Discovering and measuring association and interaction among variables is a fundamental scientific task in exploratory data mining, especially for big data.Recently, Reshef et al presented a novel association statistic called maximal information coefficient (MIC) to measuring dependency between paired variables.It is able to capture a wide range of associations regardless of linear or non-linear relationships.Selecting a parsimonious set of informative genes to build high generalization performance classifier is the most important goal for the analysis of tumor microarray expression data.However, individual-gene-ranking methods such as ignore redundancy and synergy among genes, and often result in the loss of heritability or prediction accuracy.