Figure 4

The average and median F-scores of the functional region predictions generated by 18 different criteria for sequence selection. The horizontal axis represents the sequence selection methods, DSPAC-based method, SDPfox, and Naïve approaches, with 16 different percent sequence identity thresholds. The red open rectangles with error bars indicate the averaged F-scores of the predictions when LMIC was used to detect the cluster of conserved residues. The negative part of the error bar, which would extend into the negative F-score region, is not drawn, since the F-value has a positive value by definition. The light green open triangles indicate the medians of the predictions when LMIC was used to detect the cluster of conserved residues. The blue crosses with error bars indicate the averaged F-scores of the predictions when PatchFinder was used to detect the cluster of conserved residues. The purple asterisks indicate the medians of the predictions when PatchFinder was used to detect the cluster of conserved residues. The combination of DSPAC-based sequence selection and LMIC showed the best performance among all combinations of the methods. The prediction performances by the PatchFinder and Naïve approaches at 30% sequence identity are comparable to those by the LMIC and DSPAC-based approaches.