Search

Simple Summary: Colorectal cancer is the most frequent neoplasm in Western countries, the second most frequent neoplasm after breast cancer in women and the third most frequent neoplasm after prostate and lung cancer in men. Early diagnosis and disease screening are based on a fecal occult blood test, which, if positive, is complemented by colonoscopy. Currently, efforts are underway to find alternatives to the fecal occult blood test for various reasons. First, there is an ongoing attempt to increase the participation of the population to be screened. Second, there is a need to decrease the number of false positives to reduce the number of unnecessary colonoscopies. A urine test could be more widely accepted than a fecal test, and this is the scenario for which urinary metabolomics and volatilome studies are being developed. Our review provides the first exhaustive evaluation of metabolomics and volatilomics for the determination of colorectal cancer in urine. Abstract: To increase compliance with colorectal cancer screening programs and to reduce the recommended screening age, cheaper and easy non-invasiveness alternatives to the fecal immunochemical test should be provided. Following the PRISMA procedure of studies that evaluated the metabolome and volatilome signatures of colorectal cancer in human urine samples, an exhaustive search in PubMed, Web of Science, and Scopus found 28 studies that met the required criteria. There were no restrictions on the query for the type of study, leading to not only colorectal cancer samples versus control comparison but also polyps versus control and prospective studies of surgical effects, CRC staging and comparisons of CRC with other cancers. With this systematic review, we identified up to 244 compounds in urine samples (3 shared compounds between the volatilome and metabolome), and 10 of them were relevant in more than three articles. In the meta-analysis, nine studies met the criteria for inclusion, and the results combining the case-control and the pre-/post-surgery groups, eleven compounds were found to be relevant. Four upregulated metabolites were identified, 3-hydroxybutyric acid, L-dopa, L-histidinol, and N1, N12-diacetylspermine and seven downregulated compounds were identified, pyruvic acid, hydroquinone, tartaric acid, and hippuric acid as metabolites and butyraldehyde, ether, and 1,1,6-trimethyl-1,2-dihydronaphthalene as volatiles. This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 798038

Mass spectrometry imaging (MSI) can reveal biochemical information directly from a tissue section. MSI generates a large quantity of complex spectral data which is still challenging to translate into relevant biochemical information. Here, we present rMSIproc, an open-source R package that implements a full data processing workflow for MSI experiments performed using TOF or FT-based mass spectrometers. The package provides a novel strategy for spectral alignment and recalibration, which allows to process multiple datasets simultaneously. This enables to perform a confident statistical analysis with multiple datasets from one or several experiments. rMSIproc is designed to work with files larger than the computer memory capacity and the algorithms are implemented using a multi-threading strategy. rMSIproc is a powerful tool able to take full advantage of modern computer systems to completely develop the whole MSI potential., This work has been supported by the Spanish Ministry of Economy and Competitiveness through project TEC2015-69076-P, PRs predoctoral grant no. BES-2013-065572, project BFU2017-89336-R, the Direcció General de Recerca of the Government of Catalonia through project 2017 SGR 1119 and IPT’s fellowship from the Spanish Ministry of Education, Culture and Sports grant no. FPU 14/04457.

Mass spectrometry imaging (MSI) is a label-free analytical technique capable of molecularly characterizing biological samples, including tissues and cell lines. The constant development of analytical instrumentation and strategies over the previous decade makes MSI a key tool in clinical research. Nevertheless, most MSI studies are limited to targeted analysis or the mere visualization of a few molecular species (proteins, peptides, metabolites, or lipids) in a region of interest without fully exploiting the possibilities inherent in the MSI technique, such as tissue classification and segmentation or the identification of relevant biomarkers from an untargeted approach. MSI data processing is challenging due to several factors. The large volume of mass spectra involved in a MSI experiment makes choosing the correct computational strategies critical. Furthermore, pixel to pixel variation inherent in the technique makes choosing the correct preprocessing steps critical. The primary aim of this review was to provide an overview of the data-processing steps and tools that can be applied to an MSI experiment, from preprocessing the raw data to the more advanced strategies for image visualization and segmentation. This review is particularly aimed at researchers performing MSI experiments and who are interested in incorporating new data-processing features, improving their computational strategy, and/or desire access to data-processing tools currently available. © 2016 Wiley Periodicals, Inc. Mass Spec Rev 37:281-306, 2018.

HighlightsGC×GC-MS automated compound identification and quantification in metabolomics.Orthogonal signal deconvolution (OSD) for spectra deconvolution.PCA used as an alternative to the traditional least squares approach.OSD R package used for fast and automated compound identification. Comprehensive gas chromatography-mass spectrometry (GC×GC-MS) provides a different perspective in metabolomics profiling of samples. However, algorithms for GC×GC-MS data processing are needed in order to automatically process the data and extract the purest information about the compounds appearing in complex biological samples. This study shows the capability of independent component analysis-orthogonal signal deconvolution (ICA-OSD), an algorithm based on blind source separation and distributed in an R package called osd, to extract the spectra of the compounds appearing in GC×GC-MS chromatograms in an automated manner. We studied the performance of ICA-OSD by the quantification of 38 metabolites through a set of 20 Jurkat cell samples analyzed by GC×GC-MS. The quantification by ICA-OSD was compared with a supervised quantification by selective ions, and most of the R2 coefficients of determination were in good agreement (R2>0.90) while up to 24 cases exhibited an excellent linear relation (R2>0.95). We concluded that ICA-OSD can be used to resolve co-eluted compounds in GC×GC-MS.

