Attenuated Total Reflectance Fourier Transform Infrared Spectroscopy (FTIR) and Artificial Neural Networks Applied to Investigate Quantitative Changes of Selected Soluble Biomarkers, Correlated with H. pylori Infection in Children and Presumable Consequent Delayed Growth

Helicobacter pylori infections causing gastroduodenal disorders are a common medical problem. The aim of this study was to determine the specific motives of infrared spectroscopy (IR) spectra of sera from H. pylori-infected and uninfected children applied to investigate quantitatively-selected soluble biomarkers correlated with H. pylori infection in children and presumable consequent delayed growth. Sera from 41 children infected with H. pylori (Hp(+)) and 43 uninfected (Hp(−)) under the care of the Polish Mother’s Hospital in Lodz, Poland, were analyzed. The H. pylori status was confirmed by gastroscopy, 13C urea breath testing, and anti-H. pylori IgG antibodies. Infrared spectra were measured using an FTIR/FT-NIR Spectrum 400 spectrometer (PerkinElmer). The IR spectrum was measured in the wavenumber range 3000–750 cm−1 and subjected to mathematical calculation of the first derivative. Based on the chi-square test, 10 wavenumbers of spectra correlating with H. pylori infection were selected for use in designing an artificial neural network. Ten parts of the IR spectra correlating with H. pylori infection were identified in the W2 and W3 windows associated mainly with proteins and the W4 window related to nucleic acids and hydrocarbons. Artificial neural networks for H. pylori infection were developed based on chemometric data. By mathematical modeling, children were classified towards H. pylori infection in conjunction with elevated levels of selected biomarkers in serum potentially related to growth retardation. The study concludes that IR spectroscopy and artificial neural networks may help to confirm H. pylori-driven growth disorders in children.


Introduction
Helicobacter pylori is a Gram-negative pathogenic bacterium that specifically colonizes gastric mucosa in humans (average frequency of infection 50%) and was described by Warren and Marshall in The artificial neural network learning process is based on presenting the ANN with the data included in the training set, and then calculating the errors made by the network. Then, the synaptic weights are corrected and the whole process starts all over again. After reaching the required level of the number of correct classifications, the sensitivity and specificity of the ANN is verified by presenting its cases from the validation set [36].
J. Clin. Med. 2020, 9, x FOR PEER REVIEW 3 of 14 by the network. Then, the synaptic weights are corrected and the whole process starts all over again. After reaching the required level of the number of correct classifications, the sensitivity and specificity of the ANN is verified by presenting its cases from the validation set [36].

Patients and Controls
Approval for the study was obtained from the Bioethical Committee in the Polish Mother's Memorial Hospital-Research Institute (PMMH-RI) in Lodz (RNN/134/13/KE/2-13). Healthy children-the control group (43 individuals of both sexes, mean age 11.5 ± 2.8)-and children with H. pylori-related gastritis (41 individuals of both sexes, mean age 12.5 ± 3.3) were under the care of the PMMH-RI in Lodz. Informed consent was obtained for experimentation. The right to privacy of participants was guaranteed. Blood samples were collected in a fasting state, on admission, before medical and pharmacological intervention. Serum samples were obtained within 1 h, by 30 min incubation at room temperature and subsequent 30 min incubation at 4 °C followed by centrifugation (2000× g, 10 min, 4 °C). The sera were aliquoted and stored at −80 °C for further analysis. Samples were thawed directly before being used in experiments.

Patients and Controls
Approval for the study was obtained from the Bioethical Committee in the Polish Mother's Memorial Hospital-Research Institute (PMMH-RI) in Lodz (RNN/134/13/KE/2-13). Healthy children-the control group (43 individuals of both sexes, mean age 11.5 ± 2.8)-and children with H. pylori-related gastritis (41 individuals of both sexes, mean age 12.5 ± 3.3) were under the care of the PMMH-RI in Lodz. Informed consent was obtained for experimentation. The right to privacy of participants was guaranteed. Blood samples were collected in a fasting state, on admission, before medical and pharmacological intervention. Serum samples were obtained within 1 h, by 30 min incubation at room temperature and subsequent 30 min incubation at 4 • C followed by centrifugation (2000× g, 10 min, 4 • C). The sera were aliquoted and stored at −80 • C for further analysis. Samples were thawed directly before being used in experiments.

