Multi-MoleScale: a multi-scale approach for molecular property prediction with graph contrastive and sequence...

Area of Science:

• Computational chemistry
• Machine learning
• Bioinformatics

Background:

• Machine learning models show promise in predicting molecular properties.
• Integrating molecular graph structures with sequence information remains a challenge.

Purpose of the Study:

• Introduce Multi-MoleScale, a novel multi-scale framework to address the challenge of integrating molecular graph and sequence data.
• Enhance the prediction of molecular properties by capturing both structural and contextual representations.

Main Methods:

• Combine Graph Contrastive Learning (GCL) with sequence-based models like BERT.
• Leverage GCL for intrinsic graph-based features and BERT for contextual relationships.
• Utilize contrastive learning to distinguish relevant molecular features.

Main Results:

• Multi-MoleScale consistently outperforms existing deep learning and self-supervised learning approaches on public datasets.
• Demonstrated strong performance on 12 molecular property datasets, the ADMET dataset, and 14 breast cancer cell line datasets.
• The model does not require handcrafted features, showing high adaptability.

Conclusions:

• Multi-MoleScale is a promising tool for molecular discovery tasks, including drug discovery and materials science.
• The framework effectively captures both structural and contextual molecular information.
• The approach offers enhanced predictive accuracy and versatility for various molecular research fields.