Figure 1

Gene prediction based on sequence conservation. (A) and (B) show receiver-operator characteristic curves summarizing the sensitivity and specificity of gene prediction based on COG membership when compared to the current gene annotations. In (A), an ORF is classified as a gene if it is conserved in a COG at a certain stringency; for (B), the ORF must be in a COG that contains at least one annotated gene from another species. Curves are produced by examining COGs at different stringencies. At stringency 2, tests are very sensitive but not very specific (points at upper right of each panel). As stringency increases, specificity increases and sensitivity decreases (indicated by arrow). For clarity, full ROC curves are shown for only seven of the organisms studied, and for the pooled result among all of the organisms studied. The plotting symbols and colors used in (A) and (B) are next to the organism names in (C). (C) shows the areas under the curves in (A) black bars and (B) grey bars. The ROC curve of a perfect test would enclose an area of 1, for a completely arbitrary test the area would be 0.5. The organisms in (C) are ordered by the area under the ROC curve in (B).