ArticlePDF AvailableLiterature Review

Abstract and Figures

Olfaction consists of a set of transforms from a physical space of odorant molecules, through a neural space of information processing, and into a perceptual space of smell. Elucidating the rules governing these transforms depends on establishing valid metrics for each of the three spaces. Here we first briefly review the perceptual and neural spaces, and then concentrate on the physical space of odorant molecules. We argue that the lack of an agreed-upon odor metric poses a significant obstacle toward understanding the neurobiology of olfaction, and suggest two alternative odor metrics as possible solutions.
Content may be subject to copyright.
A
vailable online at www.sciencedirect.com
Measuring smells
Rafi Haddad
1,2
, Hadas Lapid
1,2
, David Harel
2
and Noam Sobel
1
Olfaction consists of a set of transforms from a physical space
of odorant molecules, through a neural space of information
processing, and into a perceptual space of smell. Elucidating
the rules governing these transforms depends on establishing
valid metrics for each of the three spaces. Here we first briefly
review the perceptual and neural spaces, and then concentrate
on the physical space of odorant molecules. We argue that the
lack of an agreed-upon odor metric poses a significant
obstacle toward understanding the neurobiology of olfaction,
and suggest two alternative odor metrics as possible solutions.
Addresses
1
Department of Neurobiology, The Weizmann Institute of Science, Israel
2
Department of Computer Science and Applied Mathematics, The
Weizmann Institute of Science, Israel
Corresponding author: Haddad, Rafi (rhaddad@weizmann.ac.il) and
Sobel, Noam (noam.sobel@weizmann.ac.il)
Current Opinion in Neurobiology 2008, 18:438–444
This review comes from a themed issue on
Sensory systems
Edited by Tony Movshon and David P. Corey
Available online 3rd October 2008
0959-4388/$ – see front matter
#2008 Elsevier Ltd. All rights reserved.
DOI 10.1016/j.conb.2008.09.007
What was required was a perfume penetrating enough to obscure
the bouquet of rutting goat,yet not so overpowering that is called
undue attention to itself:there was little to be gained by moving
from one extreme to another on the olfactory scale
Jitterbug Perfume
Tom Robbins
Introduction
An olfactory scale, complete with the notion of units (Alo-
bars), was a trivial assumption for Tom Robbins in Jitterbug
Perfume [1], yet it has eluded both the perfumers practicing
the creation of scents, and the scientists studying the
mechanisms of their perception. The notion of probing
neural coding in a sensory space not bound by a metric is
puzzling. Imagine studying the neurobiological mechan-
isms of color vision without knowing that thecolor orange is
a reflection of a stimulus at 620 nm, red at 700 nm, and
blue at 450 nm. Moreover, imagine studying such a
system without a predictive framework that allows you
to assume that orange looks more like red than like blue...
Such is the current plight of olfaction research.
In studying sensory coding we are probing a transform
from an olfactory physicochemical space, through an
olfactory neural space, into an olfactory perceptual space.
Elucidating the rules of these transforms depends on
obtaining valid metrics for each of those spaces. Here
we will briefly comment on the two latter spaces, and then
concentrate on the notion of the olfactory physicochem-
ical metric space as a necessary component toward un-
derstanding olfactory coding.
Olfactory perceptual space
Perceptual spaces order odors such that distance in the
space confers similarity: odors near one another in the
spaces are expected to smell similar, and odorants distant
from one another are expected to smell dissimilar. Initial
efforts to develop a perceptual space assumed boundaries
defined by odor primaries, namely a small set of odors
from which all odor percepts could be composed [24].
This approach, however, failed to predict olfactory
experience, and as the number of potential primaries
grew, the strategy shifted from searching for individual
odor primaries to searching for perceptual axes, along
which these odorants may lay, and then use these axes to
define a space. In practice, this amounted to applying
various scaling methods to either similarity scores
obtained from odorant comparisons [59], or to verbal
descriptors applied to odorants [1012,13