Diagnosis of H. pylori Infection
The H. pylori status in patients and healthy donors was estimated by the 13 C urea breath test ( 13 C UBT) [37], immunoblotting for anti-H. pylori IgG (Milenia ® Blot H. pylori, DPC Biermann, GmbH, Bad Nanheim, Germany), and the enzyme-linked immunosorbent assay (ELISA) for IgG antibodies against the H. pylori antigenic complex glycine extract (GE), which was obtained by extraction with a glycine acid buffer of surface antigens from the reference H. pylori strain CCUG (Culture Collection University of Gothenburg, Sweden) 17,874, producing cytotoxin associated gene A (CagA) protein and vacuolating cytotoxin (VacA), as previously described [38]. 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), 60 kDa (Hsp), 29 kDa (UreA), between 66-22 kDa [39]. The GE protein concentration was 600 µg/mL (NanoDrop 2000c Spectrophotometer, ThermoScientific, Waltman, WY, USA) and <0.001 EU/mL of LPS, as shown by the chromogenic Limulus amebocyte lysate test (Lonza, Braine-Alleud, Belgium). The serum samples were also examined for the presence of anti-CagA IgG using recombinant CagA (rCagA, courtesy of Dr. Antonello Covacci, IRIS, Siena, Italy) as previously described [39]. Furthermore, 13 C UBT-positive children with dyspeptic symptoms underwent gastroscopy and routine histological examination of gastric tissue specimens towards Helicobacter-like organisms and inflammatory response. Short status in children was measured according to population standards [40]. A height standard deviation score (H SSD) of over −2.0 qualified children for the short status group. In the group of uninfected children there were no children with short stature. By comparison, in the group of H. pylori-infected children 36 showed a growth retardation.

The Measurement of Infrared Spectra and Their Processing
IR spectra of human sera were measured by using the attenuated total reflectance (ATR) spectroscopy technique. Measurements were performed at 20 • C at constant air humidity. Serum samples were thawed at room temperature and then shaken for 30 s. One microliter of serum was added to the spectroscope crystal and left for 5 min to evaporation. The IR spectra were measured in the range of wavenumbers 4000-650 cm −1 with a resolution of 1 cm −1 and then pre-processed in two steps: (a) calculation of the first derivative by five-point stencil, (b) normalization to the range {0, 1} [41].

Mathematical Model Development for Patient Differentiation
The chi-square statistical test was used to check the part of the IR spectra which correlated with the examined feature. Next, a number of mathematical models based on multilayer perceptrons (a type of artificial neural network (ANN)) were built. The individual artificial neural networks differed from each other in complexity-the number of input neurons (from 1 to 10) and hidden neurons (from 1 to 5), the activation functions of neurons, and the function of error. Two output neurons were selected for analysis; these neurons corresponded to two opposite states concerning H. pylori infection: Infected or uninfected. The input data for the ANNs were the absorbance values of the wavenumbers, which strongly correlated with the examined feature. The total set of spectra was randomly divided into two separate subsets: Learning and validating. The quality of the network was evaluated on the basis of the number of correct classifications in the validation set. The calculations were carried out using Statistica 12 software (StatSoft, Round Rock, TX, USA). Selected ANNs were validated in terms of sensitivity, miss rate, specificity, false positive rate, precision, false discovery rate, false omission rate, negative predictive value, positive likelihood ratio, negative likelihood ratio, accuracy, informedness, and markedness.

Statistical Analysis
Results are expressed as the median values with a range. The differences between tested variables were assessed using Statistica 12 software with a nonparametric Mann-Whitney U test. The results were considered statistically significant when p < 0.05.

