Chemometric Data Analysis for Deconvolution of Overlapped Ion Mobility Profiles

  • Behrooz Zekavat
  • Touradj SoloukiEmail author
Research Article


We present the details of a data analysis approach for deconvolution of the ion mobility (IM) overlapped or unresolved species. This approach takes advantage of the ion fragmentation variations as a function of the IM arrival time. The data analysis involves the use of an in-house developed data preprocessing platform for the conversion of the original post-IM/collision-induced dissociation mass spectrometry (post-IM/CID MS) data to a Matlab compatible format for chemometric analysis. We show that principle component analysis (PCA) can be used to examine the post-IM/CID MS profiles for the presence of mobility-overlapped species. Subsequently, using an interactive self-modeling mixture analysis technique, we show how to calculate the total IM spectrum (TIMS) and CID mass spectrum for each component of the IM overlapped mixtures. Moreover, we show that PCA and IM deconvolution techniques provide complementary results to evaluate the validity of the calculated TIMS profiles. We use two binary mixtures with overlapping IM profiles, including (1) a mixture of two non-isobaric peptides (neurotensin (RRPYIL) and a hexapeptide (WHWLQL)), and (2) an isobaric sugar isomer mixture of raffinose and maltotriose, to demonstrate the applicability of the IM deconvolution.

Key words

Chemometrics Collision-induced dissociation (CID) Data analysis Ion mobility (IM) Mass spectrometry (MS) Principle component analysis (PCA) 

1 Introduction

High throughput analysis of complex chemical mixtures with mass spectrometry (MS) often requires analyte pre-separation using techniques such as gas chromatography (GC) [1, 2] and liquid chromatography (LC) [3] for neutrals or ion mobility (IM) [4] for ions. Fast scanning MS instruments are compatible with IM and the combination provides additional structural information. Hence, IM is a highly desirable gas-phase separation technique for MS applications.

For example, IM may be used to separate conformationally dissimilar biological [4, 5, 6] or nonbiological [7, 8] ionic species in the gas phase. However, the analytical resolving powers (defined as the peak arrival time/full width at half height (FWHH)) of the current conventional IM-MS systems are limited to under ~70 [9, 10]. Often, the IM profiles of various ionic species with similar (or very close) collision cross sections (CCSs) are not completely resolved [11, 12, 13, 14, 15, 16].

The IM overlap may occur for isobaric or nonisobaric species that have comparable CCSs. Regardless of the type of IM overlapping, there are at least two challenges associated with the unresolved IM profiles. The first challenge with the unresolved IM profiles is associated with the assignment of a correct arrival time (AT) value for each of the IM overlapped species. Generally, the observed or assigned AT values are used to empirically determine CCSs for the ionic species [5] and obtain structural/conformational information. Consequently, incorrect AT values obtained from the unresolved IM profiles can result in erroneous CCS calculations. The second challenge with the unresolved IM profiles is associated with the incorrect structural characterization of the overlapped species using post IM separation fragmentation techniques (e.g., collision-induced dissociation (CID) [17] and electron transfer dissociation (ETD) [12, 13]). The gas-phase fragmentation patterns of the mobility-separated species are often used to identify the unknown species [6, 11, 12, 13, 16, 18]. The presence of unresolved peaks in the IM dimension may cause misidentification of the species that rely on post-IM fragmentation patterns for peak assignments.

The above-mentioned issues with the overlapping IM profiles can be addressed by increasing the IM resolving power of the IM-MS systems using hardware modifications [10]. However, the hardware modifications of IM-MS systems can be costly and time intensive endeavors and only yield modest improvements. The desire to seek for alternative lower cost approaches that could utilize the existing IM data and CID mass spectra for IM deconvolution propelled us to explore the possibility of using post-IM/CID MS combined with chemometric data analysis approaches.

In this report, we introduce a chemometric data analysis approach for deconvoluting the overlapping IM peaks and their corresponding mass spectra from post-IM/CID profiles. Our approach allows the calculation of both the pure CID mass spectrum and pure total IM spectrum (TIMS) profile for each of the mobility-overlapped species in the mixture. The presented data processing procedure is instrument-independent and can be implemented with other types of IM systems to enhance the utility of other highly advanced IM devices [4, 7, 8, 9, 19]. We use the post-IM/CID MS data from two binary mixtures, each with overlapping IM profiles, to demonstrate the utility of our new approach for IM deconvolution. Deconvolution results for the IM data acquired with a Waters Synapt HDMS system with IM resolving power of <10 (Waters Inc., Manchester, UK) for two binary mixtures, including (1) a mixture of two nonisobaric peptides (neurotensin (RRPYIL) and a hexapeptide (WHWLQL)) and (2) an isobaric sugar isomer mixture of raffinose and maltotriose, confirm the validity of our approach.

Employing a similar approach to conventional GC/MS peak deconvolution (viz., comparing single ion chromatograms (SIC) across a GC peak), Clemmer et al. [11] and Cooper et al. [12, 13] utilized post-IM fragmentation techniques for deconvolution of mobility-overlapped profiles. Clemmer et al. acquired post-IM/CID MS data for mass isolated mobility-overlapped isobaric precursor ions [11] and by inspecting the CID mass spectra across the mobility profiles of the mass isolated precursor ions, fragment ions specific to each species were identified. The resultant fragment IM profiles (referred to as the “extracted fragment ion drift time distributions (XFIDTD)”) represented the AT distribution of each species [11]. Similarly, Cooper et al. used precursor-specific fragment ions, generated from ETD after field asymmetric waveform IM spectrometry (FAIMS) separation, to obtain the compensation voltage profiles of the overlapped isobaric species [12, 13].

These previously reported valuable approaches [11, 12, 13] for IM (or compensation voltage) deconvolution provide practical advantages for deconvoluting the overlapping IM peaks. However, neither the XFIDTD [11] nor the FAIMS/ETD method [12, 13] can be used to generate pure TIMS profiles and CID mass spectra for the overlapped species. The approach presented in this report makes it possible to calculate TIMS profiles of the mobility-overlapped components (i.e., total ion abundances of the overlapping species for quantitative comparisons).

