Bridging the Gap between External and Internal Exposure: Challenges and a Machine Learning Approach Validated Using...

Area of Science:

• Environmental Health Sciences
• Toxicology
• Computational Biology

Background:

• Quantitative linkages between external and internal exposures are crucial for health risk assessment but face significant challenges.
• These challenges span complex exposure pathways, data gaps, biological sample limitations, and interindividual variability.

Purpose of the Study:

• To review challenges in establishing external-internal exposure linkages.
• To propose a novel machine learning framework (ML-EIExpLink) for robust external-internal exposure modeling.

Main Methods:

• Systematic review of challenges in exposure assessment.
• Development of a machine learning framework integrating diverse data sources (pollutant properties, ADME, exposure factors, environmental, socioeconomic).
• Application of gradient boosting decision trees for predictive modeling.

Main Results:

• The ML-EIExpLink framework demonstrated feasibility and reliability in a case study on benzene ring pollutants.
• The model achieved strong external validation performance (Q_ext² = 0.784).
• Key variables like air concentration and temperature showed nonlinear and threshold effects on predicted blood concentrations.

Conclusions:

• The proposed ML-based framework establishes robust and interpretable external-internal exposure linkages.
• This approach offers a valuable tool for enhancing exposure assessment and health risk evaluation.
• The findings highlight the importance of integrating multiple data dimensions for accurate exposure modeling.