Integrating single-cell and bulk RNA sequencing with ML algorithms identified MYBL2 as a candidate transcriptomic biomarker for diagnosing laryngeal squamous cell carcinoma (LSCC)

Laryngeal squamous cell carcinoma (LSCC) poses significant clinical challenges due to pronounced tumor heterogeneity. To identify robust, consistent biomarkers across cohorts, we integrated bulk RNA sequencing (RNA-seq) data from GSE127165 and GSE142083 and validated candidate biomarkers in the external cohort GSE13060. Candidate gene prioritization incorporated differentially expressed genes, weighted gene co-expression network analysis (WGCNA) results, and markers of malignant epithelial cells exhibiting high copy number variation (High-CNV) derived from single-cell RNA sequencing (scRNA-seq) (GSE206332). Additional analyses included functional enrichment, immune association (adjusted for tumor purity), quasi-temporal analysis, and CellChat inference to elucidate the biological roles of top genes. The integrated approach identified 254 differentially expressed genes (DEGs), comprising 86 upregulated and 168 downregulated genes. WGCNA revealed a tumor-associated module containing 870 genes. Intersecting these findings with high-CNV malignant epithelial markers reduced the candidate pool to 16 genes, and LASSO regression retained five key features: MYBL2, KRT17, UCHL1, PTHLH, and ODC1. Among these, MYBL2 demonstrated the strongest and most consistent performance. MYBL2 was further associated with proliferative programs, including E2F targets, the G2/M checkpoint, DNA repair, and MYC targets. Gene signatures from these subclusters were enriched in bulk tumors and positively correlated with MYBL2 expression in independent cohorts. The results also indicate that MYBL2-high malignant epithelial states are linked to distinct connectivity patterns within the tumor microenvironment.