Skip to main content
Figure 2 | Journal of Cheminformatics

Figure 2

From: A physicochemical descriptor-based scoring scheme for effective and rapid filtering of kinase-like chemical space

Figure 2

ROC plots of five databases: KBD, WDI, NCI, NPD, and MLSMR. The Y-axis indicates the fraction of true positives (FTP) and the X-axis indicates the fraction of true negatives (FTN). The table in the lower right shows the calculated maximum (upper light green line), mean (dark green line) and minimum (lower light green line) AUC values of the ROC plots based on the 10 different calculations for each database. AUC values closer to random (0.50) signify more kinase-likeness, meaning that kinase-likeness score is unable to distinguish the 1% of kinase active compounds that were mixed in from the other compounds in a given database. This indicates that those compounds have physicochemical properties that are similar to kinase active compounds. A diagonal reference line (area under the curve = 0.50) defines points where kinases are not discernible from the test database compounds. As validation, randomly selected kinase compounds were mixed in with the KBD (upper left plot) and an average AUC of 0.5 was observed.

Back to article page