|
Abstract:
|
For cohorts with omics data from multiple tissues, combining information across tissues remains a challenge. Joint analyses are often not possible due to tissue-specific model covariates (e.g. cell composition), and traditional meta-analysis methods are not appropriate because of the assumption that the summary statistics being combined are independent. To address this gap, we propose a multivariate random effects meta-analysis method to incorporate correlation between datasets, which we expect to observe when we are combining omics summary statistics from multiple tissues in the same individuals. We show in a simulation study that when datasets are independent, our summary effect estimates are highly correlated with the estimates from a standard multivariate random effects meta-analysis (r>0.99), but when datasets are correlated, the standard random effects meta-analysis underestimates the standard errors of the summary effect estimates, incorrectly decreasing the coverage of the 95% confidence intervals. We applied our method to 446 mother-infant pairs in Gen3G, testing for associations between placenta and/or cord blood DNA methylation and multiple maternal glucose measures.
|
