Clinica Chimica Acta 526 (2022) 6–13 Contents lists available at ScienceDirect Clinica Chimica Acta journal homepage: www.elsevier.com/locate/cca Breath analysis using electronic nose and gas chromatography-mass spectrometry: A pilot study on bronchial infections in bronchiectasis Luciana Fontes de Oliveira a, Celia Mallafré-Muro a, b, Jordi Giner c, Lidia Perea d, Oriol Sibila d, Antonio Pardo b, Santiago Marco a, b, * a Signal and Information Processing for Sensing Systems, Institute for Bioengineering of Catalonia (IBEC), The Barcelona Institute of Science and Technology, Baldiri Reixac 10-12, 08028, Barcelona, Spain b Department of Electronics and Biomedical Engineering, University of Barcelona, Marti I Franqués 1, 08028 Barcelona, Spain c Department of Pneumology and Allergy. Hospital de la Sta. Creu I Sant Pau. Barcelona, Spain d Respiratory Department, Hospital Clinic, IDIBAPS, Barcelona, Spain A R T I C L E I N F O A B S T R A C T Keywords: Background and aims: In this work, breath samples from clinically stable bronchiectasis patients with and without Breath analysis bronchial infections by Pseudomonas Aeruginosa- PA) were collected and chemically analysed to determine if they Bronchiectasis have clinical value in the monitoring of these patients. Signal processing Materials and methods: A cohort was recruited inviting bronchiectasis patients (25) and controls (9). Among the E-nose GC-MS former group, 12 members were suffering PA infection. Breath samples were collected in Tedlar bags and analyzed by e-nose and Gas Chromatography-Mass Spectrometry (GC-MS). The obtained data were analyzed by chemometric methods to determine their discriminant power in regards to their health condition. Results were evaluated with blind samples. Results: Breath analysis by electronic nose successfully separated the three groups with an overall classification rate of 84% for the three-class classification problem. The best discrimination was obtained between control and bronchiectasis with PA infection samples 100% (CI95%: 84–100%) on external validation and the results were confirmed by permutation tests. The discrimination analysis by GC-MS provided good results but did not reach proper statistical significance after a permutation test. Conclusions: Breath sample analysis by electronic nose followed by proper predictive models successfully differentiated between control, Bronchiectasis and Bronchiectasis PA samples. 1. Introduction and methods as randomization, restriction, or matching [19] can be used. Appropriate control of the clinical and instrumental confounding The potential advantages of breath analysis for volatolomics studies, factors on observational studies in breath analysis could improve and including the unlimited sample supply, the non-invasive way to collect decrease biased results [20–22]. samples, and the possibility deliver fast analysis results have been On the other hand, there are a variety of instrumental techniques for described in several previous works [1–7]. However, despite the abun­ breath analysis that differ on usability, cost, and retrieved chemical dant literature, the use of breath analysis for clinical applications is in its information: namely, GCxGC-MS [23], chemical sensor systems [24], infancy [8,9] and the lack of standardization on sample collection/ Proton Transfer Reaction-Mass Spectrometry (PTR-MS) [25], Selected analysis [10–13] and the complexity of data analysis step leave space to Ion Flow Tube-Mass Spectrometry (SIFT-MS) [26], Laser Spectroscopy further developments [14–18]. [27] or Gas Chromatography -Ion Mobility Spectrometry (GC-IMS) [28]. In breath analysis, there are different types of confounding factors Among the several analytical platforms available to analyse breath and the most important are the clinical (gender, age, diet, medication) samples a review on cancer detection mentions that GC-MS and e-nose and instrumental ones (time of the measurements, time from collection are the most commonly used platforms (47% and 26% of papers), while until analysis). A good design is essential to handle confounding factors their simultaneous use on the same samples appears only in 8% of the * Corresponding author. E-mail addresses: smarco@ibecbarcelona.eu, santiago.marco@ub.edu (S. Marco). https://doi.org/10.1016/j.cca.2021.12.019 Received 7 July 2021; Received in revised form 20 December 2021; Accepted 20 December 2021 Available online 23 December 2021 0009-8981/© 2021 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). L.F. Oliveira et al. Clinica Chimica Acta 526 (2022) 6–13 studies [29]. The use of two (or more) analytical distinct platforms the finds collecting a new group of samples and using external validation [30,31] on the same samples or confront the results from breath analysis methodologies and permutations test. Additionally, the same samples with another type of sample (as tissue/ sputum) [32–34] are an inter­ collected were also analyzed by GC-MS aiming to understand, if esting avenue of research in breath analysis and can be used to confirm possible, the origin of the chemical discrimination already proven by e- somehow the obtained results. noses and discuss the advantages and disadvantages of these distinct As said the raw data provided by different analytical platforms differ analytical platforms on breath analysis. In fact, prior studies using GC- in information content but also on dimensionality and data processing MS have found VOCs related to the presence of Pseudomonas Aerugi­ needs. E-noses and GC-MS datasets, particularly, are represented as a nosa (PA) in cystic fibrosis patients: methyl thiocyanate [53] and 2-ami­ vector and a matrix, for one sample, and as a matrix and a three-way noacetophenone [54] have been reported as putative biomarkers. In this array for the whole dataset, respectively. In general terms, GC-MS pro­ work, a volatolomics untargeted approach is proposed to discover po­ vides higher dimensional and richer information data, at the expense of tential signatures of PA infection in Bronchiectasis patients. requiring a more complex data processing pipeline [35]. Proper validation methodologies help to avoid overfitting and 2. Methods consequently reduce false discoveries [36–38]. Overfitting problems in GC/MS are aggravated by the curse of dimensionality [39], because A schematic representation of the applied methodology can be these datasets are highly dimensional (102–103 detected analytes) and observed in Fig. 1. This research features parallel analysis of breath most studies have limited sample size (20–100 subjects). In this context, samples by electronic nose and GC/MS. The following sections provide we advocate the use of resampling methods for model