2 Experimental

2.1 Materials and Methods

Acetic acid, hexapeptide (WHWLQL), maltotriose, methanol (HPLC grade), neurotensin (RRPYIL), raffinose, and water (HPLC grade) were purchased from Sigma-Aldrich (St. Louis, MO, USA) and used without further purification. A mixture of hexapeptide (~1.0 × 10–6 M) and neurotensin (~1.6 × 10–6 M) in methanol:water:acetic acid (~49.5:~49.5:~1) was used for the positive-ion mode electrospray ionization (ESI)/MS experiments. For the experiments involving isobaric sugars, a mixture containing raffinose (~7.5 × 10–6 M) and maltotriose (~1.0 × 10–5 M) in water was used and sprayed in the negative-ion mode ESI/MS. All of the IM-MS experimental parameters were optimized for the best IM and MS performances. Additional experimental details are provided in the relevant sections.

2.2 Ion Mobility-Mass Spectrometry (IM-MS)

The IM-MS analyses were performed using a first generation Synapt HDMS system (Waters Inc., Manchester, UK) equipped with an ESI source. The ESI voltages were set at +3.0 kV and −2.5 kV for positive- and negative-ion modes, respectively. The orthogonal time-of-flight (TOF) MS of Synapt HDMS system was operated in V-mode for maximum sensitivity. Under our experimental conditions, the maximum mass resolving power (m/Δm50%) in V-mode was ~10,000 (sufficient for our intended purposes). The IM separations were performed with a traveling-wave height of 7.5 V (wave velocity of 300 m/s) and relative potential differences of +3 V and −4 V on the trap and transfer cells with respect to the IM cell (default factory set voltages were used for all other ion guide parameters). In the post-IM/CID experiments, ions of interest were isolated before entering into the IM cell in the quadrupole assembly. The subsequent CID of the IM separated ions were conducted by increasing the electrical potential difference between the IM cell and transfer cell from −4 V to −60 V for the peptide mixtures (Figure 1a) or −35 V for the sugar mixtures (Figure 5a). Argon gas was used as both buffer gas (in the trap and transfer cells) as well as the collision gas in the transfer cell. The argon gas flow rate was set at 1.5 mL/min (corresponding to a pressure of 0.03 mbar when measured in the trap cell). During the IM separation events, a nitrogen bath gas was introduced into the IM cell at a flow rate of 25 mL/min providing a typical operating pressure of 0.59 mbar in the IM cell.
Figure 1

(a) Overlaid expanded view (between AT of 6.5 and 12.5 ms) of the original (solid line) and the preprocessed (shown with open circles [○]) post-IM/CID profiles of the protonated neurotensin (Neu) (RRPYIL) (m/z 817 Th) and hexapeptide (Hex) (WHWLQ) (m/z 882 Th) mixture. The preprocessed post-IM/CID profile was obtained by re-sampling the CID mass spectral using 105,000 m/z data points. (b) CID mass spectrum obtained by combining the mass spectra collected across the IM profile shown in panel a (solid line)

To construct a single post-IM/CID MS profile, a total number of 200 CID mass spectra at TOF “pusher” time interval of 90 μs for 18 ms were collected (across each IM profile). To enhance the signal-to-noise (S/N) ratio of the post-IM/CID MS data, 120 IM-MS profiles were averaged to construct the final post-IM/CID MS data. Therefore, a total number of 24,000 CID mass spectra were acquired and data from all spectra were collected to construct each of the post-IM/CID MS profiles shown in this report (e.g., Figures 1a and 5a). All of the CID experiments were repeated (in triplicate) to ensure the reproducibility of the collected IM-MS data within the allowed experimental error. Multiple trials yielded reproducible experimental results with total variations of less than ~10 % in IM peak area and ~0.5 % in peak AT values (see Table 1).
Table 1

Ion Mobility (IM) Profile Parameters Obtained from the Individual and Simultaneous (SIMPLISA) Isolation Experiments for the Protonated Neurotensin and Hexapeptide

IM Peak Parameter




Individual Isolation


Individual Isolation


AT {ms}

8.96 (±0.02)

8.97 (±0.05)

9.72 (±0.02)

9.74 (±0.02)

FWHH {ms}

1.24 (±0.02)

1.23 (±0.05)

1.34 (±0.02)

1.34 (±0.02)

AUC {arb. unit}

1961 (±467)

2196 (±176)

4702 (±651)

4437 (±305)

The AT, FWHH, and AUC values correspond to the average of three experimental trials and the numbers in the parentheses denote the experimental errors at the 95% confidence level.

2.3 Data Analysis

All of the data preprocessing and chemometric analyses were performed in Matlab 7.0 (The MathWorks Inc., Natick, MA, USA) software running on a 2.67 GHz Intel(R) Core i7 PC with 8.0 GB RAM desktop computer (Studio XPS 435T; Dell, Inc., Round Rock, TX, USA).

To import the collected IM-MS data (using Waters Synapt HDMS system) into the Matlab workspace, it was necessary to convert the format of the original data (.raw in Synapt HDMS system) to a Matlab compatible data format. The massWolf 4.3.1 software [20] was used to convert the IM-MS data from .raw format to .mzXML format [21] (compatible with Matlab). The massWolf 4.3.1 freeware was developed by Natalie Tasman [20].

Principle component analysis (PCA) was performed using PLS toolbox (ver. 2.0, Eigenvector Research, Inc., Wenatchee, WA, USA) in Matlab. The data preprocessing (Scheme 1) and SIMPLe-to-use Interactive Self-modeling Mixture Analysis (SIMPLISMA) [22] deconvolutions were performed using two separate in-house written Matlab user programs (included with the supplementary materials).
Scheme 1

Flow chart of the post-IM/CID MS data preprocessing platform used in this study

We used the mathematical integration across the mobility profiles with Origin 7.0 software (OriginLab Co., Northampton, MA, USA) to calculate the areas under the curves of the IM profiles.

Principle Component Analysis (PCA)

PCA is a multivariate mathematical technique that analyzes a data set representing observations described by several dependent (correlated) variables [23]. The important objective in PCA is to find possible combinations of variables that describe major trends in a data set. Therefore, as shown in Equation (1), PCA decomposes the data matrix “X” into the sum of outer product of vectors “s” (score) and “l” (loading) plus a residual matrix “E” (please note that for the listed equations throughout this report, uppercase and lowercase letters are used to represent data matrices and vectors, respectively):where symbols ⨂ and “ ′ ” denote the outer product vector multiplication and matrix transposition operators, respectively. Subscript “k” denotes the number of significant principle component(s) (PC) (or the “chemical rank” [24]) and is less than or equal to the number of original variables in the data matrix “X”. The application of the PCA to ATD and CID mass spectral data allows us to determine the number of significant components and presence (or absence) of the overlapping IM peaks in the post-IM/CID profiles.

