Lung cancer is the leading cause of cancer death among US Veterans despite recent therapeutic advances. The National Lung Screening Trial (NLST) has provided compelling evidence of the efficacy lung cancer screening, using low-dose computed tomography (LDCT), to reduce lung cancer mortality. The benefits of screening, however, must be reconciled with potential harms, including high false-positive rates and the possibility of overdiagnosis. Recent studies demonstrate that lung cancer often exhibits significant molecular heterogeneity because of genomic mutations, facilitating the initiation and expansion of diverse cell populations within the tumor, promoting resistance to targeted and immune therapies, and impacting patient survival. It is critical to characterize and predict tumor behavior in order to develop appropriate clinical management of early-stage lung cancer. We have profiled lung adenocarcinoma (LUAD) in the NLST by whole-exome sequencing (WES) and multiplex immunofluorescence (MIF). Available data includes clinical and nodule information, LDCT images acquired at three time points, and long-term clinical follow-up, which allows us to assess aggressive versus indolent behavior. We hypothesize that somatic mutations impact immune responses, shaping the tumor microenvironment (TME) and morphology, can be described by quantitative features extracted from LDCTs. Our goal is to develop a systematic approach integrating molecular signature and clinical imaging profiles to distinguish between aggressive and indolent early-stage lung cancer by utilizing data collected in the NLST study and validating findings through the analysis of publicly available data sets. We plan the following specific aims: 1) To identify the somatic mutation profiles associated with aggressive tumor behavior. In addition to analyzing the NLST data, we will extend our analysis to publicly available data sets, including The Cancer Genome Atlas (TCGA) and the Stanford Non-Small Cell Lung Cancer Radiogenomic dataset. From these two data sets, we will select subjects with equivalent clinical features to the NLST cohort. 2) To identify tumor microenvironment alteration characteristics associated with aggressive behavior in the NLST cohort. We will cross-validate our TME findings with the signatures derived from transcriptome data of publicly available data sets, as noted in Aim 1. We will deconvolute the transcriptome data to estimate the composition of cell types in TME by utilizing the cell lineage markers, which we have identified from single-cell transcriptome data of lung cancer. 3) To integrate the LDCT features and molecular profiles that differentiated aggressive from indolent behaviors in the NLST cohort. We will leverage machine learning techniques to extract and combine features from LDCT with genetic and TME signatures, as identified in Aim 1 and 2. Predicting disease aggressiveness will improve personalized patient therapy for early-stage lung cancer.