Skip to main content
Fig. 3 | BMC Research Notes

Fig. 3

From: Exploration of interaction scoring criteria in the CANDO platform

Fig. 3

Accuracies and coverages of five CANDO pipelines at various cutoffs using different compound-protein interaction scoring metrics. Random (navy) is a random control that is calculated using the average of 100 randomly generated interaction matrices. v1 (purple) is the pipeline from the first version of the CANDO platform which used the BScore to quantify compound–protein interactions, with the chemical similarity comparison done using OpenEye ROCS [7, 16]. The Best OB pipeline (magenta) uses the highest OBscore for each compound-protein interaction. Best BS (red) uses the OBscore corresponding to the fingerprint comparison between each compound and highest BSscore binding site ligand for each protein. Best OB+BS (orange) is the highest summation of OBscore and BSscore for each compound-protein interaction, and Best OBxBS (yellow) is the highest product. The average indication and pairwise accuracies for the top10 (triangle), top25 (cross), top50 (square), and top100 (X) cutoffs are shown for each pipeline. The size of the cutoff symbol corresponds to the indication coverage. By varying the interaction scoring scheme in v1.5, we are able to discern that the Best OB protocol results in the highest benchmarking performance, particularly at the top10 and top 25 cutoffs

Back to article page