This article is licensed under CC-BY 4.

pubs.acs.org/acssensors Perspective

Food for Thought: Optical Sensor Arrays and Machine Learning for
the Food and Beverage Industry
William J Peveler*

Cite This: ACS Sens. 2024, 9, 1656−1665 Read Online

ACCESS Metrics & More Article Recommendations

See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

ABSTRACT: Arrays of cross-reactive sensors, combined with statistical or

machine learning analysis of their multivariate outputs, have enabled the
holistic analysis of complex samples in biomedicine, environmental science,
Downloaded via UNIV DE CORDOBA on July 31, 2024 at 09:29:38 (UTC).

and consumer products. Comparisons are frequently made to the mammalian

nose or tongue and this perspective examines the role of sensing arrays in
analyzing food and beverages for quality, veracity, and safety. I focus on optical
sensor arrays as low-cost, easy-to-measure tools for use in the field, on the
factory floor, or even by the consumer. Novel materials and approaches are
highlighted and challenges in the research field are discussed, including sample
processing/handling and access to significant sample sets to train and test
arrays to tackle real issues in the industry. Finally, I examine whether the
comparison of sensing arrays to noses and tongues is helpful in an industry
defined by human taste.
KEYWORDS: sensing array, cross-reactive, electronic nose, machine learning, food, beverages, smell, taste

By way of example, Uhriń and co-workers have specialized in

T he food and beverage industries are a complex, multi-
trillion-dollar network of raw materials, semifinished, and
consumer products that spans hugely differing values from the
the direct NMR spectroscopy and mass spectrometry measures
of Scotch whisky demonstrating methods for congener (flavor
simplest daily staples up to the priciest Veblen goods. Within molecule) elucidation,3,4 classification against flavor and
this vast industry, sensing and measurement is crucial to monitor production method,5 and method development in high
materials safety (e.g., to check for spoilage or contamination), to resolution mass spectrometry to better analyze congeners and
assess quality and integrity (if the material is what it claims to be better understand product appearance and stability.6
and if it is from where it claims to be from) and to lead product Untargeted liquid chromatography/mass spectrometry meth-
design (taste, texture, appearance, longevity). ods have been used to analyze off-flavors in coffee beans against
Typically, the need for metrology increases with increasing professional perception scores (here, the Specialty Coffee
product value (risk of fraud), increased material/product Association cup scores). Peterson and co-workers were able to
heterogeneity (trying to “normalize” batches of a product), or use a machine learning model (orthogonal projection to latent
where human factors (consumer preference, nutritional value) structures OPLS regression, vide inf ra) to identify four key
and legal ramifications (safety requirements, allergens des- compounds (out of hundreds identified) that might negatively
ignated or protected status) are considered.1 However, sensing impact coffee flavor and quantify these in roasted and unroasted
in food and drinks can be incredibly complex. Simpler measures beans as a putative early indicator of quality.7
of color/optical density, pH, or mechanical properties are Other direct mass spectrometry methods such as matrix-
routine. But to understand the hugely complex matrix of assisted laser desorption ionization (MALDI) have been
chemistries contained in a food or beverage ingredient or demonstrated in fingerprinting the authenticity and origin of
product that contribute to flavor or spoilage, much more products, such as olive oil. Zambonin and co-workers measured
sophisticated methods are required. phospholipid profiles with MALDI mass spectrometry to
Techniques such as mass spectrometry (MS) linked with
chromatography (gas, liquid, supercritical fluid driven) or
nuclear magnetic resonance spectroscopy (NMR) can be used Received: February 1, 2024
to directly analyze as many of the chemical species in a sample as Revised: March 25, 2024
possible (“sensomics”), but given the large quantities of complex Accepted: March 29, 2024
spectrometric/spectroscopic data produced, the advent of easily Published: April 10, 2024
applied statistical and machine learning techniques (“chemo-
metrics”) has been crucial to make the most of the data.2
© 2024 The Author. Published by
American Chemical Society https://doi.org/10.1021/acssensors.4c00252
1656 ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

Figure 1. Overview of sensing arrays in food and beverage analysis. Sensing arrays offer distinct advantages in analyzing the sensomic profiles of food
and beverages in minimal format that can be included in packaging or used on the factory floor. Many different types of arrays with many different
outputs can be tailored to the application in hand and read out by a consumer or an operator monitoring a process. Sensors can be designed for quality
assurance (QA), safety and spoilage, or food fraud applications.

analyze the presence of hazelnut oil in adulterated extra virgin discussion to its use as a transducer in the read out of the second
olive oil samples,8 and Kuo et al. used MALDI to measure common approach, chemical sensing arrays.
triacylglycerol fingerprints in a wide variety of edible oil mixtures This second approach, the focus of this Perspective, is the
to classify and quantify adulterants.9 cross-reactive chemical sensing array (Figure 1). A group or
This is just a tiny sample of the work being done in this space, array of different sensors are exposed to a sample, and each
but while such leaps in matching detailed molecular structure sensor reacts with components of the sample to generate a
and concentration information (molecular fingerprints) with collective response.14 Samples are typically complex mixtures of
perception or veracity data with are hugely valuable, the barrier chemical compounds (e.g., a foodstuff or a beverage), and the
to access the required instrumental methods is high. With chemical reactivity can be one or any combination of adsorption
extensive set up and running costs and the need for highly to a surface, a redox process, a specific (bio)molecular
trained personnel, setting up these methods across production interaction, a supramolecular interaction, and so on. The key
sites or at the point-of-need can be challenging. So, the questions to this approach is that different parts of the array will react with
arising are “can simpler, lower-cost methods provide high quality different components within the mixture (cross-reactivity), and
chemical information on food and beverage samples?” and; “can measuring each element of the array holistically generates a
these approaches be applied at “point-of-need”, in goods-in, on sensing “fingerprint” for the mixture.
the factory floor, or at the point-of-sale, where there may only be
limited laboratory access or no laboratory at all?”
■ CROSS-REACTIVE SENSOR ARRAY
CONSTRUCTION, MATERIALS AND ANALYTICAL
■ ADDRESSING SENSING CHALLENGES IN THE
FOOD AND BEVERAGE INDUSTRY
OUTPUTS
The cross-reactive sensing approach has been applied widely across
Two approaches have been taken by researchers to try to address (bio)analytical chemistry, with the individual sensors in the arrays
these questions. The first approach is the use of direct constructed from reactive small molecules,15 macrocyclic or polymeric
supramolecular systems,16,17 nanoscale materials,18 or engineered
(miniaturized/portable) spectroscopy with diagnostic molec- biomacromolecules.19,20 To generate a response from the array to the
ular detection potential. Optical approaches often have a lower chemistry contained within the sample, a range of binding or bonding
setup and running costs, and are easier to shrink into a hand-held interactions can be probed, for example, dispersion forces, hydrogen
device. Examples include excitation emission spectroscopy, bonding, charge, or hydrophobic/hydrophilic interactions. Patterns of
Raman spectroscopy or Infrared spectroscopy (IR).10 In each receptors (e.g., polymers or nanoparticles with a particular repeat or
case, statistical or machine learning-enabled deconvolution of surface unit) or preorganized receptors (e.g., cavitands) can increase
the output spectra can be used to identify certain optically active specificity for elements in the array to key families of molecules in the
components within a sample or reference a sample more broadly sample.16 Chemical reactivity (redox chemistry, formation of dynamic
or stable dative or covalent bonds) can also be exploited, increasing
against a database for classification. The amount of chemical
array response and diversity, although the increased irreversibility of
“omic” information is lower than for MS-coupled methods but these reactions lend themselves best to a one-time-use sensor array.21
still high, and the approach is typically very rapid (minutes or The readout of the array can be achieved using a wide range of
seconds). There can still be a need for more expensive excitation optoelectronic transduction methods depending on the sensors used,
sources, cameras or detectors to get the most detailed including electrochemical voltammetry, amperometry or impedance; or
information, but rapid leaps in miniaturization are reducing optical absorbance, luminescence or vibrational spectroscopy.1 The
cost and making these approaches more popular in the QC output at each array element is combined into a multivariate pattern for
laboratory and even amenable to online or operando applications the sample, and many different samples can be measured with the same
(e.g., hyperspectral imaging built into a production line).11 The array to generate many different patterns based on their chemical
composition and reactivity. The similarities or differences between
ability to undertake remote or spatially offset (noninvasive) these patterns can then be interrogated using statistical or machine
testing can be a benefit, and suites of spectral analysis libraries learning tools, using the tools of chemometrics.22
and toolboxes are becoming available. A detailed discussion of The patterns derived from the sensors are largely analyzed with
the potential of direct portable optical testing can be found in computationally inexpensive linear transformation methods such as
several recent books and reviews,12,13 and here we will limit our Principal Component Analysis (PCA) to examine which elements in an

