Finding marker genes from high dimensional expression profiles: Divide-and-conquer exploiting a fuzzy rule based framework

Previously we have developed a feature selection mechanism (Fuzzy Systems - Feature Attenuating Gates, FS-FAG), which cannot deal with very high dimensional data in an efficient manner. To address this issue, in this study we introduce a divide-and-conquer strategy for selection of marker genes for high dimensional cancer microarray data. This is a hierarchical system, which can be used with very high dimensional data. To demonstrate the effectiveness of the proposed scheme, we use two sets of microarray data under different experimental conditions. We examine the variations in the selected number of genes, the number of the final sets of marker genes, and the discriminating power of cancer subtypes of the selected marker genes. Experimental results demonstrate that proposed method indeed selects marker genes with good discriminating power for different cancer subtypes.