Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=lanl20
Analytical Letters
ISSN: 0003-2719 (Print) 1532-236X (Online) Journal homepage: https://www.tandfonline.com/loi/lanl20
Real-Time Distinguishing of the Xylene Isomers
Using Photoionization and Dissociation Mass
Spectra Obtained by Femtosecond Laser Mass
Spectrometry (FLMS)
Abdullah Kepceoğlu, Yasemin Gündoğdu, Kenneth William David Ledingham
& Hamdi Sukur Kilic
To cite this article: Abdullah Kepceoğlu, Yasemin Gündoğdu, Kenneth William David Ledingham & Hamdi Sukur Kilic (2020) Real-Time Distinguishing of the Xylene Isomers Using Photoionization and Dissociation Mass Spectra Obtained by Femtosecond Laser Mass Spectrometry (FLMS), Analytical Letters, 53:2, 290-307, DOI: 10.1080/00032719.2019.1647227
To link to this article: https://doi.org/10.1080/00032719.2019.1647227
View supplementary material Published online: 27 Jul 2019.
Submit your article to this journal Article views: 163
View related articles View Crossmark data
MASS SPECTROMETRY
Real-Time Distinguishing of the Xylene Isomers Using
Photoionization and Dissociation Mass Spectra Obtained
by Femtosecond Laser Mass Spectrometry (FLMS)
Abdullah Kepceoglua , Yasemin G€undogdua, Kenneth William David
Ledinghama,b, and Hamdi Sukur Kilica,c,d
a
Faculty of Science, Department of Physics, Selc¸uk University, Konya, Turkey;bSUPA, Department of Physics, University of Strathclyde, Glasgow, Scotland;cDirectorate of High Technology Research and
Application Centre, Selc¸uk University, Konya, Turkey;dDirectorate of Laser Induced Proton Therapy Research and Application Centre, Selc¸uk University, Konya, Turkey
ABSTRACT
Distinguishing chemicals and improvement on analytical methods has a direct impact on modern chemical analysis. In this work, the dissociative ionization of xylene isomers was investigated using a femtosecond laser mass spectrometry (FLMS) method with a custom-built linear time-of-flight (TOF) instrument. Laser beams at 800 nm and 400 nm were used and intensity-dependent analysis of the obtained mass spectra was performed using principal component analysis (PCA) to distinguish the xylene isomers, which give identical mass spectra in appearance that cannot be distinguished using nor-mal mass spectrometry methods. The results show that there is a statistically highly significant difference between the xylene isomers for two principal components (1 a > 99.99%) and minimal infor-mation loss (<5%) took place during the PCA procedure. Also, the use of the k-medoid clustering method showed that the isomers may be distinguished in real-time for a wide range of ionization laser pulse powers with approximately 99% accuracy. The results suggest that real-time isomer analysis by the FLMS method is suitable for mass spectral identification applications. The FLMS method has been shown to be an important alternative to other mass spectrometric methods that use different ionization mechanisms.
ARTICLE HISTORY
Received 22 March 2019 Accepted 19 July 2019
KEYWORDS
Femtosecond laser mass spectrometry (FLMS); isomers; principal component analysis (PCA); real-time chemical analysis; time-of-flight mass spectrometer (TOF-MS)
Introduction
Distinguishing normally indistinguishable chemicals such as isomers is a very important topic in the field of mass spectrometry (MS) and improving the results has a direct impact on the area of chemical detection including the structural characterization of organisms (Alvarez-Rivera et al. 2019); distinguishing cancerous and healthy tissues (Balog et al.
2015; G€undogdu and Kilic¸2018; G€undogdu et al.2019; Vaqas et al.2019); identification of
CONTACT Hamdi Sukur Kilic hamdisukurkilic@selcuk.edu.tr Faculty of Science, Department of Physics, Selc¸uk University, Konya 42050, Turkey.
Color versions of one or more of the figures in the article can be found online atwww.tandfonline.com/lanl.
Supplemental data for this article can be accessed on the publisher’s website athttp://dx.doi.org/10.1080/00032719. 2019.1647227.
ß 2019 Taylor & Francis Group, LLC 2020, VOL. 53, NO. 2, 290–307
unknown chemicals by comparison with a mass spectral database (Samokhin, Revel’skii, and Revel’skii2011); real-time detection of explosives and their residuals that have a wide impact on homeland security (Dalgleish et al.2012); and space and extraterrestrial explor-ation (Ledingham et al.1995; Kılıc¸1997; Kilic et al.1997; Kosmidis et al.1997; Palmer and Limero2001; Balsiger et al.2007; Briois et al.2016).
Xylene is an aromatic hydrocarbon substance having two methyl group attached to a benzene ring, and it has three isomers named ortho-, meta-, and para-xylene which are shown in Figure 1. Xylenes and similar volatile organic compounds have a significant impact on living organisms (Demirel et al. 2014; Liu et al. 2014a), natural sources (Durmusoglu, Taspinar, and Karademir 2010; Mitra and Roy 2011), and atmospheric processes, i.e., ozone formation (Hoque et al. 2008; Bauri et al. 2016). In this context, an understanding of those photochemical processes is very important.
In the literature, various techniques have been used with MS. Namely, early work on the metastable ion characteristics of the o- and p-xylene isomers was investigated by means of field ionization MS that revealed that ion intensities produced very similar abundances, and therefore the o- and p-xylene isomers were not distinguished (McLafferty and Bryce 1967). Lubman et al. analyzed the isomeric species using reson-ance-enhanced multiphoton ionization spectroscopy, such as the azulene–naphthalene systems (Lubman and Kronick 1982b), as well as cresol isomers (Tembreull and Lubman 1984). In case of the one or two wavelengths resonance-enhanced multiphoton ionization spectroscopy, as a result of the analysis of xylene isomers, a sensitivity of 40 ppm in helium was estimated (Blease 1985).
Lubman and Kronick determined the xylenes using ion-mobility spectrometry and sensitivity of a few ppb was demonstrated. However, the isomers were not distinguished (Lubman and Kronick 1982a). Blease highlights that “only the species with the lowest ionization potential can ever be selectively ionized using this technique” (Blease 1985). Using Fourier transform mass spectrometer, xylene isomers were distinguished by react-ing the xylene isomers with Vþand VOþ ions (Bjarnason1996).
In a recent study, xylene molecules were investigated by means of transform-limited femtosecond (fs) pulses to distinguish isomers by probing doubly charged parent ions with circularly polarized light and probing high harmonic generation from randomly oriented isomer molecules subjected to an intense laser field (Alharbi et al. 2018).
