Bioelectronic tongue dedicated to the analysis of milk using enzymes linked to carboxylated-PVC membranes modified with gold nanoparticles

Journal Pre-proof Bioelectronic tongue dedicated to the analysis of milk using enzymes linked to carboxylated-PVC membranes modified with gold nanoparticles Clara Perez-Gonzalez, Coral Salvo-Comino, Fernando Martin-Pedrosa, Cristina García-Cabezón, María Luz Rodríguez-Méndez PII: S0956-7135(22)00618-1 DOI: https://doi.org/10.1016/j.foodcont.2022.109425 Reference: JFCO 109425 To appear in: Food Control Received Date: 27 April 2022 Revised Date: 28 July 2022 Accepted Date: 2 October 2022 Please cite this article as: Perez-Gonzalez C., Salvo-Comino C., Martin-Pedrosa F., García-Cabezón C. & Rodríguez-Méndez Marí.Luz., Bioelectronic tongue dedicated to the analysis of milk using enzymes linked to carboxylated-PVC membranes modified with gold nanoparticles, Food Control (2022), doi: https://doi.org/10.1016/j.foodcont.2022.109425. This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2022 Published by Elsevier Ltd. Authors contribution MR-M, CG-C, and FM-P conceptualized the idea and supervised the work. CP-G and CS-C performed the experiment, curated the data, and wrote the original draft. FM-P involved in software design and development. CP-G and CS-C involved in formal analysis. CG-C and MR-M acquired the funding. CP-G, CS-C, FM-P, MR-M and CG-C reviewed and edited the paper. All authors provided feedback. Biographies Clara Perez-Gonzalez obtained the Ms in Nanoscience in 2019 (U. Valladolid. Spain). She is currently working on her PhD Thesis which is dedicated to the development of electrochemical sensors for the analysis of foods. She is author of 5 scientific papers. Coral Salvo-Comino obtained the Ms in Analytical Chemistry in 2015 (U. Complutense. Madrid. Spain). She is currently working on her PhD Thesis which is dedicated to the development of electrochemical sensors for the analysis of foods. She is author of 13 scientific papers. Fernando Martin-Pedrosa is full professor at the University of Valladolid and Head of the Department of Materials Science. His research is dedicated to electrochemistry studies of different solid materials. He is author of more than 80 papers. Cristina Garcia Cabezón, is assistant professor at the Engineers school of the University of Valladolid. She is an expert in electrochemistry and impedance spectroscopy. She is author or coauthor of more than 50 papers in the field. Maria Luz Rodriguez-Mendez is Full professor of Inorganic Chemistry at the Engineers School of the University of Valladolid and Head of the group of sensors UVASens. She is leading several funded Projects devoted to the development of arrays of voltammetric nanostructured sensors and biosensors for the characterization of foods. She is author or co-author of over 165 publications (H index 44), seven books and three patents in the field. Journal Pre-proof 1 Bioelectronic tongue dedicated to the analysis of milk using enzymes 1 linked to carboxylated-PVC membranes modified with gold 2 nanoparticles 3 4 Clara Perez-Gonzalez1,2,3, Coral Salvo-Comino1,2, Fernando Martin-Pedrosa2,3, Cristina García5 Cabezón2,3*, María Luz Rodríguez-Méndez1,2* 6 7 1 Group UVASENS, Escuela de Ingenierías Industriales, Universidad de Valladolid, Paseo del Cauce, 59, 8 Valladolid 47011, Spain. 9 2 BioecoUVA Research Institute, Universidad de Valladolid, 47011 Valladolid, Spain 10 3 Department of Materials Science, Universidad de Valladolid, Paseo del Cauce, 59, 47011 Valladolid, Spain 11 * Correspondence: [email protected], [email protected] 12 Highlights 13 • A potentiometric bioET specifically dedicated to milk analysis was developed. 14 • Enzymes were covalently linked to membranes combining Carboxilated-PVC and 15 AuNPs. 16 • The effective enzymatic immobilization helped to retain the enzymatic activity. 17 • Using SVM and ensemble methods, nine physicochemical parameters can be 18 determined simultaneously. 19 20 Abstract 21 Bioelectronic tongues (bioET) made of sensors combining enzymes and nanomaterials 22 have been shown to be advantageous due to the specificity offered by the biosensors and 23 the enhanced sensitivity provided by the nanomaterials. In this work, an innovative bioET 24 for milk analysis is developed using potentiometric biosensors based on lactic 25 dehydrogenase, galactose oxidase and urease specific for the detection of compounds of 26 interest in milk (lactic acid, galactose and urea). The performance of the biosensors has 27 been fostered by covalently immobilizing the enzymes on membranes of carboxylated 28 polyvinyl chloride combined with gold nanoparticles. The design and composition of the 29 biosensors contributes to preserving the enzymatic activity, allowing limits of detection 30 in the range of 10-5 – 10-6 M with excellent sensitivity and reproducibility (variation 31 coefficients ranged from 1 to 5.1 %). 32 The three biosensors, combined in a single device and coupled to a pattern recognition 33 software, can discriminate efficiently twelve classes of milk with different fat content 34 (skimmed, semi-skimmed and whole milk) and nutritional characteristics (calcium 35 enriched, lactose free and folic acid-enriched). The bioET shows an excellent 36 classification capability with an accuracy of up to 99.7%. By applying Support Vector 37 Machine (SVM) analysis, the BioET can perform the simultaneous assessment of eight 38 physicochemical parameters (acidity, fat, proteins, lactose, density, urea, dry matter and 39 nonfat dry matter) with satisfactory correlation coefficients and low residual errors. The 40 results are further improved by implementing ensemble methodologies. The proposed 41 strategy has been demonstrated to be useful for improving the performance of bioETs in 42 the dairy industry. 43 44 Keywords 45 Bioelectronic tongue, milk, biosensor, gold nanoparticles 46 Journal Pre-proof 2 1. Introduction 47 48 In recent years, the field of electronic tongues (ETs) has driven important basic 49 developments (Juzhong, & Jie, 2020; Aouadi et al., 2020; Rodriguez-Mendez, De Saja & 50 González-Antón, 2016; Ha et al. 2015). Much of this progress is related to the design of 51 new sensors which incorporate nanomaterials that improve the sensing characteristics, 52 thanks to their high surface to volume ratio and excellent electrocatalytic properties (Li, 53 Li, Liu, & Chen, 2019; Wang & del Valle, 2021; Sobrino-Gregorio, Bataller, Soto, & 54 Escriche, 2018; Teodoro, Shimizu, Scagion, & Correa, 2019; Americo da Silva et al., 55 2019). Other advances are related to new approaches to data management, including more 56 efficient data reduction methods and improved pattern recognition algorithms and 57 classification techniques (Tian, Chen, Pan, & Deng, 2013; Prieto et al., 2013). 