e20061 Background: There has been a lack of comprehensive investigation into changes in the tissue microbial environment at various stages of lung adenocarcinoma (LUAD). This study aims to identify and analyze microbial markers that show significant differences at different stages of LUAD and to assess their impact on the tumor microenvironment. Methods: Microbiome data, gene expression, and clinical information from LUAD patient tissues were retrieved by reanalyzing the sample sequencing data from the Cancer Genome Atlas (TCGA) database, as conducted by Rob Knight’s group. The samples, categorized into early (pathological stage I) and late (stages II-IV) following the initial diagnosis, were subjected to comparative analyses. Firstly, microbial alpha diversity metrics, such as the Shannon index, were computed for each sample. Subsequently, Principal Coordinate Analysis (PCoA) and the Wilcoxon rank sum test were employed to assess differences in microbial communities between the two groups. Furthermore, the study delved into the associations between microbes and host gene expressions using Spearman rank correlation to construct a heterogeneous network encompassing microbial interaction networks at the genus level, microbe-gene interaction network, and gene co-expression network. Various network features, including the clustering coefficient, were examined. Finally, diverse machine learning algorithms were employed to develop a diagnostic model for cancer stage based on microbial and gene expression data. Results: A total of 491 samples, including 267 in the early group and 224 in the late group, were matched with microbial data, mRNA expression, and clinical information. Our findings revealed distinct differences in the microbiome of patients with LUAD at various stages, particularly in the tissues of advanced patients, where in the co-abundance network exhibited a significantly higher level of complexity. Furthermore, we identified five bacterial biomarkers (Pseudoalteromonas, Luteibacter, Caldicellulosiruptor, Loktanella, and Serratia) that were associated with the LUAD stage. Notably, Pseudoalteromonas, Luteibacter, Caldicellulosiruptor, and Serratia were significantly overexpressed in advanced patients. Utilizing the random forest model, we determined that microbial markers possess predictive capabilities for the stage of lung adenocarcinoma and can compensate for deficiencies in other omics data for determining tumor stage. Integration of the mRNA and intratumor microbe biomarkers yielded a prediction area under the Receiver Operating Characteristic curve of 0.70. Conclusions: Our study unveiled the microbial profile of LUAD patients, elucidating the intrinsic pathogenic mechanisms linking the microbiome and the disease. Additionally, we developed a multi-omics model for determining the tumor stage of LUAD.