In the literature, different statistical approaches are used for interpretation of the mass spectra. One widely used approach is the covariance mapping technique which
was applied in early work and focused on the investigation of a single-/multi-electron ionization, multiphoton multiple ionization of some small molecules, time-resolved pump-probe techniques, the understanding of underlying physical mechanisms and hid-den correlations of observed ion peaks (Fransinski, Codling, and Hatherly 1989; Frasinski, Codling, and Hatherly 1989; Frasinski, Hatherly, and Codling 1991; Codling and Frasinski1993; Pinkham2008; Frasinski 2016; Walke2016).
Principal component analysis (PCA), a dimension reduction technique, has been widely used to carry out analysis on chemometric applications and has been considered to be the first tool in multivariate analysis applications (Jolliffe 1986; Wold, Esbensen, and Geladi 1987; Bro and Smilde 2014). In recent years, distinguishing chemicals by using different spectrometric and chromatographic methods have provided very reliable and accurate results. The use of the PCA method in different experiments has been widely performed using gas chromatography (GC) results (Schwartz et al. 1987; Burns et al. 1997; Masunaga et al. 2003; Guo, Wang, and Louie 2004; Skrobot et al. 2007), mass spectrometry imaging (MSI) (Fonville et al. 2013; Race et al. 2013), and electron impact ionization mass spectrometry (EI-MS) (Hejazi et al. 2009; Samokhin and Revelsky 2011; Samokhin and Revelsky2013).
Distinguishing xylene isomers by applying PCA on mass spectra obtained using EI-MS method has been performed (Samokhin and Revelsky 2011). However, it has been concluded that in order to ensure the accuracy and reliability of the measurements, the mass spectra must be obtained under the same experimental conditions (Samokhin and Revelsky 2013).
Among the many examples of the application of the various statistical approaches to characterize chemical parameters, the analysis of the mass spectra obtained by FLMS tech-nique by PCA has not yet been studied. In this work, we have obtained and interpreted femtosecond laser dissociative ionization and data from fragmentation patterns in the mass spectra of xylene isomers using FLMS. The purpose of this work is to develop a real-time procedure to distinguish isomers of xylene based on the application of the PCA on spectra obtained by FLMS.
A novel approach has been applied to the FLMS data for determining the experimental parameters. The results show that the optimal experimental parameters may be deter-mined by simply calculating the intergroup and intragroup variances on the multidimen-sional principal component (PC) space. Maximizing the intergroup distances and minimizing the intragroup variances ensure the validation of the higher distinction rates of the investigated chemicals. PCA has been demonstrated to be helpful in unambiguously distinguishing very similar mass spectra of these isomers. In other words, the intragroup variances of different isomers were considerably smaller than the intergroup variances. In the present work, the dependences of the intragroup variance in the PCA results on the ionizing laser pulse power and wavelength (800 nm and 400 nm) have been investigated. Materials and methods
Chemicals
Xylene isomers of 99% purity were obtained from Alfa Aeser and used without further sample preparation. The samples were introduced into the vacuum chamber effusively
via a manual leak valve (ULV-150, MDC Vacuum, Hayward, CA, USA). Each of the xylene isomer samples was filled in a 5-mL volumetric flask. One of the sample flasks was randomly selected and assigned to be an unknown xylene sample.
Experimental arrangement
In the experimental procedure, a custom-designed and built TOF mass spectrometer was used (Yildirim et al. 2010; Alıc¸ et al. 2012; G€undogdu et al. 2014; Kepceoglu et al. 2018). The spectra were recorded at 1.5 106 mbar which was maintained at as con-stant as possible pressure. The background pressure was kept around 108 mbar before the sample introduction. A microchannel plate (MCP) detector (El-Mul Technologies Ltd., Israel) was used to detect the generated ions.
The recording and analyses of the mass spectra were performed using a custom-built MATLAB program (Kepceoglu and Kılıc¸ 2016). The resolution of the mass spectrometer was determined to be m/Dm 200 at a mass of 100 amu (Yildirim et al. 2010). Many results obtained using this system have been recently published (Alıc¸ et al.2012; G€undogdu
et al.2014,2019; Kepceoglu et al.2016; G€undogdu and Kilic¸2018; Kepceoglu et al.2018). A Ti:sapphire femtosecond amplifier laser system (Quantronix, Integra-C-3.5, NY, USA) was used as an energy/ionization source which delivers energies up to 3.5 mJ per pulse at a fundamental wavelength of approximately 800 nm with a 90 fs pulse duration, 1 kHz repetition rate and 8 mm output beam diameter. The second harmonic (400 nm) of the laser beam was obtained by using a femtosecond third-harmonic generation kit (FK-800, Eksma Optics, Lithuania).
The laser beam was focused down to several micrometer spot sizes on the sample plume using a biconvex lens with a 25/30 cm effective focal length at the 800 and 400 nm wavelengths used. The laser intensities were up to 0.76 1015 W/cm2at 800 nm and up to 1.06 1015 W/cm2at 400 nm. These intensities were controlled using a circu-lar variable metallic neutral density filter (100FS04DV.4, Newport, CA, USA). The tem-poral laser pulse parameters were measured using the modified frequency-resolved optical gating (FROG) technique (Grenouille/Quick Frog system from Swamp Optics, GA, USA). The spatial laser pulse parameters were monitored by using a laser beam profiler (LBP2, Newport, USA, CA, USA). The complete femtosecond laser TOF-MS system and the experimental diagram are shown in Figure 2.
Statistical method and data processing
First, mass spectra from samples were recorded using FLMS and the mass spectra (M) obtained may be expressed in the following matrix form:
Mij¼ I11 I1n ... Iij ... Im1 Inn 2 6 4 3 7 5 (1)
where m and n represent the sample number and m/z value of a corresponding ion peak, respectively. One sample can be regarded as a linear combination of mass peaks and a point in MS space with (I1, I2,… , IN) coordinates.
M1j¼ I11l1þ I12l2þ ::: I1jljþ ::: I1NlN (2)
where Iij is the ith ion peak intensity and li is m/z value of jth ion peak. The
dimen-sions of the data were reduced using PCA. Each obtained mass spectrum was presented as a point in PC subspace. Each component is also related to the linear combination of the original MS data dimensions (m/z value) as described in Figure 3a byl0N
l0
N ¼ E1l1þ E2l2þ ::: þ ENlN (3)
where EN are the eigenvectors (AN or BN in Figure 3a) and sum of squares of
eigenvec-tors equal to the unit veceigenvec-tors of corresponding principal axes. PCA may be applied to transform mass spectral coordinates (I1, I2,… , IN) to (E1, E2) coordinates in
two-dimen-sional PC subspace.
