A drug molecular classification model based on graph structure generation.

Area of Science:

• Artificial intelligence
• Cheminformatics
• Computational chemistry

Background:

• Molecular property prediction accelerates drug discovery and reduces costs.
• Graph neural networks (GNNs) are widely used but face challenges in node representation learning.
• Existing GNNs struggle with exponential node growth and noise, limiting depth exploration and key structure identification.

Purpose of the Study:

• To propose a novel graph neural network model based on structure generation for molecular property prediction.
• To address the limitations of existing GNNs in depth exploration and noise reduction.
• To improve the accuracy and efficiency of molecular property prediction for drug discovery.

Main Methods:

• A depth-first strategy is employed for key graph structure generation, enhancing exploration ability.
• A tendentious node selection method is utilized to gradually select nodes and edges, mitigating noise.
• An attention mechanism and random bias are incorporated for forward propagation and iterative optimization of structure generation.

Main Results:

• The proposed model demonstrates superior performance in molecular property prediction compared to existing methods.
• Experimental results on public datasets show improved classification accuracy.
• The model effectively addresses the challenges of limited depth exploration and noise in GNNs.

Conclusions:

• The novel GNN model based on structure generation offers a promising advancement in molecular property prediction.
• This approach has the potential to significantly speed up drug discovery and reduce associated costs.
• The model's ability to learn key structures efficiently provides a robust tool for cheminformatics and computational chemistry applications.