{"paper_id":"252acb19-08a1-4238-a443-b68b832bf051","body_text":"Helicobacter pylori  ( H. pylori ) is a Gram-negative, microaerophilic, pathogenic bacterium, colonizing the human gastric mucosa. These bacteria were isolated for the first time by Marshall and Warren in 1984 from the stomachs of patients with gastritis [ 1 ]. About 50% of the world population is infected with  H. pylori.  However, there are areas where the rate of infection reaches 80–90% [ 2 ]. The infection occurs most often in childhood and if left untreated, can persist throughout life.  H. pylori  is an etiological agent of chronic gastritis, gastric and duodenal ulcers, malignant diseases: mucosa-associated lymphoid tissue lymphoma (MALT), and gastric cancer [ 2 , 3 , 4 , 5 , 6 , 7 ].  H. pylori  induce cellular and humoral immune responses of the host. However, the chronic character of  H. pylori  infections suggests that the immune system is not able to eradicate these bacteria [ 8 , 9 ].\nSome  H. pylori  antigens, including urease and vacuolating cytotoxic (VacA) or cytotoxin associated gene A (CagA) protein increase the inflammatory response, while others such as lipopolysaccharide (LPS) inhibit the activity of immune cells [ 10 , 11 , 12 , 13 ]. Knowledge about the pathogenesis and different courses of  H. pylori  infections is insufficient. Therefore, animal models that follow the natural history of  H. pylori  infection and related inflammatory and immune processes are needed. So far, mice, Mongolian gerbils, guinea pigs ( Caviae porcellus ), dogs, cats, pigs, and monkeys have been used [ 14 , 15 ]. Guinea pigs are considered the optimal model because they share similarities with humans in terms of immune responses, complement proteins, thymus and bone marrow physiology, pulmonary physiology, and corticosteroid response [ 16 , 17 , 18 ]. Guinea pigs are not naturally infected with  Helicobacter  spp. However, they are susceptible to  H. pylori  infection, which can be confirmed by histological examination of gastric tissue specimens for inflammatory response and  Helicobacter -like organisms, detection of specific nucleotide sequences using molecular methods such as polymerase chain reaction (PCR), and serological examination of serum samples for specific anti- H. pylori  antibodies or  H. pylori  antigens in stool samples [ 19 ].\nThe varied course of  H. pylori  infections prompts researchers to search for new diagnostic methods that would enable the determination of soluble markers qualitatively and quantitatively, which would be helpful in understanding the course of infection and its consequences. Recently, fast physical methods like Fourier-transform infrared spectroscopy (FT-IR) began to be used for the diagnosis of diseases and for monitoring cellular alterations based on spectral analysis of biological fluids such as blood, serum, saliva, and urine [ 20 , 21 ].\nThere are two main types of IR spectroscopy: transmission spectroscopy, in which the intensity of radiation passing through the sample is measured and reflection spectroscopy, in which the intensity of the IR radiation reflected by the tested sample is measured. In the case of reflection spectroscopy, there are three techniques of measurement: attenuated total reflectance (ATR), diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS), and FT-IR microscopy. In the ATR technique, the IR radiation passes through the crystal and the tested sample closely adheres to its surface. As a result of the total reflection of the IR beam from the crystal surface, an evanescent wave is created which penetrates the sample. The examples of the use of ATR-IR in biological samples analysis are summarized in  Table 1 .\nFT-IR spectra of biological samples can be divided into groups of components with typical absorption bands in the wavenumber windows (W):W1—fatty acids (wavenumber range of 3000 cm −1  to 2800 cm −1 ), W2—peptides and proteins (wavenumber range of 1800 cm −1  to 1500 cm −1 ), W3—proteins, phosphate-carrying compounds and fatty acids (wavenumber range of 1500 cm −1  to 1200 cm −1 ), W4—carbohydrates (wavenumber range of 1200 cm −1  to 900 cm −1 ). The window W5 (wavenumber range of 900 cm −1  to 750 cm −1 ) corresponds to specific peaks unique for the biological samples [ 52 , 53 , 54 , 55 ]. The aim of this study was to determine the specific characteristics of IR spectra for  H. pylori  infected guinea pigs in terms of production of antibodies as well as inflammatory and metabolic biomarkers related to experimentally induced infection. The potential usefulness of this technique in the guinea pig model presumably will allow the application of this methodology to the analysis of human biological samples.