MTAG tests whether the SNP is associated with none of the traits

[Steven-ZWQ]

Dear Dr. Turley, MTAG constructs a Bonferroni-corrected test (P= n * min{p1,p2, ..., pn} ) to examine the joint null hypothesis, while a paper reported a MTAG beta with MTAG p-value for each shared variant? I'm confused. Can MTAG directly test whether the SNP is associated with none of the traits? I look forward to hearing back from you. Thank you.

Sincerely, Wenqiang

[paturley]

Hello Wenquiang,

I'm a bit confused by your question. MTAG generates betas and p-values corresponding to each trait included in the analysis. I've seen people then use these p-values to generate an aggregate p-value related to the minimum p-value among the traits, but those approaches were not tested in the MTAG paper and they are not part of the MTAG algorithm.

Does this help at all?

Patrick

On Sat, Jan 14, 2023 at 2:37 AM Steven-ZWQ @.***> wrote:

Dear Dr. Turley, MTAG constructs a Bonferroni-corrected test (P= n * min{p1,p2, ..., pn} ) to examine the joint null hypothesis, while a paper (PMID: [image: 34561436] https://user-images.githubusercontent.com/60351256/212461637-378a9887-9005-4b80-b3bd-c8fdc943c738.PNG) reported a MTAG beta with MTAG p-value for each shared variant? I'm confused. Can MTAG directly test whether the SNP is associated with none of the traits I look forward to hearing back from you. Thank you.

Sincerely, Wenqiang

— Reply to this email directly, view it on GitHub https://github.com/JonJala/mtag/issues/174, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5LTV5J3HR3JNEQAMB3WSJJTXANCNFSM6AAAAAAT3EHVOM . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[carbocation]

Not that Github Issues are the right spot for a discussion, but just to follow on here: I think this is the paper referenced by @Steven-ZWQ . In that paper, the Methods states "We implemented MTAG options that assume equal SNP heritability for each trait and perfect genetic covariance between traits."

I don't agree with (or at least, don't understand the scientific rationale for) claiming that Multiple Sclerosis and IBD share perfect genetic covariance and have equal SNP heritability. Having said that, the technical point here is that when run in this mode, MTAG produces a single set of summary statistics (e.g., it acts like meta-analysis that accommodates potentially overlapping samples). So if the thrust of the question was how to make MTAG produce one set of sumstats, I think the answer, in this case, is --equal_h2 and --perfect_gencov. (And again, I am not endorsing it for this scientific question.)

[paturley]

Ah OK. Yes. If the traits have equal h2 and rg=1, which are the assumptions that those options impose, then the MTAG solution should produce identical summary statistics for each trait and therefore MTAG only produces one set of results. Imposing those assumptions makes MTAG equivalent to meta-analysis (though accounting for potential sample overlap between sets of summary statistics).

On Tue, Jan 17, 2023 at 2:58 PM James Pirruccello @.***> wrote:

Not that Github Issues are the right spot for a discussion, but just to follow on here: I think this is the paper referenced https://www.nature.com/articles/s41467-021-25768-0#Sec11 by @Steven-ZWQ https://github.com/Steven-ZWQ . In that paper, the Methods states "We implemented MTAG options that assume equal SNP heritability for each trait and perfect genetic covariance between traits."

I don't agree with (or at least, don't understand the scientific rationale for) claiming that Multiple Sclerosis and IBD share perfect genetic covariance and have equal SNP heritability. Having said that, the technical point here is that when run in this mode, MTAG produces a single set of summary statistics (e.g., it acts like meta-analysis that accommodates potentially overlapping samples). So if the thrust of the question was how to make MTAG produce one set of sumstats, I think the answer, in this case, is --equal_h2 and --perfect_gencov. (And again, I am not endorsing it for this scientific question.)

— Reply to this email directly, view it on GitHub https://github.com/JonJala/mtag/issues/174#issuecomment-1385976985, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5J2LMZPNB77ZSXIXR3WS32YBANCNFSM6AAAAAAT3EHVOM . You are receiving this because you commented.Message ID: @.***>