Specific diagnostic model for bacterial pneumonia constructed by combining multiple omics and multi machine learning models.

Due to the difficulty in quickly discriminating bacterial pneumonia (BPs) from pneumonia caused by other pathogens in clinic, the research for new biomarkers which can specifically diagnose BPs is of great significance. Here, miRNA, proteomic, and metabolomic sequencing of 45 serum samples from BPs and 35 serum samples from non bacterial pneumonia (NBPs) were characterized. After profiling, 8 biomarkers (miR-339-3p_R-2, miR-877-5p_R+3, hnRNPK, SERPINA5, DHA, 1,3-Oxazinan-2-imine, Asparagoside_ A, and Hexanamide,N-[(1S,2R)-2-hydroxy-1-(h ydroxymethyl)heptadecyl]-6-[(7-nitro-2,1,3-benzoxadiazol-4-yl)amino]-) which owns potential to diagnose BPs were screened out. To more accurately identify BPs, six machine learning (ML) classification algorithms including logistic regression (LR), random forest (RF), support vector machine (SVM), eXtreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM) and extreme random tree (ExtraTree) were applied. Among the models, LR achieves the highest AUC value of 0.892 (95%CI: 0.723-0.992) with the sensitivity and specificity are 0.769 and 0.900, respectively. In total, this study finds 8 potential biomarkers and build a reliable diagnostic model for the BPs diagnosis.