While it might be tempting, I find it difficult convince myself that hundreds of metabolic traits have a direct causal effect on CAD and T2D, given the widespread pleiotropic effects many variants have on these traits.
Comments
Log in with your Bluesky account to leave a comment
As an example, genome-wide MR suggested that branched-chain amino acids (BCAAs) are (weakly) associated with T2D. To explore this association further, we turned to cis-MR and focussed our attention to genes that have direct effect on BCAA catabolism (BCAT2/DBT/PPM1K).
Here's the putative causal DAG. Instead of estimating the genome-wide average effect of BCAAs on T2D (which could be confounded by many factors), we are now asking if altered protein function of BCAT2/DBT/PPM1K might have a causal effect on T2D risk.
I think we need to be very careful when intepreting genome-wide MR estimates that include many 'anonymous' trans-acting genetic variants as instruments (term coined by @mendelrandom.bsky.social).
This work was led by brilliant PhD student Ralf Tambets. Many thanks to all of our co-authors and collaborators @urmovosa.bsky.social @adriaan-vd-graaf.bsky.social @zkutalik.bsky.social @genomicsdoge.bsky.social @jaanikakronberg.bsky.social @dzvinn.bsky.social
@kristafi.bsky.social
Totally agree with this Kaur. I have also been playing with the data (thanks for sharing!). I found that nearly all univariate MR associations with CAD were highly diminished or disappeared in a multivariate MR including APOB. IMO this should be the baseline analysis for this kind of data.
@kauralasoo.bsky.social what I mean by biochemical properties would be saturated lipid/unsaturated lipid, fatty acids, cholesterol, lipid n-carbon etc. The pink dots segregate directionally and should track biochemical properties, reassuring rather than puzzling imo. Tag @ericfauman.bsky.social
@kauralasoo.bsky.social the majority (90% or more) of 293 metabolites in nightingale are lipids and so this is just tagging lipid metabolism being causally linked to CAD/T2D. In this plot if you try to color the scatter by biochemical properties you’d probably see it more clearly.
I agree! One of the reasons we included this plot was to illustrate you should not treat the 249 traits as independent when performing MR (as is often done). Leaving lipid traits aside, I would still be worried about intepreting genome-wide MR that includes a bunch of anonymous trans variants.
Comments
@kristafi.bsky.social