Deciphering Reactome-based Biological Pathways in Early and Late Stages of Lung Adenocarcinoma via Association Rule Mining and Genetic Optimization

Lung adenocarcinoma (LUAD) ranks among the most common forms of lung cancer and a leading cause of cancer deaths worldwide. Identifying critical biological pathways in its early and late stages could improve diagnosis and treatment strategies. This study used computational approaches to uncover key mRNAs and related pathways in early- and late-stage LUAD. We downloaded RNA-seq and clinical data for LUAD from the TCGA repository via the GDC portal. After preprocessing, including mapping files to cases, filtering protein-coding genes, averaging expressions, and joining with clinical data, we obtained ۰۱۷ unique LUAD samples with ۱۹,۹۷۵ genes. Excluding cases with unknown stages left ۹۰ valid samples (*).early-stage [I-II], ^° late- stage [III-IV]). Preprocessing involved Z-score normalization. For feature selection, nested cross-validation (۱-fold outer, o-fold inner) with t-tests (p<*,*°) identified so differentially expressed genes. NSGA-II optimization, with self-tuning for subset size (k=۹, AUC), selected ۹. optimal mRNAs. These mRNAs were evaluated using five classifiers (SVM, NB, KNN, RF, DT) in nested CV. SVM performed best, achieving test accuracy of A, AUC of A, F- score of.,, sensitivity of.,Y, and specificity of.,۹. The ۹. mRNAs mapped to ۹۹ Reactome pathways via Ensembl BioMart. Pathway scores were calculated as mean gene expressions, normalized, and discretized (low/medium/high). FP-Growth association rule mining (min support, lift (,)) generated rules for early-stage and late-stage associations. These findings highlight dysregulated pathways in LUAD progression, offering potential biomarkers and therapeutic targets to enhance patient outcomes.

کلیدواژه ها:

Lung adenocarcinoma ، Machine learning ، Pathway analysis