Molecular Similarity in Computational Toxicology.

Area of Science:

• Cheminformatics
• Computational Toxicology
• Quantitative Structure-Activity Relationships (QSAR)

Background:

• Chemical similarity is crucial for predicting molecular properties and is fundamental to the read-across technique.
• The read-across approach estimates a chemical's properties using data from similar compounds.
• Existing similarity measurements can be refined for broader applicability.

Purpose of the Study:

• To implement and analyze various combinations of binary fingerprints and similarity metrics for chemical similarity.
• To develop a generalizable model for improving classical similarity measurements.
• To integrate the enhanced similarity approach into open-source computational toxicology tools.

Main Methods:

• Analysis of multiple binary fingerprint types.
• Evaluation of various similarity metrics.
• Implementation of a generalizable similarity model.
• Integration into CAESAR and VEGA software.

Main Results:

• Demonstrated improvement over classical similarity measurements using the developed model.
• Identified optimal combinations of fingerprints and metrics for read-across.
• Successful implementation in open-source tools, enhancing their predictive capabilities.

Conclusions:

• The proposed generalizable model significantly improves chemical similarity assessments for read-across.
• The findings provide a more robust framework for predicting chemical endpoints.
• The integration into CAESAR and VEGA facilitates wider adoption in computational toxicology.