]. In contrast to
the intuition of many, such verbal measures obtained
from humans are highly reliable and stable across time
and location [9,14], and spaces derived from such descrip-
tors have been validated using the above-noted criteria
whereby distance in the space predicted perceptual sim-
ilarity. Such a perceptual space recently developed in our
lab can be actively navigated by going to the odor space at
www.weizmann.ac.il/neurobiology/worg. Whereas many
studies have converged to suggest that the principal axis
of these spaces, and hence of human olfactory perception,
is odor pleasantness [1522], there has been only minimal
exploration of higher order axes within these perceptual
spaces.
Olfactory neural space
Olfactory neural spaces can be generated from neural
responses obtained by a variety of methods in a variety of
species [2329,30
]. The most comprehensive of these
efforts to measure the neural response to odors has given
rise to an extensive database of odorant-induced activity,
as measured with [
14
C]-2-deoxyglucose on the surface of
the rat olfactory bulb [31]. This database can be navigated
at http://leonserver.bio.uci.edu/. The results of such
measurements can then be used to formulate an olfactory
space where, again, the concept of similarity serves as a
Current Opinion in Neurobiology 2008, 18:438444 www.sciencedirect.com
guiding principal: odorants that generate similar neural
responses are proximal within such a space, and odorants
that generate divergent neural responses are distant
within the space [32,33]. Ideally, such spaces are then
further linked to behavior or perception [34
,3539].
Comparing the spaces
As previously noted, elucidating olfactory coding depends
on comparing three spaces: the physicochemical, neural,
and perceptual. Although there are various ways to
measure the physicochemical aspects of a molecule, none
of these have given rise to an agreed-upon metric that can
serve to compare one odorant molecule to another, or a
molecule to its ensuing neural activity and percept. Here
we will review several approaches to this problem, con-
centrating on what we think may be the more promising
toward an agreed physicochemical odor space.
Measuring odor quantity
Odor magnitude is odor concentration that can be con-
trolled to a reasonable extent by odorant-generating
devices known as olfactometers [4042], and measured
with analytical instruments such as photoionization
detectors. The ability to control and measure odor con-
centration allowed uncovering the mostly simple relations
between the physical, neuronal, and perceptual in this
realm, whereby increases in concentration lead to
increases in firing rate at the receptor [43,44] and spatial
extent of activity in the bulb [4547], as well as increases
in perceived intensity. The latter can be captured by a
simple logarithmic power function [48] with a slope of less
than 1; that is, successive increases in stimulus concen-
tration produce successively smaller increases in per-
ceived intensity. The degree of this compression (i.e.
the slope parameter) is odorant specific, reflecting the
odorants’ solubility in water [49]. Thus, the ability to
measure and control the olfactory stimuli gave rise to a
rule relating the physical property to neural representa-
tion, and to perception.
Measuring odor quality
Whereas the physicochemical properties that determine
odorant quantity are clear, the rules linking such proper-
ties to odor quality remain unknown. A first significant
attempt at solving this problem, also referred to as the
structure-to-odor response (SOR), was conducted by
Amoore [50]. Amoore identified benzaldehyde as a pro-
totypical molecule for an almond odor note, and then
successfully predicted the almondness of other molecules
on the basis of their three-dimensional structural fit to the
reference molecule. This initial study was followed by a
myriad of SOR efforts (comprehensively reviewed in
[51]). These studies identified several molecular proper-
ties, such as molecular weight, length, bond type, electron
donor, functional groups, and others [16,52], each of
which had a somewhat predictable influence on subsets
of odor qualities. However, all these models performed
relatively poorly when novel odorants were evaluated
[53]. Consequently, in probing the neurobiology of olfac-
tion, researchers had gravitated toward using the practical
approach of selecting odorants that differed in only one
specific attribute (in most cases the number of carbon
atoms). Although this approach has proved to be valuable
to some extent [51], it is certainly inadequate for compar-
ing any two randomly selected odorants. To summarize,
to date there is no agreed-upon olfactory metric that
enables universal odorant comparison.
Generating an olfactory physicochemical
metric
To measure and control olfactory stimuli qualitatively we
need to identify the molecular features that govern the
biological interaction. However, given the vast number of
molecular features and the diversity of olfactory receptors
across species, it is improbable that one particular mol-
ecular feature will dominate this interaction. In other
words, it is improbable that a single physicochemical
feature will influence an olfactory perceptual or neural
axis in the way that the single physical features of wave-
length or frequency dominate the perceptual axes of color
and pitch in vision and audition. One possible bypass of
this problem is to represent each odorant by a very large
number of molecular descriptors, albeit captured in a
single value. Here we describe two separate efforts we
have made in this direction.
A physicochemical odor metric that predicts
olfactory perception
Single odorants may have many physicochemical fea-
tures, and one expects these features to present them-
selves at various probabilities within the world of
molecules that have a smell. These probabilities can
be captured and represented by applying methods of
statistical dimension-reduction to detailed molecular
descriptions of odorants. To this end, in Khan et al.
[13

] we used structural chemistry software (Dragon:
http://www.talete.mi.it) to obtain 1664 molecular descrip-
tors for more than 1500 odorants. We then applied prin-
cipal components analysis (PCA), a well-established
method for dimension-reduction that generates an
orthogonal basis set for the profile space, in which each
successive dimension has the maximal possible variance.
Hence, the first principal component (PC), that can be
considered as the first new feature or dimension, is the
‘best’ one-dimensional reflection of the data. Thus, PC1
of the 1664 molecular features of the 1500 odorants can
be used as a physicochemical metric for olfaction, where
every odorant can be assigned a PC1 score.
The striking outcome of this exercise was that it revealed
a significant correlation between the primary dimension
(PC1) of physicochemical space and the primary dimen-
sion of perceptual space, namely odorant pleasantness.
This correlation allowed us to predict the pleasantness of
Measuring smells Haddad et al. 439
www.sciencedirect.com Current Opinion in Neurobiology 2008, 18:438444
odorants we had never smelled before (and that were not
part of the model-building set), on the basis of their
physicochemical structure alone (Figure 1). The main
significance of this was not in identifying pleasantness as
the primary dimension of olfactory perception (a notion
well established [1522]), nor in the ability to predict
perceptual properties from structure (a feat previously
achieved [51], albeit rarely for novel odorants not part of
the model-building set), but rather in finding that the
primary dimension of perception had a privileged link to
PC1 of structure. In other words, the single optimal axis
for explaining the variance in the physicochemical data
was the best predictor of odor pleasantness. That this
perceptual dimension is the best correlate of the most
discriminating physicochemical measures suggests that,
as with other senses, the olfactory system has evolved to
exploit a fundamental regularity in the physical world.
A physicochemical odor metric that predicts
neural response patterns
The above PC1 of physicochemical space is a single axis.
However, it is multidimensional in the sense that 1664
known features contributed to it with known weights. In
other words, we can represent each odorant as a single
value reflecting its PC1 score, or we can represent each
odorant as a vector of 1664 values. When using the former
approach, the distance between two odorants is the
difference in PC1 values. When using the latter approach,
one can compute the distance between any two odorants
by the square root of the sum of squares of the differences
between the descriptors (Euclidean distance). To ask
whether such a metric can be used to predict neural
activity in the olfactory system, in Haddad et al.[54

