DeepPSE

DeepPSE predicts polypharmacy side effects using deep learning to identify adverse effects arising from concurrent drug combinations.

Key Features:

• Dual-Module Architecture: DeepPSE employs a dual-module architecture that separates representation learning and latent feature extraction for drug pairs.
• Neural Network Fusion: The first module fuses multiple types of neural networks to learn intricate representations of drug pairs at the molecular level.
• Transformer Encoder with Self-Attention: The second module uses a transformer encoder with a self-attention mechanism to extract latent features critical for complex drug–drug interactions.
• End-to-End Prediction: Learned representations and extracted features are combined and fed into a fully connected layer to produce direct polypharmacy side-effect predictions.
• Benchmark Performance: On a dataset of 96,473 drug pairs and 964 types of polypharmacy side effects, DeepPSE demonstrated superior performance relative to five baseline or state-of-the-art methods.

Scientific Applications:

• Pharmacology: Predicting adverse effects of drug combinations to inform safety assessments in pharmacological research.
• Personalized medicine: Informing prescription decisions to mitigate patient-specific risks associated with polypharmacy and improve treatment efficacy.

Methodology:

DeepPSE fuses multiple neural network types to learn drug-pair representations, applies a transformer encoder with self-attention to refine latent features, and uses a fully connected layer for end-to-end side-effect prediction.