Fig 1.
The Template CoMFA Workflow.
Table 1.
Properties of the biological data sets.
Fig 2.
Input template alignment alternatives for the hERG target.
The three structures in each image are loratadine (green), perhexilene (yellow), and astemizole (red). Each one-letter label refers to the protocol that generated the alignment. The protocol F image also includes the best-matching 3D structure, among the training set, which the template CoMFA program generated for each of these three templates.
Fig 3.
“Predicted vs. Actual” plots of the CHEMBL_VALUEs (log affinities) for the twelve targets.
Table 2.
BACE (Beta-secretase 1): Template CoMFA model properties.
Fig 4.
Bar chart representation of the q2 values for the models from all combinations of template protocols and targets.
Yellow bars are from Xray templates (protocol A), green from Surflex-SIM alignments (B and C), cyan from the “irrational” Concord (D) or spatially dispersed (F)alignments, and red from omissions of template CoMFA (E and G).
Table 3.
Comparisons of q2 values (from leave-one-out crossvalidation) for template CoMFA models, as produced by applying protocols A-G to the input data, averaged over the twelve biological targets.
Fig 5.
Comparison of input template alignments for the BACE target.
The left panel depicts 13 ligands extracted from crystallographic structures of BACE. The right panel shows three of those ligands as optimally realigned by the Surflex-SIM program.
Fig 6.
Input template alignment alternatives for the carbonic anhydrase II target.
The three structures in each image are the ligands extracted from .pdb files 1BN3 (green), iBNM (yellow), and 1BNN (red). Each one-letter label refers to the protocol that generated the alignment. The protocol F image also includes the best-matching 3D structure, from within the training set, that the template CoMFA program generated for each of these three templates.
Table 4.
Comparison of Correct Prediction Rates, between Template CoMFA and Nearest Neighbor Tanimoto 2D Fingerprint.
Table 5.
Results from OR'ing or AND'ing the predictions of Template CoMFA with Tanimoto NN predictions.
Fig 7.
Structures with high binding affinity to the D2A target.
These identified as D2A “actives” by the template CoMFA method (3) reported as D2A “inactives” by the Tanimoto 2D fingerprint nearest neighbor method.
Fig 8.
Additional representations of the protocol B template CoMFA model for carbonic anhydrase II.
The upper panel shows the conventional “stdev*coeff” representation of a CoMFA model: the green shapes enclose spatial regions where affinity increases with steric bulk; the yellow shapes enclose regions where decreased steric bulk improves affinity; the red shapes enclose regions where more positive electrostatic potential improves affinity; the blue shapes enclose regions where more negative electrostatic potential improves affinity. The lower panel also includes structures of carbonic anhydrase II (skeleton representation) and the template CoMFA alignments of three highly active training set ligands (space-filling representations).