文档分类：
在线文档经过高度压缩，下载原文更清晰。
淘豆网网友近日为您收集整理了关于Classification of apple fruits according to their maturity state by the pattern recognition analysis of their positions：苹果果实按成熟度状态的多酚类成分的模式识别分类分析的文档，希望对您的工作和学习有所帮助。以下是文档介绍：Classication of apple fruits according to their maturity state bythe pattern recognition analysis of their positionsRosa M. Alonso-Salces a, Carlos Herrero b, Alejandro Barranco a,Luis A. Berrueta a,*, Blanca Gallo a, Francisca Vicente aaDepartamento de Qumica Analtica, Facultad de Ciencias, Universidad del Pas Vasco/Euskal Herriko Unibertsitatea,P.O. Box 644, E-48080 Bilbao, SpainbDepartamento de Qumica Analtica, Nutricion y Bromatologa, Facultad de Ciencias, Universidad de Santiago postela,Augas Ferreas s/n, Campus Universitario, 27002 Lugo, SpainReceived 28 June 2004; received in revised form 13 October 2004; accepted 13 October 2004AbstractPolyphenolic proles of cider apple cultivars were studied in order to dierentiate fruits according to their maturity state (ripe orunripe). Thiolysis and direct solvent extracts of freeze-dried apple pulps and peels were analysed by HPLC-DAD. Univariate datatreatment did not achieve
thus a multivariate approach was considered. For each apple tissue data set, severalchemometric techniques were applied to the most discriminant variables. Cluster analysis allowed a preliminary study of the datastructure. Then, supervised pattern recognition procedures, namely linear discriminant analysis, K-nearest neighbours, soft inde-pendent modelling of class analogy, and multilayer feed-forward articial works (MLF-ANN), were used to develop deci-sion rules to classify samples in the established categories. Excellent results were aorded by MLF-ANN applied to theconcentrations of total procyanidins and (+)-catechin and the average degree of polymerisation of procyanidins in apple pulp, ess in the prediction ability of 97% and 99% for unripe and ripe categories, respectively. 2004 Elsevier Ltd. All rights reserved.Keywords: P A M Pattern
Chemometrics1. IntroductionBiotransformations of cider apple fruits have to be car-ried out when they satisfy certain technological qualitycriteria. In this sense, fruits have to be processed at theiroptimum maturity state, i.e., the time when they presentan adequate position, which is responsiblefor anoleptic and nutritional properties, as wellas for the characteristics of their derived products.Several works have been developed to study thechemical constituents of apple involved in the biochem-ical transformations that take place during maturation(Ackermann, Fisher, & Amado`, 1992; Blanco, Picinelli,Gutierrez, & Mangas, 1992b; Mangas et al., 1992), inorder to establish the ontogenic stage for technologicalpurposes. Signicant changes have been detected in ap-position in the last stages of ripening, such asan accumulation of sugars and pectins, a decrease inthe content anic acids and some amino acids,and a minimum in the total polyphenol and nitrogenconcentrations reached just before the nal accumula-tion of sugars and starch degradation (Blanco et al.,1992a).Traditional criteria used for evaluating the optimummaturity state of apples are the starch content (Le Lezec/$ - see front matter
2004 Elsevier Ltd. All rights reserved.doi:10.1016/j.foodchem.*Corresponding author. Tel.: +34 94 601 5505; fax: +34 94 464 8500.E-mail address: qapbesil@lg.ehu.es (L.A. Berrueta).ate/foodchemFood Chemistry 93 (–123FoodChemistry& Babin, 1988), the total sugar/total acidity ratio (Board& Woods, 1983), the internal ethylene content (Dilley,1981), the fruit consistency, as resistance to ration(Trillot, Masseron, & Tronel, 1993), and the content ofsoluble solids (Sieguist, 1987). However, other chemicalmarkers, closely related to the quality of apple products,such as sugars, organic acids, amino acids, total poly-phenols, and pectins, together with chemometric tech-niques, have allowed a more precise characterisationof the dierent apple varieties according to their degreeof ripening (Mangas, Moreno, Picinelli, & Blanco,1998).Polyphenols play an important role in the biotrans-formation process of apples since, not only do they con-tribute to cider avour and aroma (Lea, 1995), but theyalso control anism metabolism present in themedium, controlling fermentation rates (Cowan, 1999),avoiding the development of some faults in cider (Spon-holtz, 1993), participating in cider spontaneous clarica-tion, and inhibiting enzymatic systems such asclarication enzymes (Lea, 1990). Hence, it is importantto consider pounds in order to establish theoptimum maturity state of apples to be processed.Among the main classes of apple polyphenols, avan-3-ols are preponderant, being present in monomericforms named catechins, and in oligomeric and polymericforms, known as procyanidins. The latter contribute toastringency and bitterness of apples (Lea, 1990) andtheir derived products, and to the formation of hazesand precipitates during apple juice and cider storage,due to their ability to associate with proteins (McManus et al., 1985) and harides (Renard, Bar-on, Guyot, & Drilleau, 2001). Hydroxycinnamic acidsare the next major class. Together with catechins, theyare