Function-Genes and Disease-Genes Prediction Based on Network Embedding and One-Class Classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Predicting disease and function genes is crucial for understanding biological systems.
• Existing methods struggle with a lack of negative examples, relying solely on known positive gene sets.

Purpose of the Study:

• To develop a robust machine learning algorithm for unified disease and function gene prediction.
• To address the challenge of limited negative examples in gene function and disease association studies.

Main Methods:

• Constructed a protein-protein interaction (PPI) network using experimentally validated genes.
• Employed Variational Graph Auto-Encoders (VGAE) for network embedding to represent genes.
• Utilized an improved deep learning one-class classifier (Fast Minimum Covariance Determinant, Fast-MCD) for gene prediction.

Main Results:

• VGAEMCD successfully predicts function-genes and disease-genes in a unified manner.
• The algorithm achieves high performance across Recall, Precision, F-measure, Specificity, and Accuracy.
• VGAEMCD demonstrates superior performance compared to classical one-class classification and state-of-the-art prediction algorithms.

Conclusions:

• VGAEMCD effectively learns the distribution characteristics of positive gene examples.
• The proposed method accurately identifies novel function and disease-associated genes.
• VGAEMCD offers a powerful tool for gene function and disease prediction without requiring negative data or expression profiles.