Principal component analysis of the mass spectral data
In the present work, the following algorithmic scheme from laser ionization to PCA analysis is performed as follows: ionization of molecules! ion separation ! ion detec-tion! TOF spectra recording ! performing PCA, analyzing data and finally interpret-ation of results. The TOF mass spectra were recorded using eleven and thirteen different laser pulse powers at 400 nm and 800 nm laser wavelengths. The mass spectra were recorded in a raw data matrix where each row contains consecutive mass spectra and each column contains ion intensities for the corresponding flight time for each m/z ion. After recording the mass spectra, the TOF to m/z transformation was performed. Figure 2. Schematic diagram of the experimental setup for FLMS.
Prior to PCA, the mass spectra (averaged over 1000 laser pulses) were normalized to total ion intensity (which is called as a unit variance scaling) and the normalized data were autoscaled (the average value of each row is calculated and then subtracted from data and divided by the standard deviation of data). Normalization and autoscaling were performed to all rows of the raw data matrix and then a new data set X was defined. The covariance matrix of the data matrix (X) was calculated in order to apply PCA.
In order to apply PCA to the covariance matrix, an eigen-decomposition method or singular value decomposition (SVD) algorithm was used to the autoscaled data matrix. Using this approach, eigenvectors and eigenvalues are obtained separately but mass spectral coordinates (I1, I2,… , IN) must be transformed to ðPC1; PC2Þ coordinates in
two-dimensional PC subspace by multiplying together the eigenvector matrix and the calculated covariance matrix. Thus, the score values can be plotted in a PC space.
Once the eigenvectors and eigenvalues are calculated, a new matrix was created where each column corresponds to orthogonal PCs (Kepceoglu et al. 2016). After rep-resenting (projecting) the autoscaled data matrix X into the PC space, confidence interval error ellipses can be found in the two-dimensional PC plane. The data were separated into subgroups containing 10 samples for each investigated chemical (iso-mer). After obtaining the PCA results, two-dimensional scatter plots were obtained with accompanying 95% confidence interval ellipses. While the ellipse areas were related to the intragroup variance, intergroup distances were related to distinguishing those isomer groups.
Figure 3. Schematic representation of the data processing and statistical analysis processes. An unknown xylene isomer was analyzed by FLMS and the recorded mass spectra represent N-dimen-sional information which may be difficult to interpret. PCA coordinate transformed mass spectra are represented in a lower-dimensional principal space.M is a mass spectrum, n is the highest m/z value, Iiisith ion peak intensity, and liis them/z value for ith ion.
Real-time data analysis using a k-medoids clustering method
After the determination of the optimum laser pulse power to provide minimal information loss following PCA (e.g., maximum explanation in the two-dimensional PCA space), the k-medoids clustering method (Park and Jun 2009) was applied for a real-time distinguishing of the xylene isomers.
The distance described as a difference between numbers was ordered (real numbers, polynomials, Cauchy series, etc.) (Jackson 2005). In PCA plots, the intergroup distances were measured by using the center points of each isomer group (xn, yn) where Lij is the
distance between two points i and j in a Euclidean distance and described by: jLijj ¼ Xn i¼1ðxi yiÞ2 1=2 ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffijxi xjj2þ jyi yjj2 q (4)
Eq. (4) can be expressed in the general form as Eq. (5)involving the Minkowski dis-tances (Perlibakas 2004; Varmuza and Filzmoser2016).
jLijj ¼
Xn
i¼1ðxi yiÞp
1=p
(5) When p¼ 2, it is called the Euclidean distance, while when p ¼ 1 it is named the Manhattan or city block distance.
jLijj ¼
Xn
i¼1ðxi yiÞ (6)
Following the definition of types of distances between two points, the distance between two pairs of objectsXAandXBcan be defined as
jLijj ¼ Xð B XAÞTC1ðXB XAÞ
1=2
(7) where C1 is the covariance matrix of the sample matrix X and the distance described inEq. (7) is called the Mahalanobis distance.
The relationship between the Euclidean and Mahalanobis distances in cases of ori-ginal and normalized or unnormalized data sets in PC space has presented in the litera-ture (De Maesschalck, Jouan-Rimbaud, and Massart 2000). The Mahalanobis distances of the unnormalized and normalized data set are equal to the distances calculated for the original data set.
In this study, the Mahalanobis distances are used in the k-medoid clustering method to measure distances between sample and groups plotted in two-dimensional PC space because these parameters are independent of PCA transformation.
Clustering refers to grouping data into smaller parts in which each data point exhibits high similarity with the group. The k-medoid method, introduced by Kaufman and Rousseeuw in 1987 (Kaufman and Rousseeuw 1987; Kaufman and Rousseeuw 2009), optimizes the k points with minimum dissimilarity within a group where each point is called the medoid of the cluster. Generally, L1 metric city-block distances are used in
the k-medoids method. In the present work, the Mahalanobis type distance as described inEq. (7) was employed to find each medoid. After finding the medoids, the boundaries of domains were plotted using Voronoi diagrams.
Voronoi diagrams are based on work carried out by Gauss, Dirichlet, and Voronoi (Aurenhammer 1991). Voronoi diagrams are presented in Figure 7 and divided into
three domains (i.e., Voronoi cells) that consist of the mass spectral projections of the three isomers in two-dimensional PC space. Here the center point of each group repre-sents the medoids (i.e., site) for mass spectra distribution, the intersection points of two domains are called the edges (solid blue lines), and the intersection of three domains called the vertex. Here, each pair of Voronoi cells is equidistant from the edges and the vertices are equidistant from the neighboring three sites.
The first step in the real-time analysis is to calibrate the TOF-MS system to perform time-of-flight to m/z transformation. After the calibration process is complete, the same PCA scheme was applied to perform k-medoid clustering. Each k represents a cluster (or group) of data and each isomer group consists of 10 mass spectra form a cluster. The k-medoid clustering method was performed to identify the mass spectrum of the unknown xylene isomer as belonging to one of these clusters. The analysis was per-formed using a custom-built MATLAB program (Kepceoglu and Kılıc¸2016).
