AI study suggests a quarter of uncertain STXBP1 variants may cause disease

A new AI model could help resolve a difficult question after genetic testing: whether a change in an epilepsy-linked gene actually causes disease. Northwestern Medicine researchers and collaborators developed EpiPred to assess variants in STXBP1. They report better predictions than broader tools, with potential consequences for diagnoses and access to future gene-targeted therapies.

Northwestern described the research on October 15, 2026, citing a study published in the Journal of Clinical Investigation. STXBP1 is among the genes most commonly associated with developmental and epileptic encephalopathies—rare disorders involving early-life seizures and significant developmental delays.

Genetic testing can identify changes in STXBP1 without settling whether they explain a patient's condition. Many receive the label “variant of uncertain significance,” meaning there is not enough information to classify them as disease-causing or benign.

EpiPred targets that interpretation problem, predicting whether STXBP1 missense variants, a particular class of genetic changes, are likely to cause disease. Researchers checked roughly two dozen variants using four laboratory assays measuring protein abundance, stability, solubility and interactions with other cellular components.

The NIH-funded collaboration involved Northwestern Medicine, the University of Michigan, Weill Cornell Medicine and St. Jude Children's Research Hospital.

The researchers report that EpiPred outperformed existing global variant-prediction tools in accuracy, sensitivity and specificity, and closely matched laboratory findings. Northwestern did not give numerical scores quantifying the advantage.

The team examined hundreds of known variants and predicted the impact of every possible STXBP1 missense variant. That computational coverage is not laboratory confirmation of every prediction. The study estimated that approximately one-quarter of variants currently classified as uncertain may be disease-causing.

The work also pointed in the other direction: senior author Gemma Carvill said the team identified some people whose existing STXBP1-related diagnoses it believes may be incorrect.

Because the diagnosis is not definitive, it means that you are not eligible to enroll for new precision therapies that specifically target the gene of interest.

Gemma Carvill, senior author of the study, speaking to Northwestern's News Center

Carvill said resolving uncertain results could help more patients receive definitive diagnoses and qualify for future treatments as gene-targeted therapies move toward clinical trials. Those benefits remain prospective, rather than demonstrated patient outcomes.

The investigators have made an interactive web application publicly available so clinicians, scientists and families can explore predictions for STXBP1 variants. Carvill said the team has also developed two assays intended to test every possible missense variant, assessing the effects of more than 3,000 variants on protein function.