Statistical methods have been developed in the regression setting to study the effects of clinical covariates and expression levels of genes. However, there are some limitations in statistical methods for variable set selection among genes within a pathway, and building statistical networks among genes because of not only complications in modeling unknown and strong correlations among high-dimensional genes but also overlapping genes among pathways. In this paper, we develop a semiparametric-sparse network kernel method to solve these limitations. Our approach is a unified and integrated method which can simultaneously identify important variables and build network among them. We develop our approach under a semiparametric kernel machine framework which can allow for the possibility that each gene expression effect might be nonlinear and the genes within the same pathway are likely to interact with each other in a complicated way. We demonstrate our approach using simulation study and real application on genetic pathway analysis.