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

Interesting! They're very similar in function, so I think that'd be quite the coincidence if they both appeared by chance.

Wikipedia

 

I think that's more good evidence that the function itself is involved in pathogenesis! NME1, NME2, and NME3 also all showed up in the top 115, as well as several proteasome subunits and DLGAPs. This tells me that it doesn't so much matter which part of the pathway is broken, but rather that proper functioning of the pathway itself is important.

 

Also I take back my critique about cytotoxic CD4s—I just noticed that the dot plot in figure 4 shows CD4. I must not have been able to read the tiny text when I was looking at the figures on my phone.

generally I still prefer to show markers as feature plots since only a proportion of cells in the cluster are expressing those markers and it might not be the same cells. But for these purposes I think it’s sufficient. I would, however, still like to see more figures showing the actual expression differences of the leading edge genes by cluster.

[edit: it might not be a signature that’s exclusive, or even strongest, in cytotoxic CD4s but rather that was the only cluster where the pathway enrichment was significant because of other factors]

 

I looked up the DLGAPs out of curiosity and thought it was interesting that like NLGN1 and NLGN2, the DLGAP genes (DLGAP1, DLGAP2, DLGAP3, DLGAP4 from the study) code for proteins located in a structure called the postsynaptic density. And these ones too:

SYNGAP1

From Wikipedia:

SHARPIN
Paper: Sharpin, a novel postsynaptic density protein that directly interacts with the shank family of proteins (2001, Mol Cell Neurosci)

I went through every GeneCards link and searched the pages for mentions of synapses. These ones also mention postsynaptic density: CAMK2A, GRM1, DLG2, HOMER2

And these ones mention synapses in general: CACNA2D3, ACE, PRKCZ, GABBR1, NRAS, BCL2L1, CA2, PDYN

Edit: Missed one of the DLGAPs before.

 

Interesting AMPD1 came up fairly high in the list. It's been discussed before in ME/CFS but I always assumed you needed to be have a homozygous mutation. I checked Wikipedia and it discusses that it may not necessarily be black and white regarding what causes a deficiency.

https://en.wikipedia.org/wiki/Adenosine_monophosphate_deaminase_deficiency_type_1

This is very interesting

Maybe on its own a mutation in AMPD1 is not damaging, but when combined with another disease it makes that issue worse.

Or it could be that people homozygous with AMPD1 are misdiagnosed with ME/CFS.

 

That sort of thing seems to come up a lot in chronic illnesses—a mutation that would be deleterious if homozygous, and should be fine if heterozygous, but obviously isn’t under some perfect storm of homeostatic stress. It’s always the assumption that the non-deleterious allele should be able to compensate adequately, but as in all biology it’s important to look at the edge cases.

In those situations I think it’s useful to look at situations where maximal functionality is really necessary—under circumstances like viral infection, nutrient deficiency, etc. etc. where you really need something to turn over properly

 

So what about the first 20-25 genes (cuts off most easily at 23)
Can we make a table of the proteins and functions for these and maybe check further down for sister genes where they show up?

DNMT3A. DNA methyltransferase 3 alpha Epigenetic DNA methylation -gene expression control
ADCY10 adenylyl cyclase 10. Formation of cAMP
PPP2R2A. Protein Phosphatase 2 Regulatory Subunit Balpha. Cell cycle
NLGN2. Neuroligin 2. Synapse formation
LEP Leptin Weight control/appetite
SYNGAP1. Synaptic Ras GTPase Activating Protein 1. Synapses MAPkinase signalling
AHCYL2. Adenosylhomocysteinase Like 2. Brain signalling
NLGN1. Neuroligin 1 Synapse formation
DLGAP4. DLG Associated Protein 4. Synapses
HDAC1. Histone Deacetylase 1. Regulation of gene transcription
AMPD2 adenosine monophosphate deaminase 2
AHCYL1. Adenosylhomocysteinase Like 1. Anti-inflammatory cytokine production
SHARPIN. SHANK Associated RH Domain Interactor. Signalling in auto inflammation. Synapses
NME2. NME/NM23 Nucleoside Diphosphate Kinase 2. DNA transcription. Risk factor for EBV-associated lymphoma
NME1-NME2. NME1-NME2 Readthrough. DNA transcription
CACNA2D3. Linked to brain development and autism TCR signalling
NME3. NME/NM23 Nucleoside Diphosphate Kinase 3. Goes with NME1
ZC3H13. NME/NM23 Nucleoside Diphosphate Kinase 3. RNA splicing
CAMK2A. Calcium/Calmodulin Dependent Protein Kinase II Alpha. Synapses on dendrites in brain
PIK3CA. Phosphatidylinositol-4,5-Bisphosphate 3-Kinase Catalytic Subunit Alpha. Insulin responses, brain development
MAX. MYC Associated Factor X. DNA transcription
HLA-C HLA-C (MHC I) CD8 T cell receptor and NK cell receptor recognition events
ACE. Angiotensin I Converting Enzyme

 

That is certainly a quick way to get an answer!
But it seems to have brought in all the usual memes about neuroinflammation and autoimmunity and stuff.

