Efficient Mining of Pan-Correlation Patterns from Time Course Data

There are different types of correlation patterns between the variables of a time course data set, such as positive correlations, negative correlations, time-lagged correlations, and those correlations containing small interrupted gaps. Usually, these correlations are maintained only on a subset of time points rather than on the whole span of the time points which are traditionally required for correlation definition. As these types of patterns underline different trends of data movement, mining all of them is an important step to gain a broad insight into the dependencies of the variables. In this work, we prove that these diverse types of correlation patterns can be all represented by a generalized form of positive correlation patterns. We also prove a correspondence between positive correlation patterns and sequential patterns. We then present an efficient single-scan algorithm for mining all of these types of correlations. This “pan-correlation” mining algorithm is evaluated on synthetic time course data sets, as well as on yeast cell cycle gene expression data sets. The results indicate that: (i) our mining algorithm has linear time increment in terms of increasing number of variables; (ii) negative correlation patterns are abundant in real-world data sets; and (iii) correlation patterns with time lags and gaps are also abundant. Existing methods have only discovered incomplete forms of many of these patterns, and have missed some important patterns completely.