An ischemic stroke-on-a-chip model integrated with machine learning for screening of drug candidates

Abstract

Ischemic stroke is a leading cause of death and long-term disability worldwide, characterized by the sudden loss of cerebral blood flow, resulting in energy failure, oxidative stress, inflammation, and blood–brain barrier (BBB) disruption. Despite their clinical significance, current preclinical models inadequately recapitulate the multifaceted pathophysiology of cerebral ischemia, hindering systematic therapeutic exploration. Here, we engineered a human iPSC-derived BBB-on-a-chip (iBBB-on-a-chip) platform that can simulate ischemic pathology through controlled oxygen–glucose deprivation. The model exhibited characteristic features of ischemia-induced BBB impairment, including disruption of endothelial tight junctions, pericyte fragmentation, and increased permeability. By integrating transcriptomic profiling with a weighted gene co-expression network analysis algorithm, we identified stroke-related pathways and applied machine learning (random forest and LASSO) to screen hub genes for biomarker discovery. Using the Connectivity Map database and molecular docking calculations, we identified coumarin as a potential therapeutic agent and experimentally confirmed its protective role in the iBBB-on-a-chip. This strategy establishes a novel paradigm combining organ-on-a-chip technology with machine learning-driven data analysis, creating an innovative platform for the study of cerebrovascular disease and drug screening.