Reconstruction of Complex Directional Networks with Group Lasso Nonlinear Conditional Granger Causality.

Area of Science:

• Complex Systems Science
• Network Science
• Data Science

Background:

• Reconstructing complex networks is vital across engineering and science.
• Existing methods often rely on pre-defined models or specific function types.
• Limited observations pose a significant challenge in network reconstruction.

Purpose of the Study:

• To develop a general framework for nonlinear causal network reconstruction from time-series data.
• To address nonlinearity and directionality in complex networked systems.
• To provide a robust method for network inference with limited observations.

Main Methods:

• A data-fusion strategy to obtain multi-source datasets.
• Group lasso nonlinear conditional Granger causality (NL-CG) for network reconstruction.
• Utilizing radial basis functions to approximate nonlinear interactions.
• Integrating sparsity for grouped variable selection.

Main Results:

• The proposed method was validated on simulated datasets (nonlinear VAR and dynamic models).
• Performance was further assessed using benchmark datasets from the DREAM3 Challenge.
• The approach demonstrated superior performance, indicated by a higher area under the precision-recall curve.
• The impact of data size and noise intensity on performance was analyzed.

Conclusions:

• The developed framework offers an effective approach for nonlinear causal network reconstruction.
• The group lasso NL-CG method shows promise in handling complex networked systems.
• The findings highlight the method's robustness and accuracy, even with limited and noisy data.