Knowledge-fused differential dependency network models for detecting significant rewiring in biological networks
BMC Systems Biology
|July 25, 2014
Summary
This study introduces a new method for modeling biological networks by integrating prior knowledge with data. The approach effectively identifies significant biological network rewiring and provides new mechanistic insights.
Area of Science:
- Systems biology
- Bioinformatics
- Network inference
Background:
- Biological networks are dynamic and context-specific, requiring methods to distinguish significant changes from background noise.
- Existing network inference methods struggle with large solution spaces, complex networks, imperfect prior knowledge, and lack of significance assessment.
- Integrating prior biological knowledge with data-driven approaches can enhance the robustness and relevance of network inference.
Purpose of the Study:
- To develop a robust method for inferring differential dependency networks that integrates conditional data and prior biological knowledge.
- To address challenges in biological network modeling, including small sample sizes and complex network structures.
- To systematically characterize selectively activated regulatory components and mechanisms in biological systems.
Main Methods:
- Formulated differential dependency network inference as a convex optimization problem.
- Developed an efficient learning algorithm to jointly infer conserved networks and significant rewiring across conditions.
- Utilized a novel sampling scheme to estimate error rates from prior knowledge and a strategy to leverage integrated data-knowledge.
Main Results:
- Demonstrated and validated the method's principle and performance using synthetic datasets.
- Applied the method to yeast and breast cancer microarray data, yielding biologically plausible results.
- The developed open-source R software package is freely available.
Conclusions:
- The knowledge-fused differential dependency network effectively reveals statistically significant rewiring in biological networks.
- The method robustly integrates data-driven evidence and prior knowledge, handling false positives in prior knowledge.
- Identified network rewiring events are supported by literature and offer new mechanistic insights, positioning the tool for broad bioinformatics application.
