Meta-analysis and effective sample size calculation

[Stephen Wang]

Hi all,

I'm wondering how to conduct fixed-effect meta-analysis using Genomic SEM with SNP effect? I specified the model like this:

F1 =~ 1*Pheno1 + 1* Pheno2

F1 ~ SNP

Pheno1 ~~ 0* Pheno1

Pheno2 ~~ 0* Pheno2

The "F1~SNP" should be the meta-analysis result, is it correct?

Besides, when calculated the effective sample size using the formula provided on Github, I got unreasonable results. One effective sample size for latent factor is 10 times larger than any input sumstats sample size. And another sample size is less than any input sumstats. What might be the reason?

Thank you so much.

Best,

Stephen

[agro...@gmail.com]

Hi Stephen, 

The code looks correct and you are right that F1~SNP is the meta-analytic result. An effective sample size that is slightly attenuated is not entirely unexpected. Effective Ns will be less than observed Ns due to imbalance of cases/controls if you are using binary traits, and if you are including traits that are not super highly correlated, or one trait is much lower powered, you might expect some attenuation in signal. A couple follow-up questions to help diagnose what may be going on in the case of the super inflated signal you highlight: 

1. What is the genetic correlation and h2 of the traits you include? 

2. What is the univariate LDSC intercept if you put the F1~SNP summary statistics back into ldsc? 

3. If you calculate the mean of the absolute value of the betas over the SEs for the sumstats output what do you get? I'm wondering if for the super inflated effective N you might have a trait that is getting incorrectly processed through that part of the pipeline such that Z-statistics are hugely inflated. 

Best,

-Andrew