A rule-based expert system for chemical prioritization using effects-based chemical categories.

Area of Science:

• Environmental chemistry
• Toxicology
• Computational chemistry

Background:

• Estrogen receptor (ER) binding is a key toxicological endpoint.
• Existing chemical inventories lack comprehensive ER binding data, especially for low-affinity interactions.
• Predictive models are needed for risk assessment of industrial and pesticide chemicals.

Purpose of the Study:

• To develop a mechanistically transparent rule-based expert system (ERES) for predicting chemical binding to the estrogen receptor.
• To apply the ERES to large inventories of industrial chemicals and pesticides.
• To identify chemicals with potential low-affinity ER binding.

Main Methods:

• Development of a logic-based decision tree expert system (ERES).
• Utilized "gold standard" assay data optimized for detecting low-affinity binding.
• Systematic application to two chemical inventories (>600 chemicals).

Main Results:

• The ERES was validated as a robust predictive model.
• The system identified specific chemical categories within seven major nodes.
• Approximately 5% of the chemicals in the tested inventories were predicted to bind to the ER.

Conclusions:

• The developed expert system provides a reliable method for predicting ER binding.
• The low percentage of predicted binders suggests most chemicals in the tested inventories have low ER affinity.
• This tool facilitates informed decision-making in chemical risk assessment.