Review Machine learning and multi-omics in precision medicine for ME/CFS, 2025, Huang....Armstrong

[MelbME]

https://translational-medicine.biomedcentral.com/articles/10.1186/s12967-024-05915-z



Machine learning and multi-omics in precision medicine for ME/CFS
Katherine Huang,
Brett A. Lidbury,
Natalie Thomas,
Paul R. Gooley &
Christopher W. Armstrong

Journal of Translational Medicine volume 23, Article number: 68 (2025)

Abstract
Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a complex and multifaceted disorder that defies simplistic characterisation. Traditional approaches to diagnosing and treating ME/CFS have often fallen short due to the condition’s heterogeneity and the lack of validated biomarkers. The growing field of precision medicine offers a promising approach which focuses on the genetic and molecular underpinnings of individual patients. In this review, we explore how machine learning and multi-omics (genomics, transcriptomics, proteomics, and metabolomics) can transform precision medicine in ME/CFS research and healthcare. We provide an overview on machine learning concepts for analysing large-scale biological data, highlight key advancements in multi-omics biomarker discovery, data quality and integration strategies, while reflecting on ME/CFS case study examples. We also highlight several priorities, including the critical need for applying robust computational tools and collaborative data-sharing initiatives in the endeavour to unravel the biological intricacies of ME/CFS.

 

[Utsikt]

«We need to use all of the available tools and work together to solve this».

This approach integrates biological data including genetic profiles, medical history, social, and behavioural information, to enable tailored decision-making for disease prevention, prediction, and treatment.

Why is this relevant for ME/CFS?

Although omics studies have yet to identify definitive pathways in ME/CFS, recent advancements in computational power, growing datasets (data type, volume and sample size), and more efficient machine learning algorithms can reveal previously missed or hidden underlying mechanisms.

Logical fallacy? «We haven’t found anything yet, so we need a different approach.» More like we haven’t had the funding to dig deep enough with large sample sizes..

However, it is important to acknowledge that AI and machine learning are not magic bullets capable of solving all the complexities of ME/CFS. Their success depends heavily on high-quality, relevant input data and defining specific training endpoints to be modelled. Without these, even the most advanced algorithms may struggle to produce actionable insights.

«Shit in, shit out.» I agree.

Seems to me like a hype-piece about AI and big data. They make some fair points about best practices, but I doubt this will lead to any changes from any current researches.

Maybe they are trying to take advantage of the AI hype and draw attention to ME? Could be worth a shot..

At least they got the biomed foundation of ME right, which is nice to see for a change.

 

[Trish]

Maybe they are trying to take advantage of the AI hype and draw attention to ME? Could be worth a shot..

I don't see it that way. This research group led by Chris Armstrong has been researching ME/CFS biology for some years, and show a good understanding of the difficulities of finding the cause of ME/CFS.

The social and behavioural bit puzzled me too. I'm hoping they simply mean that diagnosis includes asking the patient about the effects on their symptoms of different kinds of activity. Pacing is a behavioural strategy for managing ME/CFS to avoid PEM.

I suspect they are right that we need much bigger samples across multiple omics, which can only be analysed with the help of computer algorithms.

 

[Sasha]

Is he on the forum? Can we just ask him?

 

[Yann04]

Is he on the forum? Can we just ask him?

Yes. This thread was made by Chris.

 

[MelbME]

We were just listing all the types of information that can be collected to interpret understanding of an individual or a disease. It wasn't targeted at something specific to ME.

Relevant information to ME could be hiding in places we don't suspect. But just to give an example, there are phone apps that can monitor user behaviour. You could track how slowly someone types words, how often they make posts on a forum, how often they use the device, etc. We have played around with the idea that this could be an objective marker or a factor in a suite of objective behaviour markers (perhaps tied to step tracking or up time). From speaking large biotech companies, an objective outcome measure for clinical trials is a major hurdle for them to get involved in this area of disease. So yeah, this is an example of why that type of data could be relevant to specifically ME.

Purpose of the paper was to summarize existing efforts and to highlight how this type of work is a good fit for a complex diseases like ME/CFS. I typically get my PhD students to do a review article of some kind for their benefit and to provide a nice summary for others interested in the topic.

 

[Utsikt]

I don't see it that way. This research group led by Chris Armstrong has been researching ME/CFS biology for some years, and show a good understanding of the difficulities of finding the cause of ME/CFS.

I was not aware of that context, thank you for letting me know. Seems like I jumped to conclusions, although prior work isn’t a guarantee against future mistakes or inaccuracies so we need scrutiny for everyone (which is what you alway do!)

