA)and Genotype-Tissue Expression(GTEx)databases were used as independent external validation cohorts.Differentially expressed LYM-related genes were identified,and dimensionality reduction and diagnostic model construction were performed using univariate logistic regression combined with least absolute shrinkage and selection operator(LASSO),support vector machine-recursive feature elimination(SVM-RFE),and random forest(RF)algorithms.Model performance was evaluated using the area under the curve(AUC)of the receiver operating characteristic(ROC),nomogram analysis,and decision curve analysis(DCA).The Gene Expression Profiling Interactive Analysis(GEPIA)database was used to validate core gene expression patterns and overall survival(OS).Gene set enrichment analysis(GSEA)was performed to explore associated biological pathways.Immune infiltration characteristics and correlations were evaluated using single-sample gene set enrichment analysis(ssGSEA)and cell-type identification by estimating relative subsets of RNA transcripts(CIBERSORT).The half-maximal inhibitory concentration(IC50)values of therapeutic agents were also predicted. Results:A total of 836 differentially expressed genes were identified,and 69 candidate genes were obtained by intersecting these genes with 466 LYM-related genes.These genes were mainly enriched in extracellular matrix remodeling-related pathways.Through integrated dimensionality reduction using multiple machine learning algorithms,a gene diagnostic model consisting of serine/threonine/tyrosine kinase 1(STYK1),periostin(POSTN),NADPH oxidase 4(NOX4),endothelial cell-specific molecule 1(ESM1),and lymphatic vessel endothelial hyaluronan receptor 1(LYVE1)was established.The model achieved an AUC of 0.908 in the training cohort,and nomogram and decision curve analyses demonstrated favorable clinical utility.External validation showed that four risk genes remained highly expressed in independent cohorts.The single-gene diagnostic performance of STYK1 showed an internal AUC of 0.879 and an external AUC of 0.902.Survival analysis demonstrated that only high STYK1 expression was significantly associated with shortened OS(P<0.05).GSEA revealed that high-STYK1 expression group was mainly enriched in pathways related to the cell cycle and extracellular matrix receptor interaction.Immune infiltration analysis showed increased proportions of regulatory T cells and M0/M2 macrophages and decreased proportions of CD8+T cell infiltration in PDAC tissues.STYK1 expression was positively correlated with proportions of M0 macrophage infiltration(r=0.41)and negatively correlated with proportions of CD8+T cell infiltration(r=-0.37).Drug sensitivity prediction indicated that the high-STYK1 expression group exhibited significantly lower IC50 values for targeted agents including AZD6482,BMS-754807,Tozasertib,and AZD8055(all P<0.001). Conclusion:The 5-gene model consisting of STYK1,POSTN,NOX4,ESM1,and LYVE1 effectively differentiates PDAC from normal samples.High STYK1 expression is associated with poor prognosis and an immunosuppressive microenvironment,while also indicating potential sensitivity to specific targeted therapies.Therefore,STYK1 may serve as a biomarker for risk stratification and individualized therapeutic selection in PDAC.