Rapid differentiation between E. coli and Salmonella Typhimurium

Sensors and Actuators B 133(2008)414–419

Contents lists available at ScienceDirect

Sensors and Actuators B:

Chemical

j o u r n a l h o m e p a g e :w w w.e l s e v i e r.c o m /l o c a t e /s n

b

Rapid differentiation between E.coli and Salmonella Typhimurium using metal oxide sensors integrated with pattern recognition

Ubonrat Siripatrawan ?

Department of Food Technology,Faculty of Science,Chulalongkorn University,Bangkok 10330,Thailand

a r t i c l e i n f o Article history:

Received 18October 2007

Received in revised form 28February 2008Accepted 29February 2008Available online 6March 2008Keywords:

Metal oxide sensors

Back-propagation neural network Principal component analysis E.coli

Salmonella Typhimurium

a b s t r a c t

A rapid method to differentiate between E coli and Salmonella Typhimurium was developed.E.coli and S.Typhimurium were separately grown in super broth and incubated at 37?C.Super broth without inoc-ulation of E.coli or S.Typhimurium was used as control.Numbers of E.coli and S.Typhimurium were followed using a colony counting method.Identi?cation of the volatile metabolites produced by E.coli and S.Typhimurium was determined using solid-phase microextraction coupled with gas chromatogra-phy/mass spectrometry.An electronic nose with 12non-speci?c metal oxide sensors was used to monitor the volatile pro?les produced by E.coli and S.Typhimurium.Principal component analysis (PCA)and back-propagation neural network (BPNN)were used as pattern recognition tools.PCA was used for data exploration and dimensional reduction.PCA could visualize class separation between sample subgroups.The BPNN was shown to be capable of predicting the number of E.coli and S.Typhimurium.Good pre-diction was possible as measured by a regression coef?cient (R 2=0.96)between true and predicted https://www.360docs.net/doc/60194555.html,ing metal oxide sensors and pattern recognition techniques,it was possible to discriminate between samples containing E.coli from those containing S.Typhimurium.

?2008Elsevier B.V.All rights reserved.

1.Introduction

Detection and differentiation of foodborne pathogens is a major concern of the food industry.The public health implications of fail-ing to detect certain pathogens can cause disease or even death [1,2].Classical methods for the detection and differentiation of microorganisms have been based on traditional plate count meth-ods.These methods are both tedious and time consuming because they usually require a series of tests with the incubation of the microorganisms.The limitations of these methods have led to the research focusing on development of rapid and accurate techniques to identify pathogens in food products [3].Microorganisms can be characterized by identi?cation of speci?c metabolites generated by speci?c biochemical pathways [4,5].However,many metabo-lites may be common to several microorganism species.Therefore,the differences between samples often relate to a complex balance between patterns of volatiles rather than to a major change in one or two constituents [6,7].

Electronic nose technology has been shown to provide rapid,continuous monitoring of a wide array of different volatile chem-icals [8].A considerable number of electronic nose applications have been reported,including evaluation of the off-odor in wine

?Tel.:+6622185515;fax:+6622544314.E-mail address:ubonratana.s@chula.ac.th .

[9],detection of Salmonella Typhimurium in fat free milk [10],classi?cation of milk [11],and detection of foodborne pathogens in vegetables [12].The electronic noses can be used for the spe-ci?c analysis,identi?cation,and recognition of complex odors and volatile organic compounds [9].However,the electronic noses gen-erated multi-dimensional data that was dif?cult to handle and visualize.This underlines the importance of using multivariate data analysis to extract the speci?c information necessary to tar-get a speci?c microorganism.Pattern recognition interpretation techniques,such as principal component analysis (PCA),linear discriminant analysis (LDA)and arti?cial neural network (ANN),provide complementary information which was simply unachiev-able by conventional data analysis [13,14].