You could track how slowly someone types words, how often they make posts on a forum, how often they use the device, etc. We have played around with the idea that this could be an objective marker or a factor in a suite of objective behaviour markers (perhaps tied to step tracking or up time).

Interesting idea! Although my typing speed and posting frequency isn’t always tied to current «local» health, and it would be difficult to isolate the ME-effect from other factors. Others have also pointed out how up time increases during PEM due to e.g. brainfog and forgetting things, yet it decreases when you getting sicker long term. Maybe AI would still be able to pick out a reliable pattern across multiple variables.

Purpose of the paper was to summarize existing efforts and to highlight how this type of work is a good fit for a complex diseases like ME/CFS. I typically get my PhD students to do a review article of some kind for their benefit and to provide a nice summary for others interested in the topic.

In that case, it seems like the article fulfilled its purpose

 

[mariovitali]

Logical fallacy? «We haven’t found anything yet, so we need a different approach.» More like we haven’t had the funding to dig deep enough with large sample sizes..

Or it may be that the decisions we have been taking regarding which research will be funded or which collaborations will take place have been largely ineffective. An example here (note, I am the author) :

https://twitter.com/user/status/1874467477310287995


Having used machine learning and network analysis methods to research ME/CFS since 2015 -and not having any support from patient organisations throughout these years- I would like to give my point of view regarding the use of AI to research ME/CFS :

-Machine learning will not provide the "whole" picture that we need regarding causal mechanisms, even on a personalised level. Personalised interventions will most likely be needed once the causal mechanisms are known .

- The same holds for using features from metabolomic analysis. Knowing that a subset of metabolites is able to differentiate ME/CFS patients vs controls really doesn't say much even if high precision/recall is achieved. Is the same subset of metabolites able to differentiate ME/CFS patients vs LongCOVID patients? How about MS patients?



- What is needed is a way to understand what these analyses are telling us. The review mentions the use of Network Analysis. To give an example, here is one I made -and circulated- in 2017 (note the urea cycle node also discussed in the paper) :






What is needed is to be able to understand -as an example- why glycoproteins are situated between these two clusters of concepts appearing in the generated network. Does this mean they play a role? Yes, they could : https://twitter.com/user/status/1863564681169838363
.

-The review mentions also the use of pathway enrichment analysis : Are we sure that the representation currently used during pathway enrichment analysis software is the right one? ( there are other ways to do it more efficiently for the problem at hand)

So, unfortunately it's not only about using AI but also who uses it and how it is being used. But above everything else is who makes the decisions on which AI-assisted research moves forward.

 

[Trish]

You could track how slowly someone types words, how often they make posts on a forum, how often they use the device, etc. We have played around with the idea that this could be an objective marker or a factor in a suite of objective behaviour markers (perhaps tied to step tracking or up time).

Interesting idea. Typing speed also depends on what I'm typing about.
Factors affecting my typing speed and number of corrections I need to make include
- fatiguablity - gets slower if I go on too long - both cognitive and physical.
- what else I've done that week, day, hour and how close to crashed I am
- the subject matter and whether it requires a lot of thought or I'm just rambling or copy typing
- whether I'm touch typing on my laptop or one finger typing on my phone.
- mistake rate increases markedly as I fatigue, if pushing when crashed, understanding the words on the screen and being able to spell them deteriorates
- overall quantity of typing may not directly relate to state of health - also depends on whether there's anything I want to say, and what else I'm doing

So it's a complex picture, but I think could be useful. Perhaps it needs to be narrowed down to something like assessing typing speed and error rate on a set task. Combining and or correlating it with steps, time with feet on the floor, resting heart rate, HRV might be interesting.

 

[NelliePledge]

My posts definitely tend to get longer and more wordy when I’ve got to my limits. I often struggle to realise this until I stop typing and realise I’ve been rambling. It takes quite a bit of concentration to keep to the point.

 

[Trish]

My posts definitely tend to get longer and more wordy when I’ve got to my limits. I often struggle to realise this until I stop typing and realise I’ve been rambling. It takes quite a bit of concentration to keep to the point.

Me too. Members might not believe this, given how much I post, but I quite often delete my long incoherent rambles without posting them.

 

[Nightsong]

We also discussed typing patterns & other potential metrics on this thread a while ago:

https://www.s4me.info/threads/exper...tigueability-and-pem-discussion-thread.38585/

 

[Hutan]

Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a complex and multifaceted disorder that defies simplistic characterisation.

I'm not sure what this sentence is saying.

