Skip to main content

Advertisement

Table 6 Virtual Screening results for PDGFRβ at five different applicability levels

From: Estimation of the applicability domain of kernel-based machine learning models for virtual screening

Kernel ADE   Threshold AUROC BEDROC (α) Ligands Decoys
      100.00 53.6 32.2 20.0   
OAK KDE 50% 0.62 0.68 0.22 0.19 0.19 0.19 74 2789
   33% 0.63 0.67 0.28 0.22 0.22 0.21 55 1854
   200 0.667 0.78 0.80 0.65 0.64 0.57 7 193
   100 0.673 0.79 0.95 0.83 0.82 0.72 6 94
  wKDE 50% 0.62 0.68 0.22 0.19 0.18 0.19 74 2790
   33% 0.63 0.67 0.28 0.22 0.22 0.21 56 1853
   200 0.665 0.76 0.79 0.63 0.61 0.53 8 192
   100 0.672 0.87 0.95 0.83 0.82 0.73 6 94
  ONE 50% -11.58 0.67 0.22 0.19 0.19 0.19 76 2781
   33% -10.95 0.66 0.26 0.20 0.20 0.19 58 1851
   200 -8.93 0.78 0.80 0.65 0.64 0.57 7 192
   100 -8.64 0.78 0.95 0.83 0.82 0.72 6 94
FlexOAK KDE 50% 0.61 0.61 0.18 0.15 0.15 0.15 67 2796
   33% 0.62 0.65 0.25 0.21 0.20 0.20 52 1857
   200 0.647 0.84 0.79 0.61 0.60 0.53 9 191
   100 0.653 0.87 0.95 0.82 0.80 0.70 7 93
  wKDE 50% 0.61 0.61 0.18 0.15 0.15 0.15 61 2796
   33% 0.62 0.65 0.25 0.20 0.20 0.20 52 1857
   200 0.646 0.83 0.79 0.61 0.60 0.52 9 191
   100 0.653 0.85 0.95 0.81 0.79 0.68 8 92
  ONE 50% -12.67 0.61 0.18 0.15 0.15 0.15 67 2796
   33% -12.04 0.65 0.25 0.20 0.20 0.20 50 1859
   200 -10.33 0.82 0.79 0.61 0.59 0.52 9 191
   100 -9.97 0.86 0.95 0.82 0.80 0.70 7 93
MARG KDE 50% 0.83 0.37 0.13 0.11 0.11 0.11 52 2812
   33% 0.84 0.40 0.18 0.17 0.17 0.17 25 1884
   200 0.87 0.54 0.65 0.49 0.48 0.43 8 192
   100 0.88 0.76 0.88 0.75 0.73 0.66 4 96
  wKDE 50% 0.81 0.35 0.13 0.12 0.11 0.11 49 2814
   33% 0.82 0.38 0.17 0.17 0.17 0.17 29 1880
   200 0.86 0.57 0.65 0.49 0.48 0.42 8 192
   100 0.87 0.86 0.91 0.82 0.81 0.77 3 97
  ONE 50% -5.05 0.39 0.13 0.11 0.11 0.11 51 2812
   33% -4.58 0.42 0.19 0.18 0.18 0.18 26 1883
   200 -3.09 0.54 0.65 0.49 0.48 0.43 8 192
   100 -2.79 0.85 0.91 0.82 0.81 0.77 3 97
  1. The threshold is adjusted such that either a certain fraction (50%, 33%) of the compounds or that exactly n compounds (n = 200, 100) remain in the domain. Results, which differ significantly from random rankings (p-Value < 0.01) are shown in bold.