Multi-omics integration reveals molecular heterogeneity and constructs a machine learning survival model for sepsis-induced coagulopathy
Songzan Qian a 1, Rui Zheng d 1, Yiyi Shi e, Misha Lai a, Junhao Hu a, Mingyue Xu a, Danxiao Pang a, Hewei Ge f, Dingyuan Wang g, Jingye Pan a b c
a) Department of Intensive Care Unit, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, 325000, China
b) Key Laboratory of Intelligent Treatment and Life Support for Critical Diseases of Zhejiang Provincial, Wenzhou, 325000, Zhejiang, China
c) Zhejiang Engineering Research Center for Hospital Emergency and Process Digitization, Wenzhou, Zhejiang, 325000, China
d) Department of Critical Care Medicine, Sir Run Run Shaw Hospital, School of Medicine, Zhejiang University, Hangzhou, Zhejiang, China
e) Department of Anesthesiology, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, 325000, China
f) Department of Surgery, School of Clinical Medicine, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong
g) Department of Breast Surgery, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China
Abstract
Background
Sepsis-induced coagulopathy (SIC) is a life-threatening complication characterized by high heterogeneity and mortality. Current prognostic models relying solely on clinical indices often fail to capture complex molecular pathophysiology, limiting precise risk stratification. This study aimed to unveil the molecular landscape of SIC via multi-omics integration and develop a robust machine learning (ML) predictive model.
Methods
We conducted a comprehensive study of 878 SIC patients. A discovery cohort of 626 patients underwent blood transcriptomic profiling (RNA-seq) and a subset of 214 patients was analyzed for proteomic validation. Weighted Gene Co-expression Network Analysis (WGCNA) and Gene Set Enrichment Analysis (GSEA) were used to identify survival-associated modules and pathways. An ensemble ML framework was developed to integrate the clinical features with transcriptomic signatures for survival prediction.
Results
Clinical analysis identified age, lung infection, higher SOFA scores, and lactate levels as significant independent risk factors for mortality. Transcriptomic profiling revealed that elevated expression of GABARAPL1, PHLPP1, and KLF6 was strongly associated with an increased risk of death, whereas elevated expression of genes, including TSN, NUP155, and TTC39C, was associated with better outcomes. Functionally, GSEA enrichment analysis revealed the suppression of oxidative phosphorylation and ribosome biogenesis along with the activation of hypoxia, heme metabolism, and inflammatory pathways in non-survivors. Proteomic analyses validated the mechanistic findings. The integrated ensemble machine learning survival model (Clinical + Transcriptomics) achieved a C-index of 0.735, which significantly outperformed the clinical-only model (C-index: 0.694). Stratification based on the model successfully distinguished high-risk patients with significantly lower survival rates (p = 0.00057).
Conclusion
Our multi-omics analysis highlights metabolic reprogramming, hypoxia, and dysregulated heme metabolism as the key molecular features of SIC. The developed ensemble ML model, which integrates molecular and clinical features, offers a superior tool for early risk stratification and precision management of septic coagulopathy.