Statistical testing and annotation of gene transcriptomic-neuroimaging associations

Multiscale integration of neuroimaging and gene transcriptome is becoming a widely used approach for exploring the molecular pathways of brain structure and function, in health and disease. Statistical testing of associations between spatial patterns of imaging-based phenotypic and transcriptomic data is key in these explorations, in particular establishing that observed associations exceed 'chance level' of random, non-specific observations. We discuss options for such statistical evaluations, including commonly applied linear regression, null model based on randomized brain regions that maintain spatial relationships, and null models built upon random effects that occur from other genes. Using examples and simulations of analyses as commonly performed in literature, we explain the caveats of these statistical models and provide guidelines for using proper models to evaluate both spatial and gene specificity. The null models are presented in a web-based application called GAMBA ("Gene Annotation using Macroscale Brain-imaging Association") that is designed for exploring transcriptomic-neuroimaging associations.