Statistical methods for whole brain cellular-resolution neural data.

The work can be thought of as a combination of the following two projects, supported by their motivation:

The era of neural 'Big Data' has confronted scientists with the challenge of modelling and interpreting high dimensional and often under-sampled recordings from the brain in order to discover and enhance our knowledge about its function. Over the last decade great strides have been made through the use of statistical models and machine learning algorithms to understand data. This has been both in part due to the availability of high performance hardware, but also due to the ever increasing demand for data science methods as a whole. To this end the research will look at extensions of existing statistical models so that they can continue to be used to help make sense of the increasingly difficult neurobiological tasks of the future.

The recent developments surrounding performance hardware and the need for data science methods has directly influenced the current interest and popularity in Bayesian methods. These are methods that have yet to be sufficiently explored with neural data sets even though their quantification of uncertainty and the ability to use prior knowledge is clearly beneficial to the field. This sets the second avenue of research, that of Bayesian Memory Mapping- A methodology for modelling memory consolidation on a whole brain cellular-resolution scale.
It is well accepted that episodic memories are initially encoded in the Hippocampus before being transferred to the Neocortex over the following months after initial encoding. The full network of locations where this trace takes place is however unknown. This research looks at using hierarchical Bayesian models to understand the networks underlying memory consolidation in the brain. Here the aim is to create and study the behaviour of these models to extract whole brain memory networks, asking questions about the model such as how well it replicates that of observed data in experiment. Hierarchical Bayesian models are parametric models, so this can be taken a step further by exploring what changes in a parameter does to simulated data and interpret it in the context of memory encoding and recall in the brain. One example of this would be looking at how certain brain regions impact memory networks through simulated lesion experiments.

The timing of this research is novel, proposing to understand and extract deeper insights into complex brain datasets of which no "standard" methodology exists. Here emphasis is placed on modelling the dataset as a whole as opposed to individual sections of interest. The motivation of this "joint" formation is to allow for interactions and information sharing between brain regions and their subregions in a biologically plausible way.

This project falls within the EPSRC Information and Communication Technologies, Healthcare Technologies.
This project falls within the ESPRC research areas of 'Statistics and Applied Probability', 'Information Systems'.