]
we revisited nine previously published datasets and ana-
lyzed a novel dataset, for which we knew the odorants
used but did not know the neural response. These data-
sets consisted of different olfactory neurons (e.g. recep-
tors; glomeruli), different model systems (e.g. fly; rat),
different neuronal response measurement techniques
(e.g. imaging; electrical recording), and odorants varying
along different feature types (e.g. carbon chain-length;
functional group). We found that this multidimensional
metric generated predictions of neural activity that were
not only statistically significant, but were also significantly
better at accounting for neural responses than the particu-
lar metric used in each specific study (e.g. carbon chain-
length) (Figure 2). In other words, this approach enabled
us to use odorant structure in order to predict odorant-
induced neural activity in nonhuman animals. Thus, it
provided a generic method for comparing any number of
structurally diverse odorants without predetermining the
particular features important for each species. Moreover,
the applicability of this metric across the different species
tested, suggests that odor space is conserved across organ-
isms [55

].
The relation between the two proposed
physicochemical metrics
We have presented two metrics; both on the basis of
representing odorants using a very large number of mol-
ecular descriptors. The first metric, on the basis of the first
PC of these descriptors (the axis best explaining their
variance), enabled the prediction of perceptual attributes
(pleasantness). We will call this the variance metric. The
second metric, on thebasis of Euclidean distances between
odorants in the 1664 physicochemical space, enabled the
prediction of odorant-induced neuronal response patterns.
We will call this the distance metric. To probe the relation
between these two metrics, we tested whether the variance
metric that predicted perception in humans could similarly
predict neural activity in other animals. In four out of eight
datasets tested, we found a significant correlation between
the variance metric and PC1 of neuronal response. That no
correlation was found in the remaining four datasets may be
explained by the relatively small size of these datasets
combined with the sensitivityof the PCA method to noise.
In turn, in cases where the number of neurons sampled or
the noisiness of the measuring process renders the variance
metric inaccurate, the distance metric that is on the basis of
the full 1664-representation is less likely to err. Consistent
with this, the distance metric indeed predicted neural
response distances in all eight datasets we analyzed. Con-
versely, one can hypothesize a case where the distance
metric will be less accurate than the variance metric. For
example, if the direction of maximal variability of a set of
odorants in the physicochemical space and in the percep-
tual space is similar yet the internal distance between these
odorants in the two spaces is different (see Figure 3 for an
illustration). This situation may be consistent with a
440 Sensory systems
Figure 1
The correlation between the variance metric (first principle component of
molecular structure) and the estimated pleasantness of 90 different
odorants as assessed by 20 subjects (note that these are not the same
odorants used in Khan et al.[13

]).
Current Opinion in Neurobiology 2008, 18:438444 www.sciencedirect.com
remapping from structural to perceptual that may occur at a
the cortical level [56
,57].
Electronic measurements of odors
A hidden assumption of the above-described physico-
chemical spaces is that they accept the general framework
regarding olfactory transduction, referred to as the odo-
tope approach [5860]. Specifically, they assume that
different olfactory receptors have different affinities to
specific molecular structural physicochemical properties,
and that the differential activation of these receptors
gives rise to a spatiotemporal pattern of activity that
reflects the odor. Despite a preponderance of evidence
favoring this general framework, an alternative frame-
work suggesting that olfactory receptors measure the
molecular vibrational frequencies of molecules has been
considered [6163]. In the context of an olfactory metric,
this vibrational approach is of course very appealing,
because in its simplest form it would provide a single
axis (vibrational frequency) that could serve to predict
both perception and neural activity in the olfactory sys-
tem. Some psychophysical tests of this theory, however,
fail to support it [64]. Full consideration of this issue is
beyond the scope of this manuscript, but regardless of
how olfactory receptors do their business, this issue raises
the possibility of generating an olfactory metric using an
external odor measurement device, regardless of the
device’s mode of action. For example, we can use the
values reported by mass spectrography (MS), gas chroma-
tography (GC), IR spectra or Raman spectra, and most
recently, electronic noses (eNose).
eNoses are analytic devices that are playing an increasing
role as general-purpose odor analyzers [65]. eNoses are
Measuring smells Haddad et al. 441
Figure 2
Correlation plots of four unrelated datasets [78,79

,80,81]. Each point in the graphs represents the distances between two odorants in both the neural
space (difference in neural activity) and the distance metric (the metric used is the optimized metric described in Haddad et al.[54

]).
www.sciencedirect.com Current Opinion in Neurobiology 2008, 18:438444
cheaper than GCs, and are easier to use. The main
component of an eNose is an array of nonspecific chemi-
cal sensors. An analyte stimulates many of the sensors in
the array and elicits a characteristic response pattern. The
sensors inside eNoses can be made of a variety of tech-
nologies, but in all cases a certain physical property is
measured and a set of signals is generated. The stages of
the recognition process are similar to those of biological
olfaction, where a sensor responds to more than one
odorant and one odorant activates more than one sensor.
Together, the set of activated sensors and their signals
characterize the odor. Different eNoses can be mapped
onto one another [66] and used for odor classification [67
69] including classification of odor mixtures [70,71].
Initial efforts havebeen made to link eNose measurements
to olfactory perception [72] and activity in olfactory re-
ceptor neurons [73]. If these links are substantiated, an
eNose odor space can serve as a key tool to elucidating
coding in olfaction [74]. This will hold true only if research-
ers agree on a particular eNose and a particular analysis, in
order to allow comparisons across time and location.
Conclusions
Our approach to generating and testing olfactory spaces
was in fact quite fashionable in the late 1960s and
early 1970s [8,16,19,75,76]. However, the limited
computational powers commonly available at that time
limited the scope of these efforts. For example, the efforts
to generate physichochemical spaces typically used less
than 20 molecular descriptors, and the efforts to generate
perceptual spaces typically used only a few tens of odor-
ants. The limited applicability of such efforts rendered
this approach obsolete. The current availability of struc-
tural chemistry software offering thousands of molecular
descriptors, combined with modern computational
approaches such as PCA, and modern computing,
together have allowed us to generate spaces consisting
of thousands of odorants each described by thousands of
molecular physicochemical descriptors and hundreds of
verbal perceptual descriptors. These efforts have gener-
ated meaningful spaces, capable of predicting perception
[13

