Pharmacophore ligand-based topological

Ligand-based topological pharmacophores are a class of descriptors that attempt to simulate three-dimensional (3D) pharmacophoric representations by atomic abstraction and the application of through-graph distances as an adjunct for geometric distance through space. 

The CATS (chemically advanced template search) from Schneider et al. [23] was originally proposed as a ligand-based topological pharmacophore for scaffold hopping, a specific subset ofbioisosteric replacement, where the objective is to discover molecular scaffolds that are topologically different but retain the key functional requirements necessary for macromolecular recognition [24]. 

The work from Wagener and Lommerse [26] detailed a new ligand-based topological pharmacophore descriptor specifically for the identification of bioisosteres and can be seen as an approach to alleviate the issues of sensitivity to heteroatom replacement observed by Schuffenhauer et al. The descriptors applied in this work used an atom pair representation similar to that reported by Carhart et al. [28]. These descriptors are extracted from databases of known molecules by shredding the molecules at all deavable bonds with the attachment point being retained as a distinct atom type, X. 

Figure 13.12 A SOM-based pharmacophore road map. Different sets of ligands were projected onto a SOM that was generated by using the complete COBRA library. Black areas indicate the characteristic distributions of the compounds. Crosses indicate empty neurons in the map, i.e., areas of pharmacophore space that are not populated by the respective compound class. All molecules were encoded by a topological pharmacophore descriptor (CATS) [4], Note that each map forms a torus.

Figure 12.10 Ligand-based virtual screening examples for ion channels, (a) Calcium antagonist clopimozid from topological pharmacophore searching, IC5o< 1 pM. (b) Pancreatic Kftjp channel openers obtained from three different methods EC50 >30 pM (left), 21 pM (middle), 15 pM (right).

Figure 12.11 Ligand-based virtual screening examples for kinases, (a) GSK-3 inhibitors from topological pharmacophore searching with IC50 1.2 iM (left) and from subsequent lead...

D-ligand-based methods that create pharmacophore models capture the SAR by identifying common pharmacophoric features within a set of active molecules. These models are composed of the input molecules in a joint 3D-alignment that is based on those common features and not on 2D-topology. Hence, this approach enables the possibility to find compounds that share the same features, but are based on a different bioisosteric scaffold. This approach is being widely used in both academy and industry and has been extensively reviewed [144, 145]. 

Byvatov and Schneider compared the SVM-based and the Kolmogorov-Smirnov feature selection methods to characterize ligand-receptor interactions in focused compound libraries.Three datasets were used to compare the feature selection algorithms 226 kinase inhibitors and 4479 noninhibitors 227 factor Xa inhibitors and 4478 noninhibitors and 227 factor Xa inhibitors and 195 thrombin inhibitors. SVM classifiers with a degree 5 polynomial kernel were used for all computations, and the molecular structure was encoded into 182 MOE descriptors and 225 topological pharmacophores. In one test, both feature selection algorithms produced comparable results, whereas in all other cases, SVM-based feature selection had better predictions. 

---