H. pylori Serological Status in Children
Differences between study groups in IR spectroscopy were related to H. pylori infection and anti-H. pylori IgG. The panels of sera from healthy children (43 individuals) and from children with H. pylori-induced gastritis confirmed by histological examination of gastric tissue specimens (41 individuals) were screened ( Figure 2). The serum samples from healthy children did not contain anti-GE IgG and were free from anti-CagA IgG (
Vibration band assignment was carried out on the infrared spectra of sera by comparing the position, relative intensity, and shape of the bands with the corresponding features of bands of related molecules. Infrared spectra of serum samples from healthy children vs. children infected with H. pylori are shown in Figure 3. Absorption bands for proteins, lipids, and carbohydrates were taken into account during analysis. Particularly specific molecules, which were identified in the composition of IR spectra of sera, such as: α-tocopherol (window B1), β-carotene (window B2), folic acid (window B3), ghrelin (window B4), leptin (window B5), lycopene (window B6), lutein (window B7), vitamin B6 (window B8), vitamin B12 (window B9), and vitamin C (window B10) allowed the differentiation between H. pylori-infected and uninfected individuals ( Figure 3A-C, Table 1).
The ID column shows representative characteristic parts of infrared spectra of children's sera and molecules differentiating H. pylori-infected vs. H. pylori-uninfected children.
A vibration band assignment was done to analyze the frequencies of chemical groups present in the sample. As shown in Table 1, in the spectral region (B1) corresponding to α-tocopherol the prominent absorption peak 1472 cm −1 is due to the CH3/CH2 asymmetrical scissoring (peptides and proteins), while 1078 cm −1 corresponds to C-O symmetric stretching of glucose region (carbohydrates). The absorption band B2 (component group: amide I and amino acid) is characteristic for β-carotene absorption vibration peaks: 1650 cm −1 C=O symmetric stretching, 1457 cm −1 asymmetric C-H scissoring of CH3, 1384 cm −1 CH3 deformation, 1324 cm −1 O-H bending, and 1096 cm −1 C-C-C bending vibrations ( Figure 3, Table 1). The spectral region B3, which is typical for peptides and proteins (component group amide I, amino acid), corresponds to a vibration band characteristic for folic acid. Typical parameters of this region are as follows: N-H (amino acid) and CO2 asymmetric stretching (3050 cm −1 , 1485-1519 cm −1 ), NH2 (amide I) scissoring (1604-1619 cm −1 , 1650 cm −1 ), and C=O symmetric stretching (1693 cm −1 ) vibrations. The prominent absorption peak identified for ghrelin was 1640 cm −1 (B4) due to the NH2 scissoring (amide I) of proteins whereas the absorption peaks for leptin were identified at wavenumbers 1740 cm −1 and 1550 cm −1 (B5). The C=O stretching and N-H bending vibration strongly coupled to C-N stretching vibration of this protein. The absorption bands for lycopene (B6) at 3100 cm −1 correspond to N-H asymmetric stretching (amino acid), whereas wavenumber 2851 cm −1 is attributed to lipids C-H symmetric stretching of CH2 group, 1640 cm −1 corresponds to NH2 scissoring, 1450-1400 cm −1 to CH3 asymmetric deformation, and 1375 cm −1 to CH2
Vibration band assignment was carried out on the infrared spectra of sera by comparing the position, relative intensity, and shape of the bands with the corresponding features of bands of related molecules. Infrared spectra of serum samples from healthy children vs. children infected with H. pylori are shown in Figure 3. Absorption bands for proteins, lipids, and carbohydrates were taken into account during analysis. Particularly specific molecules, which were identified in the composition of IR spectra of sera, such as: α-tocopherol (window B1), β-carotene (window B2), folic acid (window B3), ghrelin (window B4), leptin (window B5), lycopene (window B6), lutein (window B7), vitamin B6 (window B8), vitamin B12 (window B9), and vitamin C (window B10) allowed the differentiation between H. pylori-infected and uninfected individuals ( Figure 3A-C, Table 1).
The ID column shows representative characteristic parts of infrared spectra of children's sera and molecules differentiating H. pylori-infected vs. H. pylori-uninfected children.
A vibration band assignment was done to analyze the frequencies of chemical groups present in the sample. As shown in Table 1, in the spectral region (B1) corresponding to α-tocopherol the prominent absorption peak 1472 cm −1 is due to the CH 3 /CH 2 asymmetrical scissoring (peptides and proteins), while 1078 cm −1 corresponds to C-O symmetric stretching of glucose region (carbohydrates). The absorption band B2 (component group: amide I and amino acid) is characteristic for β-carotene absorption vibration peaks: 1650 cm −1 C=O symmetric stretching, 1457 cm −1 asymmetric C-H scissoring of CH 3 , 1384 cm −1 CH 3 deformation, 1324 cm −1 O-H bending, and 1096 cm −1 C-C-C bending vibrations ( Figure 3, Table 1). The spectral region B3, which is typical for peptides and proteins (component group amide I, amino acid), corresponds to a vibration band characteristic for folic acid. Typical parameters of this region are as follows: N-H (amino acid) and CO 2 asymmetric stretching (3050 cm −1 , 1485-1519 cm −1 ), NH 2 (amide I) scissoring (1604-1619 cm −1 , 1650 cm −1 ), and C=O symmetric stretching (1693 cm −1 ) vibrations. The prominent absorption peak identified for ghrelin was 1640 cm −1 (B4) due to the NH 2 scissoring (amide I) of proteins whereas the absorption peaks for leptin were identified at wavenumbers 1740 cm −1 and 1550 cm −1 (B5). The C=O stretching and N-H bending vibration strongly coupled to C-N stretching vibration of this protein. The absorption bands for lycopene (B6) at 3100 cm −1 correspond to N-H asymmetric stretching (amino acid), whereas wavenumber 2851 cm −1 is attributed to lipids C-H symmetric stretching of CH 2 group, 1640 cm −1 corresponds to NH 2 scissoring, 1450-1400 cm −1 to CH 3 asymmetric deformation, and 1375 cm −1 to CH 2 wagging vibrations of component group amide I. The 1517-1500 cm −1 (B7) is predominantly linked to CO 2 asymmetric stretching vibrations of peptides and proteins, and characteristic for lutein. The spectral region 1280-1315 cm −1 (B8, vitamin B6) is predominantly linked to C-H/N-H deformation vibration models methyl groups, vibrations of phospholipids of proteins, phosphate-carrying compounds, and fatty acids ( Figure 3A,B, Table 1). The asymmetric and symmetric stretching CH 3 asymmetric stretching of CH 3 group and C=O band vibrations were found at 3120 cm −1 and 1670-1665 cm −1 (B9, Vitamin B12). The characteristic vibration bends were found for the vitamin C (B10): 1760 cm −1 C=O stretching, 1634 cm −1 NH 2 scissoring (amide I), and 1322 cm −1 C-H/N-H deformation vibration models methyl groups (amide III) ( Figure 3A-C wagging vibrations of component group amide I. The 1517-1500 cm −1 (B7) is predominantly linked to CO2 asymmetric stretching vibrations of peptides and proteins, and characteristic for lutein. The spectral region 1280-1315 cm −1 (B8, vitamin B6) is predominantly linked to C-H/N-H deformation vibration models methyl groups, vibrations of phospholipids of proteins, phosphate-carrying compounds, and fatty acids ( Figure 3A,B, Table 1). The asymmetric and symmetric stretching CH3 asymmetric stretching of CH3 group and C=O band vibrations were found at 3120 cm −1 and 1670-1665 cm −1 (B9, Vitamin B12). The characteristic vibration bends were found for the vitamin C (B10): 1760 cm −1 C=O stretching, 1634 cm −1 NH2 scissoring (amide I), and 1322 cm −1 C-H/N-H deformation vibration models methyl groups (amide III) ( Figure 3A-C, Table 1).

