Abstract
DNA microarrays allow simultaneous measurements of expression levels for a large number of genes across a number of different experimental conditions (samples). The algorithms for mining association rules are used to reveal biologically relevant associations between different genes under different experimental samples. This paper presents a new column-enumeration based method algorithm (abbreviated by MCR-Miner) for mining maximal high confidence association rules for up/down-expressed genes. MCR-Miner algorithm uses an efficient maximal association rules tree data structure (abbreviated by MAR-Tree). MAR-tree enumerates (lists) all genes with their binary representations, the binary representation of a gene saves the status (normal, up, and down-expressed) of a gene in all experiments. The binary representation has many advantages, scan the dataset only once, the measurements of confidences for association rules are made in one step, and it makes MCR-Miner algorithm easily finds all maximal high confidence association rules. In the experimental results on a real microarray datasets, MCR-Miner algorithm attained very promising results and outperformed other counterparts.