Uncertainty Quantification of Network Inference with Data Sufficiency

Area of Science:

• Complex systems science
• Network science
• Statistical inference

Background:

• Network inference reconstructs system connectivity from data, vital for understanding physical, biological, and chemical systems.
• Current data-driven methods often overlook the critical question of data sufficiency for accurate network topology.
• Accurate network reconstruction requires sufficient data variability to reliably infer underlying structures.

Purpose of the Study:

• To develop a statistical method for assessing data sufficiency in network inference.
• To quantify the uncertainty in inferred network connectivity based on data variability.
• To ensure that inferred network topologies accurately reflect the true underlying network structure.

Main Methods:

• Utilizing parametric confidence intervals to define bounds for true connection strengths.
• Developing a technique to assess data variability for network inference accuracy.
• Leveraging uncertainty quantification for inferred connectivity.

Main Results:

• The proposed statistical method effectively determines data sufficiency for network inference.
• Validation on Kuramoto and Stuart-Landau oscillator networks demonstrates method accuracy.
• Successful application to experimental electrochemical oscillator network data confirms predictive power.

Conclusions:

• The developed data sufficiency technique is essential for reliable network inference.
• This method enhances the trustworthiness of network topology reconstructions.
• It provides a quantitative measure to ensure sufficient data for accurate system analysis.