For lung adenocarcinoma (LUAD), patients of different stages have strong heterogeneity, and their overall prognosis varies greatly. Thus, exploration of novel biomarkers to better clarify the characteristics of LUAD is urgent. Multi-omics information of LUAD patients were collected form TCGA. Three independent LUAD cohorts were obtained from gene expression omnibus (GEO). A multi-omics correlation analysis of METTL5 was performed in TCGA dataset. To build a METTL5-associated prognostic score (MAPS). Spathial and random forest methods were first applied for feature selection. Then, LASSO was implemented to develop the model in TCGA cohort. The prognostic value of MAPS was validated in three independent GEO datasets. Finally, functional annotation was conducted using gene set enrichment analysis (GSEA) and the abundances of infiltrated immune cells were estimated by ImmuCellAI algorithm. A total of 901 LUAD patients were included. The expression of METTL5 in LUAD was significantly higher than that in normal lung tissue. And high expression of METTL5 indicated poor prognosis in all different stages (P < 0.001, HR = 1.81). Five genes (RAC1, C11of24, METTL5, RCCD1, and SLC7A5) were used to construct MAPS and MAPS was significantly correlated with poor prognosis (P < 0.001, HR = 2.15). Furthermore, multivariate Cox regression analysis suggested MAPS as an independent prognostic factor. Functional enrichment revealed significant association between MAPS and several immune components and pathways. This study provides insights into the potential significance of METTL5 in LUAD and MAPS can serve as a promising biomarker for LUAD.
Keywords: METTL5; lung adenocarcinoma; machine learning; prognosis; rRNA methylation.
Copyright © 2021 Sun, Fei, Zhang, Wang, Yang, Guo, Yang, Wang, Xue, Gao and He.