Results and discussion
Analysis of xylene mass spectra at 800 nm and 400 nm
Prior to PCA, analyses of ionization/dissociation of xylene (C8H10) isomers were
per-formed using femtosecond laser pulses as a function of laser intensity and wavelength. Possible dissociation channels were opened on xylene molecules due to the loss of H or CH3 in our results as proposed in the literature (Huang et al. 2003; Liu et al. 2014b;
Papadopoulou, Kaziannis, and Kosmidis 2016). C8H10þ h ! C½ 8H10 þþ e
! C8H9þ H½ þþ e or ! C½ 8H9 þþ H þ e
! C7H7þ CH½ 3þþ e or ! C½ 7H7 þþ CH3þ e
The ionization potentials (IP) of the xylene isomers were reported to be 8.56, 8.55, and 8.44 eV for o-, m-, and p-xylene, respectively (Lias 2005). The appearance energies (AE) of [M-H]þ ion are reported to be between 11.30 and 12.1 eV for o-xylene, 11.7 and 12.3 eV for m-xylene, and 11.35 and 12.1 eV for p-xylene. In addition, the AE of [M-CH3]þ are reported to be between 11.10 and 11.8 eV for m-xylene, 11.3 and 11.8 eV
for o-xylene, and 11.05 and 11.9 eV for p-xylene (Rosenstock et al.2011).
The femtosecond timescale dynamics of xylene isomers were revealed due to the effect of the relative positions of methyl groups, and H-loss and CH3-loss channels on
the ionization dynamics using the ultraviolet-pump/infrared-probe method (Papadopoulou, Kaziannis, and Kosmidis2016).
The literature reports that xylene isomers need to absorb at least six photons to be ionized through multiphoton ionization (MPI) at 800 nm while three photons are necessary to be absorbed for ionization at 400 nm. Figure 4a shows the mass spectra for molecular ionization performed using the 800 nm laser beam and Figure 4b
shows the mass spectra obtained at 400 nm. Figure 4a shows that a prominent molecular ion peak was detected. The ionization was followed by dissociation when the 175 mW laser pulse power (6.653 1013 W/cm2) was used. Relatively small
fragment groups were observed for the 1000 mW laser pulse power (3.802 1014 W/ cm2). The results show that molecular ions for m-xylene (1.901 1013 W/cm2) and p-xylene (5.322 1013 W/cm2) were not obtained. In contrast, the [H]þ peak is observed from m- and p-xylenes, while a prominent [C]þ peak is observed from p-xylene. It can be concluded that dissociation followed by ionization process domi-nates at lower laser intensities for m- and p-xylene.
In Figure 4b, the molecular ion peaks appear at the 50 mW (5.280 1013 W/cm2) pulse power for o-xylene and m-xylene while only 5 mW (5.280 1012 W/cm2) is required for p-xylene. These results suggest that dissociative-ionization dynamics have been taken place as pure ionization followed by dissociation due to the absence of frag-ment ion peaks in the obtained mass spectra.
The highest peak intensity is [H]þ for all isomers at the 800 nm and 400 nm wave-lengths used except for the molecular ion [M]þ peak which has the highest intensity for p-xylene at 800 nm.
The [M]þ peak may be broadened due to the laser pulse power or ion detector prop-erties. When the ion intensity of the sample is plotted on a log-log scale as a function of the laser intensity, the slope generally expresses the order of the multiphoton process. In the high-intensity regime, multiphoton ionization process may appear as to be 3/2 (i.e. one and a half multiphoton ionization process). This process is called the volume or saturation effect that originates from the saturation of the ionized species in the focal volume of a focused Gaussian beam (Levenson2012).
Figure 4. Xylene (C8H10– 106 amu) isomers (99% purity) mass spectra are given as the relative ion intensity vs. mass to charge ratio (m/z) for (a) 800 nm and (b) 400 nm laser wavelengths with an approximate 90 fs pulse width. The darker scaled mass spectra correspond to the lowest pulse power when the first ion peaks were observed.
Principal component analysis
As the first step in the isomer distinguishing procedure, dimension reduction was applied using PCA to mass spectra obtained by FLMS. PC was applied using pre-proc-essed mass spectral data to distinguish the xylene isomers. As a first result, three xylene isomers were distinguished using mass spectral data recorded as a function of laser pulse power and wavelength. The first two PCs are present with higher than 99% of the total variance in the mass spectrum for all analyses when ion peaks having intensities higher than 10% of molecular ion peak are used as seen in Figure 5. Even if only the peak intensities of the [C1Hn]þ group ions are used to discriminate isomers, greater
than 95% information (i.e., intergroup variance) may still be contained in the two-dimensional reduced data as seen inFigure 5c,d.
Figure 6shows the PCA results obtained using mass spectral data at a 1750 mW laser pulse power and 800 nm wavelength, where ion peaks intensities larger than 5% of the molecular ion were selected as the inputs.
Figure 5. PCA results obtained for the femtosecond laser mass spectra: (a) peak intensities 10% of molecular ion peaks using 800 nm and 1750 mW, (b) peak intensities 10% of molecular ion peaks using 400 nm and 1000 mW, (c) only ion peaks in the [C1Hn]þ group using 800 nm and 1750 mW, and (d) only [C1Hn]þ group ion peaks using 400 nm and 1000 mW. Color code: red: o-xylene, green: m-xylene, blue: p-xylene.
As indicated previously, most of the information gathered from the mass spectra may be presented as two-dimensional PCA results. Approximately, 99.83% of the total vari-ance is explained by the first two PCs. Loadings (coefficients) have been attributed to ion peaks corresponding to intergroup variance. The results show that o-xylene is clearly distinguished due to the variance between molecular ion peak [C8H10]þ and
smaller fragment ion peaks [C2H]þ. p-xylene was distinguished by considering smaller
ion peaks, especially peaks in the [C1Hn]þ group. Lastly, m-xylene can be distinguished
by the H-loss [C8H9]þ and CH3-loss [C7H7]þ channels, respectively.
The intergroup variances are given by measuring the Mahalanobis distances between the isomer groups determined to be 7.6984, 6.2665, and 7.5700 for the distances between the o-m, o-p, and m-p xylenes, respectively. After applying the two-tailed t-test, all isomers were statistically significantly distinguished. As a result of the t-t-test, the lowest significance level (a) (as small as a < 0.000001) was observed between o and p xylene isomers and the hypothesis is accepted within this significance level.