involved in the browning phenomena that take placeduring apple fruit processing, being responsible for theyellow or orange coloration of apple products (Amiot,hini, Aubert, & Nicolas, 1992). Dihydrochalcones,avonols and anthocyanins are ponents thatcontribute to the pigmentation of apples, and to thepotential antioxidant activity of apples and their derivedfoodstus (Ridgway, Oreilly, West, Tucker, & Wiseman,1996).Several studies of position during fruitdevelopment and maturation (Burda, Oleszek, & Lee,1990; Lancaster, 1992; Mayr, Treutter, Santos-Buelga,Bauer, & Feucht, 1995; Treutter, 2001) demonstrate thatphenolic contents of the dierent apple tissues are notexclusively dependent on the cultivar, but also on theirmaturity state. Polyphenolic proles vary during fruitgrowth and ripening. Thus, in the rst weeks of theirontogenesis, fruits present high levels of polyphenolsthat decrease throughout fruit development to a mini-mum, preceding the nal accumulation of sugars inthe last stage of maturation. Then, polyphenol concen-trations remain practically constant or slightly increase(Macheix, Fleuriet, & Billot, 1990; Blanco et al.,1992a). On the other hand, anthocyanin contents in-crease markedly during maturation. The variation ofthe procyanidin content during fruit ontogenesis is clo-sely related to the changes detected in the astringency.Main responsible factors for anoleptic parame-ter changes are ic, leading to notable dierences be-tween varieties, and physiological, which are directlyrelated to the state of maturation of the fruit. Indeed,astringency decreases or pletely duringmaturation. This decrease generally takes place alongwith a decrease in the fruit procyanidin content, or withphysicochemical changes of the procyanidin molecules,aecting their degree of polymerisation, which playsan important role in their ability to interact with pro-teins (Macheix et al., 1990).In the present work, polyphenolic content of Basquecider apple varieties, harvested at two dierent stages ofripening, were evaluated in order to dierentiate ripeand unripe fruits. The aim of this study was to achieveclassication rules that would allow prediction of fruitmaturity state (ripe or unripe) according to their poly-phenolic proles by pattern recognition analysis of thedata.2. Materials and methods2.1. Solvents and standard phenolicsMethanol (Romil Chemical Ltd, Heidelberg, Ger-many) was of HPLC grade. Water was puried on aMilli-Q system from Millipore (Bedford, MA, USA).Glacial acetic acid, formic acid, toluene-a-thiol andfuming hydrochloric acid (37%) provided by Merck(Darmstadt, Germany) and ascorbic acid, by Panreac(Barcelona, Spain), were of analytical quality. All sol-vents used were previously ltered through 0.45 lm ny-lon membranes (Lida, Kenosha, WI, USA).Polyphenol standards were supplied as follows: (+)-catechin, ()-epicatechin, rutin, phloridzin, 5-caeoyl-quinic acid and p-coumaric acid by Sigma–AldrichChemie (Steinheim, Germany); hyperoside, isoquercitrin,avicularin, quercitrin and ideain chloride by Extras-ynthe`se (Genay, France). ()-Epicatechin 4R-benzyl-thioether and 4-p-coumaroylquinic acid were kindlyprovided by Dr. Guyot (Laboratoire de RecherchesCidricoles, Biotransformation des Fruits et Legumes,INRA, Le Rheu, France), and phloretin-20-O-xylogluco-side and procyanidin B2 by Dr. F.A. Tomas-Barberan(Laboratorio de Fitoqumica, Departamento de Cienciay ologa de los Alimentos, CEBAS (CSIC), Murcia,Spain) and Dr. C. Santos-Buelga (Departamento deQumica Analtica, Nutricion y Bromatologia, Universi-dad de Salamanca, Spain), respectively. Stock standardsolutions of (+)-catechin, ()-epicatechin, ()-epicate-114 R.M. Alonso-Salces et al. / Food Chemistry 93 (–123chin 4R-benzylthioether, rutin, phloridzin, 5-caeoylqui-nic acid and p-coumaric acid, at a concentration of 1mg ml1and hyperoside, isoquercitrin, quercitrin andideain, at 0.6 mg ml1, were prepared in methanol andstored at 4 °C in darkness.2.2. SamplesPulp and peel from 14 apple cultivars used in theBasque Country for cidermaking were analysed. Thetechnological groups of the cultivars studied were asfollows: bittersweet: Geza Mina (GM), Mozoloa (MZ),Patzuloa (PT) and Ugarte (UG); semiacid: Bost Kantoi(BK), Manttoni 111 (MN111), ManttoniEM7 (MNEM7)and Urtebi Txiki (UT); acid: Errezila (ER), Goikoetxea(GK), Txalaka (TX), Udare Marroi (UM) and UrtebiHaundia (UH); and bitter-acid: Moko (MK).Apples were harvested in the Experimental Orchardof the Diputacion Foral de Gipuzkoa in Hondarribia(Guipuzcoa, Spain) during the