], and neural responses [54

] to novel odorants. It is
noteworthy that the increase in number of physicochem-
ical descriptors represents more than merely an increase
in power, but rather a shift toward describing the relevant
space much in the way the mammalian olfactory system
itself has tackled this task, with more than a thousand
receptor types [77].
To conclude, Galileo said: ‘Count what is countable,
measure what is measurable. What is not measurable,
make measurable’. Here we have highlighted two pro-
posed odor metrics. Whether it is these metrics, some
refined version of them, or some new metric, that end up
deemed representative of the world of odor, the avail-
ability of such a metric remains a crucial must if we are to
elucidate the neurobiology of olfaction.
Acknowledgements
Authors RH and NS are supported by an FP7 Ideas Grant 200850 from the
European Research Council. Author DH is supported by The John von
Neumann Minerva Center for the Development of Reactive Systems of the
Weizmann Institute of Science. We thank Arak Elite.
References and recommended reading
Papers of particular interest, published within the period of review,
have been highlighted as:
of special interest
 of outstanding interest
1. Robbins T: Jitterbug Perfume. Bantam; 1984.
2. Henning H: Der geruch. Leipzig: Barth; 1916.
3. Amoore JE: Evidence for the chemical olfactory code in man.
Ann N Y Acad Sci 1974, 237:137-143.
4. Guillot M: Anosmies partielles et odeurs fondamentales.CR
Acad Sci 1948, 226:1307-1309.
5. Moskowitz HR, Gerbers CL: Dimensional salience of odors.Ann
N Y Acad Sci 1974, 237:1-16.
6. Woskow MH: Multidimensional scaling of odors.In Theories of
Odor and Odor Measurement. Edited by Tanyolac N. Robert
College; 1968:147-191.
7. Stevens DA, Oconnell RJ: Semantic-free scaling of odor quality.
Physiol Behav 1996, 60:211-215.
8. Schiffman SS: Contributions to the physiochemical
dimensions of odor: a psychophysical approach.Ann N Y Acad
Sci 1974, 237:164-183.
442 Sensory systems
Figure 3
An example for a case where the distance metric will be less accurate
than the variance metric. The blue dots are four odorants plotted in 2D
(e.g. perceptual space). The red dots are the same odorants in a different
2D representation (e.g. neuronal space), where two odorants have
shifted considerably (dots 2 and 3). The axis of maximum variability of
the odorants in the two representations remains similar (the main
diagonal). Thus, the variance metric will provide a good fit within both
spaces. However, the pair wise distances between odorants in the two
representations differ considerably. Thus, the distance metric may fail to
find a relation in one of the spaces.
Current Opinion in Neurobiology 2008, 18:438444 www.sciencedirect.com
9. Dawes PJD, Dawes MT, Williams SM: The smell map:
commonality of odour perception confirmed.Clin Otolaryngol
2004, 29:648-654.
10. Mamlouk, A.M.: Quantifying Olfactory Perception. University of
Lubeck, Institute for Signal Processing, Lubeck, Germany. The
thesis is available online at: http://www.inb.uni-luebeck.de:8084/
publications/pdfs/Mada02.pdf.
11. Mamlouk AM, Chee-Ruiter C, Hofmann UG, Bower JM:
Quantifying olfactory perception: mapping olfactory
perception space by using multidimensional scaling
and self-organizing maps.Neurocomputing 2003, 5254:
591-597.
12. Chee-Ruiter CWJ, Bower JM: Representing odor quality space:
a perceptual framework for olfactory processing.In
Proceedings of the Sixth Annual Conference on Computational
Neuroscience: Trends in Research, 1998: Trends in Research,
1998 Table of Contents. 1998:591-598.
13.

