Detection of statistically significant network changes in complex biological networks

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Biological networks model molecular interactions and are key to discovering driver genes in cancer.
• Gene mutations during cancer progression can cause localized re-wiring of gene expression networks.
• Detecting statistically significant changes in interaction patterns can reveal novel disease signatures.

Purpose of the Study:

• To improve upon existing methods for detecting sub-network differences in pairwise labeled weighted networks.
• To develop a more efficient and accurate procedure for analyzing topological differences and their statistical significance.

Main Methods:

• Utilizing the Generalized Hamming Distance to evaluate topological differences between biological networks.
• Employing improved model selection criteria for generating p-values and estimating statistical significance.
• Developing a parallelizable algorithm for enhanced computational efficiency.

Main Results:

• The proposed method is 10-15x faster and achieves 5-10% higher AUC, Precision/Recall, and Kappa values on dense random geometric networks compared to state-of-the-art methods.
• Applied to IDH-mutant versus IDH-wild-type glioma, the method identified known master regulators and novel candidate genes.
• Demonstrated effective and efficient detection of statistically significant network re-wirings.

Conclusions:

• The network differencing procedure reliably detects significant network re-wirings across different conditions.
• The method accurately identifies key regulators in IDH-mutant and IDH-wild-type glioma subtypes.
• The approach highlights novel candidate genes missed by traditional single network analyses.