C-PUGP: A cluster-based positive unlabeled learning method for disease gene prediction and prioritization

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in genomics

Background:

• Accurate disease gene detection is crucial for understanding disease mechanisms and developing treatments.
• Machine learning methods are widely used for identifying candidate disease genes, but often face challenges with insufficient negative data.
• Semi-supervised learning on positive and unlabeled data has shown promise but can be improved.

Purpose of the Study:

• To propose a novel Positive Unlabeled (PU) learning technique for more reliable disease gene identification.
• To address the challenge of limited negative data in candidate gene detection.
• To enhance the accuracy and performance of disease gene prioritization.

Main Methods:

• A new PU learning technique combining clustering and One-Class classification is proposed.
• A three-step process is used to generate a Reliable Negative (RN) set: clustering positive data, learning One-Class classifiers, and selecting the intersection of negative data.
• A Support Vector Machine (SVM) binary classifier is employed for candidate disease gene identification and ranking.

Main Results:

• The proposed method achieved high performance metrics: 92.8% precision, 93.6% recall, and 93.1% F-measure.
• Performance, particularly in F-measure, showed an 11.7% improvement compared to existing methods.
• A notable 6% increase in prioritization results was observed.

Conclusions:

• The novel PU learning technique effectively addresses the challenge of limited negative data in disease gene detection.
• The proposed method significantly outperforms existing approaches in identifying and ranking candidate disease genes.
• This approach offers a more accurate and reliable tool for genomic research in disease understanding.