'Complex' can be a code word for 'biopsychosocial' and 'psychosomatic'. Also, we don't really know that ME/CFS is complex, it's just that we don't understand it yet.

'Multifaceted' can be read as 'biopsychosocial' also - as in, 'yes there was a biological trigger but it is perpetuated by behavioural factors'. What multiple facets are being referred to here?

'Defies simplistic characterisation' also can be read as 'biopsychosocial' and suggests that ME/CFS has been characterised, just not simply. To date, that's not true. I'm not a fan of personifying the disease by suggesting it is defiant either.

If the authors meant when they described ME/CFS as 'complex, multifaceted, and defying simplistic characterisation' that it 'has many symptoms' and 'no one understands what causes it yet', why not just say something closer to that? Or was something else meant?
Probably you don't want to say 'researchers are currently resorting to models with large numbers of parameters, that are of questionable validity, in order to separate their usually way too small ME/CFS sample from their healthy controls, and these efforts are poorly replicated'.

Traditional approaches to diagnosing and treating ME/CFS have often fallen short due to the condition’s heterogeneity and the lack of validated biomarkers.

Again, it's not clear what is being said here. There's a lack of precision in the sentence. Traditional approaches to treating ME/CFS are GET and CBT. They have not fallen short due to the condition's heterogeneity and lack of validated biomarkers. The sentence as it stands suggests that there are treatments and that they are sometimes useful. There is no evidence that those treatments work for anyone with any sort of CFS or ME/CFS diagnosis. Suggesting that there are effective treatments is wrong and problematic. We need to be saying clearly 'there are no evidence-based treatments for ME/CFS'.


Yes, I know this team means well and is doing good things. But those issues with the first two sentences of the abstract suggest a lack of care to get things right.

The abstract calls for 'collaborative data-sharing initiatives in the endeavour to unravel the biological intricacies of ME/CFS', which is great. But, did this project team include one or more people with ME/CFS familiar with the politics and the literature who could share what they know and help ensure that the many land mines of ME/CFS were not stepped on?

 

[Utsikt]

If the authors meant when they described ME/CFS as 'complex, multifaceted, and defying simplistic characterisation' that it 'has many symptoms' and 'no one understands what causes it yet', why not just say something closer to that?

The phrase "if you can't explain it simply you don't understand it well enough" comes to mind. I don’t know if that’s the case here, but it could reasonably be interpreted that way.

Nobody can expect anyone to understand ME/CFS completely or even partially, but we can expect researchers to understand what we currently understand about ME/CFS and its history.

Your comment has also reminded me of my unfondness of buzz words and tropes in academia. Don’t use definitions that requires further definitions if it’s at all avoidable.

 

[SNT Gatchaman]

Thank you @MelbME and colleagues, I think this is a great reference overview of the -omics and ML pipelines. ML in particular has a lot of unfamiliar terms.

Personally I'm happy with the characterisation of ME/CFS as "complex". It's certainly not simple for our current understanding, even if the ultimate explanation might turn out to be relatively simple once it's recognised and understood. I know BPS have co-opted the word "complex" as part of their coded obfuscation, but it can be repatriated

 

[Hutan]

Personally I'm happy with the characterisation of ME/CFS as "complex". It's certainly not simple for our current understanding, even if the ultimate explanation might turn out to be relatively simple once it's recognised and understood. I know BPS have co-opted the word "complex" as part of their coded obfuscation, but it can be repatriated

But I'm not clear what the word 'complex' is saying there -
1. that it's difficult to work out the cause, or
2. that there are many factors contributing to the disease (including possibly psychological ones), or
3. that the disease has many interacting parts, or
4. that there are lots of symptoms, or
5. something else?
Why not use a word that makes your meaning clear?

Personally I'm happy with the characterisation of ME/CFS as "complex". It's certainly not simple for our current understanding, even if the ultimate explanation might turn out to be relatively simple once it's recognised and understood. I know BPS have co-opted the word "complex" as part of their coded obfuscation, but it can be repatriated

If 'complex' is being used to say that we don't understand it, why not say 'poorly understood'? Then you avoid the problematic 'complex' word. By using 'complex' you also give governments and funders and researchers an excuse, i.e. the reason millions of people don't have a treatment for a debilitating disease is because 'it's really really complicated', whereas a substantial part of the problem is that there has not been nearly enough effort and funds expended on finding the answers.

 

[oldtimer]

If 'complex' is being used to say that we don't understand it, why not say 'poorly understood'?

I've always had a bit of a problem with 'poorly understood' which is trotted out in countless preambles and abstracts. It implies that some actual understanding of the condition exists, which doesn't seem to be the case. Why not just say that it isn't understood?

