An investigation into pharmaceutically relevant mutagenicity data and the influence on Ames predictive potential

Area of Science:

• Computational chemistry
• Drug discovery
• Toxicology

Background:

• Positive Ames tests for bacterial mutation hinder drug development.
• Previous work predicted aryl-amine genotoxicity using quantum mechanics (QM) calculations.
• This study investigates molecular descriptors to enhance QM predictions and assess external datasets for training.

Purpose of the Study:

• To improve genotoxicity predictions beyond QM reaction energy.
• To evaluate the utility of external datasets for quantitative structure-activity relationship (QSAR) model training.
• To identify key molecular descriptors for predicting Ames test outcomes.

Main Methods:

• Comparison of various QM descriptors including orbital energies, nitrenium formation energies, and anion formation energies.
• Evaluation of external datasets against internal Novartis data.
• Analysis of chemical fingerprints and substructure prevalence using unsupervised clustering and Kohonen Self-Organizing Maps.

Main Results:

• Nitrenium formation energy was the most effective single descriptor for predicting Ames test outcomes.
• External datasets showed better performance than the Novartis set, indicating significant data profile differences.
• Unsupervised clustering revealed distinct chemical profiles between Novartis and external datasets.

Conclusions:

• Developing relevant test sets for drug discovery compounds is essential.
• QSAR models require careful validation before replacing experimental data.
• A random forest model trained on combined Novartis and external data showed suitability for predicting genotoxicity across diverse chemical structures.