Nevertheless, grist to the mill.

 

There seems little doubt that the list points to signalling systems and neural systems in particular. Although it may be of note that neural synapses and immune cell 'synapses' can share mechanisms and do for some these genes.

It looks as if some of these genes do point to skeletal muscle dysfunction being relevant. But the fact that people are well before they get ME/CFS suggests maybe that these kick in once some signalling pathway has disturbed immune or neural regulation and these genes are important for muscle coping in an adverse signalling environment - as suggested already.

A lot of the genes seem to relate to high-level general regulatory processes such as DNA transcription and in that sense may not give very specific clues but there is still a sense that they may be important for 'learnt' dysregulation - rather as proposed for 'learnt' innate immune responses.

The fact that leptin has come up again seems unlikely to be a coincidence.

I think we are seeing real stuff here. And when DecodeME reports I think there must be a high chance that at least some things will chime with this, even if the different methodology may pick out different specific genes. Then we should surely be motoring. It is beginning to make sense I think. And some of the stuff from before we were uncertain about is likely to turn out to have been real. Having a gene signal just makes it so much easier to be confident about.

 

SequenceME intends to examine the DNA of up to 17,000 PwME with whole genome sequencing, presumably because they think the big numbers are necessary in order to give a useful signal. I don't understand much about what's going on in this paper but can we trust results based on only 247 cases?

 

That seems unlikely, given that that team seem desperate to get answers for PwME as soon as possible.

I wonder if they want big numbers to go for a classic approach because deep learning models might not be trustworthy. They seem very black box. I worry that we might overlook the opacity of the method if the results fit with our ideas of what ME/CFS might be. Because lots of ideas have flown about over the years, there are perhaps quite a few models to choose from.

I'm speaking from a position of pig-ignorance but should we worry that we're projecting our wishes onto these results? Confirmation bias?

 

I think that raises the question of 'why not', though. If a small, deep-learning study spits out lots of statistically significant genetic associations, do we take them seriously or not? If we have to wait for better information, is this information not reliable? And if it isn't, what's the point of doing such studies?

 

I understanding is that the bare minimum for whole genome sequencing is 1000, which I think is bigger than this study. But as with GWAS Bigger is much better. Sequencing is incredible expensive, so I’m pretty sure the largest sample size is for greater power to understand the problem. But I’d be surprised if they got funding for as many of 17,000.

 

I think it is worth remembering that these numbers for genetic studies are the numbers people think you should collect in order to be reasonably sure that you have picked up most of the significant genetic links. Some genetic links will turn up with very small numbers.

For instance, 95% of people with ankylosing spondylitis have HLA-B27 whereas in the general population it is 8%. Ten patients and ten controls would almost certainly be significant if you just looked at HLA-B. The numbers you need also depend on how many possibilities are likely to turn up with your chosen method - SNPs, rare gene panel etc.

I don't think we should discount results on 250 cases if the stats look solid. Moreover, I am pretty sure that these results are not noise. There are definite patterns and replications from before.

 

I’m pretty sure the experience of GWAS was that you do need decent sample sizes to get reliable results – not just more results. The early history was getting almost everything wrong. The exceptions are where you are looking for very big differences, like your ankylosing spondylitis example, and also APOE2 in Alzheimer’s disease. I’m sure George Davey Smith published papers on this, which led to a big change in methodology, including much bigger sample sizes.
Perhaps the telling thing was that in the GWAS field they noticed they had a replication crisis and did something about it.

I don’t know if that experience translates to these newer types of analysis

 

I’m jealous of your laptop (and have found even the smaller Gemma3 models and certainly Gemini 2.5 Flash useful for exploring topics and helping me understand roughly what papers or people are talking about at times, not sure asking them to deduce is worth much beyond a bit of fun though)

That’s key isn’t it. These are newer methods to find the signals in the data and it is hoped they will find things more efficiently. But they are newer, different methods from what has been used in the past. GWAS is a known quantity.
I really like the questioning though @Sasha and think you’re right to raise the points. Hopefully all will become clear and we’ll understand more in the coming months.

 

I think big numbers is probably more important for SNP fishing, although I know less about the rare gene approach.