Integrated multiomics machine learning and mediated Mendelian randomization investigate the molecular subtypes and...

Lung squamous cell carcinoma (LUSC) lacks specific early diagnostic markers. Given the critical role of 5'-Nucleotidase Ecto (NT5E) in immune evasion and therapy resistance of cancer cells and the involvement of Dual Specificity Phosphatase 4 (DUSP4) in tumor cell proliferation through inhibition of the ERK signaling pathway, incorporating NT5E and DUSP4 into the consensus machine learning signature (CMLS) system in this study holds significant potential for investigating the early diagnosis and immune microenvironment of LUSC. The objective of this study was to explore the prognostic targets of LUSC.

Employing integrated algorithms enhances the ability to identify molecular subtypes and key features from multiple perspectives. A combination of 10 clustering algorithms and multi-omics data from LUSC patients, merged with 10 machine learning algorithms, was used to analyze and identify high-resolution molecular subsets and develop a CMLS. Mediated Mendelian randomization (MR) was utilized to explore mediations between immune cells and metabolites associated with CMLS.

Cluster 1 demonstrated elevated infiltration of immune and stromal components, indicating an immunosuppressive microenvironment predominantly driven by tumor-associated macrophages or other inhibitory cells. In contrast, Cluster 2 displayed a metabolism-driven phenotype associated with improved prognosis. Mediated MR provided further insights into the causal relationships among CMLS, macrophages, and metabolites in LUSC. Validation of the RAS-RAF-MEK-ERK signaling pathway in conjunction with CMLS reinforced the immune characteristics of CMLS.

The integration of CMLS with multi-omics offers a robust framework for predicting prognosis, elucidating the causal interactions between the immune microenvironment and metabolic reprogramming in LUSC, and identifying patient subgroups likely to benefit from immunotherapy.