PCA is an unsupervised technique commonly used in signal processing and pattern recognition [15,16].PCA is used to reduce the dimensionality of multivariate data while preserving most of the variance,and thus is an excellent technique for observing the natural relationships between samples [17–19].Arti?cial neural network has been proved to be powerful tools for data processing [11,20].The advantages of the arti?cial neural network approach to the multivariate calibration of sensor arrays are well known.The most common neural network approach to regression-type problems is multilayer perceptrons with back-propagation neural network (BPNN).The objective is to identify a model that will correctly associate inputs with outputs.The learning data is used to train the system and to develop the calibration model.The

0925-4005/$–see front matter ?2008Elsevier B.V.All rights reserved.doi:10.1016/j.snb.2008.02.046

U.Siripatrawan/Sensors and Actuators B133(2008)414–419415

test data are then evaluated by the calibration model in order to obtain the predicted results.This technique has the advantage of producing low prediction errors[20,21].

The objective of this work was to develop a method to iden-tify and differentiate between E.coli and S.Typhimurium using12 non-speci?c metal oxide sensors integrated with PCA and BPNN.In this approach,identi?cation and differentiation was based on the determination of the volatile metabolites that signify a particular microorganism.E.coli and S.Typhimurium were used as the tar-get microorganisms in this study because outbreaks of foodborne illness caused by S.Typhimurium have been repeatedly reported [1,22,23],and the presence of E.coli in foods indicates fecal con-tamination and the presence of pathogenic microorganisms.

2.Methodology

2.1.Preparation of stock culture and test solution

The nonpathogenic strain E.coli ATCC25922obtained from the American Type Culture Collection(ATCC,Rockville,MD)was inoculated into Luria Bertani broth consisting of10g Bacto tryp-tone(Becton,Dickinson and Company,Sparks,MD),5g Bacto yeast (Becton,Dickinson and Company,Sparks,MD),and5g NaCl,and incubated at37?C in a gyratory shaker(G-25New Brunswick Sci-enti?c Corporation,New Brunswick,NJ)at100rpm.The E.coli suspension was then dispensed into sterile125ml polypropylene centrifuge bottles,and centrifuged(Rotor-GSA model RC5C Super-speed Centrifuge,Sorvall Instruments,Dupont Co.,Hoffman Estate, IL)at1600×g for10min.The supernatant was decanted,and the resulting cell pellets were resuspended in a sterile15%glycerol solution.One ml aliquots of bacterial suspension were transferred into1.5ml microcentrifugal tubes and frozen using liquid nitrogen. The frozen cultures were stored at?80?C.

Super broth consisting of32g tryptone,20g yeast,5g NaCl,and 5ml of1N NaOH per liter was used as a basal medium.Before being utilized,E.coli was transferred from the stock culture to super broth and incubated overnight at37?C.E.coli(100CFU/ml)was inocu-lated into super broth,and5ml of super broth were transferred into standard20ml headspace vials and sealed with PTFE-lined Te?on caps(Alpha M.O.S.,Hillsborough,NJ).The cultures were allowed to grow in vials at37?C in a gyratory shaker(G-25New Brunswick Scienti?c Corporation,New Brunswick,NJ)at100rpm.Test solu-tion of S.Typhimurium was prepared in the same manner.Samples were periodically analyzed after incubation at37?C for4,6,8and 10h using colony counting,solid-phase microextraction coupled with gas chromatography/mass spectrometry(SPME–GC–MS)and an electronic nose.

2.2.Colony counting method

Samples were serially diluted in sterile Butter?eld’s phosphate buffer.A series of dilutions was prepared from the stock suspen-sion.Serially diluted samples were plated in duplicate using3M Petri?lm aerobic count plates(3M Industrial Markets,St.Paul,MN) to determine total bacteria.All plates were incubated at35?C for 48±2h.Plate counts were recorded as CFU/ml.

2.3.Identi?cation of volatile compounds using SPME–GC–MS

Volatile compounds from headspace samples were collected using a polydimethylsiloxane solid-phase microextraction?ber (Supelco,Inc.,Bellefonte,PA).For headspace sampling,5ml of a liq-uid sample were placed into a20ml vial and the?ber was exposed to the head space of the media solution.Sampling temperature was maintained at37?C,and the sampling duration was10min,which Table1

