Analysis of novel biomarkers and immune cell alterations in sepsis based on single-cell sequencing and machine learning algorithms

Abstract We aim to investigate the changes of the immune milieu during sepsis and screen out novel biomarkers with favor diagnostic value using single-cell sequencing and machine learning algorithms. Two gene-chip datasets (GSE28750 and GSE95233) and two single-cell sequencing datasets (GSE167363 and GSE195965) were obtained from the GEO database. We used “ limma ” package in R software to screen out differentially expressed genes (DEGs) in GSE28750 dataset. Then, 10 key genes mostly associated with sepsis were screened out using the random forest algorithm, including SNX3, NAIP, MMP8, EVL, TRBC1, BCL11B, FAIM3, ABLIM1, SIRPG, and CD7. Results of ROC curves showed that these genes also have favor diagnostic value. Moreover, the diagnostic values of biomarkers were also validated in GSE95233 dataset. The immunological microenvironment of sepsis was analyzed using CIBERSORT algorithm, and the relationship between biomarkers and immune cells was identified by “Spearman” method. We also carried out single-cell sequencing analysis on GSE167363 dataset and found that septic T cells differentiated later than normal T cells, and expression of TRBC1 was gradually downregulated over T cell developmental trajectories. Meanwhile, through combing single-cell sequencing analysis and WGCNA analysis, we found that monocytes were upregulated and functionally activated, whereas T cells exhibited significant apoptosis and loss of function both in patients with sepsis and mouse sepsis models.