An efficient deep learning-based strategy to screen inhibitors for GluN1/GluN3A receptor.

Abstract

The GluN1/GluN3A receptor, a unique excitatory glycine receptor recently identified in the central nervous system, challenges traditional perspectives of N-methyl-D-aspartate (NMDA) receptor diversity and glycinergic signaling. Its role in emotional regulation positions it as a potential therapeutic target for neuropsychiatric disorders. However, pharmacological research on GluN1/GluN3A receptors remains at an early stage. Traditional high-throughput screening methods for ion channel drug discovery often lack efficiency, particularly when applied to large compound libraries. To address this concern, we designed a deep learning-based strategy that balances efficiency and accuracy for identifying GluN1/GluN3A inhibitors. First, a sequence-based scoring function was developed to rapidly screen a library containing 18 million compounds, reducing the pool to approximately 10⁵ candidates. Next, two complex-based scoring functions, IGModel and RTMScore, were employed to precisely score and rank the remaining candidates. Finally, an active molecule with an IC₅₀ of 2.87 ± 0.80 μM for the GluN1/GluN3A receptor was confirmed through whole-cell voltage-clamp electrophysiology. This study also presents a paradigm for integrating deep learning into rapid and precise high-throughput screening.