Accelerated estimation and permutation inference for ACE modeling
Research output: Contribution to journal › Article › Academic › peer-review
There are a wealth of tools for fitting linear models at each location in the brain in neuroimaging analysis, and a wealth of genetic tools for estimating heritability for a small number of phenotypes. But there remains a need for computationally efficient neuroimaging genetic tools that can conduct analyses at the brain-wide scale. Here we present a simple method for heritability estimation on twins that replaces a variance component model-which requires iterative optimisation-with a (noniterative) linear regression model, by transforming data to squared twin-pair differences. We demonstrate that the method has comparable bias, mean squared error, false positive risk, and power to best practice maximum-likelihood-based methods, while requiring a small fraction of the computation time. Combined with permutation, we call this approach "Accelerated Permutation Inference for the ACE Model (APACE)" where ACE refers to the additive genetic (A) effects, and common (C), and unique (E) environmental influences on the trait. We show how the use of spatial statistics like cluster size can dramatically improve power, and illustrate the method on a heritability analysis of an fMRI working memory dataset.