HuGFAN

HuGFAN reconstructs a human protein-coding gene functional association network by integrating molecular interactions from 31 diverse data sources to improve coverage and identify high-confidence gene–gene associations for functional and pathway analyses.

Key Features:

• Comprehensive Data Integration: Integrates molecular interactions from 31 diverse data sources to enhance coverage and quality of gene–gene association data.
• Feature Construction: Constructs 369 features per interaction capturing interaction-specific properties and characteristics of the involved genes.
• Machine Learning Approach: Uses a random forest classification method with multiple iterations to assign scores to interactions and identify high-confidence associations.
• Threshold Setting via Binomial Distribution: Applies a Binomial distribution–based threshold to filter interactions, yielding a network of 20,383 genes and 1,185,429 high-confidence interactions.
• Network Comparison and Validation: Compares HuGFAN network properties against networks derived from various data sources to assess functional and pathway-related characteristics.
• Application in Cancer Research: Has been applied to identify cancer driver genes using DriverNet and HotNet2, demonstrating superior performance relative to other networks.

Scientific Applications:

• Functional and pathway analysis: Supports analysis of functional associations and pathway relationships among human protein-coding genes.
• Disease mechanism and cancer driver identification: Facilitates identification of cancer driver genes and exploration of disease mechanisms, demonstrated with DriverNet and HotNet2.
• Therapeutic target prioritization: Aids prioritization of candidate therapeutic targets by highlighting high-confidence functional associations.

Methodology:

Integrates 31 data sources, constructs 369 features per interaction, employs a semi-supervised strategy to curate positive pathway interactions and generate potential negative instances, trains a random forest classifier with multiple iterations to score interactions, and applies a Binomial distribution–based threshold to define high-confidence associations.