The following is the abstract of a nearly completed manuscript. Comments
or criticisms are sought.
IDENTIFICATION OF THE COMPONENTS OF ODOR MIXTURES FROM RESPONSES OF MANY
CLASSES OF PRIMARY OLFACTORY NEURONS
William T. Nickell
ABSTRACT
The vertebrate olfactory system poorly identifies the components of an
odor mixture. This may be a consequence of olfactory coding: if primary
olfactory neurons (PONs) each respond to a substantial fraction of
presented odors, the pattern of PON responses produced by a mixture of
odors will not resemble the patterns produced by the component odors
presented separately. Because of this, identification of component odors
is an inherently difficult computational task; performance on this task
may be determined by both PON responses (coding strategies) and by the
processsing algorithms available to the nervous system.
To investigate the contributions of coding and processing algorithms to
identification of components of mixtures, we developed a model of
olfactory receptors that simulates the responses of primary olfactory
neurons to pure odors and to mixtures of these odors. This model
incorporates two features of PONof components of mixtures: non-linear
receptor response and interaction of odors at a single receptor. Odorants
were assumed to bind to receptors according to saturation kinetics. A
random process was used to produce a binding constant for each combination
of 100 odors and 100 receptors. For multiple (up to 4) odorants, binding
was calculated using saturation kinetics. The PON response was made a
non-linear function of total ligand binding.
The memories of pure odors were simulated by calculating the responses of
the 100 PON classes to the 100 pure odors presented at a standard
concentration. We then tested the capacity of selected algorithms to
distinguish the components of mixtures of these pure odors using only the
information contained in the simulated PON firing rates for the odor
mixture and the memories of the PON responses to the pure odors.
Two global measures of similarity between the test and the target response
patterns (linear correlation and Euclidean distance) distinguished the
components of a mixture about as well as humans can: Mixtures of more
than three components at equal concentrations or addition of a single
masking odor at twice the intensity of the target odor decreased the
similarity to the level of unrelated odors.
The similarity in the ability of the model and human subjects in
identification of odor components suggest that the vertebrate olfactory
system may identify odors by computing a correlation-like function between
the responses to a presented odor and memories of previously presented
odors. Experimental tests of this hypothesis appear possible.