Parallel Emerging Patterns in MicroArray

Emerging Patterns (EPs) can discover biologic gene rules from microarray. It can classify test samples, and help people developing drugs to adjust the gene expression to normal value. However, current EP can discover only one cut point to classify the two types for each gene expression array. But in fact, microarray datasets always have more than two types to predict, namely, multi-classification. For EP, combining many binary classifiers into a multi-classifier is limited, which can not be performed in parallel and waste more to train. In this paper, Parallel Emerging Patterns (PEPs), which have multiple cut points for each gene, are proposed: First, an extended parallel signal noise ratio (pSNR) index is proposed to select feature genes, Second, the theorems to discover PEPs are proved, Third, a simple parallel classifier of PEP is evaluated by using collective likelihood. Simulation results validate the feasibility of PEP, and show that PEPs can both extract useful gene rules and well classify test samples in parallel.