Inferring brain-wide interactions using data-constrained recurrent neural network models

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neuroimaging Analysis

Background:

• Understanding brain function requires analyzing coordinated activity across diverse brain regions.
• Large-scale neural recordings generate complex data, necessitating advanced computational models.
• Extracting meaningful features of inter-region communication is crucial for interpreting brain dynamics.

Purpose of the Study:

• To introduce current-based decomposition (CURBD), a novel method for inferring brain-wide interactions.
• To develop data-constrained recurrent neural network models for analyzing neural data.
• To reveal directional communication currents between multiple brain regions.

Main Methods:

• Developed CURBD, a method using data-constrained recurrent neural networks.
• Trained models to produce dynamics consistent with experimental neural data.
• Leveraged inferred functional interactions to identify directional inter-region currents.

Main Results:

• CURBD accurately isolated inter-region currents in simulated networks with known ground truth.
• Applied CURBD to multi-region neural recordings from zebrafish, mice, macaques, and humans.
• Demonstrated CURBD's applicability in untangling brain-wide interactions across species.

Conclusions:

• CURBD effectively infers directional brain-wide communication currents.
• The method provides a scalable approach for analyzing large-scale neural datasets.
• CURBD advances the understanding of inter-area communication principles underlying behavior across species.