search

MRCIEU / TwoSampleMR

R package for performing 2-sample MR using MR-Base database
https://mrcieu.github.io/TwoSampleMR
Other
420 stars 176 forks source link

don't understand clumping results #106

Open RDMelamed opened 5 years ago

RDMelamed commented 5 years ago

Hi, I'm learning how to use this package and I'm a bit confused and grateful for help. My "dat" is below. I don't understand the results from clump_dat on this, which removes rs10792021 and rs143141392. I thought the point is to remove SNPs in LD, but none of them are near any other SNP. Can anyone explain or point me to any documentation for how this works? Thanks! Rachel

          SNP chr_name chrom_start pval.exposure
1  rs61839209       10    18645086      8.32e-07
2   rs6485533       11    44507543      1.54e-07
3  rs10792021       11    68859870      6.86e-07
4 rs143141392        3    74772076      9.24e-07
5 rs143366687        4    22306890      1.43e-07
6   rs9350504        6    74457830      1.35e-10
7 rs192850832        7   110586940      9.29e-07
8   rs6948871        7   127643968      7.71e-07
9 rs139630857        7    30288256      9.54e-07
explodecomputer commented 5 years ago

It might remove them if the rs ID isn’t in the reference dataset being used, which only includes SNPs with maf > 0.01 in Europeans. Equally, if the rs ID has changed then it might remove it also.

From: RDMelamed notifications@github.com Reply to: MRCIEU/TwoSampleMR reply@reply.github.com Date: Tuesday, 13 August 2019 at 14:38 To: MRCIEU/TwoSampleMR TwoSampleMR@noreply.github.com Cc: Subscribed subscribed@noreply.github.com Subject: [MRCIEU/TwoSampleMR] don't understand clumping results (#106)

Hi, I'm learning how to use this package and I'm a bit confused and grateful for help. My "dat" is below. I don't understand the results from clump_dat on this, which removes rs10792021 and rs143141392. I thought the point is to remove SNPs in LD, but none of them are near any other SNP. Can anyone explain or point me to any documentation for how this works? Thanks! Rachel           SNP chr_name chrom_start pval.exposure 1  rs61839209       10    18645086      8.32e-07 2   rs6485533       11    44507543      1.54e-07 3  rs10792021       11    68859870      6.86e-07 4 rs143141392        3    74772076      9.24e-07 5 rs143366687        4    22306890      1.43e-07 6   rs9350504        6    74457830      1.35e-10 7 rs192850832        7   110586940      9.29e-07 8   rs6948871        7   127643968      7.71e-07 9 rs139630857        7    30288256      9.54e-07 — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

RDMelamed commented 5 years ago

Thank you! Is there some way to flag these different reasons? Or do you think that a SNP that is removed for one of these reasons should probably not be used in european populations?

explodecomputer commented 5 years ago

There isn’t a way to flag it at the moment unfortunately, but to minimise loss of variants in clumping due to this perhaps split the data to only do it on sets of SNPs for which you need to?

I do think it’s an interesting question regarding what recommendations for SNPs to use for different populations. From a practical perspective, most GWAS in MR-Base is from European populations and the clumping function is just there for convenience really, and only has European reference data. There’s a reasonable chance that a SNP not in the reference won’t be widely available in the GWASs, and proxies won’t be available either. So, the decision may be made automatically anyway. But technically there is nothing wrong with using rare variants.

Other considerations are that SNP effects in different populations might not be consistent for a whole range of reasons, e.g. different biology, different gxe, different LD. It might introduce heterogeneity when the exposure and outcome are not representatitve of the same population.

From: RDMelamed notifications@github.com Reply to: MRCIEU/TwoSampleMR reply@reply.github.com Date: Tuesday, 13 August 2019 at 15:09 To: MRCIEU/TwoSampleMR TwoSampleMR@noreply.github.com Cc: gibran hemani explodecomputer@gmail.com, Comment comment@noreply.github.com Subject: Re: [MRCIEU/TwoSampleMR] don't understand clumping results (#106)

Thank you! Is there some way to flag these different reasons? Or do you think that a SNP that is removed for one of these reasons should probably not be used?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

explodecomputer commented 5 years ago

I've added these considerations as a new feature request here #107, will try to get round to it but it might not be very soon. Thanks for your feedback!