Wavenumbers Correlating with H. pylori Infection and Mathematical Models Identifying Sera of Infected Individuals
By using the chi-square statistical test, it was possible to extract parts of IR spectra correlating with H. pylori infection in children. For designing artificial neural networks, 10 wavenumbers which strongly correlated with this infection were considered (Table 2). Six wavenumbers were presented in the W4 window, which were associated with vibrations derived from hydrocarbons and nucleic acids. Three wavenumbers correlated with H. pylori infection in the W3 window. This window is associated with different chemical bonds (the so-called mixed region); however, proteins and phosphorus-containing compounds are key for this window. The wavenumber 1557 cm −1 , localized within the W2 window, was strongly associated with H. pylori infection. According to the literature, the W2 window illustrates proteins and peptides. In H. pylori-infected individuals it may correspond to antibody production and/or an increased production of inflammatory proteins.  Satisfactory results of IR spectra analysis were obtained using ANNs. Proposed models are characterized by 83-90% accuracy. The topology of the best ANNs is presented in Table 3. However, by increasing the size of the studied groups, the differentiation capacity of the ANNs could be increased. The best neural network contained 10 entry neurons, 3 hidden neurons, and 2 output neurons (10-3-2 topology). The quality of the perceptron was assessed on the basis of 13 determinants included in Table 4. This network reached 95% sensitivity and 86% specificity, indicating the certain potential of chemometric methods in differentiating H. pylori-infected from H. pylori-uninfected children on the basis of IR spectrum analysis.

