Graph kernels for disease outcome prediction from protein-protein interaction networks

Area of Science:

• Bioinformatics
• Systems Biology
• Network Medicine

Background:

• Protein-protein interaction (PPI) networks are crucial for understanding disease mechanisms.
• Individualized PPI networks are not directly available, but gene expression data can be used to construct integrated networks.
• Current graph kernel methods struggle with the scale and missing data in biological networks.

Purpose of the Study:

• To develop novel graph kernels for comparing biological networks.
• To address the scalability and missing edge challenges in current graph kernel methods.
• To enable effective comparison of protein-protein interaction networks for disease research.

Main Methods:

• Development of novel graph kernels designed for biological network comparison.
• Implementation of methods that are computationally efficient for large-scale networks.
• Incorporation of strategies to account for biologically relevant missing interactions.

Main Results:

• The proposed graph kernels demonstrate fast computation times.
• The methods effectively handle missing interactions in biological networks.
• Practical performance was validated on two datasets of integrated gene expression/PPI networks.

Conclusions:

• The developed graph kernels offer a scalable and effective solution for comparing biological networks.
• These methods can facilitate the analysis of discrepancies in individual protein-protein interaction networks.
• This approach holds promise for advancing disease understanding and prevention through network medicine.