Machine learning approaches for assessing medication transfer to human breast milk

Area of Science:

• Pharmacology and Toxicology
• Computational Biology and Bioinformatics
• Maternal and Child Health

Background:

• The human milk/plasma (M/P) drug concentration ratio is critical for assessing drug transfer into breast milk, impacting infant safety.
• Predicting this ratio is essential for managing medication use in breastfeeding mothers.
• Existing methods for M/P ratio determination can be resource-intensive.

Purpose of the Study:

• To evaluate the efficacy of various machine learning (ML) algorithms in predicting the M/P drug concentration ratio.
• To compare the performance of K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), and Neural Networks for M/P ratio prediction.
• To provide a data-driven tool for estimating drug transfer into breast milk.

Main Methods:

• Utilized a dataset of 162 drugs with 11 predictor variables.
• Employed binary (0, 1 and ≥1) and ternary (0 to <0.5, 0.5 to <1, and ≥1) categorization for M/P ratios.
• Applied five-fold cross-validation, Principal Component Analysis (PCA) for visualization, and Bayesian Information Criterion (BIC) for KNN model selection.

Main Results:

• Neural Networks achieved the highest average accuracies: 82% for the two-category system and 76% for the three-category system.
• K-Nearest Neighbors (KNN) showed 79% and 60% accuracy, Random Forest 77% and 64%, and Support Vector Machine (SVM) 78% and 67% respectively.
• All evaluated ML models demonstrated significant potential in predicting M/P ratios.

Conclusions:

• Machine learning techniques show considerable promise for predicting M/P drug ratios, aiding in drug development and risk assessment.
• These predictive models can inform clinical decisions regarding medication safety for lactating individuals.
• Further research with larger datasets is recommended to enhance the reliability and applicability of these ML models.