Fig. 1
Construction of a prognostic HLMRG signature. A UMAP algorithm showing sample clusters before and after batch normalization. B Volcano plot showing differentially expressed HLMRGs between sepsis patients (n = 479) and healthy controls (n = 42) in the discovery set. The screening criteria were set as adjusted P < 0.05 and | log2 (fold change) |≥ 1. C Expression profile of the top 10 dysregulated HLMRGs. D Univariate Cox regression analysis to screen HLMRGs related to 28-day survival of sepsis patients in the discovery set (n = 479). Squares represent the HR of death and open-ended horizontal lines represent the 95% CI. All P-values were calculated using Cox proportional hazards analysis. E Ten-fold cross-validation for tuning parameter selection in the LASSO estimation. The partial likelihood deviance corresponding to each lambda value is shown as mean ± SD. The dotted vertical line (left) indicates the optimal value by minimum criteria. F LASSO coefficient profile of individual genes included in the estimation. G Distribution of the multivariate Cox regression coefficients of the HLMRG signature