Machine learning-driven prediction of brain metastasis in lung adenocarcinoma using miRNA profile and target gene pathway analysis of an mRNA dataset.

Abstract

Brain metastasis (BM) is common in lung adenocarcinoma (LUAD) and has a poor prognosis, necessitating predictive biomarkers. MicroRNAs (MiRNAs) promote cancer cell growth, infiltration, and metastasis. However, the relationship between the miRNA expression profiles and BM occurrence in patients with LUAD remains unclear. We conducted an analysis to identify miRNAs in tissue samples that exhibited different expression levels between patients with and without BM. Using a machine learning approach, we confirmed whether the miRNA profile could be a predictive tool for BM. We performed pathway analysis of miRNA target genes using a matched mRNA dataset. We selected 25 miRNAs that consistently exhibited differential expression between the two groups of 32 samples. The 25-miRNA profile demonstrated a strong predictive potential for BM in both Group 1 and Group 2 and the entire dataset (area under the curve [AUC] = 0.918, accuracy = 0.875 in Group 1; AUC = 0.867, accuracy = 0.781 in Group 2; and AUC = 0.908, accuracy = 0.875 in the entire group). Patients predicted to have BM, based on the 25-miRNA profile, had lower survival rates. Target gene analysis of miRNAs suggested that BM could be induced through the ErbB signaling pathway, proteoglycans in cancer, and the focal adhesion pathway. Furthermore, patients predicted to have BM based on the 25-miRNA profile exhibited higher expression of the epithelial-mesenchymal transition signature, TWIST, and vimentin than those not predicted to have BM. Specifically, there was a correlation between EGFR mRNA levels and BM. This 25-miRNA profile may serve as a biomarker for predicting BM in patients with LUAD.