Preprint Dissecting the genetic complexity of myalgic encephalomyelitis/chronic fatigue syndrome via deep learning-powered genome analysis, 2025, Zhang+

[Sasha]

My layman’s understanding is that the machine learning model doesn’t add info. The algorithm it applies might be rubbish, but it doesn’t change the underlying data.

I thought it wasn't providing that data, though. I might be wrong.

 

[jnmaciuch]

I think we can take a standard GWAS or WGS analysis as providing a true signal plus noise but do we know that about this machine-learning technique? Could it b simply rubbish plus noise?

I don’t think so, it survived cross validation plus recapitulation on an independent cohort. It’s likely not a complete story, but I don’t think there’s any reason to disregard it. The math is a little fancy but as a machine learning approach, it’s nothing that I would expect to hallucinate things out of thin air. It’s a far cry from LLMs, for example.

 

[Jonathan Edwards]

Having looked at the genes a bit and thought a bit more I am quite sure these data are telling us something real. ME/CFS involves:

Synapses
Epigenetic regulation of DNA transcription
T cell and/or related innate immune cell responses

And maybe a number of other plausible steps but the other things look more at risk of being things thrown up in any disease for rather boring reasons.

None of these are at all surprising but up to this point we have had no hard evidence of anything. We now have the evidence for a real biological process (or possibly two that look the same but are actually separate and in parallel). I still think the method used here may mean that we are getting an uneven picture of what is there - but it is still picture of what is there. If there is a real process then DecodeME ought to pick up something as well with its big numbers. The different methodology may mean that quite different genes are picked out, but that doesn't matter if they are illustrating the same overall story.

What to me is particularly exciting abut all this is that the data suggest that ME/CFS is based on a process that interferes with nerve cell biology but not necessarily through anything to do with inflammation. Which fits with the fact that nerve cells are clearly suffering in ME/CFS but we cannot find any obvious inflammation. It also puts emphasis on regulation of transcription - on misrouted signals or messages rather than tissue damage. And although some activation pathways like MAP kinase seem to come up there is nothing specifically on inflammatory cytokine involvement. The top 'cytokine' is leptin, which seems to do something altogether more subtle.

The immunological signal is not that strong and not very specific but that is not particularly unexpected. The data do not immediately point to a particular mechanism. But then nor did they in RA - it took nearly thirty years for us to work out exactly what the DR4 link meant. But it gave the confidence to pick through the relevant pathways and devise treatments.

I think the data may be telling us a lot about why we have not been finding anything so far in the effector biology and where we should be looking instead. If anything I think it confirms various thoughts people have been having on that. But the main thing is that we are no longer wandering about in a sea of uncertainty. There is definitely something on the end of the fishing line this time.

 

[mariovitali]

I submitted the genes listed earlier by @Jonathan Edwards to the framework I have been using. Below are results from one ranking method pointing mainly to DNA methylation and phosphorylation. Of interest, liver comes up again along with glycoproteins and lipid metabolism. Note also insulin and PPARGC1A which is related to mitochondrial biogenesis. Also autophagy, not sure why it takes this high ranking.




I have also been working lately in creating a reasoning engine framework - I won't get into details here but will post an update- but this work points to the cell membrane as a primary target of intervention, followed by plasmalogen+peroxisome function restoration, antioxidant and mitochondrial support. I very much agree with @ME/CFS Skeptic that we shouldn't mention to these engines any term related to ME/CFS, LongCOVID, PostViral syndromes etc.

So, I submitted the genes listed to the reasoning engine framework and ask if these genes connect with its proposed hypothesis (A peroxisome-mitochondria-lipid metabolism failure model). Here is snapshot of the process (not all information is shown):

 

[Sasha]

We now have the evidence for a real biological process [...] If there is a real process then DecodeME ought to pick up something as well with its big numbers. [...] The data do not immediately point to a particular mechanism. But then nor did they in RA - it took nearly thirty years for us to work out exactly what the DR4 link meant. But it gave the confidence to pick through the relevant pathways and devise treatments.

If these findings are looking real then the big question is, how can we use them to advance most rapidly to treatment? Are they not worth much (black box, unreplicated) until the DecodeME results come out? Are they pointing precisely enough at mechanism to start up any lab work? Are they helpful for the paper that you're writing with Jo Cambridge and Jackie Cliff?

