CHROMATOGRAPH MASS SPECTROMETER

Abstract

In order to appropriately set MS.sup.m analysis conditions, an MS.sup.m-1 analysis executer (51) makes a mass spectrometer (20) perform an MS.sup.m-1 analysis (where m is an integer from 2 to n) to acquire three-dimensional data showing an intensity for each of the N m/z values and each of the M retention times (where N and M are natural numbers). Based on the three-dimensional data, a data matrix creator (41) creates data matrix X in which intensity data are arranged in N rows which differ from each other in m/z value and M columns which differ from each other in retention-time value. A matrix factorization executer (42) determines an N×K spectrum matrix S and K×M profile matrix P (where K is a natural number) by matrix factorization based on data matrix X so that this matrix X is approximated by product SP of the matrices S and P. An m/z detector (43) detects m/z of a precursor ion originating from a sample component from the values of the matrix elements in each column of matrix S. A retention time detector (44) detects the retention time of a sample component from the values of the matrix elements in each row of matrix P. Based on the m/z and retention time, an MS.sup.m analysis execution condition determiner (45) determines an execution condition of an MS.sup.m analysis including the selection and fragmentation of a precursor ion of a sample component. An MS.sup.m analysis executer (52) makes the mass spectrometer execute an MS.sup.m analysis based on the execution condition.

Claims

1. A chromatograph mass spectrometer in which a chromatograph configured to temporally separate a sample into components is combined with a mass spectrometer having a function of an MS.sup.n analysis (where n is an integer equal to or greater than 2) in which each component in the sample separated by the chromatograph is subjected to selection and fragmentation of an ion at least one time, and ions resulting from the fragmentation are subjected to mass spectrometry, the chromatograph mass spectrometer comprising: an MS.sup.m-1 analysis executer configured to make the mass spectrometer perform an MS.sup.m-1 analysis (where m is an integer from 2 to n, inclusive) to acquire three-dimensional data showing an intensity for each of N m/z values (where N is a natural number) and each of M retention times (where M is a natural number); a data matrix creator configured to create, based on the three-dimensional data, a data matrix X in which intensity data are arranged in N rows and M columns or M rows and N columns, where the N rows or N columns of intensity data differ from each other in a value of the m/z while the M columns or M rows of data differ from each other in a value of the retention time; a matrix factorization executer configured to determine a spectrum matrix S and a profile matrix P by a technique of matrix factorization based on the data matrix X so that the data matrix X is approximated by a product SP in which the spectrum matrix S has N rows and K columns (where K is a natural number) while the profile matrix P has K rows and M columns, or by the product PS in which the spectrum matrix S has K rows and N columns while the profile matrix P has M rows and K columns; an m/z detector configured to detect the m/z of a precursor ion originating from a component contained in the sample, from values of matrix elements in each column or each row of the spectrum matrix S; a retention time detector configured to detect the retention time of a component contained in the sample, from values of matrix elements in each row or each column of the profile matrix P; an MS.sup.m analysis execution condition determiner configured to determine, based on the m/z and the retention time, an execution condition of an MS.sup.m analysis including selection and fragmentation of a precursor ion of a component contained in the sample; and an MS.sup.m analysis executer configured to make the mass spectrometer execute an MS.sup.m analysis based on the execution condition.

2. The chromatograph mass spectrometer according to claim 1, wherein the MS.sup.m analysis execution condition determiner is further configured to perform an operation, based on previously acquired data concerning a background, for removing a precursor ion corresponding to a combination of an m/z candidate and a retention-time candidate originating from the background, from a target for which the execution condition for the MS.sup.m analysis should be determined.

3. The chromatograph mass spectrometer according to claim 1, wherein the MS.sup.m analysis execution condition determiner is further configured to perform an operation, based on previously acquired data concerning a background, for setting, as a target for which the execution condition for the MS.sup.m analysis should be determined, a precursor ion corresponding to a combination of an m/z candidate and a retention-time candidate that fall within an m/z range and retention-time range which are free from an influence of the background.

4. The chromatograph mass spectrometer according to claim 1, wherein the MS.sup.m analysis execution condition determiner is further configured to perform an operation of creating divisional analysis methods by dividing an analysis method for an MS.sup.m analysis of one sample into a plurality of analysis methods so that a loop time for one MS.sup.m analysis will be equal to or less than a predetermined value.

5. The chromatograph mass spectrometer according to claim 1, wherein the MS.sup.m analysis execution condition determiner is further configured to perform an operation of setting a different level of collision energy for each component to be analyzed.

6. The chromatograph mass spectrometer according to claim 1, wherein the MS.sup.m analysis execution condition determiner is further configured to perform an operation of setting a plurality of levels of collision energy for each component to be analyzed.

