跳到正文
原文
Medical Xpress·· 2026-10-16AI 评分50

AI 工具 EpiPred 可预测 STXBP1 变异是否致病

AI tool helps identify harmful epilepsy genetic variants

AI 导读

《临床研究杂志》发表研究,西北医学院等团队开发的 AI 模型 EpiPred 可高准确率预测 STXBP1 错义变异是否致病。测试显示其优于现有通用变异预测工具,并提示约四分之一被归为意义不明的变异可能致病。研究团队已开放线上查询工具,并认为这有助于患者获得精准治疗资格。

正文
AI tool helps identify harmful epilepsy genetic variants
Supervised learning facilitates the generation of an STXBP1-specific variant effect predictor, EpiPred. Credit: Journal of Clinical Investigation (2026). DOI: 10.1172/jci207160

A new artificial intelligence tool developed by Northwestern Medicine scientists and collaborating institutions can help determine whether genetic variants associated with a severe form of childhood epilepsy will cause disease, according to a study published in the Journal of Clinical Investigation.

The findings represent a step toward connecting patients with emerging precision therapies, said Gemma Carvill, Ph.D., associate professor in the Ken and Ruth Davee Department of Neurology's Division of Epilepsy and Clinical Neurophysiology, who was senior author of the study.

In the study, Carvill and her collaborators focused on STXBP1, one of the most common genes associated with developmental and epileptic encephalopathies, a group of rare neurodevelopmental disorders characterized by seizures early in life and significant developmental delays.

Although genetic testing can identify variants in the gene, many are classified as "variants of uncertain significance," or VUS, leaving clinicians unable to determine whether they cause a patient's condition or are simply natural genetic differences.

"Variant of uncertain significance means we have insufficient information to tell whether it's disease-causing or benign," Carvill said.

For families affected by rare epilepsies, that uncertainty can have significant consequences.

"If you send off for clinical genetic testing, you get a variant of uncertain significance in this gene, it essentially precludes you from any future gene-targeting therapies," Carvill said. "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."

To address that challenge, investigators developed EpiPred, a gene-specific machine learning model trained to predict whether STXBP1 missense variants are likely to be disease-causing. The team then tested the model's predictions using multiple laboratory assays that measured protein abundance, stability, solubility and interaction with other cellular components.

The research was conducted as part of a National Institutes of Health–funded collaboration between Northwestern Medicine, the University of Michigan, Weill Cornell Medicine and St. Jude Children's Research Hospital. The initiative combined computational approaches with cellular and animal model systems to improve interpretation of epilepsy-related genetic variation.

By using the new AI-powered model, the investigators found that it outperformed existing global variant prediction tools in accuracy, sensitivity and specificity. The model also aligned closely with experimental laboratory findings and enabled investigators to identify several variants that may have been misclassified in previous studies.

"We found some misdiagnoses. In other words, individuals who have been diagnosed with STXBP1-related disorder, but we actually think that that is a false diagnosis," said Carvill, who is also an associate professor of Pharmacology and of Pediatrics.

Correcting those diagnoses could become increasingly important as gene-targeted therapies move into clinical trials.

"Being accurate in these diagnoses is really important as we move towards these precision therapies," Carvill said.

Next, the research team examined hundreds of known variants and predicted the impact of every possible STXBP1 missense variant. They found that approximately one-quarter of variants currently classified as uncertain may be disease-causing, according to the findings. This could help more patients receive definitive diagnoses and become eligible for future treatments, Carvill said.

Among the study's most significant findings was the model's ability to distinguish harmful variants with high accuracy, she said.

"We can predict with very high accuracy whether we think a missense variant is likely to be pathogenic or not," Carvill said.

The investigators have also made the tool publicly available through an interactive web application that allows clinicians, scientists and families to explore predictions for STXBP1 variants.

Now, Carvill and her team are building on the work with higher-throughput approaches that will allow investigators to evaluate thousands of variants simultaneously.

"We're now developing more high-throughput tools to be able to test every single missense variant," Carvill said.

In the original study, investigators evaluated roughly two dozen variants using four separate experimental assays. Future efforts will work to expand that scale, Carvill said.

"Now we have developed two assays where we can look at every single possible variant that could ever be made in STXBP1," Carvill said. "We can make over 3,000 variants, for instance, in this single gene and determine the impact of those variants on protein function."

Ultimately, the investigators hope their approach will extend beyond epilepsy to other rare genetic disorders, providing a framework for combining gene-specific AI models with laboratory validation. This approach could improve diagnosis, guide clinical trial enrollment and advance precision medicine.

The project's success stemmed from close collaboration among research teams with multidisciplinary expertise, Carvill said.

"I feel like this really was a good representation of team science at its best," Carvill said.

Publication details

Jeffrey D. Calhoun et al., Gene-specific machine learning model EpiPred identifies likely pathogenic variants in the epilepsy-related gene STXBP1, Journal of Clinical Investigation (2026). DOI: 10.1172/jci207160

Journal information: Journal of Clinical Investigation

Who's behind this story?

Gaby Clark

Gaby Clark

MA in English, copy editor since 2021 with experience in higher education and health content. Dedicated to trustworthy science news. Full profile →

Alexander Pol

Alexander Pol

PhD nano-engineering from Delft University. Published researcher and journal reviewer. Brings scientific insight to content standards. Full profile →

Citation: AI tool helps identify harmful epilepsy genetic variants (2026, October 15) retrieved 15 October 2026 from https://medicalxpress.com/news/2026-10-ai-tool-epilepsy-genetic-variants.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.

来源:Medical Xpress · medicalxpress.com