IN3-Structure: AI-Enhanced Ligand–Receptor Profiling for Drug Repurposing

Maria Mina, Nikolaos Kyriakos Fytilis, Minos-Timotheos Matsoukas, Theodora Katsila

Keywords: G-protein-coupled receptors, Drug Repurposing, AlphaFold3, Boltz-2, Structural Biology, Artificial Intelligence, Ligand-Receptor Affinity, Class A GPCRs, In Silico Screening, Cardiovascular Drugs.

Domains: AI for Health, AI for Science

TL;DR: IN3-STRUCTURE is an AI-driven pipeline that combines AlphaFold3 and Boltz-2 to accurately predict drug-receptor interactions, already successfully validating 61,851 GPCR-drug pairs for high-throughput drug repurposing.

Abstract:

G-protein-coupled receptors (GPCRs) constitute one of the most important protein families in human biology, regulating key physiological processes, such as metabolism, immune responses, and neuronal signalling1,2. They are also among the most widely targeted proteins in pharmacology, with approximately 40% of FDA-approved drugs acting on GPCRs. However, their intrinsic structural flexibility and the limited number of experimentally resolved structures pose significant challenges for drug discovery and repurposing.3-6 5,6
Recent advances in Artificial Intelligence (AI) have transformed structural biology7,8. Models such as AlphaFold2, and more recently AlphaFold3, enable high-accuracy prediction of protein structures and complexes, while emerging tools such as Boltz-2 allow for quantitative estimation of ligand–receptor affinity9,10. Leveraging these developments, we present IN3-STRUCTURE, an AI-driven computational pipeline that integrates structure prediction and affinity modelling to systematically explore GPCR–drug interactions.
Our approach combines AlphaFold3-based structural modelling with Boltz-2 affinity scoring to generate and evaluate GPCR–drug complexes in both active and inactive conformational states for all currently FDA-approved drugs and Class A GPCR families. As a paradigm, we focus on cardiovascular-relevant Class A GPCR families, including adrenergic, muscarinic, purinergic, sphingosine-1-phosphate, lysophosphatidic acid, prostanoid, dopamine, and cannabinoid receptors, providing a pharmacologically diverse testbed.
By integrating confidence metrics from both AI models into a unified scoring framework, we evaluated 61,851 receptor–drug pairs. The top-ranked interactions exclusively corresponded to clinically validated drug–target relationships, demonstrating high predictive precision and zero false positives within the top hits.
Overall, IN3-STRUCTURE highlights how AI can bridge 3D (structural biology) and 1D (multi-omics, translational biomarkers) data, empowering our IN3© engine (in silico, in vitro, in/ex vivo) by enabling scalable, data-driven drug repurposing.

Submission Number: 52