and 2001 sea-sons. The time lag between the harvest of unripe andripe fruits was about three weeks. Maturity state ofthe fruits was tested by the starch index (Planton,1995). Unripe fruits presented index values between 4and 8, and ripe fruits, between 9 and 10. For each vari-ety and season, two or three batches of 10 apple fruitswere mechanically peeled and cored, and sprayed withan aqueous solution of formic acid (3%, v/v) in orderto avoid polyphenol oxidation. Peels (average thickness:0.8 mm) and pulps were immediately frozen in liquidnitrogen and then freeze-dried. An aliquot of each vari-ety was used to determine the fresh/dry matter ratio. Thedried tissues were crushed in closed vials to avoid hydra-tion, using stainless-steel balls (;1 cm) and an overheadshaker (Reax 2, Heidolph, Schwabach, Germany). Ahomogeneous powder was obtained, which was storedat room temperature in a ator until analysis.2.3. Analytical procedures2.3.1. Thiolysis and direct solvent extractionFreeze-dried apple pulp and peel (0.5 g) were submit-ted to thiolysis as described by Guyot, , Sanoner,and Drilleau (2001), and to direct solvent extractionwith 30 ml of methanol–water–acetic acid (30:69:1,v/v/v) with ascorbic acid (2 g/l) in an ultrasonic bathduring 10 min (Alonso-Salces et al., 2004a). Afterwards,both the thiolysis reaction mixture and the crude solventextract were ltered through a 0.45lm PTFE lter(Waters, Milford, CA, USA) prior to injection into theHPLC system.2.3.2. Reversed-phase HPLC analysisChromatographic analysis was performed on a Hewl-ett–Packard Series 1100 system, equipped with a vac-uum degasser, a quaternary pump, a thermostattedautosampler, a thermostatted partmentand a DAD detector, connected to an HP ChemStationsoftware. A reversed phase Nova-Pak C18 (300
3.9mm i.d., 4 lm) column and a Nova-Pak C18 (10
3.9mm i.d., 4 lm) guard column (Waters, Barcelona,Spain) were used. Solvents that constituted the mobilephase were A (acetic acid–water, 10:90, v/v) and B(methanol). The elution conditions applied were: 0–10min, 0% B 10–40 min, linear gradient from0% to 15% B; 40–60 min, 15% B and nally,washing and reconditioning of the column. The owrate was 0.8 ml min1and the injection volume was 50ll of crude extracts or 10 ll of thiolysis media. The sys-tem operated at 25 °C. Flavan-3-ols and dihydrochal-cones were monitored and quantied at 280 nm,hydroxycinnamic acids at 320 nm, avonols at 370 nmand anthocyanins at 530 nm. Polyphenol identication,for which standards were available, was carried out parison of their retention times and their UV–visiblespectra with those of the standards. Some other chroma-tographic peaks were assigned to a particular polyphe-nol class according to their UV–visible spectra andbibliographic sources. In this sense, those unknownchromatographic peaks that exhibited avan-3-ol spec-tra were designated as CAT-n, and those with a spec-trum of 5-caeoylquinic acid as CAA-n, of p-coumaricas CMA-n, of dihydrochalcone as PLD-n, of avonolas QG-n and of anthocyanin as CG-n (where ‘‘n’’ is anumber). Quantication was performed by reportingthe measured integration areas in the calibration equa-tion of the corresponding standards. Thus, procyanidinB2 and the unknown avan-3-ols were quantied as (+)- phloretin-20-O-xyloglucoside and the unknowndihydrochalcones were qu avicula-rin and the unknown avonols weCAA-n species were quantied as 5-4-p-coumaroylquinic acid and CMA-n species werequantied as p-coumaric acid and the unknown antho-cyanins as ideain.2.4. Data analysis and chemometric proceduresA peel and pulp data set consisted of a 85
27 matrixand a 85
18 matrix, in which rows represented applesamples, and columns the concentrations of individualpolyphenols determined by HPLC-DAD, the total con-centration of procyanidins, and the average degree ofpolymerisation of procyanidins (DPn). Each samplewas represented in the multidimensional space by a datavector, which was an assembly of the 27 features in peeland 18 features in pulp. Data vectors belonging to thesame category (ripe or unripe) were analysed using che-mometric procedures that have been described in the lit-erature (Latorre, Pena, Garca, & Herrero, 2000;
al., 2001), such as cluster analysis (CA), linear discri-minant analysis (LDA), K-nearest neighbours (KNN),R.M. Alonso-Salces et al. / Food Chemistry 93 (–123 115soft independent modelling of class analogy (SIMCA),and multilayer feed-forward articial works(MLF-ANN). Statistic and chemometric data analysiswas performed by means of the statistical software pack-ages Statgraphics (), Parvus (Forina, Lanteri,& Armanino, 2000) SPSS () and WinNN32().Cluster analysis is a preliminary way to study datasets in the search for natural groupings among the sam-ples characterised by the values of a set of measured fea-tures. Owing to its unsupervised character, CA is apattern recognition technique that can be used to revealthe structure residing in a data set (Massart & Kaufman,1983). CA was performed on autoscaled data, samplesimilarities were calculated on the basis of the squaredEuclidean distance and the Ward hierarchical agglomer-ative method was used to establish clusters.The classication rules achieved by the supervisedchemometric techniques were validated by means of across-validation procedure, which was performed bydividing plete data set into a training set andan evaluation set. Samples were assigned randomly toa training set, consisting of 75% of them, and the testset, composed of the remaining 25% of the samples.Such a division allows for a sucient number of samplesin the training set and a representative number of mem-bers among the test set. The same process was repeatedfour times with dierent constitutions of both sets to en-sure that all the samples had the possibility of