Mass spectrometry imaging (MSI) is a molecular imaging technique that maps the distribution of molecules in biological tissues with high spatial resolution. The most widely used MSI modality is matrix-assisted laser desorption/ionization (MALDI), mainly due to the large variety of analyte classes amenable for MALDI analysis. However, the organic matrices used in classical MALDI may impact the quality of the molecular images due to limited lateral resolution and strong background noise in the low mass range, hindering its use in metabolomics. Here we present a matrix-free laser desorption/ionization (LDI) technique based on the deposition of gold nanolayers on tissue sections by means of sputter-coating. This gold coating method is quick, fully automated, reproducible, and allows growing highly controlled gold nanolayers, necessary for high quality and high resolution MS image acquisition. The performance of the developed method has been tested through the acquisition of MS images of brain tissues. The obtained spectra showed a high number of MS peaks in the low mass region (m/z below 1000 Da) with few background peaks, demonstrating the ability of the sputtered gold nanolayers of promoting the desorption/ionization of a wide range of metabolites. These results, together with the reliable MS spectrum calibration using gold peaks, make the developed method a valuable alternative for MSI applications.

New approaches are increasingly being used for studying and evaluating coronary heart disease (CHD), especially since the irruption of metabolomics. The classical approach is to use enzymatically-measured standard lipids and these are still the main markers for assessing risk of CHD. Since metabolomics relies on advanced analytical technologies, such as MS and NMR, using them to estimate standard lipids would be of great interest because there is no need for additional biochemical measures. The present study evaluates partial least squares and N-way partial least squares regression models to predict standard lipid concentrations by using serum and plasma sample sets from various clinical centres. Information provided by editing NMR techniques and 2D diffusion NMR was incorporated in these models using four different data structures. Firstly, the models were calibrated and validated with three of the four sample sets (n = 591) involved. Then the best estimation models were selected and applied to the left-out sample set. This evaluation of a new sample set gave correlation coefficients of predicted versus biochemical variables above 0.86 and %rRMSE lower than 18 %. These values are similar to those found by other studies although, in our case, the results are more general because we used a higher number of samples (n = 785) from different sample sets, different clinical centres and different blood matrices (serum and plasma). Finally, we compared the performance of NMR predicted lipids and enzymatically measured lipids in a clinical case study.

Mass spectrometry imaging (MSI) is a technique that can map analyte spatial distribution directly onto a tissue section. This enables the spatial correlation of molecular entities with a tissue morphology to be investigated. Analyte annotation in MSI is intrinsically linked to the mass accuracy of the data. Mass accuracy and analyte identification are determined by such factors as the experimental set up and the data processing workflow. We present an MSI data processing workflow that uses a label-free approach to compensate for mass shifts. The algorithms developed were designed to perform efficiently even for datasets much larger than computer's memory. Herein, we present the application of the developed processing workflow to a dataset with more than 13.000 pixels and ∼50.000 mass channels. We assessed the overall mass accuracy in the range m/z 400 to 1200 using silver and gold sputtered nanolayers. With our novel processing workflow we were able to obtain mass errors as low as 5 ppm using a TOF instrument.

The study of nutritional interventions in humans is difficult to assess because the induced metabolic changes are lower than the natural biological variability between subjects. Due to its holistic approach, 1H NMR is one of the preferred technologies for this type of studies, even though it has a very low sensitivity. This work shows how the use of several chemometric algorithms on the measured data compensates for these drawbacks and allows the study of the effects of the nutritional intervention isolating them from the natural variability inherent to human studies. Mild to moderate hypercholesterolemic patients received either placebo or soluble fiber in a low saturated fat diet. Plasma samples were collected at week 0 and week 8. Spectra obtained with NMR equipment were processed with ANOVA simultaneous component analysis (ASCA). The application of clustering techniques revealed different responses based on the patient's basal state, which allowed the identification of responders from non-responders. Results showed a triglyceride level reduction of up to 15% (p = 0.0032), with a higher reduction for those patients with a higher initial lipid profile. Moreover, line-shape fitting techniques applied to the NMR spectra allowed the conclusion that LDL (and VLDL) lipoprotein particles, and more noticeably triglycerides, moved to a profile configuration associated with lower cardiovascular risk. Results shed light on some of the metabolic modifications that husk fiber induces in humans which could not be seen with more conventional data analysis approaches. Our conclusion is that by using the right chemometric techniques it is possible to assess nutritional intervention effects in human NMR human studies despite the low sensitivity and selectivity that the technique offers today.

Summary R platform provides some packages that are useful to process mass spectrometry imaging (MSI) data; however, none of them provide an easy to use graphical user interface (GUI). Here, we introduce rMSI, an R package for MSI data analysis focused on providing an efficient way to manage MSI data together with a GUI integrated in R environment. MS data is loaded in rMSI custom format optimized to minimize the memory footprint yet maintaining a fast spectra access. The rMSI GUI is designed for simple and effective data exploration and visualization. Moreover, rMSI is designed to be integrated in the R environment through a library of functions that can be used to share MS data across others R packages. The release of rMSI for R environment establishes a novel and flexible platform for MSI data analysis, completely free and open-source. Availability and implementation The code, the documentation, a tutorial and example data are available open-source at: github.com/prafols/rMSI. Contact jesus.brezmes@urv.cat. Supplementary information Supplementary data are available at Bioinformatics online.

Gas chromatography–mass spectrometry (GC–MS) produces large and complex datasets characterized by co-eluted compounds and at trace levels, and with a distinct compound ion-redundancy as a result of the high fragmentation by the electron impact ionization. Compounds in GC–MS can be resolved by taking advantage of the multivariate nature of GC–MS data by applying multivariate resolution methods. However, multivariate methods have to be applied in small regions of the chromatogram, and therefore chromatograms are segmented prior to the application of the algorithms. The automation of this segmentation process is a challenging task as it implies separating between informative data and noise from the chromatogram. This study demonstrates the capabilities of independent component analysis–orthogonal signal deconvolution (ICA–OSD) and multivariate curve resolution–alternating least squares (MCR–ALS) with an overlapping moving window implementation to avoid the typical hard chromatographic segmentation. Also, after being resolved, compounds are aligned across samples by an automated alignment algorithm. We evaluated the proposed methods through a quantitative analysis of GC–qTOF MS data from 25 serum samples. The quantitative performance of both moving window ICA–OSD and MCR–ALS-based implementations was compared with the quantification of 33 compounds by the XCMS package. Results shown that most of the R2 coefficients of determination exhibited a high correlation (R2 > 0.90) in both ICA–OSD and MCR–ALS moving window-based approaches.

