Identifying Structure-Property Relationships through SMILES Syntax Analysis with Self-Attention Mechanism

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Structure-activity relationship (SAR) and structure-property relationship (SPR) studies are crucial for molecular design.
• Traditional methods rely on partitioning connection tables into predefined substructures.
• Linear notations like SMILES offer an alternative molecular representation.

Purpose of the Study:

• To develop a novel method for identifying SAR/SPR using SMILES syntax analysis.
• To leverage deep learning, specifically a self-attention mechanism, for interpretable SAR/SPR analysis.
• To improve the prediction of chemical properties, toxicology, and bioactivity.

Main Methods:

• Utilized SMILES linear notation as the input molecular representation.
• Applied a self-attention mechanism, an interpretable deep learning architecture, for syntax analysis.
• Evaluated the method on experimental datasets for chemical property, toxicology, and bioactivity prediction.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art models.
• Achieved high accuracy in predicting chemical properties, toxicology, and bioactivity.
• Generated chemically interpretable results, aiding compound design.

Conclusions:

• The SMILES syntax analysis with self-attention offers a powerful and interpretable approach for SAR/SPR studies.
• This method enhances the prediction of molecular properties and bioactivity.
• Facilitates the design and synthesis of improved chemical compounds.