Evaluation of dynamic causal modelling and Bayesian model selection using simulations of networks of spiking neurons

Matthew G. Thomas

2023-06-28

Abstract:Inferring the mechanisms underlying physiological and pathological processes in the brain from recorded electrical activity is challenging. Bayesian model selection and dynamic causal modelling aim to identify likely biophysical models to explain data and to fit the model parameters. Here, we use data generated by simulations to investigate the effectiveness of Bayesian model selection and dynamic causal modelling when applied at steady state in the frequency domain to identify and fit Jansen-Rit models. We first investigate the impact of the necessary assumption of linearity on the dynamics of the Jansen-Rit model. We then apply dynamic causal modelling and Bayesian model selection to data generated from simulations of linear neural mass models, non-linear neural mass models, and networks of discrete spiking neurons. Action potentials are a characteristic feature of neuronal dynamics but have not previously been explicitly included in simulations used to test Bayesian model selection or dynamic causal modelling. We find that the assumption of linearity abolishes the qualitative transitions seen as a function of the connectivity parameter in the original Jansen-Rit model. As with previous work, we find that the recovery procedures are effective when applied to data from linear Jansen-Rit neural mass models, however, when applying them to non-linear neural mass models and networks of discrete spiking neurons we find that their effectiveness is significantly reduced, suggesting caution is required when applying these methods.

Neurons and Cognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to evaluate the effectiveness and limitations of Dynamic Causal Modelling (DCM) and Bayesian Model Selection (BMS) in analyzing neural network simulation data. Specifically, researchers test these methods by using data generated from linear and nonlinear neural mass models and discrete neuronal networks with spike dynamics. The focus of the study is as follows:

1. **Impact of linear assumptions**: Explore the impact on the dynamics of the Jansen - Rit model under linear assumptions, especially how such assumptions change the dynamic characteristics of the model, for example, eliminating the qualitative transitions caused by changes in connection parameters.

2. **Effectiveness of model recovery**: Evaluate the ability of these methods to recover model parameters when applying DCM and BMS to data generated by the linear Jansen - Rit model, nonlinear neural mass models, and discrete spiking neuronal networks. The study found that the effectiveness of these methods is significantly reduced when dealing with nonlinear models and spiking neuronal network data.

3. **Impact of spike activity**: The study introduced spike activity (action potentials), which is a key feature of neural network dynamics and was not explicitly included in previous simulation tests. By comparing the results under different models, researchers hope to understand how spike activity affects the performance of DCM and BMS.

Through these evaluations, the paper aims to provide more accurate tools and methods for neuroscience research, especially in the ability to infer brain network characteristics from electrical signals recorded by electroencephalogram (EEG) and the like. This not only helps to understand the normal functions of the brain but may also provide new insights into the diagnosis and treatment of neurological diseases.