Reconstruction of human protein-coding gene functional association network based on machine learning.

Area of Science:

• Systems Biology
• Bioinformatics
• Network Science

Background:

• Molecular interaction networks are crucial for understanding organismal dynamics.
• Existing molecular interaction databases are incomplete and contain false positives from high-throughput screening.
• Accurate and comprehensive networks are needed for biological discovery.

Purpose of the Study:

• To develop a framework for reconstructing a high-coverage and high-quality human protein-coding gene functional association network.
• To integrate data from 31 sources to create a robust protein-coding gene network.
• To assess the utility of the reconstructed network in identifying cancer drivers.

Main Methods:

• Constructed 369 features for each interaction, encompassing properties of interactions and involved genes.
• Utilized a semi-supervised strategy to generate training and validation sets, using pathway interactions as positive instances.
• Applied a Random Forest classification method and a Binomial distribution threshold to score and select high-confidence interactions.

Main Results:

• Reconstructed the Human protein-coding Gene Functional Association Network (HuGFAN) with 20,383 genes and 1,185,429 high-confidence interactions.
• HuGFAN exhibits superior functional and pathway relatedness compared to existing networks.
• Demonstrated outstanding performance in identifying cancer drivers using established network-based methods (DriverNet, HotNet2).

Conclusions:

• The proposed framework effectively integrates diverse data sources to build a high-quality gene association network.
• HuGFAN provides a valuable resource for systems biology research and disease mechanism investigation.
• The network's performance in cancer driver identification highlights its potential for biomedical applications.