MGREL: A multi-graph representation learning-based ensemble learning method for gene-disease association prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying gene-disease associations is crucial for understanding disease mechanisms and finding therapeutic targets.
• Computational methods offer an effective approach but often overlook integrated data types.
• Existing methods may not fully leverage genetic, therapeutic, and network topology information.

Purpose of the Study:

• To develop an advanced computational model for predicting gene-disease associations.
• To integrate diverse data sources including genetic, therapeutic, and network information.
• To improve the accuracy and scope of gene-disease association prediction.

Main Methods:

• Re-organized a benchmark dataset with updated gene-disease associations from the OMIM database.
• Developed a multi-graph representation learning-based ensemble model (MGREL).
• Integrated knowledge extraction and graph learning channels for feature generation, followed by a 5-model ensemble classifier.

Main Results:

• MGREL demonstrated superior performance compared to five state-of-the-art methods.
• Achieved high performance metrics: AUC = 0.925 and AUPR = 0.935.
• Showcased significant improvements in predicting associations for novel genes and diseases, with relative gains of 3.24% (AUC) and 2.75% (AUPR).

Conclusions:

• MGREL effectively integrates multi-modal data for accurate gene-disease association prediction.
• The model shows promise for identifying novel therapeutic targets.
• This approach advances computational strategies in precision medicine and disease research.