HExpPredict: In Vivo Exposure Prediction of Human Blood Exposome Using a Random Forest Model and Its Application in...

Area of Science:

• Environmental health sciences
• Toxicology
• Computational chemistry

Background:

• Limited data exists on human exposure to numerous exposome substances, hindering health risk assessment.
• Quantifying all trace organics in biological fluids is challenging and costly due to high individual variability.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting chemical concentrations in human blood.
• To prioritize chemicals of potential health concern based on predicted blood levels and bioactivity data.

Main Methods:

• A machine learning model was developed using daily exposure, exposure pathway indicators, half-lives, and volume of distribution.
• Random Forest (RF) model outperformed Artificial Neural Network (ANN) and Support Vector Regression (SVR) models.
• Toxicity potential was estimated using bioanalytical equivalency (BEQ) based on predicted blood concentrations and ToxCast bioactivity data.

Main Results:

• The RF model achieved a root mean square error (RMSE) of 1.66 and mean absolute error (MAE) of 1.28.
• Human blood concentrations were successfully predicted for 7,858 ToxCast chemicals.
• Prioritization revealed food additives and pesticides as highly active compounds, surpassing widely monitored pollutants.

Conclusions:

• Accurate prediction of internal exposure from external exposure data is feasible.
• This predictive capability is valuable for chemical risk prioritization and understanding human exposure to a wide range of chemicals.