7. The chromatograph mass spectrometer according to claim 1, wherein the matrix factorization executer includes: a regularization parameter-regularization function preparer configured to prepare a plurality of regularization-parameter candidates λr (where r is a natural number from 1 to r.sub.max) and one regularization function R(S, P) which induces sparsity of a solution; a matrix candidate determiner configured to solve an optimization problem for each of the plurality of regularization-parameter candidates λr so as to determine a matrix Srt as a spectrum matrix candidate Sr which is a candidate of the spectrum matrix S and a matrix Prt as a profile matrix candidate Pr which is a candidate of the profile matrix P, where the matrices Srt and Prt are determined so as to minimize a value of a loss function L(S, P)=D(X|SP)+λrR(S, P), where D(X|SP) is a distance function expressing a degree of difference between the data matrix X and the product SP, while λrR(S, P) is a product of the regularization-parameter candidate λr and the regularization function R(S, P); a probability distribution transformer configured to determine, for each of the plurality of regularization-parameter candidates λr, a transformed value y.sub.nm=F.sub.nm(X.sub.nm|(SrPr).sub.nm) by a variable transform into a common probability distribution Pcommon for each combination of a matrix element X.sub.nm of the data matrix X and a corresponding matrix element (SrPr).sub.nm of a product SrPr of the spectrum matrix candidate Sr and the profile matrix candidate Pr, using F.sub.nm which is a function for the variable transform from a probability distribution P.sub.nm corresponding to the distance function D(X.sub.nm|(SP).sub.nm) into the common probability distribution Pcommon; a goodness-of-fit calculator configured to determine, for each of the plurality of regularization-parameter candidates λr, a goodness of fit between the transformed value y.sub.nm and a cumulative distribution function of the probability distribution Pcommon; and a matrix determiner configured to select, as the spectrum matrix S and the profile matrix P, the spectrum matrix candidate Sr and the profile matrix candidate Pr determined for a regularization-parameter candidate λr which yields a highest value of the goodness of fit among the plurality of regularization-parameter candidates λr, or a regularization-parameter candidate λr which yields the goodness of fit higher than a predetermined threshold and also has a largest value of λr.

8. The chromatograph mass spectrometer according to claim 7, wherein the matrix candidate determiner is configured to use the matrix Srt and the matrix Prt as initial values for determining a matrix Srt2 and a matrix Prt2 which minimize a value of a second loss function with no regularization term, L.sub.2(S, P)=D(X|SP), and to select the matrix Srt2 and the matrix Prt2 as the spectrum matrix candidate Sr and the profile matrix candidate Pr, instead of selecting the matrix Srt and the matrix Prt as the spectrum matrix candidate Sr and the profile matrix candidate Pr.

9. The chromatograph mass spectrometer according to claim 7, wherein the regularization term R(S, P) is L1-norm, or a linear combination of L1-norm and L2-norm, or a function which applies a trace norm, det|S.sup.TS| or log det|S.sup.TS+δI| to the matrix S (where I is a unit matrix, and δ is a hyperparameter for controlling the regularization function) while placing a constraint on a solution so that a total of values in each column of the matrix P should not exceed 1.

10. The chromatograph mass spectrometer according to claim 7, wherein the cumulative distribution function F(X|Y) is a function selected from a cumulative distribution function calculated from a function expressing a Gaussian distribution, a cumulative distribution function calculated from a function expressing a Poisson distribution, a cumulative distribution function calculated from a function expressing an exponential distribution, and a cumulative distribution function calculated from a function expressing a Tweedie distribution.

11. The chromatograph mass spectrometer according to claim 7, wherein the goodness-of-fit calculator is configured to calculate the goodness of fit by a test selected from a Kolmogorov-Smimov test, a Cramer-von Mises test, and an Anderson-Darling test.

12. The chromatograph mass spectrometer according to claim 7, wherein, when a variance σnm.sup.2 of noise in each matrix element X.sub.nm is previously known, the goodness-of-fit calculator defines the probability distribution Pcommon as a standard normal distribution, defines the cumulative distribution function F.sub.nm(X.sub.nm|(SrPr).sub.nm) as (X.sub.nm−(SrPr).sub.nm)/δ.sub.nm, and uses −|σ.sub.y.sup.2−1| as the goodness of fit, where σ.sub.y.sup.2 is given by: ${\sigma }_y^2=\frac{1}{\mathrm{NM}-1}{.Math.}_{n,m}{y}_{\mathrm{nm}}^2$ which is an unbiased variance of the transformed value y.sub.nm whose mean value is assumed to be zero.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0036] FIG. 1 is a schematic configuration diagram showing one embodiment of the chromatograph mass spectrometer according to the present invention.

[0037] FIG. 2 is a flowchart showing an operation of the chromatograph mass spectrometer according to the present embodiment.

[0038] FIG. 3 is a diagram conceptually illustrating a data matrix as well as a spectrum matrix and profile matrix, using one example of the three-dimensional data as well as the data of mass spectra and chromatograms.

[0039] FIG. 4 is a flowchart showing details of the operation for matrix factorization in the operation of the chromatograph mass spectrometer according to the present embodiment.

[0040] FIG. 5 is a diagram showing measurement data in the form of chromatograms and mass spectra which give the data matrix used in an example of the calculation of the matrix factorization performed in the chromatograph mass spectrometer according to the present embodiment.

[0041] FIG. 6 is a diagram in which measurement data which give a data matrix is shown as a superposition of a plurality of mass chromatograms.

[0042] FIG. 7 is a diagram showing an example of the result of a calculation of mass spectra and chromatograms obtained in a matrix factorization in which the regularization is insufficient due to λr being too small.

[0043] FIG. 8 is a diagram showing an example of the result of a calculation of mass spectra and chromatograms obtained in a matrix factorization in which the regularization is insufficient due to λr being too large.

