Discovery of selective GluN1/GluN3A NMDA receptor inhibitors using integrated AI and physics-based approaches.

Abstract

N-methyl-D-aspartate receptors (NMDARs) are glutamate-gated ion channels essential for synaptic transmission and plasticity in the central nervous system. GluN1/GluN3A, an unconventional NMDAR subtype functioning as an excitatory glycine receptor, has been implicated in mood regulation, with high expression in brain regions governing emotional and motivational states. However, therapeutic exploration has been significantly hindered by a lack of potent and selective modulators, limited structural data and the intrinsic complexity of ion channels. Here, we introduce a compound virtual screening pipeline that combines artificial intelligence and physical models, integrating two sequence-based deep learning prediction models (TEFDTA and ESMLigSite) with a molecular docking approach. This approach was employed to identify potential inhibitors against GluN1/GluN3A by screening a commercial database containing 18 million compounds. The strategy resulted in an impressive hit rate of 50% for discovering inhibitors, with the most promising compound exhibiting strong inhibitory activity (IC₅₀ = 1.26 ± 0.23 μM) and remarkable target specificity (>23-fold selectivity over the GluN1/GluN2A receptor). These findings highlight the effectiveness of AI-assisted strategies in addressing challenges related to unconventional ion channels and pave the way for new therapeutic exploration.