A potential XGBoost Diagnostic Score for Staphylococcus aureus bloodstream infection

JunHong Shi ¹ Lan Chen ² Xinru Yuan ¹ Jinjin Yang ¹ Yanlei Xu ¹ Li Shen ¹ Yu HUANG ¹ Wang Bingjie ^1* Fangyou Yu ^1*

• ¹ Shanghai Pulmonary Hospital, School of Medicine, Tongji University, Shanghai, China
• ² Ningbo First Hospital, Ningbo, Zhejiang Province, China

Staphylococcus aureus (S. aureus) bloodstream infection is often life-threatening, and increasing in incidence. We identified 63 differentially expressed genes (DEGs) in the GSE33341 S. aureus infection samples. Subsequently, intersecting the 63 DEGs with 950 genes from the blue module through weighted gene co-expression network analysis (WGCNA) yielded 38 genes. We leveraged Boruta and least absolute shrinkage and selection operator (LASSO) algorithms and identified5 diagnostic genes (DRAM1, UPP1, IL18RAP, CLEC4A, and PGLYRP1). Comparative analysis revealed that Extreme Gradient Boosting (XGBoost) surpassed SVM-RFE and Random Forest models, demonstrating superior diagnostic performance for S. aureus bloodstream infection (mean AUC for XGBoost =0.954; mean AUC for SVM-RFE =0.93275; mean AUC for Random Forest =0.94625). The XGBoost Diagnostic Score correlated with multiple immune cells to varying degrees, manifesting significant negative associations with CD8 T cells and CD4 naive T cells in both human and mouse samples. The diagnostic power of the Diagnostic Score was further validated by RT-qPCR results obtained from both mouse and patient samples, as well as RNA-Seq analysis conducted on mouse samples. XGBoost Diagnostic Score, consisting of DRAM1, UPP1, IL18RAP, CLEC4A, and PGLYRP1, may serve as a Diagnostic tool for S. aureus bloodstream infection.