Ensemble learning for predicting ex vivo human placental barrier permeability

Area of Science:

• Pharmacology and Toxicology
• Computational Chemistry
• Reproductive Toxicology

Background:

• The placental barrier is crucial for fetal protection and chemical risk assessment.
• In vivo permeability studies are ethically restricted.
• Ex vivo placental perfusion is effective but resource-intensive.

Purpose of the Study:

• To develop a computational model for predicting ex vivo human placental barrier permeability.
• To provide an alternative to time-consuming experimental methods.
• To support drug development and environmental chemical risk assessment.

Main Methods:

• Utilized 87 chemicals and 1444 physicochemical properties.
• Employed linear regression, random forest, and ensemble algorithms.
• Developed prediction models for ex vivo placental barrier permeability.

Main Results:

• An ensemble model achieved high predictive accuracy (R²=0.887, R²=0.825 with applicability domain).
• External validation on new chemicals showed strong correlation (R²=0.879, improved to R²=0.921).
• The model adheres to OECD guidelines for predicting placental barrier permeability.

Conclusions:

• The first OECD-compliant predictive model for ex vivo human placental barrier permeability was established.
• This model offers a valuable tool for assessing chemical placental transfer.
• Integration with developmental toxicity models can enhance fetal risk evaluation.