SIMPLe-to-use Interactive Self-modeling Mixture Analysis (SIMPLISMA)

SIMPLISMA is one of the curve resolution techniques and was first introduced by Windig and Guilment for mixture analysis [22]. The chemometric curve resolution techniques are used to extract pure chemical information from the analysis of a sample representing a mixture of components. The pure chemical information can be, for example, an analyte spectrum and/or analyte concentration profile when a mixture of chemical components undergoes changes as a function of time (e.g., chemical reactions, chromatographic separations, IM separations, etc.).

The SIMPLISMA is based on the assumption that there is a chemical variable(s) (e.g., spectral wavelength(s), wavenumber(s), m/z, etc.) in the data set under chemometric analysis that experiences a major contribution(s) from only one of the components in the mixture. Such a chemical variable is called “pure variable” and spectral intensity at the pure variable represents the analyte concentration profile. The spectrum of analyte can also be resolved by knowing the concentration profile (as explained in the following section).


SIMPLISMA is based on the mathematical “term” for pure variable (p i,j ) [22]:
$$ {p_{{i,j}}} = \left[ {\frac{{{\sigma_i}}}{{{\mu_i} + \alpha }}} \right]{\omega_{{i,j}}} $$
where “ϭ i ” and “μ i ” are defined in Equations (3) and (4), respectively:
$$ {\sigma_i} = {\left[ {\left( {\frac{1}{n}} \right)\sum\nolimits_{{j = 1}}^n {{{\left( {{X_{{i,j}}} - {\mu_i}} \right)}^2}} } \right]^{{ \frac{1}{2} }}} $$
$$ {\mu_i} = \left( {\frac{1}{n}} \right)\sum\nolimits_{{j = 1}}^n {{X_{{i,j}}}} $$

In Equations (2) to (4), subscripts “i” and “j” denote the variable and spectrum numbers, respectively, “σ i ” denotes the standard deviation of the variable number “i” in data matrix “X” with dimensions m × n (“m” denotes the number of variables (or m/z in post-IM/CID profiles in the present study) and “n” denotes the number of spectra (or IM scan numbers in the present study)); “μ i ” denotes the mean of the variable number “i”; “α” (or offset) denotes a constant value added to the mean of a variable “μ i ” to lower the purity value “p i,j ” for the μ i in the noise range (i.e., when the “μ i ” approaches zero) (typical values for “α” range from 0.5 to 10 % of the maximum of “μ i ”); “ω i,j ” denotes a determinant-based weight factor that corrects for the previously selected pure variables. For additional details on calculating “ω i,j ” please refer to the original report on SIMPLISMA [22].

The calculated purity values are represented in the form of purity spectra (i.e., plot of the purity values versus variables (e.g., Figure 4a)) and are used to guide the pure variable selection. The interactive process makes it possible to guide through the pure variable selection by changing the “α” value. The number of pure variables to be selected is prespecified and determined by the PCA.

After determining the pure variables, the original data matrix X m×n can be deconvoluted to the pure concentration profiles matrix (C n × k ) and component spectra matrix (S k × m ) as shown in Equation (5):
The intensities of pure variables in the original data matrix X and therefore C in Equation (5) are known. Hence, S can be calculated using least-squares method [25] and according to the Equation (6):
By normalizing the spectra in matrix S (obtained from Equation (6)) and calculating S norm. (i.e., the normalized spectra matrix), the concentration profiles can also be obtained by the least-squares method [25]:

3 Results and Discussion

In the following sections, we describe the details of data preprocessing and deconvolution platform using the post-IM/CID profiles collected for a two-component peptide mixture containing neurotensin (molecular weight (MW) = 816.50 Da) and a hexapeptide (WHWLQL with MW = 881.45 Da).

For the initial demonstration of the IM peak deconvolution approach, we selected the mixture of neurotensin and hexapeptide after a survey of IM profiles for several potential candidates as a model for a binary mixture. The neurotensin and hexapeptide were selected based on their seemingly indistinguishable IM profiles (viz., AT values of ~8.96 ms (FWHH ~1.24 ms) and ~9.72 ms (FWHH ~1.35 ms) for the two IM overlapped protonated peptides under our experimental conditions, see Figure S3a). Our criteria for choosing this demonstration model mixture were that the selected pair of molecular ions would have: (1) similar but not identical IM profiles, (2) distinguishable CID mass spectral patterns, and (3) different MWs (please refer to details provided in the relevant section). The selected mixture of neurotensin and hexapeptide was suitable as it satisfied all three criteria mentioned above.

The selection of species with different MW values for evaluation of our data analysis approach was important to account for different ionization efficiencies of the overlapped components and comparisons with the results of SIMPLISMA deconvolution. In other words, we wanted to consider the different ESI efficiencies for production of the two protonated peptides in the electrosprayed mixture by conducting separate experiments and isolation of each individual peptide in the quadrupole mass filter of IM-MS system. For example, the individually isolated single peptide experiments were used as references to evaluate the capability of our deconvolution technique for correctly calculating the TIMS profiles (in terms of peak area and AT values) for each peptide from the overlapped IM profiles.

To demonstrate the broader applicability of our approach, we also examined the data for deconvolution of the overlapped post-IM/CID MS profile of a two-component sugar mixture containing isobaric raffinose and maltotriose isomers (as discussed in the relevant section).

3.1 Post-IM/CID Mass Spectral Data Preprocessing Prior to Chemometric Analysis

The data preprocessing described in this section provides the details on preparing a post-IM/CID MS data matrix for subsequent chemometric data analysis. Figure 1a (solid line) shows the experimentally acquired post-IM/CID profile (expanded view between AT of 6.5 to 12.5 ms (bottom x-axis), corresponding to scan numbers between 72 and 139 (top x-axis)) from the ESI of a neurotensin and hexapeptide binary mixture. To acquire the post-IM/CID profile in Figure 1a (solid line), the protonated hexapeptide (m/z 882 Th) and neurotensin (m/z 817 Th) were simultaneously mass isolated in the quadrupole mass filter (located between the ion source and trap IM cell). Then, the CID mass spectral data were collected after mobility separation by increasing the electrical potential difference between the IM and transfer cells to −60 V.

