RESUMO
Integral membrane proteins in the G Protein-Coupled Receptor (GPCR) class are attractive drug development targets. However, computational methods applicable to ligand discovery for many GPCR targets are restricted by limited numbers of known ligands. Pharmacophore models can be developed using variously sized training sets and applied in database mining to prioritize candidate ligands for subsequent validation. This in silico study assessed the impact of key pharmacophore modeling decisions that arise when known ligand numbers for a target of interest are low. GPCR included in this study are the adrenergic alpha-1A, 1D and 2A, adrenergic beta 2 and 3, kappa, delta and mu opioid, serotonin 1A and 2A, and the muscarinic 1 and 2 receptors, all of which have rich ligand data sets suitable to assess the performance of protocols intended for application to GPCR with limited ligand data availability. Impact of ligand function, potency and structural diversity in training set selection was assessed to define when pharmacophore modeling targeting GPCR with limited known ligands becomes viable. Pharmacophore elements and pharmacophore model selection criteria were also assessed. Pharmacophore model assessment was based on percent pharmacophore model generation failure, as well as Güner-Henry enrichment and goodness-of-hit scores. Three of seven pharmacophore element schemes evaluated in MOE 2018.0101, Unified, PCHD, and CHD, showed substantially lower failure rates and higher enrichment scores than the others. Enrichment and GH scores were used to compare construction protocol for pharmacophore models of varying purposes- such as function specific versus nonspecific ligand identification. Notably, pharmacophore models constructed from ligands of mixed functions (agonists and antagonists) were capable of enriching hitlists with active compounds, and therefore can be used when available sets of known ligands are limited in number.