\n\nIn our model, the status of  H .  pylori  infection in guinea pigs was confirmed at 28 days after inoculation of animals with these bacteria, by histological, molecular, and serological methods, as previously described [ 56 , 57 ]. The gastric mucosa of guinea pigs inoculated with  H .  pylori  was colonized by bacteria, as shown by Giemsa and Warthin-Starry staining to detect  Helicobacter -like organisms. Tissue staining with hematoxylin and eosin also showed inflammation within the gastric mucosa. In the gastric tissue of infected animals,  cagA  and  ureC  sequences encoding CagA protein and subunit C of urease, respectively, were detected by polymerase chain reaction. These sequences were not detected in noninfected animals. The infected animals responded to  H. pylori  bacteria by producing anti- H .  pylori  immunoglobulins (Igs) of IgM and IgG class ( Table 1 ). The level of serum IgM and IgG antibodies towards  H. pylori  antigens, which were detected with use of glycine acid extract (GE) from the reference strain, was significantly higher in  H. pylori  infected animals compared to noninfected animals ( Table 2 ).\nThe level of C-reactive protein (CRP), which is a key systemic marker of inflammation, in serum samples from  H. pylori  infected animals, was also upregulated ( Figure 1 A). The level of tumor necrosis factor (TNF), which is a pro-inflammatory cytokine, in these samples was significantly lower than that in samples from control animals ( Figure 1 B).\nThe IR spectra of serum samples can be divided into different windows (W) corresponding to different components characterized by absorption bands: fatty acids (Window—W1: wavenumber range of 3000 cm −1  to 2800 cm −1 ); peptides and proteins (W2: wavenumber range of 1800 cm −1  to 1500 cm −1 ); proteins, phosphate-carrying compounds and fatty acid (W3: wavenumber range of 1500 cm −1  to 1200 cm −1 ); W4—carbohydrates (wavenumber range of 1200 cm −1  to 900 cm −1 ). The window W5 (wavenumber range of 900 cm −1  to 750 cm −1 ) corresponds to a specific peak unique to biological samples ( Figure 2 ).\nVibration band assignment was carried out in the infrared spectra of sera by comparing the position, relative intensity, and shape of the bands with the corresponding features of the associated molecular bands. The infrared spectra of serum samples from uninfected and  H. pylori  infected animals are shown in  Figure 2 . The absorption bands of proteins, lipids, and carbohydrates were considered during the analysis.\nSpecific molecules, which were identified in the composition of the IR spectra of guinea pig sera, such as glucose (part B1), α2-globulins (part B2), IgM (part B3), IgG1 (part B4), transferrin (part B5), IgG4 (part B6), CRP (part B7), and TNF (part B8), allowed the differentiation between  H. pylori  infected and uninfected animals ( Figure 2 A,B and  Table 3 ). A vibration band assignment was made concerning frequencies of the chemical groups present in the sample. As shown in  Table 2 , in the spectral region (B1) corresponding to glucose, the prominent absorption band at 1062–997 cm −1  is due to the C–O symmetric stretching of the glucose region and the C–O stretching in carboxylic acids. The absorption band B2 (component group: amino acid), is characterized by the α2-globulins absorption vibration peaks at: 1060–116 cm −1  C–C–C bending, C–O stretching, and C–N stretching ( Figure 2  and  Table 3 ). The spectral region B3, which is typical of proteins, phosphate-carrying compounds, and fatty acids (component group of carboxylic acid and amino acid), corresponds to the vibration band characteristics of IgM antibodies. The typical parameters of this region are as follows: N–O symmetric stretching, O–H bending, and methyl symmetric deformation. The main absorption peak identified for IgG was 1419–1361cm −1  (B4), due to C–H rocking, C–C stretching, and methyl symmetric deformations. The absorption peaks of transferrin were identified at 1428–1363 cm −1  (B5), and can include chemical bonds: CH 2  wagging and O–H bending of carboxylic acid. The absorption bands of IgG4 (B6) at 1538–1505 cm −1  correspond to C–N stretching and NH bending in amide II. The spectral region between 1541 cm −1  and 1600 cm −1  (B7, CRP) is predominantly related to C–C stretching, NH bending, and N–H in plane bending vibrations strongly coupled with C–N stretching of protein (amide II). The C=O symmetric stretching, C=C stretching, and NH 2  scissoring vibrations were found at 1652–1628 cm −1  (B8, TNF) ( Figure 2 A,B and  Table 1 ).