Pharmacophoric-constrained heterogeneous graph transformer model for molecular property prediction

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Cheminformatics

Background:

• Accurate molecular representation is vital for AI-driven drug design.
• Existing atom-based graph representations do not fully leverage pharmacophore information (functional groups, chemical reactions).
• This limits the prediction of molecular properties.

Purpose of the Study:

• To develop a more informative molecular representation for enhanced molecule property prediction.
• To introduce the Pharmacophoric-constrained Heterogeneous Graph Transformer (PharmHGT) model.

Main Methods:

• Designed a pharmacophoric-constrained multi-view molecular representation graph.
• Developed a heterogeneous graph transformer architecture (PharmHGT).
• Extensive downstream experiments on molecular property prediction tasks.

Main Results:

• PharmHGT achieved superior performance compared to state-of-the-art models.
• Improvements included up to 1.55% in ROC-AUC and 0.272 in RMSE.
• Ablation and case studies confirmed PharmHGT's ability to capture pharmacophoric and chemical information.

Conclusions:

• The proposed pharmacophoric-constrained multi-view graph representation and PharmHGT model offer enhanced molecular representation capacity.
• PharmHGT significantly advances AI-driven molecular property prediction.
• This approach holds promise for accelerating drug design and discovery.