To simultaneously isolate the two protonated peptides, the Synapt HDMS quadrupole mass filter was operated in the “profile” mode and programmed as follows: m/z 882 Th (dwell time 68 %), ramp (dwell time 20 %), m/z 817 Th (dwell time 12 %). Please note that these dwell time percentage values were selected based on the relative abundance information from the individual isolation of each protonated peptide (see relevant section for more details) to obtain post-IM/CID MS profiles with comparable intensities. Figure 1b shows the mass spectrum obtained by combining the CID mass spectral data across the IM profile (between AT of 6.5 and 12.5 ms) shown in Figure 1a (solid line). As expected, the combined CID mass spectrum in Figure 1b contains signals that correspond to fragment ions from both the protonated neurotensin (e.g., y 5 + (m/z 660 Th), [Neu + H – NH3]+ [m/z 800 Th]), and hexapeptide (e.g., b 2 + (m/z 324 Th), b 3 + (m/z 510 Th)) species.

Before chemometric analysis of the acquired post-IM/CID MS data, a few steps must be taken to assure appropriate data treatment. For example, the originally stored TOF mass spectral data via Synapt HDMS system software (at different scan numbers across the IM profiles) do not have identical number of m/z data points and, hence, are not appropriate for direct chemometric analysis. Therefore, we developed a Matlab user program to preprocess the post-IM/CID MS data prior to chemometric analysis. Scheme 1 shows the flow chart of the nine steps used in the data preprocessing platform (as explained in the following sections).

Prior to using the post-IM/CID MS data as input into the Matlab user program, the format of the data was converted from .raw (original data format from Synapt HDMS system) to .mzXML format [21] using massWolf 4.3.1 software [20] (step 1, Scheme 1). The .mzXML post-IM/CID MS data format was then used as input to the Matlab program (step 2, Scheme 1) for further data processing (steps 3–9, Scheme 1).

Data preprocessing, using the Matlab user program, started with the extraction of a mass spectrum with the maximum number of m/z data points (i.e., as a “reference” mass spectrum) from the original post-IM/CID MS data (steps 3 and 4, Scheme 1). The “reference” mass spectrum was re-sampled (steps 5–8, Scheme 1) at equal distances in m/z dimension to generate the list of the “reference” m/z values. The intensity value at each “reference” m/z number was obtained by one dimensional data interpolation in Matlab. Subsequently, all of the CID mass spectra collected across the post-IM/CID profile were re-sampled with identical number of data points at the “reference” m/z values. For example, Figure 2a shows the “reference” CID mass spectrum selected by the Matlab user program from the post-IM/CID profile of neurotensin/hexapeptide mixture (Figure 1a, solid line). Figure 2b shows the re-sampled “reference” CID mass spectrum (via collection of relative ion abundance information at equal distances of 0.006 Th in the m/z dimension) with a total number of 105,000 “reference” m/z data points between the m/z 100 Th and 900 Th. A close inspection of the two mass spectra in Figures 2a and 2b indicates successful reconstruction of the re-sampled mass spectrum in Figure 2b (i.e., containing identical number of peaks as in Figure 2a and with correct relative abundance values).
Figure 2

(a) Selected “reference” CID mass spectrum obtained by the analysis of post-IM/CID profile shown in Figure 1a (solid line) using the data analysis platform (Scheme 1). (b) Re-sampled “reference” CID mass spectrum obtained by using 105,000 m/z data points

The selection of optimum number of m/z data points is an important consideration for reconstruction of the post-IM/CID profile used for subsequent chemometric analysis. For example, under-sampling or over-sampling of the CID mass spectra can lead to inaccurate IM profiles. The number of “reference” m/z data points was optimized based on the following two comparisons: (1) similarity of the selected “reference” mass spectrum with the re-sampled “reference” mass spectrum and (2) similarity of the original post-IM/CID with the post-IM/CID obtained from the re-sampled mass spectra using “reference” m/z data points. In the current version of the data preprocessing platform, the above-mentioned criteria are checked by: (1) visual inspection of the graphical/numerical outputs of the Matlab user program and (2) calculating the root mean square error (RMSE) for the difference between the reconstructed post-IM/CID profile and the original post-IM/CID profile. The post-IM/CID profiles that match with their original post-IM/CID profiles with RMSE <5 are considered “similar” and the selected number of “reference” m/z data points for re-sampling (the CID mass spectra) is considered as “optimum” number of data points (NDP). It should be noted that small deviations from the “optimum” NDP (e.g., ± 1 %) does not change the RMSE value significantly and hence a range of acceptable NDPs (to within ± 1 % of the “optimum” value) can be used for this purpose.

To demonstrate the importance of m/z data point selection, we compared the originally acquired post-IM/CID profile for the peptide binary mixture with the post-IM/CID profile obtained after re-sampling of the CID mass spectra using three different number of data points.

The reconstructed post-IM/CID profile obtained by re-sampling the CID mass spectra with the “optimum” number of m/z data points of 105,000 is shown in Figure 1a with open circle (○) symbols. The reconstructed post-IM/CID profiles obtained by “under-sampling” (50,000 m/z data point) and “over-sampling” (200,000 m/z data points) the CID mass spectra yielded incorrect IM intensity profiles (shown in Figures S1).

For example, the calculated RMSE for re-sampling using 105,000 m/z data points was within the acceptable range of less than 5 (viz., RMSE ~1.5). On the other hand, over-sampling the CID mass spectra with a higher number of m/z data points (i.e., 200,000, Figure S1b) than the 105,000 data points resulted in a larger peak height and peak area than the original post-IM/CID profile (with the calculated RMSE of ~370.0). Therefore, 105,000 m/z data points were selected as the “optimum” number of “reference” m/z selection for reconstructing the post-IM/CID profile of the neurotensin and hexapeptide mixture. Subsequently, a data matrix with a dimension of 200 (number of scans across the post-IM/CID profile) by 105,000 (number of re-sampled m/z values) was generated (the final output (step 9) in the Scheme 1) and used for chemometric analysis.

3.2 Chemometric Analysis of the Post-IM/CID Profile from a Binary Peptide Mixture

In this section, we show that the preprocessed post-IM/CID MS data of the binary mixture of neurotensin and hexapeptide can be inspected for the presence or absence of the IM overlapping using PCA. Also, we present the results from deconvolution of the preprocessed post-IM/CID profile of the neurotensin and hexapeptide mixture using SIMPLISMA.

“Chemical Rank” Identification Using PCA

At the first glance and without any prior knowledge about the chemical sample compositions, the Gaussian looking peak shape of the IM profile in Figure 1a (solid line) suggests the presence of a single component. The number of overlapped chemical species (or “chemical rank”) in the preprocessed post-IM/CID profile in Figure 1a (profile shown with open circles (○)) can be estimated using PCA.