 

[Hutan]

Introduction
I'm continuing to get the feeling that the text is just a bit 'off'; it's not tight.

The chronic illness Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) evolves and perpetuates from a combination of biological and environmental determinants. Onset often occurs after a trigger event [1], such as viral infection, trauma, or toxin exposure, which induces a physiological response.

This seems to at once be saying things that are vaguely obvious but are also unevidenced. In some respects, doesn't every chronic illness evolve and perpetuate from a combination of biological and environmental determinants? In what way is ME/CFS different?

Such responses are typically transient, however are thought to become persistent or dysfunctional in ME/CFS, manifesting as idiopathic fatigue lasting for 3-months or longer, post-exertional malaise (PEM), brain fog, tender lymph nodes, dizziness, muscle or joint pain, digestive problems, or unrefreshing sleep [2].

In this sentence, it isn't just one physiological response, but a range of them. Reference 2 is the ICC criteria, which is an odd criteria to choose here. The sentence makes it seems as though all of those symptoms manifest in everyone with ME/CFS, all the time. This paper also talks about fatigue lasting for 3 months or longer, but the ICC does not give a time requirement and does not mention 3 months:

The Canadian Consensus Criteria were used as a starting point, but significant changes were made. The 6-month waiting period before diagnosis is no longer required. No other disease criteria require that diagnoses be withheld until after the patient has suffered with the affliction for 6 months. Notwithstanding periods of clinical investigation will vary and may be prolonged, diagnosis should be made when the clinician is satisfied that the patient has ME rather than having the diagnosis restricted by a specified time factor.

So the authors of this paper seem to have just made up the requirement for three months of idiopathic fatigue themselves.

The exact biological mechanism that results in a chronic ME/CFS state remains unidentified. However, accumulating evidence indicates anomalies in various biological systems including energy metabolism [3, 4], neuroendocrine function [5], immunology [6, 7], and autonomic regulation [8]. ME/CFS may be considered as a cluster of related, but distinct pathophysiological constructs [9], contrasting the reductionist view of it as a singular entity with stages of disease progression.

Saying 'the exact biological mechanism that results in a chronic ME/CFS state remains unidentified' suggests that the biological mechanism is mostly known. I don't think we can say that. It could be viral persistence, it could be autoimmunity, it could be something else entirely. If we can't even point with certainty to a broad description, I don't think it's reasonable to talk about 'exact biological mechanisms'. As @oldtimer suggests, it is probably best to just be upfront that we don't know.
Reference #7 is

Jahanbani F, Sing JC, Maynard RD, Jahanbani S, Dafoe J, Dafoe W et al. Longitudinal cytokine and multi-modal health data of an extremely severe ME/CFS patient with HSD reveals insights into immunopathology, and disease severity

That was that rambling paper about Whitney, i.e about one person. I don't believe that that should be being cited as evidence of immunological anomalies in ME/CFS. I don't know if it was cited due to this team being part of the OMF group?

ME/CFS may be considered as a cluster of related, but distinct pathophysiological constructs [9], contrasting the reductionist view of it as a singular entity with stages of disease progression.

The way that sentence is phrased, it is unclear if the authors are recommending that ME/CFS is considered as a cluster of related, but distinct pathophysiological constructs versus one disease that goes through a progression of stages or if they are just saying 'here are two contrasting views'. I don't think anyone would argue that there are not substantial levels of misdiagnosis in people with a label of ME/CFS.
'pathophysiological constructs'? Who is doing the constructing? Why not use 'diseases' or 'disorders'? Why tell us that it's a reductionist view - when they also tell us that the view is that there is a singular entity? You can't get much lower than 1 and still have something. And we know, ME/CFS doesn't follow set stages of progression. Not everyone ends up severe, and some people who start out severe improve to mild. The concept of one broad disease does not necessarily require the idea that the disease progresses. I'm probably not explaining well, but I don't think people who know ME/CFS well would write a sentence like that.

Additionally, the similarities between long COVID and ME/CFS pathophysiologies [10]—despite long COVID developing from a known viral origin (SARS-CoV-2 infection)—suggest that diverse symptom manifestations may be driven more by individual physiological response, rather than specific underlying causes.

I don't disagree with the point they are making, that the range of expressions of ME/CFS may be partly due host differences. But, long Covid is not really proof of that. We know that Long Covid is an umbrella term covering all sorts of illnesses, including things like lung damage, so some people don't have an ME/CFS-like illness at all.

