In Silico Prediction of Oral Acute Rodent Toxicity Using Consensus Machine Learning.

Area of Science:

• Toxicology
• Computational Chemistry
• Data Science

Background:

• Acute oral toxicity (AOT) assessment is crucial for chemical classification and labeling under the Globally Harmonized System (GHS).
• Traditional AOT studies involve animal testing, prompting a need for alternative methods aligned with the 3Rs principles (Replacement, Reduction, Refinement).
• In silico methods, particularly machine learning, present a viable alternative for predicting AOT for new chemical substances.

Purpose of the Study:

• To develop and validate a robust in silico model for predicting acute oral toxicity.
• To evaluate various molecular representations and machine learning algorithms for optimal AOT prediction.
• To establish an applicability domain for the developed consensus model.

Main Methods:

• Compilation of acute oral toxicity data from a commercial database.
• Development of a consensus classification model by assessing diverse molecular representations and machine learning algorithms.
• Evaluation of model performance using an external validation dataset sourced from scientific literature.

Main Results:

• The developed consensus model demonstrated superior predictive performance compared to existing publicly available AOT models.
• The model's efficacy was confirmed through rigorous evaluation on an independent external validation dataset.
• An applicability domain was successfully defined for the model, specifying its reliable prediction range.

Conclusions:

• The developed in silico model provides a highly accurate and reliable alternative for acute oral toxicity assessment.
• This computational approach supports the 3Rs principles by reducing reliance on animal testing for chemical safety evaluations.
• The model and its defined applicability domain facilitate informed chemical classification and labeling decisions in a regulatory context.