inclusionin the evaluation set at least once. The dierent patternrecognition techniques were applied to the four training-test sets obtained. The reliability of the classicationmodels achieved was studied in terms of recognitionability (percentage of the members of the training setcorrectly classied) and prediction ability (percentageof the members of the test set correctly classied byusing the rules developed in the training step).In KNN, the inverse square of the Euclidean distancewas used as the criterion for calculating the distance be-tween samples, and the number of neighbours (K) wasselected after studying the ess in classication withdierent K values, applying this technique to a trainingset with all the samples.The model achieved by SIMCA for each categorywas also evaluated in terms of sensitivity and specicity.The sensitivity of the model is known as the percentageof objects belonging to the category correctly identiedby the mathematical model, and its specicity, as thepercentage of objects foreign to the category classiedas foreign (Melendez, Ort z, Sanchez, Sarabia, &Iniguez, 1999).In MLF-ANN, the target output was assigned as 0 or(0,1) for unripe fruits and 1 or (1,0) for ripe ones, and asigmoidal function f(x) = 1/(1 + [exp(x)]) was used asthe transfer function. The work was trainedby means of an algorithm bined the use of anadaptative learning rate parameter (ALRP) (g) and amomentum (l) which have been described previously(Padn et al., 2001). The initial values of the weightsassociated with the connections between neurons wereselected randomly in the range 3 to 3. The maximumnumber of epochs was 2000 and the initial values of gand l were 0.2 and 0.5, the target errorwas 0.1.3. Results and discussion3.1. Analytical dataAnalytical data obtained by the chromatographicdetermination of apple polyphenols in pulp and peelfor each variety are summarised in Tables 1 and 2. Totalpolyphenols and procyanidin contents were smaller inunripe apple peels and pulps. Flavonols and anthocya-nins in peels, and hydroxycinnamic acids and avan-3-ols in pulps presented higher concentrations in ripefruits. The average degrees of polymerisation of procy-anidins (DPn) were considerably higher in unripe fruits.From the results obtained by the analysis of the poly-phenolic proles of the studied apple cultivars, it seemedthat unripe fruits were in an earlier stage of maturationbut later than the minimum reached in total polyphenolconcentration prior to the accumulation of sugars afore-mentioned, since the concentrations of certain classes ofpolyphenols were lower in unripe fruits than in ripeones. Furthermore, notable dierences were observedin the DPn, these being higher in unripe fruits, whichagreed with the observations reported by Macheixet al. (1990). In cidermaking, the use of fruits that arenot at the optimum maturity state can lead to an impor-tant increase of bitterness and astringency and the inci-dence of troubles related to procyanidins, such ascloudiness after bottling (Lea, 1990; Siebert, Carrasco,& Lynn, 1996).Bitterness of the unripe fruits was predicted by theclassication system achieved by Alonso-Salces, Herr-ero, Barranco, Berrueta, and Vicente (2004b), in orderto study the inuence of apple maturity state on the clas-sication of apple cultivars in technological groups. Inmost cases, bitterness predictions made for the unripefruits agreed with the results obtained for the ripe fruits(Alonso-Salces et al., 2004b). This allows us to concludethat dierences in position betweenbitter and non-bitter varieties are already evident beforefruits reach their optimum maturity state. However,regarding some unripe fruits, non-conclusive resultswere obtained for GM (pulp), and MK (pulp), UG(peel) and PT (peel) were misclassied as non-bitter, be-cause they had lower concentrations of polyphenolsthan ripe fruits. This can be explained by the fact thatthe polyphenol content reaches a minimum just before116 R.M. Alonso-Salces et al. / Food Chemistry 93 (–123the nal accumulation of sugars and the starch degrada-tion, which takes place at the nal stage of apple matu-ration (Blanco et al., 1992a).3.2. Univariate data analysisThe analysis of variance (ANOVA) performed on ap-ple peel and pulp data sets, constituted of the individualpolyphenol concentrations, total procyanidin contentand DPn of the fruits disclosed signicant dierencesfor all variables between ripe and unripe fruits. A leastsignicant dierence (LSD) test (p & 0.05) was also car-ried out on the data matrices of both apple tissues, in or-der to check that there were no signicant dierencesbetween seasons. Fishers test allowed us to detect themost discriminant variables between ripe and unripefruits (Sharaf, Illman, & Kowalski, 1986). In both tis-sues, peel and pulp, DPn was the feature that presenteda higher Fisher weight (p & 0.001). In pulp, the followingvariables with the highest Fisher values (p & 0.050) werethe total content of procyanidins (PC) and (+)-catechin(CAT), and in peel (p & 0.001), PLD-2 and PLD-1,which are hydroxyphloretin glycosides (Alonso-Salceset al., 2004c) and phloretin-20-O-xyloglucoside (PLXG).Hence, it seems that