Quality control of essential oils is an important topic in industrial processing of medicinal and aromatic plants. In this paper, the performance of Fuzzy Adaptive Resonant Theory Map (ARTMAP) and linear discriminant analysis (LDA) algorithms are compared in the specific task of quality classification of Rosa damascene essential oil samples (one of the most famous and valuable essential oils in the world) using an electronic nose (EN) system based on seven metal oxide semiconductor (MOS) sensors. First, with the aid of a GC-MS analysis, samples of Rosa damascene essential oils were classified into three different categories (low, middle, and high quality, classes C1, C2, and C3, respectively) based on the total percent of the most crucial qualitative compounds. An ad-hoc electronic nose (EN) system was implemented to sense the samples and acquire signals. Forty-nine features were extracted from the EN sensor matrix (seven parameters to describe each sensor curve response). The extracted features were ordered in relevance by the intra/inter variance criterion (Vr), also known as the Fisher discriminant. A leave-one-out cross validation technique was implemented for estimating the classification accuracy reached by both algorithms. Success rates were calculated using 10, 20, 30, and the entire selected features from the response of the sensor array. The results revealed a maximum classification accuracy of 99% when applying the Fuzzy ARTMAP algorithm and 82% for LDA, using the first 10 features in both cases. Further classification results explained that sub-optimal performance is likely to occur when all the response features are applied. It was found that an electronic nose system employing a Fuzzy ARTMAP classifier could become an accurate, easy, and inexpensive alternative tool for qualitative control in the production of Rosa damascene essential oil.

Nonalcoholic fatty liver disease is considered to be the hepatic manifestation of metabolic syndrome and is usually related to high-fat, high-cholesterol diets. With the rationale that the identification and quantification of metabolites in different metabolic pathways may facilitate the discovery of clinically accessible biomarkers, we report the use of (1)H NMR metabolomics for quantitative profiling of liver extracts from LDLr(-/-) mice, a well-documented mouse model of fatty liver disease. A total of 55 metabolites were identified, and multivariate analyses in a diet- and time-comparative strategy were performed. Dietary cholesterol increased the hepatic concentrations of cholesterol, triglycerides, and oleic acid but also decreased the [PUFA/MUFA] ratio as well as the relative amount of long-chain polyunsaturated fatty acids in the liver. This was also accompanied by variations of the hepatic concentration of taurine, glutathione, methionine, and carnitine. Heat-map correlation analyses demonstrated that hepatic inflammation and development of steatosis correlated with cholesterol and triglyceride NMR derived signals, respectively. We conclude that dietary cholesterol is a causal factor in the development of both liver steatosis and hepatic inflammation.

In order to improve the mass spectra (MS)-based electronic nose (E-nose) performance, we have included the retention time data given by a new E-nose configuration based on a gas chromatograph–mass spectrometer (GC–MS) as a third dimension. The primary aim of this work is to show that the addition of the third dimension is useful, and brings extra information, helping in the classification of samples. By using this extra information our second goal is to optimize the chromatographic method in order to shorten the time of the chromatographic analysis to a minimum, while still having acceptable results. An experiment was designed in the form of 20 solutions with a high degree of similarity in mass spectra and chromatographic retention times. In order to optimize the system performance and reduce the time of the measurements to a minimum two different chromatographic methods were evaluated. By analyzing these solutions with two-way and three-way PCA, PARAFAC, PLS-DA and n-PLS-DA, and concatenated with supervised Fuzzy Artmap paradigms, we show that the addition of the extra information in the form of the chromatographic separation, even when using a short chromatographic separation, improves the results obtained, compared to the two-way analysis of the mass spectra or total ion chromatogram (TIC) alone. A third goal was to see which signal processing approach was the best suited. We found that when the retention time is used as a third dimension when chromatographic peaks are well resolved, two-way methods work better than their three-way counterparts, whereas in the case of a more challenging situation (a more coeluted chromatogram, with a much shorter measurement time) three-way methods perform better than classic two-way approaches.

Monocyte chemoattractant protein-1 (MCP-1) plays a relevant role in macrophage migration but recentfindings suggest an additional role in lipid and glucose metabolism. We report the use of 1 H NMRspectroscopy as a useful complementary method to assess the metabolic function of this gene ina comparative strategy. This metabonomic analysis was rapid, simple, quantitative and reproducible, andrevealed a suggestive relationship between the expression of the MCP-1 gene and hepatic glucose andtaurine concentrations. This approach should be considered in genetically modified mice when a meta-bolic alteration is suspected, or in routine assessment of metabolic phenotype. 2009 Elsevier Masson SAS. All rights reserved. 1. IntroductionThe evaluation of the activity of candidate genes or proteinsthrough transcriptomics and proteomics is usually expensive andcomplex. With the rationale that small differences in metabolismmay result in pronounced changes in metabolite concentrations,we designed a metabonomic approach to assess the function ofselected genes in genetically modified mice using a comparativestrategy between knockout and derived double-knockout mice. Totest this approach we chose MCP-1. Although the main rolereported for MCP-1 is related to monocyte trafficking and otherimmunological functions [1,2], there is evidence to suggest that itparticipates in metabolism [3]. We have demonstrated that MCP-1is involved in the development of obesity and in impaired glucosemetabolism in LDLr-deficient mice (LDLr

In this paper, the use of a new technique to obtain transient sensor information is introduced and its usefulness to improve the selectivity of metal oxide gas sensors is discussed. The method is based on modulating the flow of the carrier gas that brings the species to be measured into the sensor chamber. In such a way, the analytes’ concentration at the surface of the sensors is altered. As a result, reproducible patterns in the sensor response develop, which carry important information for helping the sensor system, not only to discriminate among the volatiles considered but also to semi-quantify them. This has been proved by extracting features from sensor dynamics using the discrete wavelet transform (DWT) and by building and validating support vector machine (SVM) classification models. The good results obtained (100% correct identification among 5 volatile compounds and nearly a 89% correct simultaneous identification and quantification of these volatiles), which clearly outperform those obtained when the steady-state response is used, prove the concept behind flow modulation.

