Re: [PrediXcan/MetaXcan] MASHR direction of effect interpretation

114 views
Skip to first unread message

Alvaro Barbeira

unread,
Oct 9, 2020, 2:17:14 PM10/9/20
to sudha...@gmail.com, PrediXcan/MetaXcan
Hi Sudha,

For the sake of explicitness (even though I understand this is not your case): we avoid analyzing effect sizes obtained through S-PrediXcan. We focus on zscores.

Zscore direction can change depending on methodological differences on the GWAS, but in general this could mean the models are unreliable in those cohorts. 
I would investigate which variants are in play (the models are sqlite databases, and you can query them programmatically to this end for your gene/s of interest). 
The models contain variants that are at least moderately independent in the GTEx cohort, but that doesn't mean that variants couldn't be in LD in your cohort. Thus I would check the variants summary statistics in the GWAS, if they arise from imputation or were measured, etc.

Do many other genes have different directions in the two cohorts?

It makes sense to treat the second cohort as "replication", but I would investigate it further to check for any other potential source of error.

S-MultiXcan has more power to detect associations, but since it uses many models it can potentially amplify LD mismatches between GTEx and the target cohort. We typically refer to a separate colocalization measure like ENLOC for additional evidence on whether a causal mechanism underlies the gene-trait association.

Best,

Alvaro

On Thu, Oct 8, 2020 at 9:28 PM sudha...@gmail.com <sudha...@gmail.com> wrote:
Hi Alvaro,

I have a question about interpretation of direction of gene effect for any given tissue using mash-r models. Since this model borrows information from across tissues to improve prediction, is it advisable to run SPrediXcan on multiple tissues using this model or is SMultiXcan most recommended with mash-r for association? Also. how would we interpret the direction of effect? I am facing a situation where a gene is significant in two cohorts for the same trait-tissue pair using SPrediXcan but the direction of gene effect (from mashr zscore) is positive in one cohort and negative in the other. Both cohorts are EUR. How should I interpret this? Does it make sense to disregard the direction and treat the association in the second cohort as a "replication"?

Would love to hear your thoughts. Thanks!

Sudha

--
You received this message because you are subscribed to the Google Groups "PrediXcan/MetaXcan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to predixcanmetax...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/predixcanmetaxcan/6da066fd-2904-4c5e-974c-775a253951den%40googlegroups.com.

sudha...@gmail.com

unread,
Oct 10, 2020, 9:32:03 PM10/10/20
to PrediXcan/MetaXcan
Hi Alvaro,

Thanks for your response. Yes I do find other genes (~20) that have opposite effect direction between cohorts, which perhaps need to be investigated further. Does that raise a flag of some sort?

Sudha

Alvaro Barbeira

unread,
Oct 12, 2020, 10:53:53 AM10/12/20
to sudha...@gmail.com, PrediXcan/MetaXcan
Hi Sudha,

This is not necessarily worrisome or unheard of, but always worth of double checking. I would investigate the variants involved.

Best,

Alvaro



Reply all
Reply to author
Forward
0 new messages