DrIVeNN: Drug Interaction Vectors Neural Network.

Area of Science:

• Pharmacology and Bioinformatics
• Computational Drug Discovery
• Artificial Intelligence in Medicine

Background:

• Polypharmacy, the use of multiple drugs, increases the risk of adverse drug events (ADEs), particularly drug-drug interactions (DDIs).
• Predicting DDIs is challenging as clinical trials cannot test all drug combinations.
• Older adults with cardiovascular disease (CVD) are particularly vulnerable to polypharmacy-related ADEs.

Purpose of the Study:

• To identify key drug features for predicting DDIs.
• To develop and evaluate a novel computational model for DDI prediction.
• To assess the model's performance on a CVD-specific case study.

Main Methods:

• Developed a two-layer neural network, DrIVeNN (drug interaction vectors neural network).
• Incorporated drug features including molecular structure, drug-protein interactions, and mono-drug side effects.
• Utilized publicly available side effect databases for model training and evaluation.

Main Results:

• DrIVeNN achieved superior performance compared to state-of-the-art models (DGNN-DDI, KGDDI, NNPS).
• DrIVeNN demonstrated average AUROC of 0.934 and AUPRC of 0.920.
• A domain-specific CVD case study showed enhanced performance with an average AUROC of 0.979 for DDI prediction.

Conclusions:

• The DrIVeNN model shows significant potential for predicting polypharmacy ADEs.
• Domain-specific models can further improve the accuracy of DDI prediction.
• This research advances predictive modeling techniques for drug safety.