1657 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

Figure 2. A) RGB colorimetric responses of a 36-element reactive array, measured on a hand-held reader for various spirits after oxidation of their
headspace vapors. B) These patterns can be analyzed with PCA to identify factors such as Proof. Adapted with permission from Li & Suslick, ACS
Sensors. 2018, 3 (1), 121−127 Copyright 2018 American Chemical Society.40 C) Arrays of gold and aluminum nanoparticles on glass can be
multiplexed with orthogonal surface chemistries and nonoverlapping optical transmission spectra to create an array capable of discriminating spirits.
Adapted from Nanoscale 2019, 11 (32), 15216−15223 under a CC BY 3.0 DEED Unported license.33

array contribute to the similarities or differences between samples volatilize the foodstuff or beverage via heating or other means). For
observed (an unsupervised approach), or its close relation, Linear example, Swager and co-workers have classified cheeses and liquors,
Discriminant Analysis (LDA), optimized for classification with data among other foodstuffs/beverages, based on their volatile profiles, using
labels provided by the user (a supervised approach).22 Clustering arrays of CNTs decorated with 20 “selector” molecules designed to
analyses (e.g., unsupervised hierarchical clustering analysis, HCA) have increase interactions between the CNTs and the sulfur compounds,
also been widely applied and have the advantage of defining many levels alcohols, carbonyls, alkanes, and aromatic compounds in the target
of structure or similarity in the data beyond simple nearest neighbor samples. They analyzed the array outputs with nearest neighbor analysis
analysis. Regression is also increasingly valued, and methods such as (kNN) and random forests, with good success for cheese classification,
partial least-squares (or projection to latent structures) regression and moderate success for liquors and edible oils.28
(PLS) and orthogonal projections to latent structures (OPLS) are The other common approach is to create an optode array. These
increasingly applied to spectral data outputs.23,24 With the rise of arrays are constructed from different, optically active materials, that
increased computer power, and larger, more diverse data sets, there is respond to the molecules in a sample with an optical wavelength shift
also a growing move to more capable but perhaps less transparent and (change in color), or change in luminescent intensity (either steady-
more easily overfitted, supervised machine learning methods including state or lifetime changes).21 Chromophores can be modulated via
support vector machines (SVM), random forests and artificial neural change of local environment (e.g., displacement from a supramolecular
nets (ANNs).25 host by an analyte molecule),16,29 or chemical reactivity between
Arguably the earliest examples and certainly the highest TRL analytes and the array.14,25 Examples include ligation of metal-
(technology readiness level) sensing arrays applied in the food and loporphyrins to create a change in absorbance/reflectance;30 the
beverage industry are cross-reactive electrochemical gas sensors as reaction or borylated fluorophores with sugars to alter their
“chemical noses”.26 These sensors are largely based around adsorption luminescence;31 the change of local polarity and charge around
of volatile species to arrays of chemiresistive materials, with various environmentally responsive fluorophores such as coumarins, fluores-
chemical reactivities (acidic, basic, oxygen rich/poor) and filter layers cent polymers or fluorescent proteins to change their color and
added on top. The measurement of the resistivity across each array intensity;32 or the change of the size, shape or local refractive index
element comprises the sample fingerprint. These arrays originally made around a plasmonic nanoparticle to change their plasmonic color.33−35
use of metal oxide semiconductor materials, operating at high The sensor elements can be immobilized on a solid paper or glass
temperatures (several hundred degrees centigrade), but more recently substrate or suspended in solution before addition of the sample
arrays of modified carbonaceous materials such as carbon nanotubes (typically in gas or liquid form) to generate the optical response in each
(CNTs) have become popular thanks to successful operation at or close array element.
to room temperature (making them easier to build into portable The responses generated can be measured across the array by eye if
devices).27 Such sensors are best applied when there is a good set of the changes are clear and obvious enough, but more commonly, simple
volatiles available from the food or beverage (to avoid the need to cameras are used with red-green-blue (RGB) colorimetric analysis, or

