Amalia Z. Berna

Metal oxide sensors for electronic noses and their application to food analysis.

Electronic noses (E-noses) use various types of electronic gas sensors that have partial specificity. This review focuses on commercial and experimental E-noses that use metal oxide semi-conductors. The review covers quality control applications to food and beverages, including determination of freshness and identification of contaminants or adulteration. Applications of E-noses to a wide range of foods and beverages are considered, including: meat, fish, grains, alcoholic drinks, non-alcoholic drinks, fruits, milk and dairy products, olive oils, nuts, fresh vegetables and eggs.

Geographical origin of Sauvignon Blanc wines predicted by mass spectrometry and metal oxide based electronic nose.

Analysis of 34 Sauvignon Blanc wine samples from three different countries and six regions was performed by gas chromatography-mass spectrometry (GC-MS). Linear discriminant analysis (LDA) showed that there were three distinct clusters or classes of wines with different aroma profiles. Wines from the Loire region in France and Australian wines from Tasmania and Western Australia were found to have similar aroma patterns. New Zealand wines from the Marlborough region as well as the Australian ones from Victoria were grouped together based on the volatile composition. Wines from South Australia region formed one discrete class. Seven analytes, most of them esters, were found to be the relevant chemical compounds that characterized the classes. The grouping information obtained by GC-MS, was used to train metal oxide based electronic (MOS-Enose) and mass spectrometry based electronic (MS-Enose) noses. The combined use of solid phase microextraction (SPME) and ethanol removal prior to MOS-Enose analysis, allowed an average error of prediction of the regional origins of Sauvignon Blanc wines of 6.5% compared to 24% when static headspace (SHS) was employed. For MS-Enose, the misclassification rate was higher probably due to the requirement to delimit the m/z range considered.

Bio-Benchmarking of Electronic Nose Sensors

Background Electronic noses, E-Noses, are instruments designed to reproduce the performance of animal noses or antennae but generally they cannot match the discriminating power of the biological original and have, therefore, been of limited utility. The manner in which odorant space is sampled is a critical factor in the performance of all noses but so far it has been described in detail only for the fly antenna. Methodology Here we describe how a set of metal oxide (MOx) E-Nose sensors, which is the most commonly used type, samples odorant space and compare it with what is known about fly odorant receptors (ORs). Principal Findings Compared with a flys odorant receptors, MOx sensors from an electronic nose are on average more narrowly tuned but much more highly correlated with each other. A set of insect ORs can therefore sample broader regions of odorant space independently and redundantly than an equivalent number of MOx sensors. The comparison also highlights some important questions about the molecular nature of fly ORs. Conclusions The comparative approach generates practical learnings that may be taken up by solid-state physicists or engineers in designing new solid-state electronic nose sensors. It also potentially deepens our understanding of the performance of the biological system.

Analysis of Breath Specimens for Biomarkers of Plasmodium falciparum Infection

Abstract Currently, the majority of diagnoses of malaria rely on a combination of the patient’s clinical presentation and the visualization of parasites on a stained blood film. Breath offers an attractive alternative to blood as the basis for simple, noninvasive diagnosis of infectious diseases. In this study, breath samples were collected from individuals during controlled malaria to determine whether specific malaria-associated volatiles could be detected in breath. We identified 9 compounds whose concentrations varied significantly over the course of malaria: carbon dioxide, isoprene, acetone, benzene, cyclohexanone, and 4 thioethers. The latter group, consisting of allyl methyl sulfide, 1-methylthio-propane, (Z)-1-methylthio-1-propene, and (E)-1-methylthio-1-propene, had not previously been associated with any disease or condition. Before the availability of antimalarial drug treatment, there was evidence of concurrent 48-hour cyclical changes in the levels of both thioethers and parasitemia. When thioether concentrations were subjected to a phase shift of 24 hours, a direct correlation between the parasitemia and volatile levels was revealed. Volatile levels declined monotonically approximately 6.5 hours after initial drug treatment, correlating with clearance of parasitemia. No thioethers were detected in in vitro cultures of Plasmodium falciparum. The metabolic origin of the thioethers is not known, but results suggest that interplay between host and parasite metabolic pathways is involved in the production of these thioethers.

