Biologically realistic computational models of brain networks observed in human Neuroscience data (Williams et al. (2023a), NeuroImage)

Brain networks are key to executing all cognitive functions such as speaking, listening, remembering, planning, and decision-making. However, we do not know precisely how these networks are generated.  In this study, we investigated the role of conduction delays between brain regions in generating large-scale brain networks. To do this, we applied Approximate Bayesian Computation (ABC) methods to compare biologically realistic computational models with different delay assumptions, using human Magnetoencephalographic (MEG) data (see below figure). We found that "distance-dependent" conduction delays likely underlie empirically observed large-scale brain networks, thus providing neuroscientific insight into how these networks might be generated. The study's technical relevance lies in using ABC methods to inform biologically realistic models with empirical data, thus achieving a closer integration between modelling and experimentation than has been previously achieved in Cognitive Neuroscience.

I completed this study, in collaboration with Prof. Samuel Kaski and Prof. Matias Palva.

Please also see a news item summarising our study for a general audience.

Systems-level neurophysiology work on identifying modules in brain connectomes (Williams et al. (2023), NeuroImage)

Brain oscillations from different brain regions exhibit synchronization between their phases, and this synchronization is considered to mediate communication between regions. Modules in connectomes of phase synchronization, i.e., sets of strongly synchronized regions, might represent functional systems responsible for tasks in specific cognitive domains, but methodological limitations have left these modules unidentified. In this study, we combined a large cohort of intra-cranial Electroencephalographic (EEG) recordings with novel computational methods, to identify modules in brain connectomes of phase synchronization. We found the identified modules to comprise spatially proximal brain regions and further, we found evidence in support of each module representing functional systems specialised for tasks in distinct cognitive domains such as memory, executive function and language (see below figure). The study's neurophysiological significance is in potentially having identified an entirely new class of functional systems, compared to those previously identified by other neuroimaging modalities such as functional Magnetic Resonance Imaging.

I completed this study, in collaboration with Prof. Matias Palva.

Computational methods to analyse time-varying brain networks in human Neuroscience data (Williams et al. (2018), Front. Comput. Neurosci.)

A plethora of computational methods have been developed to characterise brain networks from human electrophysiological data, such as Electroencephalography(EEG) and Magnetoencephalography (MEG).  However, these methods assume brain networks to remain constant in time, while brain networks are known to change rapidly as a cognitive task is being performed. In this study, we used a novel combination of a sparse Multi-Variate Auto-Regressive model and a Markov Model, to characterise time-varying brain networks from EEG task data. We showed that the temporal pattern of variation in brain networks captured by the method, was statistically different for two cognitive tasks (see below figure).  The study's significance lies in demonstrating that our Markov model-based method is sensitive to task-specific temporal pattern of variation in brain networks, thus paving the way for its use by Cognitive Neuroscientists to track rapid changes in interaction patterns as a task is being performed.

I completed this study, in collaboration with Prof. Slawomir Nasuto and Dr. Ian Daly.

I released an open-source MATLAB toolbox implementing this method, for use by the EEG and MEG community.