miRNA-mRNA Multi-omics Integration in Breast Cancer Staging

The tumor-node-metastasis (TNM) anatomic system is the current clinical standard for breast cancer staging, yet it inadequately captures the molecular heterogeneity that drives disease progression, motivating the development of expression-based biomarker classifiers. We integrated microRNA (miRNA) and messenger RNA (mRNA) expression profiles from 1,081 primary TCGA-BRCA tumors (early-stage, Stages I–II, n = 822; late-stage, Stages III–IV, n = 259) and benchmarked nine machine learning algorithms using a stratified 70/15/15 train–validation–test split with class-imbalance weighting. On the held-out test set, XGBoost, selected after Bayesian hyperparameter optimization on the validation set, achieved AUC = 0.687 (95% CI: 0.622–0.748), accuracy = 79.8%, and good model calibration (Hosmer-Lemeshow p = 0.31), significantly outperforming mRNA-only (AUC = 0.654; p = 0.028) and miRNA-only (AUC = 0.612; p = 0.007) models. Differential expression analysis identified 15 significant miRNAs and 194 significant mRNAs (FDR < 0.05), and regulatory network analysis revealed three co-expression modules governing epithelial-mesenchymal transition, metabolic reprogramming, and immune evasion, with 15 hub miRNAs regulating 97–516 predicted targets. Multi-omics integration captures progression-associated molecular signatures beyond anatomic staging and could, pending external validation, complement conventional TNM staging by providing a molecularly informed risk-stratification layer; the framework provides a reproducible basis for exploratory translational biomarker discovery.