\nThe regions of the infrared spectra that are correlated with  H. pylori  infection in guinea pigs were determined by using the chi-square statistical test. To perform the cluster analysis (HCA), 10 wavenumbers strongly correlated with this infection were considered ( Table 4 ).\nThe HCA analysis of the W2–W4 regions of IR spectra and their first derivatives showed two distinct clusters ( Figure 3 ). Cluster I corresponds to noninfected animals, while cluster II is representative for  H. pylori  infected animals. One wavenumber appears in the W4 window, which is associated with the vibrations generated by hydrocarbons and nucleic acids. The next five wavenumbers were correlated with  H. pylori  infection in the window W3. This window is associated with vibrations of different bonds (so-called mixed region); however, proteins and phosphorus-containing compounds are key to this window.\nThe wavenumbers of 1552 cm −1 , 1541 cm −1 , and 1630 cm −1 , localized within the W2 window were strongly associated with  H. pylori  infection. According to the literature this window is characteristic for proteins and peptides. The peaks in this region of the spectrum for  H. pylori  infected animals may reflect production of antibodies and/or an increased production of inflammatory proteins. In  H. pylori  infected animals, it may correspond to the production of antibodies and/or greater production of inflammatory proteins. The entire IR spectra range (W2–W4) has been used to develop a k-nearest neighbor predictive model to discriminate  H. pylori  infected from uninfected animals. Our model is characterized by a sensitivity of 97% and a specificity of 100% ( Table 5 ).\nRegardless of this model, we also showed that all the wavenumbers considered in the analysis IR spectra of the analyzed sera included in  Table 3  were also useful for differentiation of serum samples. The average values of χ2-square test or  p -value for the wavenumbers differ significantly ( p   <  0.0001,  t -test) between  H. pylori  infected and noninfected animas. According to the Shapiro–Wilk test, the distribution of the wavenumber values in both groups was normal ( p   =  0.59 for  H. pylori  infected animals and  p   =  0.22 for uninfected animals).\nH. pylori  infected animals were distinguished from the control group by the following rule: if the value of the first-order derivative of all wavenumbers in the IR spectra of  H. pylori  infected animals vs. uninfected animals, included in  Table 4 , is <0.01, then the animals will be treated as  H. pylori  infected.\n\nThe gold standard for detection of  H. pylori  infection in humans consists of an invasive endoscopic technique, during which gastric tissue samples are taken for further examination by a rapid urease test (RUT), histological staining for the presence of  Helicobacter -like organisms (HLO) and inflammatory cells and/or culture [ 62 , 63 ]. Among non-invasive methods, the  13 C urea breath test (UBT) is generally accepted as sufficiently specific and sensitive for primary diagnosis and confirming the effectiveness of eradication therapy [ 64 , 65 ]. Many variants of ELISA test are used for detection of specific anti- H. pylori  antibodies in serum samples. Detection of specific  H. pylori  DNA sequence in the gastric tissue and  H. pylori  antigens in stool samples was considered as a diagnostic tool [ 63 ]. However, a comprehensive method is needed to quantify selected markers to track the systemic effects of the infection that result from  H. pylori  colonization of gastric mucosa. Potentially, the determination of such markers may be helpful in predicting the course of infection and its consequences. It may also contribute to the confirmation of the relationship between local  H. pylori  infection and the development of extragastric diseases, including coronary heart disease, anemia, and type 2 diabetes [ 66 , 67 ].