58 The emergence of bioelectronic tongues (bioETs) combining classical unspecific sensors 59 with biosensors has been a breakthrough in the field, because these systems 60 simultaneously provide global information about the sample (as in classical ETs) plus 61 information about specific compounds obtained from the biosensors (Wasilewski, 62 Kamysz, & Gebicki, 2020; Skladal, 2020; Ghasemi-Vamankhasti et al., 2012; Yhan et 63 al., 2021; Ha et al., 2017). The performance of electrochemical biosensors can be further 64 improved by combining enzymes or other biological bioreceptors with nanomaterials. 65 Nanomaterials provide an effective platform for the immobilization of biomolecules, 66 inducing unique performance characteristics in terms of sensitivity and specificity. Some 67 examples of voltammetric bioETs based on combinations of enzymes and nanomaterials 68 have recently been reported. For instance, an array formed by phenol oxidases and 69 glucose oxidase combined with nanoparticles has been successfully used to analyze 70 grapes and musts (Garcia-Cabezón et al., 2020; Garcia-Hernandez et al., 2019). Human 71 taste receptors combined with carbon nanotubes (CNTs) or polypyrrole nanotubes have 72 been used to form a field effect transistor with human-tongue-like selectivity (Kim et al., 73 2011; Song et al., 2012). 74 Milk is a complex mixture that contains many different compounds, including 75 carbohydrates (mainly lactose), fats, proteins (casein or whey), minerals (such as calcium) 76 and many other miscellaneous constituents. E-tongues have been developed and applied 77 to the dairy industry in quality control, evaluation of taste or freshness, detection of 78 adulterations, origin recognition, etc. (Ciosek, 2016). These previous works have used 79 different types of electrodes and materials (Winquist et al. 1998; Wei, Wang, & Jin, 2013; 80 Pascual et al., 2018; Yu et al., 2015; Li et al., 2015; Ciosek, & Wroblewski, 2015; Tazi 81 et al., 2018; Dias et al., 2009; Pérez-González et al., 2021; Yang et al., 2021; Hruškar et 82 al., 2010; Collier et al. 2003; Valente et al., 2018; Scagiona et al., 2016). Only a few 83 attempts have been made to introduce nanomaterials in ETs applied to the dairy industry. 84 They include an array of voltammetric electrodes modified with nanostructured Layer85 by-Layer films (Salvo-Comino et al. 2018), a potentiometric ET using sensors modified 86 with nanoparticles (Mercante et al., 2015) and an impedimetric ET using electrospun 87 nanofibers (Ohlson et al., 2017). However, due to the complexity of milk, the analysis 88 using ETs is not a completely solved problem and new developments in the field are 89 required. 90 The proposal here is to take a step forward in the field of bioETs by developing novel 91 sensors combining enzymes specific to compounds present in milk (galactose, urea and 92 lactic acid) with nanomaterials. Galactose and its content is an important indicator of milk 93 quality and its content can be measured with individual galactose oxidase (GaOx) 94 biosensors (Ohlson et al., 2017; Kanyong, Krampa, Aniweh, & Awandare, 2019; Mangan 95 et al., 2018; Nguyen et al. 2016). Few examples can be found in the literature, where 96 Journal Pre-proof 3 GaOx has been combined with nanomaterials such as graphene (Çakıroğlu et al. 2019) or 97 nanoparticles (Migliorini et al., 2018). The detection of urea is also of prime importance 98 to assess the nutritional program of cows and can indicate underlying pathological 99 problems. Few examples of individual nanobiosensors for the detection of urea have been 100 reported. They are based on the combination of urease with nanoparticles (Jakhar & 101 Pundir, 2018) or nanofibers (Jia et al. 2011). Finally, the control of lactic acid is essential 102 to evaluate the fermentation of lactose due to lactic bacteria. Over the last few years, some 103 examples of biosensors based on lactate dehydrogenase (LDH) combined with 104 nanomaterials have been reported (Rahman et al. 2009). 105 In enzyme-based biosensors, the use of an adequate method to immobilize the enzymes 106 is crucial to preserve the enzymatic activity and avoid leakages (Nguyen, & Kim, 2017). 107 Covalent immobilization has the advantage of high surface loading and low protein loss 108 (Zucca, & Sanjust, 2014; Lee et al., 2017). Our proposal here is to develop an 109 immobilization membrane using carboxylated PVC (C-PVC) -instead of the classical 110 PVCwhere enzymes can be covalently linked using a covalent reaction between 111 carboxyl groups of the C-PVC and the amines on the protein. 112 In the ETs, it is also important to select the best chemometric methods to process the data. 113 Unsupervised and supervised analysis methods, such as principal component analysis 114 (PCA), linear discrimination analysis (LDA), support vector machines (SVM) or weighed 115 k-nearest neighbor analysis (KKNN), have been extensively applied (Skladal, 2020). One 116 of the emerging trends in data analysis is the combined use of statistical algorithms 117 through ensemble methodologies, where the outputs of the different algorithms are 118 combined in a decision fusion strategy to create a single response for a given problem 119 (Zhou, 2012). However, this strategy has barely been applied in the field of ETs, where 120 they could represent a great advance in complex media analysis such as milk. 121 In summary, the aim of this work was to develop a potentiometric bioET based on 122 membranes made of carboxylate PVC modified with nanoparticles. The carboxylate PVC 123 is used to covalently link the enzymes able to detect compounds in milks: GaOx, LDH 124 and Ure, which have been selected for their ability to detect important components in 125 milk. Once prepared and characterized, the sensing units are combined in a single device 126 to obtain a bioET that is used to analyze and classify 12 classes of milk with different 127 nutritional characteristics and to predict the eight physicochemical parameters most 128 commonly used in the dairy industry for quality control. In this work, a first approach to 129 an ensemble methodology routine is also proposed for the correlation of data obtained 130 with the bioET with physicochemical parameters. 131 132 2. Material and methods 133 All the reactants were of analytical grade and were used without further purification. They 134 were purchased form Sigma-Aldrich (St.Louis, USA). All the solutions were prepared in 135 MilliQ deionized water (Merck, KGaA, Darmstadt, Germany). 136 137 2.1 Milk samples 138 A set of 120 milk samples corresponding to 12 types of commercial milk types (ten 139 replicas from each milk) were included in the study. This set was formed by milks with 140 different fat content (skimmed, semi-skimmed and whole milk) and nutritional content 141 (lactose-free, calcium-enriched, and folic acid-enriched milk). The milks were analyzed 142 using traditional standard chemical methods: the titratrion method for acidity (ISO 143 22113:2012), the Hydrometer method for density (ISO 2449:1974), the Gravimetry Röse144 Journal Pre-proof 4 Gottlieb method for fat content (ISO 1211:2010), the Kjeldahl method for protein content 145 (ISO 8968-1:2014), HPLC to determine the lactose content (ISO 22662:2007), and 146 Infrared spectroscopy for the urea content (ISO 9622:2013). Total dry matter (DM) and 147 non-fat dry matter (NFDM) were also analyzed (ISO 6731:2010) (International 148 Organization For Standardization, 2021). The physicochemical data are summarized in 149 Table 1. 