Machine learning integration of bulk and single-cell RNA-Seq data reveals COPS2 as a central immune regulator in deep vein thrombosis.

Deep vein thrombosis (DVT) is a life-threatening condition in which dysregulated immune-endothelial interactions drive thrombus formation, yet upstream molecular regulators remain poorly understood. Employing an integrative multi-omic strategy - combining proteome-wide Mendelian randomization, immune cell mediation analysis, bulk and single-cell transcriptomics, and machine learning - we identified 132 plasma proteins with genetically inferred causal effects on DVT. Among these, COP9 signalosome subunit 2 (COPS2) emerged as the most consistent candidate, with approximately 20% of its effect mediated through effector memory cluster of differentiation 8-positive (CD8+) T cells. Transcriptomic profiling localized COPS2 expression to CD8+ T cells and endothelial cells, and functional validation in endothelial cells demonstrated that COPS2 directly upregulates tissue factor (TF), intercellular adhesion molecule-1 (ICAM-1), and Bcl-2-associated X protein (BAX), enhances procoagulant activity, and activates the Janus kinase-signal transducer and activator of transcription (JAK-STAT) signaling pathway. Machine learning models consistently identified COPS2 as a robust diagnostic feature across independent cohorts. In vivo, Cops2 deficiency reduced thrombus burden, attenuated activation of effector memory CD8+ T cells, and decreased intrathrombotic levels of interferon-gamma (IFN-γ) and granzyme B (GZMB). These findings establish COPS2 as a key regulator of immune-endothelial crosstalk in DVT and highlight its potential as a biomarker and therapeutic target.