The odd sentence that is a bit loose would be fine; it just might be different interpretations or something. But, there are so many sentences that aren't clear or quite logical. The paper needed a good proof-read.

 

[Hutan]

There are currently no definitive laboratory tests, and diagnosis is made based on exclusion.

I don't believe that there are laboratory tests of even some utility in positively diagnosing ME/CFS, never mind 'definitive'.

While comorbid conditions like long COVID, fibromyalgia (FM), and postural orthostatic tachycardia syndrome (POTS) are common in ME/CFS, the more critical diagnostic complication arises from symptom presentations that resemble pre-malignant states, and undiagnosed rheumatic diseases, neurological diseases and endocrinopathies, increasing the risk of misdiagnosis.
This not only delays proper treatment but also makes downstream data analysis more difficult by introducing unknown sources of heterogeneity into the patient population.

Again, the authors don't seem to have recognised that long Covid is simply the existence of new persistent symptoms after a Covid-19 infection. If someone meets the criteria for ME/CFS after a Covid-19 infection, they have ME/CFS. Someone with ME/CFS after a Covid-19 infection doesn't really have 'long Covid' as a co-morbidity, anymore than someone with ME/CFS after EBV has 'long EBV' as a co-morbidity.

It is not clear why some non-ME/CFS post-Covid pathology such as lung damage or kidney damage is less of a problem in terms of muddying the multi-omic waters than an undiagnosed rheumatic disease. I think the point that there is a high risk of misdiagnosis in people with ME/CFS labels could have been made more simply and compellingly.

 

[Hutan]

The heterogeneous nature and healthcare burden of ME/CFS present a timely opportunity for precision medicine, which aims to understand the molecular and biological factors that initiate and progress human diseases at the individual level [13]. This approach integrates biological data including genetic profiles, medical history, social, and behavioural information, to enable tailored decision-making for disease prevention, prediction, and treatment [14]. For example, in oncology, precision medicine has transformed care through targeted therapies for patients with specific molecular markers, such as HER2 protein in breast cancer [15] and EGFR gene mutations in non-small cell lung cancer [16].

I am skeptical about the concept of precision medicine when applied to ME/CFS right now. It works in oncology because clinicians have moved beyond 'cancer' to knowing specific types of cancer and how to treat them. And there is information about how people with specific genes respond to different treatments. Basically, it works because so much is known, and oncologists are now fine-tuning the knowledge.

Whereas, with ME/CFS, we are still at the 'a trigger might cause some sort of unknown physiological reaction that causes a chronic state as a result of some unknown mechanism'. I don't think you can usefully do 'precise', without doing 'accurate' first. For our disease, I think precision medicine is a trendy buzzword. Perhaps this paper helps build a case for ME/CFS research being eligible for pots of 'precision medicine' funding, in which case, all power to the authors. But, other than that, I'm not yet sure why 'precision medicine' is in the title of this paper. Maybe it will become clear.


The next paragraph seems to essentially say what I have said:

However, applying precision medicine to ME/CFS is more challenging due to the lack of well-defined pathology, reproducible biomarkers [17], and identifiable treatment targets. The goal in ME/CFS is to move beyond symptom-based classifications and focus on the biological mechanisms driving the disease [13]. This shift requires advanced computational tools such as machine learning and bioinformatic approaches to model the complex, multi-dimensional data and uncover the key pathways involved in the diverse ME/CFS presentations. Although omics studies have yet to identify definitive pathways in ME/CFS, recent advancements in computational power, growing datasets (data type, volume and sample size), and more efficient machine learning algorithms can reveal previously missed or hidden underlying mechanisms. Once these pathways are identified, the application of precision medicine can be fully realised through endpoints like (differential) biomarker-based diagnostics, patient subgrouping, and personalised treatments targeting specific pathways. This review outlines the essential machine learning steps, key multi-omics findings, and necessary data requirements for future ME/CFS studies implementing these computational frameworks.

That is, first we need to get an idea of what the ME/CFS pathology is.

But, that's in contrast to the sentences in the abstract:

The growing field of precision medicine offers a promising approach which focuses on the genetic and molecular underpinnings of individual patients. In this review, we explore how machine learning and multi-omics (genomics, transcriptomics, proteomics, and metabolomics) can transform precision medicine in ME/CFS research and healthcare.

Something has to exist before you transform it.

I've seen 'precision medicine' used to describe things as diverse as wastewater testing for Covid-19, genetic testing of newborns and apps that are supposed to help people quit smoking or lose weight. It can be used for anything technological. It's a vague term that I think is best avoided. Terms that really mean something, like 'computational biology' and 'individualised treatment' are better.