(MRC DTP) Data-driven null models for big data networks

When seeking to understand big, complex datasets, medical researchers are increasingly turning to the tools of network science. In representing their data as a network, the nodes are the objects in the data and the links between nodes are the presence and strength of relationships between those objects. We can use networks to analysis brain activity from fMRI and EEG recordings in both healthy and patient populations, where the nodes are brain areas and the links are the correlated activity between those brain areas. We can use them to analyse time-series of multiple neuron recordings, such as from epilepsy patients with implanted devices. We can use them to describe social groups – the people and the contacts between them – providing a basis for analysing the spread of infections and other contagious phenomena. Capturing big medical data in network form can thus allow us to visualise, analyse, and test that data in one unified description. But there is a problem common to all such medical applications: in data-driven networks, how do we know what structure in the network is signal and what is noise?

The aim of this project is to develop an analysis framework for separating signal from noise in data-driven networks. We will develop new, flexible data-driven null models for network structure – models that will predict noise. Using these models, we will create powerful ways of extracting the signal network from observed data. With these signal networks, we will then explore how they change the conclusions we can draw from our observed data, what large-scale structures they uncover in the networks, and how they can track changes to networks over time.

http://www.systemsneurophysiologylab.ls.manchester.ac.uk


This is a potential studentship to be funded via the MRC Doctoral Training Programme. Projects under this scheme are competitively funded; i.e. there are more projects advertised than available.