Skip to main content

Advertisement

Table 2 Adjusted Rand Indices, BEDROC, and AUCROC values.

From: Application of 3D Zernike descriptors to shape-based ligand similarity searching

Comparison Method     
Descriptors Metric Order Adjusted Rand Index BEDROC (α = 160.9) BEDROC (α = 32.2) ROC AUC
3DZD Correlation coefficient 4 0.299 0.506 0.453 0.694
   6 0.256 0.590 0.494 0.708
   8 0.422 0.739 0.654 0.738
   10 0.465 0.614 0.536 0.724
   12 0.487 0.697 0.630 0.732
   14 0.442 0.697 0.618 0.733
  Euclidean 4 0.278 0.338 0.329 0.680
   6 0.278 0.526 0.446 0.704
   8 0.357 0.678 0.621 0.738
   10 0.395 0.594 0.553 0.722
   12 0.487 0.658 0.610 0.730
   14 0.372 0.717 0.622 0.743
  Manhattan 4 0.270 0.318 0.329 0.686
   6 0.260 0.484 0.427 0.703
   8 0.328 0.698 0.619 0.732
   10 0.408 0.591 0.619 0.736
   12 0.393 0.637 0.591 0.732
   14 0.213 0.656 0.598 0.748
USR Correlation coefficient 12 0.213 0.697 0.617 0.695
   16 (Kurtosis) 0.227 0.721 0.651 0.707
  Euclidean 12 0.270 0.760 0.639 0.708
   16 0.270 0.760 0.642 0.709
  Manhattan 12 0.343 0.762 0.661 0.718
   16 0.328 0.782 0.675 0.720
SIMCOMP (Maximal Common Subgraph) - 0.400 0.847 0.779 0.808
EVA - σ = 100 cm-1 0.442 - - -
   σ = 50 cm-1 0.388 - - -
   σ = 20 cm-1 0.381 - - -
UNITY2D - - 0.247 - - -
MACCS (Tanimoto) 166 bit key 0.364 0.778 0.659 0.742
MOLPRINT2D (Tanimoto)   0.516 0.848 0.755 0.806
Atom Count -   0.400 0.467 0.460 0.850
  1. For the 3DZD, orders of expansion 4-14 have been applied and similarities calculated using the correlation coefficient, the Euclidean and the scaled Manhattan distances. For the USR the fourth statistical moment kurtosis has also been calculated. Hierarchical clustering was done using Ward's method. Results for the EVA and UNITY2D descriptors (for 7 clusters) have been taken from Table 2 in the article by Takane and Mitchell.