NNPS

NNPS predicts polypharmacy side effects by integrating mono side effects and drug-protein interaction data within a neural network model to identify adverse interactions in multi-drug regimens.

Key Features:

• Novel Feature Vectors: Integrates mono side effects and drug-protein interaction data into feature vectors for each drug.
• Neural Network Architecture: Employs a neural network architecture tailored to model interactions in polypharmacy data.
• Performance Metrics: Outperforms Decagon, Concatenated drug features, Deep Walk, DEDICOM, and RESCAL with AUROC improved by approximately 9.2%, AUPRC increased by about 12.8%, F-score enhanced by 8.6%, Accuracy boosted by 10.3%, and Matthews Correlation Coefficient (MCC) improved by 18.7%.
• Efficiency in Computation: Achieves approximately 8 hours per fold of cross-validation compared to Decagon's reported 15 days per fold.
• Robust Validation: Validated on a benchmark dataset of 964 polypharmacy side effects using 5-fold cross-validation over 50 iterations.

Scientific Applications:

• Pharmacology: Predicting adverse drug interactions in polypharmacy scenarios to support risk assessment of multi-drug regimens.

Methodology:

Constructs feature vectors from mono side effects and drug-protein interaction data; trains a neural network model; conducts comparative evaluation against Decagon, Concatenated drug features, Deep Walk, DEDICOM, and RESCAL; and validates results using 5-fold cross-validation repeated over 50 iterations on a dataset of 964 polypharmacy side effects.