Brilliant - I never would have found that.
For example, if ALL pathological ME variants are absent on the UKB Axiom array, how sure can we be that imputation against the general population will cause them to be associated with pwME?
Preprint Initial findings from the DecodeME genome-wide association study of myalgic encephalomyelitis/chronic fatigue syndrome, 2025, DecodeMe Collaboration
Brilliant - I never would have found that.
I don't think I understand imputation well enough to answer. In any case, we know there is a lot of the genome data missing in DecodeME. Imputation can only get you so far, and some of the imputed variants might be wrong.
A whole genome sequencing study, AKA SequenceME, would be very valuable so that we could actually measure all the positions instead of making educated guesses about all the non-measured positions.
Jonathan Edwards
Senior Member (Voting Rights)
My impression is that 'imputation' is used to cover more than one inferential process, at least one involving individual SNP profiles (imputing intervening SNPs) and another making use of population weightings of SNP forms (imputing risk genes from minihaplotypes of SNPs).
On the TNFSF4 (also known as OX40L):
a paper on lupus (SLE) quoted by Forestglip said that an up-regulation of TNFSF4/OX40L (the OX40 ligand) predisposes to lupus. I think TNFSF4 activates CD4+ t-cells:
BUT DecodeME seems to be suggesting the opposite for ME/CFS - a decreased expression of TNFSF4, at least in some defined tissues
Members have noted that some drugs that reduce expression of TNFSF4 are coming to market and there was a suggestion that it could be tried in ME/CFS.
If I have understood things correctly, (and I may well not have, in which case, let me know), knocking down expression of TNFSF4 in people with ME/CFS would be very unlikely to help. The genetic variant found by DecodeME that reduces TNFSF4 expression might only be relevant to increasing the risk of developing ME/CFS, so an OXO40L monoclonal that reduces TNFSF4 expression might not make symptoms worse, but there doesn't seem to be a good rationale to try it.
?
a paper on lupus (SLE) quoted by Forestglip said that an up-regulation of TNFSF4/OX40L (the OX40 ligand) predisposes to lupus. I think TNFSF4 activates CD4+ t-cells:
Another quoted paper suggests that blocking the ligand and receptor can improve lupus:
BUT DecodeME seems to be suggesting the opposite for ME/CFS - a decreased expression of TNFSF4, at least in some defined tissues
Members have noted that some drugs that reduce expression of TNFSF4 are coming to market and there was a suggestion that it could be tried in ME/CFS.
If I have understood things correctly, (and I may well not have, in which case, let me know), knocking down expression of TNFSF4 in people with ME/CFS would be very unlikely to help. The genetic variant found by DecodeME that reduces TNFSF4 expression might only be relevant to increasing the risk of developing ME/CFS, so an OXO40L monoclonal that reduces TNFSF4 expression might not make symptoms worse, but there doesn't seem to be a good rationale to try it.
?
Last edited:
I think there are often situations where the same variant can be associated with increases or decreases of gene expression depending on the tissue, or maybe depending on other factors like age.
For example, a variant from a past study:
Apart from that, the gene expression database isn't comprehensive of every tissue that might be affected, both because there might be too low of statistical power and because not every possible cell type/tissue type is included in the database.
So just noting that it's possible this variant might lead to increased expression where it matters for ME/CFS. Or of course you might be right and it could be a dead end.
V.R.T.
Senior Member (Voting Rights)
So I may be misremembering but the Sanofi OX40L monoclonal was claimed by the company to have a t cell calming/soothing effect, and JE and others have theorised T Cells may be central to the mechinism of MECFS. So I was going from that and the connections I saw iirc.
But i didnt realise it was down not up in DecodeME.
Kitty
Senior Member (Voting Rights)
Thank you! Bookmarked.
I have spent an hour or two grappling with phasing and imputation, but realize the task is very large and complex. It just bothers me that most of the SNPs in the preprint are inferred by fiendish arithmetic.
As you say, SequenceME will be a blessing. Is there currently ANY whole genome data for ME? Is SequenceME the only WGS project on the horizon?
I think just much smaller studies or studies that didn't use as strict of definitions. For example, the UK BioBank has WGS data for their chronic fatigue syndrome phenotype, and it was included in a rare variant study, but the cohort is smaller than DecodeME and the definition of CFS is more permissive.
There's this study on only 20 cases of severe ME/CFS: A Network Medicine Approach to Investigating ME/CFS Pathogenesis in Severely Ill Patients: A Pilot Study, 2024, Hung, Davis, Xiao
This study included a few hundred cases with WGS data obtained from three cohorts (Stanford, CureME, and Cornell): Dissecting the genetic complexity of myalgic encephalomyelitis/chronic fatigue syndrome via deep learning-powered genome analysis
Just the ones I can think of right now, there might be more. I'm not sure about other future projects.
