GRAPH NEURAL NETWORK-GUIDED VIRTUAL SCREENING AND CORE-HOPPING-BASED DESIGN OF NOVEL PPAR-γ MODULATORS
TL;DR
An integrated GNN-guided virtual screening and core-hopping workflow tailored specifically to PPAR-γ modulator design is proposed, built around the goal of identifying selective partial agonists and biased ligands that retain insulin-sensitizing efficacy while minimizing helix-12-driven adverse transcriptional programs.
Abstract
Peroxisome proliferator-activated receptor gamma (PPAR-γ) remains one of the most extensively pursued nuclear receptor targets in the treatment of type 2 diabetes mellitus, metabolic syndrome, and, increasingly, inflammatory and oncological disease. Decades of thiazolidinedione-based drug development established proof of therapeutic concept but simultaneously exposed the mechanistic liabilities of full agonism at this receptor, most notably fluid retention, weight gain, and cardiovascular risk arising from complete stabilization of helix 12 and constitutive coactivator recruitment. The emergence of two complementary computational paradigms is now reshaping how medicinal chemists approach this target. Graph neural networks (GNNs), which represent small molecules as attributed molecular graphs rather than fixed-length descriptor vectors, have demonstrated superior performance in virtual screening, binding-affinity prediction, and pharmacokinetic property estimation across diverse drug targets. In parallel, core-hopping and scaffold-hopping methodologies allow medicinal chemists to replace a validated pharmacophoric core with topologically distinct but electronically and geometrically compatible alternatives, offering a route to novel intellectual property, improved selectivity, and mitigated toxicity liabilities without discarding validated structure-activity knowledge. This review synthesizes the structural biology of the PPAR-γ ligand-binding domain, surveys classical structure-based design campaigns undertaken against this receptor and critically examines the architectures and applications of GNNs in contemporary virtual screening. We then propose an integrated GNN-guided virtual screening and core-hopping workflow tailored specifically to PPAR-γ modulator design, built around the goal of identifying selective partial agonists and biased ligands that retain insulin-sensitizing efficacy while minimizing helix-12-driven adverse transcriptional programs. Design considerations including selectivity across the PPAR α/β/δ/γ subfamily, avoidance of Ser273 phosphorylation, trans repression-biased anti-inflammatory signalling, and generative-model-assisted lead optimization are discussed in a tabulated, comparative format intended as a practical reference for computational and medicinal chemists developing next-generation PPAR-γ therapeutics.