The results from two different chemical rank determination techniques [24], including the (a) plot of eigenvalues versus the PC numbers (Figure S2) and (b) Malinowski’s factor indicator (MFI) function (Figure S2) [26] confirmed the presence of two significant PCs in the PCA model of the reconstructed post-IM/CID profile in Figure 1a (profile shown with open circles (○)). The MFI is a chemical rank determination technique known to have a larger dynamic range and is more tolerant for using “noisy” experimental data [26]. The two significant PCs (i.e., PC #1 and #2) accounted for a total accumulative variance of ~97 % of the analyzed post-IM/CID MS data. The estimated chemical rank of “two” for the preprocessed post-IM/CID MS profile of the neurotensin and hexapeptide mixture conforms to the physical reality (i.e., the presence of two IM overlapping peptides in the original mixture).

The PCA score plot of the preprocessed post-IM/CID profiles (Figure 3) can also be used to inspect the presence or absence of mobility overlapping and the regions of IM overlapping for the preprocessed post-IM/CID profile of the binary mixture. The interpretation of the PCA score plots for the preprocessed post-IM/CID MS profiles is similar to the interpretation of the datascope plots introduced by Kvalheim et al. [27] for the overlapped spectroscopy data. Briefly, in the two-dimensional PCA score plots (e.g., PC #1 versus PC #2), the mobility data points that lie on the straight lines passing through the origin (coordinate [0, 0]) of the score plot represent the pure (i.e., no overlapping) IM regions. In contrast, the mobility data points that do not lie on the two straight lines passing through the origin of the score plots represent the regions of the IM overlapping. The labels on the data points in the PCA score plot in Figure 3 denote the scan numbers across the post-IM/CID profile (i.e., scan numbers on the top x-axis of Figure 1a). The added trend lines in Figure 3 (solid black lines) indicate the linear segments of the score plot (data points are shown as open circles in red). Thus, the score plots can be visually inspected for the presence or absence of mobility-overlapped regions in the preprocessed post-IM/CID profiles. For example, the three representative mobility data points (i.e., scan numbers 90, 106, and 120 in Figures 1a and 3) correspond to pure (scans #90 and #120) and overlapped (scan #106) mobility regions.
Figure 3

PCA score plot of the preprocessed post-IM/CID profile shown in Figure 1a (with open circles [○]). The labels on the data points denote the scan numbers. The significance of the straight solid lines in the score plot, as the regions corresponding to the “pure” mobility regions, is explained in the text

Deconvolution of the Post-IM/CID MS Profile from a Binary Peptide Mixture

For deconvolution of the IM profiles and the CID mass spectra, the number of overlapped components was determined by the PCA and used as an input for the SIMPLISMA program.

The first step in SIMPLISMA deconvolution of the mobility profiles is to determine the most “pure” m/z values for the different components of the mixture. To determine the purest m/z values, the purity values of all of the re-sampled m/z values across the post-IM/CID profile were calculated using Equation (2).

Figure 4a shows the two purity spectra (i.e., plot of purity values versus the re-sampled m/z values) obtained for the preprocessed post-IM/CID profile of the neurotensin and hexapeptide mixture. The purest m/z values for each of the overlapping components correspond to the peaks with the highest purity (p) value in each spectrum (purity spectra #1 and #2). The highest value in the purity spectrum #1 and #2 correspond to the peaks labeled with the star symbols (*) in Figure 4a (viz., m/z 324 and m/z 817 in the left and right panels, respectively).
Figure 4

(a) Left and right panels: the first two purity spectra calculated by SIMPLISMA. (b) SIMPLISMA-deconvoluted TIMS profiles (bottom panel) and CID mass spectra (top panels) of the protonated neurotensin and hexapeptide. The star symbols (*) in panel (a) denote the purest m/z values selected by SIMPLISMA for deconvolution

The purity spectra can also be used to select the species-specific fragment ions to obtain the selected ion mobility spectrum (SIMOS) profiles of the IM overlapped species. The above-mentioned approach for generating SIMOS is more efficient than the previously reported approach [11]. For example, in our approach, constructing the SIMOS profiles and finding species-specific fragment ions are not based on the manual/visual inspection of the CID spectra across the post-IM/CID profiles and can be achieved by inspecting the purity spectra (i.e., identifying the purest m/z or highest p values [22]).

The two purest m/z values selected from the purity spectra #1 and #2 in Figure 4a correspond to [M + H]+ of the neurotensin (i.e., m/z 817 Th) and b 2 + fragment of the hexapeptide (i.e., m/z 324 Th). We used intensity profiles of these species-specific fragment ions for IM deconvolution (using Equations (5) to (7)). Figure 4b (bottom) shows the SIMPLISMA-deconvoluted TIMS profiles for the protonated neurotensin (Figure 4b solid line with the maximum AT value of ~8.97 ms) and hexapeptide (Figure 4b dash line with the maximum AT value of ~9.74 ms). The SIMPLISMA-deconvoluted CID mass spectra of neurotensin and hexapeptide are also shown as insets in Figure 4b (top). As expected, the deconvoluted CID mass spectra for the two peptides look significantly different from each other and the identities of all fragment ions in Figure 4b (top), corresponding to each peptide, can be assigned. The deconvoluted CID mass spectra (Figure 4b, top) for the neurotensin and hexapeptide (from the binary mixture) are identical to their individual CID mass spectra acquired under similar experimental conditions (CID mass spectra included with the supplementary materials).

Please note that the “purity” and the “SIMPLISMA-deconvoluted” CID mass spectra are constructed for different purposes (using different equations). The former or the “purity” mass spectra show the variations in the purity values (using Equation (2)) as a function of m/z (e.g., purity spectra in Figure 4a). A higher purity value for a certain m/z value in the purity spectrum (e.g., the peak at m/z 324 in purity spectrum #1 of Figure 4a or the peak at m/z 817 in purity spectrum #2 of Figure 4a) indicates larger variations in the ion intensity, for that particular m/z value, across the IM profile (Figure 1a). However, the SIMPLISMA-deconvoluted CID mass spectra are simply the deconvoluted mass spectra (constructed by using Equation (6)) and should resemble the corresponding CID mass spectra for each individual species.

