Fatigue is one of the most universally reported symptoms in medicine. As a subjective experience, it is incredibly difficult to quantify and even harder to define in scientific terms. This problem becomes even more pronounced when fatigue is not just a symptom but the core feature of  Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS). In ME/CFS, fatigue  is an overwhelming, unrelenting exhaustion that does not resolve with rest. The cause of ME/CFS remains elusive and is one of the most misunderstood, overlooked medical conditions. There is no definitive diagnostic test, and for many patients, the diagnosis process can take years—sometimes even decades—marked by frustration, misdiagnoses, and dismissal by healthcare providers.

The urgency to address ME/CFS is growing. With COVID-19 and the increasing prevalence of long COVID, the number of individuals with conditions resembling ME/CFS is likely to rise significantly. Both conditions share overlapping features, such as debilitating fatigue and “brain fog”, with long COVID bringing renewed attention to the long-term impact of post-viral syndromes. Understanding ME/CFS not only holds the key to improving care for millions already suffering but could also provide critical insights into managing long COVID.

Diagnosing, treating and researching a complex illness like ME/CFS is no small feat. We often find ourselves navigating a labyrinth of overlapping symptoms and comorbidities, trying to identify biomarkers that can truly define this enigmatic condition. Our study tackled this challenge by using machine learning to develop multi-variable biological signatures that can distinguish heterogeneous ME/CFS patients from non-ME/CFS individuals with similar comorbidities.

Behind the Scenes: The Research Journey

When we began this project, our primary goal was to see beyond ME/CFS as an isolated diagnosis. We wanted to understand how comorbid conditions—such as hypertension, depression, asthma, irritable bowel syndrome, hay fever, hypothyroidism, and migraine—interact with ME/CFS and whether these interactions could inform more accurate differential diagnostic tools. The UK Biobank dataset was a core resource, providing blood plasma nuclear magnetic resonance (NMR) metabolomics profiles of deeply phenotyped individuals with and without ME/CFS. Having access to such a comprehensive dataset was critical, as independently collecting data on this many comorbid conditions, along with their unique patterns, while ensuring enough statistical power for meaningful analysis, would have been an almost impossible undertaking. For an overview of our research, check out this 3-minute summary below.

The Machine Learning Edge

ME/CFS patients often present with multiple overlapping symptoms, making it difficult to discern whether these symptoms stem from ME/CFS itself or from co-existing conditions. For example, a patient with ME/CFS and depression might share common biochemical pathways with a non-ME/CFS individual with depression. Teasing apart these nuances was like solving a complex puzzle. 

A breakthrough came when we began to analyse lipoproteins, surface lipids and ketone bodies—biomarkers that offered unique insights into ME/CFS pathophysiology. These biomarkers showed distinct patterns, not just for ME/CFS but also with specific comorbid conditions like migraine and irritable bowel syndrome. This was an exciting moment for the research team, as it underscored the importance of considering comorbidities in diagnostic research. 

We developed a predictive model that could classify heterogenous ME/CFS patients and non-ME/CFS individuals with the same comorbid conditions correctly 83% of the time, and accurately identify ME/CFS individuals 70% of the time. This model incorporated 19 baseline characteristics (like age, blood pressure, blood count) and nine NMR metabolomic biomarkers. The model’s performance was promising; however,  there was room for improvement. The path forward involves validating our findings in larger, more diverse cohorts and exploring additional biomarkers that may further enhance predictive accuracy.

Take home message

By first analysing data from large cohorts can we begin to understand the complexities of ME/CFS at the individual level. This cohort-level study brings us closer to identifying patterns that will lay the foundation for developing personalised approaches in diagnosing and treating this disease