150 151 Table 1. Milk samples and physicochemical parameters established by traditional 152 standard methods 153 Sample Fat content Nutritional description Acidity (ºD) Density (g/ml) Fat (%m) Proteins (%m) Lactose (%m) NFDM (%m) DM (%m) Urea (mg/ml) S1 Skimmed Classic 12.55 1031.55 0.31 3.3 5 9.02 9.33 387 S2 Skimmed Calcium 15.82 1039.47 0.29 3.93 5.59 10.51 10.8 724 S3 Skimmed Lactose Free 12.66 1033.57 0.32 3.29 0.36 9.02 9.33 <10 S4 Skimmed Folic Acid 12.57 1033.7 0.40 3.29 4.95 9.04 9.43 586 S5 SemiSkimmed Classic 12.55 1031.6 1.56 3.27 4.91 8.91 10.47 355 S6 SemiSkimmed Calcium 16.06 1037.29 1.55 3.9 5.49 10.40 11.95 597 S7 SemiSkimmed Lactose Free 12.19 1032.09 1.59 3.31 0.42 8.99 10.57 <10 S8 SemiSkimmed Folic Acid 12.95 1032.38 1.64 3.21 4.93 8.94 10.58 638 S9 Whole Classic 12.17 1029.38 3.56 3.21 4.85 8.78 12.33 388 S10 Whole Calcium 15.86 1035.71 3.55 3.91 5.54 10.45 14.0 769 S11 Whole Lactose Free 11.98 1029.4 3.59 3.23 0.31 8.82 12.41 <10 S12 Whole Folic Acid 12.72 1030.55 3.1 3.18 4.94 8.92 12.02 792 154 155 2.2 Sensors and biosensors 156 Gold nanoparticles were synthetized by reduction of tetrachloroauric in the presence of 157 trisodium citrate as the reducing agent, using the classical Turkevich method (Kimling et 158 al., 2006). The colloid obtained was characterized by UV-Vis, showing a maximum at 159 537 nm. The concentration of AuNPs was calculated by Beer’s law, with a particle 160 concentration result of 5.98 x 10-11 M and a diameter of 52.1 nm (Haiss, Nguyen, 161 Aveyard, & Fernig, 2007). 162 163 Sensors were based on polymeric membranes made of carboxylated PVC [poly (vinyl 164 chloride) carboxylate] (C-PVC) as the polymeric matrix. The C-PVC was mixed with an 165 additive (oleyl alcohol) and a plasticizer [(bis(1-butylpentyl) adipate (named plasticizer 166 A), tris(2-ethylhexyl) phosphate (named plasticizer B) or 2-nitrophenyl-octylether 167 (named plastizicer C)] using tetrahydrofurane as the solvent. A second set of sensors was 168 prepared by introducing gold nanoparticles in the membrane. 169 The membranes described above were modified with galactose oxidase (GaOx) from 170 Dactylium dendroides (Sigma-Aldrich, St. Louis, USA), lactate dehydrogenase (LDH) 171 from Mus musculus (Roche diagnostics, Indianapolis, USA), and urease (Ure) from 172 Journal Pre-proof 5 Canavalia ensiformis (Sigma-Aldrich, St. Louis, USA). The enzymes were covalently 173 linked to the surface of the polymeric membrane using the carbodiimide method 174 (Kazenwadel, Wagner, Rapp, & Franzreb, 2015). The reaction was carried out in two 175 steps. First, the carboxylic groups of the C-PVC were activated by means of EDC (1176 Ethyl-3-(3-dimethylaminopropyl) carbodiimide. Then, the enzyme was added and a 177 peptide bond was formed between the carboxylic groups on the C-PVC and the superficial 178 amino side chains of the enzyme. As a result of the combination of the six membranes 179 with each of the three enzymes (GaOx, Ure and LDH) a set of 24 membranes were 180 obtained (Table 2). 181 182 Table 2. Composition of the sensors. 183 184 Sensor C-PVC (w/w%) Additive (w/w %) Plasticizer (P) Type (w/w%) AuNPs (w/w%) Enzyme A 32 3 A 65 - - A-GaOx GaOx A-Ure Ure A-LDH LDH B 32 3 B 65 - - B-GaOx GaOx B-Ure Ure B-LDH LDH C 32 3 C 65 - - C-GaOx GaOx C-Ure Ure C-LDH LDH A-AuNP 32 3 A 55 10 - A-AuNP-GaOx GaOx A-AuNP-Ure Ure A-AuNP-LDH LDH B-AuNP 32 3 B 55 10 - B-AuNP-GaOx GaOx B-AuNP-Ure Ure B-AuNP-LDH LDH C-AuNP 32 3 C 55 10 - C-AuNP-GaOx GaOx C-AuNP-Ure Ure C-AuNP-LDH LDH 185 186 The bioET was designed using a methacrylate tube, in which 24 holes (0.3 cm diameter) 187 were drilled. The holes were half-filled with an epoxy silver resin (EPO-TEK, Billerica, 188 USA) and the resin was covered with one of the 24 membranes. The inner part of the 189 silver epoxy resin was connected to a data acquisition system (Agilent Data Acquisition 190 Switch Unit 34970A). In all measurements, the Ag/AgCl electrode was used as the 191 reference electrode. Figure 1 shows the scheme of the designed bioET system. 192 193 Journal Pre-proof 6 194 195 196 Figure 1. Scheme of the bioET designed in this work. A) Data Acquisition Switch; B) 197 Reference electrode; C) Electronic tongue body; D) Enzyme covalently linked; E) C-PVC 198 membrane; F) Silver epoxy resin and copper wire. 199 200 The potentiometric measurements were carried out by immersing the sensor array in a 201 100 ml glass cell containing the standard solutions or the milk samples. Standard solutions 202 of compounds usually found in milk (KCl, CaCl2, galactose, urea and lactic acid) were 203 prepared in a phosphate buffer (0.1M, pH 7) with concentrations ranging from 1 × 10−4 204 to 1 × 10−2 M. The milks were diluted 1:1 in phosphate buffer and measured without 205 further modification. In addition, nicotinamide adenine (NAD+) (Roche diagnostics, 206 Indianapolis, USA) was added to the standard solutions in order to simulate the levels 207 usually present in milk (final concentration 12 mM) (Fox, & McSweeney,1998). After 208 immersing the electrodes in the corresponding sample, the membrane potentials were 209 registered every three seconds. The signals were stabilized after 5 minutes (average 210 variation of 1.6 mV/decade between each reading). 211 The potentials obtained from the sensor array were used as the input variables for 212 multivariate analysis. Principal Component Analysis (PCA) was used to estimate the 213 discrimination ability of the multisensory system. A Support Vector Machine (SVM) 214 was applied to establish correlations with the physicochemical parameters obtained using 215 traditional methods (Theodore, & Robin, 2006; Cortes, & Vapnik, 1995). Additionally, 216 the SVM was applied to elaborate classification models. Finally, an approach towards 217 ensemble methods was implemented by applying Stochastic Gradient Bosting for 218 regression (Friedman, 2002). The statistical analysis was performed by using Matlab 219 R2020b (The Mathworks Inc., Natick, USA), RKWard 0.7.1, and the Caret package 220 (Kuhn, 2008). 221 222 3. Results and discussion 223 224 3.1 Development and optimization of the sensor array 225 In order to obtain efficient potentiometric biosensors, the immobilization of the enzymes 226 on the polymeric membrane was accomplished using carboxylated PVC (C-PVC) instead 227 of the bare PVC classically used to fabricate potentiometric sensors (Tazi et al., 2018; 228 Dias et al., 2009). Using C-PVC, the enzymes can be immobilized by establishing a 229 covalent link between the carboxylate groups of the membrane and the amine groups of 230 Journal Pre-proof 7 the enzymes. In addition, membranes were doped with AuNPs to further increase the 231 intensity of the signals. As observed in Figure 2, the membrane potential increased with 232 the content of AuNPs in the membrane. For instance, the sensitivity values obtained from 233 the slopes of the calibration curves towards galactose were 17.23 mV for sensor A 234 (without AuNPs), 19.13 mV for sensor A-Au containing 5% of AuNPs, and 32.22 mV 235 for sensor A-Au containing 10% of AuNPs. Higher concentrations of AuNPs did not 236 produce any further improvement in the sensitivity values. Based on these findings, the 237 decision was taken to set the AuNPs content at 10%. 