This title appears in the Scientific Report :
2018
Please use the identifier:
http://hdl.handle.net/2128/17943 in citations.
Comparison of statistical methods for spatio-temporal patterns detection in multivariate point processes: an application to neuroscience
Comparison of statistical methods for spatio-temporal patterns detection in multivariate point processes: an application to neuroscience
The understanding of how information is processed and how neurons coordinate their activity is still a challenge in neuroscience. The brain network is highly interconnected, and the firing activity of single neurons is influenced by the behavior of their peers: Hebb’s hypothesis states that the evol...
Saved in:
Personal Name(s): | Stella, Alessandra (Corresponding author) |
---|---|
Contributing Institute: |
Jara-Institut Brain structure-function relationships; INM-10 Computational and Systems Neuroscience; IAS-6 Computational and Systems Neuroscience; INM-6 |
Imprint: |
2017
|
Physical Description: |
105 |
Dissertation Note: |
Masterarbeit, University of Turin, 2017 |
Document Type: |
Master Thesis |
Research Program: |
Connectivity and Activity |
Link: |
OpenAccess OpenAccess |
Publikationsportal JuSER |
The understanding of how information is processed and how neurons coordinate their activity is still a challenge in neuroscience. The brain network is highly interconnected, and the firing activity of single neurons is influenced by the behavior of their peers: Hebb’s hypothesis states that the evolution of neuronal activity is driven by the activation sequence of groups of neurons called assemblies. Moreover, recent technological developments in neuroscience made electrophysiological recordings of large neuronal population possible. Correlated activity is thought to be the expressing signature of the activation of a cell assembly, thus many methods have been designed to extract correlation and dependency structures from this high-dimensional data. Some of these methods look specifically to significant repetitions of firing patterns, although they are based on different hypothesis and have different performances.This present work addresses the question by studying two statistical methods in order to compare and contrast the hypothesis behind them, and both their statistical and computational performances. The purpose is to identify, in our specific model, in which settings one method is more performing than the other in order to obtain a possible workflow for the analysis of suitable electrophysiological data.We present the mathematical objects generally used to represent electrophysiological data, defining respectively a continuous and discrete time model.We show the biological framework and the neuroscientific questions behind this work, in particular the relevance of spatio-temporal patterns in spike train data in neuronal encoding.We present the statistical model used to generate artificial data, and the two statistical methods we will be comparing.Finally, we give an overview of the differences between the two statistical methods, both in terms of statistical hypothesis and parameter input. We then compare how the two methods are able to capture pairwise correlations in data from an analytical point of view. Then, we describe in detail the ground truth data we generate, and we present the results that the two methods obtain from the analysis in different settings, evaluating their performance in terms of False Positives and False Negatives. Finally, we assess their computational performance in terms of runtime in different contexts (i.e. datasets with and without Spatio-Temporal Patterns). |