Sensor types and volatile descriptors

Sensors Volatile description

P101,P102,SYGCT Non-polar volatiles:methane,propane

PA2,SYAA,T301Organic solvents:polar compounds,ethanol

T702Alcohol and aromatic compounds

SYG,SYGH Amines and amine containing compounds and ammonia

derivatives

SYGCTI Ammonia and sulfur

P401,SYLG Fluoride and chloride:?uorinated and chlorinated

compounds,aldehydes

was suf?cient to permit the establishment of a near equilibrium for the compounds tested.A GC(HP-6890,Hewlett-Packard Co.,Wilm-ington,DE)was used for the analysis of the compounds.Volatiles were separated using a capillary column(SPB5,30m×0.1mm i.d., 0.25?m coating thickness).The carrier gas used was ultrapuri?ed helium(99.99%purity)at a?ow rate of0.5ml/min.The tempera-ture program was isothermal for2min at40?C and raised to240?C at a rate of50?C/min.Electron impact ionization(FCD-650,LECO Corp.,St Joseph,MI)was used by the time-of-?ight(TOF)mass spec-trometer in evaluation of the volatiles.Mass spectra were collected at a rate of40spectra/s over a range of30–400m/z.The ioniza-tion energy was70eV.Identi?cation of volatile components was determined by comparison of collected mass spectra with those of authenticated standards and spectra in the National Institute for Standards and Technology(NIST)mass spectral library.

2.4.Volatile analysis using metal oxide sensors

An electronic nose(Fox3000,Alpha M.O.S.,Hillsborough,NJ) with12metal oxide sensors(SYLG,SYG,SYAA,SYGH,SYGCTI,SYGCT, T301,P101,P102,P401,T702,and PA2)was used for monitoring changes in volatiles produced by E.coli in a super broth medium. The descriptors associated with the sensors are shown in Table1. The volatile analysis system combines a measurement chamber for generating the volatile compounds and a detection system made up of12metal oxide sensors.This instrument was linked to an auto-sampler capable of analyzing a total of64samples.Samples were placed in glass vials and sealed with crimped PTFE/metal septa.

Samples were placed in the HS100auto-sampler in arbitrary order.Prior to analysis,the vial was removed from the sample tray and placed in a temperature-controlled chamber.The vial temper-ature was held at37?C while being spun in order to produce an equilibrated headspace.The time the vial remains in this chamber is the headspace generation time.The automatic injection unit heats the samples to37?C using an incubation time of300s.The temper-ature of the injection syringe was42?C.The injector needle then removes2.5ml of headspace and injects it into the sensor chamber. The delay time between two injections was300s.Each injection was repeated,with separate samples(three times for all variations per day)for seven days.The electronic signals from the sensors were digitized and then transferred to the control computer.

Each sensor element changes in resistance(R max)when exposed to volatile compounds.In order to produce consistent data for the classi?cation,the sensor response was presented with a volatile chemical relative to the base resistance in air,which is the maxi-mum change in the sensor electrical resistance divided by the initial resistance,as follows

Relative resistance change=

R max

R0(1) where R max=R max?R0is the maximum change in the sensor electrical resistance and R0the initial baseline resistance of the sensor.The baseline of the sensors was acquired in a synthetic air

416U.Siripatrawan/Sensors and Actuators B133(2008)414–419

Table2

Growth of E.coli and S.Typhimurium in super broth

Incubation time(h) E.coli log(CFU/ml)S.Typhimurium log(CFU/ml)

4EC1 3.54±0.49ST1 3.67±0.30 6EC2 5.68±0.26ST2 6.44±0.37 8EC38.72±0.38ST37.30±0.54 10EC49.83±0.21ST49.49±0.42

saturated steam water at?xed temperature.The relative resistance change value was used for data evaluation because it gives the most stable result,and is more robust against sensor baseline variation [12].

2.5.Data analysis

Data were made up of108samples from9subgroups(super broth(control,CTR)and super broth inoculated with E.coli(EC)or S.Typhimurium(ST)).EC1,EC2,EC3and EC4were samples with E.coli incubated at37?C for4,6,8and10h,respectively.ST1,ST2, ST3and ST4were samples with S.Typhimurium incubated for4, 6,8and10h,respectively.Each subgroup had12replicate samples collected from several cultivations.Each sample was analyzed using 12metal oxide sensors.The sensor responses of all108samples were arranged in a108×12matrix.