optimization and methodological information. external validation for performance assessment [40,41]. We have to remind that for small datasets, all samples can be used for external 2.1. Cohort selection and experimental design validation using double cross-validation methodologies [42]. Even if external validation is unbiased, small sample datasets do still feature Observational studies are always suspect of bias. In order to block performance estimators with large variance and a permutation test potential confounding factors, we carried out a proper experimental should be used to confirm the obtained results [43]. design. To prevent gender as a confounding factor a restriction strategy In the last decade, a number of VOCs in breath have been found to be was applied. It is known that in non-cystic fibrosis bronchiectasis has helpful in the diagnostics of several diseases including respiratory dis­ more prevalence among females than among males. Additionally, fe­ eases and cancer [5,44–46] Among the several diseases that can be males suffer more severe diseases and with worse prognoses in terms of evaluated by breath analysis the development of bronchial infections on poorer lung function and survival [55]. For these reasons, only females bronchiectasis patients has been described and evaluated by previous were included in the cohort. work. In this case, electronic nose has been used to identify airway Breath samples from woman subjects were collected, all of them bacterial colonization in Chronic Obstructive Pulmonary Disease were not currently smokers and the ones with prescribed drugs therapies (COPD) patients [47,48]. Furthermore, e-nose technology has been were asked to stop medication 1 day before sample collection and food- proven successful to identify Pseudomonas aeruginosa infection in bron­ drink intake at least 12 h before. All patients signed the informed con­ chiectasis [49]. sent form to participate in the study (Ethical approvement code: Institut In the clinical stability phase the presence of potential pathogens d’Investigació Biomédica Sant Pau- IIBSP-PRI-2018-105). Diagnosis of bacteria in the airway of bronchiectasis patients are common (30–70%) Bronchiectasis was performed according to current European guidelines being mainly Pseudomonas aeruginosa [50]. Furthermore, aggravations [56]. Bronchial infection was determined using a quantitative sputum as faster lung function loss, pulmonary and systemic inflammation are culture prior to breath samples collection. PA infection was diagnosed serious concerns. Bronchial infection is the reason for 60–70% of these using sputum culture that was not performed in healthy controls because aggravations [51], that has a direct relation with mortality increase in PA only affects patients with pulmonary diseases who had chronic Bronchiectasis [52] and the reasons why bronchiectasis patients are sputum production [50]. The procedure used to diagnose PA was well more susceptible to developing a bronchial infection are still unknown. validated and previously described [57], besides that no other patho­ In this work, we build upon previous works to test the adequacy of gens differently of PA were detected. Bacterial colonization was breath sampling to monitor infections in bronchiectasis patients, considered when patients had PA infection and clinical stability, defined particularly with Pseudomonas aeruginosa. Previous studies have re­ by the absence of increased symptoms that required changes in baseline ported success on e-nose applications using linear discriminant analysis treatment during 4 weeks [57]. and leave-one-out internal validation. In this new study, we amplified To block instrumental shifts often found in e-noses [37] and even in Fig. 1. Overview of sample collection and analysis. 7 L.F. Oliveira et al. Clinica Chimica Acta 526 (2022) 6–13 GC-MS, we matched the collection and analysis of samples belonging to injector. The chromatographic column used was type DB-624 (60 m × the different groups as depicted in Fig. 2. Supplementary Material 1 0.320 mmID × 1.8µm – Agilent). The temperature of the column was shows specific information for the subjects in the study. maintained at 40 ◦ C for 2 min and then subjected to a temperature ramp of 10 ◦ C/min till 250 ◦ C and stayed at this temperature for 5 additional minutes. The carrier gas used was helium in a constant flow of 1.7 mL 2.2. Breath sample collection min− 1. The temperatures of the injector and the transfer line were set to 250 ◦ C and 230 ◦ C, respectively. Ion source temperature was set to Three liters Tedlar® bags were used to collect the total amount of 200 ◦ C and the mass scan range was from 40 to 400 m/z. exhaled air by the patients. Two-valve Tedlar® bags were cleaned before use by flushing with argon and baking at 45 ◦ C during 15 min (repeated three times) [58]. All samples were collected in the same room after the 2.4. Data analysis patients breathing through a Hans-Rudolph valve during 3 min as described in previous work [49]. Also, a biological filter was used for 2.4.1. E-nose each patient to avoid pathogens entering the bags and cross contami­ A non-linear transformation (arctangent transformation) was used to nation between patients. improve data gaussianity [60]. Data normality was then confirmed with Samples were collected in the hospital and e-nose measurements the Shapiro-Wilk test at the 5% risk with Benjamini-Hochberg multitest were done a few minutes after the patients filled the bags. Then, the bags correction [61]. Variance of inter-replicates for each sample and robust were carried out to the laboratory and analyzed on the same day of PCA [62] was used for outlier detection. Specifically, the algorithm sample collection by GC-MS. For each day of sample collection, ambient ROBPCA (available in the rospca package in R) was used and outliers air, controls, and cases samples were collected. The final sample set were selected based on the robust score distance and the robust analyzed consisted of 8 ambient air samples, 9 Controls (healthy orthogonal distance. Proper cutoff values for those statistics are given by women), 13 bronchiectasis patients (Bronch), and 12 bronchiectasis Hubert et al. [63]. After outlier removal, data was autoscaled and subjects with bronchial infection by Pseudomonas aeruginosa (Bro_PA). inspected by classical PCA. Subject classification was based on the K-NN algorithm (available in 2.3. Sample analysis the class package in R [64]) plus a majority voting over the replicate measurements. K-NN classifier optimization and performance assess­ 2.3.1. E-nose ment were based on double cross-validation [42] using leave one subject The e-nose device Cyranose 320® (Smith Detections, Pasadena, CA), out (LOSO). By LOSO we mean that all the replicate measurements from that features a nanocomposite sensor array with 32 sensors, was con­ the same subject are treated as a single indicator to decide the final label nected to the breath Tedlar bag for 5 min and each measurement con­ given to the subject. In the inner loop (internal validation) the number of sisted of 5 replicates. Nitrogen was used as carrier gas and a constant neighbors was optimized, while performance assessment was carried out flow rate of 120 mL/min was used during 60 s and 40 s for baseline in the external loop (external validation). In both cases, the chosen recording and sample analysis, respectively, followed by an increase of figure of merit was classification accuracy (classification rate: CR). Final the flow for 180 mL/min for sample line purging and air intake. A Tedlar class assignment to each subject was based on the joint classification of bag with ambient air collected in the day of sample analysis was ana­ all the replicates through a voting mechanism. To check that the ob­ lysed in parallel every day as background measurement. Supplementary tained value cannot be obtained by random choice (null hypothesis) we Materials shows actual pictures of the breath sampling process (Figure 2 calculated a permutations test [65] with 500 iterations. Supplementary in supplementary materials.) Material 3 shows a block diagram of the e-nose data analysis (Figure 3 in supplementary materials.) 2.3.2. Gas chromatography Solid phase micro extraction (SPME) sample preconcentration was 2.4.2. Gas chromatography carried out using a 75 µm carboxen®/ Polydimethylsiloxane (CAR/ Features from the raw chromatograms were extracted using the PDMS) fiber [59]. The fiber was exposed inside the bags for 30 min at XCMS package in R [66,67]. On XCMS matched filter algorithm was ambient temperature and immediately after it was desorbed into the GC used for peak detection followed by peak clustering and alignment. Data Fig. 2. Distribution of cases, controls, and quality controls samples between the days of collection. 8 L.F. Oliveira et al. Clinica Chimica Acta 526 (2022) 6–13 imputation was used to fill missing values based on the integration of the Table 1 peak position. Robust Principal component analysis (RobPCA) was used Summary of the KNN models performance in external validation for the e-nose to explore the data and verify the presence of outliers in the same dataset (confidence limits 95% in brackets, calculated according to the binomial manner described above. distribution). The extracted features were then corrected using log transformation All replicates and PQN normalization [68]. The AlpsNMR package [69] was used to Models best Sensitivity Specificity CR(%) p- create Partial Least Squares – Discriminant Analysis (PLS-DA) [70] k value classification models followed by permutations test. Furthermore, a Control vs 7 – – 78 0.002 second strategy was used applying variable selection based on Wilcoxon Bronch vs and binary problems were built using the same strategy applied to the e- Bro_PA nose data (double cross-validation [42] using leave one subject out Control vs 7 0.86 0.9 89(84,96) 0.002 Bronch (0.73,0.92) (0.76,0.96) (LOSO)), and PLS-DA and KNN models were built. All data analysis for e- Control vs 5 0.94 0.9 92(83,95) 0.004 nose and GC-MS was done in RStudio 4.0.3. Bro_PA (0.82,0.97) (0.76,0.96) Bronch vs 5 0.86 0.92 89(81,93) 0.002 3. Results Bro_PA (0.71,0.92) (0.83,0.97) Majority vote 3.1. E-nose analysis Models best Sensitivity Specificity CR (%) p- k value Control vs 7 84 0.002 E-nose analysis aim is to record a breath-print of a wide range of – – Bronch vs gases and vapours (mixture of compounds) present in each breath Bro_PA sample not focusing on a specific compound or class of compounds, in Control vs 7 0.92 1(0.66,1.00) 95 0.002 the case of Cyranose 320® this is done by 32 nanocomposite sensor Bronch (0.64,1.00) (77,100) arrays. Each patient breath measurement is represented as a matrix, Control vs 5 1(0.75,1.00) 1(0.66,1.00) 100 0.004 Bro_PA (84,100) having 32 columns, one per sensor, and as many rows as replicates have Bronch vs 5 0.75 1(0.74,1.00) 87(69,97) 0.002 been measured (typically five). Supplementary Material 4 shows a Bro_PA (0.43–0.95) heatmap where on rows is showed the different replicates of each pa­ tient and in the columns the sensors. The colors correspond to the value of the sensor’s response after preprocessing. The rows are arranged ac­ with dimensions 42 samples × 409 features. The outlier detection step cording to the class of the patient and the columns are ordered according did not show any anomalous sample and all subjects were kept in the to hierarchical clustering of the sensor responses. After the non-linear data set. The Total Ion Chromatograms (TICs) in log scale for the GC-MS transformation, all sensors except numbers 6 and 28 were normally analysis for all collected samples and the outlier detection step plot are distributed and we could not reject the null hypothesis of a normal shown on Supplementary Material 6(i) and 6(ii). distribution using the Shapiro-Wilk hypothesis test with Benjamini- Fig. 4 (i) shows the score plots of a PLS-DA model for a binary Hochberg multitest correction. Three entire samples (including all rep­ classification problem. However, it is well known that scoreplots are licates) were considered outliers (see methods) and were removed from overoptimistic. In this spirit, we gave better credit to the evaluation of the dataset before the construction of the models. After preprocessing, the classifiers in external validation. The best results were obtained for data were visually inspected by PCA and the score plot (PC1xPC2) is the discrimination between Control versus Bronchiectasis_PA. PLS-DA shown in Fig. 3. No clear data separation was observed at this point. models presented good classification rates above 0.75 on external vali­ The best number of neighbors k (optimized in the internal validation dation. However, permutation tests were applied for all binary PLS-DA loop), the values of classification rates (in external validation), and the models, and it was not possible to reject the null hypothesis (see Fig. 4 p-value after permutation tests for all constructed models are showed in (ii)). Table 1 (including three class and two class models). Furthermore, Table 2 shows a summary of the obtained results for the GC-MS after Supplementary Material 5 shows the confusion matrix for the three class feature selection applying Wilcoxon test (binary models) and then K-NN models. applying the same strategy used to the e-nose (double cross-validation using leave one subject out). The last step on an untargeted approach is the compound identifi­ 3.2. Gas chromatographic analysis cation and although several important compounds were already iden­ tified and described as potential to be related with specific diseases on The application of XCMS to the raw data provided a feature table breath samples [71,72] the untargeted methodology used here was not able to reach the annotation step, in other words, even though some PLS- DA presented good results neither model was able to overcome per­ mutations tests for statistical significance. 