Large-scale directed network inference with multivariate transfer entropy and hierarchical statistical testing.

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Network inference algorithms are crucial for analyzing complex neuroimaging datasets.
• Multivariate transfer entropy effectively captures nonlinear and lagged dependencies in time series data.
• Existing greedy algorithms face challenges with high-dimensional data, false positives, and computational demands.

Purpose of the Study:

• To present a novel network inference algorithm implemented in the IDTxl open-source software.
• To address limitations of previous methods by controlling the family-wise error rate and enabling parallelization.
• To validate the algorithm's performance on large-scale synthetic datasets with varying dynamics.

Main Methods:

• The IDTxl algorithm employs hierarchical statistical tests for efficient network inference.
• It utilizes multivariate transfer entropy to identify directed network models from time series.
• The method is designed for high-dimensional data and allows for parallel computation.

Main Results:

• The algorithm achieved high precision, recall, and specificity (>98% on average) for 10,000 time samples.
• Performance improved with longer time series, demonstrating robust accuracy.
• Validation on networks up to 100 nodes, an order of magnitude larger than previous studies, confirmed feasibility.

Conclusions:

• The IDTxl algorithm offers a scalable and accurate solution for network inference in large neuroimaging datasets.
• Hierarchical testing and parallelization enhance computational efficiency and statistical rigor.
• The method is suitable for analyzing typical electroencephalography (EEG) and magnetoencephalography (MEG) data.