Transcriptome-Based Immune Phenotyping and Survival Stratification in Lung Adenocarcinoma

Lung adenocarcinoma (LUAD) is a major subtype of lung cancer and remains a leading contributor to cancer-related mortality. Only a subset of patients derives durable benefit from immunotherapy, in part because of heterogeneity in the tumor immune microenvironment. This preprint presents a computational framework for stratifying LUAD tumors and evaluating survival differences using bulk RNA-seq data. quanTIseq and EpiDISH deconvolution outputs were integrated across 515 TCGA-LUAD tumor profiles. Immune landscape features were used to define three phenotypes: Immune-Inflamed (24%), Immune-Suppressed (20%), and Immune-Desert (56%). An attention-based model reproduced these phenotype assignments with 97.1% validation accuracy. Kaplan–Meier analysis yielded a log-rank p-value of 0.0091, indicating an association between the inferred phenotypes and overall survival. An exploratory Immune Cell Interaction Network Dynamics analysis produced a disruption score of 0.056 in tumor samples relative to a normal-reference score of 0. These findings are based on retrospective, internally evaluated analyses and require validation in independent cohorts. The framework is intended for research and hypothesis generation and is not a clinical decision tool. Analysis code, processed result summaries, manuscript source, and publication figures are available at: https://github.com/JayadityaVetsa/luad-immune-phenotyping