1658 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

UV-visible-IR absorbance or luminescence spectroscopy, to measure array were collected on a hand-held reader and analyzed by
the output of each array element. The array format lends itself well to HCA and SVM to identify 14 spirits from around the world, with
high throughput read-out methods, such as wide-field and/or the outputs able to distinguish alcoholic strength (proof) as well
hyperspectral imaging, or well-plate-style, serialized readout formats.36 as sample dilution by as little as 1%.40 Suslick has pioneered this
The optical response of multiple luminescent sensor elements can be
highly multiplexed in a single location, further reducing the number of
style of gas-sensing optical array and previously demonstrated
array elements that need to be measured and the volume of sample the breadth of the approach, discriminating coffees, beers and
required (potentially down to a whole array in one few-μL well of a 384 many other foodstuffs.43,44
well plate).25,37 The spectroscopic methods required enable the use of Using an alternative approach featuring an array of duplexed
portable spectroscopy via simple illumination sources and basic lenses plasmonic elements of orthogonally functionalized Au and Al
and gratings for spectroscopic analysis with a CCD or PMT nanoparticles, Clark and co-workers have created a “plasmonic
photodetector, or even cellphone cameras1,38 tongue” (Figure 2C). Each plasmonic element generates an
The analysis of the optode signals generated by the array can be LSPR response that is modified by the local dielectric
achieved with the chemometric tools described above, with either point environment, and by coating the plasmonic elements with
color (RGB)/wavelength changes analyzed with supervised or
unsupervised discriminant or clustering methods or a full spectral
different chemistries, differential interactions with a sample can
analysis with partial least-squares methods. The “depth” of data versus be achieved. This red or blue shifts the plasmon response, giving
the number of different samples and groupings is worth considering cross-reactive responses from a measurement of the array
when choosing an analysis method, to ensure the method used is transmission spectra. The multiplexed nature of the array
suitable for the data acquired (and assumptions of the method are not increased the dimensionality, and discriminating power for age
violated, or overfitting does not occur).39 and style, when tested against seven whiskies as well as vodka
A final consideration when sensor arrays and statistical/machine and 40% ethanol solution.33
learning are applied to food and beverages is how the array and analyte Baijiu samples have been discriminated using a variety of
are placed in contact with each other. Natural volatiles can be sampled, different arrays based around reactive chromophore gener-
but if the material is a liquid or solid, it may require dissolution,
concentration, or dilution. Liquids can contain the whole sample,
ation,45 aggregation, growth or etching of silver and gold
including dissolved volatiles, and often have higher concentrations of nanoparticles,45−47 luminescent response of lanthanide con-
chemical analytes than those of vapors. Vapor sample concentration is taining metal organic frameworks,48 and the luminescence
entirely dependent on the relative vapor pressures of the analytes and quenching/enhancement of colloidal quantum dots. In each
the sampled headspace. However, vapors do typically present a “matrix case, the optical signals were analyzed with PCA, LDA, and
free” sample, whereas liquid samples can contain a large background of HCA and in one case a neural network to distinguish between 12
uninteresting solvent (e.g., water or alcohol) that dilutes and interferes and 22 brands of baijiu. In several cases the underlying chemistry
with analyte-sensor interactions. Additionally, liquids or vapors may of the sample impacting the array (e.g., caproic acid, butyric acid,
benefit from some kind of pretreatment to improve the sensing ethyl acetate etc.) was identified and studied in isolation at
response.40 For liquids in particular, (micro)fluidic delivery across a
relevant concentrations.45
surface enables many spatially separate sensor elements to continuously
respond to the sample at once, while minimizing the volume of analyte The spirits described above are often easily distinguished by
required, and is becoming a popular approach, alongside microwells.41 eye/smell/taste, so in many of these examples, and many more
besides,49 the need for the sensor array is justified by the need to

■ OPTICAL SENSOR ARRAYS FOR CHALLENGES IN

FOOD AND BEVERAGES
detect forgery or lower quality products from more expensive or
exclusive examples that might be mislabeled.42 However, in such
cases, if the chemistry sampled by the array is not specific
Classification and Forgery Detection. Many sensor enough to the principal differences between the different classes
arrays, combined with statistical learning, have been demon- of product (i.e. the chemical differences between brands, makers,
strated for distinguishing between brands, classes, or styles of a styles, or even batches, is greater than the difference between
particular food or beverage (which may or may not be high- or low-quality products), then the sheer scale of the
immediately obvious to the eye, nose, or tongue of the relatively pattern library required to identify all possible products may
well-educated taster/consumer). cause overlaps between “good” and “bad” products.
Spirits are a popular target, with many examples of arrays that Furthermore, in many works arrays are trained and tested
distinguish between whiskies, baijiu and other fermented and (clustered or discriminated) against the individual groups or
distilled beverages.42 The array chemistry should interact with production styles rather than examples of good or bad products,
molecules in the beverage under study to detect subtle limiting the proof of utility for forgery detection. While “quick
differences between brands or production styles. For example, and simple” antiforgery applications of sensing arrays in food
whisky is perhaps best distinguished via chemistry arising from and beverages are attractive, they are only viable where the need
its wood aging (tannins, polyphenols, lactones), whereas baijiu to detect forgery is justifiable and practicable: where there are
can be distinguished via the chemistry arising from the expensive (Veblen) goods and an extant forgery/black market;
fermentation method and grain used, prior to distillation (esters high batch to batch consistency of chemistry or clear chemical
and organic acids). differences between real and fake goods; and opportunities to
Li and Suslick demonstrated headspace analysis on various sample products on import/export/sale/consumption to
spirits, including whisky, bourbon, and brandy. The volatile actually find the forgeries. Thus, this is not the only area
alcohols and carbonyls in the headspace were passed over a where sensor array research should focus.
reactive array of 36 chromophore elements, with a partial Food and Beverage Quality Assurance. There are many
preoxidation step (Figure 2A,B). The elements were composed foodstuff and beverage production methods where the
of an array of pH, oxidation/reduction, and base/acid chemistry of the sample (and hence the taste and smell) is
responsive molecular complexes that change color in a altered via ingredients choice and product processing, and
differential fashion when exposed to various common functional sensing arrays provide a method for rapidly assessing the quality
groups. RGB chromatic shifts (before versus after) across the of input materials that may impact on the downstream process
1659 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

Figure 3. A) Grouped responses to three classes of tea (30 examples) by 14 different proteinaceous barrels (B) with a displaced indicator dye. C) SVM
analysis of the fluorescence changes across the array could be used to successfully identify the teas. Adapted Nat Commun 2023, 14 (1), 383 under a
CC BY 4.0 license.19 D) Large chromatic changes across RGB space by gold nanoparticles in response to off flavors in maple sap and syrup samples
enable an array of responses from a single sensor element. The simplicity of the test enables this to be carried out at the point-of-need in rural Canada.
Reproduced from Anal. Methods 2020, 12 (19), 2460−2468 with permission from the Royal Society of Chemistry.57 E) Structural color created by
arrays of functionalized bacteriophages is used to monitor the volatiles created by ripening bananas over 15 days. RGB extraction and HCA analysis can
be used to follow the process. Adapted from Kim et al. Sensors and Actuators B 2022, 362, 131763 copyright Elsevier (2022).58

