MS-BACL: enhancing metabolic stability prediction through bond graph augmentation and contrastive learning

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Accurate prediction of molecular metabolic stability is vital for drug safety and efficacy.
• Current deep learning models, particularly graph neural networks, are limited by focusing on atom-atom interactions and neglecting bond relationships.

Purpose of the Study:

• To develop a novel model for enhanced molecular metabolic stability prediction.
• To address the limitations of existing methods by incorporating bond-level information.

Main Methods:

• Proposed the MS-BACL model utilizing bond graph augmentation and contrastive learning.
• Constructed an 'atom-bond-atom' bond graph to capture both atom and bond information.
• Employed contrastive learning on molecular and bond graphs for robust molecular representation.

Main Results:

• The MS-BACL model effectively predicts molecular metabolic stability.
• Incorporating bond-to-bond relationships is a novel approach for this task.
• Experimental results on public datasets demonstrate superior performance compared to state-of-the-art models.

Conclusions:

• The MS-BACL model offers a more accurate and reliable method for predicting molecular metabolic stability.
• Considering bond-level information significantly enhances molecular representation and predictive power.
• This approach advances the application of deep learning in drug discovery and development.