Scalable Non-Linear Graph Fusion for Prioritizing Cancer-Causing Genes

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression and protein-protein interaction (PPI) networks are crucial for understanding disease-associated genes.
• Existing methods often analyze these data sources separately, limiting comprehensive insights into complex diseases like cancer.

Purpose of the Study:

• To develop a novel gene prioritization algorithm for identifying and ranking cancer-causing genes.
• To integrate complementary information from gene expression and PPI network data for enhanced accuracy.
• To introduce scalable methods for learning functional similarity and evaluating network quality.

Main Methods:

• A new quantitative index was developed to assess gene importance, considering differential expression and PPI network connectivity.
• A scalable non-linear graph fusion technique (ScaNGraF) was proposed to learn disease-specific functional similarity networks from co-expression and PPI data.
• A new measure, DiCoIN, was introduced to evaluate the quality of the learned affinity networks.

Main Results:

• The proposed ScaNGraF technique efficiently combines information from multiple data sources with reduced computational cost.
• The gene prioritization algorithm effectively identifies and ranks cancer-associated genes.
• Extensive comparisons on cancer datasets demonstrate superior performance compared to existing methods.

Conclusions:

• The integrated approach using gene expression and PPI data significantly enhances the identification of cancer-causing genes.
• ScaNGraF offers an efficient and scalable solution for learning functional gene similarities.
• The developed methods provide valuable tools for cancer research and biomarker discovery.