Structural Modeling of Bitter Taste Receptors Employing A Multi-Template Homology Approach for Ligand Development: The TAS2R14 Case Study Validated Against CryoEM Structures

The Taste 2 Receptor family (TAS2Rs), a subfamily of G protein-coupled receptors (GPCRs), mediates bitter taste perception and is proposed to exist to detect potentially harmful compounds. TAS2R14, although the best structurally characterized member of the class, remains poorly defined due to the flexible nature of extracellular loops, particularly extracellular loop 2 (ECL2). Here, we present a useful TAS2R14 homology model (R14HM) developed using a multi-template approach including MODELLER for core building and Rosetta Kinematic Closure for loop refinement. Validation through membrane-embedded molecular dynamics (MD) simulations, MolProbity analysis, and stereochemical assessments confirmed its structural accuracy and stability. Recently released CryoEM structures of TAS2R14 enabled comparisons, corroborating the predictive accuracy of the model and the methods used to generate it. Despite the availability of Cryo-EM structures, these alone are unable to resolve ECL2 or to provide a dynamic model able to sample the membrane-specific behaviour of these GPCRs—computational models remain necessary to ask these questions. To investigate ligand interactions, a curated library of bitter compounds, including phytochemicals, endocrine ligands, and humulone analogs, was docked into the R14HM, followed by MD refinement and MM-GBSA binding energy calculations. Finally, both the R14HM and the available Cryo-EM structures are then compared to the models generated using popular physics-free approaches such as AlphaFold 3, and the limitations and inaccuracies in the AlphaFold models are discussed. This work establishes a validated framework for TAS2R14 structure-based ligand discovery and highlights key considerations for modeling low-homology Class A GPCRs, bridging computational predictions and experimental findings and re-emphasizes the need for caution when using structures provided by pattern recognition software.