Network inference from short, noisy, low time-resolution, partial measurements: Application to C. elegans neuronal...

Area of Science:

• Computational neuroscience
• Systems biology
• Network science

Background:

• Network link inference from time series data is vital for understanding complex systems.
• Applications include estimating neuronal synaptic connectivity from calcium imaging data.
• Existing methods face challenges due to data limitations like noise and limited duration.

Purpose of the Study:

• To systematically assess the impact of data acquisition limitations on network link inference techniques.
• To evaluate how limited duration, sampling rate, noise, and partial measurements affect link inference scores.
• To compare the performance of Granger causality, transfer entropy, and a machine learning method under these limitations.

Main Methods:

• Utilized synthetic data from coupled chaotic systems.
• Employed experimental data from *Caenorhabditis elegans* neural activity.
• Applied Granger causality, transfer entropy, and a machine learning-based method for network inference.
• Assessed the utility of surrogate data for determining statistical confidence.

Main Results:

• Data limitations significantly influence the character of link inference scores.
• The performance of different inference techniques varies under specific data constraints.
• Surrogate data analysis provides a method for assessing statistical confidence in inferred links.

Conclusions:

• Experimental data limitations critically affect the reliability of network link inference.
• Choosing appropriate inference methods and validation techniques is essential for accurate network reconstruction.
• This work provides insights into robust network inference strategies for real-world, noisy datasets.