Network-based prediction and knowledge mining of disease genes

Area of Science:

• Bioinformatics
• Systems Biology
• Network Medicine

Background:

• High-throughput methods generate vast protein interaction data, revealing complex biological relationships.
• Network analysis and data mining uncover indirect molecular connections, crucial for understanding interdependent systems.
• Evaluating proteins within networks can identify relationships not apparent through isolated studies.

Purpose of the Study:

• To analyze the human protein interaction network in relation to human diseases.
• To develop a classifier for identifying disease-associated proteins using network topology.
• To explore novel disease-protein associations through computational prediction and literature validation.

Main Methods:

• Utilized the Disease Ontology to examine the human protein interaction network.
• Calculated topological metrics and trained an alternating decision tree (ADTree) classifier.
• Employed bootstrapping to identify conserved protein characteristics and validated predictions with literature evidence.

Main Results:

• Achieved 79% AUC in predicting disease-related genes, demonstrating classifier efficacy.
• Identified key network characteristics (degree centrality, disease neighbor ratio, etc.) distinguishing disease proteins.
• Discovered potential novel disease-related proteins and provided supporting literature evidence.

Conclusions:

• Protein interaction networks and neighbor disease associations are key indicators of disease relevance.
• The classifier identified unannotated proteins as potentially disease-related based on network properties.
• This approach offers a valuable tool for identifying new protein targets for drug repositioning.