DeSIDE-DDI

DeSIDE-DDI predicts drug-drug interactions using deep learning on drug-induced gene expression signatures to identify potential adverse interactions during polypharmacy.

Key Features:

• Deep Learning Framework: Employs a deep learning framework that leverages drug-induced gene expression signatures to predict DDIs.
• Dynamic Drug Feature Engineering: Incorporates a gating mechanism and an attention mechanism to dynamically engineer drug features and model co-administration effects by focusing on relevant genes.
• Latent Space Projection: Projects side effects into a latent space using translating embedding techniques to capture complex relationships between drugs and interactions.
• Interpretability: Provides interpretable predictions by reporting gene expression levels implicated in predicted DDIs.
• High Predictive Accuracy: Reports an Area Under the Curve (AUC) of 0.889 and an Area Under the Precision-Recall Curve (AUPR) of 0.915 for predicting unseen interactions.
• Generalization to New Compounds: Predicts potential DDIs involving compounds not present in the training dataset.

Scientific Applications:

• Polypharmacy Risk Mitigation: Early identification of adverse DDIs to reduce unexpected harmful side effects in polypharmacy.
• Mechanistic Insight: Interpretation of drug-induced gene expression changes to provide molecular insights into DDI mechanisms.
• Drug Development Support: Prediction of interactions to inform compound prioritization and safety assessment during drug development.

Methodology:

Uses drug-induced gene expression signatures as input; employs a gating mechanism and attention mechanism for dynamic feature engineering to model co-administration effects; applies translating embedding techniques to project side effects into a latent space; trains and evaluates a deep learning model with performance assessed by AUC and AUPR for unseen interaction prediction.