Computational tools for neuroscience at different scales

Sicurella, Emanuele 2023. Computational tools for neuroscience at different scales. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (11MB) | Preview
	PDF (Cardiff University Electronic Thesis and Dissertation Form) - Supplemental Material Restricted to Repository staff only Download (384kB)

Abstract

Nowadays, computational tools are essential and ubiquitous in basically every research field. In particular, they enormously helped Psychology and Neuroscience by providing computational models of cognition, description of neurobiological phenomena, data analysis tools and many more. In this thesis, I will propose a classification method based on the scale of the problem under investigation, providing the results of applying three different computational tools for neuroscience at different scales. First, I perform graph theory-based analysis of whole human structural (DWI) and functional (MEG and fMRI) connectomes using a new approach based on small, induced, connected subgraphs called graphlets. I will show that graphlet provides an elegant and effective way to represent and characterise topological information of brain networks without the need for numerous classical graph-theory measures. Second, I apply deep learning for parameter recovery of a perceptual decisionmaking model simulating the evidence accumulation of the LIP area of the brain. Deep learning offers a valuable tool for parameter recovery of more complex biologically plausible models. However, I also stress that successful parameter recovery depends not only on the choice of the tool but also on the careful design of the experiment to avoid the recovery of parameters that have a similar effect on the output, thus making parameter recovery difficult. Finally, I perform neuronal decoding of spiking neuron activity during a 2D reaching task performed by two monkeys. Velocity decoding performances are generally better than position or acceleration decoding. I also study the effect of PCA on dimensionality reduction of neural data and consequent neuronal decoding, showing that general performances on reduced data are lower except for position decoding from PMd cortex activity of many neurons. Following previous research results, I hypothesise this is caused by the encoding of different processes in the PMd cortex not related to the simple forward motor output. Overall, this work explores the use of powerful computational tools to solve problems in neuroscience at different scales.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Psychology
Date of First Compliant Deposit:	31 May 2024
Last Modified:	31 May 2024 10:33
URI:	https://orca.cardiff.ac.uk/id/eprint/169348

Actions (repository staff only)

Edit Item