Fig. 1

The workflow of our study. A After log-cpm and Voom-SNM normalization of TCGA genome and microbiome data, respectively, WGCNA networks were separately constructed and the correlation between their modules was analyzed. B The mRNA modules with the highest correlation to microbiome modules in each type of tumor were extracted for Reactome gene enrichment analysis. C The OAM of each type of tumor was tested for its ability to discriminate among primary tumors using the gradient boosting machine algorithm. D Analysis of microorganisms in OAM for correlation with CD8+ T cells or TAM1 cells. E Prognosis-related OAM microbes and prognosis-related mRNAs with the highest correlation to microbiome modules were integrated for each tumor type using four machine learning algorithms: “Coxnet”, “Random Forest”, “Xgboost”, and “Coxboost” for survival analysis. The KM curves demonstrate the prognostic value of the prognostic microorganism with the highest CARS score in each tumor. F The relative abundance of 2511 tumor microorganisms at the genus level was included for TP53 wild-type and mutant differentiation at the pan-cancer level. log-cpm log-counts per Million, TCGA the cancer genome atlas, WGCNA weighted gene coexpression network analysis, mRNA messenger ribonucleic acid, OAM oncogenome associated microbiome module, CD cluster of differentiation, TAM1 tumor-associated macrophages 1, CARS correlation-adjusted regression survival, TP53 tumor protein P53