4. Discussion 4.1. Sample collection and analysis Breath samples can be collected and analyzed online and offline, the main reason for choosing one or another method will depend on the final aim of the work. Although, analyzing breath directly and the use of cartridges are preferential for e-nose and GC-MS, respectively, Tedlar® bags fits very well when the objective is to analyze the same sample with two or more analytical platform [58]. Furthermore, when a patient has a pulmonary disability, identifying and collecting the end-tidal breath it is Fig. 3. Score plot for the Principal Component Analysis (PC1xPC2) containing not a simple task and for this reason and, to follow the same protocol for all samples from the e-nose measurements. all involved subjects in the study, whole breath samples were collected 9 L.F. Oliveira et al. Clinica Chimica Acta 526 (2022) 6–13 Fig. 4. (i) Scoreplot PLS-DA model obtained from alpsNMR using as class Control versus Bronchiectasis_PA (ii) permutation test for the predictive model. When trying to classify individual measurements, K-NN models Table 2 presented very good classification rates on external validation, and the Summary of results obtained for the GC-MS after feature selection applying values varied between 78% and 92%. Permutation tests were used for all Wilcoxon test (binary models). models and in all cases, the classification rates were considered statis­ PLS-DA KNN tically significant (risk level 0.05) compared with the distribution of the Number of selected features CR (%) CR (%) null hypothesis. The three-class problem resulted in a smaller CR (78%) Control vs Bronch 39 40 62 but still statistically significant. For all the other binary problems the CR Control vs Bro_PA 42 62 52 ranged between 89 and 92%, but those differences were not statistically Bronch vs Bro_PA 13 48 58 significant due to the limited cardinality of the different groups. Results improved significantly when we used the majority vote and the methodology was previous validated [49]. mechanism to classify a subject using the five consecutive replicates. In Furthermore, the CAR/PDMS fiber and the column selected possess this case, the CR for the three-class problem improved up to 84%, while intrinsic characteristics that will allow the pre-concentration and anal­ we got perfect classification (100)% for the Control vs Bronchiectasis ysis of a delineated group of compounds presented in breath samples with PA infection. The next model in terms of good performance was the analyzed by GC-MS. However, the preconcentration step is mandatory discrimination between Controls and Bronchiectasis, while the presence since many compounds will be present in very low concentrations inside of PA infection in Bronchiectasis performed a bit lower but still with an the bags mainly when whole breath samples are collected as in this excellent 87% classification rate. In general, models presented better work. Breath sampling methodologies, advantages, and disadvantages specificity than sensitivity, however, the latter still ranged from 92% to are described in the literature [73]. 100% (see Table 1). The current study indicates that the e-nose was able to classify the breath samples not only in internal validation as previously described 4.2. E-nose analysis but also in external validation. Furthermore, the class separation is not linear requiring non-linear decision functions to obtain good results. Initial visual inspection of the sensor response distribution indicated While these results are encouraging, they should be further validated a strong lack of normality. The histogram presented long right-side tails with more subjects (due to the risk of over adjustment related with the but also the presence of negative values. This is known to have a small sample conditions), during a longer study, and eventually in a negative effect on data analysis techniques based on the analysis of multicenter study. It should be independently tested with additional e- variance. Data normality was greatly improved using a non-linear nose units. Another direction of study is to investigate if this very good transformation. To be able to deal with negative values we selected separation is specific to the sensing technology used with the presently the arctangent transformation instead of the most common logarithmic used device or if they can be replicated with electronic noses of different transform. technologies. It is possible to observe that samples do not appear linearly separable in an unsupervised exploration based on the PCA scoreplot (Fig. 3). K- NN and PLS-DA classifiers were evaluated. PLS-DA is one of the most 4.3. Gas chromatographic analysis common classifiers in metabolomics, but it provides only linear parti­ tions of the input space. To test a more flexible input space partition K- While the score plot shows a good separation between the two NN models were chosen as a simple an model-free alternative. All K-NN studied classes, we have to take into account that PLS-DA score plots are models constructed for the e-nose dataset presented better results than easily overoptimistic [36,74]. Additionally, the apparent good result in the PLS-DA ones and for this reason, only the KNN results are shown. classification rate is unable to overcome the additional permutation test The classification rates were calculated using two different ap­ due to the large variance of the CR estimator probably linked to the proaches. First, considering each individual replicates and second using small number of samples compared to the input data dimensionality. the most voted class using all replicates from the same individual. This Regarding the second strategy applied (Table 2), these results agree last procedure does not represent in fact additional costs since all the with the obtained results for the e-nose in the sense that KNN models replicates are just consecutive analyses from the same bag as explained presented a better performance than PLS-DA (exception for the Control in Section 2.4.1. Using all the replicates approach significantly improved vs Bro_PA). However, in this work, all the predictive models constructed the performance of the classifier, except