An Improved Framework for Drug-Side Effect Associations Prediction via Counterfactual Inference-Based Data...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Pharmacovigilance

Background:

• Accurate detection of drug-side effect associations (DSAs) is crucial for drug development.
• Network-based computational methods are widely used for DSA prediction but face challenges due to data scarcity.
• Existing data augmentation techniques often randomly manipulate networks, neglecting causal relationships and hindering prediction accuracy.

Purpose of the Study:

• To propose a novel counterfactual inference-based data augmentation method to improve drug-side effect association prediction.
• To address the data scarcity issue in predicting DSAs.
• To enhance the performance of computational models for DSA prediction.

Main Methods:

• Constructed a heterogeneous information network (HIN) by integrating diverse biomedical data.
• Applied community detection on the HIN to develop a counterfactual inference method for deriving augmented links, creating an augmented HIN.
• Utilized a meta-path-based graph neural network (GNN) for learning drug and side effect representations to predict DSAs.

Main Results:

• The proposed counterfactual inference-based data augmentation method effectively addresses data scarcity in DSA prediction.
• The augmented HIN, generated through causal inference, leads to improved prediction accuracy.
• Experimental results demonstrate the significant effectiveness of the proposed method over existing approaches.

Conclusions:

• Counterfactual inference-based data augmentation offers a promising strategy to overcome data scarcity in DSA prediction.
• The integration of HINs and GNNs, augmented with causal reasoning, enhances the reliability of computational drug safety assessments.
• This approach provides a robust framework for advancing computational methods in pharmacovigilance and drug development.