Ensemble positive unlabeled learning for disease gene identification.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Numerous genes are confirmed as causative for human diseases.
• Machine learning can identify novel disease-associated genes from existing data.
• Positive unlabeled learning (PU learning) is effective for this task, using confirmed disease genes (P) and candidate genes (U).

Purpose of the Study:

• To develop an effective PU learning framework for disease gene identification.
• To integrate multiple biological data sources and machine learning classifiers.
• To enhance prediction accuracy and robustness by minimizing bias from single data sources or methods.

Main Methods:

• Proposed an ensemble-based PU learning framework (EPU).
• Integrated data from multiple biological sources for training PU learning classifiers.
• Employed an ensemble of PU learning classifiers to combine predictions.

Main Results:

• EPU demonstrated significantly improved performance compared to state-of-the-art methods.
• Evaluations across six disease groups confirmed EPU's accuracy and robustness.
• The integrated approach minimized bias from individual data sources and algorithms.

Conclusions:

• EPU provides an effective framework for disease gene identification.
• Integrating diverse data and ensemble methods enhances prediction accuracy.
• The framework can incorporate future biological and computational resources for improved predictions.