One of the fields where the idea of developing artificial electronic mammalian sense devices might be widely applied is in food analysis. In this field, one very complex attribute yet of great importance is that of freshness. For instance, freshness is one of the most significant parameters of fish quality in most markets. The state of fish freshness can be assessed by various indicators but most of them are costly, time-consuming and lack of aptness for in situ or on site monitoring. This makes that the development of simple and undemanding systems for the evaluation of fish freshness is still an unresolved goal. Based on this demand, we focused our attention on the use of unselective sensors for the potential development of low-cost and easy-to-use devices for the estimation of freshness. After a preliminary study with a number of different electrodes, simple Au and Ag wires were used for the analysis of the evolution of sea (Sparus aurata) bream fish stored throughout time. PCA and neuronal network analyses showed that it is possible to determine the post-mortem time elapsed in minced gilthead sea bream using simple potentiometric measurements, suggesting that this method could be used for the analysis and evaluation of fish freshness. In order to somehow validate the potentiometric protocol for freshness assessment, the method has been compared with the determination of fish freshness based on the analysis of the concentration of certain bio-molecules as a function of time after the death of the fish. Among different possibilities the K1 index that simultaneously measures the variation in the ATP metabolites inosine 5′-phosphate (IMP), inosine (Ino) and hypoxanthine (Hx) was used. A remarkable correlation has been found between the K1 index determined during 7 days and the potentials of the Au and Ag electrodes.

A new multisensor system based on a thermal desorption pre-concentrator has been used jointly with a tin oxide gas sensor array to asses the contamination of carbon dioxide by benzene. The results obtained show that using this type of pre-concentrator, it is possible to detect benzene at ppb levels even in the presence of other pollutants at ppm levels. The system uses a Carbopack B unit that absorbs benzene traces and then releases them concentrated in a factor greater than 200 and is capable of classifying benzene samples down to 10 ppb in a CO2 flow with different volatile interferences. The e-nose reached a 100% success rate in the classification and identification of the benzene, CO2 and pollutants. These results show that e-nose technology can be a useful tool as quality control in the beverage industry.

For the past 15 years, several research groups have studied the feasibility to use electronic olfactory systems to measure fruit quality and ripeness in a nondestructive manner. This chapter reviews the different attempts, strategies, configurations, and goals that have been explored using electronic noses for fruit quality and ripeness assessment. They are compared to established methods and a final discussion on what the future holds for these instruments is included.

The selectivity of metal oxide gas sensors can be improved by operating the sensors in a temperature-modulated mode. Although the selection of optimal modulating frequencies deserves accurate attention, this aspect has been generally overlooked. In this paper, a systematic method to determine which are the optimal temperature-modulation frequencies to solve a given gas analysis problem has been introduced, discussed in detail and fully validated for the first time. The optimization method is based on the use of multi-level pseudo-random sequences. These sequences share some properties with white noise and allow for the impulse response of the sensor–gas system to be estimated. Using this strategy, it is shown that the best temperature-modulating frequencies to discriminate and quantify gases using an array of four metal oxide gas sensors are identified. The process is illustrated solving a practical application: the quantitative analysis of acetaldehyde, ethylene, ammonia and their binary mixtures (monitoring climacteric fruit during cold storage). By using a multi-sinusoidal temperature-modulating signal, the frequencies of which are a reduced set of the optimal ones, the gases and gas mixtures were discriminated with a 100% success rate. In gas identification, features from the sensors’ dynamic response extracted via the fast Fourier transform (FFT) were used together with a fuzzy ARTMAP neural network. After the identification process, the concentration of the different species was accurately predicted by PLS-based calibration models. These results compare favorably with the ones obtained when the sensor array was operated in a steady-state mode. The optimization method is shown to be consistent and effective, since the process of determining optimal modulation frequencies and the validation process were conducted using different metal oxide gas sensor micro-arrays (of the same type) and different measurement sets.

In this paper, a new variable selection technique inspired in sequential forward selection but specifically designed to work with support vector machines is introduced. The usefulness of the variable selection coupled to support vector machines for solving classification and regression problems is assessed by analysing two different databases. The first database corresponds to different concentrations of vapours and vapour mixtures measured with a metal oxide gas-sensor e-nose and the second database corresponds to different Iberian hams measured with a mass-spectrometry based e-nose. Using a reduced set of important variables (i.e. reducing the dimensionality of input space by the variable selection procedure) results in support vector machines with better performance. For example, the success rate in ham classification (11-class problem) rises from 79.91% (when all the variables available are used) to 90.30% (when a reduced set of input variables is used). Furthermore, a quantitative analysis of ham samples with good accuracy is shown to be possible: when the variable selection process introduced is coupled to support vector machine regression models, the correlation coefficients of actual versus predicted humidity, water activity and salt in ham samples are 0.975, 0.972 and 0.943, respectively. This compares favourably with the correlation coefficients obtained when no variable selection is performed (0.937, 0.924 and 0.894).

