An insight into biological datamining based on rarity and correlation as constraints

Souad Bouasker, Sadok Ben Yahia, Gayo Diallo

Abstract: Association-rules mining techniques have been widely applied to identify differently expressed gene expressions among micro-array data. Rare correlated patterns are identified as efficient in generating accurate association rules. To our knowledge, no any algorithm which is able to perform this challenging task is currently available. Therefore, we designed CPMiner, a new generic method for mining interesting correlations from data. It performs the extraction of the sets of both frequent correlated and rare correlated patterns as well as their associated condensed representations according to two distinct measures: all-confidence and bond. CPMiner has been applied on processing biological data. To this end we developed CoRaM, the first unified framework dedicated to the extraction of a generic basis of Correlated-Rare association rules from gene expression data. It relies on CPMiner extracting the rare correlated patterns and both a specific discretization method and the derivation of the generic basis of the rare correlated association rules. Our proposed approach has been successfully applied on a breast-cancer Gene Expression Matrix (GSE1379) with very promising results.