Xiong et al. (2025)
- Authors: Ruoyun Xiong, Elizabeth Aiken, Ryan Caldwell, Suzanne D. Vernon, Lina Kozhaya, Courtney Gunter, Lucinda Bateman, Derya Unutmaz, Julia Oh
- Institutes: Duke University, Durham, NC, USA; The University of Connecticut Health Center, Farmington, CT, USA; The Jackson Laboratory, Farmington, CT, USA.
- Publisher: nature medicine
- Link: DOI
Summary
This study introduces a powerful AI tool that integrates vast amounts of biological data from the gut, blood, and immune system to create a more comprehensive and personalized picture of ME/CFS. By connecting specific biological abnormalities to individual symptoms, it moves beyond a simple “sick vs. healthy” comparison and begins to explain the disease’s profound variability. The findings reinforce that ME/CFS involves a significant breakdown in the communication between the gut microbiome and the immune system, leading to chronic inflammation that correlates with symptom severity. This approach provides a new framework for discovering more precise biomarkers and generating targeted hypotheses for future treatment research.
What was researched?
This study developed and applied a deep neural network, named BioMapAI, to a large longitudinal, multi-omics dataset from ME/CFS patients and healthy controls. The objective was to integrate diverse biological data—including gut metagenomics, plasma metabolomics, and immune profiling—to predict clinical symptom severity, identify specific biomarkers, and create a comprehensive map of the biological interactions underlying the disease.
Why was it researched?
ME/CFS is a complex illness with highly variable symptoms, which makes diagnosis and treatment challenging. Previous studies often focused on identifying one or two key disease indicators, an approach that fails to capture the multifaceted nature of the condition. This research was motivated by the need for methods that can link a wide range of biological data types to the complex matrix of patient-specific symptoms, enabling a more personalized understanding of the disease.
How was it researched?
This was a 4-year longitudinal study that collected data from a cohort of 153 ME/CFS patients and 96 healthy controls. The researchers gathered multiple data types: gut microbiome sequencing, plasma metabolomics, immune cell profiling, standard blood lab results, and detailed clinical questionnaires. They used this extensive dataset to train BioMapAI, an explainable AI model, to map the biological profiles to 12 core clinical symptom scores. The model’s ability to classify ME/CFS was validated using both held-out data from the cohort and four independent, external patient datasets.
What has been found?
The BioMapAI model successfully classified ME/CFS patients from healthy controls with high accuracy (91% AUC). The study identified both disease-specific biomarkers (consistent across most symptoms) and, more uniquely, symptom-specific biomarkers. Key findings include altered associations in ME/CFS between gut microbial metabolism (e.g., of short-chain fatty acids like butyrate, BCAAs, and tryptophan) and a heightened inflammatory response from specific T cell subsets (MAIT and T cells). These disruptions in the microbe-immune network were strongly correlated with the severity of symptoms like fatigue, pain, gastrointestinal issues, and sleep disturbances.
Discussion
The authors acknowledge several limitations, including that the model was not tested for its ability to differentiate ME/CFS from other conditions with overlapping symptoms, such as fibromyalgia. The study cohort was from a single geographic location and predominantly white, which may limit the generalizability of the findings. Furthermore, the 4-year study period may not be sufficient to capture the full, often decades-long, nonlinear progression of ME/CFS.
Conclusion & Future Work
The authors conclude that their AI-driven, multi-omics approach provides unprecedented systems-level insights into ME/CFS, refining existing hypotheses and generating new ones about the disrupted connections between the microbiome, metabolome, and immune system. While the findings are not yet ready for direct therapeutic application, they offer numerous avenues for future mechanistic studies. The researchers suggest future work should include longer follow-up periods and incorporate other data types, like host RNA sequencing, to further unravel the disease’s complexity.