Property-Guided Few-Shot Learning for Molecular Property Prediction With Dual-View Encoder and Relation Graph...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Molecular property prediction is crucial for drug discovery.
• Deep learning methods struggle with limited experimental data for novel molecules or rare diseases.
• Accurate prediction is essential for efficient drug development.

Purpose of the Study:

• To propose PG-DERN, a novel few-shot learning model for molecular property prediction.
• To address the challenge of limited data in deep learning for drug discovery.
• To improve the accuracy and efficiency of predicting molecular properties.

Main Methods:

• Developed a dual-view encoder for integrated node and subgraph molecular representation.
• Introduced a relation graph learning module to enhance information propagation and prediction accuracy.
• Employed a MAML-based meta-learning strategy with a property-guided feature augmentation module.

Main Results:

• PG-DERN demonstrated superior performance compared to state-of-the-art methods on four benchmark datasets.
• The model effectively handles limited data scenarios in molecular property prediction.
• The integrated approach improved the comprehensiveness of molecular feature representation.

Conclusions:

• PG-DERN offers a robust solution for molecular property prediction with limited data.
• The proposed methods enhance molecular representation learning and information propagation.
• This advancement holds significant potential for accelerating drug discovery pipelines.