1660 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

Figure 4. A) Pesticides can be detected and classified by preprocessing samples with a sensitive acetylcholine esterase (AChE) before exposing the
products to an array of five compounds that react calorimetrically to any H2O2 or thiocholine produced by the enzyme. Differential actions of the
enzyme and the array are analyzed by HCA to classify different pesticides in apple juice and tea. Adapted with permission from Anal. Chem. 2015, 87
(10), 5395−5400, Copyright 2015 American Chemical Society.62 B) Paper strips containing an array of indicator dyes within sealed meat and fish
packaging react to volatile compounds to signal spoilage. Analysis of the RGB images by PCA allowed for classification into “safe”, “warning”, and
“hazard” groups. Adapted from Foods 2020, 9 (5), 684 under a CC BY license.66 C) MOF-based colorimetric sensors for spoilage also react to emitted
volatile compounds, and kNN analysis of image color can be used to monitor freshness of milk and meat. Adapted from Adv. Mater. Interfaces 2023, 10
(28), 202300329 under a CC BY 4.0 DEED license.67 D) An alternative transduction approach to measuring array-sample interactions is to use surface
enhanced Raman spectroscopy on an array of surface modified gold nanopillars. The surface modification differentially alters the interaction between
surface and target, and therefore what Raman signals are observed before analysis with PCA and LDA. Adapted from Nat Commun 2020, 11 (1), 207
under a CC BY 4.0 license.68

or the success of processes.2 This need is greatest where the the nanoparticles formed and the rate of their formation,
processes are lengthy, expensive, or otherwise hard to monitor; collectively analyzed by HCA. Analysis of a semifinished
where they cause clear chemical changes in the sample; and product, a single cask sampled over time, allowed for matching
rapid go/no-go decisions might save time, effort, and money. of the developing chemistry with the sensor array output and
Areas where assurance might be useful (alongside anticounter- development of the sensor as a measure of cask age.52
feiting, vide supra), include rapidly assessing the nutrient content The quantity of antioxidant compounds, particularly
of raw ingredients (vitamins, minerals, antioxidants etc.);50 flavonoids, tannins, and other phenolic substances, are also
monitoring batch to batch variation in production inside sealed important in assessing the quality green and black teas against
containers (e.g., inside a cask or barrel); or assuring a product grading scales, as well as their putative health benefits. For
has met legal minima, such as minimum aging requirements.51 example Huo et al. used a version of Suslick’s color changing
Returning to the example of whisk(e)y, this is a product where arrays to grade and identify geographical origin of nine green
the major processing step is long aging in a closed wooden cask teas.53 Similarly, the high antioxidant content of a luxury tea
for a legally mandated minimum period (3 years in the case of (Tieguanyin) was leveraged in an array by Yang et al. to measure
Scotch Whisky, and often much longer) before blending (or different polyphenols in the tea with a peroxidase mimicking
vatting) of multiple casks together into a batch for retail. As metal organic particles and a TMB color changing output, and
different casks age at different rates, depending on the wood, discriminate genuine from adulterated samples.54
cask condition, and storage conditions, they impart different Many other examples of sensing teas exist using a versatile set
flavor compounds to the spirit. To monitor how the different of sensing array construction methods. Ni et al. recently
casks are aging, every cask could be taste tested, but over leveraged the reactivity of boronates with catechols to create an
hundreds of casks in a warehouse or rickhouse, this takes a long indicator displacement assay consisting of combinations of two
time. It is also often impractical or even unsafe to have a local GC indicator fluorophores and three multidentate binder/quench-
or similar tool. Sensing arrays can give a quick holistic ers for various plant derived polyphenols in tea. Various
impression of a cask based on the reducing chemical content statistical or ML techniques were applied to the data with LDA
in the aging whisky that stems from the wood contact, including found to be the most successful in discriminating the 16 teas
organic acids, polyphenols, furfurals etc. We recently demon- under test.55 Bunz and co-workers used a library of conjugated
strated a multidimensional sensor that consists of reacting gold polymers, some quenched by macrocyclic and cucurbit[8]uril to
or silver salts with different whiskies to create plasmonic measure amino acids or xanthines in teas. Molecules such as
nanoparticles as fingerprints for the whisky based on the color of caffeine and theobromine interacted with the polymers and
1661 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

macrocycle to trigger a differential turn on fluorescent response and Au NPs that aggregated and changed their plasmonic color
that could separate the 22 different teas.56 Finally, Woolfson and in the presence of biogenic amines in extracted meat samples
co-workers utilized an array of 14 different coiled coil peptide and would be similarly readable with a portable spectrometer.34
barrels and an environmentally responsive dye (1,6-diphenyl- Unpowered, solid-state optical arrays that can be read with a
1,3,5-hexatriene) to create differential fluorescent responses for camera or by eye offer a huge advantage. They can be
30 teas in three classes (earl gray vs black tea vs green tea) via a incorporated within packaging as externally readable sensors
support vector classifier (Figure 3A-C).19 While these later to indicate if the contents have spoiled. Arrays of reactive color-
examples focus mostly on discriminating green from black tea, changing ink spots responsive to pH (volatile organic acids and
oolong tea, or other obviously different teas, they do highlight amines) and thiols have been exploited by Magnaghi et al. for
how versatile chemical approaches, particularly using host− defining “safe” “warning” or “hazard” categorizations to meat
guest or dynamic covalent interactions can be combined and and fish, that had been left at room temperature, from within the
applied to create sensing arrays for the same target chemistries. packet (Figure 4B). RGB analysis on smartphone camera images
Maple syrup is an expensive product thanks to the remoteness coupled with PCA could warn of spoilage, although these
of the raw materials (in the forests of North America) and the categorizations were not externally defined by other gold
long processing (boiling) step to concentrate tree sap into the standard testing or microbiology.66 Heinke and co-workers
final sugary syrup. Sap quality is hugely important because any developed an array of thin metal organic framework (MOF)
contaminants or “off-flavors” in the raw sap are unavoidably films that formed Fabry-Pérot cavities when oversputtered with
concentrated into the final syrup, lowering the quality or spoiling metal (Figure 4C). The porous cavity filling between metal
the batch. Compounding the difficulties, sap is harvested and layers means gases entering the MOF material change the
processed in large volumes far from laboratories, so a point-of- observed color of the cavity, allowing for colorimetric gas
need testing solution is useful for assessing sap quality before sensing based on the gases present and their interaction with the
batching and boiling. Masson and co-workers used a cross- varied MOF material. Read out was possible with a smartphone
reactive gold-nanoparticle aggregation assay to assess the off- camera, RGB analysis and kNN clustering, and the devices were
note content of sap and finished syrup. Crucially the simple assay used to measure the spoilage of milk and meat samples inside
has a multicolor read out (Figure 3D, so could be considered a containers.67
single element array) and can be performed at the point of need
in the sugar shacks in rural Quebec. The sample amino acid
content was identified as the key consideration for syrup quality
■ THE FUTURE OF SENSOR ARRAYS FOR FOOD AND
BEVERAGES?
and could be measured by the red shifting of gold nanoparticles In the selection of examples above, I have tried to illustrate the
mixed with sap, using an end-point colorimetric index − possibilities that chemical sensing arrays, combined with
COLORI, combined with a mixed effects statistical model to statistical or machine learning can offer sensing in the food
predict likely syrup quality, based on assay data from over 29,000 and beverage industry, tackling challenges in quality assurance,
sap and syrup samples.57,59 production, and safety. New materials and approaches to sensing
A final example of quality assurance where arrays have been in liquids and gases offer a wealth of cross-reactive or targeted
applied is in monitoring fruit ripening. Many volatiles are arrays, and a problem-driven approach is key to making useful
released during fruit ripening (including ethylene, 2-pentanone, progress for the industry.
and 3-methyl-1-butanol) and need to be monitored in transit to Arrays combined with hand-held/portable electronics and
avoid spoilage. Oh and colleagues have pioneered colorimetric optics are far more suited to point of need testing than the gold-
sensing arrays created from bacterial phages bearing distinct standard omic techniques such as LC, GC, MS and NMR.
surface peptides, that self-assemble into microscale architectures Although arrays cannot deliver the same untargeted molecular
and display structural color.20 The interaction between different precision as these methods, they can be easily tuned for the
phage-based materials and various gases causes material swelling samples if the underlying chemical content is taken into
and changes the observed iridescence.60 Five differently consideration when designing and constructing the sensing
functionalized phage materials were exposed to fruit ripening elements. A move to embracing optical arrays allows for simple
gases and ripening bananas, with their color changes monitored analysis using hand-held devices, including smartphones, and
by camera and processed to RGB shifts, to successfully follow the computational power required to collect and analyze array
the ripening process via HCA analysis (Figure 3E).58 data against a pretrained model or a library of preclassified
Food and Beverage Safety. A final exemplar area where samples is well within the capabilities of such devices for more
sensing arrays can be usefully applied is food and beverage simple linear methods such as LDA. Even if more computa-
consumer safety. Arrays have been used to detect toxic tionally expensive image/video analysis is required or the data
contaminants such as pesticide residues or heavy metals (Figure load is very high (many samples or time points), calculations can
4A),61,62 illegally introduced dopants such as melamine in infant be performed via remote data services.
milk,63 or food-borne pathogens.64 A current challenge is that many arrays, from the original “e-
Bacterial growth and spoilage of food items such as meat, fish, noses” to the latest plasmonic sensor arrays, rely on a differential
and dairy poses a high risk to consumers. Markers of spoilage can adsorption or an increase in local molecular mass at each array
include biogenic amines such as spermine, organic acids, or element to generate the signals. This can lead to relatively low
thiols. To detect biogenic amines, arrays comprising metal “orthogonality” in the data generated, limiting the discrim-
complexes that change their UV-visible spectra have been shown inatory power of the array as the number of different samples
to be effective. Singh et al. built a liquid array and a portable increases and the similarity between them increases. Chemically
microplate reader that could detect spermine and tryptamine reactive arrays that have diverse responses to different chemical
contamination in meat and cheese, although sample preparation moieties are a counter example, but are inherently “one-time”
was laborious.65 In a similar fashion Hormozi-Nezhad and co- use, which can be a limitation in certain circumstances.
workers built a liquid array based on eight surface modified Ag Fluorophores that respond in color and intensity to different
1662 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

