Figure 3

Evaluation of distance measures. Four of the most commonly used gene co-expression distance measures were evaluated in terms of their ability to detect biologically relevant gene associations according to biological process Gene Ontology annotations (A, B, C) and the KEGG PATHWAY map annotations (D, E, F). We incorporated gene pairs, starting with the highest co-expressed ones, continuing successively with weaker interactions, and predicted their accuracies. The impact of this incorporation on the prediction accuracies for the three species S. cerevisiae (yeast) (A, D), D. melanogaster (fly) (B, E), and C. elegans (worm) (C, F) and for four commonly used distance measures are shown. The background accuracy 'b' (grey horizontal lines) is the average over all gene association and represents the expected accuracy for a randomly chosen gene pairs. For the S. cerevisiae, D. melanogaster, and C. elegans datasets, the analysis is based on 1771, 2065, and 2065 genes, respectively (see also Figure 2). The top 8000 interactions obtained using the Spearman correlation, which was used in the conservation of co-expression study shown in Figure 4, contain 1131, 1308, and 1117 genes for S. cerevisiae, D. melanogaster, and C. elegans, respectively.