Data reduction was performed using PCA.Prior to PCA,the data matrix(108samples×12measurement variables)was standard-ized to zero mean and unit variance.This procedure ensures that the variables have the same weighting in the principal component(PC) model.The whole of the standardized data matrix was then submit-ted to the PCA.PCA mathematically decomposes independent data, X matrix into two latent matrices.These latent variables are con-structed in such a way that they contain the maximum amount of information given by the original variables.The latent variables can be written as a weighted sum of the original variables,as follows X=ZP T+E(2) where Z is a scores matrix,P is a loadings matrix(the superindex T indicates the transposed matrix),and E is the residual error matrix [18,26].

For the BPNN algorithm,the input data can be conductometric responses or,as in this study,scores from the PCA,and the output data were the number of E.coli and S.Typhimurium in the sam-ples.The data sets were partitioned into2subsets:a training set (70%of the data)used to adjust the network weights,and a valida-tion set(the remaining30%of the data)used to validate the model. First,data patterns of known identity were presented to the neu-ral network as exemplars.The network was trained by comparing predicted E.coli and S.Typhimurium numbers with their known identify.The neural network was trained using the Bayesian reg-ularization algorithm.Once the neural network was trained,any data pattern could be presented to it and the output analyzed to ?nd the most likely identity of the pattern.The performance of the neural network was con?rmed by measuring its performance on an independent set of data called a validation set.All calculations were carried out using MATLAB5.2routines written by the author,and making use of the toolbox provided by Mathworks(Mathworks, Inc.,Natick,MA,USA).

3.Results and discussion

3.1.Colony counts

The population of E.coli and S.Typhimurium grown in closed vials containing super broth is shown in Table2.The data were from 12replicates.Cultures exhibited exponential growth and numbers increased rapidly from～102to～109CFU/ml.

3.2.SPME–GC–MS

From library matches,the speci?c compounds produced by E. coli grown in super broth included dimethyl disul?de,ethanol, 2-nonanone,2-heptanone,pentyl-cyclopropane and indole.The volatile compounds signifying S.Typhimurium included hydrogen sul?de,ethanol,carbon disul?de,dimethyl cyclopropane and1-propanol.The basal media used in this research was super broth, which has a high concentration of tryptophan.Trytophan is broken down into indole due to the enzymatic activity(tryptophanase)of E.coli.Tryptophanase degrades tryptophan to indole,pyruvate and ammonia[24].Salmonella is negative for indole but can produce hydrogen sul?de.The alcohols detected in the samples are prob-ably from amino acid decarboxylation and deamination.Dimethyl disul?de may be present due to degradation of sulfur containing amino acids such as methionine and cysteine[25].As a result of metabolic activity,microorganisms produce various metabolites and intermediates in their growth environment.The results from SPME–GC–MS showed that there were differences in the volatile organic compounds produced by E.coli and S.Typhimurium.

3.3.Metal oxide sensors

The data matrix comprised108samples from9subgroups(CTR, EC1,EC2,EC3,EC4,ST1,ST2,ST3and ST4)as analyzed using the12 sensors(SYLG,SYG,SYAA,SYGH,SYGCTI,SYGCT,T301,P101,P102, P401,T702,and PA2).Fig.1shows the average responses of all sam-ples in9subgroups to the12sensors.The patterns of the samples differ due to the different patterns of volatile metabolites gener-ated by E.coli and S.Typhimurium produced during the incubation period.The control sample(super broth without inoculation of E. coli or S.Typhimurium)had

low conductometric response from the sensor array.Different sensor signal intensities between samples are important to discriminate between samples.

The sensor array has the potential to be a sensitive,fast,one-step method to characterize volatile compounds in a sample.However, the electronic sensor array generated multi-dimensional data that

Fig.1.The average responses of all samples in9subgroups to the12sensors.The samples are labeled using the following scheme:CTR is uninoculated samples;EC1, EC2,EC3and EC4are samples inoculated with E.coli and incubated for4,6,8and 10h,respectively;ST1,ST2,ST3and ST4are samples inoculated with S.Typhimurium and incubated for4,6,8and10h,respectively.

U.Siripatrawan/Sensors and Actuators B133(2008)414–419

417

Fig.2.PCA loadings plot of the data from electronic sensors with percent of total variance explained by PC180.87%and PC212.79%.

was dif?cult to handle and visualize.Therefore,PCA and BPNN were used as pattern recognition technique to discriminate between dif-ferent sample groups based on conductometric responses.