238 239 240 241 242 Figure 2. Response of sensor A (without AuNPs), A-AuNPs5% and A-AuNPs10% to 243 increasing concentrations of galactose. 244 245 Once the composition of the membranes had been optimized, the enzymes GaOx, Ure 246 and LDH were immobilized at the membrane surface and the responses of the obtained 247 biosensors were analyzed. As observed in Figure 3, the intensity of the responses 248 produced by a bare C-PVC membrane were lower than those obtained when the enzymes 249 were covalently linked to the membrane. Taking the case of urea as an example, the 250 measured voltage increased from 0.057 V in the bare C-PVC sensor (sensor A) to 0.119 251 V in the AuNP modified sensor (A-AuNP). The enzyme addition increased the intensity 252 of the responses (0.156 V in A-Ure); and they increased even further to 0.276 V in A253 AuNP-Ure when the enzyme was combined with AuNPs. Similar results were obtained 254 for LDH or GaOx. 255 These results indicate that the enzymes are properly immobilized at the surface of the 256 membrane and the enzymatic activity is retained. The synergistic effect observed when 257 C-PVC and AuNPs are combined in the support membrane is also worth noting. 258 259 Journal Pre-proof 14 within each group of milks according to its nutritional content. Therefore, by 402 incorporating enzymes and AuNPs, the discrimination capacity of the system has been 403 increased. 404 405 3.3 Analysis of milk with the bioET: Classification models 406 The milk classification analysis was based on the features from the nine sensors that make 407 up the simplified bioET by applying the Support Vector Machine classification method 408 (SVMC). The Support Vector Machine (SVM) is a kernel-based supervised pattern 409 recognition technique, established by Cortes and Vapnik and based on statistical learning 410 theory (Cortes, & Vapnik, 1995). Compared with other approaches, SVM possesses the 411 advantages of avoiding overfitting, is capable of establishing non-linear correlations 412 between data sets and of dealing with high-dimensional input. 413 The SVM classification chosen was based on the radial basis function (RBF) as a 414 nonlinear kernel approximation, defined as 415 416 𝑲(𝒙𝒊− 𝒙𝒋) = 𝒆𝒙𝒑⁡((−𝜸 ∥ 𝒙𝒊− 𝒙𝒋∥𝟐),⁡⁡⁡⁡⁡⁡𝜸 > 0 417 where xi and xj are the training vectors of the input data, and γ is the kernel parameter. 418 Before the validation stage, to achieve a better performance, the kernel function penalty 419 parameter (C) and the kernel parameter γ in the SVM were optimized. To optimize these 420 parameters, the grid search method was applied, where approaches were made using 421 log2C and log2γ, varying from [10, 10] at one interval (Cortes, & Vapnik, 1995). The grid 422 points of (C, γ) were confirmed through the validation accuracy in the [10, 10] grid. The 423 results showed that the best validation accuracy was achieved when C=1 and γ=0.1. Due 424 to the relatively small number of samples available, the leave-one-out cross-validation 425 method was used to better evaluate the true success rate that can be reached with the 426 SVM. 427 The classification of the samples was carried out in two steps. Initially, a study was 428 proposed aimed at determining whether the milk samples could be classified based on 429 their lactose content (presence or absence of lactose), as well as folic acid and calcium 430 content (samples with or without enrichments in calcium or folic acid). This led to the 431 development of three different classification models. 432 The results obtained for each of the models were the following: 98.2% calibration 433 accuracy and 97.2% validation accuracy for milk samples with or without lactose; 96.3% 434 calibration accuracy and 95.8% validation accuracy for samples with folic acid 435 enrichments; and finally, 97.8% accuracy for the calibration and 97.1% in the validation 436 was achieved in the classification model to determine which milk samples were enriched 437 in calcium. All the classification models developed in this approach were able to establish 438 mathematical models with high accuracy values. 439 A second approach was taken as an attempt to classify the analyzed milk samples 440 according to their nutritional composition and their fat content, which resulted in a total 441 of twelve categories. By applying SVMC, the results obtained for the simplified bioET 442 showed 99.7% accuracy in the calibration and 98.4% accuracy in the validation. These 443 results determined that the electronic tongue developed with nine sensors was able to 444 classify milk samples according to their nutritional content as well as for their fat content. 445 446 3.4 Prediction of chemical parameters: Correlations between electronic tongue and 447 chemical analysis 448 Journal Pre-proof 15 One of the main advantages of ETs is the possibility to predict the concentration of several 449 components in a single measurement. For this purpose, mathematical models must be 450 developed to establish correlations between data provided by the sensor array and 451 physicochemical data measured by traditional methods. It is expected that the presence 452 of biosensors could help to achieve good correlations with specific compounds. 453 The simplified bioET developed here was used to predict parameters commonly used to 454 assess the gross composition of milks, including the total amount of fats, total proteins 455 (casein or whey), carbohydrates (lactose), urea, and total solids (dry matter and non-fat 456 dry matter) which is the residue left when water and gases are removed. Only few 457 attempts to use ETs to evaluate the chemical composition of milk have been reported 458 previously (Hruskar et al. 2010; Salvo-Comino et al. 2018; Pérez-González et al. 2021). 459 Support Vector Machine regression was used to determine the nature of the relationships 460 between the data collected by the bioET and the physicochemical parameters. To forecast 461 acidity, density, percentage of protein, lactose, fat, DM and NFDM, the Radial Basis 462 Function was chosen as the core function, since it can handle non-linear interactions 463 between the sensor inputs and the target characteristics. The regression models were 464 created using SVM Regression (epsilon SVM, kernel type: radial basis function, C value: 465 1, cross validation segments size: 15, and standard deviation weighting process in all 466 cases). 