LGGA-MPP: Local Geometry-Guided Graph Attention for Molecular Property Prediction

Area of Science:

• Computational chemistry
• Cheminformatics
• Artificial intelligence in drug discovery

Background:

• Molecular property prediction is crucial for drug discovery.
• Deep learning models are increasingly used but often neglect 3D molecular information.
• Existing self-supervised learning (SSL) methods focus on global geometry, missing local structural details.

Purpose of the Study:

• To develop a novel approach for molecular representation learning that captures local 3D geometry.
• To improve the accuracy and interpretability of molecular property prediction models.
• To enhance the capabilities of graph neural networks (GNNs) for drug discovery tasks.

Main Methods:

• Proposed Local Geometry-Guided Graph Attention (LGGA) model.
• Integrated local molecular geometry into the attention mechanism and message-passing of GNNs.
• Developed a new method for modeling molecules to capture intricate local structural details.

Main Results:

• LGGA significantly improved molecular property prediction performance across various datasets.
• The integration of local geometry demonstrably enhanced model results.
• The proposed model outperformed existing state-of-the-art methods.

Conclusions:

• LGGA offers a promising approach for molecular property prediction by effectively utilizing local 3D geometric information.
• The method enhances the ability of GNNs to capture complex molecular structures.
• LGGA shows potential as a valuable tool in accelerating drug discovery and related research.