Fig. 2

ROC (Receiver operator curve) of weighted-scoring, quality-score, PSIPred, SEG and GlobPlot sequence predictor against 6599 Pfam-to-PDB mappings. For each sequence predictor, 19 data points form each plot (Additional file 2: Tables S1 and S2). However, in the case of PSIPred, SEG and GlobPlot, several of their data points coincide to the same positions for the lower thresholds of between 0.05 and around 0.50. As such, the sensitivity and specificity for PSIPred, SEG and GlobPlot cannot be improved beyond certain thresholds. Quality-score by itself is a better predictor than SEG and GlobPlot but worse than PSIPred for a limited range. Meanwhile, the weighted-scoring scheme performs the best for false-positive rate (i.e. 1-specificity) of less than 0.235 (see vertical dotted line, Fig. 3). Beyond that, the quality-score takes over as the better predictor. However, since larger false-positive rate values are typically avoided, the slight inferior performance of the weighted-scoring scheme beyond the false-positive rate of 0.31. Generally speaking, the better performance of the weighted-scoring scheme than any single predictor is due to the statistical weighing step where the contributions of the better predictors are made more significant