467 As observed in Table 5, the values obtained for the coefficients of correlation and errors 468 for the calibration and the prediction reached values of R2 above 0.98 for calibration and 469 prediction, with low errors (RMSE) between 0.101 and 0.139. These high correlation 470 coefficients could be due to the specificity induced by the presence of the biosensors. In 471 fact, the biosensor containing galactose oxidase provides data about galactose; LDH can 472 give information on lactic acid, which is in turn related to the acidity of the milk; while 473 urease can account for the levels of urea. The good correlations with acidity, density fat 474 and dry matter can be attributed to the fact that the enzymes contained in the array are 475 sensitive to pH. In addition, potentiometic measurements are sensitive to the percentage 476 of water (which is inversely proportional to density) and to the fat content (directly related 477 to the conductivity and the double layer at the electrode surface). These results show that 478 the reduced bioET is capable of establishing good correlations with the physicochemical 479 parameters thanks to the selection of the suitable sensors in previous steps of this work. 480 If we compare the results of the regression with the previous work (Pérez-González et al. 481 2021) we observe an increase of the correlation coefficients as well a reduction in the 482 errors (RMSE). The increase in R2 for lactose and acidity is especially remarkable. 483 Correlation coefficients have improved from values of 0.96 and 0.90 respectively in the 484 validation, to 0.99 in both cases. These results demonstrate the effectiveness of the use of 485 biosensors in the composition of a ET providing specific information on compounds of 486 interest in milk, such as lactose, without losing global information of the sample. 487 488 Table 5. Correlation parameters from the SVM regression analysis. 489 490 491 492 Parameters Acidity Density %Proteins %Fat %Lactose %DM %NFDM Urea SVM R2C 0.9953 0.9910 0.9941 0.9956 0.9924 0.9933 0.9991 0.9915 RMSEC 0.1177 0.1376 0.1088 0.1018 0.1093 0.1233 0.1146 0.1187 R2P 0.9946 0.9903 0.9944 0.9951 0.9928 0.9927 0.9982 0.9902 RMSEP 0.1181 0.1396 0.1092 0.1055 0.1113 0.1281 0.1151 0.1193 Journal Pre-proof 16 3.5 Ensemble method development 493 Although the SVM was very capable of establishing mathematical models for the correct 494 classification of the milk samples and the prediction of the physicochemical parameters; 495 here, we aimed to go a step further by establishing correlation models using ensemble 496 methodologies. 497 In the context of machine learning, ensemble methods are commonly defined as a 498 machine learning system, designed with a set of independent models working in parallel, 499 whose outputs are combined with a decision fusion strategy to create a single response 500 for a given problem. Therefore, an ensemble method aims to combine several separate 501 models to achieve a better result than each individual method in terms of consistency and 502 accuracy (Zhou, 2012). 503 The first step in developing an ensemble method is to select the individual methods. 504 Different algorithms may lead to different results for the same data by imposing a specific 505 structure for it. Moreover, there is no single algorithm able to perform consistently well 506 for different problems and there are no clear rules to follow while selecting individual 507 algorithms for a given problem. 508 In principle, any individual models could be used as long as they are suitable for the 509 dataset. In this work case, the Caret Package developed for R is used to select the 510 individual algorithms (Kuhn, 2008). The generated models should be as different from 511 each other as possible. A high level of diversity means that they will be able to capture 512 different information about the data and can overcome the weaknesses of single 513 techniques, since each technique handles the error made by the others. 514 Starting with the SVM regression model (svmRadial), five models were selected to ensure 515 their diversity. For this, the "max.dissim" function of the Caret Package was used, in 516 which the Jaccard dissimilarity function was selected as the diversity criterion. The 517 models selected were: Support Vector Machine (svmRadial), Quasi-recurrent Neural 518 Networks (qrnn), Cubist Regression Model (cubist), Weighed k-nearest neighbor (kknn), 519 and Bagged Earth (bagEarth). 520 The support vector machine was chosen as the starting model due to its great performance 521 in the previous section of this work. Furthermore, SVM is a powerful method widely used 522 in the development of ensemble models. 523 Once the models had been selected, the original data were split into two sets: a training 524 set containing 75% of the original data to be used in the calibration process of each 525 algorithm, and a testing set covering the remaining 25% of the data for validation. It is 526 essential to verify that both sets of data are representative of all the recognized categories; 527 consequently, the percentage of each set is computed in relation to the total data as well 528 as the amount of data in each category. 529 Each algorithm was executed individually, but the control parameters for all of them were 530 established beforehand. Validation was performed using repeated 10-fold cross531 validation, to establish reasonable values for the tuning parameters and random search 532 was established as the preferred method. After each individual algorithm was applied, the 533 Stochastic Gradient Boosting (gbm) method was used to generate the ensemble through 534 the Caret Ensemble package in R (Kuhn, 2008). 535 Stochastic Gradient Boosting is a machine learning algorithm, able to perform 536 classification and regression problems. Gradient Boosting is especially convenient, 537 because of its computational efficiency and robustness to overfitting, as a simple 538 technique to develop ensemble decision trees by creating training trees on subsamples of 539 the training dataset (Friedman, 2002). 540 Journal Pre-proof 17 Table 6 shows the values obtained for the correlation and error coefficients for the 541 calibration and prediction obtained by the ensemble. The coefficients of correlation and 542 mean errors for the calibration and prediction reached values of R2 above 0.9992 for both 543 calibration and prediction, with low errors (RMSE) between 0.0033 and 0.0172. 544 545 Table 6: Correlation parameters from the ensemble regression analysis. 546 547 Considering the high values of the correlation parameters achieved with SVM regression, 548 it was expected that the result of the regression ensemble would reach nearly 100% 549 precision while establishing correlations, since there is a very reduced number of errors 550 in the original model. However, the intention in this section is not to ensure the capability 551 of the simplified bioET to establish correlations with the studied parameters, but to 552 demonstrate the possibility of combining the developed system with ensemble 553 methodologies that could be applied in the study of future and more complex samples, 554 where the settings may not be as good as they should be. 555 556 557 4. Conclusions 558 In this work, a bioET with improved characteristics was developed and used to predict 559 the chemical characteristics of milk with unprecedented accuracy. The system 560 incorporates biosensors based on membranes of carboxylated PVC (C-PVC) containing 561 gold nanoparticles (AuNPs), where GaOx, LDH and Ure were effectively immobilized. 