MTAG with perfect genetic correlation option and equal heritability

[bdilor]

Hello,

I would like to ask for your feedback on the usage of the perfect genetic correlation with or without the equal heritability options in MTAG.
Briefly: I am trying to perform a MTAG analysis on three traits that are three definitions of the same condition. The definitions are obtained via two different questionnaires (one is more accurate than the other), and ICD codes. I am running MTAG with the perfect genetic correlation option (since rg among traits is > 0.7) but not sure about the h2 option. The h2 of the three traits is significant but not the same across them (spanning from 4% to 14%).
What puzzles me is the fact that I get identical results for all the three traits after MTAG under the perfect_gencov option, at least in terms of significant loci, and these are mostly not present in the original GWAS of the three traits. When the equal_h2 option is added, 8 signals are lost and 8 are gained in comparison to the results obtained with the perfect_gencov option, but most of the obtained signals are the same of the single trait analyses. Summary figure included as attachment.

I was wondering what is actually driving the observed differences among the used options? Since the condition under examination is the same, but identified by three different definitions, what is in your opinion the best assumption to make for running MTAG (only genetic correlation or genetic correlation with equal heritability)?

Thanks!

[paturley]

Hello, If you use the perfect genetic correlation option, I believe that you should get identical p-values for each phenotype, but the beta coefficients will differ by a constant to reflect the differences in heritability. (Logically, this should be the case because any SNP that is associated with one trait should be associated with all the traits if they are truly perfectly genetically correlated.) If you assume equal heritability and it's true, then this assumption will lead to slightly more statistical power. If you assume it and it's false, then I think you will get a weighted average of effect sizes across the three traits, with greater weight on the (truly) more heritable phenotype. The best assumption depends on what you believe is mostly likely to be true. If you are unsure, I generally prefer to make the least restrictive assumption (non-equal h2 in this case).

…

On Mon, Dec 16, 2024 at 11:05 AM bdilor ***@***.***> wrote: Hello, I would like to ask for your feedback on the usage of the perfect genetic correlation with or without the equal heritability options in MTAG. Briefly: I am trying to perform a MTAG analysis on three traits that are three definitions of the same condition. The definitions are obtained via two different questionnaires (one is more accurate than the other), and ICD codes. I am running MTAG with the perfect genetic correlation option (since rg among traits is > 0.7) but not sure about the h2 option. The h2 of the three traits is significant but not the same across them (spanning from 4% to 14%). What puzzles me is the fact that I get identical results for all the three traits after MTAG under the perfect_gencov option, at least in terms of significant loci, and these are mostly not p resent in the original GWAS of the three traits. When the equal_h2 option is added, 8 signals are lost and 8 are gained in comparison to the results obtained with the perfect_gencov option, but most of the obtained signals are the same of the single trait analyses. Summary figure included as attachment. I was wondering what is actually driving the observed differences among the used options? Since the condition under examination is the same, but identified by three different definitions, what is in your opinion the best assumption to make for running MTAG (only genetic correlation or genetic correlation with equal heritability)? Thanks! MTAG.png (view on web) <https://github.com/user-attachments/assets/fc39753a-1feb-47b9-8821-f3802c4e528c> — Reply to this email directly, view it on GitHub <#226>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBUB5OADDDTQSFJAOS65ZD2F325HAVCNFSM6AAAAABTWPFCKCVHI2DSMVQWIX3LMV43ASLTON2WKOZSG42DENZYHE3DAOA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>