Inference of complex biological networks: distinguishability issues and optimization-based solutions

Area of Science:

• Systems Biology
• Computational Biology
• Network Science

Background:

• Biological network inference from high-throughput data is a key challenge in systems biology.
• Existing methods often fail due to data limitations and algorithmic deficiencies.
• A significant, often overlooked, issue is the lack of uniqueness in inferred network models.

Purpose of the Study:

• To address the lack of uniqueness in biological network inference.
• To develop methods for distinguishability analysis of chemical reaction network (CRN) models.
• To investigate the impact of prior assumptions and data richness on network identifiability.

Main Methods:

• Developed novel methods based on linear programming (LP) for efficient CRN analysis.
• Applied LP to analyze CRNs with hundreds of complexes and reactions.
• Integrated new tools with existing methods to assess network topology from literature data.

Main Results:

• Demonstrated that unique network topologies are often indeterminable, even with complete data.
• Showcased that certain mechanisms can be missed or incorrectly identified despite model consistency.
• Found that sparsity-promoting methods alone are insufficient without additional prior information.

Conclusions:

• Biological network inference is inherently difficult, even with ideal experimental data.
• Incomplete, noisy, or dynamically limited measurements exacerbate inference challenges.
• Structural uniqueness and identifiability can be achieved by incorporating extra constraints, verifiable through computational methods.