DeepGCGR: an interpretable two-layer deep learning model for the discovery of GCGR-activating compounds

Area of Science:

• Pharmacology and Computational Chemistry
• Artificial Intelligence in Drug Discovery
• Metabolic Disorders Therapeutics

Background:

• The glucagon receptor (GCGR) is a key therapeutic target for Type 2 Diabetes Mellitus (T2DM) and obesity.
• Developing effective GCGR agonists is challenging due to resource-intensive synthesis and screening processes.
• Artificial intelligence (AI) offers potential to accelerate drug discovery through efficient prediction of compound bioactivity.

Purpose of the Study:

• To develop and validate DeepGCGR, a novel two-layer deep learning model for identifying GCGR agonists.
• To expedite the screening of large chemical libraries and predict the functional effects of compounds on GCGR signaling.
• To discover new GCGR-regulating compounds from traditional Chinese medicine (TCM) for T2DM treatment.

Main Methods:

• Utilized a two-layer deep learning architecture, integrating graph convolutional networks (GCN) with a multiple attention mechanism.
• The first layer predicted compound bioactivity against GCGR, filtering chemical libraries.
• The second layer classified bioactive compounds based on their functional effects (agonistic/antagonistic) on GCGR signaling.

Main Results:

• DeepGCGR successfully predicted bioactivity and functional effects of compounds targeting GCGR.
• The model efficiently filtered large chemical libraries, identifying promising drug candidates.
• Novel GCGR-regulating compounds for T2DM treatment were identified from TCM natural products.

Conclusions:

• DeepGCGR provides an effective strategy for discovering functional GCGR agonists.
• The AI model accelerates the identification of potential therapeutics for T2DM.
• This approach offers new insights into developing novel treatments for metabolic disorders.