Abstract

Integrated multiomics identifies a four-gene plasma signature for brain hemorrhage. • Machine learning reliably distinguishes hemorrhage from stroke mimics and hypertension. • Molecular shifts in core biomarkers manifest prior to vascular rupture. • Animal models and prospective cohorts validate the prehemorrhagic signatures. • Immune profiling reveals a unique proinflammatory microenvironment in hemorrhage. Hypertensive intracerebral hemorrhage (ICH) is a devastating stroke subtype with high mortality and disability, yet reliable early risk biomarkers remain elusive. This study integrated protein microarray and transcriptomics with machine learning to identify a 4-gene panel ( ADAM17 , TMPRSS5 , PLAU , and ADAMTS13 ). A LightGBM model with Nested Cross-Validation demonstrated robust discriminative power, achieving AUCs of 0.94 and 0.89 in independent validation cohorts for distinguishing ICH from hypertension and ischemic stroke, respectively. SHAP analysis identified PLAU as the most influential predictor, validated by RT-qPCR and ELISA in an independent replication cohort. Notably, prospective evaluation and animal models revealed a significant PLAU surge prior to vascular rupture. Functional inhibition of PLAU using UK122 significantly delayed ICH onset in hypertensive mice, independent of blood pressure. Furthermore, GSEA and immune infiltration analysis revealed specific enrichment of the Nod-like receptor signaling pathway and a uniquely proinflammatory microenvironment (activated dendritic cells and neutrophils) in ICH.