[0044] FIG. 9 is a diagram showing an example of the result of a calculation of mass spectra and chromatograms obtained in a matrix factorization in which the regularization is appropriately performed using an optimum value of λr.

DESCRIPTION OF EMBODIMENTS

[0045] One embodiment of the chromatograph mass spectrometer according to the present invention is hereinafter described using FIGS. 1-9.

(1) Configuration of Chromatograph Mass Spectrometer According to Present Embodiment

[0046] FIG. 1 shows the configuration of the main components of a liquid chromatograph/ion trap time-of-flight mass spectrometer (LC/IT-TOFMS) 1 which is an embodiment of the present invention. This LC/IT-TOFMS 1 is roughly divided into a liquid chromatograph (LC) unit 10, mass spectrometry (MS) unit 20, data processing unit 40 and analysis control unit 50.

[0047] The LC unit 10 includes a mobile phase container 11, liquid supply pump 12, injector 13, and column 14. The mobile phase container 11 is used for storing a mobile phase. The liquid supply pump 12 is configured to draw the mobile phase from the mobile phase container 11 and supply it to the injector 13 at a constant flow rate. The injector 13, which includes an autosampler, is configured to automatically select one of the prepared samples and injects a predetermined volume of the sample into the mobile phase at a predetermined timing. When a sample is injected from the injector 13 into the mobile phase, the sample is carried by the mobile phase and introduced into the column 14. While the sample is passing through the column 14, the various components in the sample are separated from each other and exit from the outlet end of the column 14 in a temporally separated form, to be introduced into the MS unit 20.

[0048] The MS unit 20 includes an ionization chamber 21 to be maintained at atmospheric pressure, and an analysis chamber 29 to be maintained at a high degree of vacuum by being evacuated by a turbo molecular pump (not shown). A first-stage intermediate vacuum chamber 24 and second-stage intermediate vacuum chamber 27, with their degrees of vacuum increased in a stepwise manner, are provided between the ionization chamber 21 and analysis chamber 29. The ionization chamber 21 communicates with the first-stage intermediate vacuum chamber 24 through a thin desolvation tube 23. The first-stage intermediate vacuum chamber 24 communicates with the second-stage intermediate vacuum chamber 27 through an orifice of a small diameter bored at the apex of a conical skimmer 26. A first ion guide 25 and second ion guide 28 are arranged within the first-stage intermediate vacuum chamber 24 and second-stage intermediate vacuum chamber 27, respectively.

[0049] The ionization chamber 21 is equipped with an ESI nozzle 22 as the ion source. The ESI nozzle 22 is configured to be supplied with an eluate containing sample components from the LC unit 10 and spray the eluate into the ionization chamber 21 in the form of droplets while electrically charging the droplets by a high DC voltage applied from a high voltage source (not shown). The electrically charged droplets collide with gas molecules of atmospheric origin and are thereby broken into even smaller droplets, which are quickly dried (desolvated), leaving sample molecules in a gas state. Those sample molecules are ionized through ion evaporation. The droplets containing the resultant ions are drawn into the desolvation tube 23 by the pressure difference between the ionization chamber 21 and the first-stage intermediate vacuum chamber 24. While passing through the desolvation tube 23, the droplets further undergo desolvation and produce more ions. It should be noted that the method for ionizing sample molecules is not limited to the electrospray ionization (ESI) described in this paragraph; for example, an atmospheric pressure chemical ionization (APCI) or atmospheric pressure photoionization (APPI) can also be used.

[0050] The ions which have passed through the desolvation tube 23 travel through the first-stage and second-stage intermediate vacuum chambers 24 and 27 while being converged by the first and second ion guides 25 and 28, and are sent into the analysis chamber 29.

[0051] The analysis chamber 29 contains an ion trap 30, time-of-flight mass separator (TOF) 31 as the mass separator, and ion detector 33.

[0052] Within the ion trap 30, the ions are temporarily captured and accumulated by a quadrupole electric field created by radio-frequency voltages respectively applied from a power source (not shown) to the electrodes. The various ions accumulated within the ion trap 30 are simultaneously given kinetic energy at a predetermined timing and thereby ejected from the ion trap 30 into the TOF 31.

[0053] Additionally, as shown in FIG. 1, the ion trap 30 can be supplied with a collision induced dissociation (CID) gas, such as argon. This allows the ions accumulated within the ion trap 30 to be fragmented into product ions by CID. In the case of an MS.sup.2 analysis, after the various ions have been accumulated within the ion trap 30, the voltages applied to the electrodes are controlled so that an ion having a specific m/z among those ions will be selectively retained as a precursor ion. The CID gas is subsequently introduced into the ion trap 30 to promote the fragmentation of the precursor ion. The resultant product ions are simultaneously ejected from the ion trap 30 toward the TOF 31 at a predetermining timing.

[0054] The TOF 31 includes a reflectron electrode 32 to which a DC voltage is applied from a DC power source (not shown). Due to the effect of the thereby created DC electric field, the ions are returned and reach the ion detector 33. Among the ions which have been simultaneously ejected from the ion trap 30, an ion having a smaller m/z flies at a higher speed. Consequently, the ions separately reach the ion detector 33, having temporal differences according to their m/z values. The ion detector 33 produces, as a detection signal, an electric current corresponding to the number of ions arriving at the detector.