Khan R, Luk C, Flinker A, Aggarwal A, Lapid H, Haddad R, Sobel N:
Predicting odor pleasantness from odorant structure:
pleasantness as a reflection of the physical world.J Neurosci
2007, 27:10015-10023.
This study describes a systematic link between an odor molecular
structure and its perceived pleasantness, suggesting that olfactory plea-
santness is partly written into nature much like visual color and auditory
pitch.
14. Dravnieks A: Odor quality: semantically generated multi-
dimensional profiles are stable.Science 1982, 218:799-801.
15. Richardson JT, Zucco GM: Cognition and olfaction: a review.
Psychol Bull 1989, 105:352-360.
16. Schiffman SS: Physicochemical correlates of olfactory quality.
Science 1974, 185:112-117.
17. Godinot N, Sicard G: Odor categorization by human-
subjects — an experimental approach.Chem Senses 1995,
20:101.
18. Berglund B, Berglund U, Engen T, Ekman G: Multidimensional
analysis of 21 odors.Scand J Psychol 1973, 14:131-137.
19. Schiffman S, Robinson DE, Erickson RP: Multidimensional-
scaling of odorants — examination of psychological and
physiochemical dimensions.Chem Senses Flavour 1977, 2:
375-390.
20. Steiner JE: Human facial expressions in response to taste and
smell stimulation.Adv Child Dev Behav 1979, 13:257-295.
21. Soussignan R, Schaal B, Marlier L, Jiang T: Facial and autonomic
responses to biological and artificial olfactory stimuli in
human neonates: re-examining early hedonic discrimination
of odors.Physiol Behav 1997, 62:745-758.
22. Engen T: The Perception of Odors. New York: Academic Press;
1982.
23. Johnson BA, Woo CC, Leon M: Spatial coding of odorant
features in the glomerular layer of the rat olfactory bulb.
J Comp Neurol 1998, 393:457-471.
24. Meister M, Bonhoeffer T: Tuning and topography in an odor
map on the rat olfactory bulb.J Neurosci 2001, 21:
1351-1360.
25. Mori K, Nagao H, Yoshihara Y: The olfactory bulb: coding and
processing of odor molecule information.Science 1999,
286:711-715.
26. Manzini I, Brase C, Chen TW, Schild D: Response profiles to
amino acid odorants of olfactory glomeruli in larval Xenopus
laevis.J Physiol 2007, 581:567-579.
27. Hallem EA, Carlson JR: The odor coding system of Drosophila.
Trends Genet 2004, 20:453-459.
28. Galizia CG, Sachse S, Rappert A, Menzel R: The glomerular code
for odor representation is species specific in the honeybee
Apis mellifera.Nat Neurosci 1999, 2:473-478.
29. Laurent G: Olfactory processing: maps, time and codes.Curr
Opin Neurobiol 1997, 7:547-553.
30.
Davison IG, Katz LC: Sparse and selective odor coding by
mitral/tufted neurons in the main olfactory bulb.J Neurosci
2007, 27:2091.
This study used more than 300 odorants to probe olfactory mitral cell
responses. It provides insight on the relationship between odor structure
and neural response.
31. Johnson BA, Woo CC, Hingco EE, Pham KL, Leon M:
Multidimensional chemotopic responses to n-aliphatic
acid odorants in the rat olfactory bulb.J Comp Neurol 1999,
409:529-548.
32. Doving KB: Odorant properties correlated with physiological
data.Ann N Y Acad Sci 1974, 237:184-192.
33. Duchamp A, Revial MF, Holley A, Leod P: Odor discrimination by
frog olfactory receptors.Chem Senses 1974, 1:213-233.
34.
Youngentob SL, Johnson BA, Leon M, Sheehe PR, Kent PF:
Predicting odorant quality perceptions from multidimensional
scaling of olfactory bulb glomerular activity patterns.Behav
Neurosci 2006, 120:1337-1345.
Together with Ref. [55

] this study demonstrates how odors that elicit
similar neural response tend to elicit similar behavior. This specific study
used odors that differfrom each profoundly and thus the observedbehavior
could not be simply predicted from the differences in odor structure.
35. Kent PF, Youngentob SL, Sheehe PR: Odorant-specific spatial
patterns in mucosal activity predict perceptual differences
among odorants.J Neurophysiol 1995, 74:1777-1781.
36. Kent PF, Mozell MM, Youngentob SL, Yurco P: Mucosal activity
patterns as a basis for olfactory discrimination: comparing
behavior and optical recordings.Brain Res 2003, 981:1-11.
37. Cleland TA, Morse A, Yue EL, Linster C: Behavioral models of
odor similarity.Behav Neurosci 2002, 116:222-231.
38. Linster C, Johnson BA, Yue E, Morse A, Xu Z, Hingco EE, Choi Y,
Choi M, Messiha A, Leon M: Perceptual correlates of neural
representations evoked by odorant enantiomers.J Neurosci
2001, 21:9837-9843.
39. Ho SL, Johnson BA, Chen AL, Leon M: Differential responses to
branched and unsaturated aliphatic hydrocarbons in the rat
olfactory system.J Comp Neurol 2006, 499:519-532.
40. Bodyak N, Slotnick B: Performance of mice in an automated
olfactometer: odor detection, discrimination and odor
memory.Chem Senses 1999, 24:637-645.
41. Kobal G, Hummel C: Cerebral chemosensory evoked potentials
elicited by chemical stimulation of the human olfactory and
respiratory nasal mucosa.Electroencephalogr Clin Neurophysiol
1988, 71:241-250.
42. Johnson BN, Sobel N: Methods for building an olfactometer
with known concentration outcomes.J Neurosci Methods 2007,
160:231-245.
43. Duchamp-Viret P, Chaput MA, Duchamp A: Odor response
properties of rat olfactory receptor neurons.Science 1999,
284:2171-2174.
44. Duchamp-Viret P, Duchamp A, Chaput MA: Peripheral odor
coding in the rat and frog: quality and intensity specification.
J Neurosci 2000, 20:2383-2390.
45. Rubin BD, Katz LC: Optical imaging of odorant representations
in the mammalian olfactory bulb.Neuron 1999, 23:499-511.
46. Spors H, Grinvald A: Spatio-temporal dynamics of odor
representations in the mammalian olfactory bulb.Neuron 2002,
34:301-315.
47. Johnson BA, Leon M: Modular representations of odorants in
the glomerular layer of the rat olfactory bulb and the effects of
stimulus concentration.J Comp Neurol 2000, 422:496-509.
48. Stevens SS: On the psychophysical law.Psychol Rev 1957,
64:153-181.
49. Cain WS: Odor intensity. Differences in exponent of
psychophysical function.Percept Psychophys 1969, 6: 349.
50. Amoore JE: Stereochemical and vibrational theories of odour.
Nature 1971, 233:270-271.
Measuring smells Haddad et al. 443
www.sciencedirect.com Current Opinion in Neurobiology 2008, 18:438444
51. Rossiter KJ: Structureminus signOdor relationships.Chem Rev
1996, 96:3201-3240.
52. McGill JR, Kowalski BR: Intrinsic dimensionality of smell.Anal
Chem 1977, 49:596-602.
53. Sell CS: On the unpredictability of odor.Angew Chem Int Ed
2006, 45:6254-6261.
54.