Furthermore, two-dimensional PCA results are presented in Supplementary Figures S1 and S2 in the supporting information using infrared (800 nm) and ultraviolet (400 nm) ionization MS where the peak intensities 10%. In addition, laser pulse power-dependent PCA intergroup and intragroup variances and Heron’s area values are provided in Supplementary Tables S1 and S2. The results show better performance for 800 nm laser radiation for the PCA results and isomer distinguishing process. Also, in preliminary work (Kepceoglu et al.2016), the dependence of PCA results on laser power and gas pressure has been characterized using the FLMS method. These results showed that fluctuations in these parameters do not affect the ability of the technique to distin-guish xylene isomers.
Real-time distinguishing of xylene isomers
Figure 7 shows k-medoid clustering to create k subgroups applied to PCA results of mass spectral data of xylene isomers obtained for laser pulse powers equal to 350, Figure 6. Biplot of score and loading values obtained by principal component analysis of mass spec-tra obtained using a 1750 mW laser pulse power and 800 nm laser wavelength. Color code: red: o-xylene, green: m-xylene, blue: p-xylene.
1000, 1250, 1500, 1750, and 2000 mW. In this example, k¼ 3 since xylene has three isomers. Each group has 10 spectra, represented as a dot in the k-medoids plots and clustering of thirty spectral data gave three different medoids. o-xylene, m-xylene, p-xylene, and the unknown xylene isomer are represented as red, green, blue, and black, respectively.
The classification of real-time acquired, unknown xylene isomer spectra (which is regarded as the 31st of the data or in k-medoid plots represented as the unknown xylene isomer) is carried out by calculating the distance of this data point presented in two-dimensional PC subspace to that stored in the database. The partitioning around medoids (PAM) algorithm (Kaufman and Rousseeuw 2009) was used to find medoids and Mahalanobis-type distances are used to calculate the group distances. The entire process required less than three seconds but the time may vary depending upon the averaging process executed by oscilloscope and the k-medoid algorithm (Park and Jun
2009). In addition, two-dimensional PC space was divided into one million points (each PC axis is divided into 1000 points). The k-medoid clustering method was applied for each point to precisely represent the Voronoi areas. As a result of the k-medoids clus-tering for each point, the Voronoi areas of the o-xylene, m-xylene, and p-xylene isomers are represented by red, green, and blue backgrounds, respectively.
The intergroup and intragroup variances values for first 30 samples are provided in
Table 1 for the PCA results shown in Figure 7. The intragroup variances of the xylene isomers are obtained by calculating the 95% confidence interval of ellipse areas. The intergroup variances are determined by measuring the Mahalanobis distances between three isomers (o-m, o-p, and m-p). The valida values of the two-tailed t-test results are given in parenthesis for PC1 and PC2, respectively. The areas of the intergroup varian-ces were calculated by Heron’s Formula using three intergroup distanvarian-ces. Intergroup variance increases with laser pulse power and reaches a maximum between 1500 and 1750 mW.
Similarly, the intragroup variance decreases with increasing laser pulse power and reaches a maximum from 1250 to 1750 mW. Due to the two-tailed t-test results, all iso-mers were significantly distinguished on either the PC1 or PC2 axis for all laser pulse power values. The highest a values were observed between the m and p xylene isomers Table 1. Laser pulse power-dependent PCA intergroup and intragroup variance values. The laser wavelength is 800 nm and ion intensities 10% of the molecular ion intensity.
Laser pulse power (mW)
PCA variances (800 nm and 10% intensities)
Intergroup variances Intragroup variances
jo-mj jo-pj jm-pj Area o m p Sum
350 17.8771 (1012–1020) 18.4794 (1015–1018) 12.5111 (0.40–1025) 106.6460 2.4827 3.9930 1.4510 7.9267 1000 18.4415 (1012–1020) 17.9976 (1015–1018) 12.7169 (0.40–1025) 108.4981 1.9712 4.3029 1.5003 7.7744 1250 17.6849 (1012–1018) 13.0755 (1015–1019) 18.6010 (0.44–1025) 110.3735 2.3500 3.9719 1.3316 7.6535 1500 14.2029 (1011–1013) 16.4772 (1015–1021) 18.6418 (0.001–1019) 112.7130 2.2814 4.6781 1.7186 8.6781 1750 14.1137 (1012–1012) 16.5979 (1014–1022) 18.4544 (0.001–1017) 112.2269 2.3091 4.0280 1.4491 7.7862 2000 18.3826 (1011–1015) 13.9243 (1012–1023) 16.5257 (0.01–1020) 110.4373 3.1933 5.0058 1.2824 9.4815
on the PC1 axis. Therefore, the maximum distinction is expected for 1500–1750 mW laser pulse power values.
Using the coordinates (PC1, PC2) of the unknown xylene isomer in a PC plane pre-sented in Figure 7, a thousand random numbers were generated using built-in MATLAB random number generator having a bivariate normal distribution with a mean value of mu (31st point coordinates) and sigma (average covariance value of the three xylene isomer groups). Following the random data point generation around the real-time data, leave-one-out cross-validation was applied for the evaluation of the accuracy (ratio of the correctly identified samples out of total sample measurements) and the error rate of the real-time distinguishing process. The misclassification rate was determined to be zero when the sigma value was lower than 10 times the average sigma value. When sigma value is equal to 10, only one point was misclassified of 1030 sam-ples using a laser pulse power of 350 mW. Therefore, this approach exhibited 99.90% accuracy. Similarly, 99.90%, 99.80%, 99.51%, 99.03%, and 99.71% accuracies were com-puted for laser pulse powers of 1000 mW, 1250 mW, 1500 mW, 1750 mW, and 2000 mW, respectively.
Conclusions
The goal of this study is to show that the FLMS method is suitable and powerful for identification applications such as cancer diagnostic research, material analysis, and Figure 7. Distinguishing the xylene isomers using the k-medoid clustering method for various laser pulse powers at the 800 nm laser wavelength. The Voronoi areas are presented where the unknown xylene isomer is regarded as belonging to that group. The black asterisks represent each Voronoi cen-ter in a Voronoi cell. The Voronoi edges are plotted with dotted lines. Color code: red: o-xylene, green: m-xylene, turquoise: p-xylene, black: unknown xylene isomer.
chemical detection, and may be an important alternative to other MS methods such as electrospray ionization MS, electron impact ionization MS, inductively coupled plasma MS, desorption electrospray ionization MS, and rapid evaporative ionization MS.