Last edited:
Seems like the 9 genes they identified ACADL, BRCA1, CFTR, COX10, HABP2, MFRP, PCLO, PRKN, and ZFPM2 do not show up in DecodeME.
I WILL look at this (and the 24 pages of comments), but might you have a TLDR?
Thanks so much for your help - all your hard efforts are obviously appreciated by many many s4me folks, and I just wanted to add my appreciation.
Quick and potentially not totally accurate summary: They created a machine learning model designed to predict whether participants were cases or controls based on rare variants in their DNA. The prediction algorithm was connected to an external database of known protein interactions (STRING) which influenced the training of the model.
I think it went something like: if cases tend to have variants in a certain gene (gene A), then the model will train to also look for variants in other related genes (related based on STRING database) for later predictions, based on the assumption that if cases have variants in gene A, and if gene A is highly related to gene B, then future cases are also more likely to also have variants in B, even if the specific cases that were tested did not necessarily have any variants in B.
After training, the researchers looked at the model weights (the inner workings of the trained model) to see which genes it prioritized for making predictions. In the paper they highlighted the 115 most highly prioritized genes as being the "ME/CFS genes" (genes copied to this post). But the limitation above applies that some of these high priority genes might not have actually had many rare variants in the cases tested, so some of these genes may or may not actually be relevant to ME/CFS.
Then they looked to see what kinds of processes these 115 genes (or subsets of them) tend to be involved in, and found that these genes seem to be involved in several areas, with the two main ones highlighted in the paper being synaptic function and proteasome function.
Andy
Senior Member (Voting rights)
Once the ongoing remaining analyses are complete. Current guess when that might be? Some time next year, dependent on the peer review process.
Jonathan Edwards
Senior Member (Voting Rights)
I don't think this matters. Formal publication used to ensure wide dissemination and that the study was as well written up as could be expected. Those are already sorted for studies like this.
Robert 1973
Senior Member (Voting Rights)
Do you anticipate that any of ongoing analysis will be made available before it is formally published in a peer reviewed journal?
Andy
Senior Member (Voting rights)
Good question and something that hasn't been discussed internally yet. Personally, unless there were to be good reasons not to, I would be in favour of publishing a revised version of the current preprint, adding the results of the additional analyses, before submitting it for peer review.
You should be writing these papers - what a clear and crisp summary. I am very grateful - you have inspired me to now start a close reading of the paper and comments. Thanks again.
Dolphin
Senior Member (Voting Rights)
Update VHS conference 2025: DecodeME poster finalist.
Go to,
https://vhscotland.org.uk/vhs-poster-competition-finalists/
https://vhscotland.org.uk/wp-conten...VHS-conference-poster-Oct-2025.jpg-scaled.jpg
Go to,
https://vhscotland.org.uk/vhs-poster-competition-finalists/
https://vhscotland.org.uk/wp-conten...VHS-conference-poster-Oct-2025.jpg-scaled.jpg
Responding to this from another thread.
Relevant part of DecodeME paper:
So my understanding is that SNP-based heritability, as reported above, only captures the genetic influence from what was measured in the study: common variants. That means it misses any genetic influence of rare variants. And I don't think it captures structural changes like copy number variants and inversions.
Also, I think GWAS tests assume a certain model, which I think is most commonly the additive model, meaning having one copy of the risk allele increases risk by some amount and having two alleles doubles the risk. As opposed to a dominance model, where having one or two alleles causes equal increases in risk, or a recessive model, where only having both risk alleles can increase risk. Regenie documentation says you can choose the model to use. If you set it to additive, I think you would miss a lot of the influence from recessive or dominant mutations.
Here is an image from a paper visualizing the genetic contributions to autism spectrum disorder (ASD), schizophrenia (SCZ), and Alzheimer's disease (AD). For autism, for example, the genetic heritability is estimated at 82% of total liability (liability includes genetic and environmental influences). Yet the portion based on common variants is only around 12%. There's also some heritability due to rare variants, non-additive variants, plus 59% of liability is genetic heritability which has not been attributed to a specific factor.
From variants to mechanisms: Neurogenomics in the post-GWAS era, 2025, Neuron
I think total genetic heritability will be better captured by family studies, though I'm not sure how exactly they disentangle shared environment.
Jonathan Edwards
Senior Member (Voting Rights)
Yes, that is my understanding. However, my memory is that other data fit with ~10% and Chris gave the impression (I thought) that there were reasons to think this was most of the risk. My guess is that if genes are really rare they don't make much difference and if they aren't rare then it is a bit unfair if none of them show up on the GWAS, but others will know more than I do. I would be surprised if the h2 value was above 20%. Which means we have enough of a ball park estimate to guide both theory building and clinical advice.