Efficient Mining Maximal Subspace Differential Co-expression Patterns in Matrix Datasets: a General Earthquake Analysis Approach

Abstract

The electromagnetic anomaly observations before earthquake, have been confirmed by many cases of strong earthquakes. The analysis of earthquake magnetic anomaly is an effective approach for seismo-precursor detection. Traditional frequent mining methods for electromagnetic matrix datasets analysis often find the co-related items. However, these methods may miss the items which are differential co-related patters under different datasets. Mining these differential co-related patterns is more valuable for inferring potential knowledge. In this paper, we develop an algorithm, MSPattern, to mine maximal subspace differential co-expression patterns. MSPattern constructs a weighted undirected item-item relational graph firstly. Then all the maximal co-related patterns would be mined using item-growth method in above graph. MSPattern also utilizes several techniques for producing maximal patterns without candidate patterns maintenance. Evaluated by real electromagnetic matrix datasets and the gene expression datasets, the experimental results show our algorithm can find some potential knowledge for earthquake analysis, and MSPattern algorithm is more efficient than traditional ones. The performance of MSPattern is also evaluated by empirical p-value and gene ontology, the results show our algorithm can find statistical significant and biological differential co- expression patterns.