in the case of the three-class for the GC-MS data the classification rates were not good enough to problem. distinguish between the classes, and consequently it was not possible to 10 L.F. Oliveira et al. Clinica Chimica Acta 526 (2022) 6–13 discover the compounds that are important to class separation. analysis, Resources, Writing – review & editing, Supervision. Santiago It is interesting to confront the successful results of the predictive Marco: Conceptualization, Methodology, Formal analysis, Resources, models built with the e-nose measurements in opposition to the failure Writing – review & editing, Supervision, Project administration, Fund­ obtained using GC-MS data. We can point several underlying reasons ing acquisition. behind these results. First, the e-nose measurements have replicates (5 per sample) while a single GC/MS analysis is carried out per bag. Sec­ Declaration of Competing Interest ondly and as expected, the dimensionality of the e-nose is much smaller than the GC-MS leading to curse of dimensionality problems. This is The authors declare that they have no known competing financial more important when facing binary problems (for the GC-MS) since the interests or personal relationships that could have appeared to influence sample count is even smaller. the work reported in this paper. Furthermore, the signal processing pipeline for GC-MS is more complex than for e-nose data. The large number of peaks, sometimes Acknowledgements with strong coelution, baseline instabilities, and slight shifts in retention time leading to alignment problems, makes the whole data processing This work is part of the BEST Postdoctoral Program, funded by the workflow a real challenge, particularly if in addition we have a limited European Commission under Horizon 2020 Marie Skłodowska-Curie supply of examples for the machine learning step. This is in agreement Actions COFUND scheme (Grant Agreement no. 712754) and by the with previous research that combined GC-MS and e-nose analysis on the Severo Ochoa program of the Spanish Ministry of Science and same samples for cancer screening [75,76]. The GC-MS results obtained Competitiveness (Grant SEV-2014-0425 (2015-2019)). We would like to in this study sign that, even though the use of experimental design and acknowledge, the Departament d’Universitats, Recerca i Societat de la good analytical chemistry practices are essential, good validations Informació de la Generalitat de Catalunya (expedient 2017 SGR 1721); techniques in the development of the models are key to avoiding false the Comissionat per a Universitats i Recerca del DIUE de la Generalitat discoveries in complex data. de Catalunya; and the European Social Fund (ESF). Additional financial support has been provided by the Institut de Bioenginyeria de Catalunya 5. Conclusions (IBEC), Spain. IBEC is a member of the CERCA Programme/Generalitat de Catalunya, Spain. This work has been additionally funded by Spanish This study showed that e-noses were able to differentiate bronchi­ MINECO Project TENSOMICS (RTI2018-098577-B-C22). This work was ectasis and bronchiectasis with bronchial infections, produced by pseu­ supported also by Sociedad Española de Neumología y Cirugía Torácica domonas aeruginosa, patients from controls with good results in external (SEPAR), Societat Catalana de Pneumologia (SOCAP), Fundació Cata­ validation and the results were confirmed by permutation tests. lana de Pneumologia (FUCAP) and Instituto de SaludCarlos III - Fondos We would like to highlight a number of methodological factors that FEDER (PI18/00311). Finally, we want to acknowledge Prof. Francisco support the results and the validity of the conclusions. First, the proper Javier Santos, from the Department of Chemical Engineering and experimental design to block the most important confounding factors. Analytical Chemistry at the University of Barcelona, for his help in Second, the evaluation of the predictive models in external validation regards the GC-MS analysis of breath samples. using double leave one subject out and the additional permutation tests to explore if the obtained results can just be obtained due to the large Appendix A. Supplementary data variance of performance estimators in small sample conditions. Results for e-nose improved significantly after non-linear signal transformation, Supplementary data to this article can be found online at https://doi. and the use of majority voting over measurement replicates. org/10.1016/j.cca.2021.12.019. The use of GC-MS to explore the important compounds for the class differentiation was not successful. The main reasons for that were the small sample counting, the lack of replicates and the complexity of the References obtained signals. We consider that more strict validation methodologies [1] S. Das, M. Pal, Review—non-invasive monitoring of human health by exhaled should be in use to avoid false discoveries in breath analysis. breath analysis: a comprehensive review, J. Electrochem. Soc. 167 (3) (2020) Despite the good results obtained by electronic nose, the fact that this 037562, https://doi.org/10.1149/1945-7111/ab67a6. approach does not allow to identify condition specific compounds is a [2] M.D. Davis, S.J. Fowler, A.J. Montpetit, Exhaled breath testing – a tool for the clinician and researcher, Paediatr. Respir. Rev. 29 (2019) 37–41, https://doi.org/ clear limitation of this approach since it does not bring additional in­ 10.1016/j.prrv.2018.05.002. formation for the understanding of the underlying mechanisms below [3] J.D. Beauchamp, C.E. Davis, J. Pleil (Eds.), Breathborne Biomarkers and the the observed discrimination. Human Volatilome, Elsevier, Amsterdam, 2020, https://doi.org/10.1016/c2018-0- 04980-4. The obtained results should be considered as a positive indication [4] W. Miekisch, J.K. Schubert, G.F.E. Noeldge-Schomburg, Diagnostic potential of supporting the validity of the proposed methodology. However, studies breath analysis—focus on volatile organic compounds, Clin. Chim. Acta 347 (1-2) with larger cohorts, from different geographical areas and recruitment (2004) 25–39. [5] Y. Saalberg, M. Wolff, VOC breath biomarkers in lung cancer, Clin. Chim. Acta 459 hospitals are needed to give additional support to the findings reported (2016) 5–9, https://doi.org/10.1016/j.cca.2016.05.013. in this work. [6] M. Rodríguez-Aguilar, L. Díaz de León-Martínez, P. Gorocica-Rosete, R.P. Padilla, I. Thirión-Romero, O. Ornelas-Rebolledo, R. Flores-Ramírez, Identification of breath-prints for the COPD detection associated with smoking and household air CRediT authorship contribution statement pollution by electronic nose, Respir. Med. 163 (2020) 105901, https://doi.org/ 10.1016/j.rmed.2020.105901. Luciana