Can MTAG-enhanced GWAS for Trait A (jointly analyzed with Trait B) be used in genetic correlation and MR with Disease 2?

[zllfly-hub]

Subject: Inquiry Regarding the Use of MTAG for Cross-Phenotype GWAS Data in Downstream Genetic Analyses

Dear MTAG Development Team,

I hope this message finds you well.

I am writing to seek your expert guidance on a methodological question concerning the appropriate use of MTAG in a specific analytical context.

In our research, we are studying a complex disorder (“Disease 1”) that manifests through two correlated phenotypic traits—Trait A and Trait B. We are interested in leveraging MTAG to jointly analyze GWAS summary statistics for Trait A and Trait B, with the goal of obtaining an enhanced GWAS for Trait A that incorporates genetic information shared with Trait B. Subsequently, we would like to use this MTAG-derived Trait A summary statistic in downstream analyses—specifically, (1) estimating genetic correlation with a distinct condition (“Disease 2”) using LD Score Regression, and (2) conducting Mendelian randomization (MR) analyses to assess potential causal effects of Trait A on Disease 2.

However, we are concerned about potential biases that might arise if the MTAG procedure inadvertently introduces signal from Trait B (which may itself be associated with Disease 2) into the Trait A summary statistics. This could, in principle, lead to inflated genetic correlations or spurious causal estimates in MR, even if Trait A has no true association with Disease 2.

Could you kindly advise whether this application aligns with MTAG’s intended use? Specifically:

Is it methodologically appropriate to use MTAG-enhanced summary statistics for one trait (e.g., Trait A) in genetic correlation or MR analyses when the other MTAG input trait (Trait B) is biologically related to the outcome (Disease 2)?
Are there known caveats or recommended sensitivity analyses we should consider in such a scenario?
We greatly appreciate the development of MTAG and your ongoing contributions to the field. Any insights you could share would be invaluable to ensuring the rigor of our analyses.

Thank you very much for your time and consideration.

Warm regards,
linlin Zhao

[paturley]

Thanks for your question. MTAG can sometimes affect genetic correlation estimates. The reason for this is that estimation error in MTAG summary statistics can be correlated with the effect sizes for other traits that are input into MTAG. This was illustrated nicely in Hill et al. https://www.nature.com/articles/s41380-017-0001-5 In your case, if you are interested in the genetic correlation between trait A and trait C, and you use MTAG to boost the power of the trait A summary statistics, the estimated genetic correlation will be some combination of the genetic correlation between A and C and between B and C. So as a worst case scenario, if rg_AC = 0 and rg_BC > 0, then using MTAG summary statistics may produce rg estimates that are driven by the true rg_BC value. We didn't really consider this problem when the paper was written, so we didn't fully test how big this problem is and under what circumstances. As a result, I don't know that I have any advice other than acknowledging this limitation. Best, Patrick

…

On Mon, Dec 1, 2025 at 8:52 AM zllfly-hub ***@***.***> wrote: *zllfly-hub* created an issue (JonJala/mtag#246) <#246> Subject: Inquiry Regarding the Use of MTAG for Cross-Phenotype GWAS Data in Downstream Genetic Analyses Dear MTAG Development Team, I hope this message finds you well. I am writing to seek your expert guidance on a methodological question concerning the appropriate use of MTAG in a specific analytical context. In our research, we are studying a complex disorder (“Disease 1”) that manifests through two correlated phenotypic traits—Trait A and Trait B. We are interested in leveraging MTAG to jointly analyze GWAS summary statistics for Trait A and Trait B, with the goal of obtaining an enhanced GWAS for Trait A that incorporates genetic information shared with Trait B. Subsequently, we would like to use this MTAG-derived Trait A summary statistic in downstream analyses—specifically, (1) estimating genetic correlation with a distinct condition (“Disease 2”) using LD Score Regression, and (2) conducting Mendelian randomization (MR) analyses to assess potential causal effects of Trait A on Disease 2. However, we are concerned about potential biases that might arise if the MTAG procedure inadvertently introduces signal from Trait B (which may itself be associated with Disease 2) into the Trait A summary statistics. This could, in principle, lead to inflated genetic correlations or spurious causal estimates in MR, even if Trait A has no true association with Disease 2. Could you kindly advise whether this application aligns with MTAG’s intended use? Specifically: Is it methodologically appropriate to use MTAG-enhanced summary statistics for one trait (e.g., Trait A) in genetic correlation or MR analyses when the other MTAG input trait (Trait B) is biologically related to the outcome (Disease 2)? Are there known caveats or recommended sensitivity analyses we should consider in such a scenario? We greatly appreciate the development of MTAG and your ongoing contributions to the field. Any insights you could share would be invaluable to ensuring the rigor of our analyses. Thank you very much for your time and consideration. Warm regards, linlin Zhao — Reply to this email directly, view it on GitHub <#246>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBUB5N5HE4RA2VBV4RYOTT37RBYRAVCNFSM6AAAAACNVWQX32VHI2DSMVQWIX3LMV43ASLTON2WKOZTGY4DCMJRHA2TCNY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[zllfly-hub]

Thank you very much for your reply. I hope this message finds you well.

We are currently interested in estimating the genetic correlation between Disease 1 and Disease 2. However, we do not have access to GWAS summary statistics for Disease 1 itself. Instead, we know from prior evidence that Disease 1 is clinically and genetically characterized by two core phenotypic components: Trait A and Trait B.

Given this, we are considering using MTAG to jointly analyze GWAS summary statistics for Trait A and Trait B—thereby generating an enhanced GWAS for Trait A that incorporates shared genetic information from Trait B—and then using this MTAG-derived Trait A summary statistic as a proxy for Disease 1 in downstream genetic correlation and possibly Mendelian randomization analyses with Disease 2.

We would greatly appreciate your expert insight on the potential methodological concerns with this approach. Specifically:

Does this strategy risk overestimating or misrepresenting the true genetic relationship between Disease 1 and Disease 2?
Are there alternative strategies you would recommend under these data constraints?
We recognize that MTAG is designed to boost power under shared genetic architecture, but we are cautious about unintended bias when using the output as a surrogate for a broader clinical syndrome.

Thank you very much for your time and guidance. We greatly value your contributions to the field and would be grateful for any advice you can offer.

Warm regards,
linlin Zhao