Mining Time-delayed Gene Regulation Patterns from Gene Expression Data

Abstract

Discovered gene regulation networks are very helpful to predict unknown gene functions. The activating and deactivating relations between genes and genes are mined from microarray gene expression data. There are evidences showing that multiple time units delay exist in a gene regulation process. Association rule mining technique is very suitable for finding regulation relations among genes. However, current association rule mining techniques cannot handle temporally ordered transactions. We propose a modified association rule mining technique for efficiently discovering time-delayed regulation relationships among genes.
By analyzing gene expression data, we can discover gene relations. Thus, we use modified association rule to mine gene regulation patterns. Our proposed method, BC3, is designed to mine time-delayed gene regulation patterns with length 3 from time series gene expression data. However, the front two items are regulators, and the last item is their affecting target. First we use Apriori to find frequent 2-itemset in order to figure backward to BL1. The Apriori mined the frequent 2-itemset in the same time point, so we make the L2 split to length one for having relation in the same time point. Then we combine BL1 with L1 to a new ordered-set BC2 with time-delayed relations. After pruning BC2 with the threshold, BL2 is derived. The results are worked out by BL2 joining itself to BC3, and sifting BL3 from BC3. We use yeast gene expression data to evaluate our method and analyze the results to show our work is efficient.