Multi-Omics AI Detects Chronic Fatigue Syndrome Biomarkers

Millions living with chronic fatigue syndrome (ME/CFS) may finally be closer to scientific validation—and, potentially, treatment—thanks to a new study that maps the invisible biological disruptions behind the condition.

Researchers from The Jackson Laboratory, Duke University, and the Bateman Horne Center have used artificial intelligence and multi-omics data to identify distinct signatures in gut microbes, immune cells, and metabolites that distinguish ME/CFS from healthy controls with 90% accuracy. The findings, published July 25 in Nature Medicine, may also shed light on long COVID, a closely related post-viral illness.

A Complex Condition with No Simple Test

ME/CFS affects up to 3.3 million Americans and costs the U.S. economy as much as $51 billion each year, yet diagnosis remains a challenge. No single test can confirm it, and many patients are misdiagnosed or dismissed. Symptoms span a wide range—fatigue, cognitive impairment, pain, dizziness, digestive issues, and sleep disturbances—and vary significantly between individuals.

“Some physicians doubt it as a real disease due to the absence of clear laboratory markers,” said Dr. Derya Unutmaz of The Jackson Laboratory. “Our study achieved 90% accuracy in distinguishing individuals with chronic fatigue syndrome.”

The research team developed BioMapAI, a deep learning platform trained on four years of data from 153 patients and 96 healthy controls. The model integrates gut microbiome sequences, plasma metabolomics, immune cell profiling, blood tests, and patient-reported symptoms.

Key Findings from the Multi-Omics Approach

By linking thousands of biological datapoints to 12 symptom categories, the AI model uncovered biomarkers that aligned with individual symptoms—not just the overall disease state. Among the most notable discoveries:

• Immune cell profiles were the best predictors of fatigue, pain, and general health decline.
• Gut microbiome data most accurately predicted gastrointestinal issues, emotional dysregulation, and sleep disturbances.
• Patients with ME/CFS showed depleted levels of beneficial fatty acids like butyrate and branched-chain amino acids.
• Elevated tryptophan and benzoate levels suggested a disrupted gut–brain axis and altered microbial metabolism.
• MAIT cells, key regulators of gut–immune communication, were unusually active and inflammatory.

“We integrated clinical symptoms with cutting-edge omics technologies to identify new biomarkers of ME/CFS,” said Dr. Julia Oh, lead author and microbiologist at Duke University.

Time Matters: Early Versus Long-Term Illness

The research also showed that patients with recent onset (<4 years) of ME/CFS had less severe biological disruption than those with illness lasting over a decade. Long-term patients displayed more entrenched immune-microbiome imbalances, suggesting that early intervention might prevent worsening over time.

“Our data indicate these biological disruptions become more entrenched over time,” Unutmaz noted. “That doesn’t mean longer-duration ME/CFS can’t be reversed, but it may be more challenging.”

Implications for Long COVID and Beyond

Because ME/CFS and long COVID share clinical similarities—both often follow viral infections like Epstein-Barr virus or SARS-CoV-2—this systems-level approach may offer a model for other post-viral conditions.

BioMapAI’s predictions held up across independent datasets, achieving about 80% accuracy in external validation using microbiome and metabolomics data from other ME/CFS cohorts. This cross-study reproducibility suggests the disease signatures are real and not artifacts of a single dataset or location.

“Despite diverse data collection methods, common disease signatures emerged in fatty acids, immune markers, and metabolites,” Oh said. “That tells us this is not random. This is real biological dysregulation.”

Mapping the Path Forward

The team plans to release the BioMapAI tool and dataset to other researchers, enabling broader exploration of how gut bacteria and immune responses shape chronic illness. They believe this AI-powered framework could eventually help guide targeted therapies or lifestyle interventions.

“The microbiome and metabolome are dynamic,” Oh added. “That means we may be able to intervene—through diet, lifestyle, or targeted therapies—in ways that genomic data alone can’t offer.”

For patients who have long struggled without validation, the hope is that science is finally catching up to experience. Chronic fatigue may still be difficult to see—but it’s getting harder to ignore.

Journal Information

Journal: Nature Medicine
DOI: 10.1038/s41591-025-03788-3
Publication Date: July 25, 2025