has anyone tackled mendelian randomization before with good ole’ summary stats? i’m gonna dive into that realm with my stats crew who are way more capable than me when it comes to analysis but i was curious what folks think about it. is it appealing or frowned upon in the genetics/genomics realm?
Hey, @ecebayram, thanks for sharing! From what I’ve heard/read, Mendelian randomization is commonly the method of choice for genetics group statistics. A relatively recent example in a PD genome context: Finding genetically-supported drug targets for Parkinson’s disease using Mendelian randomization of the druggable genome.
Being that my PhD was in tinnitus and did not involve genetics (
), I’m curious what other community members with the relevant experience would say about using Mendelian randomization for group statistics? @betamaro , @fbbriggs , @VesnavM, @lucasf, @ViniciusCrr, @sherinechan, @maylabel
1 Like
Yes, I’ve published a handful of MR studies (e.g. Telomere Length and Parkinson’s Disease Traits: A Mendelian Randomization Study ). It is definitely a useful tool for hypothesis generating and/or causal inference. The challenge are the causal assumption that it is based on - and using genetic instrumental variables for “exposures” that only reflect a component of the exposure. For example, many MR studies using smoking GWAS hits as GIVs have been done, but what most fail to interpret is these smoking GIVs only reflect smoking behavior/initiation (nicotine addiction) and not the full breadth of what smoking does (i.e. oxidative stress, DNA damage, post-translation modifications, etc). There are several great primers out there: A Guide to Understanding Mendelian Randomization Studies - PMC . Happy to follow-up!
2 Likes
This is awesome, thank you so much for sharing the paper! Do you feel like there’s a sample size restriction? For instance, I focus primarily on Lewy body dementia and the sample size for the GWAS so far is relatively small (in thousands and not 10 thousands), would it be worthwhile at all?
You are set with effect estimates from a few thousands as MR uses the SE, but best practice is to calculate the F-statistic for the potential GIV beforehand (>5 value is ok, but >10 is a safe bet).
1 Like