The results of this work have shown that when PCA is used, the intergroup variances of isomers are greater than the intragroup variance, i.e., the differences between isomers were responsible for the extension of the data cloud in multidimensional space. On the other hand, the differences in isomers originated from the sum of possible experimental fluctuations including the laser power, background pressure, electronic fluctuations of the electrode voltages, and detector power supply. It can clearly be argued that the opti-mal experimental parameters can be determined by simply calculating the intergroup and intragroup variances in the multidimensional PC space.
Furthermore, we have shown that using the k-medoid clustering method isomers may be distinguished in real-time, at less than three seconds depending on the duration of the data collection and averaging processes. Also, the real-time analysis method pre-sented in this work may be used to construct a mass spectral database to open new pos-sibilities for the analysis of fragmentation data using computational methods (Hufsky and B€ocker 2017). A more extensive and comprehensive study that includes a larger set of molecules in addition to the three isomers studied here will provide a more accurate picture of the actual potential of the method to distinguish isomeric species.
Data availability statement
The data that support the findings of this study are openly available in Mendeley Data athttp://dx.doi.org/10.17632/y5k6xhhw6v.1.
Funding
This work was supported by the Scientific and Technical Research Council of Turkey (TUBITAK) under Grant No. 1649B031405880 and Scientific Research Projects Coordination Unit of Selc¸uk University, Project No. 14201085.
ORCID
Abdullah Kepceoglu http://orcid.org/0000-0002-4743-5517
References
Alharbi, A., A. E. Boguslavskiy, D. Austin, N. Thire, D. Wood, P. Hawkins, F. McGrath, A. S. Johnson, I. Lopez-Quintas, B. Schmidt., et al. 2018. Femtosecond laser mass spectrometry and high harmonic spectroscopy of xylene isomers. Scientific Reports 8(1):3789. doi: 10.1038/s41598-018-22055-9.
Alıc¸, T. Y., H. S¸€uk€ur Kılıc¸, H. Durmus¸, M. Dogan, and K. W. D. Ledingham. 2012. A mass spec-trometric investigation of isomers of butane. Rapid Communications in Mass Spectrometry 26(8):893–905. doi:10.1002/rcm.6184.
Alvarez-Rivera, G., D. Ballesteros-Vivas, F. Parada-Alfonso, E. Iba~nez, and A. Cifuentes. 2019. Recent applications of high resolution mass spectrometry for the characterization of plant nat-ural products. TrAC Trends in Analytical Chemistry 112:87–101. doi:10.1016/j.trac.2019.01.002.
Aurenhammer, F. 1991. Voronoi diagrams—a survey of a fundamental geometric data structure. ACM Computing Surveys (CSUR) 23(3):345–405. doi:10.1145/116873.116880.
Balog, J., Kumar, S. J. Alexander, O. Golf, J. Huang, T. Wiggins, N. Abbassi-Ghadi, A. Enyedi, S. Kacska, J. Kinross, G. B., et al. 2015. In vivo endoscopic tissue identification by rapid evapora-tive ionization mass spectrometry (REIMS). Angewandte Chemie International Edition 54(38): 11059–11062. doi:10.1002/anie.201502770.
Balsiger, H., Altwegg, K. P. Bochsler, P. Eberhardt, J. Fischer, S. Graf, A. Jackel, E. Kopp, U. Langer, M. Mildner, J., et al. 2007. Rosina–Rosetta orbiter spectrometer for ion and neutral analysis. Space Science Reviews 128(1–4):745–801. doi:10.1007/s11214-006-8335-3.
Bauri, N., P. Bauri, K. Kumar, and V. K. Jain. 2016. Evaluation of seasonal variations in abundance of BTXE hydrocarbons and their ozone forming potential in ambient urban atmosphere of Dehradun (India). Air Quality, Atmosphere & Health 9(1):95–106. doi:10.1007/s11869-015-0313-z.
Bjarnason, A. 1996. Xylene isomer mass spectral identification through metal ion chemistry in an FTICR. Analytical Chemistry 68(21):3882–3883. doi:10.1021/ac960597u.
Blease, T. G. 1985. Laser multiphoton ionization spectroscopy and analysis of the xylene isomers. PhD thesis, The University of Edinburgh, UK.
Briois, C., R. Thissen, L. Thirkell, K. Aradj, A. Bouabdellah, A. Boukrara, N. Carrasco, G. Chalumeau, O. Chapelon, F. Colin., et al. 2016. Orbitrap mass analyser for in situ characterisa-tion of planetary environments: Performance evaluacharacterisa-tion of a laboratory prototype. Planetary and Space Science 131:33–45. doi:10.1016/j.pss.2016.06.012.
Bro, R., and A. K. Smilde. 2014. Principal component analysis. Analytical Methods 6(9): 2812–2831. doi:10.1039/C3AY41907J.
Burns, W. A., P. J. Mankiewicz, A. E. Bence, D. S. Page, and K. R. Parker. 1997. A principal-component and least-squares method for allocating polycyclic aromatic hydrocarbons in sedi-ment to multiple sources. Environsedi-mental Toxicology and Chemistry 16(6):1119–1131. doi:10. 1002/etc.5620160605.
Codling, K., and L. J. Frasinski. 1993. Dissociative ionization of small molecules in intense laser fields. Journal of Physics B: Atomic, Molecular and Optical Physics 26(5):783. doi: 10.1088/0953-4075/26/5/005.
Dalgleish, J. K., K. Hou, Z. Ouyang, and R. G. Cooks. 2012. In situ explosive detection using a miniature plasma ion source and a portable mass spectrometer. Analytical Letters 45(11): 1440–1446. doi:10.1080/00032719.2012.675485.
De Maesschalck, R., D. Jouan-Rimbaud, and D. L. Massart. 2000. The mahalanobis distance. Chemometrics and Intelligent Laboratory Systems 50(1):1–18. doi: 10.1016/S0169-7439(99)00047-7.
Demirel, G., €O. €Ozden, T. D€ogeroglu, and E. O. Gaga. 2014. Personal exposure of primary school children to BTEX, NO2and ozone in Eskis¸ehir, Turkey: Relationship with indoor/outdoor con-centrations and risk assessment. Science of the Total Environment 473:537–548. doi:10.1016/j. scitotenv.2013.12.034.
Durmusoglu, E., F. Taspinar, and A. Karademir. 2010. Health risk assessment of BTEX emissions in the landfill environment. Journal of Hazardous Materials 176(1–3):870–877. doi:10.1016/j. jhazmat.2009.11.117.