charge, solvation, and bonding environments can also increase detecting aging fruit, taking this biomimetic approach to a new
orthogonality, but weaker binding interactions require higher extreme.71
concentrations of target to generate an effect. A sensing array that is well aligned to human tastes and
New materials being used to construct sensing arrays allow for preferences could be invaluable in taste testing new products,
new modes of operation that may add orthogonality to the array aligning the taste of products with the particular taste of a
outputs. Steady state luminescence has been thoroughly distinct population, and measuring the consistency of the flavor
exploited, but lower cost integrated single photon avalanche of a batch produced product when then raw materials are subject
detector (SPAD) arrays have the potential to make lumines- to change. Many of these measures are currently achieved by
cence lifetime imaging more easily available, multiplying the human tasting panels or skilled individuals (master blenders at a
outputs of suitably designed luminescent array that varies in distillery, for example), and there is a degree of reticence in many
intensity, emission color, and lifetime(s). Established plasmonic parts of the industry that these could or should ever be replaced.
materials (e.g., gold nanoparticles) and newer Raman-active 2D And perhaps they cannot; after all, perception of food and drink
materials (e.g., MXenes) open up the potential for surface is so subjective there is always the need for the human element.
enhanced Raman arrays where surface chemistry interacts with However, a versatile technology that is well aligned to taste and
molecular targets to not only give a Raman spectrum of the smell, and can be tailored to the product under test, can make
target but alter the enhanced spectrum of the surface coating, taster’s lives easier. It could enable optimization and
giving rise to a highly multidimensional output from the array parallelization, particularly when working with very strong
with potential for “fingerprinting” as well as a degree of direct flavors, semifinished products (that do not yet taste “good”), or
detection via enhancements in key Raman regions.68 when working with potential toxins where humans cannot
Array stability, whether the shelf life of a one-time use array or operate. It is in this space that I propose sensing arrays and
long-term regenerability of a reusable array, is also a challenge. machine learning techniques will really impact industry in the
Library collection and statistical training is only valid if the array future.
and its outputs are reproducible over time, and so simple
construction and built-in standardization (either unchanging
elements, measurement of standards, or pre/post exposure
■ AUTHOR INFORMATION
Corresponding Author
measurements) are useful to ensure usable data is collected. This William J Peveler − School of Chemistry, Joseph Black Building,
is particularly relevant in the food and beverage industries, where University of Glasgow, Glasgow G128QQ, U.K.; orcid.org/
in-line testing is attractive for many processes, so arrays will have 0000-0002-9829-2683; Email: william.peveler@
to remain useable after extended periods in potentially harsh glasgow.ac.uk
environments without degrading or fouling. Much can be
Complete contact information is available at:
learned from the application of arrays in medical diagnostics,69
https://pubs.acs.org/10.1021/acssensors.4c00252
however the number of assays required is greater and budgets
are typically tighter in food production, so cost and reusability is
Notes
a major consideration.
The author declares no competing financial interest.
To better train arrays for current problems or challenges in the
industry, access to relevant and ideally large and varied sample
sets is key. This is particularly true for raw ingredients or
semifinished products, or spoiled or otherwise contaminated
■ ACKNOWLEDGMENTS
The author gratefully acknowledges the University of Glasgow
products that would be hard to collect or mimic/spike without for funding and thanks K. Suslick and J-F. Masson for the useful
stakeholder input. Such samples will enable researchers to go discussions. For the purpose of open access, the author has
beyond the simple classification of finished consumer products applied a Creative Commons Attribution (CC BY) license to
bought in the local store, and begin to tackle more complex any Author Accepted Manuscript version arising from this
challenges, as exemplified well by Masson’s work with the maple submission.
syrup industry in Quebec.59
Given the analogy between chemical sensor arrays with
machine learning, and the mammalian olfaction system perhaps
■ REFERENCES
(1) Dincer, C.; Bruch, R.; Costa-Rama, E.; Fernández-Abedul, M. T.;
the most exciting and challenging application in food and Merkoçi, A.; Manz, A.; Urban, G. A.; Güder, F. Disposable Sensors in
beverages would be to generate sensors that can truly mimic Diagnostics, Food, and Environmental Monitoring. Adv. Mater. 2019,
31 (30), e1806739.
human smell and taste.70 Human noses do not operate like a GC (2) Granvogl, M.; Schieberle, P. The Sensomics Approach: A Useful
− we simply cannot distinguish that many different but similar Tool to Unravel the Genuine Aroma Blueprint of Foods and Aroma
compounds, and many of the compounds detected are not Changes during Food Processing. Compr. Anal. Chem. 2022, 96, 41−68.
necessarily the compounds a human would use to “taste” or (3) Kew, W.; Bell, N. G. A.; Goodall, I.; Uhrín, D. Advanced Solvent
distinguish a food product or beverage. By relying on cross- Signal Suppression for the Acquisition of 1D and 2D NMR Spectra of
reactivity, identifying key families or members of families of Scotch Whisky. Magn. Reson. Chem. 2017, 55 (9), 785−796.
compounds, and linking these to common descriptors, we can (4) Stockwell, M.; Goodall, I.; Uhrín, D. Quantification of Whisky
holistically sample and distinguish complex mixtures just as a Congeners by 1H NMR Spectroscopy. Anal Sci. Adv. 2020, 1 (2), 132−
sensing array can. So can we (and should we) try to align the 140.
(5) Kew, W.; Goodall, I.; Uhrín, D. Analysis of Scotch Whisky by 1H
natural and the artificial? Recent work by Fan and co-workers
NMR and Chemometrics Yields Insight into Its Complex Chemistry.
demonstrates a massively increased number of cross-reactive Food Chem. 2019, 298, 125052.
elements that can be electrochemically surveyed on a chip from (6) Kew, W.; Mackay, C. L.; Goodall, I.; Clarke, D. J.; Uhrín, D.
tens to hundreds or even thousands. Combining this this with a Complementary Ionization Techniques for the Analysis of Scotch
neural network and computer vision they created their version of Whisky by High Resolution Mass Spectrometry. Anal. Chem. 2018, 90
a “robot dog”, capable of sniffing out different foodstuffs and (19), 11265−11272.

