A Comparison of Partial Information Decompositions Using Data from Real and Simulated Layer 5b Pyramidal Cells

Area of Science:

• Computational Neuroscience
• Information Theory

Background:

• Partial information decomposition (PID) is a framework for analyzing how information is shared among multiple variables.
• Understanding information flow in neural systems is crucial for deciphering complex computations.

Purpose of the Study:

• To compare five different partial information decomposition methods.
• To evaluate these methods using neural data from layer 5b pyramidal cells in two distinct experimental/modeling studies.

Main Methods:

• Application of five distinct partial information decomposition techniques.
• Analysis of electrophysiological data from layer 5b pyramidal cells (Study 1: somatic AP amplification, dendritic inhibition; Study 2: compartmental model of synaptic inputs).

Main Results:

• Two decomposition methods yielded significantly higher estimates of synergy and shared information compared to others.
• Differences in unique information asymmetry were observed, particularly under dendritic inhibition.
• Three methods supported the conclusion that dendritic inhibition reduces apical amplification.
• Analysis of subsets revealed a bifurcation in unique information asymmetry, suggesting shifts in information drive (apical to basal).

Conclusions:

• The choice of partial information decomposition method can influence the interpretation of neural information processing, especially concerning synergistic and shared information.
• Dendritic inhibition plays a significant role in modulating information processing within pyramidal cells.
• Cooperative context-sensitivity properties are supported by all tested decomposition methods to varying degrees.