DE-INTERACT: A machine-learning-based predictive tool for the drug-excipient interaction study during product...

Area of Science:

• Pharmaceutical Sciences
• Computational Chemistry
• Drug Development

Background:

• Drug-excipient compatibility is vital for pharmaceutical formulation stability.
• Conventional analytical methods can identify potential drug-excipient interactions.
• Machine learning offers a promising approach for predicting these interactions.

Purpose of the Study:

• To develop a predictive machine learning model for drug-excipient incompatibility.
• To utilize PubChem Fingerprints as molecular descriptors for drugs and excipients.
• To create a reliable tool for excipient selection in formulation design.

Main Methods:

• Employed PubChem Fingerprints (881-bit binary) as compound descriptors.
• Constructed a dataset of over 3500 drug-excipient instances from research papers.
• Trained an Artificial Neural Network (ANN) model, named DE-Interact, optimizing for accuracy, loss, and precision.

Main Results:

• The DE-Interact model achieved high training (0.9930) and validation (0.9161) accuracies.
• The model successfully predicted three known incompatible drug-excipient pairs: paracetamol-vanillin, paracetamol-methylparaben, and brinzolamide-polyethyleneglycol.
• Experimental validation using DSC, FTIR, HPTLC, and HPLC confirmed the model's predictions.

Conclusions:

• The developed DE-Interact tool provides a reliable and efficient method for predicting drug-excipient incompatibility.
• This tool can aid formulators in the early stages of drug development by facilitating excipient selection.
• Machine learning, specifically ANNs with PubChem Fingerprints, is effective for predicting complex chemical interactions in pharmaceuticals.