Predicting disease genes based on multi-head attention fusion.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying disease-related genes is crucial for human disease diagnosis and treatment.
• Developing accurate computational methods for predicting disease-causing genes remains challenging due to data sparsity and complexity.
• Existing multi-feature fusion models struggle with the intricate nature of biomedical data.

Purpose of the Study:

• To propose an effective multi-feature fusion model for predicting disease-causing genes.
• To leverage heterogeneous biological information networks for enhanced gene-disease association prediction.
• To address the limitations of current computational methods in identifying pathogenic genes.

Main Methods:

• Constructing heterogeneous biological information networks by integrating multiple biomedical knowledge databases.
• Employing graph representation learning algorithms to extract feature vectors from gene-disease pairs.
• Fusing extracted features using a multi-head attention mechanism within the MHAGP model.
• Utilizing a multi-layer perceptron for the final prediction of gene-disease associations.

Main Results:

• The proposed MHAGP model demonstrated superior performance compared to existing methods in comparative experiments.
• Case studies confirmed the capability of MHAGP in predicting genes potentially associated with diseases.
• The model effectively fuses multi-source biological data for robust gene-disease association prediction.

Conclusions:

• The MHAGP model offers a significant advancement in disease-gene prediction accuracy.
• MHAGP shows potential for expansion to other prediction tasks, such as gene-drug and drug-disease associations.
• Future work can enhance MHAGP by incorporating additional biological entity association data to improve prediction accuracy.