A knowledge graph embedding based approach to predict the adverse drug reactions using a deep neural network.

Area of Science:

• Pharmacology
• Bioinformatics
• Artificial Intelligence

Background:

• Adverse Drug Reactions (ADRs) pose significant risks in drug development, leading to injuries and fatalities.
• Accurate prediction of ADRs is crucial for drug safety but remains challenging due to data limitations.

Purpose of the Study:

• To develop a novel and effective method for predicting Adverse Drug Reactions (ADRs) using Artificial Intelligence.
• To address the limitations of existing methods in predicting ADRs, especially with sparse data.

Main Methods:

• Constructed a Knowledge Graph (KG) integrating drugs, ADRs, target proteins, indications, pathways, and genes.
• Employed Node2Vec algorithm for embedding KG nodes into a feature space.
• Designed and trained a custom Deep Neural Network (DNN), termed Knowledge Graph DNN (KGDNN), for ADR classification.

Main Results:

• The KGDNN model achieved a high AUROC score of 0.917.
• The proposed method significantly outperformed existing approaches for ADR prediction.
• Case studies demonstrated the model's efficacy in predicting liver injury and COVID-19 drug-related ADRs.

Conclusions:

• Knowledge Graph embedding combined with Deep Neural Networks offers a powerful approach for predicting Adverse Drug Reactions.
• The KGDNN model represents a significant advancement in computational drug safety and discovery.
• This method holds promise for enhancing the safety profile of new and existing medications.