Abstract
The study of schizophrenia over the past decades has generated a sea of data. This is
not entirely surprising in view of the vast size of the subject and the fact that it covers
so many disciplines. Unfortunately, all this data has not led us to a single set of con¬
clusions regarding the origin and pathophysiology of schizophrenia. Consequently
we have seen a number of theories attempting to describe the underlying biological
and psychological processes which are dysfunctional in schizophrenics. This thesis
attempts to reconcile some of these theories through the use of computational models
which allow us to investigate the links between biological and psychological processes.
Existing artificial neural network models of schizophrenia generally have poor bio¬
logical validity (chapter 4). They also rely on the interpretation of mental states as
binary patterns of activity, making it difficult to realistically represent complex men¬
tal phenomena such as those that occur in schizophrenia. To try and overcome these
problems I have focussed on the information processing properties of certain brain
structures implicated in schizophrenia, namely the prefrontal cortex (PfCx) and nu¬
cleus accumbens (NAcc), and I have used models which operate at several different
levels. Much work has been done on the structure and function of the PfCx and its
involvement in schizophrenia. However, an understanding of the role of dopamine
(DA) in the PfCx is still lacking. I have suggested a possible mechanism for the ac¬
tion of DA in the PfCx and illustrated this with biologically plausible models which
can be interpreted at cellular and pharmacological levels. I have then related this to
schizophrenia. Dysfunctions between brain regions are also suggested to underly the
symptoms of schizophrenia. This is the other theme of the thesis, where I have used
a reinforcement learning based model to examine interactions between brain regions
and the effects of variations in DA transmission on these interactions.
More specifically, I will show how oscillations between pyramidal cells and GABA
cells in the PfCx may arise (chapter 5), and how disruption of this information pro¬
cessing capacity can occur through multiple different pathologies. The existence of
oscillations is shown through simulations and theoretically by modelling neurotrans¬
mitter interactions within the mesocortical and mesolimbic dopamine (DA) systems
(chapter 6). This work reveals the conditions under which oscillations will occur, and
shows how DA can act as a control parameter in initiating oscillations. I have mod¬
elled a high level cognitive process, the Tower of London task, using a rule-based
model to represent PfCx function (chapter 7). Finally, a reinforcement learning model
is presented to illustrate putative NAcc function (chapter 8). The interaction between
the two models is investigated and illustrations of the possible origins of the positive
symptoms of schizophrenia are given. In all of these models, the role of DA has been
crucial. One conclusion from this work is that the symptoms of schizophrenia may
arise through inappropriate fluctuations in DA levels in the NAcc and the PfCx. The
work is based on a
large amount of neurobiological data and follows theories presented by Friston (1998) and Goldman-Rakic and Selemon (1997) amongst others.