Multi - Scale Gaussian Mixtures for Cross - species Study

Cross-species studies using microarray gene expressions help discover genetic diversity among different species, the results of which are fundamental to comparative genomics. Various approaches have been used for cross-species studies, such as homogeneity test and cluster analysis. A homogeneity test provides a homogeneity significance ranking for each gene pair whilst cluster analysis looks at the discovery of co-expressed meta-genes. We propose a unified method to extract both homogeneous and heterogeneous expression patterns across species. The basic idea is to model the sum and difference of expressions across species using multi-scale Gaussians which reveal information about homogeneous and heterogeneous expression patterns respectively. We show using both simulated and real data that the proposed method is suited to identifying both homogeneous and heterogeneous gene expression patterns.