3.4.PCA

When the data matrix was submitted to PCA,the number of PCs,PC loading,and PC scores were calculated.To?nd each suc-ceeding axis,two conditions were imposed:(1)as much variance as possible is accounted for by each PC,and(2)the newest axis is mutually orthogonal to the set of axes already located.In this study, two PCs were calculated and included in the model.The number of principal components is the number of substantively meaningful independent(uncorrelated)patterns,among the variables.The?rst PC accounts for a major fraction(80.87%)of the total variance of the data.The second principal axis is orthogonal to the?rst eigenvector and accounts for12.79%of the variation not accounted for by the ?rst PC.These two PCs together represent94%of the information in the overall data set.

PCA decomposes the data matrix(108×12)into two smaller matrices of PC scores and loading scores,and displays them in a loading plot(Fig.2)and score plot(Fig.3).The loading plot displays the relationship between the descriptor or measurement variables (12sensors).The score plot shows the relationship between objects (108samples).

The loading display(Fig.2)facilitates an exploration of the con-tributions of the variables to each PC from the positive and negative values of the weights.All sensors are important because of their high negative/positive loadings.High positive loadings of SYLG, P101,P102,P401,PA2,T301,and T702responses relative to those with high negative loadings of SYGH,SYGCT,SYGCTI on PC1dis-criminated samples with different number of the microorganisms. On the other hand,sensors that have high positive loadings(SYAA, SYG,SYGCT,SYGCTI,P101,P102and T301)and high negative load-ings(SYGH and SYLG)on PC2are suitable to discriminate samples with EC from those with ST.

Patterns can be separated from each other using a scatter plot of principal component scores.The score plot in Fig.3shows6distinct clusters with positive and negative scores for both PCs.Samples in the same subgroups are grouped together under the same area in the PCA score plot.Samples having divergent numbers are

located Fig.3.PCA score plot of samples.The samples are labeled using the scheme given in the caption of Fig.1.

further apart.The control media and nutrient media inoculated with E.coli and S.Typhimurium incubated for4h overlapped.The medium is indistinguishable from the early inoculants(the num-ber of E.coli and S.Typhimurium were3.54and3.67log(CFU/ml), respectively);all of these points are grouped under the same area in the PCA score plot.This indicates that the electronic sensors were not able to detect a difference in the volatile metabolites at the early growth stage.Only after4h of incubation do the sensor responses become statistically distinguishable from the control medium.As a result,in the method reported here,E.coli and S.Typhimurium could be detected and differentiated when the number was above 5log(CFU/ml).Groups of E.coli and S.Typhimurium samples incu-bated for6,8,and10h occupied different areas,indicating that the electronic sensors could differentiate between these samples.

PCA is ideal for preliminary evaluation of the data in that it examines only the measured variables in order to perform the anal-ysis,but does not take the class structure of the data into account. PCA could qualitatively differentiate the sample but did not provide quantitative information of the number of the microorganisms. Therefore,BPNN was used to predict the number of E.coli and S. Typhimurium present in the sample.

3.5.BPNN

The network architecture created for the E.coli and S. Typhimurium data matrix includes an input layer,one hidden layer of neurons,and an output layer,with a hyperbolic tangent sigmoidal transfer function in the hidden layer and a linear transfer function in the output layer(Fig.4).The indices j,k,and l refer to the

input

Fig.4.The architecture of BPNN.

418U.Siripatrawan /Sensors and Actuators B 133(2008)414–419

signals (j =1,...,m )in the input layer,the neurons (k =1,...,p )in the hidden layer,and the neuron (l =1,...,q )in the output layer,respectively.There were 2neurons (m =2)in the input layer,2neu-rons (p =2)in the hidden layer,and one neuron (q =1)in the output layer.The transfer function,f (v k ),in the hidden layer was a hyper-bolic tangent (Eq.(3))and a linear function was used in the output layer (Eq.(5)).y k =f (v k )=

1?exp(v k )1+exp(v k )

(3)

with v k being computed as v k =b k +

m j =1

w kj x j

(4)

where y k is the output of the hidden layer,f (v k )is the transfer func-tion associated with the neuron k in the hidden layer,v k is the sum

of weighted input of neuron k ,b k is the bias,and x j is the input https://www.360docs.net/doc/60194555.html,e of bias b k has the effect of applying an af?ne transformation in the model.A linear function can be calculated as follows.y l =f (v l )=b l +

p k =1

w lk y k

(5)

where y l is the output of the output layer,f (v l )is the transfer func-tion associated with neuron l in the output layer,y k is the input to the neuron l ,v l is the sum of weighted input of neuron l ,b l is the bias,and w lk is the weight connection of neuron k and neuron l .

