Abstract
Lung cancer is the third most common cancer in the US with a 5-year survival rate of 17%. Non-small cell lung cancer, especially adenocarcinoma, prevails. Therefore, early detection and biomarker discovery are extremely important. This study uses deep learning to find new biomarkers for lung adenocarcinoma. RNA-Seq data from 522 samples, including 506 lung adenocarcinoma patients and 16 healthy controls, were analyzed. DEGs were identified after strict preprocessing, and deep learning algorithms predicted markers. Functional annotation, pathway, and protein interaction analyses elucidated the biological importance of DEGs. Clinical relevance was assessed by correlation with clinical parameters and survival analysis. External validation was carried out using GDAC and GEO datasets. Blood samples from 30 lung adenocarcinoma patients and 30 healthy people were analyzed by real-time PCR to validate the expression levels of key genes. Among 522 participants(506 cases, 16 controls), the mean age was 62.95 ± 15.71 years. Normalized data showed 3,513 DEGs. The deep learning model had a predictive accuracy of 98.44%, Brier score (probability MSE) = 0.0013, and AUC of 1.0. CYP3A7 had the highest effect size. ROC analysis found diagnostic genes A2M, CYP2C9, and SIRPD (Ensembl ID: 128646) with a sensitivity of 0.96. Real-time PCR showed upregulated CYP2C9, KRT14, and PECAM1 and downregulated A2M in patients compared to controls(P < 0.001). Bioinformatics-identified genes are potential markers for early lung adenocarcinoma detection and management. RT-PCR validation shows AI's effectiveness in identifying biomarkers, enabling prompt treatment to halt disease progression.