Discovering Temporal Associations among Significant Changes in Gene Expression

One of the most demanding problems in mining temporal data is to identify how multivariate change associations might be discovered and used to better understand data interactions and dependencies. This paper introduces a framework to mine associations among significant changes in multivariate time-series data. Building on statistical methods, we detect significant changes in time-series data and use marginal change rates to qualify the direction of change at significant change points. Furthermore, a propositional confirmation-guided rule discovery method is used to discover associations among these significant changes. We apply our approach to gene expression data measured in yeast cell cycles and demonstrate that our method can learn novel and high-quality significant change associations among different genes. Such associations can be used to cluster genes and build gene interaction networks.