Fontes de Oliveira: Conceptualization, Methodology, [7] L. Díaz de León-Martínez, M. Rodríguez-Aguilar, P. Gorocica-Rosete, C. A. Domínguez-Reyes, V. Martínez-Bustos, J.A. Tenorio-Torres, O. Ornelas- Software, Validation, Formal analysis, Investigation, Resources, Data Rebolledo, J.A. Cruz-Ramos, B. Balderas-Segura, R. Flores-Ramírez, Identification curation, Writing – original draft, Writing – review & editing, Visuali­ of profiles of volatile organic compounds in exhaled breath by means of an zation, Project administration, Funding acquisition. Celia Mallafré- electronic nose as a proposal for a screening method for breast cancer: a case- control study, J. Breath Res. 14 (4) (2020) 046009, https://doi.org/10.1088/1752- Muro: Software, Data curation, Writing – review & editing, Visualiza­ 7163/aba83f. tion. Jordi Giner: Validation, Investigation, Resources, Writing – review [8] A. Bikov, Z. Lázár, I. Horvath, Established methodological issues in electronic nose & editing. Lidia Perea: Validation, Investigation, Writing – review & research: how far are we from using these instruments in clinical settings of breath editing. Oriol Sibila: Conceptualization, Methodology, Validation, analysis? J. Breath Res. 9 (3) (2015) 034001, https://doi.org/10.1088/1752-7155/ 9/3/034001. Investigation, Resources, Writing – review & editing, Supervision, [9] S.F. Solga, L.A. Spacek, T.H. Risby, Challenges in clinical breath research Project administration, Funding acquisition. Antonio Pardo: Formal development, in: J.D. Beauchamp, C.E. Davis, J.D. Pleil (Eds.), Breathborne 11 L.F. Oliveira et al. Clinica Chimica Acta 526 (2022) 6–13 Biomarkers Hum. Volatilome, Elsevier B.V., Amsterdam, 2020, pp. 601–613, analysis by UV-irradiated electronic nose and GC/QTOF/MS, Sensors Actuators B https://doi.org/10.1016/b978-0-12-819967-1.00036-0. Chem. 311 (2020) 127932, https://doi.org/10.1016/j.snb.2020.127932. [10] A. Amann, W. Miekisch, J. Pleil, T. Risby, J. Schubert, Methodological issues of [32] Y. Sun, Y. Chen, C. Sun, H. Liu, Y. Wang, X. Jiang, Analysis of volatile organic sample collection and analysis of exhaled breath, in: I. Horvath, J.C. de Jongste compounds from patients and cell lines for the validation of lung cancer (Eds.), Exhaled Biomarkers, European Respiratory Society, 2010, pp. 96–114, biomarkers by proton-transfer-reaction mass spectrometry, Anal. Methods. 11 (25) https://doi.org/10.1183/1025448x.00018509. (2019) 3188–3197. [11] P. Sukul, P. Trefz, J.K. Schubert, W. Miekisch, Immediate effects of breath holding [33] E. Gashimova, A. Temerdashev, V. Porkhanov, I. Polyakov, D. Perunov, maneuvers onto composition of exhaled breath, J. Breath Res. 8 (3) (2014) A. Azaryan, E. Dmitrieva, Investigation of different approaches for exhaled breath 037102, https://doi.org/10.1088/1752-7155/8/3/037102. and tumor tissue analyses to identify lung cancer biomarkers, Heliyon 6 (6) (2020) [12] J.D. Beauchamp, W. Miekisch, Breath sampling and standardization, in: J. e04224, https://doi.org/10.1016/j.heliyon.2020.e04224. D. Beauchamp, C.E. Davis, J.D. Pleil (Eds.), Breathborne Biomarkers Hum. [34] D. Grove, G. Miller-Atkins, C. Melillo, F. Rieder, S. Kurada, D.M. Rotroff, A. Volatilome, Elsevier B.V., Amsterdam, 2020, pp. 23–41, https://doi.org/10.1016/ R. Tonelli, A.R. Tonelli, R.A. Dweik, R.A. Dweik, Comparison of volatile organic b978-0-12-819967-1.00002-5. compound profiles in exhaled breath versus plasma headspace in different diseases, [13] Z.-C. Yuan, W. Li, L. Wu, D. Huang, M. Wu, B. Hu, Solid-phase microextraction J. Breath Res. 14 (2020), https://doi.org/10.1088/1752-7163/ab8866. fiber in face mask for in vivo sampling and direct mass spectrometry analysis of [35] A. Smolinska, A.-C. Hauschild, R.R.R. Fijten, J.W. Dallinga, J. Baumbach, F.J. Van exhaled breath aerosol, Anal. Chem. 92 (17) (2020) 11543–11547, https://doi. Schooten, Current breathomics – a review on data pre-processing techniques and org/10.1021/acs.analchem.0c0211810.1021/acs.analchem.0c02118.s001. machine learning in metabolomics breath analysis, J. Breath Res. 8 (2) (2014) [14] X. Li, D. Huang, J. Zeng, C.K. Chan, Z. Zhou, Positive matrix factorization: A data 027105, https://doi.org/10.1088/1752-7155/8/2/027105. preprocessing strategy for direct mass spectrometry-based breath analysis, Talanta [36] R. Rodríguez-Pérez, L. Fernández, S. Marco, Overoptimism in cross-validation 192 (2019) 32–39, https://doi.org/10.1016/j.talanta.2018.09.020. when using partial least squares-discriminant analysis for omics data: a systematic [15] K. Rosenthal, D.M. Ruszkiewicz, H. Allen, M.R. Lindley, M.A. Turner, E. Hunsicker, study, Anal. Bioanal. Chem. 410 (23) (2018) 5981–5992, https://doi.org/10.1007/ Breath selection methods for compact mass spectrometry breath analysis, J. Breath s00216-018-1217-1. Res. 13 (4) (2019) 046013, https://doi.org/10.1088/1752-7163/ab34d4. [37] S. Marco, The need for external validation in machine olfaction: emphasis on [16] A. Smolinska, A.-C. Hauschild, R.R.R. Fijten, J.W. Dallinga, J. Baumbach, F.J. van health-related applications, Anal. Bioanal. Chem. 406 (16) (2014) 3941–3956, Schooten, Current breathomics - A review on data pre-processing techniques and https://doi.org/10.1007/s00216-014-7807-7. machine learning in metabolomics breath analysis, J. Breath Res. 8 (2) (2014) [38] R. Rodríguez-Pérez, M. Padilla, S. Marco, The need of external validation for 027105, https://doi.org/10.1088/1752-7155/8/2/027105. metabolomics predictive models, in: R. Cumeras, X. Correig (Eds.), Volatile Org. [17] G. Stavropoulos, D. Salman, Y. Alkhalifah, F.-J. van Schooten, A. Smolinska, Compd. Anal. Biomed. Diagnosis Appl, first ed.,, Apple Academic Press, Boca Preprocessing and analysis of volatilome data, in: J.D. Beauchamp, C.E. Davis, J. Raton, 2018, pp. 197–223, https://doi.org/10.1201/9780429433580-8. D. Pleil (Eds.), Breathborne Biomarkers Hum. Volatilome, Elsevier B.V., [39] D.L. Donoho, The Curses and Blessings of Dimensionality, in: Am. Math. Soc. Lect. Amsterdam, 2020, pp. 633–647, https://doi.org/10.1016/b978-0-12-819967- Challenges 21st Century, Los Angeles, 2000: pp. 1–33. https://www.dl.icdst.org/ 1.00038-4. pdfs/files/236e636d7629c1a53e6ed4cce1019b6e.pdf (accessed November 8, [18] J.D. Pleil, J.R. Sobus, Mathematical and Statistical Approaches for Interpreting 2017). Biomarker Compounds in Exhaled Human Breath, in: A. Amman, D. Smith (Eds.), [40] N. Fens, A.C. Roldaan, M.P. van der Schee, R.J. Boksem, A.H. Zwinderman, E. Volatile Biomarkers Non-Invasive Diagnosis Physiol. Med., Elsevier, Oxford, 2013, H. Bel, P.J. Sterk, External validation of exhaled breath profiling using an pp. 3–18. electronic nose in the discrimination of asthma with fixed airways obstruction and [19] W.B. Dunn, I.D. Wilson, A.W. Nicholls, D. Broadhurst, The importance of chronic obstructive pulmonary disease, Clin. Exp. Allergy 41 (10) (2011) experimental design and QC samples in large-scale and MS-driven untargeted 1371–1378, https://doi.org/10.1111/j.1365-2222.2011.03800.x. metabolomic studies of humans, Bioanalysis 4 (18) (2012) 2249–2264, https://doi. [41] M. Schumacher, N. Holländer, W. Sauerbrei, Resampling and cross-validation org/10.4155/bio.12.204. techniques: a tool to reduce bias caused by model building? Stat. Med. 16 (1997) [20] R. Rodríguez-Pérez, R. Cortés, A. Guamán, A. Pardo, Y. Torralba, F. Gómez, 2813–2827, https://doi.org/10.1002/(SICI)1097-0258(19971230)16:24<2813:: J. Roca, J.A. Barberà, M. Cascante, S. Marco, Instrumental drift removal in GC-MS AID-SIM701>3.0.CO;2-Z. data for breath analysis: the short-term and long-term temporal validation of [42] P. Filzmoser, B. Liebmann, K. Varmuza, Repeated double cross validation, putative biomarkers for COPD, J. Breath Res. 12 (3) (2018) 036007, https://doi. J. Chemom. (2009) 160–171, https://doi.org/10.1002/cem.1225. org/10.1088/1752-7163/aaa492. [43] F. Lindgren, B. Hansen, W. Karcher, M. Sjöström, L. Eriksson, Model validation by [21] L. Blanchet, A. Smolinska, A. Baranska, E. Tigchelaar, M. Swertz, A. Zhernakova, J. permutation tests: applications to variable selection, J. Chemom. 10 (1996) W. Dallinga, C. Wijmenga, F.J. van Schooten, Factors that influence the volatile 521–532, https://doi.org/10.1002/(SICI)1099-128X(199609)10:5/6<521::AID- organic compound content in human breath, J. Breath Res. 11 (1) (2017) 016013, CEM448>3.0.CO;2-J. https://doi.org/10.1088/1752-7163/aa5cc5. [44] N.M. Zetola, C. Modongo, O. Matsiri, T. Tamuhla, B. Mbongwe, K. Matlhagela, [22] A.C. Hauschild, T. Frisch, J.I. Baumbach, J. Baumbach, Carotta: Revealing hidden E. Sepako, A. Catini, G. Sirugo, E. Martinelli, R. Paolesse, C. Di Natale, Diagnosis of confounder markers in metabolic breath profiles, Metabolites 5 (2015) 344–363, pulmonary tuberculosis and assessment of treatment response through analyses of https://doi.org/10.3390/metabo5020344. volatile compound patterns in exhaled breath samples, J. Infect. 74 (4) (2017) [23] D. Zanella, J.-F. Focant, J.E. Hill, P.-H. Stefanuto, Comprehensive gas 367–376, https://doi.org/10.1016/j.jinf.2016.12.006. chromatography-mass spectrometry, in: J.D. Beauchamp, C.E. Davis, J.D. Pleil [45] C. Lourenço, C. Turner, Breath analysis in disease diagnosis: methodological (Eds.), Breathborne Biomarkers Hum. Volatilome, Elsevier, Amsterdam, 2020, considerations and applications, Metabolites 4 (2014) 465–498, https://doi.org/ pp. 239–251, https://doi.org/10.1016/b978-0-12-819967-1.00015-3. 10.3390/metabo4020465. [24] H. Haick, R. Vishinkin, C. Di Natale, S. Marco, Sensor systems, in: J.D. Beauchamp, [46] M. Rodríguez-Aguilar, L. Díaz de León-Martínez, P. Gorocica-Rosete, R. Pérez- C.E. Davis, J.D. Pleil (Eds.), Breathborne Biomarkers Hum. Volatilome, Elsevier B. Padilla, C.A. Domínguez-Reyes, J.A. Tenorio-Torres, O. Ornelas-Rebolledo, V., Amsterdam, 2020, pp. 201–220, https://doi.org/10.1016/b978-0-12-819967- G. Mehta, B.N. Zamora-Mendoza, R. Flores-Ramírez, Application of chemoresistive 1.00013-x. gas sensors and chemometric analysis to differentiate the fingerprints of global [25] C.A. Mayhew, J. Herbig, J.D. Beauchamp, Proton transfer reaction–mass volatile organic compounds from diseases. Preliminary results of COPD, lung spectrometry, in: J.D. Beauchamp, C.E. Davis, J.D. Pleil (Eds.), Breathborne cancer and breast cancer, Clin. Chim. Acta 518 (2021) 83–92, https://doi.org/ Biomarkers Hum. Volatilome, Elsevier B.V., Amsterdam, 2020, pp. 155–170, 10.1016/J.CCA.2021.03.016. https://doi.org/10.1016/b978-0-12-819967-1.00010-4. [47] O. Sibila, L. Garcia-Bellmunt, J. Giner, J.L. Merino, G. Suarez-Cuartin, A. Torrego, [26] D. Smith, P. Španěl, G.B. Hanna, R. Dweik, Selected ion flow tube mass I. Solanes, D. Castillo, J.L. Valera, B.G. Cosio, V. Plaza, A. Agusti, Identification of spectrometry, in: J.D. Beauchamp, C.E. Davis, J.D. Pleil (Eds.), Breathborne airway bacterial colonization by an electronic nose in Chronic Obstructive Biomarkers Hum. Volatilome, Elsevier, Amsterdam, 2020, pp. 137–153, https:// Pulmonary Disease, Respir. Med. 108 (11) (2014) 1608–1614. doi.org/10.1016/b978-0-12-819967-1.00009-8. [48] H. Shafiek, F. Fiorentino, J.L. Merino, C. López, A. Oliver, J. Segura, I. de Paul, [27] C. Wang, P. Sahay, Breath analysis using laser spectroscopic techniques: Breath O. Sibila, A. Agustí, B.G. Cosío, K. Kostikas, Using the electronic nose to identify biomarkers, spectral fingerprints, and detection limits, Sensors 9 (2009) airway infection during COPD exacerbations, PLoS One. 10 (9) (2015) e0135199, 8230–8262, https://doi.org/10.3390/s91008230. https://doi.org/10.1371/journal.pone.013519910.1371/journal.pone.0135199. [28] D. Salman, G.A. Eiceman, D. Ruszkiewicz, V. Ruzsanyi, E. Brodrick, C.L.P. Thomas, g00110.1371/journal.pone.0135199.g00210.1371/journal.pone.0135199. Ion mobility spectrometry, in: J.D. Beauchamp, C.E. Davis, J.D. Pleil (Eds.), g00310.1371/journal.pone.0135199.g00410.1371/journal.pone.0135199. Breathborne Biomarkers Hum. Volatilome, Elsevier B.V., Amsterdam, 2020, t00110.1371/journal.pone.0135199.t00210.1371/journal.pone.0135199. pp. 171–183, https://doi.org/10.1016/b978-0-12-819967-1.00011-6. t00310.1371/journal.pone.0135199.t00410.1371/journal.pone.0135199.s001. [29] A. Krilaviciute, J.A. Heiss, M. Leja, J. Kupcinskas, H. Haick, H. Brenner, Detection [49] G. Suarez-Cuartin, J. Giner, J.L. Merino, A. Rodrigo-Troyano, A. Feliu, L. Perea, of cancer through exhaled breath: a systematic review, Oncotarget 6 (36) (2015) F. Sanchez-Reus, D. Castillo, V. Plaza, J.D. Chalmers, O. Sibila, Identification of 38643–38657, https://doi.org/10.18632/oncotarget.5938. Pseudomonas aeruginosa and airway bacterial colonization by an electronic nose [30] P.H. Stefanuto, D. Zanella, J. Vercammen, M. Henket, F. Schleich, R. Louis, J. in bronchiectasis, Respir. Med. 136 (2018) 111–117, https://doi.org/10.1016/j. F. Focant, Multimodal combination of GC × GC – HRTOFMS and SIFT – MS for rmed.2018.02.008. asthma phenotyping using exhaled breath, Sci. Rep. (2020) 1–12, https://doi.org/ [50] A. Rodrigo-Troyano, O. Sibila, The respiratory threat posed by multidrug resistant 10.1038/s41598-020-73408-2. Gram-negative bacteria, Respirology 22 (7) (2017) 1288–1299. [31] T. Saidi, M. Moufid, K. de Jesus Beleño-Saenz, T.G. Welearegay, N. El Bari, A. Lisset [51] J.A. Wedzicha, T.A. Seemungal, COPD exacerbations: defining their cause and Jaimes-Mogollon, R. Ionescu, J.E. Bourkadi, J. Benamor, M. El Ftouh, B. Bouchikhi, prevention, Lancet 370 (2007) 786–796. Non-invasive prediction of