1. Interpretation of multi-parametric statistical models of large scale neuronal network activity
- Author
-
Cronin, Joseph Thomas, Hennig, Matthias, and Rochefort, Nathalie
- Subjects
612.8 ,multi-parametric statistical models ,neuronal networks ,Ising model ,Fisher information matrix ,CNQX ,mouse model ,computational modelling - Abstract
Multi-parametric statistical models have been shown to be successful in characterising the statistical properties of neuronal network activity for large scale populations, in a variety of in vitro and in vivo experimental settings. In this thesis, preliminary work explored one of the primary modelling techniques used to reproduce firing pattern distributions in binary networks, the Ising model, and outlined a novel manipulation to this analysis in the form of a distribution rescaling and renormalisation. The aim of this extension was to extract additional information about the correlation structure of typically sparse neural activity, which can often be lost due to the dominant effect that firing rate has on determining multi-neuronal firing patterns. Results here successfully uncovered a selection of simulated scenarios in which the rescaled Ising model could quantify network changes that were missed by the original model. Additionally, the interpretation of these models was also investigated through the eigendecomposition of the model’s Fisher information matrix, which has been shown to offer further insights into network mechanisms through a characterisation of model parameter sensitivity structure. From here, larger scale analysis was then employed by utilising the population tracking model and the diagonal Fisher approximation, to investigate a network-wide remodelling effect in a set of multi-electrode array recordings of dissociated hippocampal cultures. Resulting entropy and divergence measures obtained from the models indicated a permanent remodelling effect upon the addition of a known chemical inhibitor, CNQX. Further to this, the use of the diagonal Fisher information approximation created an efficient framework for analysing large scale populations while maintaining sufficient information on the structure of the data, demonstrating some highly desireable qualities for use in future work. Finally, research was undertaken to characterise neural population coding impairments in a mouse model of autism spectrum disorder (ASD), by using two-photon CA2+ imaging of the primary visual cortex in awake behaving SynGAP+/− mice, in response to a number of visual stimulation protocols and an induced plasticity paradigm (monocular deprivation). Analysis techniques here involved primary response quantifications such as orientation selectivity and ocular dominance, as well as the modelling techniques presented in the earlier chapters of this thesis. While there are known physiological and behavioural differences between the two genotypes, results indicated no significant differences in the average activity, selectivity and coding capabilities of the two animal groups. However, characteristic response curves were found to exhibit different features in the SynGAP+/− animals, with neurons typically responding earlier with respect to stimulus onset and displaying a lack of habituation to repeated stimulation. In summary, this study has expanded and advanced the range of analytical techniques available for the computational modelling of neural spiking activity and also contributed to refinements in sensitivity analysis techniques used for the interpretation of large scale activity of neuronal networks, as well as presenting novel results from network remodelling effects in cultured networks and visual representations in a mouse model of ASD.
- Published
- 2020
- Full Text
- View/download PDF