The BPNN was trained using selected parameters from data sets from several cultivations and was subsequently validated using independent data sets.The training started with different initial random weights,and was optimized during training.The differ-ence between the target value and neural network output was propagated back through the network to the input.The error was minimized by adjusting the weight.The learning process contin-ued epoch-by-epoch until the synaptic weights and bias level of the network stabilized and the average square error over the entire training set converged to the minimum value.

The performance of the trained network was evaluated by mea-suring the errors in the validation sets.A low mean square error (MSE =0.126)between the predicted and measured values indicates that there was low error when using BPNN to predict the number of E.coli and S.Typhimurium.Fig.5shows the predictions

versus

Fig.5.The predicted and true numbers of E.coli and S.Typhimurium using BPNN.

true values of numbers of E.coli and S.Typhimurium in the sam-ples using metal oxide sensors and BPNN algorithm.The correlation coef?cient,R 2,between the outputs and targets was a measure of how well the variation in the output was explained by the targets and outputs.A regression coef?cient (R 2=0.96)indicates a good ?t between actual and predicted data.This study has shown that electronic nose analysis results and application of neural network can provide accurate information on the number of E.coli and S.Typhimurium.4.Conclusion

E.coli and S.Typhimurium could be identi?ed by their volatile metabolites which re?ect the metabolism of individual species.The electronic nose appears to be able to discriminate between both species.PCA could visualize class separation between different sample subgroups.The BPNN could provide accurate information on the number of E.coli and https://www.360docs.net/doc/60194555.html,pared to colony counting methods,the electronic nose integrated with pattern recognition technique can accomplish in less time.The system has potential to be applied to food products or incorporated into safety and quality control systems in the food industries.Acknowledgements

The author is highly grateful to Prof.Dr.Bruce R.Harte,the School of Packaging,Michigan State University for providing the metal oxide sensors and Prof.Dr.John E.Linz,Department of Food Science and Human Nutrition,Michigan State University for the microbiological facility.References

[1]Centers for Disease Control and Prevention.Infections Associated with Eat-ing Seed Sprouts.Available from:https://www.360docs.net/doc/60194555.html,/ecoli/2006/september/,2006.

[2]https://www.360docs.net/doc/60194555.html,zcka,F.Javier Del Campo,F.Xavier Munoz,Pathogen detection:a per-spective of traditional methods and biosensors,Biosens.Bioelectron.22(2007)1205–1217.

[3]Council for Agricultural Science and Technology,Foodborne Pathogens:

Review of Recommendations.Available from:https://www.360docs.net/doc/60194555.html,/ncidod/EID/vol9no4/99-0519.html ,1999.

[4]O.Canhoto,F.Pinzari,C.Fenelli,N.Magan,Application of electronic nose tech-nology for the detection of fungal contamination in library paper,Int.Biodeter.Biodegr.54(2004)303–309.

[5]L.Marilley,S.Amupuero,T.M.G.Zesiger,Casey,Screening of aroma-producing

lactic acid bacteria with electronic nose,Int.Dairy J.14(2004)846–849.

[6]T.Borjesson,Electronic nose for odor classi?cation of grains,Cereal Chem.73

(1996)457–461.

[7]K.E.Sapsford,M.M.Ngundi,M.H.Moore,https://www.360docs.net/doc/60194555.html,ssman,L.C.Shriver-Lake,C.R.

Taitt,F.S.Ligler,Rapid detection of foodborne contaminants using an array biosensor,Sens.Actuators B:Chem.113(2006)599–607.

[8]M.P.Mart,O.Busto,J.Guasch,R.Boqu ′e,

Electronic noses in the quality control of alcoholic beverages,Trends Anal.Chem.24(2005)57–66.

