Repurposing drugs for multi-target treatment of cytokine storm using machine learning

Abstract

Background: Cytokine Release Syndrome (CRS) is a major driver of acute organ failure in critical care settings, affecting approximately 49 million people and killing approximately 11 million people annually. Targeting multiple dysregulated feedback loops via single-molecule polypharmacology offers a promising strategy to blunt excessive immune activation and treat CRS-induced medical conditions such as sepsis. Methods: We used a computational workflow that combined machine-learning classification and molecular docking to identify multi-target modulators of NLRP3, IKBKB, JAK1, CCR5, and SIGMAR1. To eliminate data leakage, feature selection was strictly embedded within a 10-fold cross-validation loop. Candidates were structurally evaluated for binding affinity and active-site contact residues. Results: The model achieved an accuracy of 95.0% to 99.6% and an area under the receiver-operating curve (AUROC) of 0.995 to 1.000 across all five proteins, demonstrating robust generalization. Multiple compounds with average model-predicted scores above 0.50 across all five proteins were discovered. Docking analyses confirmed that the top candidates had favorable binding energetics, displaying stronger binding affinity than known protein inhibitors. Namely, Simeprevir had an average docking score of -10.3 kcal/mol. Shared key contact residues across drug candidates further support the validity of these docking scores and promise interesting in vitro work. Conclusion: Single-drug, multi-target inhibition effectively modulates interconnected inflammatory loops that drive the sepsis-related cytokine storm. This approach provides an alternative to multi-drug combination therapy in intensive care medicine.