Theory-driven approaches: advancing understanding of disease
processes
Cognitive tasks and neuroimaging are combined with theoretical
approaches in computational psychiatry to learn more about differences
in behaviour, which in extremis are associated with symptoms of mental
illness. Computational models are created with the aim of explaining
observed behaviour (e.g. responses on tasks) in a mathematical
framework. These models are composed of free variables, i.e. parameters
which vary on a per individual basis and represent unobservable
neurobiological processes, such as learning or decision-making. When
these models are applied to imaging and to cognitive tasks, the
resulting parameters can illustrate individual differences in
performance at a mechanistic level, capturing the heterogeneity that is
a cardinal feature of complex conditions.
Reinforcement learning models represent a promising example of this
approach. Under this framework, the relationship between two modes of
learning about rewards can be assessed e.g., goal-directed vs. habitual.
Fitting these models to each individual’s data produces parameters,
i.e., summaries of the learning process, which can be compared to reveal
if there is a significant difference between two predefined groups or if
novel groups exist within the population.
These models have a biological basis, and provide further mechanistic
insight into the processes underlying aberrant behaviour. For example,
mismatches between what is expected and what is actually observed by an
individual generates prediction errors, which are encoded by
dopaminergic neurons in subcortical regions [1]. Deficits in
goal-directed learning, and therefore a tendency towards habits, have
been found across disorders such as binge eating, addiction, social
anxiety and obsessive-compulsive disorder (OCD), and have also been
related to a specific transdiagnostic factor - compulsive behaviour and
intrusive thought - which underlies all of these disorders [3].