CASE: Using computational modelling to understand the effects of pharmacological manipulations on cognitive function
Jonathan Roiser, Francesca Cormack, Cambridge Cognition
This collaborative project is a partnership between Prof Jonathan Roiser (UCL Institute of Cognitive Neuroscience) and Cambridge Cognition Ltd. Prof Roiser’s group focuses on understanding the cognitive and neural mechanisms underpinning mental health problems, especially depression. Over the past 5 years his work increasingly adopted a computational approach, using generative models to better understand patterns of behaviour (and brain responses) observed when human participants perform cognitive tasks.
Cambridge Cognition is a world leader in developing computerised cognitive testing, and it sells the well-known Cambridge Neuropsychological Test Automated Battery (CANTAB) software. It develops, often in partnership with academics, computerised tests that allow the measurement of specific domains of cognition, for example executive function, memory, attention and decision making. The tests it develops are based on a wealth of neuroscientific data from humans and animals.
The analysis of data derived from computerised cognitive tests is an active area of development, and over the past decade it has become increasingly common to use a computational approach to create parameters that summarise specific cognitive processes. This approach involves specifying, in mathematical form, precisely how the experimenter believes that the participant is completing the task. This is known as a “generative model” (because, given the same task as the human would perform, the model can generate behavioural responses). The behaviour of the model is governed by specific parameters, which can be interpreted in a cognitive framework, and by fitting the model to data collected in human participants it is possible to estimate parameters. These parameters then serve as summary statistics, in contrast to the traditional descriptive approach to data analysis, which usually involves simply calculating means or differences between conditions. The major advantage of the computational approach is that it can capitalise on the richness inherent in the data, which is usually overlooked in the traditional approach (for example, processes that evolve gradually over time which are difficult to capture using a purely descriptive approach).
A major aim of this project is to develop computational models for CANTAB tests, which are currently analysed using traditional descriptive approaches. Cambridge Cognition owns a wealth of data collected in the general population (both face-to-face and using its online platform) that can be used for this purpose. Following the development of the models, experiments will be conducted to understand how the model parameters are influenced by symptoms of mental health problems (especially depression), and by pharmacological interventions, particularly with cognitive enhancing drugs.
The first 6-9 months of the PhD will be spent at the ICN learning how to implement computational models (with support from Valton, a post-doctoral research in Roiser’s group), taking mandatory classes for first-year ICN PhD students, and completing a modelling course at the Gatsby Computational Neuroscience Unit (http://www.gatsby.ucl.ac.uk/teaching/courses/ml1/). A minimum amount of quantitative understanding and computer programming ability is required for this course and this will be a selection criterion.
Following the initial training period, the student will spend 3 months at Cambridge Cognition Ltd applying computational models to existing large datasets. The specific tests of interest will be the Delayed Match to Sample (episodic memory), Rapid Visual Information Processing (attention) and Cambridge Gambling Task (decision-making). Cambridge Cognition Ltd has data on several hundred participants who have also provided measures of depressive symptoms.
- Candidates should have a strong undergraduate (minimum high 2:1) or postgraduate (minimum merit, including in an empirical project) in one of the following subject areas: psychology; neuroscience; or a quantitative discipline such as mathematics, physics, computer science or engineering
- Candidates with an academic background in psychology or neuroscience will need to demonstrate excellent quantitative skills, including statistical training, computer programming ability and, ideally, familiarity with computational analysis of behavioural or neuroscientific data
- Candidates with an academic background in a quantitative discipline will need to demonstrate a strong interest in psychology and/or neuroscience, in particular relating to mental health
Adams RA, Huys QJ, Roiser JP (2016). Computational Psychiatry: towards a mathematically informed understanding of mental illness. Journal of Neurology, Neurosurgery and Psychiatry 87 (1): 53-63