We address the problem of variable selection within the pattern recognition engine of multisensor systems. This problem arises when there is high dimensionality in the data used for training (e.g. because a high number of sensors are used, or because many features are extracted from the response of each sensor, or both). Different variable selection techniques (including deterministic and stochastic methods) have been coupled with neural network-based classifiers. The usefulness of each technique implemented is benchmarked by evaluating its performance in terms of three objective parameters: the success rate in classification, the dimensionality of the final set of variables used for training and the time needed to complete the variable selection procedure. The database consisted of 96 measurements of ammonia, acetone and o-xylene vapours and their binary mixtures gathered using a metal oxide gas sensor array. Each measurement was described by 120 variables (12 sensors and 10 parameters each). A new strategy for variable selection, which is based on a two-step approach, is introduced that leads to the building of parsimonious classification models based on either the fuzzy ARTMAP or the probabilistic neural networks. For example, a 91.66% success rate in the simultaneous identification and quantification of the species and their mixtures was obtained using nine input variables only (out of the 120 available). This process of variable selection was conducted in two-step, i.e. a coarse selection based on a variance criterion followed by a simulated annealing process. This two-step variable selection took about 10 min to complete in a Pentium 4 PC platform.

Active layers consisting of rf sputtered WO3 were deposited on microhotplate substrates. The films were doped with seven different materials (Pt, Au, Ag, Ti, SnO2, ZnO and ITO (indium tin oxide)). The eight types of sensors (including pure tungsten oxide ones) were tested in the presence of ammonia, hydrogen sulphide, nitrogen dioxide, carbon monoxide and methane. It was found that gold improved the sensitivity to H2S. On the other hand, doping with Ag and Pt led to higher responses to NO2 and NH3, respectively. No response to CH4 was observed. The sensitivity to CO was very low. The influence of the working temperature on the sensor response was also studied. Our study proves that selective gas detection is possible combining a few tungsten oxide sensors with different dopants.

In recent years, modulating the working temperature of metal-oxide gas sensors has been one of the most widely used methods to enhance sensor selectivity. When the working temperature of a gas sensor is modulated, the kinetics of the gas-sensor interaction are altered, and this leads to characteristic response patterns. Many works have shown that it is possible to identify and determine the concentration of gases in simple mixtures, even using a single temperature-modulated metal-oxide gas sensor. However, the selection of the frequencies used to modulate temperature remains an empirical process. In this paper, we introduce a method, borrowed from the field-of-system identification, to systematically determine the optimal set of modulating frequencies to solve a given gas-analysis application. The method consists of using maximum-length pseudorandom binary sequences to modulate the working temperature of metal-oxide gas sensors. Since these signals have a flat power spectrum (i.e., like white noise) in a wide frequency range, an estimate of the impulse response of each gas-sensor pair can be computed by the cross correlation of the excitatory and response sequences. Studying the impulse response estimates, the set of modulating frequencies that are useful to discriminate between different gases and to estimate gas concentration, is obtained in a systematic way. The method is demonstrated with tungsten oxide micro-hotplate gas sensors applied to detect ammonia, nitrogen dioxide, and their binary mixtures at different concentrations. It is shown that it is possible to find temperature-modulating frequencies to obtain high gas identification and quantification rates (95.55% and 100%, respectively).

The gas sensing properties of pure and indium-doped nanoparticle WO3 thick films were studied. Sensors were prepared using commercial WO3 nanopowders and powder mixtures with different concentrations of In (1.5, 3.0 and 5.0 wt.%). The gas sensing properties of the sensors to nitrogen dioxide, carbon monoxide, ammonia and ethanol were investigated. It was observed that the pure nanoparticle WO3 sensing layers began to react with nitrogen oxide at room temperature. These sensors had maximum sensitivity to NO2 at 100 °C. The indium-doped WO3-based sensors were selective to NO2 and CO at 200 and 300 °C, respectively. A mechanism for such behaviour is discussed.

We show how it is possible to optimise a multi-frequency signal to be used in the modulation of the operating temperature of an integrated gas sensor microarray. In the first step, a multilevel pseudo random sequence (ML-PRS), which allows for modulating the operating temperature of the sensors in a wide frequency range, is used to obtain an estimate of the impulse response of each microsensor-gas system. ML-PRS are interesting because they help to reduce the effects of noise and non-linearity as experienced with gas sensors. In the second step, by analysing the spectral components of the impulse response estimates, the modulating frequencies that better help in discriminating and quantifying the gases and gas mixtures considered are found. Finally, by selecting a subset of the best modulating frequencies, an optimised multi-frequency temperature-modulating signal can be synthesised. The method is illustrated by analysing different concentrations of NH 3 ,N O 2 and their mixtures using a microarray of WO3-based gas sensors, but it can be further applied to any given gas analysis problem. © 2005 Elsevier B.V. All rights reserved.

We report on the use of either reactive magnetron sputtering or screen printing to deposit tin and tungsten-oxide gas-sensitive layers onto integrated micromachined arrays. The procedures allow the deposition of the sensing layers before membranes have been etched, which leads to gas microsensors with an excellent fabrication yield. The microstructure of the sensitive films is analyzed by means of SEM and EDX. The response of the different microarrays to ethanol, acetone, and ammonia vapors and their binary mixtures, and toxic gases such as NO/sub 2/ and CO, is studied at different operating temperatures. The response of the different sensors to ambient humidity is also investigated. Finally, it is shown that by using PCA and fuzzy ARTMAP neural networks, it is possible to simultaneously identify and quantify the toxic gases with a 100% success rate. A 95% success rate is obtained in the semi-quantitative analysis of vapors and vapor mixtures. These results prove the viability and usefulness of the techniques introduced to obtain integrated sensor microarrays that are suitable for battery-powered gas/vapor monitors.

