Causal network reconstruction from time series: From theoretical assumptions to practical estimation.

Area of Science:

• Causal inference and network analysis
• Time series analysis
• Computational biology and systems science

Background:

• Causal network reconstruction from time series data is crucial for understanding complex systems.
• Distinguishing direct from indirect dependencies and common drivers in multivariate time series is a key challenge.

Purpose of the Study:

• To review the foundations and practical problems of time series-based causal discovery.
• To illustrate causal network inference, including time lags, from multivariate time series.
• To stimulate further methodological developments in the field.

Main Methods:

• Recapitulation of causal assumptions and practical estimation problems.
• Illustration with examples covering unobserved variables, sampling issues, and nonlinearity.
• Comparison studies of common causal reconstruction methods highlighting effects of noise, autocorrelation, and high dimensionality.

Main Results:

• Discussion of challenges such as unobserved variables, sampling issues, nonlinearity, and measurement error.
• Highlighting the impact of dynamical noise, autocorrelation, and high dimensionality on causal reconstruction methods.
• Suggestion of method performance evaluation approaches and criteria.

Conclusions:

• Time series causal discovery involves complex assumptions and practical challenges.
• Understanding these challenges is vital for accurate causal network inference.
• Further methodological development is needed to improve causal discovery from time series data.