Evaluating disease similarity based on gene network reconstruction and representation

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Quantifying disease associations is crucial for understanding disease biology, diagnosis, and drug development.
• Existing methods often rely on gene set overlaps or ontology distances, failing to capture deep semantic information.
• Deep representation learning offers a promising avenue for more effective disease similarity computation.

Purpose of the Study:

• To develop a novel disease representation model for accurate disease similarity computation.
• To leverage deep representation learning by integrating gene-disease interactions and gene ontology hierarchies.
• To enhance understanding of disease biology and facilitate drug discovery through improved disease association analysis.

Main Methods:

• Constructed a novel gene network using HumanNet and Gene Ontology resources.
• Employed deep representation learning to learn gene and disease representations.
• Computed disease similarity using cosine similarity of learned disease representations.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.8074, a 10.1% improvement over the most competitive baseline.
• Demonstrated significant improvement in disease similarity computation accuracy.
• Validated the model's ability to learn effective disease representations.

Conclusions:

• The proposed deep learning model effectively computes disease similarity by integrating gene networks and ontology hierarchies.
• This approach offers significant advancements in disease association analysis, potentially aiding in gene-related disease prediction and drug development.
• The model's ability to capture deep semantic information represents a substantial step forward in bioinformatics research.