A Machine Learning Model to Predict Drug Transfer Across the Human Placenta Barrier

Area of Science:

• Computational chemistry
• Pharmacology
• Toxicology

Background:

• Assessing chemical transfer across the placental membrane is crucial for developing safe therapeutics during pregnancy.
• Existing methods for evaluating placental drug transfer can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop a low-dimensional machine learning model for classifying compounds based on their ability to cross the placental barrier.
• To create a predictive tool that aids in the early stages of drug discovery and virtual screening.

Main Methods:

• Compiled a database of 248 compounds with experimental placental transfer data.
• Characterized compounds using approximately 5.4 thousand physicochemical and structural descriptors.
• Employed a genetic algorithm for feature selection within a five-cross-validation framework.
• Evaluated various machine learning classifiers, optimizing for minimal false positives.

Main Results:

• A Linear Discriminant Analysis model utilizing only four structural features demonstrated high robustness.
• The optimized model achieved a minimal false positive rate, with only one instance across all testing folds.
• The model effectively classifies compounds regarding their placental barrier permeability.

Conclusions:

• The developed machine learning model is a valuable tool for predicting placental drug transfer.
• This predictive capability can serve as an efficient filter for chemical libraries in virtual screening campaigns.
• The model contributes to the development of safer therapeutic options during pregnancy.