View article

A pressing question in current genetic biology is which genes interact; improvements in the prediction of interactions between genes can help researchers in generating new hypotheses of how particular genes act together. The availability of affordable microarray technology has spurred the development of bioinformatics techniques aimed at extracting such genetic interactions automatically from expression data. As these techniques have employed a number of different measures to estimate which pairs of genes interact, a question resulting from this development is how these different measures compare. We study the performance of five different measures of genetic interactions on four different microarray datasets, using four different biological databases to evaluate the results: BIND [1], HPRD [8], KEGG [7] and Reactome [6]. Performance is measured by measuring the area under the curve (AUC) of the ROC …