One of the unique capabilities of the presented post-IM/CID deconvolution approach is that one can obtain the correct TIMS profile (in terms of maximum AT and peak area) for the IM overlapped components. For example, we collected a second set of post-IM/CID MS profiles from the same two-component peptide mixture using individually isolated peptides and both sets of the experiments yielded statistically identical results. Table 1 shows the summary of the IM profile parameters including AT, FWHH, and area under the curve (AUC) for protonated neurotensin and hexapeptide from the individual and simultaneous isolation experiments. Based on the case 2 t-test (for n1 = n2 = 3) evaluation of our replicate examinations, the FWHH and AUC values for the protonated neurotensin and hexapeptide from the individual isolation experiments (n1 = 3 for each peptide) and SIMPLISMA-deconvoluted procedure (n2 = 3) matched at the 95 % confidence level (CL). Also, the AT differences for the two peptides from the SIMPLISMA-deconvoluted profiles and individual isolation experiments were statistically negligible for neurotensin (98 % CL) and hexapeptide (95 % CL).

Additional data analyses were carried out to confirm that when an incorrect chemical ranking was used as an input to SIMPLISMA program, the deconvolution process did not yield acceptable TIMS profiles and/or CID mass spectra. For instance, forcing the SIMPLISMA program to calculate two (or more) sets of TIMS profiles and CID mass spectra for a single-component post-IM/CID profile (e.g., protonated hexapeptide in the individual isolation experiment) resulted in identification of the original component plus an additional IM profile and associated CID mass spectrum containing negative intensities (with negative values larger than 1 %, which we set as a criteria to reject the deconvolution results). Similarly, the deconvolution resulted in negative values and erroneous TIMS distributions if a binary mixture were processed as having three or more components (e.g., 4, 10, and 20). Specifically, both TIMS profiles and CID mass spectra for the assumed third and/or fourth (or additional) components of the binary peptide mixture showed negative intensities confirming that the SIMPLISMA inputs were incorrectly assigned. Therefore, checking for negative intensities (in the TIMS profile and CID spectrum) was a helpful diagnostic tool to check and validate the deconvolution.

3.3 Deconvolution of the Post-IM/CID MS Profile from a Binary Isobaric Sugar Mixture

To demonstrate the applicability of the present data analysis approach for deconvolution of IM overlapped profiles of isobaric isomers/species, we used a mixture of raffinose (Raf) and maltotriose (Mal) sugars.

The negative ion mode ESI of the raffinose and maltotriose mixture generated [Raf – H] and [Mal – H] (at m/z 503 Th) with insufficient S/N ratios. However, in the presence of chloride ions (Cl) in the sample, the chlorine adduct forms of the two sugars (at m/z 539 Th) were generated with higher intensity than [Raf – H] and [Mal – H] species. No chloride ion was added to the chemical sample mixture, and these chlorine adduct ions were formed because the original samples and/or solvent (water) contained low level salt contaminations (e.g., NaCl and KCl estimated to be under micro-molar levels). Therefore, we took advantage of the enhanced S/N ratio and employed the “advantageous” chlorine adduct forms of raffinose ([Raf + Cl]) and maltotriose ([Mal + Cl]) for the post-IM/CID MS studies.

To acquire the post-IM/CID profiles of the chloride adduct forms of raffinose and maltotriose, the species at m/z 539 Th corresponding to [Raf + Cl]- and [Mal + Cl]- were mass isolated in the quadrupole mass filter. The mass isolated species were injected into the IM cell and CID experiments of the IM separated species were conducted by increasing the electrical potential difference between the IM and transfer cells to −35 V.

The acquired post-IM/CID profile (expanded view between AT of 3 to 11 ms) of the binary mixture of chlorine adduct forms of raffinose and maltotriose is shown in Figure 5a (solid line, labeled as “Original”). Figure 5b shows the mass spectrum obtained by combining the CID mass spectra collected across the post-IM/CID profile of the mixture of [Raf + Cl] and [Mal + Cl]. The post-IM/CID profile obtained by re-sampling the CID mass spectral data across the original post-IM/CID profile is also shown as overlaid profile with open circle (○) symbols in Figure 5a (labeled as “Preprocessed”). The “optimized” number of “reference” m/z data points (in the m/z range of 100 Th to 550 Th) to reconstruct the post-IM/CID profile shown in Figure 5a was 80,000. The resulting 200 (number of scan across the mobility profile) by 80,000 (number of re-sampled “reference” m/z data points) preprocessed data matrix was used to deconvolute the overlapped IM profiles of the two-component sugar mixture.
Figure 5

(a) Overlaid view of the experimentally obtained (solid line) and re-sampled (using 80,000 m/z data points, in the m/z range of 100 Th-550 Th) (open circles [○]) post-IM/CID profiles of the mass isolated chlorine adduct forms of raffinose (Raf) ([Raf + Cl]) and maltotriose (Mal) ([Mal + Cl]). (b) Mass spectrum obtained by combining the CID mass spectra across the post-IM/CID profile shown in panel (a) (with solid line)

The inspection of the preprocessed post-IM/CID profile of the mixture of [Raf + Cl] and [Mal + Cl] using PCA (i.e., eigenvalue and Malinowski’s factor indictor plots, Figure S4) revealed the presence of two significant PCs with total accumulative variance of ~95 %.

The calculated purity spectra (not shown), using the preprocessed post-M/CID profile shown in Figure 5a (profile shown with open circle (○)), indicated the presence of m/z values of 383 Th and 503 Th as the two purest variables for the two IM overlapped components. The identities of the two selected pure m/z values were X 2 (m/z 383 Th) and [Mal(or Raf) – H] (m/z 503 Th, loss of HCl) fragments from [Mal(or Raf) + Cl].