\nIn our study we used IR spectroscopy, which has been adopted for the characterization of biological materials, like tissue sections, cytological and histological specimens or biofluids, particularly serum samples. Using this technique to analyze serum samples, Ghimire et al. [ 68 ] identified mouse lymphoma and melanoma, while Gajjaee et al. [ 28 ] showed spectral changes in blood samples from patients with ovarian cancer. IR spectroscopy was useful for the estimation of serum cortisol levels in athletes [ 69 ], whereas Lechowicz et al. [ 29 ] used FT-IR spectroscopy to enable the differentiation between sera from patients with rheumatoid arthritis (RA) and non-RA sera. Zhou et al. [ 70 ] showed spectral changes in the HL60 cell line during various stages of cell differentiation and apoptosis. Erukhimovitch et al. [ 27 ] analyzed human plasma by FT-IR to determine spectral parameters for monitoring patients with leukemia. Shen et al. [ 71 ] used FT-IR to develop a noninvasive method to measure glucose in whole blood samples from diabetic patients [ 71 ], whereas Sankari et al. [ 40 ] analyzed serum immunoglobulins (IgA, IgM, and IgG) in patients with myeloma. FT-IR can be potentially useful for monitoring chronic infections based on changes in the systemic concentration of selected exogenous and endogenous molecules. We showed that FT-IR may help to investigate quantitative changes of selected soluble biomarkers, correlated with  H. pylori  infection in children and presumable consequent delayed growth [ 35 ].\nPreviously we developed an experimental model of  H. pylori  infection in guinea pigs, which was followed by the inflammatory response in gastric tissue, similarly to humans infected with these bacteria [ 56 , 57 , 72 ]. This model allowed us to study the outcome of  H .  pylori -driven inflammatory and immune responses as well as induction of regeneration processes in the gastric tissue in response to this infection [ 72 , 73 ]. The effects of  H. pylori  infections can vary from gastritis to gastric ulcer, duodenal ulcer, and gastric cancer, but the reasons for this are unknown. Therefore, new diagnostic methods need to be developed to assess  H. pylori  infection and its potential consequences. Such possibilities are offered by the experimental model of  H. pylori  infection in guinea pigs.\nThe aim of this study was to combine HCA analysis to determine the fragments of IR spectra of guinea pig sera characteristic for  H. pylori  infection. For this purpose, we used sera of uninfected animals ( n  = 20) or animals experimentally infected with  H. pylori  ( n  = 40).  H. pylori  infection was confirmed in all guinea pigs that received  H. pylori,  as previously described [ 56 , 57 , 73 ]. Infrared spectra were measured using an FT-IR Spectrum 400 spectrometer in the range of wavenumber 3000–750 cm −1  and then subjected to mathematical pre-processing (calculation of first derivative).\nSera from  H. pylori  infected animals showed lower absorbance values for TNF, α2-globulins, and glucose, while higher absorbance values were detected for CRP, transferrin, as well as IgM, IgG1/IgG4 immunoglobulins, than in sera from  H. pylori  uninfected animals.\nThe CRP protein is a widely accepted systemic marker of the inflammatory response, in  H. pylori  infection [ 74 , 75 ]. We showed elevated levels of CRP in serum samples of  H. pylori  infected compared to uninfected animals, which was consistent with the development of an inflammatory response in the gastric tissue colonized with these bacteria. By comparison, levels of TNF in serum from  H. pylori  infected animals were significantly lower than in uninfected animals, which remains in line with a lower maximum absorption of the characteristic wavenumber of this cytokine.\nThis remains in line with a lower maximum absorption of the characteristic wavenumber for this cytokine. It is worth mentioning that in humans,  H. pylori  status was found to correlate with serum levels of CRP. In turn, increased levels of CRP were associated with elevated levels of total cholesterol and triglycerides, which are the risk factors for CHD [ 76 ]. The production of CRP in the liver is regulated by proinflammatory TNF. However, in this study we showed using FT-IR that TNF was lower in serum samples of  H. pylori  infected animals than uninfected animals. It could be due to neutralization of this cytokine by soluble TNF receptor. As shown by Shibata et al., during  H. pylori  infection the gastric epithelial cells undergo TNF-driven apoptosis and due to this the host responds by elevated production of soluble TNF receptor I in order to neutralize TNF and prevent its deleterious effects [ 77 ]. It is possible that in this study the lower level of TNF in the sera of guinea pigs infected with  H. pylori  as compared to uninfected animals may be due to neutralization of this cytokine by soluble TNF receptor. Previously we showed that  H. pylori  infection in guinea pigs was related to increased oxidative stress and apoptosis, and that the serum level of TNF in  H. pylori  infected animals determined by ELISA was lower than in uninfected animals [ 56 , 57 ].