Prediction of postoperative infection through early-stage salivary microbiota following kidney transplantation using machine learning techniques.

Kidney transplantation (KT) is an effective treatment for end-stage renal disease; however, the lifelong immunosuppressive regimen increases the risk of infection, presenting significant clinical, and economic challenges. Identifying predictive biomarkers for infection onset is critical. In this study, 122 postoperative saliva samples from 39 KT recipients were analyzed using 16S rRNA sequencing, with 16 developing infections within one year. The composition of the salivary microbiota differed significantly between the infection and control groups, with notable variations at the Phylum level. Infected patients exhibited higher alpha diversity and 12 dominant taxa. A random forest model, utilizing five-fold three-times repeated cross-validation and incorporating differential biomarkers, significantly outperformed baseline peripheral blood lymphocyte subpopulation (PBLS) counts in predicting infections (area under the curve, 85.97% ± 10.64% vs. 67.03% ± 15.54%, p = 0.0008). Stepwise logistic regression, integrating clinical data, PBLS counts, and microbiome information, identified Mogibacterium as a significant predictor. The relative abundance of Mogibacterium correlated significantly with the ratio of plateau PBLSs to baseline PBLSs. Early-stage salivary microbiota profiles were predictive of post-KT infections within one year, reflecting lymphocyte reconstitution dynamics.