SApredictor: An Expert System for Screening Chemicals Against Structural Alerts.

Area of Science:

• Computational toxicology
• cheminformatics
• Medicinal chemistry

Background:

• Accurate chemical toxicity evaluation is crucial for safety assessment.
• Many computational models lack interpretability, acting as 'black boxes'.
• Identifying specific toxicophores (structural alerts) is key to understanding toxicity mechanisms.

Purpose of the Study:

• To identify and collect structural alerts (SAs) associated with significant toxicity endpoints.
• To develop an interpretable tool for predicting chemical toxicity.
• To aid medicinal chemists in chemical structure optimization.

Main Methods:

• Systematic identification and curation of structural alerts (SAs) for various toxicity endpoints.
• Development of a database for efficient storage and retrieval of SAs.
• Creation of a user-friendly web server, SApredictor (www.sapredictor.cn), for chemical screening against SAs.

Main Results:

• A collection of SAs linked to key toxicity endpoints was established.
• The SApredictor web server was successfully developed and deployed.
• The server provides intuitive visualization of toxic substructures within chemical compounds.

Conclusions:

• SApredictor enhances chemical safety assessment by providing interpretable toxicity predictions.
• The tool facilitates the identification of specific toxicophores, guiding structural modifications.
• This approach improves the understanding of structure-toxicity relationships for safer chemical design.