The researchers developed a prognostic GMGscore model for sepsis based on 6 genes associated with gut microbiota (CYP1A2, FFAR2, IL4R, MUC1, RORA, ASPM), which achieved high accuracy (AUC = 0.903 in the training set GSE65682 and AUC = 0.901 in the validation set GSE95233). The model was created by analyzing differentially expressed genes from the GEO database (GSE154918), stretched with 248 genes from the GutMGene database, identifying 34 relevant genes enriched in pathways such as inflammatory bowel disease and IL-17 signaling. A high GMGscore was associated with poor survival, increased neutrophil degranulation, and decreased neutrophil counts. The key gene RORA was consistently suppressed in sepsis, with the highest diagnostic value (high AUC), expressed mainly in effector T cells and NK cells, where it was positively correlated with the infiltration of CD8+ T cells (R = 0.419) and NK cells (R = 0.352). RORA virtual knockout reduced the expression of cytotoxic genes. Molecular docking showed stable binding of RORA with metabolites from Collinsella (citric acid, sedoheptulose, tricarballylic acid). These findings were confirmed by RT-qPCR in patient blood and scRNA-seq data (GSE167363).