An electronic nose for classification of olive oil samples is presented. Principal component analysis and a modified fuzzy artmap neural network where applied to data acquired from 12 sensors. A custom designed variable selection technique was also used to boost performance. Ten different samples of olive oils were classified with 78% accuracy, and confusion occurred mostly between similar olive oils. Defective samples were separated from defect-free olive oil with 97% accuracy. These results show that careful variable selection, coupled to a modified fuzzy artmap algorithm, can significantly improve electronic nose performance.

Volatile production, sensory acceptability and electronic nose responses were measured in 'Fuji' apples (Malus x domestica Borkh.) to evaluate the effects of different storage conditions, storage periods and days of shelf life. Apples harvested at 185 days after full bloom were analysed after 5 and 7 months of cold storage in normal atmosphere (AIR: 21% O 2 + 0.03% CO 2 ) and in two controlled atmospheres (CA: 1% O 2 + 2% CO 2 and 3% O 2 + 2% CO 2 ). During post-storage ripening, apples were kept at 20°C for 1, 5 and 10 days before analytical and sensory measurements were made. Volatile production was measured by means of headspace/gas chromatography and electronic nose and the acceptability of apples by a semi-trained panel. Sensor responses registered by 21 different metal oxide sensors of an electronic nose (EN) were used to classify apples using Principal Component Analysis (PCA). The PCA model containing data from all fruits (at-harvest and after storage) showed that it was possible to identify fruits recently harvested and under different storage conditions. Principal components 1 (PC1) and 2 (PC2) accounted for 63 and 30% of the total variability respectively. On the other hand, a PCA involving volatile production, sensory acceptability and EN signals corresponding to all fruits allowed a better differentiation between fruits stored at 1% O 2 + 2% CO 2 and 3% O 2 + 2% CO 2 , with an explained variance of 58% (two first PCs).

A new method based on the use of an SPME MS e-nose, to assess oxidative and hydrolytic rancidity in crisps is introduced. The method can become an alternative tool to the traditional methods such as the Rancimat and ADV tests. Because it does not rely on a previous step of oil extraction from the samples to be analysed, sample preparation is easier and the analysis is speeded up (e.g., the Rancimat method can take several hours to produce a result if the oil under analysis is of good quality). The effectiveness of the e-nose in the assessment of the quality of crisps was demonstrated by developing two different applications. First, the e-nose was used to classify crisp samples according to their rancidity stage. When the e-nose data was processed by a fuzzy ARTMAP classifier, the success rate in classification was estimated to be around 93% (validation results). The system could discriminate 100% of the fresh crisps (i.e., class A) from the rancid ones (B, C and D). The e-nose was trained to predict the results of the ADV and Rancimat tests by building quantitative PLS models. A good correlation existed between our instrument and the results of the Rancimat and ADV tests (the correlation coefficients were 0.98 and 0.97, respectively). The best results were obtained when the dimensionality of input data was reduced by applying variable selection procedures based on principal component analysis and genetic algorithms.

Well-established methods to assess rancidity in potato crisps such as the Rancimat or the acid degree value are time-consuming and labour-intensive. Here, we report on alternative methods, based on e-nose technology, to assess rancidity directly from potato crisps without any previous oil extraction step. This simplifies sample preparation, avoids the use of solvents or high temperatures and significantly speeds up the measurement process (from several hours down to 25 min). Two different e-noses were implemented. One was based on SPME coupled to fingerprint MS and the other one was based on dynamic headspace sampling and an array of metal oxide gas sensors. The two e-noses were used to classify crisps according to four stages of oxidative rancidity. While the MS e-nose reached a 100% success rate in this classification, the success rate of the GS e-nose was 68%. These results show that e-nose technology can be a useful tool for the crisp industry. We show that it is possible to reliably assess rancidity in potato crisps by either a mass spectrometry or a gas sensor-based electronic nose. The two approaches are presented and their performance compared in the framework of this application.

The main goal of our study was to see whether an artificial olfactory system can be used as a nondestructive instrument to measure fruit maturity. In order to make an objective comparison, samples measured with our electronic nose prototype were later characterized using fruit quality techniques. The cultivars chosen for the study were peaches, nectarines, apples, and pears. With peaches and nectarines, a PCA analysis on the electronic nose measurements helped to guess optimal harvest dates that were in good agreement with the ones obtained with fruit quality techniques. A good correlation between sensor signals and some fruit quality indicators was also found. With pears, the study addressed the possibility of classifying samples regarding their ripeness state after different cold storage and shelf-life periods. A PCA analysis showed good separation between samples measured after a shelf-life period of seven days and samples with four or less days. Finally, the electronic nose monitored the shelf-life ripening of apples. A good correlation between electronic nose signals and firmness, starch index, and acidity parameters was found. These results prove that electronic noses have the potential of becoming a reliable instrument to assess fruit ripeness.

This paper presents the design, optimization, and evaluation of a mass spectrometry-based electronic nose (MS e-nose) for early detection of unwanted fungal growth in bakery products. Seven fungal species (Aspergillus flavus, Aspergillus niger, Eurotium amstelodami, Eurotium herbariorum, Eurotium rubrum, Eurotium repens, and Penicillium corylophillum) were isolated from bakery products and used for the study. Two sampling headspace techniques were tested: static headspace (SH) and solid-phase microextraction (SPME). Cross-validated models based on principal component analysis (PCA), coupled to discriminant function analysis (DFA) and fuzzy ARTMAP, were used as data treatment. When attempting to discriminate between inoculated and blank control vials or between genera or species of in vitro growing cultures, sampling based on SPME showed better results than those based on static headspace. The SPME-MS-based e-nose was able to predict fungal growth with 88% success after 24 h of inoculation and 98% success after 48 h when changes were monitored in the headspace of fungal cultures growing on bakery product analogues. Prediction of the right fungal genus reached 78% and 88% after 24 and 96 h, respectively.

