Machine Learning Uncovers Adverse Drug Effects on Intestinal Bacteria

Area of Science:

• Microbiology
• Pharmacology
• Computational Biology

Background:

• The human gut microbiome is crucial for health and influenced by diet, lifestyle, and medications.
• Drugs can significantly impact gut microbiome composition and function, potentially leading to disease.
• Current drug development processes do not routinely assess drug effects on the gut microbiome.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting drug-induced impairment of gut bacterial growth.
• To identify effective machine learning algorithms for predicting drug-microbiome interactions.

Main Methods:

• A machine learning model was trained on over 18,600 drug-bacteria interaction datasets.
• Thirteen distinct machine learning models, including tree-based, ensemble, and neural network techniques, were compared.
• Hyperparameter tuning and multi-metric evaluation (AUROC, recall, precision, f1-score) were performed to select the best model.

Main Results:

• A tuned extra trees algorithm was selected as the lead machine learning model.
• The model achieved an AUROC of 0.857 (±0.014), recall of 0.587 (±0.063), precision of 0.800 (±0.053), and f1-score of 0.666 (±0.042).
• The model demonstrated strong predictive performance for drug effects on bacterial growth.

Conclusions:

• The developed machine learning model can predict drug-induced impairment of gut bacterial growth.
• This tool can aid the pharmaceutical industry in drug development by assessing potential microbiome impacts.
• The model has potential applications in clinical settings for personalized medicine.