avan-3-ols in pulp and dihydroch-alcones in peel are the classes of polyphenols that under-go greater changes in the last stage of maturation,whereas the remaining polyphenol concentrations re-main practically constant, as noted before in the litera-ture (Burda et al., 1990; Mayr et al., 1995). Despitethe dierences in these features between ripe and unripeapple fruits, their box and whisker plots showed anoverlap between the two classes, indicating insucientdiscriminatory ability. Thus, none of the variables meas-ured was able to discriminate between ripe and unripecategories by itself. Therefore, it was necessary to moveon to a multivariate data analysis.3.3. Multivariate data analysis3.3.1. Cluster analysisWhen CA was applied to plete set of varia-bles, no clear groupings of the samples according totheir maturity (ripe or unripe) could be observed, eitherin peel or in pulp. However, considering only the mostdiscriminant variables in each apple tissue, enoughinformation was provided by these features to achievea classication of fruits in the established categories.Thus, the variables regarded in peel were DPn, PLD-2,PLD-1 and PLXG, and in pulp, DPn, PC and CAT. Re-sults attained for each apple tissue are presented as adendrogram (Fig. 1). In pulp, at a similarity level of0.30, three clusters were identied as follows: A, madeup of unripe fruits, B, containing all ripe fruits and theunripe fruits of cultivars GK and MNEM7, and C, con-sisting of samples of UG variety. In peel, at a similarityTable 1Concentrations (mg kg1of apple) of avan-3-ols, hydroxycinnamic acids, dihydrochalcones and avonols and the average degree of polymerisationof procyanidins (DPn) in apple pulpsPolyphenol Fruit maturity stateUnripe RipeMean S.D. Min Max Mean S.D. Min MaxFlavan-3-olesCAT 14 23 nd 91 23 29 0.7 113EC 86 57 26 209 115 63 48 227PB2 87 61 32 231 117 74 46 324CAT-2 9 5 3 22 11 6 5 27PC 990 545 500 3 672 3041DPn 7 1 5 9 4.7 0.5 4.0 5.6Hydroxycinnamic acidsCQA 240 177 40 630 311 200 61 724PCQ 19 17 1 59 29 32 1 120CAA-1 23 19 4 58 24 15 8 58CMA-2 0.5 1.0 nd 3.4 0.2 0.5 nd 1.5DihydrochalconesPLXG 19 13 3 45 29 19 5 62PLG 19 25 5 100 17 12 7 56PLD-1 6 6 0.4 20 3 2 0.7 7PLD-2 5 9 1 34 3 2 1 11FlavonolsHYP 0.2 0.6 nd 2.2 t nd tIQC 0.4 0.3 nd 1.0 0.4 0.3 nd 0.8QCI 1 1 nd 5 2 1 0.5 5QG-1 0.6 0.4 nd 1.6 0.6 0.5 t 1.6R.M. Alonso-Salces et al. / Food Chemistry 93 (–123 117level of 0.75, six clusters were found: clusters A, D, Eand F contained unripe fruits, whereas clusters B andC were constituted of ripe samples. Two apple varieties(MNEM7 and PT) presented their unripe samples in acluster of ripe fruits, and ripe samples of ER varietywere inside cluster E of unripe fruits. These samples,that are in clusters not corresponding to their category,were situated in the overlapped region of both classes ina multidimensional plot of the samples in the space de-ned by the most discriminant variables (Fig. 2). Inthese plots, a natural separation between unripe and ripeapple fruits could be observed, these results being ordance with those obtained by CA.3.3.2. Supervised pattern recognition methods3.3.2.1. General. LDA, KNN, SIMCA and a MLF neu-work were applied to the autoscaled data matrixof each apple tissue, formed of unripe and ripe samples(85 pulp samples and 85 peel samples) and the most dis-criminant features (3 variables for pulp and 4 variablesfor peel), in order to achieve a prediction rule for classi-fying apple fruits according to their maturity state: ripeor unripe.In KNN, the optimum number of neighbours (K) wasstudied. For both apple tissues, pulp and peel, the sameresults were attained for the K-values assayed (3, 5, 7and 9); none of the samples was misclassied, thereforeK = 5 was selected.When using MLF-ANN with the purpose of makingpredictions, some empirical preliminary trials have to beperformed in order to determine an adequate MLF-ANN structure. The MLF-ANN was applied to an in-put pattern consisting of the autoscaled data matrix.The neural architecture which gave better results thenothers was a MLF-ANN with three layers: for pulp,an input layer and one hidden layer with 3 neurons each,and an output layer consisting of a neuron with a binaryoutput, and, for peel, an input layer with 4 neurons, oneTable 2Concentrations (mg kg1of apple) of avan-3-ols, hydroxycinnamic acids, dihydrochalcones, avonols and anthocyanins and the average degree ofpolymerisation of procyanidins (DPn) in apple peelsPolyphenol Fruit maturity stateUnripe RipeMean S.D. Min Max Mean S.D. Min MaxFlavan-3-olesCAT 3 3 0.2 11 4 4 0.3 17EC 38 28 12 94 37 25 11 81PB2 33 19 12 72 32 17 8 56CAT-2 4 2 2 7 4 2 2 6PC 574 171 355 882 629 181 360 929DPn 9 2 7 16 5.8 0.7 4.9 7.1Hydroxycinnamic acidsCQA 20 24 2 81 28 21 5 67PCQ 2 2 nd 5 3 3 0 12CAA-1 4 3 1 12 5 3 1 11CAA-2 1.1 0.7 0.3 2.4 0.4 0.6 nd 1.8CMA-2 0.5 0.5 nd 2.0 0.5 0.7 nd 2.7DihydrochalconesPLXG 7 4 1 13 15 9 2 28PLG 30 23 6 82 44 34 7 123PLD-1 9 6 2 23 4 3 1 11PLD-2 27 20 4 66 9 6 1 21FlavonolsHYP 23 17 1 53 34 20 4 68IQC 7 5 1 18 9 6 2 24AVI 18 13 5 49 21 14 9 52QCI 9 6 2 22 11 8 3 29QG-1 11 7 3 23 14 7 5 28QG-2 1 1 0 6 2 1 0.4 6QG-3 0.4 1 nd 4 0.5 1 nd 4AnthocyaninsIDE 