[0055] An analogue-to-digital (A/D) converter 34 for converting the detection signal into a digital value is connected to the ion detector 33. After the conversion by the A/D converter 34, the detection signal is sent to the data processing unit 40.

[0056] The data processing unit 40 includes a data matrix creator 41, matrix factorization executer 42, m/z detector 43, retention time detector 44, and MS.sup.2 analysis execution condition determiner 45 (which corresponds to the MS.sup.m analysis execution condition determiner with m=2). The matrix factorization executer 42 includes a regularization parameter-regularization function preparer 421, matrix candidate determiner 422, probability distribution transformer 423, goodness-of-fit calculator 424, and matrix determiner 425. Details of those components will be described later. The data processing unit 40 is connected to a storage unit 61.

[0057] The analysis control unit 50 is configured to control the components of the LC unit 10 and MS unit 20 so as to perform an LC/MS analysis and LC/MS.sup.2 analysis. It includes an LC/MS analysis executer 51 (which corresponds to the MS.sup.m-1 analysis executer with m=2) and LC/MS.sup.2 analysis executer 52 (which corresponds to the MS.sup.m analysis executer with m=2).

[0058] The data processing unit 40 and analysis control unit 50 are embodied by a personal computer (PC) on which predetermined controlling-and-processing software is installed. The storage unit 61 is embodied by a hard disk drive, solid state drive or other types of storage devices provided in or for the PC. The PC also has a display unit 62 as well as an operation unit 63 including a keyboard, mouse, touch panel and/or other devices.

(2) Operation of Chromatograph Mass Spectrometer (LC/IT-TOFMS) According to Present Embodiment

[0059] An operation of the LC/IT-TOFMS 1 according to the present embodiment is hereinafter described using FIGS. 2 and 3. The functions of the components in the data processing unit 40 will also be described.

[0060] Initially, an operator using the operation unit 63 performs a predetermined operation to initiate a measurement. In response to this operation, the LC/MS analysis executer 51 in the LC/IT-TOFMS 1 begins to control the components of the LC/IT-TOFMS 1 to conduct an LC/MS analysis for a target sample, as will be described later (Step 1). The target sample injected from the injector 13 into the mobile phase is thereby sent into the column 14, and the eluate from the column 14 is introduced into the MS unit 20, which repeatedly performs a mass spectrometric analysis of the eluate. The detection signals produced by the ion detector 33 in the MS unit 20 are converted into digital values by the A/D converter 34 and sent to the data matrix creator 41 in the data processing unit 40.

[0061] In the data matrix creator 41, N signals obtained at each m/z within a predetermined m/z range as a result of one ejection of ions from the ion trap 30 are acquired as the values of N matrix elements to be included in one column of a data matrix X with N rows and M columns. The data acquisition is similarly performed for each of the M ejections of ions performed at intervals of time. Based on those data, the data matrix X with N rows and M columns as shown below is created (Step 2).

[0062] Each matrix element X.sub.nm of the data matrix X (where n is an integer from 1 to N, while m is an integer from 1 to M) indicates the intensity detected at the n-th m/z within the aforementioned m/z range as well as at the m-th ion ejection (which corresponds to the retention time). Each matrix element X.sub.nm of the data matrix X has a value of zero or positive value (non-negative value).

[0063] Next, the matrix factorization executer 42 performs a matrix factorization by a method which will be described later (in “(3) Details of Data Analysis Method (Operation of Matrix Factorization) According to Present Embodiment”) to determine a spectrum matrix S with N rows and K columns as well as a profile matrix (also called a “chromatogram matrix”) P with K rows and M columns so that their product SP approximates to the data matrix X (Step 3). The spectrum matrix S and profile matrix P can be expressed as follows:

[0064] Each matrix element s.sub.nk of the spectrum matrix S (where n is an integer from 1 to N, while k is an integer from 1 to K) indicates the intensity at one m/z value in a mass spectrum originating from one of the K kinds of components contained in a sample (this component is hereinafter called the “k-th component”). Similarly, each matrix element p.sub.km of the profile matrix P indicates the intensity at one retention time in the chromatogram originating from the k-th component. In other words, each set of matrix elements surrounded by the broken line in the above spectrum matrix S shows a mass spectrum of one component, while each set of matrix elements surrounded by the broken line in the above profile matrix P shows a chromatogram of one component. Each of the matrix elements s.sub.nk of the spectrum matrix S and the matrix elements p.sub.km of the profile matrix P has a value of zero or positive value (non-negative value). FIG. 3 conceptually illustrates the data matrix X as well as the spectrum matrix S and profile matrix P, using one example of the three-dimensional data 71 as well as the data of the mass spectra 72 and chromatograms 73.

[0065] Next, the m/z detector 43 performs a peak-detecting operation for each column of the obtained spectrum matrix S (i.e., for each value of k from 1 to K), including the steps of detecting one or more peaks from the mass spectrum in the k-th column of the spectrum matrix S and determining the m/z values corresponding to those peaks (Step 4). The m/z values corresponding to those peaks will be the candidates of the m/z value of the precursor ion originating from the k-th component contained in the target sample. Similarly, the retention time detector 44 performs a peak-detecting operation for each row of the obtained profile matrix P (i.e., for each value of k from 1 to K), including the steps of detecting one or more peaks from the chromatogram in the k-th row of the profile matrix P and determining the retention times corresponding to those peaks (Step 5). The retention times corresponding to those peaks will be the candidates of the retention time of the k-th component contained in the target sample.

