Host gene expression in cooperation with the microbiome has been discovered to play a large role in disease progression and response. In particular, changes in host gene expression may have a marked impact on bacterial species diversity and abundance which in turn can trigger response from the host. Utilizing the advances in information science we propose a new procedure to select mediation mechanisms in integrated –omics data for predicting clinical outcomes. A comprehensive simulation study shows that the proposed method outperforms existing nonparametric mediation methods. In addition, the new method is applied in a real data analysis and we get some interesting findings.