Either by screen-printing or sputtering, sensitive layers of tin oxide were deposited on silicon micromachined substrates (four-element integrated microarrays). The procedures allow the deposition of sensing layers before membranes have been etched, which leads to gas micro-sensors with an excellent fabrication yield. The response of the different microarrays to ethanol, acetone and ammonia vapours and their binary mixtures was studied. The influence of ambient humidity was also investigated. By using an integrated array of four integrated microsensors operated at two different temperatures and a fuzzy ARTMAP neural network, it was possible to simultaneously identify and quantify the different species (success rate was 100%). These results confirm the viability of the techniques introduced to obtain micromachined sensors suitable for battery-powered gas/vapour monitors.

This paper describes how thick-film technology was used to fabricate small, robust, sensitive, and selective semiconductor metal oxide (SMO) sensors to detect traces of ethanol vapours in air. The sensing parameters of several active layers were studied including: sensitivity, response repeatability, stability and selectivity. The response to different species of five differently doped SnO2 sensors and a commercially available one were measured at concentrations between 1 ppb and 1000 ppm. Interaction was extremely high when ethanol came into contact with the Pt-doped SnO2 surface. The change in resistance of the Pt-doped sensors was between 2 and 55 times higher than the change in the commercial one. The Pt-doped SnO2 material is less resistant, more sensitive and shows faster response to ethanol than pure SnO2. Since the detection limit for ethanol is at sub-ppb level, the fabricated sensors could be used for alcoholmeters or for on-line monitoring and controlling the concentration of ethanol in fruit ripening storage chambers.

Considerable attention has been given to strategies for variable selection in gas analysis using multisensor systems. Of the stochastic methods, genetic algorithms (GAs) have been found to be useful for variable selection, although they do not prevent irrelevant variables from being selected. Here we introduce the cascaded genetic algorithm, a new and simple approach that removes initially selected irrelevant variables. We coupled the method to the fuzzy ARTMAP classifier and used it to analyze single vapors and binary mixtures of three volatile organic compounds using a 12-element metal oxide gas sensor array. The method allowed to reduce the number of variables used to build the classifiers from 120 down to nine, which significantly increased the generalization ability of the fuzzy ARTMAP classifier. Success rates of 91.67 and 88.33% were reached in the simultaneous identification, quantification of single vapors, and in the identification of single vapors and their binary mixtures, respectively.

In this work, we study the influence of the introduction method of Pt atoms on the sensitivity to traces of ethanol of Pt-doped SnO2 sensors. The tin oxide films were obtained by a screen-printing process. Two different methods were employed to introduce Pt atoms on SnO2 films. In the first one, the Pt atoms were added to the screen-printed tin oxide layer by using RF magnetron sputtering and a subsequent thermal treatment. The second method consisted of mixing SnO2 and Pt pastes before the screen-printing process. The different active layers (including un-doped tin oxide) were carefully examined relative to their sensitivity to ethanol at different working temperatures. Sensors prepared by the second method showed sensitivity to ethanol four times higher than one of the sensors prepared by the first method and 12 times higher than un-doped sensors. XPS and scanning electron microscopy (SEM) measurements showed that this behaviour could be associated with the spatial distribution of the doping elements within the tin oxide film. While in Pt-sputtered sensors most of the Pt atoms were found at the surface of the active layer, for the sensors made by mixing Pt and SnO2 pastes, a homogeneous distribution of the Pt atoms was observed. These sensors show high sensitivity and fast response time to ethanol vapours, with a detection limit in the ppb range.

By means of a screen-printing technique, sensitive layers of nanopowder tin oxide were deposited on silicon micromachined substrates. The thickness of the sensing layers was 5 μm and particle size was around 40 nm. Each chip contains four thin silicon nitride membranes, on the centre of which a polysilicon heating resistor, insulating layers, platinum electrodes and sensitive layer are stacked. Unlike in previously reported works, the technological procedure reported here allows the deposition of the sensing layers before the membranes have been etched. This avoids damaging the membranes during film deposition, which leads to gas sensor microsystems with an excellent fabrication yield. The deposition method overcomes disadvantages such as low porosity and low surface area, generally associated to chemical vapour deposition (CVD) or sputtering methods, and keeps power consumption low (80 mW for a working temperature of 480 °C). As an example, the sensor response to ethanol, acetone and ammonia vapours and their binary mixtures was studied. The sensors were very sensitive to ammonia vapours. The influences of the sensor operating temperature and the electrode geometry were also investigated. By using an integrated array of four microsensors operated at two different temperatures and a fuzzy ARTMAP neural network, it was possible to identify the different species measured (success rate was higher than 91%). It was also possible to determine the concentration of the samples with a success rate higher than 84%. These results confirm the viability of the technique introduced to obtain micromachined sensors and sensor microsystems suitable for battery-powered gas/vapour monitors.

To gain some insight into the conductance response of temperature-modulated metal oxide gas sensors, we introduce a model for the physicochemical processes involved in the sensing operation. For this, we consider the interactions that take place at the sensor surface in the presence of reducing and oxidising species. Then we validate the model against experimental responses in the presence of ppm levels of CO and NO 2 in air. A sinusoidal voltage drives a resistive platinum heater and modulates the temperature of a micromachined tin oxide gas sensor; the resulting variation in conductance is analysed. Excellent agreement between theoretical and experimental responses is achieved. The model developed was used to compute the conductance response of a temperature-modulated sensor in the presence of different concentrations of CO and NO 2 . Features from the simulated response transients were extracted using the discrete wavelet transform and classified using a principal component analysis. A linear separation between CO and NO 2 was obtained, which is in good agreement with our previous experimental results.

