Spontaneous activity emerging from an inferred network model captures complex spatio-temporal dynamics of spike data

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Network Dynamics

Background:

• Inference methods are crucial for building models from data, but their application to neuroscience network dynamics is underexplored.
• Existing models often fail to capture the complex, non-stationary, and non-linear activity observed in biological neural networks.

Purpose of the Study:

• To develop and validate a novel inference method for capturing the structural and dynamic properties of neural networks.
• To investigate the dynamics of inferred models at the network level, addressing a gap in current neuroscience research.

Main Methods:

• A modified generalized linear model (GLM) was developed, incorporating a saturating transfer function and a super-Poisson spike generation mechanism.
• The model's parameters were learned from in-vitro spike data of a biological neural network.

Main Results:

• The modified GLM successfully captured prominent features of non-stationary and non-linear dynamics in spontaneous neural activity.
• The model outperformed the reference GLM in reflecting fine-grained spatio-temporal dynamical features.
• The saturating transfer function enhanced robustness to noise, while the super-Poisson mechanism accounted for network undersampling and activity fluctuations.

Conclusions:

• The proposed modified GLM provides a more accurate and robust approach for inferring neural network dynamics.
• Key model components, including the saturating transfer function and super-Poisson mechanism, are critical for capturing biological neural activity.
• This work advances computational neuroscience by offering a powerful tool for understanding complex neural systems.