Feature selection for chemical sensor arrays using mutual information.

We address the problem of feature selection for classifying a diverse set of chemicals using an array of metal oxide sensors. Our aim is to evaluate a filter approach to feature selection with reference to previous work, which used a wrapper approach on the same data set, and established best features and upper bounds on classification performance. We selected feature sets that exhibit the maximal mutual information with the identity of the chemicals. The selected features closely match those found to perform well in the previous study using a wrapper approach to conduct an exhaustive search of all permitted feature combinations. By comparing the classification performance of support vector machines (using features selected by mutual information) with the performance observed in the previous study, we found that while our approach does not always give the maximum possible classification performance, it always selects features that achieve classification performance approaching the optimum obtained by exhaustive search. We performed further classification using the selected feature set with some common classifiers and found that, for the selected features, Bayesian Networks gave the best performance. Finally, we compared the observed classification performances with the performance of classifiers using randomly selected features. We found that the selected features consistently outperformed randomly selected features for all tested classifiers. The mutual information filter approach is therefore a computationally efficient method for selecting near optimal features for chemical sensor arrays.

Functional coupling of a nematode chemoreceptor to the yeast pheromone response pathway.

Sequencing of the Caenorhabditis elegans genome revealed sequences encoding more than 1,000 G-protein coupled receptors, hundreds of which may respond to volatile organic ligands. To understand how the worms simple olfactory system can sense its chemical environment there is a need to characterise a representative selection of these receptors but only very few receptors have been linked to a specific volatile ligand. We therefore set out to design a yeast expression system for assigning ligands to nematode chemoreceptors. We showed that while a model receptor ODR-10 binds to C. elegans Gα subunits ODR-3 and GPA-3 it cannot bind to yeast Gα. However, chimaeras between the nematode and yeast Gα subunits bound to both ODR-10 and the yeast Gβγ subunits. FIG2 was shown to be a superior MAP-dependent promoter for reporter expression. We replaced the endogenous Gα subunit (GPA1) of the Saccharomyces cerevisiae (ste2Δ sst2Δ far1Δ) triple mutant (“Cyb”) with a Gpa1/ODR-3 chimaera and introduced ODR-10 as a model nematode GPCR. This strain showed concentration-dependent activation of the yeast MAP kinase pathway in the presence of diacetyl, the first time that the native form of a nematode chemoreceptor has been functionally expressed in yeast. This is an important step towards en masse de-orphaning of C. elegans chemoreceptors.

Feature selection in Enose applications

In my presentation I will summarise results for feature selection from a number of Enose applications ranging from general chemical classification and breath analysis with metal-oxide based Enoses to work scoping the use of biological receptors, in particular receptors of the fruit fly Drosophila, for applications in wine making and explosives detection. The common thread in all applications is the availability of high-dimensional data which is often noisy and in all of its details not necessarily very information rich for any particular application. The challenge for building useful classification systems is to define, extract and select the most “informative” features from the highdimensional data. I will give an overview over our work using exhaustive “wrapper” approaches and a brief comparison to information-theory based methods. I will conclude by highlighting the most pertinent open questions encountered in this research area.

Is CYBERNOSE®an instrument for measuring odor space?

