Table 2.
Performance of different COMBINE modelsafor the whole set of inhibitors in fitting and prediction
|
|
Data set 1b |
Data set 2b |
||||||
---|---|---|---|---|---|---|---|---|---|
Electrostatic Modelsc | No. of PCs | r2 | SDEC | q2 | SDEP | r2 | SDEC | q2 | SDEP |
1 |
1 |
0.709 |
0.78 |
0.653 |
0.948 |
0.588 |
0.929 |
0.51 |
1.153 |
|
2 |
0.82 |
0.614 |
0.713 |
0.914 |
0.793 |
0.659 |
0.701 |
0.816 |
|
3 |
0.871 |
0.521 |
0.74 |
1.131 |
0.891 |
0.477 |
0.703 |
1.074 |
|
4 |
0.917 |
0.417 |
0.713 |
0.989 |
0.91 |
0.435 |
0.761 |
1.096 |
|
5 |
0.944 |
0.344 |
0.691 |
1.00 |
0.923 |
0.402 |
0.775 |
0.98 |
2 |
1 |
0.252 |
1.336 |
0.149 |
1.609 |
0.295 |
1.215 |
0.106 |
1.571 |
|
2 |
0.718 |
0.994 |
0.528 |
1.116 |
0.78 |
0.679 |
0.704 |
0.913 |
|
3 |
0.856 |
0.848 |
0.657 |
0.931 |
0.824 |
0.607 |
0.719 |
0.785 |
|
4 |
0.898 |
0.805 |
0.691 |
1.097 |
0.896 |
0.466 |
0.747 |
1.06 |
|
5 |
0.921 |
0.824 |
0.677 |
1.131 |
0.92 |
0.41 |
0.786 |
0.988 |
3 |
1 |
0.212 |
1.37 |
0.105 |
1.623 |
0.26 |
1.245 |
0.045 |
1.621 |
|
2 |
0.773 |
0.845 |
0.66 |
0.85 |
0.79 |
0.663 |
0.725 |
0.949 |
|
3 |
0.864 |
0.819 |
0.68 |
1.088 |
0.819 |
0.617 |
0.731 |
0.796 |
|
4 |
0.922 |
0.889 |
0.623 |
1.007 |
0.897 |
0.465 |
0.728 |
0.728 |
5 | 0.956 | 1.002 | 0.521 | 0.87 | 0.916 | 0.419 | 0.772 | 0.816 |
aAbbreviations: r2, correlation coefficient; SDEC, standard deviation of errors of correlation; q2, predictive correlation coefficient; SDEP, standard deviation of errors of prediction.
bData set 1, all the BACE-1/inhibitor complexes were built from the A chain of 1 W51 and each of the co-crystallized ligands. All the co-crystallized ligands were translocated directly to the binding pocket of 1 W51 by a superimposition method using the Cα atoms (1 W51 structure as the reference), after this process, each BACE-1/inhibitor complex was energy minimized using the AMBER 9.0 program.
Data set 2, all the BACE-1/inhibitor complexes were the actual complexes present in the PDB and energy minimized with the AMBER 9.0 program.
cThe three types of electrostatic models. Model 1: a distance-dependent dielectric constant model. Model 2: a uniform dielectric constant model. Model 3: a sigmoidal model. The values in bold highlight the best quality models.