1.1 2.2 nd 7.5 1.6 3.1 nd 9.7CG-1 0.05 0.1 nd 0.4 0.1 0.1 nd 0.5CG-2 nd t nd tCG-3 0.02 0.07 nd 0.27 0.03 0.06 nd 0.22CG-4 0.02 0.07 nd 0.27 0.02 0.06 nd 0.21 118 R.M. Alonso-Salces et al. / Food Chemistry 93 (–123hidden layer with 9 neurons, and a binary output neuron(Table 3).Table 4 shows the recognition and prediction abilitiesaorded by each multivariate technique applied on thepulp and peel data, respectively.3.3.2.2. Apple pulp. LDA and KNN achieved highly sat-isfactory classications of ripe apples, with recognitionand prediction abilities of 100%. For unripe apples,these abilities were slightly worse. Thus, the models pro-posed by these techniques were selecthat is, the probability that ripe apples were classiedas unripe was hardly nought. However, there existed acertain likelihood that unripe fruits were misclassied(8% in LDA and 14% in KNN). These results agreedwith those obtained by CA, where some unripe sampleswere found inside clusters of ripe fruits. The classica-tion plished by SIMCA achieved better resultsfor the unripe pulps than for the ripe ones, the recogni-tion and prediction abilities being 98.8% and 96.4%, andof 95.0% and 94.4%, respectively. The unripe model pre-sented a sensitivity of 91% and a specicity of 92%,which meant that it accepted 91% of the unripe samplesand 8% of the ripe fruits. The ripe model recognised 56%Fig. 1. Dendrograms of cluster analysis for apple data. Sample codes:1, 2, ripe fruits.DPnCAT (mg/kg apple)PC(mg/kgapple)2 4 6 8
4000(a) pulpDPn PLD-2 (mg/kg apple)PLD-1(mg/kgapple)0 612 180 (b) peelFig. 2. Projection of apple samples on the multidimensional spacedened by the most discriminant features between the two categories.Sample codes: ¤, , ripe fruits.Table 3MLF-ANN architectures assayed and their prediction abilities for ripeand unripe apple fruitsApplematerialMLF-ANNarchitecturePredictionability (%)RMSEPulp 3, 2, 1 92.7 0.04 3, 3, 1 97.8 0.02 3, 5, 1 97.1 0.02 3, 3, 2 96.3 0.04Peel 4, 3, 1 88.4 0.09 4, 5, 1 88.4 0.08 4, 7, 1 88.4 0.05 4, 9, 1 92.0 0.05 4, 11, 1 90.2 0.09 4, 7, 2 91.1 0.09 4, 9, 2 90.2 0.05R.M. Alonso-Salces et al. / Food Chemistry 93 (–123 119of the ripe pulps and rejected 84% of unripe ones, thesepercentages being its sensitivity and specicity, respec-tively. Hence, there existed a probability (of 8% for ripefruits and of 16% for unripe fruits) that they werewrongly classied by SIMCA. Fig. 3 represents SIMCAresults as a Coomans plot for the squared SIMCA dis-tances obtained from the data set. Classication resultsaorded by the work were excellent for bothcategories, the prediction abilities being above 96%.3.3.2.3. Apple peel. LDA results obtained were similarfor both categories, showing recognition and predictionpercentages above 90%. Most samples misclassied werelocalised in the overlapped region of the two classes(Fig. 2). KNN achieved better classications thanLDA, presenting for ripe fruits, recognition and predic-tion abilities close to 100%. With both techniques, LDAand KNN, results were slightly better for the ripe cate-gory. Thus, the likelihood that ripe apples were classiedas unripe was smaller than in the opposite case. Theseresults tally with those of CA, where unripe fruits oftwo apple cultivars were present in a group of ripe fruits,and ripe fruits of one other cultivar was included in anunripe cluster. On the other hand, SIMCA aorded con-siderably more satisfactory results for unripe apples,with esses in recognition and prediction of 94.1%and 92.3%, than for ripe fruits, with 88.8% and 79.2%,respectively. In terms of sensitivity and specicity, SIM-CA models for both categories presented a sensitivity of93%, but the unripe model had a specicity of 63%,whereas the model for the ripe category was of 86%.Hence, the probabilities that SIMCA misclassied sam-ples was considerably high for both categories (37% ofripe peels and 14% of unripe peels). Fig. 3 representsSIMCA results as a Coomans plot for the squared SIM-CA distances. MLF-ANN achieved notably better pre-dictions for the unripe fruits (96.0%) than for the ripeones (88.7%).Fig. 3. Coomans plot for the squared SIMCA distances for apple data. Codes: Training set: 1, 2, ripe class. Test set: ¤, ,ripe fruits.Table 4Classication results for the supervised pattern recognition techniques applied to apple dataTechnique Class Pulp PeelRecognitionability (%)Predictionability (%)Recognitionability (%)Predictionability (%)LDA Unripe 93.1 92.0 93.8 90.3Ripe 100.0 100.0 95.1 91.3KNN (K = 5); inverse squared Euclidean distance Unripe 94.1 85.7 94.1 92.9Ripe 100.0 100.0 98.8 100.0SIMCA; a = 0.05 Unripe 98.8 96.4 94.1 92.3Ripe 95.0 94.4 88.8 79.2MLF-ANN; g = 0.2; l = 0.5; sigmoidal transfer functionaUnripe 100.0 96.6 100.0 96.0Ripe 100.0 98.7 100.0 88.7aMLF-ANN architecture: (3
