Relationship between major depressive disorder and [ME/CFS]: a two-sample mendelian randomization study analysis, 2025, Zhu et al

I did find that the lead author, Wenjing Song, has done at least two mendelian randomization studies before.

For Daytime napping and the incidence of Parkinson’s disease: a prospective cohort study with Mendelian randomization, I looked at the SNPs in the paper, and compared them to the SNPs in the 23andMe paper they cite on daytime napping, and most of them match.

For the other, I didn't see any SNPs listed.

 

Ok thanks, that's useful to know.

 

Seriously impressive digging again on here.

I’m afraid I can’t help, to pass on a couple of things Chris has mentioned to me in general. First, because UK buying bank data is so accessible, it’s used in a lot of GWAS Studies, and some of these are of questionable quality. Second, Mendelian randomisation studies are sophisticated and hard to get right.

These are general comments and don’t tell us anything about this study, but the apparent errors are not a great sign. Hopefully, the authors will clarify things

 

Did they ever respond? If not, it might be worthwhile to post something to PubPeer. I drafted this up and could post it:

Edit: Or feel free to post something yourself since it was your discovery.

 

Yes but not sure if I can copy their reply so I'll try to summarise.

They basically said that they did not use the 102 significant SNPs highlighted in the publication by Howard et al. (2019). Instead they screen the full dataset where summary data for all tested SNPs is listed. They don't say this explicitly but I believe this info is available in the text file named ‘PGC_UKB_depression_genome-wide.txt’ on this webpage, which the Howard 2019 publication refers to:
https://datashare.ed.ac.uk/handle/10283/3203

But even then, the data don't seem to match from what I can tell. The SNP listed in the supplementary material still do not seem to be associated with depression and the p-values don't match. To take one example: the first SNP in the supplementary material is rs1000940 for which Song et al. report a p-value (pval.exposure) of 1.63E-08. But in the text file listed by Howard et al. 2019 I found a p-value of 0.5313.

 

I'm concerned by their response. It sounds like they're saying they independently found 57 new SNPs associated with MDD from a GWAS already analyzed by others, which sounds like a study worth a paper of its own, but its just a small section of another study that then uses those SNPs in mendelian randomization.

I ran a script to get all 57 rows from that PGC text file that match Song et al's SNPs:

 

Thanks, got the same, only one that met the 5*10^-8 threshold.

Have replied this to the researchers. Curious what the explanation will be.