ROC: Is Twitter also useful in search?
Figure 1 ROC curve: i. i.d source signal.
The SEN, SPC, and ACC values were determined from the optimal cut point of the ROC curve, i.e., the point closest to (0,1).
The optimal prediction performance corresponds to the point closest to the top left corner of the ROC plot (i.e. 100% true viable rate, 0% false viable rate).
The resulting metagene expression values are then used in ROC analyses, i.e. by ranking the samples according to their metagene expression values.
Hence, cut-off points for the ESR1 methylated promoter was established from the receiver operating characteristics (ROC) curves i.e. selecting values that gave the maximal likelihood ratio (in current case the cut-of value was 0.02 relative units) [ 20].
For initial multiple sequence alignments and for level of divergence ×2 and ×3 in Figure 4, the up-right head points of these two ROC curves (i.e. each corresponding to threshold τ = 0.02) have tpr < 1; this means that there exists some true branches with BioNJ bootstrap-based confidence value < 0.02.
Indeed, in Figure 4, the down-left tail point of each ROC curve (i.e. corresponding to threshold τ = 0.98) always has highest tpr (y-axis) for trimmed alignments than for initial ones; this shows that a larger proportion of true branches with BioNJ bootstrap-based confidence values >0.98 is observed with trimmed alignments than with initial alignments.
We computed the area under the ROC curve (i.e., the AUC score), where 1 is returned for a perfect inference and 0.5 is returned for a random inference, and the area under the PR curve (i.e., AUPR score), where 1 is returned for a perfect inference and the ratio of positive examples against all samples in the gold standard data is returned for a random inference.
We perform power calculations for various settings and present them through ROC-curves, i.e. as plots with significance levels versus power with respect to a set of underlying score thresholds (Selin, 1965; Bradley, 1996).
The mean confidence rating correlates very well with the confidence bias reflected by the ROC curve: B ROC = ln ∑ i = 1 3 Y i + 1 - X i 2 - Y i - X i + 1 2 ∑ i = 4 6 Y i + 1 - X i 2 - Y i - X i + 1 2 (r = 0.94, p < 10−18).
