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

On looking again, I realized I misunderstood this.

C2_v2
COVID-19 positive (controls include untested)

C2_v2_england_controls
COVID-19 positive (controls include untested), only patients from centers in England

There's nothing different about the groups in terms of lab testing, the second one is the same thing, just only including participants in England. I got confused because "controls" got added to the code for the England group for some reason.

So both groups include the cases group where people tested positive for COVID and the control group where people either tested negative or didn't get tested.

 

On the one hand I’ve had both at different times and found they’re very very different conditions, almost polar opposites in some ways. So I find people talking about similarities difficult to understand.

On the other I’ve had both at different times so there may be a genetic predisposition.

It’s certainly an interesting one.

Thanks for getting your head around this on our behalf! I found that whole chunk of the paper confusing tbh. I get their goal but am unsure if it tells us much. There could just be lots of people in the UK Biobank flagged in their records as having had depression or covid, both very high prevalence conditions over time (and with very wide ranging severity).

 

I was depressed when I got covid and ME/CFS as a result of that.

To me, there are no similarities at all, other than reduced activity in general. I got help with my depression during the first year of being ill, and it ‘s night and day.

 

But would you not expect that a decent chunk of the people in the depression group in the BioBank might have ME/CFS? It might feel different when you know the differences, but if a patient has ME/CFS but they or their doctor don't know about that condition, I think many people would assume they have depression instead, mainly based on the drastically reduced activity levels.

 

Yes, there is overlap because depression questionnaires are quite bad and would flag many normal traits for sick people as depression.

I think we were talking about the experience of actually being depressed vs having ME/CFS.

 

Yes, @Utsikt puts it well.

I agree there may be some crossover in groups but that is generally a failure of those giving the diagnosis. To me there are not similarities in the illnesses but misunderstanding and poor diagnosis.

When I was depressed I was physically able to do things but didn’t see the point and to be honest would have been fine with my life just stopping.

Now I have ME/CFS I am physically unable to do things but really want to and am fighting to make the most of life and one day have more of it back.

In my view/experience these are trivial for anyone with an understanding of both conditions to differentiate between. Probably with a few simple questions.

I’m simplifying a bit and there are crossovers and complexities, people fan have both, but that they are often seen as similar is a I think a failure to understand either condition by medical practitioners.

 

I have the same experience. And I’ve also got the assessment of a therapist that followed me the first 2.5 years after getting sick.

She was adamant that there is nothing wrong with my mental health now, yet I would have scored high on most surveys.

That’s the issue with most psych labels - they don’t consider the circumstances or alternative explanations for the results.

 

On the statistical point, given depression is reported by what, 20-25% of people? Maybe more over a lifetime. And IBS is very common and almost everyone got covid, the associations with them just seem very sketchy to me. It’s like looking for a correlation between people that had lunch today and people with these gene variants.

Maybe I’m being overly cynical or missed something they did to factor this in but it seems like a stretch to me to layer this on top of their earlier findings. It detracts rather than adds IMHO.

 

It's been a week since you submitted your comment and it's still not up.

 

Sorry, I had actually deleted the comment after the author replied by email the first time. (No second reply yet but I followed up.) But I made a new comment on MedRxiv yesterday (still in moderation) since a response on there would probably be better so that anyone who views the preprint can see the genes.

Edit: I removed the MedRxiv comment to not annoy the authors. If they can share it, they'll get back to me by email.

 

Sorry, I haven't been following the thread! Do you feel we're further forward with understanding what they found and what its implications are?

 

I don't think much more has been figured out in the thread, just some speculation.

 

The authors have sent me Supplementary Table 2 and allowed me to share it here. They also informed me that they've uploaded it (and I assume the other tables) to MedRxiv, but it'll take some time for it to be made visible. I attached the spreadsheet to this post which appears to have every gene they tested, but here are the top 115 that were used in the model:

Edit: Links to GeneCards added.

 

Thank you! Very excited to dig into it. Several of the top markers are very interesting, DMNT3A, HDAC1, [edit: CHD8] etc. are mediators of certain epigenetic marks.

 

Yes, a very interesting list.
Lots of things relevant to central nervous and immune regulation.
I think these are going to turn out to be of real significance. It is just a bit difficult to judge to what extent the order is telling us something and to what extent it depends more on chance technical issues relating to individual rare genes.

 

I think we were all hoping for some specific top hits, but it does seem to be a little all over the board. There's hits for calcium signaling, epigenetic markers, protein kinases, enzymes involved in cAMP and MAPK signaling, purine metabolism, a lot of random one-offs like gonadotrophin releasing hormone and transcription factors associated with MYC.

Hard to find a throughline between many of them, though it might be possible to connect some to other findings in the field.

 

I was hoping that their model attention score would be higher for a select few rather than a gradient like we're seeing. I'd suggest filtering above a q-value cut off and then looking at the ranking by attention difference rather than p-value.

 

Insulin and leptin both being in the top is also very striking...

 

I see that both NLGN1 and NLGN2 are very high up. Is it that they're in LD and that's likely why both appeared?

 

It looks like the two genes are on different chromosomes so LD is not likely. My understanding is that rare variant analyses are generally much less susceptible to LD confounding