Targeting mitochondrial quality control using AI-Driven drug repurposing in drug-resistant epilepsy

To the Editor, Drug-resistant epilepsy (DRE) is a major neurological challenge. It happens when seizures continue despite trials of at least two appropriate anti-seizure medications (ASMs). DRE affects about 30% of people with epilepsy worldwide and carries a higher mortality when compared with responsive epilepsy. Patients with DRE have frequent seizures, disability, and poor quality of life. The condition also increases the risk of status epilepticus, injuries, and sudden unexpected death in epilepsy. DRE pathogenesis involves a mix of genetic factors, ion channel dysfunction, neuroinflammation, and distinct cellular mechanisms such as mitochondrial dysfunction that alter neuronal energy balance and excitability. Mitochondria normally regulate ATP production, calcium homeostasis, and reactive oxygen species; when they fail, neuronal hyperexcitability increases, and drugs do not work well. Diagnosis requires clinical history, Electroencephalogram (EEG), and imaging. Standard treatments include ASMs, ketogenic diets, neurostimulation, and surgery in some cases, but control remains inadequate in many patients[1,2]. Drug repurposing is the strategy of finding new uses for existing drugs. In epilepsy, repurposing aims to identify compounds that can help patients who fail traditional ASMs. AI and deep learning can accelerate this process. A recent study developed an in silico drug repurposing pipeline that integrates deep learning with molecular binding affinity screening to identify potential anti-seizure candidates from approved drugs. This pipeline screened over 11 000 compounds against 24 targets associated with epileptogenesis, including mitochondrial pathways. It identified several candidates with strong predicted binding and blood-brain barrier permeability. The core idea is that AI models can learn complex drug-target interactions and propose new uses for safe, approved drugs. This approach has been used in epilepsy, cancer, and other Central Nervous System (CNS) diseases and shows promise as a cost-effective way to find novel treatments[3,4]. Mitochondrial dysfunction is increasingly recognized as a key mechanism in DRE. Neuronal mitochondrial defects, impaired mitophagy, and increased oxidative stress contribute to uncontrolled seizures. Therapies targeting mitochondrial integrity, such as antioxidants and metabolic modulators, have shown benefit in preclinical models and early studies. Some clinical reports suggest that interventions like dietary therapies and mitochondria-targeted antioxidants can reduce seizure burden in subsets of patients with mitochondrial disease-associated epilepsy. AI-guided repurposed drugs, such as cholesterol-lowering drugs with predicted multi-target effects, represent a novel class under investigation. These agents may modulate key pathways such as oxidative stress, ion channel function, or metabolic balance and thus reduce seizure frequency in drug-resistant cases. However, the data are early and require clinical validation[5]. Despite the promise, there are key challenges. First, AI predictions require rigorous experimental and clinical validation. Many in silico candidates never reach human trials. Second, mitochondrial therapies often lack specificity, and their mechanisms in epilepsy are not fully understood. Third, data quality and diversity for AI models are limited, leading to potential bias and reduced generalizability. Clinician awareness of AI-guided repurposing is still low, and regulatory pathways for repurposed drugs are complex and uncertain. These limitations slow translation into patient care[3]. In conclusion, AI-assisted drug repurposing targeting mitochondrial dysfunction offers a novel avenue for DRE. Future work should include large collaborative trials, improved model validation, and clinician education. Combining computational predictions with biological insights may help bring new therapies to patients who currently have few options. This letter to the editor adheres to the Transparency in the Reporting of Artificial Intelligence in Research (TITAN) guideline[6]. Sciencerly,