Automatic knowledge extraction from chemical structures: the case of mutagenicity prediction

Area of Science:

• Computational chemistry and cheminformatics
• Toxicology and drug discovery

Background:

• Structure-activity relationship (SAR) studies are crucial for understanding molecular interactions.
• Identifying structural alerts is key for predicting biological activity and potential toxicity.
• Existing methods may lack transparency or user control in fragment mining.

Purpose of the Study:

• To develop a novel SAR approach for mining molecular fragments associated with biological activity.
• To create a system that aligns with human reasoning for improved reliability and user customization.
• To validate the approach using the mutagenicity endpoint.

Main Methods:

• Development of a new SAR methodology focused on fragment identification.
• Integration of human-interpretable reasoning into the prediction process.
• Application and testing of the approach on mutagenicity data.

Main Results:

• The proposed SAR approach demonstrated significant predictive capabilities for mutagenicity.
• The method successfully identified known structural alerts from existing literature.
• New evidence and previously unrecognized molecular fragments were uncovered.

Conclusions:

• The novel SAR approach offers a reliable and user-controllable method for identifying biologically active molecular fragments.
• This technique aids in the discovery of toxicological and pharmacological knowledge.
• The approach holds promise for applications in drug discovery and safety assessment.