\nThe absorbance values of glucose and α2-globulins were also lower in  H. pylori  infected animals than in uninfected animals. The relationship between  H. pylori  infection and lower fasting blood glucose level has been previously reported [ 78 , 79 , 80 ]. Higher absorbance values shown for IgM, IgG1, IgG4, and transferrin when testing sera from  H. pylori  infected guinea pigs are consistent with higher levels of anti- H. pylori  antibodies in this group, as evaluated by ELISA. In  H. pylori  infected patients, a higher systemic level of transferrin has been observed in conjunction with iron deficiency [ 81 , 82 ].\nThe study designed a predictive mathematical model: k-NN algorithm to distinguish  H. pylori  infected animals from uninfected animals with a sensitivity of 97% and a specificity of 100%. The proposed model can be a useful tool to detect and track the course of  H. pylori  infection, considering the serological and inflammatory parameters. Only one serum from the group of  H. pylori  infected animals was misclassified based on IR spectra. This could have been caused by the water in the samples which may affect the IR spectrum in the entire spectral range. Therefore, further studies are necessary to develop standardized protocols for the samples used for analysis.\n\nIn vivo experiments were developed according to the EU directive (Directive 2010/63/EU of the European Parliament and of the Council of 22 September 2010 on the protection of animals used for scientific purposes (Dz.U. L 276 z 20.10.2010, s. 33–79), and were approved by the resolution of the Local Ethics Committee (LKE9) for Animal Experiments of the Medical University of Lodz, Poland, which was established by the Ministry of Science and Higher Education in Poland (Decision 58/ŁB45/2016).\nThree-month-old, male Himalayan guinea pigs (400–600 g), free of pathogens, were housed in the Animal House at the Faculty of Biology and Environmental Protection, University of Lodz (Poland), kept in cages with free access to drinking water and fed with standard chow. The animals were inoculated  per os  with bacterial suspension of the reference  H. pylori  strain CCUG 17874 (Culture Collection University of Gothenburg, Sweden), three times in two day intervals as previously described [ 56 , 57 ]. Then, 28 days after the last  H. pylori  inoculation, the animals were euthanized according to a protocol approved by an ethics committee, and the gastric tissue was collected for analyses, whereas blood samples were processed to obtain serum and then stored at −80 °C.  H. pylori  infection was confirmed by microscopic imaging of HLOs, scoring tissue inflammation in thin layer preparations stained by routine histological staining, polymerase chain reaction (PCR) to detect  H. pylori cagA  and  ureC  gene sequences, as previously described [ 56 ].\nThe laboratory ELISA for anti- H. pylori  IgM and IgG antibodies against the  H. pylori  antigenic complex, glycine extract (GE), which was obtained by extraction with glycine acid buffer of surface antigens from the reference  H. pylori  17874 strain CCUG, producing CagA protein and VacA, as previously described [ 83 ]. Major proteins in GE recognized by the reference sera from  H. pylori  infected patients were as follows: 120 kDa (CagA), 87 kDa (VacA), 66 kDa (UreB, subunit B of urease), 60 kDa (heat shock protein—Hsp), 29 kDa (UreA), between 66–22 kDa [ 65 ]. The GE protein concentration was 600 μg/mL (NanoDrop 2000c Spectrophotometer, ThermoScientific, Wlatman, MA, USA) and <0.001 EU/mL of LPS, as shown by the chromogenic Limulus amebocyte lysate test (Lonza, Braine-Alleud, Belgium) as previously described [ 83 , 84 ].\nIn total, 60 animals were used in the study: 20 uninfected (control) and 40 infected with  H. pylori . The levels of anti- H. pylori  GE IgM and IgG were showed as mean ± standard deviation (SD) for each group.