An AI-integrated pharmacophore and transcriptomic framework for rapid discovery of therapeutic leads for ischemic stroke

Ischemic stroke (IS) is one of the main reasons causing death and disability worldwide, but there are no effective pharmacological treatments available. There is an urgent need to rapidly discover translatable therapeutic candidates. Artificial intelligence (AI) can be applied to accelerate therapeutic lead identification and mechanistic characterization. In this paper, we present a multi-layered AI-based discovery approach that combines a multi-pathway machine learning (ML) model, a Latent Gene Expression Graph Neural Network (LGE-GNN) model, and the Drug Decompose Net (DDN) model to perform structural pharmacophore-level analysis. Using this framework, we identified cannabidiol (CBD) as a proof-of-concept repurposed lead and further validated its therapeutic relevance in IS. Mechanism studies involving LGE-GNN prediction, DeepD2V modeling, and CUT&Tag sequencing indicated that the compound restores antioxidant status and induces angiogenesis through the NRF2/BMAL1 signaling axis. Moreover, DDN-directed structure–activity relationship (SAR) analysis prioritized several putative favorable substructures, including the (+)-dipentene-containing region of CBD, thereby providing a hypothesis-generating framework for future rational chemical optimization. It is presented in the form of an innovative screening system that can be highly integrated with artificial intelligence, medicinal chemistry, and state-of-the-art biological techniques. As indicated by the proposed framework, it can be an efficient strategy for rapid drug repurposing in ischemic stroke. By enabling the efficient identification of repurposable therapeutic leads, the framework can substantially reduce the time, cost, and labor required for early-stage drug development and may be further extended to support future drug discovery efforts.