lung cancer histological types through exhaled breath [52] J.D. Chalmers, A. Filonenko, M. Shteinberg, P.C. Goeminne, M.J. McDonnell, S. Aliberti, T.C. Fardon, D. Obradovic, E. Operschall, C. Gerlinger, G. Sotgiu, 12 L.F. Oliveira et al. Clinica Chimica Acta 526 (2022) 6–13 K. Dimakou, E. Polverino, A. De Soyza, A. Hill, Characterisation of the frequent [63] M. Hubert, P.J. Rousseeuw, K. Vanden Branden, ROBPCA: A new approach to exacerbator phenotype in bronchiectasis: Data from the friends cohort, Am. J. robust principal component analysis, Technometrics 47 (2005) 64–79, https://doi. Respir. Crit. Care Med. 195 (2017) rccm.201711-2202OC. http://www.atsjournals. org/10.1198/004017004000000563. org/doi/abs/10.1164/ajrccm-conference.2017.195.1_MeetingAbstracts.A7305% [64] B.D.B.D. Ripley, W.N.W.N. Venables, in: Modern Applied Statistics with S, 0Ahttp://ovidsp.ovid.com/ovidweb.cgi? Springer-Verlag, New York, 2002, https://doi.org/10.1198/tech.2003.s33. T=JS&PAGE=reference&D=emed18&NEWS=N&AN=617704054. [65] M. Ojala, Permutation tests for studying classifier performance, J. Mach. Learn. [53] V. Shestivska, A. Nemec, P. Dřevínek, K. Sovová, K. Dryahina, P. Španěl, Res. 11 (2009) 1833–1863, https://doi.org/10.1109/ICDM.2009.108. Quantification of methyl thiocyanate in the headspace of Pseudomonas aeruginosa [66] C.A. Smith, E.J. Want, G. O’Maille, R. Abagyan, G. Siuzdak, XCMS: processing mass cultures and in the breath of cystic fibrosis patients by selected ion flow tube mass spectrometry data for metabolite profiling using nonlinear peak alignment, spectrometry, Rapid Commun. Mass Spectrom. 25 (17) (2011) 2459–2467, https:// matching, and identification, Anal. Chem. 78 (2006) 779–787. doi.org/10.1002/rcm.5146. [67] N.G. Mahieu, J.L. Genenbacher, G.J. Patti, A roadmap for the XCMS family of [54] A.J. Scott-Thomas, M. Syhre, P.K. Pattemore, M. Epton, R. Laing, J. Pearson, S. software solutions in metabolomics, Curr. Opin. Chem. Biol. 30 (2016) 87–93, T. Chambers, 2-Aminoacetophenone as a potential breath biomarker for https://doi.org/10.1016/j.cbpa.2015.11.009. Pseudomonas aeruginosa in the cystic fibrosis lung, BMC Pulm. Med. 10 (2010) 56, [68] F. Dieterle, A. Ross, G. Schlotterbeck, H. Senn, Probabilistic quotient normalization https://doi.org/10.1186/1471-2466-10-56. as robust method to account for dilution of complex biological mixtures. [55] C. Vidaillac, V.F.L. Yong, T.K. Jaggi, -.-M. Soh, S.H. Chotirmall, Gender differences Application in1H NMR metabonomics, Anal. Chem. 78 (13) (2006) 4281–4290, in bronchiectasis: a real issue? Breathe 14 (2) (2018) 108–121, https://doi.org/ https://doi.org/10.1021/ac051632c. 10.1183/20734735.000218. [69] F. Madrid-Gambin, S. Oller-Moreno, L. Fernandez, L. Fernandez, S. Bartova, M. [56] E. Polverino, P.C. Goeminne, M.J. McDonnell, S. Aliberti, S.E. Marshall, M. P. Giner, C. Joyce, F. Ferraro, I. Montoliu, S. Moco, S. Marco, S. Marco, AlpsNMR: R. Loebinger, M. Murris, R. Cantón, A. Torres, K. Dimakou, A. De Soyza, A.T. Hill, An R package for signal processing of fully untargeted NMR-based metabolomics, C.S. Haworth, M. Vendrell, F.C. Ringshausen, D. Subotic, R. Wilson, J. Vilaró, Bioinformatics 36 (2020) 2943–2945, https://doi.org/10.1093/bioinformatics/ B. Stallberg, T. Welte, G. Rohde, F. Blasi, S. Elborn, M. Almagro, A. Timothy, btaa022. T. Ruddy, T. Tonia, D. Rigau, J. Chalmers, European Respiratory Society guidelines [70] M. Barker, W. Rayens, Partial least squares for discrimination, J. Chemom. 17 (3) for the management of adult bronchiectasis, Eur. Respir. J. 50 (3) (2017) 1700629, (2003) 166–173, https://doi.org/10.1002/cem.785. https://doi.org/10.1183/13993003.00629-201710.1183/13993003.00629-2017. [71] J. Rudnicka, M. Walczak, T. Kowalkowski, T. Jezierski, B. Buszewski, Supp110.1183/13993003.00629-2017.Supp2. Determination of volatile organic compounds as potential markers of lung cancer [57] O. Sibila, G. Suarez-Cuartin, A. Rodrigo-Troyano, T.C. Fardon, S. Finch, E. by gas chromatography-mass spectrometry versus trained dogs, Sensors Actuators, F. Mateus, L. Garcia-Bellmunt, D. Castillo, S. Vidal, F. Sanchez-Reus, M.I. Restrepo, B Chem. 202 (2014) 615–621, https://doi.org/10.1016/j.snb.2014.06.006. J.D. Chalmers, Secreted mucins and airway bacterial colonization in non-CF [72] A. Amann, B.d.L. Costello, W. Miekisch, J. Schubert, B. Buszewski, J. Pleil, bronchiectasis, Respirology 20 (7) (2015) 1082–1088, https://doi.org/10.1111/ N. Ratcliffe, T. Risby, The human volatilome: volatile organic compounds (VOCs) resp.12595. in exhaled breath, skin emanations, urine, feces and saliva, J. Breath Res. 8 (3) [58] J. Beauchamp, J. Herbig, R. Gutmann, A. Hansel, On the use of Tedlar® bags for (2014) 034001, https://doi.org/10.1088/1752-7155/8/3/034001. breath-gas sampling and analysis, J. Breath Res. 2 (4) (2008) 046001, https://doi. [73] O. Lawal, W.M. Ahmed, T.M.E. Nijsen, R. Goodacre, S.J. Fowler, Exhaled breath org/10.1088/1752-7155/2/4/046001. analysis: a review of ‘breath-taking’ methods for off-line analysis, Metabolomics 13 [59] B. Buszewski, T. Ligor, J. Rudnicka, Clinical application of SPME: analysis of VOCs (10) (2017), https://doi.org/10.1007/s11306-017-1241-8. in exhaled breath as cancer biomarkers, Isocyanate Sampl. (2012) 17–18. [74] R.G. Brereton, G.R. Lloyd, Partial least squares discriminant analysis: taking the http://dialnet.unirioja.es/servlet/articulo? magic away, J. Chemom. 28 (2014) 213–225, https://doi.org/10.1002/cem.2609. codigo=4192463&orden=388902&info=link#page=17. [75] H.H. Nir Peled, Meggie Hakim, Paul A. Bunn, York E. Miller, Timothy C. Kennedy, [60] R.A. van den Berg, H.CJ. Hoefsloot, J.A. Westerhuis, A.K. Smilde, M.J. van der Jane Mattei, John D. Mitchell, Fred R. Hirsch, Non-invasive breath analysis of Werf, Centering, scaling, and transformations: improving the biological pulmonary nodules, J. Thorac. Oncol. 7 (2013) 1528–1533, https://doi.org/ information content of metabolomics data, BMC Genom. 7 (1) (2006), https://doi. 10.1097/JTO.0b013e3182637d5f.Non-Invasive. org/10.1186/1471-2164-7-142. [76] G. Peng, M. Hakim, Y.Y. Broza, S. Billan, R. Abdah-Bortnyak, A. Kuten, U. Tisch, [61] Y. Benjamini, Y. Hochberg, Controlling the false discovery rate: a practical and H. Haick, Detection of lung, breast, colorectal, and prostate cancers from exhaled powerful approach to multiple testing, J. R. Stat. Soc. Ser. B. 57 (1) (1995) breath using a single array of nanosensors, Br. J. Cancer 103 (4) (2010) 542–551, 289–300, https://doi.org/10.1111/j.2517-6161.1995.tb02031.x. https://doi.org/10.1038/sj.bjc.6605810. [62] T.-N. Yang, S.-D. Wang, Robust algorithms for principal component analysis, Pattern Recognit. Lett. 20 (9) (1999) 927–933, https://doi.org/10.1016/S0167- 8655(99)00060-4. 13