Combining DNA methylation features and clinical characteristics predicts ketamine treatment response for PTSD.
Amir Valizadeh, John D Roache, Xinyu Zhang, Ying Hu, Ralitza Gueorguieva, Lynnette A Averill, Mohini Ranganathan, Zuoheng Wang, Douglas E Williamson, Paulo R Shiroma, Matthew J Girgenti, Ismene L Petrakis, Argelio L López-roca, Stacey Young-McCaughan, Terence M Keane, Alan L Peterson, Chadi G Abdallah, John H Krystal, Ke Xu
iScience January 16, 2026 Peer reviewed DOI: 10.1016/j.isci.2025.114445 via PubMed
Summary
The largest randomized trial of ketamine for PTSD found no overall benefit over placebo, highlighting the need to identify which patients might respond. Using pre-treatment blood DNA methylation profiles and clinical data from that trial, machine learning predicted treatment response. A model based on 1,208 methylation sites outperformed clinical variables alone, and combining both improved accuracy. The methylation-derived score distinguished responders with 92.9% accuracy. Predictive sites were near genes involved in glutamatergic signaling, immune regulation, and known PTSD risk loci, suggesting blood methylation patterns can identify individuals likely to benefit from ketamine.
Study at a glance
| Design | randomized controlled trial |
|---|---|
| Population | participants with post-traumatic stress disorder (PTSD) |
| Key finding | Peripheral DNA methylation patterns can identify individuals likely to benefit from ketamine for PTSD, advancing precision pharmacotherapy. |
Abstract
Post-traumatic stress disorder (PTSD) exhibits extensive clinical and biological variability, making treatment challenging. The Consortium to Alleviate PTSD (CAP)-ketamine trial, the largest randomized study of ketamine for PTSD, found no overall benefit of ketamine over placebo, underscoring the necessity to identify responsive subgroups. Using pre-treatment blood DNA methylation profiles and clinical measures from the CAP-ketamine trial, we applied machine learning to predict treatment response. A model based on 1,208 methylation sites achieved higher predictive accuracy than models using clinical variables alone, and combining both data types further improved performance. The methylation-derived score distinguished responders with 92.9% accuracy. The predictive CpGs were enriched near genes involved in glutamatergic signaling and immune regulation, as well as established PTSD risk loci. These findings suggest that peripheral DNA methylation patterns can identify individuals likely to benefit from ketamine, advancing precision approaches to PTSD pharmacotherapy.