Are You Sure You Have a Good Model?
Nicolas Triballeau, Hugues‐Olivier Bertrand, Francine Acher · Methods and principles in medicinal chemistry · 2006
This chapter contains sections titled: Introduction Validation Methods: Different Answers Brought to Different Questions Software-related Validation Methods Ligand-based Pharmacophore Research Protein Structure-based Pharmacophore Research Critical Remarks Regarding Structure-based Pharmacophore Models Visual Inspection Consistency with Structure - Activity Relationships Some Limitations of Computer Programs Retained Chemical Features Spatial Arrangement 3D-QSAR Pharmacophore Models External Data to Back Up a Pharmacophore Model Biophysical Data Other Published Pharmacophore Models The “Test Set” Approach and the Kubinyi Paradox Database Mining Some Metrics to Assess Screening Performances The ROC Curve Approach A Successful Application: the Ultimate Validation Proof Validation of Pharmacophore Models for Virtual Screening Which Validation Method Should One Insist On? Validation of Pharmacophore Models to Guide Medicinal and Computational Chemistry Validation of Pharmacophore Models for Activity Prediction Which Validation Method Should One Insist On? Case Study: a New Pharmacophore Model for mGlu4R Agonists Metabotropic Glutamate Receptors as Potential Therapeutic Targets Pharmacology of Metabotropic Glutamate Receptor Subtype 4 (mGlu4) Training Set Elaboration Strategy for Perceiving the Pharmacophore Four Criteria to Validate our Pharmacophore Model Results of Our Pharmacophore Model Research with Catalyst-HypoGen and HypoRefine Description of the Two Retained Pharmacophore Models Hypothesis 1 (Catalyst-HypoRefine with Variable Weights) Hypothesis 2 (Catalyst-HypoRefine with Variable Weights and Tolerances) Comparison of the Two Retained Hypotheses Further Validation: Virtual Screening of the CAP Database Conclusion Acknowledgments References