Group graph: a molecular graph representation with enhanced performance, efficiency and interpretability.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Navigating chemical space is crucial for identifying new chemical entities.
• Atom-level molecular representations like SMILES can obscure substructure information.
• Substructure-level representations offer richer insights into molecular properties and interactions.

Purpose of the Study:

• To develop a novel substructure-level molecular representation for intact and simple encoding of entire molecular structures.
• To address the limitations of existing molecular representations in capturing complex chemical information.
• To enhance the prediction of molecular properties and drug-drug interactions.

Main Methods:

• Developed a novel substructure-level molecular representation termed 'group graph'.
• Utilized graph isomorphism network (GIN) for property prediction and analysis.
• Evaluated the group graph's performance against other molecular graph models.

Main Results:

• The group graph effectively represents molecular structural features with minimal information loss.
• GIN models based on group graphs demonstrated superior accuracy and efficiency in predicting molecular properties and drug-drug interactions.
• The group graph facilitated the prediction of structural modifications to improve blood-brain barrier permeability (BBBP).

Conclusions:

• The group graph is a powerful tool for representing both local and global molecular features.
• This novel representation enhances the interpretability and efficiency of predicting molecular properties and drug-drug interactions.
• The group graph shows significant potential for advancing drug discovery and development.