Effect Directionality and Similarity of Results Across Traits

[zjppdozen123]

Thank you for providing such a useful package! I have three questions about MTAG:

1. I am running MTAG on GWAS of two biomarkers. The effects of these biomarkers point in opposite directions with respect to disease risk: for biomarker 1, higher levels indicate higher risk, while for biomarker 2, higher levels indicate lower risk. The original GWAS for each trait was performed using their native directionality. When applying MTAG, do I need to align the effect directions across traits? For example, if SNP x has a positive beta (risk allele) for trait A but a negative beta (also risk allele) for trait B, should I flip one so they are consistent?

2. When I run MTAG with only two traits, I find that the signals identified in the MTAG-adjusted GWAS for trait 1 and trait 2 are almost identical, showing the same loci with only small differences in p-values. These signals seem to be primarily driven by trait 2 (they are weak in the original GWAS of trait 1 but strong in trait 2). Is this expected behavior (genetic correlation ≈ 0.3), and how should I interpret it?

3. When I run MTAG with three traits, I noticed an interesting pattern. For one SNP that was not significant in the original GWAS of trait 1, the MTAG-corrected GWAS for trait 1 became significant after including traits 2 and 3 (p-value changed from 5e-03 to 5e-08). However, the same SNP was significant in the original GWAS of trait 3, but in the MTAG-corrected GWAS for trait 3 it became much less significant (p-value changed from 5e-07 to 0.05). How can this pattern be explained?

[paturley]

Hi! Sorry for the slow response here. It's been grant season. 1) MTAG takes the sign of the genetic correlation into account, so it shouldn't matter if there are sign differences for the two traits. (In the example from the original paper, we used depressive symptoms and subjective well-being, which also have opposite-signed effects.) 2) With a genetic correlation of 0.3, I'd expect only a small amount of information sharing across phenotypes. If there is a big power imbalance between the two phenotypes and a moderate genetic correlation, I'd be a little worried about the false discovery rate. I generally recommend people calculate the maxFDR in their setting and see if the maximum potential FDR is small enough for them to be confident in their new results. Best Patrick

…

On Tue, Sep 30, 2025 at 3:25 PM zjppdozen123 ***@***.***> wrote: *zjppdozen123* created an issue (JonJala/mtag#243) <#243> Thank you for providing such a useful package! I have two questions about MTAG: 1. I am running MTAG on GWAS of two biomarkers. The effects of these biomarkers point in opposite directions with respect to disease risk: for biomarker 1, higher levels indicate higher risk, while for biomarker 2, higher levels indicate lower risk. The original GWAS for each trait was performed using their native directionality. When applying MTAG, do I need to align the effect directions across traits? For example, if SNP x has a positive beta (risk allele) for trait A but a negative beta (also risk allele) for trait B, should I flip one so they are consistent? 2. When I run MTAG with only two traits, I find that the signals identified in the MTAG-adjusted GWAS for trait 1 and trait 2 are almost identical, showing the same loci with only small differences in p-values. These signals seem to be primarily driven by trait 2 (they are weak in the original GWAS of trait 1 but strong in trait 2). Is this expected behavior (genetic correlation ≈ 0.3), and how should I interpret it? — Reply to this email directly, view it on GitHub <#243>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFBUB5JE4Q2CLIEVLJUQTCD3VLKKJAVCNFSM6AAAAACH52ECPCVHI2DSMVQWIX3LMV43ASLTON2WKOZTGQ3TANZZG43TINA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>