Mechanisms of pattern processing in olfactory bulb-like circuits

The scope of brain research spans several orders of magnitude ranging from
small groups of amino acid residues in ion channels to fMRI signals reflecting
activity averaged over thousands of neurones. From a theoreticians point of
view very interesting questions arise at an intermediate level of cellular but
not sub-cellular resolution. How do neuronal units interact to process
information? Is it possible to find general laws or a repertoire of
computational motifs that would allow mastering the enormous challenge posed
by the brain’s sheer complexity?
Here I took advantage of the zebrafish olfactory bulb which combines a number
of features that make it an ideal target for theoretical analysis. Firstly,
the primary input to the olfactory bulb is known and can be administered by
the experimenter, allowing for both, control over and an obvious
interpretation of evoked activity. Secondly, due to the small size of the
olfactory bulb (20.000—30.000 neurones) a substantial fraction of all neurones
participating in an odour response can be recorded from in a single
experiment. Finally, the synaptic architecture of the olfactory bulb is
comparatively well-understood and simple.
In this study I used computational models to identify the structural features
of the olfactory bulb that are essential to its function. In order to
mechanistically understand this relation I complemented computer simulations
with mathematical analysis.
It is known from large-scale imaging experiments that peripheral odour
representations consisting of overlapping spatial patterns of afferent
activity are transformed into less overlapping representations carried by
mitral and tufted cells, the output elements of the olfactory bulb. It is
hypothesised that in refining odour representations for the benefit of
downstream circuits this pattern decorrelation serves an important function
(see chapter 1). Interestingly, a minimalistic circuit model (chapter 1) was
sufficient to reproduce most aspects of experimentally observed mitral cell
responses suggesting that decorrelation in the olfactory bulb is a network
phenomenon rather than a consequence of sophisticated computational properties
of individual neurones. In addition, the model was mathematically tractable
which allowed me to describe to a high level of detail and stringency the
mechanism by which this circuit achieves universal pattern decorrelation. In
the course I could explain why sparse connectivity and a high mitral cell
spontaneous activity lead to effective pattern decorrelation.
In simulations I also observed that symmetric connectivity further improves
decorrelation performance. In chapter 2 I present partial results towards a
theoretical analysis of this effect.
I also performed computer simulations with more detailed models consisting of
integrate-and-fire units. These were mostly exploratory in nature and are
therefore not described in this thesis. I did, however, include technical
documentation for the simulator I programmed (appendices 4 and 5) in the hope
that it will be useful.
The final chapter makes a simple observation regarding odour categorisation.