JonJala
Using local regions to conduct MTAG is ok ?
Hi, I want to use local regions that the local genetic correlation are more than 0.7 to conduct MTAG between two traits. Is it could lead to bias? The other question i want to ask is that the beta_mtag is equal to beta of single GWAS? I am extremely grateful for your assistance in resolving this issue!!
Thank you, ChaoWang
Depends a bit on how you select the regions. If it is based just on the local rg, then you'll get a number of regions that have a high rg by chance even if the true rg is low. This would lead to you oversharing information across traits than is optimal and some potential bias. Ultimately, however, this is just my conjecture. You'd have to test it out in simulations or something if you want to implement MTAG in a way different than has been formally tested in the paper.
The beta_mtag should be the same as the beta in a single GWAS if the GWAS was done for a quantitative trait that had been standardized. Otherwise, the betas will be a constant multiple of the betas for the GWAS.
Best, Patrick
On Thu, Oct 12, 2023 at 4:45 AM cwnag-c @.***> wrote:
Hi, I want to use local regions that the local genetic correlation are more than 0.7 to conduct MTAG between two traits. Is it could lead to bias? The other question i want to ask is that the beta_mtag is equal to beta of single GWAS? I am extremely grateful for your assistance in resolving this issue!!
Thank you, ChaoWang
— Reply to this email directly, view it on GitHub https://github.com/JonJala/mtag/issues/190, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5JYVVIAX3NQRSSZFS3X66U3LANCNFSM6AAAAAA55H7RPY . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Thank you, if i use the local regions that are both high rg and P<0.05, is that better? And i want to know the results (Z_mtag for same snp) used whole genome whether is the same as used local region?
Any selection on rg will potentially lead to bias, but I don't know how much bias. Maybe if you restrict to highly significant regions, it won't be as bad, but it's something that you will have to test.
I don't understand your second question. Can you clarify?
On Thu, Oct 12, 2023 at 10:50 PM cwnag-c @.***> wrote:
Thank you, if i use the local regions that are both high rg and P<0.05, is that better? And i want to know the results (Z_mtag for same snp) used whole genome whether is the same as used local region?
— Reply to this email directly, view it on GitHub https://github.com/JonJala/mtag/issues/190#issuecomment-1760683656, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5N4VALN3YTW3QJEWH3X7CUBDAVCNFSM6AAAAAA55H7RP2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONRQGY4DGNRVGY . You are receiving this because you commented.Message ID: @.***>
For example, i just select snps in ch1 to conduct MTAG (MTAG_chr1) between two traits , while MTAG uses all chrs. rs3748595 is in chr1. Is the beta_mtag of rs3748595 in MTAG same as the beta_mtag of rs3748595i n MTAG_chr1 in theory?
It won't necessarily be. MTAG estimates the genetic correlation between the two traits using the available SNPs. Given the sample of SNPs will be different for the chr1 analysis, the genetic correlation will be different and therefore the beta will be different. If you estimate the Omega and Sigma matrix using all of the SNPs and feed those directly into MTAG, then you should get the same beta whether you use the full genome or just a single chromosome.
On Fri, Oct 13, 2023 at 9:42 PM cwnag-c @.***> wrote:
For example, i just select snps in ch1 to conduct MTAG (MTAG_chr1) between two traits , while MTAG uses all chrs. rs3748595 is in chr1. Is the beta_mtag of rs3748595 in MTAG same as the beta_mtag of rs3748595i n MTAG_chr1 in theory?
— Reply to this email directly, view it on GitHub https://github.com/JonJala/mtag/issues/190#issuecomment-1762467516, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5OMVAGOUFRPV7XYZ73X7HUZTAVCNFSM6AAAAAA55H7RP2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONRSGQ3DONJRGY . You are receiving this because you commented.Message ID: @.***>