All human senses, other than smell, can be readily represented by definable physical surrogates. Our sense of hot and cold approximates temperature as measured by a thermometer; touch equates to pressure; balance measures orientation and acceleration; vision integrates measures of the wavelengths and intensities of a portion of the electromagnetic spectrum with the four dimensions of space and time; hearing fundamentally equates to the amplitude and frequency of vibrations. Taste, apart from our bitter sense, measures the concentrations of acids, salts, a few amino acids and a number of sugars. Our sense of smell is different. There is no instrument that can measure the fundamental properties that our noses respond to indeed we don’t really know what those properties are. Whilst there is some loose relation between sensory perception of smells and chemical structures, this only seems to apply to closely connected chemical species and breaks down once we move beyond that. In principle we could objectively define an odor space, to sit alongside the human or animal percept, that could be measured with a set of odorant receptors, precisely definable as strings of amino acids expressed under standard conditions. Such odorant receptors would define the dimensions of odorant space. Of course, currently, there are many practical obstacles to employing biological odorant receptors for this purpose. One of the difficulties is the existence of many millions of different odorant receptor sequences belonging to at least four molecular families and distributed among millions of different metazoan species. We could define a human receptor-based odor space or a dog or a fly or nematode worm receptor-based odor space. They would differ, in some respects profoundly. Nevertheless, solving the aforementioned practical problems would not only support many practical applications but also allow us to reproducibly define the dimensions of odor space and make “smell” an objective measurable property. Our approach to this is through machine olfaction. Currently we are working with off-the shelf electronic noses as well as developing our own bioelectronic nose, the CYBERNOSE instrument. In the former case, we looked to biology, specifically to published data on Drosophila olfaction, to guide our understanding of how solid state sensors encode odor space [1]. The take home message appears to be that the inadequacies of current solid state sensors prevent them being very useful for characterising and defining odor space, as proposed by Shurmer & Whitehouse (1993). On the other hand, we believe that development of the CYBERNOSE instrument, which incorporates odorant biosensors, offers a viable approach to characterise defined, standardised odor spaces, based on but not limited to what is available to the nematode worm Caenorhabditis elegans.

Malaria detection using breath biomarkers.

Although there were almost 200 million cases of malaria in 2013, resulting in over half a million deaths, this lethal infection is in retreat.1 Better control through prevention with insecticide-treated nets and more effective drugs (including artemisinin, for the discovery of which the 2015 Nobel Prize in Physiology or Medicine was awarded) mean that the ambition of elimination is again on the agenda. Sensitive diagnosis of malaria is becoming increasingly important, particularly for lowlevel and asymptomatic cases. Currently, most diagnoses of malaria use microscopy, which is not sufficiently sensitive to enable elimination and is dependent on highly trained operators with good equipment.

Classifying chemical sensor data using GPU-accelerated bio-mimetic neuronal networks based on the insect olfactory system

Chemosensing “e-nose” technology has great potential applications in everyday life, ranging from drug detection to food quality assessment and even the diagnosis of illness. However, odour detection and classification remains a highly challenging domain, characterized by high dimensionality, unknown organization of the vast “odourant space” of volatile chemicals and complex, turbulent odour plumes. To compound these difficulties, current sensor technology continues to exhibit distinct shortcomings in speed, sensitivity, selectivity, recovery, and drift avoidance. In the research reported here we turn to a range of recent neuronal models [1-6] that were developed to describe the insect olfactory system, and have been shown to perform well across a range of classic, static classification tasks such as the MNIST handwritten digit set and the Sigma-Aldrich scent database. The insect olfactory system has been extensively studied and has been shown to be both fast and highly effective in complex natural conditions despite its limited size and complexity (when compared to the mammalian system) [2]. Insects such as moths, honey bees, locusts and fruit flies are capable of odour detection and classification tasks well beyond the abilities of current e-nose technology and machine learning algorithms [6]. We present results of applying an insect-inspired approach to the design of a learning spiking neural network that receives synchronized time series data from up to 12 metal-oxide based gas sensors, comprising an optimised [7] combination of classical doped tin oxide and novel zeolite-coated chromium titanium oxide sensors. We have collected sample data sets for classification tasks ranging from “easy” (single chemical identification presented under laboratory conditions) through to “hard” (identification of indicators of infectious diseases in breath samples taken from patients). To address slow sensor response we consider a range of transient-based processing before applying self-organisation techniques to most effectively locate “virtual receptors” (VR) in sensor space. We look to address decorrelation (separation in feature space) and the supervised association of responses with rewards by using correlates of the insect antennal lobe (AL) and mushroom body (MB) structures whilst applying reward-based spike-timing dependent plasticity mechanisms. Classification accuracy is compared with support vector machine (SVM) learning which we also look to match for speed through the use of GPU accelerated neural simulation [5] via the NVidia CUDA TM -based GeNN platform (http://sourceforge.net/projects/genn/).