Haddad R, Khan R, Takahashi YK, Mori K, Harel D, Sobel N: A
metric for odorant comparison.Nat Methods 2008, 5:425-429.
This study provides a metric for organizing odorants in a multidim ensional
space. This metric enables measuring distances between any two odor-
ants so that close odorants will elicit similar neural response. This metric
is shown to be applicable across different species and thus suggests that
odor space is conserved across species (has also been shown in Ref.
[55

]).
55.

Kreher SA, Mathew D, Kim J, Carlson JR: Translation of sensory
input into behavioral output via an olfactory system.Neuron
2008, 59:110-124.
This study measured the response profiles of olfactory receptors in
Drosophila larva and showed that the activity of small number of recep-
tors correlates with the strength of the behavioral response and that odor
spaces is conserved across larva and adult Drosophila.
56.
Gottfried JA, Winston JS, Dolan RJ: Dissociable codes of odor
quality and odorant structure in human piriform cortex.Neuron
2006, 49:467-479.
This work suggests that there are two regions for odor coding in the
piriform cortex, whereby posterior regions encode quality (but not struc-
ture) and anterior regions encode structure (but not quality).
57. Wilson DA: Rapid, experience-induced enhancement in
odorant discrimination by anterior piriform cortex neurons.
J Neurophysiol 2003, 90:65-72.
58. Buck LB: Information coding in the vertebrate olfactory
system.Annu Rev Neurosci 1996, 19:517-544.
59. Axel R: The molecular logic of smell.Sci Am 1995, 273:
154-159.
60. Firestein S: How the olfactory system makes sense of scents.
Nature 2001, 413:211-218.
61. Brookes JC, Hartoutsiou F, Horsfield AP, Stoneham AM: Could
humans recognize odor by phonon assisted tunneling? Phys
Rev Lett 2007, 98:38101.
62. Wright RH: Odor and molecular vibration: neural coding of
olfactory information.J Theor Biol 1977, 64:473-502.
63. Turin L: A spectroscopic mechanism for primary olfactory
reception.Chem Senses 1996, 21:773-791.
64. Keller A, Vosshall LB: A psychophysical test of the vibration
theory of olfaction.Nat Neurosci 2004, 7:337-338.
65. Gardner JW, Bartlett PN: Electronic noses. Principles and
applications.Meas Sci Technol 2000, 11:1087.
66. Shaham O, Carmel l, Harel D: On mapping between electronic
noses.Sens Actuators B: Chem 2005, 106:76-82.
67. Dutta R, Hines EL, Gardner JW, Boilot P: Bacteria classification
using Cyranose 320 electronic nose.Biomed Eng Online 2002,
1:4.
68. Turner AP, Magan N: Electronic noses and disease diagnostics.
Nat Rev Microbiol 2004, 2:161-166.
69. Mandenius CF: Electronic noses for bioreactor monitoring.Adv
Biochem Eng Biotechnol 2000, 66:65-82.
70. Carmel l, Sever N, Harel D: On predicting responses to mixtures
in quarz microbalance sensors.Sens Actuators B: Chem 2005,
106:128-135.
71. Carmel l, Harel D: Mix-to-mimic odor synthesis for electronic
noses.Sens Actuators B: Chem 2007, 125:635-643.
72. Burl MC, Doleman BJ, Schaffer A, Lewis NS: Assessing the ability
to predict human percepts of odor quality from the detector
responses of a conducting polymer composite-based
electronic nose.Sens Actuators: B Chem 2001, 72:149-159.
73. Haddad R, Carmel L, Sobel N, Harel D: Predicting the receptive
range of olfactory receptors.PLoS Comput Biol 2008, 4:e18.
74. Harel D, Carmel L, Lancet D: Towards an odor communication
system.Comput Biol Chem 2003, 27:121-133.
75. Amoore JE, Venstrom D: Correlations between steriochemical
assessments and organoleptic analysis of odourus
compounds.In Olfaction and Taste II. Edited by Hayashi T.
Pergamon Press; 1967:3-17.
76. Amoore JE, Palmieri G, Wanke E: Molecular shape and odour:
pattern analysis of PAPA.Nature 1967, 216:1084-1087.
77. Buck L, Axel R: A novel multigene family may encode odorant
receptors: a molecular basis for odor recognition.Cell 1991,
65:175-187.
78. Manzini I, Peters F, Schild D: Odorant responses of Xenopus
laevis tadpole olfactory neurons: a comparison between
preparations.J Neurosci Methods 2002, 121:159-167.
79.

Hallem EA, Carlson JR: Coding of odors by a receptor
repertoire.Cell 2006, 125:143-160.
This is the first large-scale profiling of olfactory receptor neurons using
more than 100 odorants. This dataset can be used to unravel the relation-
ship between odor structure and neural response as was done in Refs.
[54

