An 11-gene blood transcriptomic signature reflects a sepsis-associated host-response pattern across public cohorts

Sepsis is a clinically defined syndrome caused by a dysregulated host response to infection, yet the biological architecture underlying this response remains highly heterogeneous. Blood transcriptomic studies have shown that patients with sepsis can be subdivided into reproducible molecular states; however, translation has been hindered by inconsistent gene sets, cohort-specific modeling strategies, and insufficient external validation across public datasets. We aimed to derive a compact blood transcriptomic signature to characterize a sepsis-associated host-response pattern and to evaluate its portability across independent public cohorts and clinically distinct validation settings.

Using GSE65682 as the discovery cohort, we identified differentially expressed genes in the Abdominal_Sepsis vs. GI_Control comparison, prioritized hub genes through STRING-PPI analysis, and entered the top 20 degree-ranked candidates into cross-validated LASSO-logistic modeling. The final 11-gene signature was defined according to the nonzero coefficient rule. Because the discovery coefficients were estimated from standardized variables, external fixed-score evaluation was performed after the retained genes were transformed using cohort-specific StandardScaler procedures within each external analysis cohort. External analyses were performed using GSE236713 across three Day 1 settings: OOHCA-SIRS vs. sepsis, abdominal vs. pulmonary sepsis, and survival among patients with sepsis. GSE54514 was used as a supplementary, non-confirmatory cohort.

In GSE236713, the standardized fixed-coefficient score showed moderate discrimination between Day 1 OOHCA-SIRS and sepsis (AUC = 0.7676; Mann-Whitney P = 4.74 × 10^-7), whereas the abdominal vs. pulmonary sepsis and Day 1 survival analyses showed near-null discrimination (AUC = 0.5126 and 0.5291; P = 0.8119 and 0.6334, respectively). GSE54514 was retained as a supplementary, non-confirmatory cohort.

We identified an 11-gene blood transcriptomic score that may reflect a sepsis-associated host-response pattern. The score showed moderate discrimination between Day 1 sepsis and OOHCA-SIRS, whereas the available Day 1 analyses did not demonstrate clear discrimination by infection source or short-term survival under the current analysis settings.