Prediction Of Lung Cancer and Analysis of Lung Cancer Related Gene Expressions Through Logistic Regression, PCA Random Forest, And LASSO Regression

Research Background: Lung cancer is frequently associated with the expression levels of certain genes, where both over-expression and under-expression may indicate the presence of the disease. It was hypothesized that specific genes might exhibit a more pronounced correlation with lung cancer. To explore this, gene expression data from patients with and without lung cancer were obtained and initially screened using volcano plots to identify genes with significant differential expression. Study Contributions: The refined dataset was subjected to comprehensive analysis using five distinct methodologies: principal component analysis (PCA), random forest, logistic regression, least absolute shrinkage and selection operator (LASSO), and a method employing lambda cross-validation for selecting influential genes. Each model’s performance was assessed by its area under the curve (AUC) value, which then informed a weighting system to prioritize the findings. A weighted table was subsequently developed to finalize the diagnosis of lung cancer. The synthesis of these approaches not only enhanced the accuracy of the diagnostic model, confirmed to be precise in at least 96 percent of cases, but also led to the identification of 75 genes significantly associated with lung cancer. Among these, CLDN18, GKN2, LYVE-1, GPIHBP1, and CLIC5 were determined to be the most closely linked to the disease.