1663 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

(7) Sittipod, S.; Schwartz, E.; Paravisini, L.; Tello, E.; Peterson, D. G. (26) Schaller, E.; Bosset, J. O.; Escher, F. ‘Electronic Noses’ and Their
Identification of Compounds That Negatively Impact Coffee Flavor Application to Food. LWT - Food Sci. Technol. 1998, 31 (4), 305−316.
Quality Using Untargeted Liquid Chromatography/Mass Spectrome- (27) Khatib, M.; Haick, H. Sensors for Volatile Organic Compounds.
try Analysis. J. Agric. Food Chem. 2020, 68 (38), 10424−10431. ACS Nano 2022, 16, 7080.
(8) Calvano, C. D.; Ceglie, C. D.; D’Accolti, L.; Zambonin, C. G. (28) Schroeder, V.; Evans, E. D.; Wu, Y.-C. M.; Voll, C.-C. A.;
MALDI-TOF Mass Spectrometry Detection of Extra-Virgin Olive Oil McDonald, B. R.; Savagatrup, S.; Swager, T. M. Chemiresistive Sensor
Adulteration with Hazelnut Oil by Analysis of Phospholipids Using an Array and Machine Learning Classification of Food. ACS Sens. 2019, 4
Ionic Liquid as Matrix and Extraction Solvent. Food Chem. 2012, 134 (8), 2101−2108.
(2), 1192−1198. (29) Wu, D.; Sedgwick, A. C.; Gunnlaugsson, T.; Akkaya, E. U.; Yoon,
(9) Kuo, T.-H.; Kuei, M.-S.; Hsiao, Y.; Chung, H.-H.; Hsu, C.-C.; J.; James, T. D. Fluorescent Chemosensors: The Past, Present and
Chen, H.-J. Matrix-Assisted Laser Desorption/Ionization Mass Future. Chem. Soc. Rev. 2017, 46 (23), 7105−7123.
Spectrometry Typings of Edible Oils through Spectral Networking of (30) Rakow, N. A.; Suslick, K. S. A Colorimetric Sensor Array for
Triacylglycerol Fingerprints. ACS Omega 2019, 4 (13), 15734−15741. Odour Visualization. Nature 2000, 406 (6797), 710−713.
(10) Mishra, M. Spectroscopic Techniques for the Analysis of Food (31) Pushina, M.; Penavic, A.; Farshbaf, S.; Anzenbacher, P.
Quality, Chemistry, and Function. In Advanced Spectroscopic Techniques Fluorescent Sensor Array for Quantitative Determination of Saccha-
for Food Quality; Shukla, A. K., Ed.; Royal Society of Chemistry, 2022; rides. ACS Sensors 2021, 6 (11), 4001−4008.
pp 1−22. (32) Han, J.; Ma, C.; Wang, B.; Bender, M.; Bojanowski, M.; Hergert,
(11) Qin, J.; Kim, M. S.; Chao, K.; Chan, D. E.; Delwiche, S. R.; Cho, M.; Seehafer, K.; Herrmann, A.; Bunz, U. H. F. A Hypothesis-Free
B.-K. Line-Scan Hyperspectral Imaging Techniques for Food Safety Sensor Array Discriminates Whiskies for Brand, Age, and Taste. Chem.
and Quality Applications. Appl. Sci. 2017, 7 (2), 125. 2017, 2 (6), 817−824.
(12) Portable Spectroscopy and Spectrometry; Crocombe, R., Leary, P., (33) Macias, G.; Sperling, J. R.; Peveler, W. J.; Burley, G. A.; Neale, S.
Kammrath, B., Eds.; Wiley, 2021. DOI: 10.1002/9781119636489 L.; Clark, A. W. Whisky Tasting Using a Bimetallic Nanoplasmonic
(13) McVey, C.; Elliott, C. T.; Cannavan, A.; Kelly, S. D.; Tongue. Nanoscale 2019, 11 (32), 15216−15223.
Petchkongkaew, A.; Haughey, S. A. Portable Spectroscopy for High (34) Abbasi-Moayed, S.; Orouji, A.; Hormozi-Nezhad, M. R.
Throughput Food Authenticity Screening: Advancements in Technol- Multiplex Detection of Biogenic Amines for Meat Freshness
ogy and Integration into Digital Traceability Systems. Trends Food Sci. Monitoring Using Nanoplasmonic Colorimetric Sensor Array.
Technol. 2021, 118, 777−790. Biosensors 2023, 13 (8), 803.
(14) Peveler, W. J.; Yazdani, M.; Rotello, V. M. Selectivity and (35) Bigdeli, A.; Ghasemi, F.; Golmohammadi, H.; Abbasi-Moayed,
Specificity: Pros and Cons in Sensing. ACS Sensors 2016, 1 (11), 1282− S.; Nejad, M. A. F.; Fahimi-Kashani, N.; Jafarinejad, S.; Shahrajabian,
1285. M.; Hormozi-Nezhad, M. R. Nanoparticle-Based Optical Sensor
(15) Motiei, L.; Margulies, D. Molecules That Generate Fingerprints:
Arrays. Nanoscale 2017, 9 (43), 16546−16563.
A New Class of Fluorescent Sensors for Chemical Biology, Medical
(36) Lu, X.; Suslick, K. S.; Li, Z. Nanoparticle Optical Sensor Arrays:
Diagnosis, and Cryptography. Acc. Chem. Res. 2023, 56, 1803.
Gas Sensing and Biomedical Diagnosis. Anal. Sens. 2023, 3 (2).
(16) Sedgwick, A. C.; Brewster, J. T.; Wu, T.; Feng, X.; Bull, S. D.;
DOI: 10.1002/anse.202200050.
Qian, X.; Sessler, J. L.; James, T. D.; Anslyn, E. V.; Sun, X. Indicator
