Skip to main content
Fig. 5 | Journal of Cheminformatics

Fig. 5

From: Comparative evaluation of methods for the prediction of protein–ligand binding sites

Fig. 5

Binding pocket characterisation. Ligand binding site predictions by eleven methods with default parameters (d) are compared to the LIGYSIS reference dataset. A black dashed line separates LIGYSIS from the eleven ligand binding site predictors. Predictions by DeepPocketRESC and fpocketPRANK are not included in this analysis as they are re-scored and re-ranked fpocket predictions and their plotting would be redundant. Data points farther than four standard deviations (SD) from the mean are considered outliers. The limit of the Y axis is the maximum non-outlier value plus a buffer value. This way, only the most extreme outliers are hidden, which maximises visual interpretation of the data whilst minimising the number of data points not shown. Within the violin plots are box plots representing the underlying distribution. Line represents the median, box contains the interquartile range (IQR), and whiskers extend to 1.5 × IQR. A Number of pockets per protein; B pocket radius of gyration Rg (Å); C minimum inter-pocket centroid distance (MCD) (Å). This is a measure of how close predicted pockets are to each other within a protein; D Maximum inter-pocket residue overlap (MRO). Residue overlap was calculated as Jaccard Index. This is a measure of how much pockets overlap in terms of binding residues

Back to article page