In this work, we analysed the effects produced by the presence of humidity on the discriminating ability of a single resistive gas sensor working under the dynamic operation mode. The gases discriminated were CO, NO 2 and their mixtures. The working temperature of the sensor was modulated between 190 and 380 °C and its dynamic response to different concentrations of CO, NO 2 and CO+NO 2 mixtures was monitored either at 0 or 50% relative humidity (at 30 °C). Either the fast Fourier transform (FFT) or the discrete wavelet transform (DWT) was used to extract important features from the sensor response. These features were then input to different (statistical and neural) pattern recognition methods. The species considered could be discriminated with a success rate of 85.5% using a radial basis function neural network, which is very similar to the success rate (91.1%) reached in the classification of measurements performed in dry conditions.

Comprehensive gas chromatography - mass spectromety (GCxGC-MS) has become a promising tool in metabolomics. However, algorithms for GCxGC-MS data processing are needed in order to automatically process the data and extract the most pure information about the compounds appearing in the complex biological samples. This study shows the capability of orthogonal signal deconvolution (OSD), a novel algorithm based on blind source separation, to extract the spectra of the compounds appearing in GCxGC-MS samples. Results include a comparison between OSD and multivariate curve resolution - alternating least squares (MCR-ALS) with the extraction of metabolites spectra in a human serum sample analyzed through GCxGC-MS. This study concludes that OSD is a promising alternative for GCxGC-MS data processing.

We demonstrate that NO2 can be quantitatively analysed in the presence of CO using a single tungsten oxide based resistive gas sensor. The working temperature of the sensor was modulated between 190 and 380 degrees C and its dynamic response to different concentrations of CO, NO2, and CO + NO2 mixtures was monitored. Either the fast Fourier transform (FFT) or the discrete wavelet transform (DWT) was used to extract important features from the sensor response. These features were then input to different (statistical and neural) pattern recognition methods. The species considered can be discriminated with a success rate higher than 90% using a Fuzzy ARTMAP or a radial basis function neural network. The concentrations of the gases studied can be accurately predicted, by using the DWT coupled to partial least squares (PLS) models. The correlation coefficients of the predicted versus real concentrations were 0.923, 0.870 and 0.866 for CO, NO2, and NO2 in CO + NO2 mixtures, respectively.

It is shown that a single thermally-modulated tin oxide based resistive microsensor can discriminate between two different pollutant gases (CO and NO2) and their mixtures. The method employs a novel feature-extraction and pattern classification method, which is based on a 1-D discrete Wavelet transform and a Fuzzy ARTMAP neural network. The wavelet technique is more effective than FFT in terms of both data compression and drift rejection. Furthermore, Fuzzy ARTMAP networks lead to a 100% success rate in gas recognition in just 2 training epochs, which is significantly lower than the number of epochs required to train the back-propagation network.

One of the main challenges in nuclear magnetic resonance (NMR) metabolomics is to obtain valuable metabolic information from large datasets of raw NMR spectra in a high throughput, automatic, and reproducible way. To date, established software packages used to match and quantify metabolites in NMR spectra remain mostly manually operated, leading to low resolution results and subject to inconsistencies not attributable to the NMR technique itself. Here, we introduce a new software package, called Dolphin, able to automatically quantify a set of target metabolites in multiple sample measurements using an approach based on 1D and 2D NMR techniques to overcome the inherent limitations of 1D (1)H-NMR spectra in metabolomics. Dolphin takes advantage of the 2D J-resolved NMR spectroscopy signal dispersion to avoid inconsistencies in signal position detection, enhancing the reliability and confidence in metabolite matching. Furthermore, in order to improve accuracy in quantification, Dolphin uses 2D NMR spectra to obtain additional information on all neighboring signals surrounding the target metabolite. We have compared the targeted profiling results of Dolphin, recorded from standard biological mixtures, with those of two well established approaches in NMR metabolomics. Overall, Dolphin produced more accurate results with the added advantage of being a fully automated and high throughput processing package.

In this work, an electronic nose is used to assess the ripeness state of pinklady apples through their shelf-life. In order to evaluate the electronic nose performance, fruit quality indicators are also obtained to compare results from both techniques. Pinklady apples were harvested at their optimal date so that electronic nose measurements and fruit quality measurements could be performed on the fruit samples during their ripening process. A PCA analysis, a well-known linear classification technique, does not show any clustering behaviour that might be attributed to ripening. On the other hand, Fuzzy art, an unsupervised neural network classification algorithm, shows a tendency to classify measurements regarding to their shelf-life period. Finally, electronic nose signals are correlated with classical fruit quality parameters such as firmness, starch index and acidity. Good correlation coefficients are obtained, a clear indication that electronic nose signals are related to the ripening process of apples.

Electronic noses consist of an array of non-selective gas sensors with a pattern recognition engine. The sensors and pattern recognition methods depend on the specific application. The design process of electronic noses usually involves time-consuming measurements in a non-standard trial and error process. This paper addresses the problem by using an electronic nose simulation tool centred on PSpice. Using previously developed generic PSpice models for the response of gas sensors, a four-element tin oxide sensor array was simulated. The sensor model parameters were adjusted using calibrated response data from ethanol, methane and their mixtures, detected with real tin oxide sensors. To study the performance of the array, statistical error modelling and PSpice simulations were used in a Monte Carlo analysis coupled with principal component analysis. The results show that this simulation strategy is useful for analysing the effects of sampling errors, the changes in operation temperature, random errors and sensor drift. The importance of these errors is discussed in terms of the array discrimination ability. We conclude that this simulation strategy can help to systematise the design of electronic noses.

In this paper we present a new PSpice behavioural model of metal oxide resistive gas sensors. The sensor model, which is based on previously developed theoretical and experimental analyses, depicts the static and dynamic responses of the sensors to multicomponent gas mixtures. The dynamic response is obtained either when the gas concentration or the operating temperature of the sensor are modulated. The model developed is in good agreement with experimental data. PSpice also gets equivalent results up to 10 times faster (on the same PC platform) than when the diffusion-reaction equations that model the response of sensors are solved numerically.