Discussion
IR spectroscopy has become a suitable method for the characterization of biological materials, like tissue sections, cytological and histological specimens, or biofluids. Liu et al. identified leukemia cells CLL through the level of DNA/RNA and lipids [51]. Zhou et al. showed spectral changes in the HL60 cell line during differentiation and apoptosis [52]. Erukhimovitch et al. applied FTIR to the analysis of human plasma samples in order to detect spectral parameters that might serve as biomarkers for the monitoring and identification of patients with leukemia [53]. Shen et al. developed a noninvasive method for the measurement of glucose concentration in whole blood samples of diabetes patients [54]. Sankari et al. analyzed serum immunoglobulins IgA, IgM, and IgG in patients with myeloma using FTIR [55], whereas Lechowicz et al. used FTIR spectroscopy to differentiate rheumatoid arthritis (RA) from non-RA sera [41]. FTIR can potentially be useful for monitoring chronic infections and related consequences on the basis of systemic changes in concentration of exo-and endogenous molecules.
H. pylori induces gastritis, gastric or duodenal ulcers, and even gastric cancers in infected individuals. About 20% of infected people develop clinical symptoms. The reason for such different responses to H. pylori is not well understood. In children H. pylori infection can cause developmental disorders. It can affect the balance of nutritional status in children and correlates with malnutrition and growth retardation [20,[56][57][58][59]. Symptomatic infections are correlated with deficiencies of macro-and micronutrients (e.g., iron, zinc, selenium, vitamin C, vitamin A, α-tocopherol, vitamin B12, and folic acid) and essential minerals [60][61][62][63][64][65][66]. It is therefore advisable to follow the changes in the concentration of these molecular determinants in connection with H. pylori infection. Fourier transform infrared spectroscopy in conjunction with artificial neural networks enable the analysis of molecular parameters in various biological materials, including blood serum. Taking this into account, the purpose of this study was to use this specialized tool to determine the specific motives of IR spectra of serum samples from H. pylori-infected children (n = 41), including those with growth disorders vs. uninfected children (n = 43). Infrared spectra were measured using an FTIR/FT-NIR Spectrum 400 spectrometer. The IR spectrum was measured in the wavenumber range of 3000-750 cm −1 and then subjected to mathematical pre-processing (calculation of first derivative).
For sera obtained from H. pylori-positive donors, lower absorbance values were observed for individual molecules than when using sera from H. pylori-negative donors. These differences concerned the following components: vitamin A, vitamin B6, vitamin B12, vitamin C, α-tocopherol, folic acid, β-carotene, and lutein as well as the hormone peptides ghrelin and leptin. Lower levels of these molecules in the blood, which were estimated by biochemical or serological methods, were associated with H. pylori infection [59][60][61][62][63][64][65][66]. The low levels of vitamin B12 in H. pylori-infected patients were linked with an impairment of vitamin B12 absorption from food and the development of pernicious anemia [61,62,67]. The lower concentration of β-carotene (pro-vitamin A) was shown in patients with gastric atrophy and intestinal metaplasia due to H. pylori infection [59,63]. Similarly, vitamin C concentration in plasma was 20% lower in H. pylori-infected subjects than it was in negative controls, even after correction for confounding factors such as smoking and dietary behavior. H. pylori infection may cause an irreversible inactivation of ingested vitamin C in the intestinal lumen prior to its absorption [68,69].
Furthermore, H. pylori infection is related to a lower secretion of appetite-regulating peptides and hormones, including ghrelin, which is involved in the hypothalamic regulation of metabolism and energy balance, as well as leptin and gastrin [70][71][72][73][74][75].
Concerning the clinical implications of this study, it seems that lowering all the molecular markers shown using the FTIR technique in conjunction with artificial neural networks may have a negative impact on the health and development of children; it may, however, be helpful in detecting growth delay driven by H. pylori infection in children. This combined analysis may also help to predict, on the basis of soluble systemic markers, the potential deleterious effects of H. pylori infection on the level of gastric barrier. Further research is needed to standardize this technique for the proposed application. It is expected that it will be possible to propose additional molecular markers to differentiate H. pylori-infected patients from uninfected ones in terms of the consequences of this infection.

Conclusions
Based on the chi-square test, 10 wavenumbers of IR spectra correlating with H. pylori infection were selected for ANN design. The sensitivity and specificity of this model were 95% and 86%, respectively, whereas accuracy reached 90%. This study shows that the combination of infrared spectroscopy and ANN techniques may be useful for differentiating serum samples from children exposed or unexposed to H. pylori. Particularly, components selected for analysis including vitamin A, vitamin B6, vitamin B12, vitamin C, α-tocopherol, folic acid, β-carotene, and lutein, as well as the hormone peptides ghrelin and leptin can help to diagnose children with growth delay due to H. pylori infection. Further study on the larger number of samples will make it possible to standardize this method for investigating molecular level changes correlated with H. pylori infection in children and presumable consequent delayed growth. In future research the relationship between the IR parameters determined in this study with the biomarkers of local and systemic inflammatory response in H. pylori-infected children should also be considered. Introduction of these additional parameters will enable a more comprehensive analysis of the health status of H. pylori-infected patients and could possibly enable the application of specific treatment.