Integrating Bayesian variable selection with Modular Response Analysis to infer biochemical network topology.

Area of Science:

• Systems Biology
• Computational Biology
• Genomics and Proteomics

Background:

• Large quantitative datasets from genetics and proteomics present challenges for reverse engineering biochemical networks.
• Inferring network topologies from biological data, particularly steady-state responses to perturbations, remains a significant problem.

Purpose of the Study:

• To develop a robust computational algorithm for inferring functional interactions and network topologies in cellular signaling and gene regulatory networks.
• To identify direct and indirect molecular interactions, even when network components are unknown.

Main Methods:

• A hybrid approach combining Modular Response Analysis (MRA), a deterministic network inference method, and Bayesian Variable Selection, a statistical model selection algorithm.
• Application to simulated perturbation data from signaling pathways and gene regulatory networks (DREAM challenge).
• Exploration of the ERBB-regulated G1/S transition pathway in drug-resistant breast cancer cells using experimental perturbation data.

Main Results:

• The proposed method demonstrates robustness against noise and scalability for large biological networks.
• Accurate inference of network topologies is achievable even with incomplete perturbation datasets.
• The algorithm successfully identified known interactions in the ERBB pathway and revealed novel interactions contributing to drug resistance in breast cancer.

Conclusions:

• The developed algorithm offers a robust, scalable, and cost-effective solution for inferring biological network topologies from perturbation data.
• This approach has potential applications in exploring novel pathways involved in diseases like cancer.