Perceptron ensemble of graph-based positive-unlabeled learning for disease gene identification

Area of Science:

• Bioinformatics
• Computational Biology
• Genetics

Background:

• Disease gene identification is crucial in biomedical research.
• Machine learning, particularly positive-unlabeled learning, shows promise for this task.
• Existing methods require further refinement for accuracy and robustness.

Purpose of the Study:

• To develop and evaluate a novel Perceptron ensemble of graph-based positive-unlabeled learning (PEGPUL) method for disease gene identification.
• To integrate diverse biological attributes including gene ontologies, protein domains, and protein-protein interaction networks.
• To enhance the accuracy and robustness of computational disease gene identification.

Main Methods:

• Employed a co-training schema to extract reliable positive and negative gene sets.
• Constructed a gene similarity graph using metric learning and a multi-rank-walk method.
• Developed a Perceptron ensemble integrating multilevel support vector machine, k-nearest neighbor, and decision tree classifiers.

Main Results:

• PEGPUL demonstrated reasonable performance in identifying disease genes across six disease classes.
• The method effectively incorporated statistical properties of gene data and evaluated biological features.
• The multilevel schema contributed to the noise robustness of the PEGPUL method.

Conclusions:

• PEGPUL offers a robust and effective approach for disease gene identification by leveraging graph-based positive-unlabeled learning and diverse biological data.
• The study highlights the importance of integrating multiple biological attributes and advanced machine learning techniques.
• The developed method shows potential for advancing biomedical and bioinformatics research in disease gene discovery.