Fig. 1

Overview of the data used for constructing PPB. Distribution of a target type as defined in ChEMBL and b source of targets. c Distribution of targets as per number of associated bioactive compounds. d Histogram of city block distances (log scale) calculated for 50 million random pairs of compounds from ChEMBL 21 using six molecular fingerprints. e APfp, MQN, SMIfp, Sfp and ECfp4 fingerprints were scaled with respect to Xfp to adjust to the value of the most frequent distance. Scaling factors are shown in parentheses. f, g Enrichment of 40 set of DUD actives from corresponding decoys set by six different fingerprints (APfp, Xfp, MQN, SMIfp, Sfp and ECfp4) and four similarity fusion methods (Ffp1–4). City block distance was used as sorting function. Data is represented as average of f Area under ROC curve and g Enrichment factor at 1% of screen database for 40 targets from DUD. h, i Example of p value calculation. h Observed (red) and fitted (black) random distance distributions for the muscarinic acetylcholine receptor M1 (CHRM1, CHEMBL216) in MQN fingerprint space. City block distances were calculated for 1788 ligands of CHRM1 with respect to random compounds from ZINC database. Negative binomial distribution was used for curve fitting. i Cumulative density plot indicating area under fitted curve in h