CMMS-GCL: cross-modality metabolic stability prediction with graph contrastive learning.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Machine Learning

Background:

• Metabolic stability is critical for drug development, influencing candidate screening and lead optimization.
• Experimental assessment of metabolic stability is costly and time-consuming.
• In silico prediction offers an alternative, but robust and interpretable methods are limited.

Purpose of the Study:

• To develop a novel computational model for predicting molecular metabolic stability.
• To enhance the interpretability of predictions by identifying key functional groups.
• To provide an efficient and accurate tool for drug discovery and lead optimization.

Main Methods:

• Developed a cross-modality graph contrastive learning model (CMMS-GCL).
• Utilized deep learning for feature extraction from SMILES sequences (BiGRU encoder) and molecular graphs (graph contrastive learning encoder).
• Integrated sequence and structure representations using fully connected neural networks.

Main Results:

• CMMS-GCL outperformed seven state-of-the-art methods on two benchmark datasets.
• Demonstrated the model's interpretability through case studies and statistical analyses, identifying crucial functional groups.
• Achieved consistent improvements in predicting metabolic stability.

Conclusions:

• CMMS-GCL is an effective and interpretable tool for predicting drug metabolic stability.
• The model facilitates efficient drug candidate screening and lead compound optimization.
• Identifies critical functional groups, providing valuable insights for medicinal chemists.