Contrastive learning-based drug screening model for GluN1/GluN3A inhibitors

Area of Science:

• Neuroscience
• Computational Chemistry
• Drug Discovery

Background:

• GluN3A-containing NMDA receptors are key targets for neurological disorder treatments.
• Traditional drug screening methods face limitations in identifying potent modulators.
• Discovering effective GluN1/GluN3A receptor modulators is crucial for therapeutic advancement.

Purpose of the Study:

• Introduce CLG-DTA, a novel graph contrastive learning method for predicting drug-target affinity.
• Enhance drug discovery for the GluN1/GluN3A receptor using advanced computational techniques.
• Improve molecular representation by integrating natural language supervision and numerical knowledge graphs.

Main Methods:

• Developed CLG-DTA, a graph contrastive learning framework incorporating natural language supervision.
• Transformed regression labels into textual representations for enhanced molecular understanding.
• Utilized a numerical knowledge graph to refine continuous text embeddings for complex interactions.

Main Results:

• Screened 18 million compounds using CLG-DTA, identifying 12 potential drug candidates.
• Experimental validation confirmed significant activity in five compounds.
• Boeravinone E showed the highest potency with an IC50 of 3.40 ± 0.91 μM.

Conclusions:

• CLG-DTA effectively accelerates the identification of GluN1/GluN3A receptor modulators.
• The method demonstrates potential for advancing drug discovery in neurological disorders.
• This work provides a strong foundation for developing novel therapeutics targeting GluN1/GluN3A receptors.