,73].
80. Sato T, Hirono J, Tonoike M, Takebayashi M: Tuning specificities
to aliphatic odorants in mouse olfactory receptor neurons and
their local distribution.J Neurophysiol 1994, 72:2980-2989.
81. Sachse S, Rappert A, Galizia CG: The spatial representation of
chemical structures in the antennal lobe of honeybees: steps
towards the olfactory code.Eur J Neurosci 1999, 11:3970-3982.
444 Sensory systems
Current Opinion in Neurobiology 2008, 18:438444 www.sciencedirect.com
Conference Paper
Discovering the alliance between the molecular arrangement and its corresponding smell has proved to be a laborious task in the field of neuroscience, olfactory research, perfumery, psychology, and chemistry. One of the constraints to demonstrate structureodor alliance is the fuzzy and incon-clusive molecular descriptors with distinct sources of smell-based molecular compounds. The graph models tend to suffer from the problem of over-smoothing whenever the number of layers is increased, node representations become fuzzy, performance degrades, and graph representations cannot be distinguished. To overcome this problem, we have proposed the Graphormer model- Bidirectional Encoder Representations from Transformers(BERT) with graph methods(GraphSAGE, Graph Isomorphism Network(GIN), Graph Attention Networks(GAT), and Graph Convolution Network(GCN)). It uses Atomic Co-ordinates(AC), Substring(SS), Structural Images(IMG) representations for the prediction of odor classes and predicts sharp smells using SMILES input. Multi-label prediction models used 5300 chemical compounds with 110 smell insights. Graphormer(GraphSAGE)+ HYB model got the highest accuracy of 98.59% and 98% AUC while with AC+SS it attained 97.8% accuracy and 98% AUC. Hence, the proposed Graphormer paradigms with physico-chemical properties motivate us to pin-point the broad correlation between the structure and smell percepts by outperforming the existing baseline models.
Article
The sense of smell is based on sensory detection of the molecule(s), which is then further perceptually interpreted. A possible measure of olfactory perception is an odor independent olfactory perceptual fingerprint (OPF) defined by Snitz et al. We aimed to investigate, whether OPF can distinguish patients with olfactory dysfunction due to COVID-19 from controls and which perceptual descriptors are important for that separation. Our study included 99 healthy controls and 41 patients. They rated ten odors using eight descriptors 'pleasant', 'intense', 'familiar', 'warm', 'cold', 'irritating', ‘edible', and ‘disgusting'. An unsupervised machine learning method, hierarchical cluster analysis, showed that OPF can distinguish patients from controls with accuracy of 83%, sensitivity of 51%, and specificity of 96%. Furthermore, a supervised machine learning method, random forest classifier, showed that OPF can distinguish patients and controls in the testing dataset with accuracy of 86%, sensitivity of 64%, and specificity of 96%. Principal component analysis and random forest classifier showed that familiarity and intensity were the key qualities to explain the variance of the data. In conclusion, people with COVID-related olfactory dysfunction have a fundamentally different olfactory perception.
Article
Full-text available
Replicating specific scents by mixing ‘base odors’ is the main idea behind the concept of odor reproduction. Just as primary colors can be combined to generate a wide range of colors in visual perception, the hope is to find a similar principle in olfaction. However, defining such ‘primary odors’ is difficult due to the complexity of our sense of smell, which is much different from vision. In spite of these challenges, numerous scientists have endeavored to identify a combination of basic smell components that can mimic various scents. Currently, the suggested combinations of smell components are typically only successful at recreating specific, intended scents. The range of reproducible scents differs based on the selection of base odors, yet these studies have provided valuable knowledge about the nature of these primary odors. This review paper will focus on the latest studies related to odor reproduction. Understanding odor space, odor reproduction, and the challenges encountered when expanding the range of odors will be explained. © 2023 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.
Article
Full-text available
The spread of the SARS‐CoV‐2 virus produces a new disease termed COVID‐19, the underlying physiological mechanisms of which are still being understood. Characteristic of the infection is the compromising of taste and smell. There is a persistent need to discriminate the dysfunctions and correlation between taste and smell, which are probably epiphenomena of other concealed conditions. Anosmic and ageusic long‐term COVID‐19 patients were re‐evaluated after 1 year using a Volabolomic approach with an e‐nose recording system coupled with olfactometric and gustometric tests. Here a range of sensory arrangements was found, from normal taste and smell to complete losses. The following patterns of olfactory threshold (OT)‐taste threshold‐olfactory uni‐ and cross‐modal perception were found anosmia‐severe hypogeusia‐anosmia; hyposmia‐hypogeusia‐severe hyposmia; normosmia‐ageusia‐hyposmia; severe hyposmia ‐normogeusia‐normosmia. There is a strong correlation between OT and olfactory uni‐ and cross‐modal perception, a moderate correlation between olfactory and taste threshold and no correlation between OT and taste threshold. In conclusion, this study provides evidence for the feasibility of testing the chemical senses to directly objectify function in order to discriminate taste from olfactory impairment. Furthermore, it allows to hypothesize a long‐term effect of the virus due to neuroinvasion through, probably, the olfactory system with injury in the related multisensory areas of taste and smell. Anosmic and ageusic long‐term COVID‐19 patients were re‐evaluated after 1 year. The following patterns of long‐term COVID‐19 were found: anosmia‐anosmia‐severe hypogeusia; hyposmia‐severe hyposmia‐hypogeusia; normosmia‐hyposmia‐ageusia; severe hyposmia‐normosmia‐normogeusia. A long‐term effect of the virus due to neuroinvasion through, probably, the olfactory system with lesions in the related multisensory areas of taste and smell is therefore conceivable.
Article
Negative symptoms are among the greatest sources of functional impairment for individuals with schizophrenia, yet their mechanisms remain poorly understood. Olfactory impairment is associated with negative symptoms. The processing of pleasant olfactory stimuli is subserved by reward-related neural circuitry while unpleasant olfactory processing is subserved by emotion-related neural circuitry, suggesting that these two odor dimensions may offer a window into differential mechanisms of negative symptoms. We examined whether pleasant and unpleasant odor identification bears differential relationships with avolition and inexpressivity dimensions of negative symptoms, whether these relationships are transdiagnostic, and whether pleasant and unpleasant odor processing also relate differently to other domains of functioning in a sample of individuals diagnosed with schizophrenia (N = 54), other psychotic disorders (N = 65), and never-psychotic adults (N = 160). Hierarchical regressions showed that pleasant odor identification was uniquely associated with avolition, while unpleasant odor identification was uniquely associated with inexpressivity. These relationships were largely transdiagnostic across groups. Additionally, pleasant and unpleasant odor identification displayed signs of specificity with other functional and cognitive measures. These results align with past work suggesting dissociable pathomechanisms of negative symptoms and provide a potential avenue for future work using valence-specific olfactory dysfunction as a semi-objective and low-cost marker for understanding and predicting the severity of specific negative symptom profiles.