(37) Peveler, W. J.; Algar, W. R. More Than a Light Switch:
Displacement Assays (IDAs): The Past, Present and Future. Chem. Soc.
Engineering Unconventional Fluorescent Configurations for Biological
Rev. 2021, 50 (1), 9−38.
Sensing. ACS Chem. Biol. 2018, 13 (7), 1752−1766.
(17) Tian, J.-H.; Hu, X.-Y.; Hu, Z.-Y.; Tian, H.-W.; Li, J.-J.; Pan, Y.-C.;
(38) Peveler, W. J.; Algar, W. R. Toward Clinical Applications of
Li, H.-B.; Guo, D.-S. A Facile Way to Construct Sensor Array Library
via Supramolecular Chemistry for Discriminating Complex Systems. Smartphone Spectroscopy and Imaging. Portable Spectroscopy and
Nat. Commun. 2022, 13 (1), 4293. Spectrometry 2021, 199−226.
(18) Behera, P.; De, M. Surface-Engineered Nanomaterials for Optical (39) Li, Z.; Fang, M.; LaGasse, M. K.; Askim, J. R.; Suslick, K. S.
Array Based Sensing. ChemPlusChem. 2023, e202300610. Colorimetric Recognition of Aldehydes and Ketones. Angew. Chem., Int.
(19) Dawson, W. M.; Shelley, K. L.; Fletcher, J. M.; Scott, D. A.; Ed. 2017, 56 (33), 9860−9863.
Lombardi, L.; Rhys, G. G.; LaGambina, T. J.; Obst, U.; Burton, A. J.; (40) Li, Z.; Suslick, K. S. A Hand-Held Optoelectronic Nose for the
Cross, J. A.; Davies, G.; Martin, F. J. O.; Wiseman, F. J.; Brady, R. L.; Identification of Liquors. ACS Sensors 2018, 3 (1), 121−127.
Tew, D.; Wood, C. W.; Woolfson, D. N. Differential Sensing with (41) Sperling, J. R.; Poursat, B. A. J.; Savage, L.; Christie, I.; Cuthill, C.;
Arrays of de Novo Designed Peptide Assemblies. Nat. Commun. 2023, Aekbote, B.; McGuire, K.; Karimullah, A. S.; Robbie, J.; Sloan, W. T.;
14 (1), 383. Gauchotte-Lindsay, C.; Peveler, W. J.; Clark, A. W. A Cross-Reactive
(20) Oh, J.-W.; Chung, W.-J.; Heo, K.; Jin, H.-E.; Lee, B. Y.; Wang, E.; Plasmonic Sensing Array for Drinking Water Assessment. Environ. Sci.:
Zueger, C.; Wong, W.; Meyer, J.; Kim, C.; Lee, S.-Y.; Kim, W.-G.; Nano 2023, 10 (12), 3500−3508.
Zemla, M.; Auer, M.; Hexemer, A.; Lee, S.-W. Biomimetic Virus-Based (42) Yu, Y.; Shi, F.; Zhang, Y.; Li, F.; Han, J. Optical Sensor Array for
Colourimetric Sensors. Nat. Commun. 2014, 5, 3043. the Discrimination of Liquors. J. Futur. Foods 2024, 4 (1), 48−60.
(21) Li, Z.; Askim, J. R.; Suslick, K. S. The Optoelectronic Nose: (43) Suslick, B. A.; Feng, L.; Suslick, K. S. Discrimination of Complex
Colorimetric and Fluorometric Sensor Arrays. Chem. Rev. 2019, 119 Mixtures by a Colorimetric Sensor Array: Coffee Aromas. Anal. Chem.
(1), 231−292. 2010, 82 (5), 2067−2073.
(22) Stewart, S.; Ivy, M. A.; Anslyn, E. V. The Use of Principal (44) Zhang, C.; Bailey, D. P.; Suslick, K. S. Colorimetric Sensor Arrays
Component Analysis and Discriminant Analysis in Differential Sensing for the Analysis of Beers: A Feasibility Study. J. Agric. Food Chem. 2006,
Routines. Chem. Soc. Rev. 2014, 43 (1), 70−84. 54 (14), 4925−4931.
(23) Oliveri, P.; Malegori, C.; Casale, M. Chemometrics: multivariate (45) Chen, Y.; Wang, S.; Hu, Y.; Chen, H.; Fu, H.; Zhou, C.; She, Y.
analysis of chemical data. In Chemical Analysis of Food, 2nd ed; Pico, Y. Novel Colorimetric Sensor Array for Identification of Baijiu Using
Academic Press, 2020; pp 33−76. Color Reactions of Flavor Compounds. Microchem. J. 2021, 167,
(24) Analytical Methods Committee, AMCTB No. 100. Multivariate 106277.
Statistics in the Analytical Laboratory (1): An Introduction. Anal. (46) Zhang, J.; Huang, R.; Zhang, J.; Zhou, J.; Qin, H.; Yang, P.; Hou,
Methods 2020, 13 (2), 274−277. C.; Huo, D. A Novel Colorimetric Sensor Array for Baijiu
(25) Fargher, H. A.; d’Oelsnitz, S.; Diaz, D. J.; Anslyn, E. V. Pushing Discrimination Based on Gold Triangular Nanoplates Etching. Sens.
Differential Sensing Further: The Next Steps in Design and Analysis of Actuators B: Chem. 2021, 337, 129789.
Bio-Inspired Cross-Reactive Arrays. Anal. Sens. 2023, DOI: 10.1002/ (47) Jia, J.; Zhang, S.; Ma, L.; Wang, S.; Shen, C.; She, Y. Gold
anse.202200095. Nanobipyramid Colorimetric Sensing Array for the Differentiation of

1664 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665
ACS Sensors pubs.acs.org/acssensors Perspective