Fonville, J. M., C. L. Carter, L. Pizarro, R. T. Steven, A. D. Palmer, R. L. Griffiths, P. F. Lalor, J. C. Lindon, J. K. Nicholson, E. Holmes, and J. Bunch. 2013. Hyperspectral visualization of mass spectrometry imaging data. Analytical Chemistry 85(3):1415–1423. doi:10.1021/ac302330a. Fransinski, L. J., K. Codling, and P. A. Hatherly. 1989. Multiphoton multiple ionisation of N2 probed by covariance mapping. Physics Letters A 142(8–9):499–503. doi: 10.1016/0375-9601(89)90522-7.
Frasinski, L. J. 2016. Covariance mapping techniques. Journal of Physics B: Atomic, Molecular and Optical Physics 49(15):152004. doi:10.1088/0953-4075/49/15/152004.
Frasinski, L. J., K. Codling, and P. A. Hatherly. 1989. Covariance mapping: A correlation method applied to multiphoton multiple ionization. Science 246(4933):1029–1031. doi:10.1126/science. 246.4933.1029.
Frasinski, L. J., P. A. Hatherly, and K. Codling. 1991. Multiphoton multiple ionisation of N2O probed by three-dimensional covariance mapping. Physics Letters A 156(5):227–232. doi:10. 1016/0375-9601(91)90145-X.
G€undogdu, Y., H. Alptekin, P. Karabaglı, M. S¸ahin, and H. S¸. Kilic. 2019. Discrimination of can-cerous and healthy colon tissues: A new laser-based method. Lasers in Surgery and Medicine 51(4):363–369. doi:10.1002/lsm.23033.
G€undogdu, Y., A. Kepceoglu, H. S. Kilic, M. Akbas¸, M. Dogan, K. W. D. Ledingham. 2014. Femtosaniye Laser Pulsları Yardımıyla Cnþ (n¼ 1 … 60) Topak _Iyonlarının Hem C60 Hemde Grafit Hedefler Kullanılarak Ayrı Ayrı €Uretimi ve €Olc¸€um€u. Afyon Kocatepe €Universitesi Fen Ve M€uhendislik Bilimleri Dergisi 14(2):1–9.
G€undogdu, Y., and H. S¸. Kilic¸. 2018. A new method for investigation of different tissues using femtosecond laser mass spectrometry. Photomedicine and Laser Surgery 36(10):548–554. doi:10. 1089/pho.2018.4496.
Guo, H., T. Wang, and P. K. Louie. 2004. Source apportionment of ambient non-methane hydro-carbons in Hong Kong: Application of a principal component analysis/absolute principal com-ponent scores (PCA/APCS) receptor model. Environmental Pollution 129(3):489–498. doi:10. 1016/j.envpol.2003.11.006.
Hejazi, L., D. Ebrahimi, M. Guilhaus, and D. B. Hibbert. 2009. Discrimination among geometrical isomers of alpha-linolenic acid methyl ester using low energy electron ionization mass spec-trometry. Journal of the American Society for Mass Spectrometry 20(7):1272–1280. doi:10.1016/ j.jasms.2009.02.027.
Hoque, R. R., P. S. Khillare, T. Agarwal, V. Shridhar, and S. Balachandran. 2008. Spatial and tem-poral variation of BTEX in the urban atmosphere of Delhi. Science of the Total Environment 392(1):30–40. doi:10.1016/j.scitotenv.2007.08.036.
Huang, C.-L., J. C. Jiang, Y. T. Lee, and C. K. Ni. 2003. Photoisomerization and photodissoci-ation of m-xylene in a molecular beam. The Journal of Physical Chemistry A 107(20): 4019–4024. doi:10.1021/jp022387t.
Hufsky, F., and S. B€ocker. 2017. Mining molecular structure databases: Identification of small molecules based on fragmentation mass spectrometry data. Mass Spectrometry Reviews 36(5): 624–633. doi:10.1002/mas.21489.
Jackson, J. E. 2005. A user’s guide to principal components. Vol. 587. Hoboken, NJ: John Wiley & Sons.
Jolliffe, I. T. 1986. Principal component analysis and factor analysis. In Principal component ana-lysis. Berlin: Springer.
Kaufman, L., and P. J. Rousseeuw. 1987. Clustering by means of medoids. Amsterdam, North-Holland: Elsevier.
Kaufman, L., and P. J. Rousseeuw. 2009. Finding groups in data: An introduction to cluster ana-lysis. Vol. 344. Hoboken, NJ: John Wiley & Sons.
Kepceoglu, A., N. K€okl€u, Y. G€undogdu, €O. Dereli, and H. S. Kilic. 2018. Analysis of the xylenol isomers by femtosecond laser time of flight mass spectrometry. Canadian Journal of Physics 96(7):711–715. doi:10.1139/cjp-2017-0762.
Kepceoglu, A., Y. G€undogdu, K. W. D. Ledingham, and H. S. Kilic. 2016. Identification of the isomers using principal component analysis (PCA) method. AIP Conference Proceedings 1722(1):060004.
Kepceoglu, A., and H. S¸. Kılıc¸. 2016. Chemical analysis program for the time of flight mass spec-trometry system. International Journal of Applied Mathematics, Electronics and Computers 4(Special Issue 1):57–60.
Kilic, H. S., K. W. D. Ledingham, C. Kosmidis, T. McCanny, R. P. Singhal, S. L. Wang, D. J. Smith, A. J. Langley, and W. Shaikh. 1997. Multiphoton ionization and dissociation of nitro-methane using femtosecond laser pulses at 375 and 750 nm. The Journal of Physical Chemistry A 101(5):817–823. doi:10.1021/jp962495d.
Kılıc¸, H. S¸. 1997. Comparison of nanosecond and femtosecond laser mass spectrometry (FLMS). PhD thesis, Glasgow University, Glasgow, UK.
Kosmidis, C., K. W. D. Ledingham, H. S. Kilic, T. McCanny, R. P. Singhal, A. J. Langley, and W. Shaikh. 1997. On the fragmentation of nitrobenzene and nitrotoluenes induced by a femtosec-ond laser at 375 nm. The Journal of Physical Chemistry A 101(12):2264–2270. doi:10.1021/ jp963187i.
Ledingham, K. W. D., R. M. Deas, A. Marshall, T. McCanny, R. P. Singhal, H. S. Kilic, C. Kosmidis, A. J. Langley, and W. Shaikh. 1995. A comparison of femtosecond and nanosecond multiphoton ionization and dissociation for some nitro-molecules. Rapid Communications in Mass Spectrometry 9(15):1522–1527. doi:10.1002/rcm.1290091512.