The SIMPLISMA deconvolution using the two purest m/z values (i.e., m/z 383 Th and 503 Th) for sugar compounds resulted in the two TIMS profiles shown in Figure 6 (bottom). The SIMPLISMA-deconvoluted CID mass spectra corresponding to the two deconvoluted TIMS profiles are also shown in Figure 6 (top). The identities of the two deconvoluted TIMS profiles in Figure 6 were assigned by comparing the deconvoluted AT values (~6.65 ms for [Raf + Cl] and ~7.14 ms for [Mal + Cl]) with IM profiles acquired for the chemical samples containing either pure raffinose or pure maltotriose (~6.60 ms for [Raf + Cl] and ~7.08 ms for [Mal + Cl], Figure S5a). Likewise, the identities of the CID mass spectra shown in Figure 6 insets were assigned based on their similarities to the CID mass spectra of each individual sugar (i.e., [Raf + Cl] or [Mal + Cl], Figure S5b).
Figure 6

SIMPLISMA-deconvoluted TIMS profiles (bottom) and CID mass spectra (top) of [Raf + Cl] and [Mal + Cl] obtained from the deconvolution of the post-IM/CID profile shown in Figure 5a (open circles [○]). The star symbols (*) in the CID mass spectra (top) denote the purest m/z values selected by SIMPLISMA for deconvolution

Note that the purpose of acquiring post-IM/CID profiles (e.g., Figure S5a) was to obtain the AT values and CID mass spectra of the chlorine adduct forms of raffinose and maltotriose for the evaluation of SIMPLISMA results. Although the relative ion abundances of each chlorine adduct sugar species (i.e., [Raf + Cl] and [Mal + Cl]) can be retrieved from the deconvoluted TIMS, more accurate relative ion comparisons will require prior knowledge about other relevant parameters such as ionization efficiencies. In other words, because the ionization efficiencies of the two sugar species when present in the solution as a single component versus in the mixture may not be identical, the peak areas of the IM profiles in Figure 6 cannot be used for quantitative comparisons.

3.4 Limitations of the Data Analysis Approach for IM Deconvolution

The first step in using SIMPLISMA for deconvolution of the overlapped mobility profiles is to extract the purest CID fragment ions for each overlapping component. The purest fragments are then used to obtain the pure IM profiles and CID mass spectra for each of the overlapped components. Although the data analysis approach presented here offers several advantages for TIMS profile and CID mass spectral deconvolutions, there are two limitations associated with the approach.

The first limitation is related to the selection of the pure m/z values for the mobility overlapped components. According to the Equations (2) to (4), successful selection of the pure m/z values depends strongly on the standard deviation (σ i ) of the intensity values (of each m/z) across the mobility profiles. For an m/z value to be selected as a “pure variable,” the intensity values of the target m/z must vary (across IM profile) in such a way that the calculated “σ” value for the selected m/z is larger than the “σ” value(s) for the other m/z values in the background noise. At collision energies in which the CID fragmentation patterns of the overlapped conformers do not vary significantly across the IM profiles, SIMPLISMA cannot be used to calculate TIMS profiles or perform CID mass spectral deconvolution. This limitation can be addressed by: (1) chemometric analyses of post-IM/CID profiles collected at different collision energies and/or (2) utilizing other post-IM fragmentation techniques such as ETD or photodissociation (which may generate more isomer/conformation specific fragment ions).

The second limitation is related to the percentage of the mobility overlap in the post-IM/CID profiles. The “overlap percentage” refers to the percentage of the data points across a post-IM/CID profile that contains fragment ions related to all the IM overlapped components. For example, a 100 % value for the “overlap percentage” of a binary mixture means that each data point across the overlapped post-IM/CID profile contains fragment ions from both species in the binary mixture. Conversely, a 0 % value for the “overlap percentage” means that the two components present in the mixture have two completely resolved post-IM/CID profiles. The current approach cannot be utilized for the deconvolution of overlapped post-IM/CID profiles with 100 % mobility overlapping. If there are no CID fragment ions with variable intensities across the post-IM/CID MS profile, both XFIDTD and SIMPLISMA will fail to deconvolute the mobility profiles. We are exploring the possibility of using energy-resolved CID to address the deconvolution of totally overlapped IM profiles.

With the above-mentioned requirements, SIMPLISMA combined with the post-IM fragmentation should allow deconvolution of the IM overlapped profiles containing more than two components.

4 Conclusions

A data analysis approach, based on post-IM/CID MS combined with chemometrics, can be used for deconvolution of unresolved IM profiles. The utility of the combined use of the post-IM/CID MS and chemometrics for IM profiles and CID mass spectral deconvolution was successfully demonstrated using (a) a binary non-isobaric peptide mixture and (b) a two-component isobaric sugar mixture using a Synapt HDMS IM-MS system with IM resolving power of <10. The new approach provides the capability to detect the presence of IM overlapping and calculate the pure TIMS profiles and CID mass spectra of the IM overlapped components. The two independent chemometric techniques, i.e., PCA and SIMPLISMA, yield complementary results and add to the validity of the present data analysis approach.

The IM deconvolution approach can be extended and used for the quantitative analysis of complex chemical mixtures in x-omics studies such as metabolomics, petroleomics, and proteomics/peptidomics.



The authors thank Baylor University for financial support. They thank Mahsan Miladi for her assistance in data acquisition and Dr. Christopher Becker for helpful technical discussions on IM-MS, and providing the sugar samples used in this study.

Supplementary material

13361_2012_471_MOESM1_ESM.docx (141 kb)
ESM 1 (DOCX 140 kb)