Strong Aroma-Type Baijiu with Different Geographical Origins. Food (67) Zhan, K.; Jiang, Y.; Qin, P.; Chen, Y.; Heinke, L. A Colorimetric
Chem. 2024, 432, 137197. Label-Free Sensor Array of Metal−Organic-Framework-Based Fabry−
(48) Liu, X.; Zhang, J.; Ma, Y.; Zhao, D.; Huo, D.; Luo, H.; Li, J.; Luo, Pérot Films for Detecting Volatile Organic Compounds and Food
X.; Hou, C. One Minimalist Fluorescence MOF Sensor Array for Baijiu Spoilage. Adv. Mater. Interfaces 2023, 10 (28), 202300329.
Discrimination. Anal. Methods 2022, 14, 1585. (68) Kim, N.; Thomas, M. R.; Bergholt, M. S.; Pence, I. J.; Seong, H.;
(49) Han, J.; Bender, M.; Seehafer, K.; Bunz, U. H. F. Identification of Charchar, P.; Todorova, N.; Nagelkerke, A.; Belessiotis-Richards, A.;
White Wines by Using Two Oppositely Charged Poly(P- Payne, D. J.; Gelmi, A.; Yarovsky, I.; Stevens, M. M. Surface Enhanced
phenyleneethynylene)s Individually and in Complex. Angew. Chem., Raman Scattering Artificial Nose for High Dimensionality Finger-
Int. Ed. 2016, 55 (27), 7689−7692. printing. Nat. Commun. 2020, 11 (1), 207.
(50) Yang, M.; Zhang, M.; Jia, M. Optical Sensor Arrays for the (69) Geng, Y.; Peveler, W. J.; Rotello, V. M. Array-Based “Chemical
Detection and Discrimination of Natural Products. Nat. Prod. Rep. Nose” Sensing in Diagnostics and Drug Discovery. Angew. Chem., Int.
2023, 40 (3), 628−645. Ed. 2019, 58 (16), 5190−5200.
(51) Kumar, A.; Castro, M.; Feller, J.-F. Review on Sensor Array- (70) Wilson, A. D.; Baietto, M. Applications and Advances in
Based Analytical Technologies for Quality Control of Food and Electronic-Nose Technologies. Sensors 2009, 9 (7), 5099−5148.
Beverages. Sensors 2023, 23 (8), 4017. (71) Wang, C.; Chen, Z.; Chan, C. L. J.; Wan, Z.; Ye, W.; Tang, W.;
(52) Gracie, J.; Zamberlan, F.; Andrews, I. B.; Smith, B. O.; Peveler, W. Ma, Z.; Ren, B.; Zhang, D.; Song, Z.; Ding, Y.; Long, Z.; Poddar, S.;
J. Growth of Plasmonic Nanoparticles for Aging Cask-Matured Whisky. Zhang, W.; Wan, Z.; Xue, F.; Ma, S.; Zhou, Q.; Lu, G.; Liu, K.; Fan, Z.
ACS Appl. Nano Mater. 2022, 5 (10), 15362−15368. Biomimetic Olfactory Chips Based on Large-Scale Monolithically
(53) Huo, D.; Wu, Y.; Yang, M.; Fa, H.; Luo, X.; Hou, C. Integrated Nanotube Sensor Arrays. Nat. Electron. 2024, 7, 157−167.
Discrimination of Chinese Green Tea According to Varieties and
Grade Levels Using Artificial Nose and Tongue Based on Colorimetric
Sensor Arrays. Food Chem. 2014, 145, 639−645.
(54) Yang, X.; Bi, Z.; Yin, C.; Zhang, S.; Song, D.; Huang, H.; Li, Y. A
Colorimetric Sensor Array Based on Peroxidase Activity Nanozyme for
the Highly Efficient Differential Sensing of Tea Polyphenols and
Tieguanyin Adulteration. Food Chem. 2024, 432, 137265.
(55) Ni, W.; Yu, Y.; Gao, X.; Han, Y.; Zhang, W.; Zhang, Z.; Xiao, W.;
Hu, Q.; Zhang, Y.; Huang, H.; Li, F.; Chen, M.; Han, J. Multilocus
Distance-Regulated Sensor Array for Recognition of Polyphenols via
Machine Learning and Indicator Displacement Assay. Anal. Chem.
2024, 96, 301.
(56) Wang, B.; Han, J.; Bender, M.; Hahn, S.; Seehafer, K.; Bunz, U. H.
F. Poly(Para-Phenyleneethynylene)-Sensor Arrays Discriminate 22
Different Teas. ACS Sens. 2018, 3 (2), 504−511.
(57) Forest, S.; Théorêt, T.; Coutu, J.; Masson, J.-F. A High-
Throughput Plasmonic Tongue Using an Aggregation Assay and
Nonspecific Interactions: Classification of Taste Profiles in Maple
Syrup. Anal. Methods 2020, 12 (19), 2460−2468.
(58) Kim, C.; Kim, S.-J.; Lee, Y.; Nguyen, T. M.; Lee, J.-M.; Moon, J.-
S.; Han, D.-W.; Oh, J.-W. A Phage- and Colorimetric Sensor-Based
Artificial Nose Model for Banana Ripening Analysis. Sens. Actuators B:
Chem. 2022, 362, 131763.
(59) Forest, S.; Coutu, J.; Montiel-Léon, J. M.; Beniani, I.; Yu, Z. S.;
Craig, M.; Masson, J.-F. Prediction of Maple Syrup Quality from Maple
Sap with a Plasmonic Tongue and Ordinal Mixed-Effects Modeling.
ACS Food Sci. Technol. 2023, 3 (4), 635−647.
(60) Kim, S.-J.; Lee, Y.; Choi, E. J.; Lee, J.-M.; Kim, K. H.; Oh, J.-W.
The Development Progress of Multi-Array Colourimetric Sensors
Based on the M13 Bacteriophage. Nano Converg. 2023, 10 (1), 1.
(61) Chen, X.; Yao, C.; Li, Z. Microarray-Based Chemical Sensors and
Biosensors: Fundamentals and Food Safety Applications. TrAC Trends
Anal. Chem. 2023, 158, 116785.
(62) Qian, S.; Lin, H. Colorimetric Sensor Array for Detection and
Identification of Organophosphorus and Carbamate Pesticides. Anal.
Chem. 2015, 87 (10), 5395−5400.
(63) Yang, L.; Huo, D.; Jiang, Y.; Hou, C.; Zhang, S. Monitoring the
Adulteration of Milk with Melamine: A Visualised Sensor Array
Approach. Food Addit. Contam.: Part A 2013, 30 (5), 786−795.
(64) Feng, N.; Wang, S.; Wei, L.; Wang, Q.; Cheng, X.; Lu, P.; Peng,
X.; Wang, X.; Zhan, C.; Dong, Y.; Chen, Y. Artificial Intelligence-Based
Imaging Transcoding System for Multiplex Screening of Viable
Foodborne Pathogens. Anal. Chem. 2023, 95 (22), 8649−8659.
(65) Singh, H.; Singh, G.; Kaur, N.; Singh, N. Pattern-Based
Colorimetric Sensor Array to Monitor Food Spoilage Using Automated
High-Throughput Analysis. Biosens. Bioelectron. 2022, 196, 113687.
(66) Magnaghi, L. R.; Capone, F.; Zanoni, C.; Alberti, G.; Quadrelli,
P.; Biesuz, R. Colorimetric Sensor Array for Monitoring, Modelling and
Comparing Spoilage Processes of Different Meat and Fish Foods. Foods
2020, 9 (5), 684.

1665 https://doi.org/10.1021/acssensors.4c00252
ACS Sens. 2024, 9, 1656−1665