Graph clustering AI pinpoints inflammation subtypes in severe epilepsy NORSE

Cryptogenic new-onset refractory status epilepticus (cNORSE) is a severe, unexplained seizure disorder with poor outcomes. Emerging evidence points to immune dysregulation, but the marked heterogeneity in inflammatory signatures makes choosing the right immunotherapy extremely difficult. To solve this, researchers applied graph-based clustering to serum cytokine profiles—measuring 96 different cytokines in a cohort of 62 cNORSE patients. The algorithm identified distinct inflammatory subgroups, or clusters, that are biologically validated.

Building on those clusters, the team developed a predictive model that takes a new patient’s cytokine panel and outputs the most likely inflammatory cluster, an attribution probability, and a statistical confidence score. The framework also supports longitudinal monitoring, tracking how a patient’s inflammatory profile evolves over time. The result is a clinician-friendly tool that translates complex immune data into actionable insights, helping neurologists choose targeted immunomodulatory therapies with greater precision. This study, published on arXiv (2606.24351), bridges graph-based AI and neuroimmunology to address a critical clinical need.