FIGURE 2 

The distribution of correlation coefficients between actual and MED-predicted gene expression derived from crossvalidation (see Materials and methods section). The blue curve, whose average is 0.52 presented as the blue diamond, is the distribution for all 5719 genes in the S. cerevisiae genome, whereas the red curve, whose average is 0.72 presented as the red circle, is the distribution for about 2600 genes earlier work (Beer and Tavazoie, 2004) used for comparison purpose. the black square represents the average of 10 average correlation coefficients derived from 10 crossvalidation runs with the input expression data whose rows were permuted. The superiority of MED lies in its ability not only to produce good prediction, but also to reduce bad prediction (i.e. genes with little or even negative correlation). It is worth noting that these 2600 genes used by early work (Beer and Tavazoie, 2004) stand out automatically as an outcome of MED without the need for heuristically selecting them out in the first place.