[0066] Based on the candidates of the m/z in the k-th column of the spectrum matrix S obtained in Step 4 and those of the retention time in the k-th row of the profile matrix P obtained in Step 5, the MS.sup.2 analysis execution condition determiner 45 creates a precursor-ion list L for each value of k from 1 to K (i.e., for each component contained in the target sample), where each item of the list consists of one candidate of the m/z of the precursor ion paired with one candidate of the retention time (Step 6).

[0067] In the case where the LC/MS data of a background with no sample has been acquired beforehand, the MS.sup.2 analysis execution condition determiner 45 may additionally perform, based on the background data, a selecting operation in which all pairs of the m/z candidate and retention-time candidate originating from the background are removed from the precursor-ion list L, and the remaining candidate pairs are selected as new pairs of the m/z candidate and retention-time candidate (Step 7). As another possibility, the previously described operation in Step 7 may be replaced by a selecting operation based on the background data in which only the pairs of the m/z candidate and retention-time candidate that fall within an m/z range and retention-time range which are free from the influence of the background are selected as new pairs of the m/z candidate and the retention-time candidate. These operations in Step 7 may be omitted.

[0068] Based on the obtained precursor-ion list L (after the removal of the pairs of the m/z candidate and retention-time candidate originating from the background if Step 7 is carried out), the MS.sup.2 analysis execution condition determiner 45 determines execution conditions of an MS.sup.2 analysis (MS.sup.2 analysis method) including the selection and fragmentation of the precursor ion of a component contained in the sample (Step 8). In most MS.sup.2 analyses, this type of analysis method is previously known for each component. Therefore, the known analysis methods can be previously stored in the storage unit 61 so that the MS.sup.2 analysis execution condition determiner 45 can retrieve an appropriate analysis method from the storage unit 61 based on the information concerning the candidates of the m/z and retention time in the precursor-ion list L.

[0069] In the process of determining an analysis method, the analysis method for an MS.sup.2 analysis of one sample may be divided into a plurality of analysis methods so that the loop time (sampling interval) for one MS.sup.2 analysis will be equal to or less than a predetermined value. This ensures a sufficiently high sampling rate and improves the sensitivity of the quantitative determination.

[0070] The process of determining an analysis method may allow the setting of a different level of collision energy for each component to be analyzed. This allows an analysis of each component to be more appropriately performed when an appropriate level of collision energy for each component is previously known. As another possibility, a plurality of levels of collision energy may be set for each component to be analyzed. This allows a tentative analysis to be performed using multiple levels of collision energy to determine an optimum fragmentation condition when an appropriate level of collision energy for each component is unknown.

[0071] Based on the analysis method thus determined, the LC/MS.sup.2 analysis executer 52 in the LC/IT-TOFMS 1 controls each component of the LC/IT-TOFMS 1 to perform an LC/MS.sup.2 analysis (Step 9). The LC/MS.sup.2 analysis is performed at all retention times included in the precursor-ion list L. After the LC/MS.sup.2 analyses at all retention times have been completed, the entire sequence of operations of the LC/IT-TOFMS 1 is discontinued.

(3) Details of Operation of Matrix Factorization

[0072] Next, using FIG. 4, details of the operation of the matrix factorization (Step 3) which is performed in the matrix factorization executer 42 will be described along with the functions of the components in the matrix factorization executer 42.

[0073] Initially, the regularization parameter-regularization function preparer 421 prepares a plurality of regularization-parameter candidates λr (in the present case, there are b candidates, where b is a natural number) and one regularization function R(S, P) (Step 31). The regularization function R(S, P) used in the present embodiment is the sum of the L1-norm of the matrix S and that of the matrix P, i.e., R(S, P)=|S|.sub.1+|P|.sub.1. The L1-norm of a matrix means the sum of all matrix elements in the matrix. As for the regularization-parameter candidates λr, a plurality of positive real numbers are appropriately selected.

[0074] Next, for each of the b regularization-parameter candidates λr, the matrix candidate determiner 422 determines spectrum and profile matrices Srt and Prt which minimize the value of the loss function L(S, P)=D(X|SP)+λrR(S, P) (Step 32). The distance function D(X|SP)=Σ.sub.n,mD(X.sub.nm|(SP).sub.nm) is the total of the distances between the matrix elements X.sub.nm of the data matrix X and the corresponding matrix elements (SP).sub.m of the product SP. This function represents the degree of difference between the matrix elements of the data matrix X and those of the product SP. In the present embodiment, the generalized KL divergence D.sub.KL(x|y)=x log(x/y)−(x−y) is used as the distance function D(x|y) for each element.

[0075] The matrix candidate determiner 422 further determines a spectrum matrix candidate Sr and profile matrix candidate Pr for each regularization-parameter candidate λr by one of the following two methods. In the first method, the combination of the temporary candidates (Srt, Prt) are directly selected as a spectrum matrix candidate Sr and profile matrix candidate Pr (Step 33).

[0076] In the second method, the following operations are performed in place of Step 33. Using Srt and Prt as the initial values, matrices Srt2 and Prt2 which minimize the value of a second loss function having no regularization term, L.sub.2(S, P)=D(X|SP), are determined (Step 33-1). These matrices Srt2 and Prt2 are selected as a spectrum matrix candidate Sr and profile matrix candidate Pr (Step 33-2).

