Computational inference of neural information flow networks

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Bioinformatics

Background:

• Understanding neural information flow is crucial for deciphering perception, learning, and behavior in animals.
• Existing linear computational methods are insufficient for capturing the known nonlinear nature of neural interactions.

Purpose of the Study:

• To adapt and validate a dynamic Bayesian network (DBN) inference algorithm for inferring nonlinear neural information flow from electrophysiology data.
• To compare the performance of the DBN algorithm against linear methods in mapping neural pathways.

Main Methods:

• Applied a DBN inference algorithm, originally developed for gene expression data, to electrophysiology data from the songbird auditory pathway.
• Collected electrophysiology data using microelectrode arrays.
• Compared inferred neural networks with known anatomical pathways and system timing.

Main Results:

• The DBN algorithm successfully inferred nonlinear neural information flow networks, restricted to known anatomical paths.
• Inferred networks revealed the significance of reciprocal feedback in auditory processing.
• Demonstrated greater information flow to higher-order auditory areas with natural sounds compared to synthetic sounds.

Conclusions:

• The DBN algorithm provides a biologically validated method for inferring nonlinear neural information flow.
• Linear methods applied to the same data produced inaccurate network inferences, including non-neural connections.
• This study highlights the importance of nonlinear network inference for accurately understanding brain function.