562 The developed biosensors and the associated methodology have resulted in a bioET where 563 the enzymes can work simultaneously while also preserving the enzymatic activity. 564 Nanoparticles have proven to have a potential to amplify the electrochemical signals. 565 The biosensors have shown excellent sensitivity and reproducibility towards standard 566 solutions of compounds usually found in milk (CaCl2, KCl, urea, lactic acid and 567 galactose), with excellent sensitivity and reproducibility, showing LODs of 10-6 M. 568 The bioET was successfully used to discriminate between milks by applying PCA based 569 on their nutritional content. The bioET shows an excellent classification capability and 570 can classify milk with different compositions by applying SVM with accuracies above 571 95%. The system can predict the acidity, density, %proteins, %lactose, %fat and dry 572 matter with low errors and high correlation coefficients. The results show that the SVM 573 models constructed with the e-tongue and physicochemical parameters have potential for 574 use in simultaneously assessing 8 parameters, thus reducing the time of analysis. 575 Moreover, it has been proved that applying ensemble methodologies can further improve 576 the correlation between the bioET data and the physicochemical parameters. 577 Investigations into the efficiency of the prototype devices can create new application 578 possibilities and suggest successful implementations in real applications. 579 580 Declaration of Competing Interest 581 The authors declare that they have no known competing financial interests or personal 582 relationships that could have appeared to influence the work reported in this paper. 583 Parameters Acidity Density %Proteins %Fat %Lactose %DM %NFDM Urea Ensemble R2C 0.9997 0.9999 0.9999 0.9994 0.9999 0.9999 0.9999 0.9999 RMSEC 0.0097 0.0053 0.0041 0.0164 0.0042 0.0037 0.0033 0.0040 R2P 0.9994 0.9997 0.9998 0.9992 0.9998 0.9998 0.9998 0.9998 RMSEP 0.0102 0.0075 0.0056 0.0172 0.0051 0.0045 0.0041 0.0048 Journal Pre-proof 18 Funding: 584 This work was supported by MICINN-FEDER (RTI2018-097990-B-100), Consejería de 585 Educación JCyLFEDER (VA202P20), EU-FEDER program (CLU-2019-04) and 586 «Infraestructuras Red de Castilla y Leon (INFRARED)» 587 588 Authors contribution 589 MR-M, CG-C, and FM-P conceptualized the idea and supervised the work. CP-G and 590 CS-C performed the experiment, curated the data, and wrote the original draft. FM-P 591 involved in software design and development. CP-G and CS-C involved in formal 592 analysis. CG-C and MR-M acquired the funding. CP-G, CS-C, FM-P, MR-M and CG-C 593 reviewed and edited the paper. All authors provided feedback. 594 References 595 Americo da Silva, T., Braunger, M.L., Neris Coutinho, M.A., Rios do Amaral, L., 596 Rodrigues, V. & Riul, A. (2019) 3D-Printed Graphene Electrodes Applied in an 597 Impedimetric Electronic Tongue for Soil Analysis. Chemosensors, 7, 4, 50. 598 http://dx.doi.org/10.3390/chemosensors7040050 599 Aouadi, B., Zinia Zaukuu, J.L., Vitális, F., Bodor, Z., Fehér, O., Gillay, Z., Bazar G. & 600 Kovacs, Z. (2020) Historical Evolution and Food Control Achievements of Near Infrared 601 Spectroscopy, Electronic Nose, and Electronic Tongue—Critical Overview, Sensors, 602 20(19), Article 5479. http://dx.doi.org/10.3390/s20195479 603 Çakıroğlu, B., Demirci, Y.C., Gökgöz, E. & Özacar M. (2019) A Photoelectrochemical 604 Glucose and Lactose Biosensor Consisting of Gold Nanoparticles, MnO2 and g-C3N4 605 decorated TiO2. Sensors and Actuators B: Chemical, 282. 606 http://dx.doi.org/10.1016/j.snb.2018.11.064 607 Chiang, H., Wang, Y., Zhang, Q., & Levon, K. (2019). Optimization of the 608 Electrodeposition of Gold Nanoparticles for the Application of Highly Sensitive, Label609 Free Biosensor. Biosensors, 9(2), 50. https://doi.org/10.3390/bios9020050 610 Ciosek, P. & Wroblewski, W. (2015) Potentiometric And Hybrid Electronic Tongues For 611 Bioprocess Monitoring-An Overview. Analytical Methods 7-9, 3958-3966 612 http://dx.doi.org/10.1039/c5ay00445d. 613 Ciosek, P. (2016) Milk and Dairy Products Analysis by Means of an Electronic Tongue. 614 In Electronic Noses and Tongues in Food Science. Ed. Rodriguez-Mendez. Academic 615 Press. (Chapter 212016), 209-223. http://dx.doi.org/10.1016/B978-0-12-800243616 8.00021-4 617 Collier, W.A., Baird, D.B., Park-Ng, Z.A., More, N. & Hart, A.L. (2003) Discrimination 618 among milks and cultured dairy products using screen-printed electrochemical arrays and 619 an electronic nose, Sensors and Actuators B: Chemical 92, 232–239. 620 https://doi.org/10.1016/S0925-4005(03)00271-5 621 Cortes, C. & Vapnik, V. (1995) Support-vector networks. Machine Learn 20, 273–297; 622 http://dx.doi.org/10.1007/BF00994018 623 Dias, L.A., Peres, A.M., Veloso, A.C.A., Reis, F.S., Vilas-Boas, M. & Machado 624 A.A.S.C. (2009) An electronic tongue taste evaluation: identification of goat milk 625 adulteration with bovine milk. Sensors and Actuators B: Chemical 136, 209–217. 626 http://dx.doi.org/10.1016/J.SNB.2008.09.025 627 Fox, P.F. & McSweeney, P.L.H. (1998) Dairy Chemistry and Biochemistry, Springer 628 Science & Business Media. 629 Journal Pre-proof 19 Friedman, J.H. (2002) Stochastic gradient boosting, Computational Statistics & Data 630 Analysis, 30, 4, 367-378. http://dx.doi.org/10.1016/S0167-9473(01)00065-2 631 Garcia-Cabezón, C., Gobbi, G., Dias, L.G., Salvo-Comino, C., García-Hernandez, C., 632 Rodriguez-Mendez, M.L. & Martin-Pedrosa, F. (2020) Analysis of Phenolic Content in 633 Grape Seeds and Skin by means of a BioElectronic Tongue Using Electrochemical 634 Modified Biosensors with Phthalocyanines, NiO Nanoparticles and Tyrosinase. Sensors, 635 20, 15, Article 4176. http://dx.doi.org/10.3390%2Fs20154176 636 Garcia-Hernandez, C., Garcia-Cabezon, C., Martin-Pedrosa, F., de Saja, J.A. & 637 Rodriguez-Mendez, M.L. (2019) Analysis of Musts and Wines with a bio-electronic 638 tongue based on Tyrosinase and Glucose oxidase using a Polypyrrole/gold nanoparticle 639 composite as electron mediator. Food Chemistry, 289, 751-756. 640 http://dx.doi.org/10.1016/j.foodchem.2019.03.107 641 Ghasemi-Vamankhasti, M., Rodríguez-Méndez, M.L., Mohtasebi, S.S., Apetrei, C., 642 Lozano, J., Razavi, S.H., Ahmadi, H. & de Saja, J.A. (2012) Monitoring the aging of 643 beers using a bioelectronic tongue, Food Control, 25, 216-224; 644 http://dx.doi.org/10.1016/j.foodcont.2011.10.020 645 Ha, D., Sun, Q.Y., Su, K., Wan, H., Haibo, L., Xu, N., Sun, F., Zhuang, L., Hu, N. & 646 Wang, P., (2015) Recent achievements in electronic tongue and bioelectronic tongue as 647 taste sensors. Sensors and Actuators B: Chemical, 207, 1136-1146; 648 http://dx.doi.org/10.1016/j.snb.2014.09.077 649 Haiss, W., Nguyen, T.T.K., Aveyard, J. & Fernig, D.G. (2007) Determination of Size and 650 Concentration of Gold Nanoparticles from UV−Vis Spectra, Analytical Chemistry 651 (2007), 4215-4221; http://dx.doi.org/10.1021/ac0702084 652 Hruškar, M., Major, N., Krpan, M. & Vahčić, N. (2010) Simultaneous determination of 653 fermented milk aroma compounds by a potentiometric sensor array. Talanta. 