Quantifying Analogue Suitability for SAR-Based Read-Across Toxicological Assessment.

Area of Science:

• Toxicology and Cheminformatics
• Computational Chemistry
• Regulatory Science

Background:

• Structure-activity relationship (SAR)-based read-across is crucial for toxicological safety assessment.
• Justifying analogue selection and prediction reliability in SAR read-across remains a significant challenge.
• Current methods often rely solely on structural similarities, neglecting biological relevance.

Purpose of the Study:

• To develop a quantitative approach for assessing analogue suitability in SAR read-across for systemic toxicity predictions.
• To incorporate biological and toxicological features beyond structural comparisons.
• To enhance the transparency, consistency, and regulatory acceptance of read-across assessments.

Main Methods:

• Development of fingerprint keys to quantitatively compare metabolism, reactivity, and physicochemical properties.
• Application of these fingerprints to assess similarity between target chemicals and potential analogues across 14 case studies.
• Utilization of machine learning to determine the contribution and importance of each similarity attribute.

Main Results:

• The proposed non-structural similarity scores align with expert judgment for read-across suitability.
• Machine learning analysis revealed the relative importance of different similarity attributes for various chemical classes.
• The approach provides a numerical score and interpretable fingerprints for quantifying analogue differences and ranking quality.

Conclusions:

• This quantitative, biologically informed approach improves the justification and reliability of SAR read-across predictions.
• The method enhances transparency and consistency, facilitating broader implementation and regulatory acceptance.
• Incorporating metabolism, reactivity, and physicochemical properties is key to robust read-across for systemic toxicity.