\nTotal CPR or TNF concentrations in guinea pig serum samples were determined by ELISA with a sensitivity of 4 pg/mL for CRP (Cloud-Clone, Katy, TX, USA) and 8 pg/mL for TNF (ThermoScientific, Waltham, MA, USA), as recommended by the manufacturer. The absorbance was estimated using a Victor2 reader at a wavenumber of 450 nm. For each group the median values were calculated.\nIR spectra of animal sera were evaluated using the ATR-FTIR technique using the FT-IR/FT- near-infrared spectroscopy (NIR) Spectrum 400 spectroscope (Perkin Elmer, Waltham, MA, USA). The spectrometer was equipped with an ATR single reflection diamond crystal. Serum samples were stored at −80 °C until the measurement was made. Sera samples were thawed at 20 °C using a CH-100 (BIOSAN, Riga, Latvia) thermoblock and then shaken for 30 s using LabDancer vario (IKA, Staufen im Breisgau, Germany). Measurements were made at 20 °C with constant air humidity. Before each measurement, the ATR crystal was thoroughly cleaned with 95% alcohol and a baseline measurement was made. One microliter of serum was added to the spectroscope crystal by using a disposable sterile pipette tip and was left to evaporate the water. The amount of evaporated water was estimated based on the intensity of a band at 3400–3200 cm −1 . It was assumed that the water was evaporated from the sample when the band intensity at wavenumber 3300 cm −1  decreased to about 40% of the original value. In such conditions, the process of water evaporation lasted about 5 min. The high water content in the sample suppresses the intensity of the bands throughout the IR spectrum. After removing most of the water from the sample, the spectrum of the sediment remaining on the ATR crystal was measured. IR spectra were measured in the wavenumber range of 4000–650 cm −1  with a resolution of 1 cm −1  and then preprocessed in two steps: (a) calculation of the first derivative by a five-point stencil, (b) normalization to this range {0.1} [ 29 ].\nThe first derivatives were used for mathematical analysis. To develop the dendrograms, we used the Manhattan metric and Ward’s method. Serum samples were collected from each guinea pig in four test tubes. One IR spectrum for each serum sample was collected as a result of 25 running scans. A total number of 240 IR spectra were measured ( Figure 4 .)\nFor developing a prediction model we used k-NN algorithm. The k-NN classifier is one of the well-known and simple classification algorithms. It was first introduced by Fix and Hodges as a non-parametric algorithm, i.e., it does not make any assumptions on the input data distribution; thus, it is widely used in different applications [ 85 ]. In the k-NN classifier, an unknown sample is classified based on the similarity to the known, trained, or labeled samples by computing the distances between the unknown sample and all labeled samples. k-nearest samples are then selected as the basis for classification; and the unknown sample is assigned to the class which has the most samples among the k-nearest samples. For that, the k-NN classifier algorithm depends on: (a) integer k (number of neighbors)—changing the values of k parameter may change the classification result; (b) a set of labeled training data—thus, adding or removing any samples to the training samples will affect the final decision of k-NN classifier; and (c) a distance metric. In k-NN, the Euclidean distance is often used as the distance metric to measure the distance between two samples. k-NN classifier is analytically traceable and simple to implement, but one of the main problems of the k-NN algorithm is that it needs all the training samples to be in memory at run-time; for this reason, it is called memory-based classification [ 85 , 86 ].\nThe set of 240 spectra was randomly divided into two subsets: learning subset (180 cases) and validation subset (60 cases). The k-NN model is based on the spectral windows W1–W5. Calculations were performed using Statistica 12. The quality of the model was evaluated on the basis of quality indicators presented in  Table 1 .