This article is in the news archive.
Congratulations Dr Petra Vértes on award of MRC Fellowship
Dr Petra Vértes from the Department of Psychiatry has recently been awarded a Fellowship by the Medical Research Council to extend her current work on modelling the development of complex brain networks.
Dr Vértes (pictured right) completed both her undergraduate degree and PhD in Physics at the University of Cambridge. During her PhD she examined neural networks and how the architecture or shape of the neural networks determines how good they are at pattern recognition. After her PhD she joined Professor Ed Bullmore’s lab in the Brain Mapping Unit and her interest shifted to building large-scale brain networks based on real neuroimaging data.
She is interested in looking at structural connections between brain regions using magnetic resonance imaging (MRI) recordings, and functional connectivity using data from functional MRI (fMRI) and magnetoencephalography (MEG) recordings. These data are collected as part of the NSPN and Cam-CAN projects.
The Department of Psychiatry’s news team interviewed Dr Vértes to congratulate her and asked her why modelling the development of brain networks is important.
“There are several potential reasons. What people have mostly been doing so far is to build these networks from neuroimaging data, look at their characteristics and see the differences between groups of people with and without a psychiatric condition such as Alzheimer’s disease or schizophrenia, and so the importance of that might be diagnostic, or possibly prognostic. What I am interested in now is to explore and describe how these networks actually evolve over time. I then aim to build models that will explain and predict how brain networks change over time during periods of large-scale brain network reorganization such as adolescence or ageing. And if the models are simple enough, for example if they only have one or two parameters, that might actually help you understand what is driving those changes, especially if the parameters you are using are biologically plausible.”
During the collection of data for network analysis, participants are not usually requested to do any task, but asked not to think of anything in particular. The idea is that the scanner then measures how neurons in the brain fire during everyday thinking and a network can be constructed to describe which regions of the brain usually communicate with each other and how strong these connections are. These data can then be used to explore if there are differences between groups of individuals with and without a psychiatric condition or, for example, if connectivity is linked to measures like intelligence (IQ).
“I got started with this about year ago, when colleagues and I built a model that generates networks which look very much like normal healthy brain networks. It is a surprisingly simple computerised model; you basically take the known positions of each of about 200 brain regions and the model predicts which links are likely to form. You then build, or ‘grow’, a network based on these predictions, which you can compare with real networks that you have extracted from the neuroimaging data. What we found is that they actually correspond or look very similar in terms of lots of mathematical characteristics, in spite of the model having only two parameters. In addition, these parameters are biologically plausible. For example, in our case, the parameters encoded a trade-off between two factors: one was a distance penalty, which means that it is more costly to form long-range connections, and the other factor had to do with two regions being more likely to connect if they share many neighbours. This is reminiscent of the Hebbian rule that if regions share a lot of common input then they are likely to be active together and that would cause them to connect.”
“Perhaps the most exciting aspect of our model is that if you tune these two parameters you can go from growing the healthy-looking network to the kind of network that you observe in individuals with schizophrenia. The reason this kind of modelling is interesting to me is that you can start thinking, ‘what is it about these parameters, if they really have a biological interpretation, that can cause something like schizophrenia?’”
For an example of what a brain network looks like click here.
“Now this model was really simple and it just generated a network that ended up looking like a brain network but it did not necessarily resemble the brain during every step of the generation. So it was a generative model, but it was not a growth model, it did not reproduce every step of how a brain network might develop. But now that we have data on how large-scale brain networks are changing over time, for instance during adolescence, that is what I would like to do. I would like to build models where every step of the way I can ask ‘how is the network changing now?’ By doing this we will hopefully find simple rules like these trade-offs between cost and function that can reproduce the changes during adolescence. Mapping these changes from childhood to early adulthood is particularly interesting and important since many psychiatric conditions such as schizophrenia are thought to be developmental in origin.”
“Network approaches to neuroimaging are moving away from focussing only on diagnosis and are moving towards building models which will hopefully allow us to understand the underlying mechanisms.”
If you are interested in reading more about network analysis please go to the following links to read an article by Dr Vértes on simple models of human brain functional networks and recent reviews by Professor Ed Bullmore on graph theory and the economy of brain network organisation.
Adapted from Department of Psychiatry News
Posted on 15/07/2013
Go to the news index page.