Integrating node embeddings and biological annotations for genes to predict disease-gene associations

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate prediction of disease causative genes is crucial for understanding human diseases and guiding treatment.
• Existing computational methods can be enhanced by leveraging diverse biological information available for genes.

Purpose of the Study:

• To present N2VKO, an integrative framework for accurate disease gene prediction.
• To demonstrate the utility of node embeddings from protein-protein interaction networks combined with biological annotations.

Main Methods:

• Learning gene node embeddings from protein-protein interaction networks using node2vec.
• Integrating node embeddings with biological annotations to create rich feature representations.
• Employing binary classification models and oversampling techniques to address data imbalance in disease gene prediction.

Main Results:

• The N2VKO framework significantly outperforms four state-of-the-art methods across seven diseases.
• Integrating node embeddings with biological annotations enhances classification model performance.
• Oversampling techniques further improve prediction accuracy by correcting data imbalance.

Conclusions:

• Node embeddings from protein-protein interaction networks are effective for disease gene prediction.
• Combining node embeddings with biological annotations and using oversampling techniques boosts prediction performance.
• N2VKO demonstrates effectiveness in identifying potential disease genes, supported by literature validation.