Data Science

There is now a growing array of analysis methods allowing researchers to move beyond univariate group comparisons or pairwise associations between variables.

In many projects we use contemporary analysis methods from data science. For example, using network science to capture complex interrelationships between variables. The resulting network can open the door to a new toolbox of analytical techniques, like graph theory. Rather than inferring the presence of singular latent factors, these approaches capture various different ways in which individual measures can be related over developmental time. For example, how are cognitive skills, literacy, numeracy and mental health related over developmental time?

Unsupervised machine learning techniques – like artificial neural networks – can also capture the multidimensional space in which children may differ. These algorithms are highly flexible, and the resulting models can easily accommodate non-linear relationships, identify sub-populations, make predictions about unseen data, be combined with simulations, incorporate different datatypes and open the way to tools for testing generalisation, like cross-validation.

Within the group we have recently used these tools to identify sub-populations of children with different cognitive or brain profiles, irrespective of diagnosis, and to map the non-linear relationships between brain organisation and cognitive difficulties.

If you would like find out more about this work then take a look at the papers appearing in the slider, or get in touch via the website.  


Key collaborators:

Dr Joe Bathelt (Royal Holloway, University of London)

Dr Joni Holmes (University of Cambridge)

Professor Anna Vignoles (University of Cambridge)

Current Funders

Follow us: