An Explainable Framework for Predicting Drug-Side Effect Associations via Meta-Path-Based Feature Learning in...

Area of Science:

• Biomedical Informatics
• Computational Pharmacology
• Drug Discovery

Background:

• Accurate identification of drug side effects (DSAs) is crucial for drug development and safety surveillance.
• Traditional methods for DSA identification are time-consuming, costly, and potentially incomplete.
• Existing computational methods often fail to fully integrate multiple databases, capture complex semantics, or provide explainability.

Purpose of the Study:

• To develop a novel computational method for predicting drug-side effect associations (DSAs).
• To address limitations of existing methods, including underutilization of data, inadequate semantic capture, and lack of explainability.
• To enhance drug safety surveillance and drug development through accurate DSA prediction.

Main Methods:

• Construction of a heterogeneous information network (HIN) integrating multiple biological datasets.
• Application of a meta-path-based feature learning module to capture complex drug-side effect semantics within the HIN.
• Development of a prediction module utilizing learned features for DSA prediction and explainability.

Main Results:

• The proposed MPGNN-DSA model demonstrated significant effectiveness in predicting drug-side effect associations.
• The meta-path-based approach successfully captured complex semantics among drugs and side effects.
• The method provided explainability for the predicted DSAs, enhancing interpretability.

Conclusions:

• MPGNN-DSA offers a feasible and effective solution for drug-side effect association prediction.
• The model's ability to integrate diverse data and provide explainable predictions advances computational pharmacology.
• This approach holds promise for improving drug safety and accelerating drug development processes.