  1. 1.
    Szulejko, J.E., Solouki, T.: Potential analytical applications of interfacing a GC to an FT-ICR MS: fingerprinting complex sample matrixes. Anal. Chem. 74, 3434–3442 (2002)CrossRefGoogle Scholar
  2. 2.
    Solouki, T., Szulejko, J.E., Bennett, J.B., Graham, L.B.: A preconcentrator coupled to a GC/FTMS: advantages of self-chemical ionization, mass measurement accuracy, and high mass resolving power for GC applications. J. Am. Soc. Mass Spectrom. 15, 1191–1200 (2004)CrossRefGoogle Scholar
  3. 3.
    Tipton, J.D., Tran, J.C., Catherman, A.D., Ahlf, D.R., Durbin, K.R., Lee, J.E., Kellie, J.F., Kelleher, N.L., Hendrickson, C.L., Marshall, A.G.: Nano-LC FTICR tandem mass spectrometry for top-down proteomics: routine baseline unit mass resolution of whole cell lysate proteins up to 72 kDa. Anal. Chem. 84, 2111–2117 (2012)CrossRefGoogle Scholar
  4. 4.
    Zinnel, N.F., Pai, P.J., Russell, D.H.: Ion mobility-mass spectrometry (IM-MS) for top-down proteomics: increased dynamic range affords increased sequence coverage. Anal. Chem. 84, 3390–3397 (2012)CrossRefGoogle Scholar
  5. 5.
    Ruotolo, B.T., Benesch, J.L., Sandercock, A.M., Hyung, S.J., Robinson, C.V.: Ion mobility-mass spectrometry analysis of large protein complexes. Nat. Protoc. 3, 1139–1152 (2008)CrossRefGoogle Scholar
  6. 6.
    Badman, E.R., Myung, S., Clemmer, D.E.: Gas-phase separations of protein and peptide ion fragments generated by collision-induced dissociation in an ion trap. Anal. Chem. 74, 4889–4894 (2002)CrossRefGoogle Scholar
  7. 7.
    Fernandez-Lima, F.A., Becker, C., McKenna, A.M., Rodgers, R.P., Marshall, A.G., Russell, D.H.: Petroleum crude oil characterization by IMS-MS and FTICR MS. Anal. Chem. 81, 9941–9947 (2009)CrossRefGoogle Scholar
  8. 8.
    Becker, C., Qian, K., Russell, D.H.: Molecular weight distributions of asphaltenes and deasphaltened oils studied by laser desorption ionization and ion mobility mass spectrometry. Anal. Chem. 80, 8592–8597 (2008)CrossRefGoogle Scholar
  9. 9.
    Zucker, S.M., Lee, S., Webber, N., Valentine, S.J., Reilly, J.P., Clemmer, D.E.: An ion mobility/ion trap/photodissociation instrument for characterization of ion structure. J. Am. Soc. Mass Spectrom. 22, 1477–1485 (2011)CrossRefGoogle Scholar
  10. 10.
    Giles, K., Williams, J.P., Campuzano, I.: Enhancements in traveling wave ion mobility resolution. Rapid Commun. Mass Spectrom. 25, 1559–1566 (2011)CrossRefGoogle Scholar
  11. 11.
    Lee, S., Li, Z., Valentine, S.J., Zucker, S.M., Webber, N., Reilly, J.P., Clemmer, D.E.: Extracted fragment ion mobility distributions: A new method for complex mixture analysis. Int. J. Mass Spectrom. 309, 154–160 (2012)Google Scholar
  12. 12.
    Creese, A.J., Cooper, H.J.: Separation and identification of isomeric glycopeptides by high field asymmetric waveform ion mobility spectrometry. Anal. Chem. 84, 2597–2601 (2012)CrossRefGoogle Scholar
  13. 13.
    Xuan, Y., Creese, A.J., Horner, J.A., Cooper, H.J.: High-field asymmetric waveform ion mobility spectrometry (FAIMS) coupled with high-resolution electron transfer dissociation mass spectrometry for the analysis of isobaric phosphopeptides. Rapid Commun. Mass Spectrom. 23, 1963–1969 (2009)CrossRefGoogle Scholar
  14. 14.
    Riba-Garcia, I., Giles, K., Bateman, R.H., Gaskell, S.J.: Evidence for structural variants of a- and b-type peptide fragment ions using combined ion mobility/mass spectrometry. J. Am. Soc. Mass Spectrom. 19, 609–613 (2008)CrossRefGoogle Scholar
  15. 15.
    Riba Garcia, I., Giles, K., Bateman, R.H., Gaskell, S.J.: Studies of peptide a- and b-type fragment ions using stable isotope labeling and integrated ion mobility/tandem mass spectrometry. J. Am. Soc. Mass Spectrom. 19, 1781–1787 (2008)CrossRefGoogle Scholar
  16. 16.
    Polfer, N.C., Bohrer, B.C., Plasencia, M.D., Paizs, B., Clemmer, D.E.: On the dynamics of fragment isomerization in collision-induced dissociation of peptides. J. Phys. Chem. A 112, 1286–1293 (2008)CrossRefGoogle Scholar
  17. 17.
    Hoaglund-Hyzer, C.S., Li, J., Clemmer, D.E.: Mobility labeling for parallel CID of ion mixtures. Anal. Chem. 72, 2737–2740 (2000)CrossRefGoogle Scholar
  18. 18.
    Badman, E.R., Hoaglund-Hyzer, C.S., Clemmer, D.E.: Dissociation of different conformations of ubiquitin ions. J. Am. Soc. Mass Spectrom. 13, 719–723 (2002)CrossRefGoogle Scholar
  19. 19.
    Wyttenbach, T., Kemper, P.R., Bowers, M.T.: Design of a new electrospray ion mobility mass spectrometer. Int. J. Mass Spectrom. 212, 13–23 (2001)CrossRefGoogle Scholar
  20. 20.
    massWolf4.3.1. Available at: Last accessed March 05, 2012
  21. 21.
    Pedrioli, P.G., Eng, J.K., Hubley, R., Vogelzang, M., Deutsch, E.W., Raught, B., Pratt, B., Nilsson, E., Angeletti, R.H., Apweiler, R., Cheung, K., Costello, C.E., Hermjakob, H., Huang, S., Julian, R.K., Kapp, E., McComb, M.E., Oliver, S.G., Omenn, G., Paton, N.W., Simpson, R., Smith, R., Taylor, C.F., Zhu, W., Aebersold, R.: A common open representation of mass spectrometry data and its application to proteomics research. Nat. Biotechnol. 22, 1459–1466 (2004)CrossRefGoogle Scholar
  22. 22.
    Windig, W., Guilment, J.: Interactive self-modeling mixture analysis. Anal. Chem. 63, 1425–1432 (1991)CrossRefGoogle Scholar
  23. 23.
    Abdi, H., Williams, L.J.: Principal component analysis. WIREs Comp. Stat. 2, 433–459 (2010)CrossRefGoogle Scholar
  24. 24.
    Meloun, M., Capek, J., Mikšík, P., Brereton, R.G.: Critical comparison of methods predicting the number of components in spectroscopic data. Anal. Chim. Acta 423, 51–68 (2000)CrossRefGoogle Scholar
  25. 25.
    Kowalski, B.R., Reebe, K.R.: An introduction to multivariate calibration and analysis. Anal. Chem. 59, 1007A–1017A (1987)CrossRefGoogle Scholar
  26. 26.
    Malinowski, E.R.: Determination of the number of factors and the experimental error in a data matrix. Anal. Chem. 49, 612–617 (1977)CrossRefGoogle Scholar
  27. 27.
    Kvalheiml, O.: M.; Liang, Y.-Z.: Heuristic evolving latent projections: resolving two-way multicomponent data. 1. Selectivity, latent-projective graph, datascope, local rank, and unique resolution. Anal. Chem. 64, 936–946 (1992)CrossRefGoogle Scholar

Copyright information

© American Society for Mass Spectrometry 2012

Authors and Affiliations

  1. 1.Department of Chemistry and BiochemistryBaylor UniversityWacoUSA

Personalised recommendations