Figure 1

Nucleotide level sensitivity (nSn) and specificity (nSp). We restrict to coding regions, and display performance as a function of protein level identities in the aligned regions. Every data point represents 658 of the 10395 mRNAs from the mouse-human alignments. Obviously, the results for CAT, est_genome and GeneWise are hard to distinguish from each other when it comes to sensitivity. In plotting the figure, we discard the worst 5% of pairs where the fraction of aligned regions in respect to the length of full CDS is too small. These 5% of orthologous pairs tend to be wrongly predicted in the HomoloGene database due to their short aligned regions. Discarding them yields more consistent curves.