82, 1292– 654 1297. http://dx.doi.org/10.1016/j.talanta.2010.06.048. 655 International Organization For Standardization (2021) ISO/TC 34/SC 5 - Milk and Milk 656 Products. https://www.iso.org/committee/47878.html. 657 Jakhar, S. & Pundir, C.S. (2018) Preparation, characterization and application of urease 658 nanoparticles for construction of an improved potentiometric urea biosensor. 659 Biosensors and Bioelectronics. 100, 242-250. 660 http://dx.doi.org/10.1016/j.bios.2017.09.005 661 Jia, W., Su, L. & Lei Y. (2011) Pt nanoflower/polyaniline composite nanofibers based 662 urea biosensor. Biosensors and Bioelectronics. 30, 1, 158-164. 663 http://dx.doi.org/10.1016/j.bios.2011.09.006 664 Juzhong & T., Jie, X. (2020). Applications of electronic nose (e-nose) and electronic 665 tongue (e-tongue) in food quality-related properties determination: A review, Artificial 666 Intelligence in Agriculture, 4, 104-115. http://dx.doi.org/10.1016/j.aiia.2020.06.003 667 Kanyong, P., Krampa, F.D., Aniweh, Y. & Awandare, G.A. (2019) Enzyme-based 668 amperometric galactose biosensors: A review. Microchimica acta, 184, 10, 3663-3671; 669 http://dx.doi.org/10.1007/s00604-017-2465-z 670 Kazenwadel, F., Wagner, H., Rapp, B.E., Franzreb, M. (2015) Optimization Of Enzyme 671 Immobilization On Magnetic Microparticles Using 1-Ethyl-3-(3672 Journal Pre-proof 20 Dimethylaminopropyl) Carbodiimide (EDC) As A Crosslinking Agent. Analytical 673 Methods. 7, 24, 10291-10298, http://dx.doi.org/10.1039/c5ay02670a 674 Kim, T.H., Song, H.S., Jin, H.J., Lee, S.H., Namgung, S., Kim, U.K., Park, T.H. & 675 Hong, S. (2011) Bioelec-tronic super-taster device based on taste receptor-carbon 676 nanotube hybridstructures. Lab on a Chip, 11, 2262–2267. 677 http://dx.doi.org/10.1039/c0lc00648c 678 Kimling, J., Maier, M., Okenve, B., Kotaidis, V., Ballot, H. & Plech, A. (2006) Turkevich 679 Method for Gold Nanoparticle Synthesis Revisited. J. Phys. Chem. B. 110, 15700-15707. 680 http://dx.doi.org/10.1021/jp061667w 681 Kuhn, M. (2008) Building Predictive Models in R Using the caret Package. Journal of 682 Statistical Software, 28, 5, 1–26, http://dx.doi.org/10.18637/jss.v028.i05 683 Lee, J., Lee, I., Nam, J., Hwang, D.S, Yeon, K.M. & Kim, J. (2017) Immobilization And 684 Stabilization Of Acylase On Carboxylated Polyaniline Nanofibers For Highly Effective 685 Antifouling Application Via Quorum Quenching, ACS Applied Materials & Interfaces, 686 9,18, 15424-15432; http://dx.doi.org/10.1021/acsami.7b01528 687 Li, L., Yu, Y., Yang, J., Yang, R., Dong, G. & Jin, T. (2015) Voltammetric electronic 688 tongue for the qualitative analysis of milk adulterated with urea combined with multi689 way data analysis. International Journal of Electrochemistry. 10, 5970–5980. 690 http://dx.doi.org/10.1016/j.jfoodeng.2018.09.022 691 Li, X., Li, S., Liu, Q., & Chen, Z. (2019) Electronic-Tongue Colorimetric-Sensor Array 692 for Discrimination and Quantitation of Metal Ions Based on Gold-Nanoparticle 693 Aggregation. Anal. Chem., 91, 9, 6315–6320. 694 http://dx.doi.org/10.1021/acs.analchem.9b01139 695 Mangan, D., McCleary, B.V., Culleton, H., Cornaggio, C., Ivory, R., McKie, V.A., 696 Delaney, E. & Kargelis, T. (2018) A novel enzymatic method for the measurement of 697 lactose in lactose-free products. Journal of the Science of Food and Agriculture. 99, 947698 956; http://dx.doi.org/10.1002/jsfa.9317 699 Mercante, L.A., Scagion, V.P., Pavinatto, A., Sanfelice, R.C., Mattoso, L.H.C. & Correa, 700 D.S. (2015) Electronic tongue based on nanostructured hybrid films of gold nanoparticles 701 and phthalocyanines for milk analysis, Journal of Nanomaterials, 16, Article 402; 702 https://doi.org/10.1155/2015/890637 703 Migliorini, F.L., Sanfelice, R.C., Mercante, L.A., Andre, R.S., Mattoso, L.H.C. & Correa, 704 D.S. (2018) Urea impedimetric biosensing using electrospun nanofibers modified with 705 zinc oxide nanoparticles. Applied Surface Science. 443, 18-23. 706 http://dx.doi.org/10.1016/j.apsusc.2018.02.168 707 Nguyen, B.H., Nguyen, B.T., Van Vu, H., Van Nguyen, C., Nguyen, D.T., Nguyen, L.T, 708 Thi Vu, T. & Tran, L.D. (2016) Development of label-free electrochemical lactose 709 biosensor based on graphene/poly(1,5-diaminonaphthalene) film. Current Applied 710 Physics Volume 16, 2, 135-140. http://dx.doi.org/10.1016/j.cap.2015.11.004 711 Nguyen, H.H. & Kim, M. (2017) An overview of techniques in enzyme immobilization. 712 Appl. Sci. Converg. Technol. 26, 157-163. 713 http://dx.doi.org/10.5757/ASCT.2017.26.6.157 714 Ohlsson, J.A., Johansson, M., Hansson, H., Abrahamson, A., Byberg, L., Smedman, A., 715 Lindmark-Månsson, H. & Lundh, Å. (2017) Lactose, glucose and galactose content in 716 Journal Pre-proof 21 milk, fermented milk and lactose-free milk products, International Dairy Journal, 73, 151717 154. http://dx.doi.org/10.1016/j.idairyj.2017.06.004 718 Ohlsson, J.A., Johansson, M., Hansson, H., Abrahamson, A., Byberg, L., Smedman, A., 719 Lindmark-Månsson, H. & Lundh, Å. (2017) Lactose, glucose and galactose content in 720 milk, fermented milk and lactose-free milk products, International Dairy Journal. 73, 721 151-154. http://dx.doi.org/10.1016/j.idairyj.2017.06.004 722 Pascual, L., Gras, M., Vidal-Brotóns, D., Alcañiz, M., Martínez-Máñez, R. & Ros-Lis, 723 J. V. (2018) A voltammetric e-tongue tool for the emulation of the sensorial analysis and 724 the discrimination of vegetal milks. Sensors and Actuators B: Chemical, 270, 231-238. 725 http://dx.doi.org/10.1016/j.snb.2018.04.151 726 Pérez-González, C., Salvo-Comino, C., Martin-Pedrosa, F., Dias, L., Rodriguez-Perez, 727 M.A., Garcia-Cabezón, C. & Rodriguez-Mendez, M. L. (2021) Analysis Of Milk Using 728 A Portable Potentiometric Electronic Tongue Based On Five Polymeric Membrane 729 Sensors, Frontiers In Chemistry 9. http://dx.doi.org/10.3389/fchem.2021.706460. 730 Prieto, N., Oliveri, P., Leardi, R., Gay, M., Apetrei, C., Rodriguez-Méndez, M.L. & de 731 Saja, J.A. (2013) Application of a GA-PLS strategy for variable reduction of electronic 732 tongue signals. Sensors and Actuators B: Chemical, Volume 183, 5257. 733 http://dx.doi.org/10.1016/j.snb.2013.03.114. 734 Rahman, M.M., Shiddiky, M.J.A., Rahman, M.A. & Shim, Y. (2009) A lactate biosensor 735 based on lactate dehydrogenase/nictotinamide adenine dinucleotide (oxidized form) 736 immobilized on a conducting polymer/multiwall carbon nanotube composite film. 737 Analytical Biochemistry. 384, 1, 159-165. http://dx.doi.org/10.1016/j.ab.2008.09.030 738 Rodriguez-Mendez, M.L., De Saja, J.A. & González-Antón, R. (2016) Electronic Noses 739 and Tongues in Wine Industry. Frontiers in bioengineering and biotechnology, 4, 81. 