paturley
commented
3 years ago We never explored this very fully in the paper, so I don't have any careful analysis to point to justify this claim, but based on a handful of subsequent examples that I've seen, the genetic correlation of MTAG estimates obtained from the same MTAG analysis tend to be biased upwards due to the correlation of the error between the set of summary statistics. Also, as you say, there is also a risk of false positives when the true genetic correlation between a pair of phenotypes is moderate and difference in mean chi2 between the two samples is large. I'd recommend you calculate the maxFDR to calculate the worst case scenario FDR in your particular setting. That will hopefully give you a better sense of whether the results you are finding are just trait1 loci or if they are feasibly trait 2 loci.
Best, Patrick
On Tue, Nov 17, 2020 at 12:06 PM Kijoung Song notifications@github.com wrote:
Hi,
I ran MTAG with two traits (data1=better power, but less specific, data2=less power, but more specific). The genetic correlation of both gwas results was 62% (p=1.8e-47). However, after mtag analysis, data1 gwas and data2 mtag showed almost perfect genetic correlation = 99% (p=0). Both manhattan plots are very similar.. (almost identical).. question? In this case, I think the mtag results are more relevant to the genetic contributions to data1 than they are to the genetic contributions to data2. What do you think of it? Summary of MTAG results:
Trait ... GWAS equiv. (max) N 1 data1.txt ... 442804 2 data2.txt ... 40218
[2 rows x 7 columns]
Estimated Omega: [[9.676e-07 1.276e-06] [1.276e-06 4.481e-06]]
(Correlation): [[1. 0.613] [0.613 1. ]]
Estimated Sigma: [[0.928 0.114] [0.114 0.984]]
(Correlation): [[1. 0.119] [0.119 1. ]]
MTAG weight factors: (average across SNPs) [0.994 2.708]
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/JonJala/mtag/issues/119, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5MTXNV7XDUV4AOG4JLSQKUSZANCNFSM4TY3BPEA .
Hi,
I ran MTAG with two traits (data1=better power, but less specific, data2=less power, but more specific). The genetic correlation of both gwas results was 62% (p=1.8e-47). However, after mtag analysis, data1 gwas and data2 mtag showed almost perfect genetic correlation = 99% (p=0). Both manhattan plots are very similar.. (almost identical)..
question? In this case, I think the mtag results are more relevant to the genetic contributions to data1 than they are to the genetic contributions to data2. What do you think of it?
Summary of MTAG results:
Trait ... GWAS equiv. (max) N 1 data1.txt ... 442804
2 data2.txt ... 40218
[2 rows x 7 columns]
Estimated Omega: [[9.676e-07 1.276e-06] [1.276e-06 4.481e-06]]
(Correlation): [[1. 0.613] [0.613 1. ]]
Estimated Sigma: [[0.928 0.114] [0.114 0.984]]
(Correlation): [[1. 0.119] [0.119 1. ]]
MTAG weight factors: (average across SNPs) [0.994 2.708]