Levenson, M. 2012. Introduction to Nonlinear Laser Spectroscopy 2e. Amsterdam, Netherlands: Elsevier, p. 191.
Lias, S. G. 2005. Ionization energy evaluation. NIST Chemistry WebBook, NIST Standard Reference Database 69.
Liu, F. F., B. I. Escher, S. Were, L. Duffy, and J. C. Ng. 2014a. Mixture effects of benzene, tolu-ene, ethylbenztolu-ene, and xylenes (BTEX) on lung carcinoma cells via a hanging drop air expos-ure system. Chemical Research in Toxicology 27(6):952–959. doi:10.1021/tx5000552.
Liu, Y.-Z., K. Gregor, S.-R. Xiao, and G. Thomas. 2014b. Ultrafast imaging of electronic relax-ation in ortho-xylene: New features from fragmentrelax-ation-ion spectroscopy. Chinese Physics Letters 31(12):127802. doi:10.1088/0256-307X/31/12/127802.
Lubman, D. M., and M. N. Kronick. 1982a. Discrimination of isomers of xylene by resonance-enhanced two-photon ionization. Analytical Chemistry 54(13):2289–2291. doi:10.1021/ ac00250a033.
Lubman, D. M., and M. N. Kronick. 1982b. Mass spectrometry of aromatic molecules with reson-ance-enhanced multiphoton ionization. Analytical Chemistry 54(4):660–665. doi:10.1021/ ac00241a014.
Masunaga, S., Y. Yao, I. Ogura, T. Sakurai, and J. Nakanishi. 2003. Source and behavior analyses of dioxins based on congener-specific information and their application to Tokyo Bay basin. Chemosphere 53(4):315–324. doi:10.1016/S0045-6535(03)00009-2.
McLafferty, F. W., and T. A. Bryce. 1967. Metastable-ion characteristics: Characterization of iso-meric molecules. Chemical Communications 23:1215–1217. doi:10.1039/c19670001215.
Mitra, S., and P. Roy. 2011. BTEX: A serious ground-water contaminant. Research Journal of Environmental Sciences 5(5):394–398. doi:10.3923/rjes.2011.394.398.
Palmer, P. T., and T. F. Limero. 2001. Mass spectrometry in the US space program: Past, present, and future. Journal of the American Society for Mass Spectrometry 12(6):656–675. doi:10.1016/ S1044-0305(01)00249-5.
Papadopoulou, C. C., S. Kaziannis, and C. Kosmidis. 2016. On the dynamics of xylene isomers excited in the vacuum-ultraviolet (VUV) region. Chemphyschem 17(15):2415–2423. doi:10. 1002/cphc.201600215.
Park, H.-S., and C. H. Jun. 2009. A simple and fast algorithm for K-medoids clustering. Expert Systems with Applications 36(2):3336–3341. doi:10.1016/j.eswa.2008.01.039.
Perlibakas, V. 2004. Distance measures for PCA-based face recognition. Pattern Recognition Letters 25(6):711–724. doi:10.1016/j.patrec.2004.01.011.
Pinkham, D. W. 2008. Ultrafast control of the dynamics of diatomic molecules. PhD thesis, University of Virginia, USA.
Race, A. M., R. T. Steven, A. D. Palmer, I. B. Styles, and J. Bunch. 2013. Memory efficient princi-pal component analysis for the dimensionality reduction of large mass spectrometry imaging data sets. Analytical Chemistry 85(6):3071–3078. doi:10.1021/ac302528v.
Rosenstock, H. M., K. Draxl, B. W. Steiner, and J. T. Herron. 2011. Ion energetics data. NIST Chemistry WebBook, NIST Standard Reference Database 69
Samokhin, A., and I. Revelsky. 2011. Distinguishing by principal component analysis of o-xylene, m-xylene, p-xylene and ethylbenzene using electron ionization mass spectrometry. European Journal of Mass Spectrometry 17(5):477–480. doi:10.1255/ejms.1152.
Samokhin, A., and I. Revelsky. 2013. Reliable comparison of electron ionization mass spectra recorded under identical experimental conditions. Spectroscopy Letters 46(2):117–121. doi:10. 1080/00387010.2012.692755.
Samokhin, A. S., A. I. Revel’skii, and I. A. Revel’skii. 2011. Comparison of electron ionization mass spectra of some organic compounds (MW< 200 Da) recorded on mass spectrometers of various types. Journal of Analytical Chemistry 66(14):1432–1436.
Schwartz, T. R., Stalling, D. L. Cy, and L. Rice. 1987. Are polychlorinated biphenyl residues adequately described by Aroclor mixture equivalents? Isomer-specific principal components analysis of such residues in fish and turtles. Environmental Science & Technology 21(1):72–76. doi:10.1021/es00155a008.
Skrobot, V. L., E. V. R. Castro, R. C. C. Pereira, V. M. D. Pasa, and I. C. P. Fortes. 2007. Use of principal component analysis (PCA) and linear discriminant analysis (LDA) in gas chromato-graphic (GC) data in the investigation of gasoline adulteration. Energy & Fuels 21(6): 3394–3400. doi:10.1021/ef0701337.
Tembreull, R., and D. M. Lubman. 1984. Use of resonant two-photon ionization with supersonic beam mass spectrometry in the discrimination of cresol isomers. Analytical Chemistry 56(11): 1962–1967. doi:10.1021/ac00275a048.
Vaqas, B., S. J. Cameron, J. L. Alexander, K. S. O’Neill, J. M. Kinross, and Z. Takats. 2019. The iKnife: Development and clinical applications of rapid evaporative ionization mass spectrom-etry. In The handbook of metabolic phenotyping, 219–236. Amsterdam, Netherlands: Elsevier. Varmuza, K., and P. Filzmoser. 2016. Introduction to multivariate statistical analysis in
chemomet-rics. Boca Raton, FL: CRC Press.
Walke, D. J. 2016. New instrumentation and methods for ultrafast pump-probe spectroscopy. PhD thesis, Imperial College London, UK.
Wold, S., K. Esbensen, and P. Geladi. 1987. Principal component analysis. Chemometrics and Intelligent Laboratory Systems 2(1–3):37–52. doi:10.1016/0169-7439(87)80084-9.
Yildirim, M., O. Sise, M. Dogan, and H. S. Kilic. 2010. Designing multi-field linear time-of-flight mass spectrometers with higher-order space focusing. International Journal of Mass Spectrometry 291(1–2):1–12. doi:10.1016/j.ijms.2009.12.014.