\nSelected k-NN were validated in terms of: false negative (FN), true positive (TP), informedness, markedness, condition positive (P)—number of real positive cases in the data, condition negative (N)—number of real negative cases in the data. Also,\n sensitivity as well as true positive rate (TPR): TTR =  TP P  =  TP TP + FN  = 1 − FNR specificity as well as true negative rate (TNR):\nTNR =  TN N  =  TN TN + FP  = 1 − FPR miss rate as well as false negative rate (FNR): FNR =  FN P  =  FN FN + TP  = 1 − TPR false positive rate (FPR): FPR =  FP N  =  FP FP + TN = 1 − TNR  where FP is a number of false positives, TN is the number of true negatives and N = FP + TN precision (PPV): PPV =  TP TP + FP false omission rate (FOR): FOR =  FN FN + TN  = 1 − NPV negative predictive value (NPV): NPV =  TN TN + FN  = 1 − FOR false discovery rate (FDR): FDR =  FP FP + TP  = 1 − PPV positive likelihood ratio (PLR): PLR =  TPR 100 − TNR negative likelihood ratio (NLR): NLR =  100 − TPR TNR accuracy: accuracy =  TN + TP TN + FP + TP + FN\nsensitivity as well as true positive rate (TPR): TTR =  TP P  =  TP TP + FN  = 1 − FNR\nspecificity as well as true negative rate (TNR):\nTNR =  TN N  =  TN TN + FP  = 1 − FPR\nmiss rate as well as false negative rate (FNR): FNR =  FN P  =  FN FN + TP  = 1 − TPR\nfalse positive rate (FPR): FPR =  FP N  =  FP FP + TN = 1 − TNR\nwhere FP is a number of false positives, TN is the number of true negatives and N = FP + TN\nprecision (PPV): PPV =  TP TP + FP\nfalse omission rate (FOR): FOR =  FN FN + TN  = 1 − NPV\nnegative predictive value (NPV): NPV =  TN TN + FN  = 1 − FOR\nfalse discovery rate (FDR): FDR =  FP FP + TP  = 1 − PPV\npositive likelihood ratio (PLR): PLR =  TPR 100 − TNR\nnegative likelihood ratio (NLR): NLR =  100 − TPR TNR\naccuracy: accuracy =  TN + TP TN + FP + TP + FN\nHierarchical cluster analysis can be conceptualized as being agglomerative or divisive. Agglomerative hierarchical clustering separates each case into its own individual cluster in the first step so that the initial number of clusters equals the total number of cases [ 87 ]. At successive steps, similar cases—or clusters—are merged together until every case is grouped into one single cluster. Divisive hierarchical clustering works in the reverse manner with every case starting in one large cluster and gradually being separated into groups of clusters until each case is in an individual cluster [ 87 ].\nHCA, an unsupervised method consisting of grouping the spectra with the same degree of similarity using the software Statistica 12, was also conducted to differentiate the serum spectra for noninfected or  H. pylori  infected animals. The chi-square statistical test was used to check the part of the IR spectra which correlated with the examined feature. Best predictors for  H. pylori  infection were used in HCA. Manhattan metric and Euclidean distances (inter-spectral distance) between all data were calculated and Ward’s algorithm was used to construct the dendrogram displaying the grouping of spectra into clusters based on heterogeneity values. Euclidean distances allow for the distance between two cases to be calculated across all variables and reflected in a single distance value. At each step in the procedure, the squared Euclidean distance between all pairs of cases and clusters is calculated and shown in a proximity matrix. At each step, the pair of cases or clusters with the smallest squared Euclidean distance will be joined with one another. Ward’s minimum variance method calculates the distance between cluster members and the centroid. The centroid of a cluster is defined as the point at which the sum of squared Euclidean distances between the point itself and each other point in the cluster is minimized [ 88 ].\n\nBased on the chi-square test, 10 wavenumbers of IR spectra correlating with  H. pylori  infection were selected for HCA analysis by the use of the k-NN model. The sensitivity and specificity of this model were 97% and 100%, respectively, whereas the accuracy reached 98%. This study shows that the combination of infrared spectroscopy and HCA methods may potentially be useful for the differentiation of serum samples from guinea pigs uninfected or infected with  H. pylori  on the basis of TNF, CRP, transferrin, glucose, α2-globulins, and IgM, IgG1/4. Further study on a larger number of samples will allow standardization of this method for the analysis of guinea pig sera regarding  H. pylori  exposure in conjunction with the development of inflammatory response and antibody production as well as some metabolic markers such as glucose. The results obtained from the experimental model of  H. pylori  infection in guinea pigs will be helpful in adjusting this methodology to study the systemic effects of  H. pylori  local gastric infection in humans on the basis of different levels of selected soluble determinants in serum samples.","source_license":"CC-BY-4.0","license_restricted":false}