[0077] After the combination (Sr, Pr) of the candidates of the spectrum matrix S and profile matrix P has been determined by one of the two methods, the probability distribution transformer 423 prepares a cumulative distribution function F.sub.nm of the probability distribution corresponding to the distance function D(X.sub.nm|(SP).sub.nm) for each of the b regularization-parameter candidates λr (Step 34). Then, for each of the b regularization-parameter candidates λr as well as for each combination (X.sub.nm, (SrPr).sub.nm) of the matrix element X.sub.nm of the data matrix X and the corresponding matrix element (SrPr).sub.nm of the product SrPr of the spectrum matrix candidate Sr and profile matrix candidate Pr, the probability distribution transformer 423 substitutes the values of those matrix elements into the cumulative distribution function FM to determine y.sub.nm=F.sub.nm(X.sub.nm|(SrPr).sub.nm) which is expected to show a standard uniform distribution (Step 35). It is commonly known that the cumulative distribution function F(X|SP) corresponding to the generalized KL divergence D.sub.KL(X|SP)=X log(X/SP)−(X−SP), which is the loss function used in the present embodiment, is a cumulative distribution function of a Poisson distribution.

[0078] Next, the goodness-of-fit calculator 424 calculates the goodness of fit between the empirical distribution y.sub.nm=F.sub.nm(X.sub.nm|(SrPr).sub.nm) determined by the probability distribution transformer 423 for each of the b regularization-parameter candidates λr and the cumulative distribution function of a standard uniform distribution (Step 36). For the calculation of the goodness of fit, commonly known methods for calculating the goodness of fit in statistics are available, such as a Kolmogorov-Smimov (KS) statistic, Cramer-von Mises statistic, or Anderson-Darling statistic.

[0079] Subsequently, the matrix determiner 425 compares the values of the goodness of fit respectively calculated for the regularization-parameter candidates λr, and selects, as the spectrum matrix S and profile matrix P, the spectrum matrix candidate Sr and profile matrix candidate Pr corresponding to the regularization-parameter candidate λr which yields the highest value of the goodness of fit (Step 37). In place of the spectrum and profile matrices corresponding to the regularization-parameter candidate λr which yields the highest value of the goodness of fit, the spectrum matrix candidate Sr and profile matrix candidate Pr corresponding to the largest value of the regularization parameter λr among the regularization-parameter candidates λr which yield the values of the goodness of fit equal to or greater than a predetermined threshold may be selected as the spectrum matrix S and profile matrix P. Thus, the operation of the matrix factorization is completed.

(4) Example of Calculation of Matrix Factorization

[0080] An example of the calculation of the matrix factorization using the chromatograms shown in the left section of FIG. 5 and the mass spectra shown in the right section of FIG. 5 is hereinafter described. In the left section of the drawing, a chromatogram acquired by one measurement is divided into five chromatograms showing four peaks originating from four kinds of components and the background (BG). Each of the five chromatograms corresponds to one row of a data matrix. Accordingly, in the present example, the value of j in the actual data (obtained by an experiment) is five. The right section of FIG. 5 shows five mass spectra which correspond to the five chromatograms, respectively.

[0081] FIG. 6 shows a large number of mass chromatograms acquired at different m/z values from the actual data (obtained by an experiment). This diagram shows the entire information held in the data matrix X. More specifically, one set of intensity values at a number of retention times in one mass chromatogram corresponds to one set of values of the matrix elements in one row of the data matrix X. On the other hand, one set of intensity values on the large number of mass chromatograms at one retention time corresponds to one set of values of the matrix elements in one column of the data matrix X.

[0082] For this data matrix X, a spectrum matrix candidate Sr and profile matrix candidate Pr were determined for three values of λr (1, 256 and 512) by the matrix factorization according to the present embodiment, and the goodness of fit was calculated for each case.

[0083] FIG. 7 shows the result obtained in the case of λr=1, where the chromatograms (left section) correspond to the rows of the matrix elements of the profile matrix candidate Pr, while the mass spectra (right section) correspond to the columns of the matrix elements of the spectrum matrix candidate Sr. Similarly, FIG. 8 shows chromatograms and mass spectra obtained in the case of λr=512, and FIG. 9 shows chromatograms and mass spectra obtained in the case of λr=256. The calculated value of the KS statistic is also shown in FIGS. 7-9. The KS statistic is a numerical value obtained by the Kolmogorov-Smimov (KS) test. The smaller the numerical value is, the higher the goodness of fit is.

[0084] In the case of λr=1 (FIG. 7), the KS statistic is 0.0924. The value of j in the determined profile matrix candidate Pr and spectrum matrix candidate Sr is 7, which is greater than the actual value (j=5). This means that the regularization was insufficient due to the too small value of λr. As a matter of fact, it is obvious that the obtained chromatograms and mass spectra do not agree with the actual data (FIG. 5).

[0085] In the case of λr=512 (FIG. 8), the KS statistic is 0.2652. The value of j in the determined profile matrix candidate Pr and spectrum matrix candidate Sr is 2, which is smaller than the actual value. This means that the regularization had an excessive effect due to the too large value of λr. It is obvious that the obtained chromatograms and mass spectra do not agree with the actual data (FIG. 5).

