Machine learning-based approaches for disease gene prediction

Area of Science:

• Biomedical research
• Computational biology
• Genomics

Background:

• Disease gene prediction is crucial in biomedical research.
• Early methods relied on annotations.
• High-throughput technologies enable network-based and machine learning approaches.

Purpose of the Study:

• To provide a roadmap of machine learning-based disease gene prediction methods.
• To compare the performance and efficiency of various prediction algorithms.
• To analyze the advantages, disadvantages, interpretability, and trust of these methods.

Main Methods:

• Review of machine learning techniques for disease gene prediction.
• Examination of binary, unary, and semi-supervised classification approaches.
• Analysis of advanced methods including ensemble learning, matrix factorization, and deep learning.
• Comparative analysis of 12 representative methods based on prediction performance and running time.

Main Results:

• Disease gene prediction has evolved from basic classification to sophisticated machine learning models.
• Different machine learning methods offer varying trade-offs in prediction accuracy, speed, and interpretability.
• Advanced techniques like deep learning show promise for improved disease gene identification.

Conclusions:

• Machine learning offers powerful tools for disease gene prediction.
• The choice of method depends on specific research needs and data availability.
• Further analysis of interpretability and trust is essential for clinical application.