I'm wondering whether Zhang's approach has been used in other disease areas and has had its findings replicated by traditional WGS studies.

 

[Jonathan Edwards]

If these findings are looking real then the big question is, how can we use them to advance most rapidly to treatment?

I am not sure that this particular set of data on their own point to specific drug targets but there are already ideas on that and once we have results from DecodeME and have shaken things down a bit I think potential targets will show themselves. Fluge's continued work on antibody targeting may have some merit but it would good to get a clearer idea what role antibody plays if it does.

The Zhang data don't really need replicating. It is more a question of having a few more jigsaw puzzle pieces so that they can be fitted together to get the centre of the picture.

 

[Sasha]

The Zhang data don't really need replicating. It is more a question of having a few more jigsaw puzzle pieces so that they can be fitted together to get the centre of the picture.

Why don't the Zhang data need replicating? Normally with studies, especially small ones, we want everything potentially worthwhile replicated, don't we?

 

[Jonathan Edwards]

Why don't the Zhang data need replicating? Normally with studies, especially small ones, we want everything potentially worthwhile replicated, don't we?

We mostly want replication when there are significant fears of bias in sampling or lab method or some such. For genetic data like this that isn't going to be biased because a machine churns out one thing and not another thing (i.e. there is an internal control comparison) and by and large nobody is rooting for any particular result. The machines are pretty reliable - having quality controls built in. And as @jnmaciuch points out, the study already includes some replication.

And finally, I don't think there is more than one chance in a million that a set of genes with so much overlap in function should turn up by mistake. It's synapses. It must be.

 

[Utsikt]

It's synapses. It must be.

Do we have any idea where the synapses are located or what type of synapse that are affected?

 

[Sasha]

We mostly want replication when there are significant fears of bias in sampling or lab method or some such.

We're not worried about bias in selection of cases?

 

[hotblack]

Do we have any idea where the synapses are located or what type of synapse that are affected?

Maybe it depends? We don’t all have exactly the same symptoms…

 

[Jonathan Edwards]

We're not worried about bias in selection of cases?

Not really. If selection is poorly defined that should just dilute results. The only real worry is that the cases all had some there illness but the results do not seem to point to that.

 

[Jonathan Edwards]

Do we have any idea where the synapses are located or what type of synapse that are affected?

I don't think so.

 

[Sasha]

Do we have any idea where the synapses are located or what type of synapse that are affected?

I don't think so.

Tons of us have OI. Does anyone know what synapses would have to be messed up for OI to occur? Are any OI-type genes showing up in the Zhang results?

 

[V.R.T.]

I don't think so.

So is it possible the errant signaling might be affecting any and all synapses? Just sort of randomly throwing spanners in the works? That could explain why PEM symptoms can be so variable from person to person and even crash to crash.

Or does it have to be somewhere specific in the brain or nervous system? And if that is the case, can we treat it without knowing precicely which synapses are affected? I suppose if there are already some ideas about drug targets the answer is possibly.

 

[Robert 1973]

I don't think there is more than one chance in a million that a set of genes with so much overlap in function should turn up by mistake. It's synapses. It must be.

Sorry if this is a dumb question but could it be an antigen acting on the synapses which is causing symptoms?

 

[Kitty]

It's synapses. It must be.

So the biochemical signals that get passed across the space, rather than electrical signals? Or could it be either?

(I've long had my ten bob on an ion channel disorder, but I guess there are all sorts of things that could go wrong in that space.)

 

[Jonathan Edwards]

Does anyone know what synapses would have to be messed up for OI to occur?

We don't know the mechanism of the OI so that's hard to say.

 

[Jonathan Edwards]

Sorry if this is a dumb question but could it be an antigen acting on the synapses which is causing symptoms?

Did you mean antibody? An antigen is just any molecule that in the given context reacts with an antibody.

There are antibodies to neurotransmitter receptors but their levels aren't very different from normals. I think this is an unlikely scenario. The genes coming up seem to have to do with housekeeping processes that regulate patterns of synapses put in use or taken out of use.

 

[Jonathan Edwards]

So the biochemical signals that get passed across the space, rather than electrical signals? Or could it be either?

These genes control the maintenance of the whole synaptic machinery I think. That includes neurotransmitter action and generation of electrical potentials through ion channel opening and closing.

I think this is an area of biology we haven't really got our heads around on the forum and it is very complicated. Hopefully we will be able to follow the story as it unfolds.