paturley
commented
3 years ago Hi KJ,
That is a great question. While at an individual SNP level, MTAG estimates should be unbiased and have standard errors with correct coverage (as long as MTAG's assumptions are satisfied), when you look at aggregated genome-wide summary statistics, the error in MTAG estimates can be correlated in funny ways that produce biased estimates. In general, the bias is in the direction of the corresponding statistic for the included traits. For example, if the heritability for trait 1 is h2_1 and the heritability for trait 2 is h2_2, then the heritability estimated using MTAG summary statistics will generally be something between h2_1 and h2_2.
All this said, I haven't investigated this very carefully. This is just based on informal observations of the results I've seen in the years since the paper was published.
Best, Patrick
On Wed, Apr 7, 2021 at 3:48 PM Kijoung Song @.***> wrote:
Hi, I'd like to hear your opinion in terms of this concern. One of my colleagues brought the following up. "GCTA-COJO computes an estimate of the phenotypic variance by looking at the distribution of effect sizes and standard errors genome-wide. I have no intuition at the moment what this will return with MTAG estimates, will these pheno variance estimates be the same, inflated, or under-estimated… this then impacts model selection in GCTA-COJO. Have you thought about this? " I’m eager to receive your feedback. Thank you, KJ
— 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/129, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5IUDL7VU4IKANBHXXDTHSZJPANCNFSM42RNQVZA .
Hi, I'd like to hear your opinion in terms of this concern. One of my colleagues brought the following up. "GCTA-COJO computes an estimate of the phenotypic variance by looking at the distribution of effect sizes and standard errors genome-wide. I have no intuition at the moment what this will return with MTAG estimates, will these pheno variance estimates be the same, inflated, or under-estimated… this then impacts model selection in GCTA-COJO. Have you thought about this? " I’m eager to receive your feedback. Thank you, KJ