Abstract
The performance of the AutoDock, GOLD and FlexX docking programs was evaluated for docking of dicarboxylic acid inhibitors into metallo-beta-lactamases (MBLs). GOLD provided the best overall performance, with RMSDs between experimental and docked structures of 1.8-2.6 A and a good correlation between the experimentally determined MBL-inhibitor affinities and the GOLD scores. GOLD was selected for a test including a broad spectrum of inhibitors for which experimental MBL-inhibitor binding affinities are available. This study revealed that (1) for most compound classes (dicarboxylic acids, tetrazoles, sulfonylhydrazones, and peptide-like compounds) there is a good correlation between the experimentally determined MBL-inhibitor affinities and the GOLD scores, (2) the correlation only holds within a given class, that is, scores of compounds from different classes cannot be directly compared, (3) for some compound classes (e.g. small sulphur compounds) there is no direct correlation between the experimentally determined MBL-inhibitor affinities and the GOLD scores. Using partial least squares methods, a model with R2 = 0.82 and Q2 = 0.78 for the training set was obtained based on the GOLD score and descriptors associated with binding of the IMP-1 inhibitors to the enzyme. The external Q2 for the test set is 0.73. This final model for prediction of IMP-1 MBL-inhibitor affinity handled all known classes of MBL-inhibitors, except small sulphur compounds.
| Original language | English |
|---|---|
| Journal | Journal of Computer - Aided Molecular Design |
| Volume | 18 |
| Issue number | 4 |
| Pages (from-to) | 287-302 |
| Number of pages | 16 |
| ISSN | 0920-654X |
| Publication status | Published - 2004 |
Keywords
- Computational Biology
- Enzyme Inhibitors
- Ligands
- Models, Molecular
- Protein Binding
- Software
- Succinic Acids
- beta-Lactamases