[0086] On the other hand, in the case of λr=256 (FIG. 9), the KS statistic is 0.0164, which is the lowest value among the three candidates. This means the highest goodness of fit among the three candidates. Accordingly, the profile matrix candidate Pr and spectrum matrix candidate Sr in the case of λr=256 should be selected as the profile matrix P and spectrum matrix S from the three candidates. The value of j in the obtained profile matrix candidate Pr and spectrum matrix candidate Sr is 5, which agrees with the actual data. The chromatograms and mass spectra obtained from the profile matrix P and spectrum matrix S are approximate to the actual data (FIG. 5).

[0087] In the previously described example, there were three regularization-parameter candidates λr, from each of which a profile matrix candidate Pr and spectrum matrix candidate Sr were obtained and shown in the form of chromatograms and mass spectra. The number of regularization-parameter candidates λr is not limited to three. The larger the number of the regularization-parameter candidates λ is, the more accurate the ultimately obtained profile matrix P and spectrum matrix S will be.

(5) Other Notes

[0088] In the previous embodiment, the data matrix X is defined as a matrix with k rows and n columns, the spectrum matrix S is defined as a matrix with k rows and j columns, and the profile matrix P is defined as a matrix with j rows and n columns. It is also possible to define the data matrix X as a matrix with n rows and k columns, the spectrum matrix S as a matrix with j rows and k columns, and the profile matrix P as a matrix with n rows and j columns. In that case, the product PS should be used in place of the product SP.

[0089] The configuration of the chromatograph mass spectrometer is not limited to that of the previously described TOFMS 1. For example, the present invention can also be applied in a chromatograph mass spectrometer which includes the combination of a mass filter (e.g., a quadrupole mass filter) and a collision cell in place of the ion trap 30 used in the previous embodiment, as well as an orthogonal acceleration TOF in place of the TOF 31 used in the previous embodiment. Furthermore, the present invention is not limited to TOFMSs but is also applicable in other types of chromatograph mass spectrometers.

[0090] In addition, it is needless to say that the present invention is not limited to the previous embodiment but can be changed or modified in various forms.

Modes of Invention

[0091] A person skilled in the art can understand that the previously described illustrative embodiment is a specific example of the following modes of the present invention.

[0092] (Clause 1)

[0093] A chromatograph mass spectrometer according to Clause 1 is a chromatograph mass spectrometer in which a chromatograph configured to temporally separate a sample into components is combined with a mass spectrometer having the function of an MS.sup.n analysis (where n is an integer equal to or greater than 2) in which each component in the sample separated by the chromatograph is subjected to the selection and fragmentation of an ion at least one time, and ions resulting from the fragmentation are subjected to mass spectrometry, the chromatograph mass spectrometer including:

[0094] an MS.sup.m-1 analysis executer configured to make the mass spectrometer perform an MS.sup.m-1 analysis (where m is an integer from 2 to n, inclusive) to acquire three-dimensional data showing an intensity for each of N m/z values (where N is a natural number) and each of M retention times (where M is a natural number);

[0095] a data matrix creator configured to create, based on the three-dimensional data, a data matrix X in which intensity data are arranged in N rows and M columns or M rows and N columns, where the N rows or N columns of intensity data differ from each other in the value of the m/z while the M columns or M rows of data differ from each other in the value of the retention time;

[0096] a matrix factorization executer configured to determine a spectrum matrix S and a profile matrix P by a technique of matrix factorization based on the data matrix X so that the data matrix X is approximated by the product SP in which the spectrum matrix S has N rows and K columns (where K is a natural number) while the profile matrix P has K rows and M columns, or by the product PS in which the spectrum matrix S has K rows and N columns while the profile matrix P has M rows and K columns;

[0097] an m/z detector configured to detect the m/z of a precursor ion originating from a component contained in the sample, from the values of the matrix elements in each column or each row of the spectrum matrix S;

[0098] a retention time detector configured to detect the retention time of a component contained in the sample, from the values of the matrix elements in each row or each column of the profile matrix P;

[0099] an MS.sup.m analysis execution condition determiner configured to determine, based on the m/z and the retention time, an execution condition of an MS.sup.m analysis including the selection and fragmentation of a precursor ion of a component contained in the sample; and

[0100] an MS.sup.m analysis executer configured to make the mass spectrometer execute an MS.sup.m analysis based on the execution condition.

[0101] The chromatograph mass spectrometer according to Clause 1 can specify a precursor ion based on the three-dimensional data acquired by an MS.sup.m-1 analysis, and appropriately set MS.sup.m analysis conditions, without requiring an operator to manually set analysis conditions.

[0102] (Clause 2)

[0103] In the chromatograph mass spectrometer according to Clause 2, which is one mode of the chromatograph mass spectrometer according to Clause 1, the MS.sup.m analysis execution condition determiner is further configured to perform an operation, based on previously acquired data concerning a background, for removing a precursor ion corresponding to a combination of an m/z candidate and a retention-time candidate originating from the background, from the target for which the execution condition for the MS.sup.m analysis should be determined.

[0104] (Clause 3)

[0105] In the chromatograph mass spectrometer according to Clause 3, which is one mode of the chromatograph mass spectrometer according to Clause 1, the MS.sup.m analysis execution condition determiner is further configured to perform an operation, based on previously acquired data concerning a background, for setting, as a target for which the execution condition for the MS.sup.m analysis should be determined, a precursor ion corresponding to a combination of an m/z candidate and a retention-time candidate that fall within an m/z range and retention-time range which are free from an influence of the background.

