Towards safer and efficient formulations: Machine learning approaches to predict drug-excipient compatibility

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Drug Development

Background:

• Accurate prediction of drug-excipient compatibility is essential for successful pharmaceutical formulation.
• Existing methods for predicting drug-excipient interactions have limitations in accuracy and scope.

Purpose of the Study:

• To develop and validate an advanced machine learning model for enhanced drug-excipient compatibility prediction.
• To create a user-friendly web platform for accessible drug-excipient compatibility assessment.

Main Methods:

• Utilized Mol2vec and 2D molecular descriptors as input features.
• Employed a stacking ensemble technique to improve predictive model performance.
• Integrated machine learning models into an interactive web application.

Main Results:

• Achieved high predictive performance with accuracy (0.98), precision (0.87), recall (0.88), AUC (0.93), and MCC (0.86).
• The stacking model demonstrated superior performance compared to the DE-INTERACT benchmark, correctly identifying 10/12 incompatibility cases.
• A web platform was launched for user-friendly input and prediction generation.

Conclusions:

• The developed machine learning approach offers a significant advancement in predicting drug-excipient compatibility.
• The web-based tool provides accessible and rapid compatibility predictions.
• Further model refinement is recommended for practical clinical application.