Computing Integrated Information (Φ) in Discrete Dynamical Systems with Multi-Valued Elements

Area of Science:

• * Cognitive Science
• * Theoretical Neuroscience
• * Computational Neuroscience

Background:

• * Integrated Information Theory (IIT) quantifies consciousness and cause-effect structures in physical systems using integrated information (Φ).
• * The PyPhi software package previously enabled IIT analysis for discrete dynamical systems with binary elements.
• * Analyzing multi-valued systems is crucial for a more comprehensive understanding of complex causal structures.

Purpose of the Study:

• * To extend the PyPhi software package to accommodate discrete, multi-valued elements in dynamical systems.
• * To enable the analysis and comparison of causal properties in networks with binary, ternary, quaternary, and mixed-valued nodes.
• * To evaluate the impact of binarization methods on preserving causal structure in multi-valued systems.

Main Methods:

• * Modification and extension of the PyPhi Python package to handle multi-valued elements.
• * Generation and analysis of random networks composed of various node types (binary, ternary, quaternary, mixed).
• * Application of the enhanced PyPhi tools to a non-binary p53-Mdm2 regulatory network model.

Main Results:

• * Successful extension of PyPhi to analyze multi-valued discrete dynamical systems.
• * Demonstrated ability to compare causal properties across networks with diverse node valuations.
• * Identified limitations of binarization methods in maintaining the original causal structure of multi-valued systems.

Conclusions:

• * The updated PyPhi software significantly broadens the scope of IIT applications to more complex systems.
• * Analysis of multi-valued systems provides a more nuanced understanding of causal relationships than binary approximations.
• * Careful consideration of binarization techniques is necessary when applying IIT to biological or other complex networks.