Protein-ligand-based pharmacophores: generation and utility assessment in computational ligand profiling

Ligand profiling is an emerging computational method for predicting the most likely targets of a bioactive compound and therefore anticipating adverse reactions, side effects and drug repurposing. A few encouraging successes have already been reported using ligand 2-D similarity searches and protein-ligand docking. The current study describes the use of receptor-ligand-derived pharmacophore searches as a tool to link ligands to putative targets. A database of 68,056 pharmacophores was first derived from 8,166 high-resolution protein-ligand complexes. In order to limit the number of queries, a maximum of 10 pharmacophores was generated for each complex according to their predicted selectivity. Pharmacophore search was compared to ligand-centric (2-D and 3-D similarity searches) and docking methods in profiling a set of 157 diverse ligands against a panel of 2,556 unique targets of known X-ray structure. As expected, ligand-based methods outperformed, in most of the cases, structure-based approaches in ranking the true targets among the top 1% scoring entries. However, we could identify ligands for which only a single method was successful. Receptor-ligand-based pharmacophore search is notably a fast and reliable alternative to docking when few ligand information is available for some targets. Overall, the present study suggests that a workflow using the best profiling method according to the protein-ligand context is the best strategy to follow. We notably present concrete guidelines for selecting the optimal computational method according to simple ligand and binding site properties.