Inhalation Toxicity Screening of Consumer Products Chemicals using OECD Test Guideline Data-based Machine Learning...

Area of Science:

• Environmental Chemistry
• Toxicology
• Computational Chemistry

Background:

• Consumer products like air fresheners and fragrances contain numerous chemicals.
• Assessing the inhalation toxicity of these chemicals is crucial for public health.
• Existing regulatory frameworks may have gaps in addressing inhalation risks from consumer products.

Purpose of the Study:

• To screen inhalation toxicity of chemicals in consumer products using machine learning.
• To identify chemicals of high concern within consumer product ingredients.
• To evaluate the potential of AI in chemical risk assessment and prioritization.

Main Methods:

• Manually curated inhalation toxicity data for 1709 chemicals from OECD databases.
• Trained machine learning models using ten algorithms and four molecular fingerprints/descriptors.
• Validated models on occupational chemicals to assess prediction uncertainty.
• Applied validated models to screen chemicals used in consumer products.

Main Results:

• Machine learning models achieved F1 scores from 51% to 91% in test datasets.
• Virtual screening identified 79 consumer product chemicals as high concern for inhalation toxicity.
• Prioritized chemicals often lacked GHS classifications for inhalation toxicity.
• Models demonstrated high sensitivity (75%) but low specificity (23%) for conservative screening.

Conclusions:

• AI models can effectively screen consumer product chemicals for inhalation toxicity.
• A potential regulatory blind spot exists for inhalation risks of consumer product chemicals.
• This study underscores the utility of AI in chemical safety assessment and regulatory prioritization.