A cross-platform validated biomarker signature for mucoepidermoid carcinoma identified by integrated WGCNA and machine learning.

Mucoepidermoid carcinoma (MEC), the most prevalent malignant salivary gland cancer, remains incompletely characterized at the molecular level. An integrated analytical pipeline was employed, combining differential expression analysis, weighted gene co-expression network analysis (WGCNA), resampling-based stability selection, machine learning, and cross-platform validation to identify robust diagnostic biomarkers for MEC. Two public microarray datasets (GSE169753 and GSE262344) were harmonized to form a discovery cohort of 49 samples (39 MEC and 10 normal salivary gland tissues) comprising 19,565 genes. Candidate biomarkers were identified by intersecting WGCNA hub genes with differentially expressed genes identified within the same 8,000-gene expression subset and further refined through 100 resampling iterations, retaining genes selected in at least 60% of the iterations. Elastic-net, random forest, and linear support vector machine models were trained, and the final gene signature was externally validated in an independent RNA-seq cohort (GSE282430). Eleven genes were identified: HTN3, MUCL1, GPR45, PLIN5, PAIP2B, PART1, CD109, PIP, ABCC6, CSN1S1, and KLK1. This signature demonstrated strong discrimination between MEC and normal salivary gland tissue and showed good cross-platform reproducibility in an independent RNA-seq cohort. Functional enrichment analyses revealed downregulation of pathways associated with normal sensory and secretory salivary functions in MEC. These results support the potential diagnostic utility of an 11-gene biomarker panel for MEC.