Integrated QSAR Models to Predict Acute Oral Systemic Toxicity.

Area of Science:

• Toxicology
• Computational Chemistry
• Regulatory Science

Background:

• Regulatory agencies require reliable methods to assess acute oral systemic toxicity.
• Existing methods often rely on animal testing, which is costly and ethically challenging.
• The need for validated in silico models to support regulatory decision-making is increasing.

Purpose of the Study:

• To develop and validate in silico Quantitative Structure-Activity Relationship (QSAR) models for predicting acute oral systemic toxicity.
• To create models capable of classifying chemicals as very toxic (LD50 < 50 mg/kg) or nontoxic (LD50 >= 2,000 mg/kg).
• To support regulatory needs by providing alternative methods to experimental toxicity testing.

Main Methods:

• Development of new QSAR models using a Bayesian consensus approach integrating three classification algorithms.
• Application of models to a large dataset of 8992 chemicals, adhering to OECD principles for regulatory QSAR.
• Validation of model performance using a blind set of external molecules.

Main Results:

• The developed Bayesian consensus QSAR models demonstrated robust and predictive performance for both very toxic and nontoxic endpoints.
• Blind validation confirmed the reliability of the models on unseen external data.
• Integration of predictions allowed for the identification of compounds with medium toxicity and assessment of prediction consistency.

Conclusions:

• The developed in silico QSAR models are robust and predictive for acute oral systemic toxicity.
• These models show promise for reducing or replacing experimental toxicity tests in regulatory settings.
• The consensus approach and validation strategy contribute to the regulatory acceptance of in silico predictions.