The Role and Machine Learning Analysis of Mitochondrial Autophagy-Related Gene Expression in Lung Adenocarcinoma

Binyu Wang ¹ Di Liu ² Danfei Shi ¹ Xinmin Li ³ Yong LI ^1,4*

• ¹ The First Affiliated Hospital of Huzhou University, Huzhou, China
• ² Huzhou Maternity and Child Health Care Hospital, Huzhou, China
• ³ Chongqing Hospital of Traditional Chinese Medicine, Chongqing, China
• ⁴ The First People's Hospital of Huzhou, Huzhou, China

Objective: Lung adenocarcinoma (LUAD) continues to be a primary cause of cancer-related mortality globally, highlighting the urgent need for novel insights finto its molecular mechanisms. This study aims to investigate the relationship between gene expression and mitophagy in LUAD, with an emphasis on identifying key biomarkers and elucidating their roles in tumorigenesis and immune cell infiltration.Methods: We utilized datasets GSE151101 and GSE203609 from the Gene Expression Omnibus (GEO) database to identify differentially expressed genes (DEGs) associated with lung cancer and mitophagy. DEGs were identified using GEO2R, filtered based on criteria of P < 0.05 and log2 fold change ≥ 1. Subsequently, Weighted Gene Co-expression Network Analysis (WGCNA) was conducted to classify DEGs into modules. Functional annotation of these modules was performed using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses. Gene Set Enrichment Analysis (GSEA) was applied to the most relevant module, designated as the greenyellow module. To identify critical biomarkers, machine learning algorithms including Random Forest, Least Absolute Shrinkage and Selection Operator (LASSO) regression, and Support Vector Machine (SVM) were employed. Validation of the findings was conducted using The Cancer Genome Atlas (TCGA) database, Human Protein Atlas (HPA), quantitative PCR (qPCR), and immune cell infiltration analysis via CIBERSORTx.Results: Our analysis identified 11,012 overlapping DEGs between the two datasets. WGCNA revealed 11 modules, with the green-yellow module exhibiting the highest correlation. Functional enrichment analysis highlighted significant associations with FOXM1 signaling pathways and retinoblastoma in cancer. Machine learning algorithms identified COASY, FTSJ1, and MOGS as pivotal genes. These findings were validated using TCGA data, qPCR experiments, which demonstrated high expression levels in LUAD samples. Immunohistochemistry from HPA confirmed consistency between protein levels and RNA-seq data. Furthermore, pan-cancer analysis indicated that these genes are highly expressed across various cancer types. Immune infiltration analysis suggested significant correlations between these genes and specific immune cell populations.Conclusion: COASY, FTSJ1 and MOGS have emerged as critical biomarkers in LUAD, potentially influencing tumorigenesis through mitophagy-related mechanisms and immune modulation. These findings provide promising avenues for future research into targeted therapies and diagnostic tools, thereby enhancing LUAD management.