Machine learning-driven prognostic and diagnostic models for lung adenocarcinoma using intratumor heterogeneity and multi-omics data

Intratumor heterogeneity (ITH) significantly impacts cancer prognosis and treatment response. Focusing on lung adenocarcinoma (LUAD), this study investigates the relationship between ITH and clinical outcomes, and constructs machine learning-based prognostic and diagnostic models. ITH scores were calculated using the DEPTH2 package, and weighted gene co-expression network analysis (WGCNA) was applied to identify ITH-associated core genes. A 19-gene prognostic model was developed using Elastic Net (Enet), and a 7-gene diagnostic model was built through a combination of LASSO and Random Forest (RF). The prognostic model was validated across six independent datasets, while the diagnostic model was tested in three. ITH was found to correlate significantly with clinical characteristics such as gender, M stage, and overall survival. WGCNA revealed the black and lightgreen modules as key to ITH, contributing 126 core genes. Both models demonstrated strong predictive performance and generalizability, accurately stratifying LUAD patients and distinguishing them from healthy controls. These findings underscore the clinical value of incorporating ITH and multi-omics data into model construction to enhance precision in LUAD diagnosis and prognosis.