Predicting adverse drug effects: A heterogeneous graph convolution network with a multi-layer perceptron approach

Area of Science:

• Computational drug discovery
• Pharmacovigilance
• Bioinformatics

Background:

• Identifying potential drug side effects is crucial for patient safety and drug development.
• Existing methods for predicting drug side effects can be time-consuming and may miss subtle associations.
• Integrating diverse drug and side effect data is essential for comprehensive analysis.

Purpose of the Study:

• To develop and evaluate a novel in silico method, GCNMLP, for predicting potential drug side effects.
• To leverage heterogeneous graph convolution networks (GCN) and multi-layer perceptrons (MLP) for enhanced side effect prediction.
• To demonstrate the efficiency and superiority of the GCNMLP method compared to traditional approaches.

Main Methods:

• Utilized heterogeneous graph convolution networks (GCN) combined with multi-layer perceptrons (MLP) for drug side effect prediction.
• Integrated data from SIDER, OFFSIDERS, and FAERS datasets, focusing on drug characteristics and side effect networks.
• Employed network inference to identify relationships between similar drugs and their associated side effects.

Main Results:

• The GCNMLP method demonstrated superior performance compared to non-negative matrix factorization (NMF) and other machine learning methods across multiple evaluation metrics.
• The model successfully predicted potential side effects for specific drugs, including Vancomycin, Amlodipine, Cisplatin, and Glimepiride.
• The study successfully identified novel, previously undocumented side effects for investigated drugs.

Conclusions:

• The GCNMLP approach offers a significant advancement in computational drug side effect discovery.
• This in silico method can accelerate the identification of unseen side effects, reducing development time and improving drug safety.
• The findings underscore the importance of exploring drug mechanisms through well-documented data and advanced network analysis.