1) in peel.120 R.M. Alonso-Salces et al. / Food Chemistry 93 (–123When apples are processed to obtain a nal productwith adequate sensory and nutritional qualities, it is ofgreat importance to ascertain that raw material is atthe optimum condition of maturation, so that unripe ap-ples are not used. To plish this, suitable tools thatcan classify unripe apples inside their category with highpercentages of hits are required, thus minimising the riskof considering unripe fruits as ripe. In this particularcase, the fact that ripe apples could be misclassied isnot so relevant. In this sense, the chemometric techniquethat aorded the better results was the work inboth apple tissues, presenting prediction abilities for un-ripe apples above 96% and implying little probability(4%) that unripe apple were mispredicted. These classi-cation systems presented better prediction abilities thanother decision rules previously reported (Mangas et al.,1998), which used sugars, organic acids, amino acids, to-tal polyphenols, and pectins as chemical variables, andchemometrics in order to classify cider apples accordingtheir degree of ripening.4. ConclusionsThe results attained in this study allow us to concludethat certain polyphenols (procyanidins and (+)-catechinin pulp and dihydrochalcones, such as the xyloglucosideof phloretin and hydroxyphloretin glycosides in peel)and the average degree of polymerisation of procyani-dins, jointly with several chemometric techniques, areappropriate for dierentiating ripe and unripe applefruits and for making predictions of the maturity stateof apples.The procedure proposed for establishing the maturitystate (ripe/unripe) of the apple cultivars studied, involvesthe determination of the (+)-catechin concentration, thetotal procyanidin content and the average degree ofpolymerisation of procyanidins in apple pulp, and theuse of the classication rule developed by the -work for performing the prediction. This nal proposal isdue to the fact that apple position depends, to agreat extent, on climatology (Lea, 1990) and sun exposiveof the fruit, causing dierences according to the positionof the fruit on the tree and even, in the same fruit, be-tween sun-exposed parts and shaded parts (Awad, De Ja-ger, & Van Westing, 2000). Hence, since pulp predictionswere more homogeneous and did not depend so much onexternal factors, they were considered to be more reliablethan those of peel.AcknowledgementsThis research was supported by Gobierno Vasco/Eu-sko Jaurlaritza (project number PI-1997-19 and PI-) and Universidad del Pas Vasco/Euskal Her-riko Unibertsitatea (Project Number 171.310-EB013/98). The authors express their gratitude to the Dip-utacion Foral de Gipuzkoa for providing apple samples.Rosa MaAlonso-Salces wishes to thank Gobierno Vas-co/Eusko Jaurlaritza for a PhD Grant.Appendix AAbbreviationsAVI avicularinCQA 5-caeoylquinic acidCAA-1-2 unknown hydroxycinnamic acids withcaeic acid UV spectraCAT (+)-catechinCAT-2 unknown avan-3-olCG-1,-2,-3, -4 unknown anthocyaninsCMA-2 unknown hydroxycinnamic acid withp-coumaric acid UV spectraDPn average degree of polymerization ofprocyanidinsEC ()-epicatechinHYP hyperosideIDE ideainIQC isoquercitrinPB2 procyanidin B2PC total procyanidinsPCQ 4-p-coumaroylquinic acidPLD-1 hydroxyphloretin diglycosidePLD-2 hydroxyphloretin monoglycosidePLG phloridzinPLXG phloretin-20-O-xyloglucosideQCI quercitrinQG-1-2-3 unknown avonolsCA Cluster analysisKNN K-nearest neighboursLDA linear discriminant analysisMLF-ANN multilayer feed-forward-workRMSE root medium square errorSIMCA soft independent modelling of classanalogyDAD Diode array detectorHPLC High Performance Liquid Chromatog-raphynd not detectedt tracesSD standard deviationBK Bost KantoiER ErrezilaGM Geza MinaGK GoikoetxeaMN111 Manttoni 111MNEM7 ManttoniEM7MK MokoR.M. Alonso-Salces et al. / Food Chemistry 93 (–123 121MZ MozoloaPT PatzuloaTX TxalakaUM Udare MarroiUG UgarteUH Urtebi HaundiaUT Urtebi TxikiReferencesAckermann, J., Fisher, M., & Amado`, R. (1992). Changes in sugars,acids, and amino acids during ripening and storage of apples (Cv.Glockenapfel). Journal of Agricultural Food Chemistry, 40,.Alonso-Salces, R. M., Barranco, A., Corta, E., Berrueta, L. A., Gallo,B. & Vicente, F. (2004a). A validated solid–liquid extractionmethod for the HPLC determination of polyphenols in appletissues. Comparison with pressurised liquid extraction. Talanta, inpress.Alonso-Salces, R. M., Herrero, C., Barranco, A., Berrueta, L. A.,Gallo B. & Vicente, F. (2004b). Technological classication ofBasque cider apple cultivars according to their polyphenolicproles by pattern recognition analysis. Journal of AgriculturalFood Chemistry, in press.Alonso-Salces, R. M., Ndjoko, K., Queiroz, E. F., Ioset, J. R.,Hostettmann, K., Berrueta, L. A., et al. (2004c). On-line charac-terisation of apple polyphenols by high performance liquidchromatography coupled with mass spectrometry and ultravioletabsorbance detection. Journal of Chromatography A, 0.Amiot, M. J., hini, M., Aubert, S., & Nicolas, J. (1992).position and browning susceptibility of variousapple cultivars at maturity. Journal of Food Science, 57,958–962.Awad, M. A., De Jager, A., & Van Westing, L. M. (2000). Flavonoidand chlorogenic acid levels in apple fruit: characterisation ofvariation. Science Horticulturae, 83, 249–263.Blanco, D., Moran, M. J., Gutierrez, M. D., Moreno, J., Dapena, E.,& Mangas, J. J. (1992a). Biochemical study of the ripening of ciderapple varieties. Zeitschrift fu¨r Lebensmitteluntersuchung und -Forschung, 194, 