740 http://dx.doi.org/10.3389/fbioe.2016.00081 741 Salvo-Comino, C., García-Hernández, C., García-Cabezón, C. & Rodríguez-Méndez, 742 M.L. (2018) Discrimination of Milks with a Multisensor System Based on Layer-by743 Layer Films, Sensors, 18, 8, Article 2716. https://doi.org/10.3390/s18082716 744 Scagiona, V.P., Mercante, L.A., Sakamoto, K.Y., Oliveira, J.E., Fonseca, F.J., Mattoso, 745 L.H.C., Ferreira, M.D. & Correa, D.S. (2016) An electronic tongue based on conducting 746 electrospun nanofibers for detecting tetracycline in milk samples, RSC Adv., 6. 747 http://dx.doi.org/10.1039/C6RA21326J 748 Skladal, P. (2020) Smart bioelectronic tongues for food and drinks control. Trac-Trends 749 In Analytical Chemistry, 127, Article 115887. 750 http://dx.doi.org/10.1016/j.trac.2020.115887 751 Sobrino-Gregorio, L., Bataller, R., Soto, J. & Escriche, I. (2018) Monitoring honey 752 adulteration with sugar syrups using an automatic pulse voltammetric electronic tongue. 753 Food Control, 91, 254-260. https://doi.org/10.1016/j.foodcont.2018.04.003. 754 Song, H.S., Kwon, O.S., Lee, S.H., Park, S.J., Kim, U.K., Jang, J. & Park, T.H. (2012) 755 Human taste receptor-functionalized field effect transistor as a human-like 756 nanobioelectronic tongue. Nano Letters, 13, 172–178. 757 http://dx.doi.org/10.1021/nl3038147. 758 Tazi, I., Triyana, K., Siswanta, D., Veloso, A.C.A., Peres, A.M. & Dias, L.G. (2018) 759 Dairy products discrimination according to the milk type using an electrochemical 760 Journal Pre-proof 22 multisensor device coupled with chemometric tools, Journal of Food Measurement and 761 Characterization 12, 2385–2393. http://dx.doi.org/10.1007/s11694-018-9855-8. 762 Teodoro, K.B.R., Shimizu, F.M., Scagion, V.P. & Correa, D. S. (2019) Ternary 763 nanocomposites based on cellulose nanowhiskers, silver nanoparticles and electrospun 764 nanofibers: Use in an electronic tongue for heavy metal detection. Sensors and Actuators 765 B: Chemical, 290, 387-395; http://dx.doi.org/10.1016/j.snb.2019.03.125 766 Theodore, B.T. & Robin, C.G. (2006) Robust classification and regression using 767 support vector machines, European Journal of Operational Research, 173, 3, 893-909. 768 http://dx.doi.org/10.1016/j.ejor.2005.07.024. 769 Tian, S., Chen, M., Pan, W. & Deng, S. (2013) A comparison study of three nonlinear 770 multivariate data analysis methods in smartongue: Kernel PCA LLE and Sammon 771 mapping. J. Chem. Pharm. Res, 5, 575-783. ISSN : 0975-7384. 772 Vaghela, C., Kulkarni, M., Haram, S., Aiyer, R., & Karve, M. (2018). A novel inhibition 773 based biosensor using urease nanoconjugate entrapped biocomposite membrane for 774 potentiometric glyphosate detection. International Journal Of Biological 775 Macromolecules, 108, 32-40. https://doi.org/10.1016/j.ijbiomac.2017.11.136 776 Valente, N.I.P., Rudnitskaya, A., Oliveira, J.A.B.P., Gaspar, E.M.M. & Gomes, M.T.S.R. (2018) Quartz 777 crystals Cheeses Made from Raw and Pasteurized Cow’s Milk Analysed by an Electronic Nose and an 778 Electronic Tongue, Sensors 18,8, Article 2415. https://doi.org/10.3390/s18082415 779 Wang, Q. & del Valle, W. (2021) Determination of Chemical Oxygen Demand (COD) 780 Using Nanoparticle-Modified Voltammetric Sensors and Electronic Tongue Principles. 781 Chemosensors, 9, 3, 46. http://dx.doi.org/10.3390/chemosensors9030046 782 Wasilewski, T., Kamysz, W. & Gebicki, J. (2020) Bioelectronic tongue: Current status 783 and perspectives, Biosensors & Bioelectronics, 150, 11, Article 1923; 784 http://dx.doi.org/10.1016/j.bios.2019.111923 785 Wei, Z., Wang, J. & Jin, W. (2013) Evaluation of varieties of set yogurts and their 786 physical properties using a voltammetric electronic tongue based on various potential 787 waveforms Sensors and Actuators B: Chemical. 177, 684-694. 788 http://dx.doi.org/10.1016/j.snb.2012.11.056 789 Winquist, F., Krantz-Rulcker, C., Wide, P. & Lundstrom, I. (1998) Monitoring of fresh 790 milk, by an electronic tongue on the basis of voltammetry. Measurement Science and 791 Technology, 9, 1937–1946Z. https://doi.org/10.1088/0957-0233%2F9%2F12%2F002 792 Xiaoyan, W.P.Z. (2003) Model Selection of SVM with RBF Kernel and its Application. 793 Computer Engineering and Applications, 24, 021. http://dx.doi.org/10.1007/11875581_6 794 Yang, Z., Miao, N., Zhang, X., Li, Q., Wang, Z., Li, C., Sun, X. & Lan,Y. (2021) 795 Employment of an electronic tongue combined with deep learning and transfer learning 796 for discriminating the storage time of Pu-erh tea, Food Control. 121, 107608. 797 https://doi.org/10.1016/j.foodcont.2020.107608. 798 Yhan, S.M., Rosario, D., Silva, L.R.G., Santos, F.D., Santos, L.P., Janegitz, B.C., 799 Filgueiras, P.R., Romão, W., Ferreira, R.Q. & Conte-Junior, C.A. (2021) Portable 800 electronic tongue based on screen-printed electrodes coupled with chemometrics for rapid 801 differentiation of Brazilian lager beer. Food Control, 127, Article 108163. 802 https://doi.org/10.1016/j.foodcont.2021.108163 803 Yu, Y., Zhao, H., Yang, R., Dong, G., Li, L., Yang, J., Jin, T., Zhang, W., Liu & Y. (2015) 804 Journal Pre-proof 23 Pure milk brands classification by means of a voltammetric electronic tongue and 805 multivariate analysis. International Journal of Electrochemistry, 10, 4381–4392. 806 Zhou, Z. H. (2012) Ensemble methods: foundations and algorithms. CRC press. 807 Zucca, P. & Sanjust, E. (2014) Inorganic Materials As Supports For Covalent Enzyme 808 Immobilization: Methods And Mechanisms. Molecules. 19, 9, 14139-14194. 809 http://dx.doi.org/10.3390/molecules190914139 810 811 812 Biographies 813 814 Clara Perez-Gonzalez obtained the Ms in Nanoscience in 2019 (U. Valladolid. Spain). 815 She is currently working on her PhD Thesis which is dedicated to the development of 816 electrochemical sensors for the analysis of foods. She is author of 5 scientific papers. 817 818 Coral Salvo-Comino obtained the Ms in Analytical Chemistry in 2015 (U. Complutense. 819 Madrid. Spain). She is currently working on her PhD Thesis which is dedicated to the 820 development of electrochemical sensors for the analysis of foods. She is author of 13 821 scientific papers. 822 823 Fernando Martin-Pedrosa is full professor at the University of Valladolid and Head of the 824 Department of Materials Science. His research is dedicated to electrochemistry studies of 825 different solid materials. He is author of more than 80 papers. 826 827 Cristina Garcia Cabezón, is assistant professor at the Engineers school of the University 828 of Valladolid. She is an expert in electrochemistry and impedance spectroscopy. She is 829 author or coauthor of more than 50 papers in the field. 830 831 Maria Luz Rodriguez-Mendez is Full professor of Inorganic Chemistry at the Engineers 832 School of the University of Valladolid and Head of the group of sensors UVASens. She 833 is leading several funded Projects devoted to the development of arrays of voltammetric 834 nanostructured sensors and biosensors for the characterization of foods. She is author or 835 co-author of over 165 publications (H index 44), seven books and three patents in the 836 field. 837 838 839 840 841 842 843 844 845 846 847 Journal Pre-proof