Covariance-based sample selection for heterogenous data: Applications to gene expression and autism risk gene detection
share
Risk for autism can be influenced by genetic mutations in hundreds of genes across the genome. Based on findings showing that genes with highly correlated gene expressions are functionally interrelated, "guilt by association" methods such as DAWN have been developed to identify autism risk genes. Previous research in this direction analyze the BrainSpan dataset, which contains gene expression of brain tissues from varying brain regions and developmental periods. Because the covariance among gene expression has been shown to vary with respect to the spatiotemporal properties of brain tissue, previous research was restricted to the subset of samples originating from a particular brain region and developmental period known to be associated with autism. While this was done to avoid the issue of heterogeneity, it also led to a potential loss of statistical power when detecting risk genes. In this article, we develop a new method to find a subset of samples that share the same population covariance matrix to retain a larger and more homogenous set of samples for the downstream DAWN analysis. Based on this analysis, we identify a risk gene set with greater enrichment in an independent study.
READ FULL TEXT