33–37.Blanco, D., Picinelli, L., Gutierrez, M. D., & Mangas, J. J. (1992b).Determination of amino acids in ripening apples by high perfor-mance liquid chromatography. Zeitschrift fu¨r Lebensmitteluntersu-chung und -Forschung, 194, 134–138.Board, P. W., & Woods, H. J. (1983). Compositional variations andsensory acceptability of apple juice drink. Journal of FoodTechnology, 18, 763–769.Burda, S., Oleszek, W., & Lee, C. Y. (1990). pounds andtheir changes in apples during maturation and cold storage. Journalof Agricultural Food Chemistry, 38, 945–948.Cowan, M. M. (1999). Plant products as anti-microbial agents. ClinicalMicrobiology Reviews, 12, 564–582.Dilley, D. R. (1981). Assessing fruit maturity and ripening andtechniques to delay ripening in storage. Annual Report of MichiganState Horticultural Society, 110, 132–146.Forina, M., Lanteri, S. & Armanino, C. (2000). Q-Parvus 3.0. Anextendable package of programs for data explorative analysis,classication and regression analysis. Dipartimento Chimica ologie Farmaceutiche ed Alimentari, University of Genova,Genova, Italy.Guyot, S., , N., Sanoner, P., & Drilleau, J. F. (2001). Directthiolysis on crude apple materials for the HPLC characterizationand quantication of polyphenols in cider apple tissues and juices.Methods in Enzymology, 335, 57–70.Lancaster, J. E. (1992). Regulation of skin color in apples. CriticalReviews in Plant Science, 10, 487–502.Latorre, M. J., Pena, R., Garca, S., & Herrero, C. (2000).Authentication of Galician (N.W. Spain) honeys by multivar-iate techniques based on metal content data. The Analyst, 125,307–312.Le Lezec, M. & Babin, J. (1988). Test de regresion de lalmidon despommes. In: LArboriculture du Val de Loire, November 88 (pp. 18–19).Lea, A. G. H. (1990). Bitterness and astringency: The procyanidins offermented apple ciders. In R. L. Rousself (Ed.), Bitterness in foodand beverages (pp. 123–143). Oxford: Elsevier.Lea, A. G. H. (1995). Cidermaking. In A. G. H Lea & J. R. Piggott(Eds.), Fermented beverage production (pp. 66–96). London: BlackieAcademic & Professional.Macheix, J. J., Fleuriet, A., & Billot, J. (1990). Changes andmetabolism of pounds in fruits. In Fruit phenolics(pp. 149–237). Boca Raton: CRC Press.Mangas, J. J., Dapena, E., Rodrguez, M. S., Moreno, J., Gutierrez,M. D., & Blanco, D. (1992). Changes in pectic fractions duringripening of cider apples. Hortscience, 27, 328–330.Mangas, J. J., Moreno, J., Picinelli, A., & Blanco, D. (1998).Characterization of cider apple fruits according to their degree ofripening. Journal of Agricultural Food Chemistry, 46, .Massart, L., & Kaufman, L. (1983). The interpretation of analyticalchemical data by the use of clusters analysis. New York: WileyInterscience.Mayr, U., Treutter, D., Santos-Buelga, C., Bauer, H., & Feucht, W.(1995). Developmental changes in the phenol concentrations of‘‘Golden Delicious’’ apple fruits and leaves. Phytochemistry, 38,.Mc Manus, J. P., Davis, K. G., Beart, J. E., Ganey, S. H.,Lilley, T. H., & Haslam, E. (1985). Polyphenol interactions.Part I. Introduction: some observations on the plexation of polyphenols with proteins and harides.Journal of the Chemical Society, Perkin Transactions II,.Melendez, M., Ortz, M. C., Sanchez, M., Sarabia, L., & Iniguez, M.(1999). Chemometric characterization of the clarets and rose winesof the certied denomination of origin Rioja using Cielab param-eters. Qumica Analtica, 18, 119–126.Padn, P. M., Pena, R. M., Garca, S., Iglesias, R., Barro,S., & Herrero, C. (2001). Characterization of Galician(N.W. Spain) quality brand potatoes: parison studyof several pattern recognition techniques. The Analyst, 126,97–103.Planton, G. (1995). Le point sur le test almidon des pommes pourlaide a` la decision de recolte. Aout 1995, No. 6. Paris: Infos-CTIFL.Renard, C. M. G. C., Baron, A., Guyot, S., & Drilleau, J. F. (2001).Interactions between apple cell walls and native apple polyphenols:quantication and some consequences. International Journal ofBiological Macromolecules, 29, 115–125.Ridgway, T., Oreilly, J., West, G., Tucker, G., & Wiseman, H. (1996).Potent antioxidant properties of novel apple-derived mercial potential as food additives. Biochemical SocietyTransactions, 24, S391–S391.Sharaf, M., Illman, D., & Kowalski, B. R. (1986). Chemometrics. NewYork: John Wiley & Sons.Siebert, K. J., Carrasco, A., & Lynn, P. Y. (1996). Formation ofprotein-polyphenol haze in beverages. Journal of Agricultural FoodChemistry, 44, .Sieguist, R. (1987). Dates de recolte des pommes de garde: Jonalgold,Gloster, Jonnee, et Idared. Revue Suisse de Viticulture, Arbor-iculture, Horticulture 19, 295–298.122 R.M. Alonso-Salces et al. / Food Chemistry 93 (–123播放器加载中，请稍候...
该用户其他文档
下载所得到的文件列表Classification of apple fruits according to their maturity state by the pattern recognition analysis of their positions：苹果果实按成熟度状态的多酚类成分的模式识别分类分析.pdf
文档介绍：
Classication of apple fruits according to their maturity state bythe pattern recognition analysis of their positionsRosa M. Alonso-Salces a, Carlos Herrero b, Alejandro Barranco a,Luis A. Berrueta a,*, Blanca Gallo a, Francisca Vicente aaDepartamento de Qumica Analtica, Facultad de Ciencias, Univers...
内容来自淘豆网转载请标明出处.}