Article
Full-text available
This paper introduces a novel dual-aspect theory of consciousness that is based on the principle of holographic-duality in modern physics and explores the prospects of making philosophically significant empirical discoveries about the physical correlates of consciousness. The theory is motivated by an approach that identifies certain anti-physicalist problem intuitions associated with representational content and spatial location and attempts to provide these with a consciousness-independent explanation, while suspending questions about the hard problem of consciousness and the more problematic “phenomenal character”. Providing such topic neutral explanations is “hard” enough to make a philosophical difference and yet “easy” enough to be approached scientifically. I will argue that abstract algorithms are not enough to solve this problem and that a more radical “computation” that is inspired by physics and that can be realized in “strange metals” may be needed. While speculative, this approach has the potential to both establish necessary connections between structural aspects of conscious mental states and the physical substrate “generating” them and explain why this representational content is “nowhere to be found”. I will end with a reconsideration of the conceivability of zombies.
Article
Full-text available
A meta-learning algorithm, conventionally used for visual recognition, was applied to the recognition and classification of aroma oils. A printable chemiresistive sensor array was fabricated, based on composites of carbon black with various active materials. Standard aromatherapy kits with 30 types of essential oils were used as targets in an odor sensing experiment. Benefiting from the pattern recognition ability of the fabricated sensor array, a high-quality dataset was obtained with 30 aroma oil classes, in which each class had nine replicate samples. A deep metric learning model, based on a Siamese neural network and a multilayer perceptron, was used to perform the N-way k-shot meta-learning. A test accuracy of over 98.7% was obtained for 31-way 9-shot learning, on discriminating whether the input pair samples were taken from similar or dissimilar classes. The model was effective in extracting meta-features of the aroma oils; this was proved by the improved clustering effect of samples in the spaces of principal components analysis and t-distributed stochastic neighbor embedding. The 30 aroma oils were divided into two datasets according to 6-fold cross-validation: 25 aroma oil classes (plus one blank class) as seen classes for constructing 26-way 9-shot learning models and the remaining five aroma oils as unseen classes for prediction. Average accuracies of 93.5% and 93.9% were achieved for recognition of the unseen aroma oils from the seen classes and classification of the unseen aroma oils themselves, respectively, demonstrating the effectiveness of the developed sensor and model for odor recognition and classification.
Article
Full-text available
Background: Diminished sense of smell impairs the quality of life but olfactorily disabled people are hardly considered in measures of disability inclusion. We aimed to stratify perceptual characteristics and odors according to the extent to which they are perceived differently with reduced sense of smell, as a possible basis for creating olfactory experiences that are enjoyed in a similar way by subjects with normal or impaired olfactory function. Methods: In 146 subjects with normal or reduced olfactory function, perceptual characteristics (edibility, intensity, irritation, temperature, familiarity, hedonics, painfulness) were tested for four sets of 10 different odors each. Data were analyzed with (i) a projection based on principal component analysis and (ii) the training of a machine-learning algorithm in a 1000-fold cross-validated setting to distinguish between olfactory diagnosis based on odor property ratings. Results: Both analytical approaches identified perceived intensity and familiarity with the odor as discriminating characteristics between olfactory diagnoses, while evoked pain sensation and perceived temperature were not discriminating, followed by edibility. Two disjoint sets of odors were identified, i.e., d = 4 "discriminating odors" with respect to olfactory diagnosis, including cis-3-hexenol, methyl salicylate, 1-butanol and cineole, and d = 7 "non-discriminating odors", including benzyl acetate, heptanal, 4-ethyl-octanoic acid, methional, isobutyric acid, 4-decanolide and p-cresol. Conclusions: Different weightings of the perceptual properties of odors with normal or reduced sense of smell indicate possibilities to create sensory experiences such as food, meals or scents that by emphasizing trigeminal perceptions can be enjoyed by both normosmic and hyposmic individuals.
Article
Full-text available
Carbon chain length in several classes of straight-chain aliphatic odorants has been proposed as a model axis of similarity for olfactory research, on the basis of successes of studies in insect and vertebrate species. To assess the influence of task on measured perceptual similarities among odorants and to demonstrate that the systematic similarities observed within homologous odorant series are not task specific, the authors compare 3 different behavioral paradigms for rats (olfactory habituation, generalization, and discrimination). Although overall patterns of odorant similarity are consistent across all 3 of these paradigms, both quantitative measurements of perceptual similarity and comparability with 2-deoxyglucose imaging data from the olfactory bulb are dependent on the specific behavioral tasks used. Thus, behavioral indices of perceptual similarity are affected by task parameters such as learning and reward associations.
Chapter
For most people, olfaction is a sense which is often ignored or used unconsciously during day-to-day activities. Yet it is a remarkably powerful sense, providing us with the means to remotely monitor our chemical environment. The main components of olfactory perception (odor quality, odor intensity, and hedonic value) are all active areas of research in olfactory behavior and psychophysics. Nevertheless, the nature and extent of olfactory experience remain elusive, and this especially hampers efforts to approach olfaction computationally.
Article
The human nose is often considered something of a luxury, but in the rest of the animal world, from bacteria to mammals, detecting chemicals in the environment has been critical to the successful organism. An indication of the importance of olfactory systems is the ...
Article
No model exists in olfaction which strictly relates quantitative measures of olfactory quality with quantitative physicochemical measures. Although each of the more notable olfactory theories makes useful suggestions, none provides a basis for developing a complete model which could accurately order all olfactory stimuli.
Article
ChemInform is a weekly Abstracting Service, delivering concise information at a glance that was extracted from about 200 leading journals. To access a ChemInform Abstract, please click on HTML or PDF.