[9]J.A.Ragazzo-Sanchez,P.Chalier,C.Ghommidh,Coupling gas chromatography

and electronic nose for dehydration and desalcoholization of alcoholized bev-erages application to off-?avour detection in wine,Sens.Actuators B:Chem.104(2005)301–308.

[10]https://www.360docs.net/doc/60194555.html,kshmanan,R.Guntupalli,J.Hu,V.A.Petrenko,J.M.Barbaree,B.A.Chin,

Detection of Salmonella typhimurium in fat free milk using a phage immobilized magnetoelastic sensor,Sens.Actuators B:Chem.126(2007)544–550.

[11]K.Brudzewski,S.Osowski,T.Markiewicz,Classi?cation of milk by means of an

electronic nose and SVM neural network,Sens.Actuators B:Chem.98(2004)291–298.

[12]U.Siripatrawan,Self-organizing algorithm for classi?cation of packaged fresh

vegetable potentially contaminated with foodborne pathogens,Sens.Actuators B:Chem.128(2008)435–441.

[13]A.Hernandez Gomez,G.Hu,J.Wang,A.Garc ′ya

Pereira,Evaluation of tomato maturity by electronic nose,Comput.Electron.Agric.54(2006)44–52.

[14]U.Siripatrawan,J.E.Linz,B.R.Harte,Electronic sensor array coupled with arti?-cial neural network for detection of S.Typhimurium,Sens.Actuators B:Chem.119(2006)64–69.

[15]J.Goschnick,D.Haeringer,I.Kiselev,Multicomponent quanti?cation with a

novel method applied to gradient gas sensor microarray signal patterns,Sens.Actuators B:Chem.127(2007)237–241.

U.Siripatrawan/Sensors and Actuators B133(2008)414–419419

[16]S.Kokot,M.Grigg,H.Panayioton,T.D.Phunong,Data interpretation by

some common chemometrics methods,Electronanlysis10(1998)1081–1088.

[17]T.Aishima,Correlating sensory attributes to gas chromatography–mass spec-

trometry pro?les and e-nose responses using partial least squares regression analysis,J.Chromatogr.A1054(2004)39–46.

[18]M.O’Connell,G.Valdor,G.Peltzer,R.M.Negri,A practical approach for?sh

fresness determinations using a portable electronic nose,Sens.Actuators B: Chem.80(2001)149–154.

[19]E.Oja,A simpli?ed neuron model as a principal component analyzer,J.Math.

Biol.15(1982)267–273.

[20]M.A.Martin,J.P.Santos,J.A.Agapito,Application of arti?cial neural networks

to calculate the partial gas concentration in a mixture,Sens.Actuator B:Chem.

77(2001)468–471.

[21]S.Haykin,Communication System:A Comprehensive Foundation,second ed.,

Prentice Hall,New Jersey,1999.

[22]Economic Research Service,U.S.Department of Agriculture,Economics of food-

borne disease:Estimating the bene?ts of reducing foodborne disease.Available from:https://www.360docs.net/doc/60194555.html,/Brie?ng/FoodborneDisease/,2000.[23]J.Gill,W.E.Keene,J.E.Mohle-Boetani,J.A.Farrar,P.L.Waller,C.G.Hahn,R.P.

Cieslak,Alfalfa seed decontamination in Salmonella outbreak,Emerg.Infect.

Dis.9(2003)474–479.

[24]A.G.Moat,J.W.Foster,Microbial Physiology,second ed.,Willey&Sons,New

York,1988.

[25]U.Siripatrawan, B.R.Harte,Solid phase microextraction/Gas chromato-

graph/mass spectrometer integrated with chemometrics for detection of Salmonella Typhimurium contamination in a packaged fresh vegetable,Anal.

Chim.Acta581(2007)63–70.

[26]D.Lee,J.Jung,J.Lim,J.Hus,D.Lee,Recognition of volatile organic compounds

using SNO2sensor array and pattern recognition analysis,Sens.Actuators B: Chem.77(2001)228–236.

Biography

Ubonrat Siripatrawan obtained her Ph.D.degree in packaging from Michigan State University.She is an assistant professor in the Department of Food Technology, Faculty of Science,Chulalongkorn University.