Learning the gene coexpression pattern is a central challenge for high-dimensional gene expression analysis. Recently, sparse singular value decomposition (SVD) has been used to achieve this goal.