How LD clumping was performed to select eQTLs for candidate gene?

[Shicheng-Guo]

Dear Mike,

I am wondering How LD clumping was performed to select eQTLs for candidate gene in eQTL-based two-sample Mendelian Randomization? How to handle pleiotropy eQTL issue? for example, a SNP is eQTLs for candidate gene, but also for multiple other genes? Do we have any strategy to select gene-specific eQTL?

Thank.

Shicheng

[mikelove]

hi Shicheng,

Thanks for the questions.

In the paper, we used clumping on the eQTL data (both for the sim and real data examples, same parameters):

https://github.com/mikelove/mrlocusPaper/blob/master/sim/Snakefile#L212-L228

We are currently also using conditional QTL as provided by e.g. APEX or other upstream tools. I think conditional QTL will likely give better instruments and better gene-to-trait estimation.

Horizontal pleiotropy is not explicitly modeled in MRLocus, but instead we rely on multiple instruments with low correlation to attempt to control FP of gene-to-trait effects. With few instruments, our credible intervals grow large anyway. We have simulations of horizontal pleiotropy in the paper (Supp Fig H gives an example of such a sim where a locus contains variants that are trait-only effects mixed in). We performed well there -- at least in the simulation framework, which we didn't create but just tweaked to allow for HP.

MRLocus does not currently have any methods to deal with multiple genes in a locus, each with overlapping regions contributing to heritability of expression level AND also > 1 gene mediating trait heritability. TMWR does this, but we don't have this currently.

MRLocus is designed for investigating loci with strong causal gene candidates, whereas other methods that estimate gene-to-trait effects for many genes in a locus simultaneously may have less biased estimation of the effect, when a strong gene candidate is not present. Future work on the MRLocus model may involve estimation of the mediating effect of candidate causal genes in the context of other relevant genes in a pathway and a polygenic background

[mikelove]

Oh, in practice, we generally start with TWAS genes, as we need TWAS significance (speaking roughly) in order for the gene to also be MRLocus significant. Another note is that, in practice, multi-mapping reads and overlapping genes can be a real issue (i.e. Saha and Battle's paper on multi-mapping and QTL analysis).

[Shicheng-Guo]

Thank you Mike. It totally make sense!!

Shicheng

[Shicheng-Guo]

I am wondering do we have public available independent identified eQTL for GTEx-V8 based on APEX or GCTA?
Thanks
Mike