[0106] The chromatograph mass spectrometer according to Clause 2 or 3 removes an influence of the background based on previously acquired background data. Therefore, the execution condition for the MS.sup.m analysis can be more appropriately determined.

[0107] (Clause 4)

[0108] In the chromatograph mass spectrometer according to Clause 4, which is one mode of the chromatograph mass spectrometer according to one of Clauses 1-3, the MS.sup.m analysis execution condition determiner is further configured to perform the operation of creating divisional analysis methods by dividing an analysis method for an MS.sup.m analysis of one sample into a plurality of analysis methods so that the loop time for one MS.sup.m analysis will be equal to or less than a predetermined value.

[0109] The chromatograph mass spectrometer according to Clause 4 can ensure a sufficiently high sampling rate and improves the sensitivity of the quantitative determination.

[0110] (Clause 5)

[0111] In the chromatograph mass spectrometer according to Clause 5, which is one mode of the chromatograph mass spectrometer according to one of Clauses 1-4, the MS.sup.m analysis execution condition determiner is further configured to perform the operation of setting a different level of collision energy for each component to be analyzed.

[0112] (Clause 6)

[0113] In the chromatograph mass spectrometer according to Clause 6, which is one mode of the chromatograph mass spectrometer according to one of Clauses 1-4, the MS.sup.m analysis execution condition determiner is further configured to perform the operation of setting a plurality of levels of collision energy for each component to be analyzed.

[0114] The chromatograph mass spectrometer according to Clause 5 can perform an analysis of each component more appropriately when an appropriate level of collision energy for each component is previously known. The chromatograph mass spectrometer according to Clause 6 can perform a tentative analysis using multiple levels of collision energy to determine an optimum fragmentation condition when an appropriate level of collision energy for each component is unknown.

[0115] (Clause 7)

[0116] In the chromatograph mass spectrometer according to Clause 6, which is one mode of the chromatograph mass spectrometer according to one of Clauses 1-6, the matrix factorization executer includes:

[0117] a regularization parameter-regularization function preparer configured to prepare a plurality of regularization-parameter candidates λr (where r is a natural number from 1 to r.sub.max) and one regularization function R(S, P) which induces sparsity of the solution;

[0118] a matrix candidate determiner configured to solve an optimization problem for each of the plurality of regularization-parameter candidates λr so as to determine a matrix Srt as a spectrum matrix candidate Sr which is a candidate of the spectrum matrix S and a matrix Prt as a profile matrix candidate Pr which is a candidate of the profile matrix P, where the matrices Srt and Prt are determined so as to minimize the value of a loss function L(S, P)=D(X|SP)+λrR(S, P), where D(X|SP) is a distance function expressing the degree of difference between the data matrix X and the product SP, while λrR(S, P) is the product of the regularization-parameter candidate λr and the regularization function R(S, P);

[0119] a probability distribution transformer configured to determine, for each of the plurality of regularization-parameter candidates λr, a transformed value y.sub.nm=F.sub.nm(X.sub.nm|(SrPr).sub.nm) by a variable transform into a common probability distribution Pcommon for each combination of a matrix element X.sub.nm of the data matrix X and a corresponding matrix element (SrPr).sub.nm of the product SrPr of the spectrum matrix candidate Sr and the profile matrix candidate Pr, using F.sub.nm which is a function for the variable transform from a probability distribution P.sub.nm corresponding to the distance function D(X.sub.nm|(SP).sub.nm) into the common probability distribution Pcommon;

[0120] a goodness-of-fit calculator configured to determine, for each of the plurality of regularization-parameter candidates λr, a goodness of fit between the transformed value y.sub.nm and a cumulative distribution function of the probability distribution Pcommon; and

[0121] a matrix determiner configured to select, as the spectrum matrix S and the profile matrix P, the spectrum matrix candidate Sr and the profile matrix candidate Pr determined for a regularization-parameter candidate λr which yields the highest value of the goodness of fit among the plurality of regularization-parameter candidates λr, or a regularization-parameter candidate λr which yields the goodness of fit higher than a predetermined threshold and also has the largest value of λr.

[0122] The chromatograph mass spectrometer according to Clause 7 can determine a spectrum matrix S and profile matrix P whose factor number K is appropriate and close to the number of kinds of components contained in a sample even when the number of kinds of components is unknown.

[0123] (Clause 8)

[0124] In the chromatograph mass spectrometer according to Clause 8, which is one mode of the chromatograph mass spectrometer according to Clause 7, the matrix candidate determiner is configured to use the matrix Srt and the matrix Prt as initial values for determining a matrix Srt2 and a matrix Prt2 which minimize the value of a second loss function with no regularization term, L.sub.2(S, P)=D(X|SP), and to select the matrix Srt2 and the matrix Prt2 as the spectrum matrix candidate Sr and the profile matrix candidate Pr, instead of selecting the matrix Srt and the matrix Prt as the spectrum matrix candidate Sr and the profile matrix candidate Pr.

[0125] In the chromatograph mass spectrometer according to Clause 8, the matrices Srt and Prt which have been determined so as to minimize the value of the loss function L(S, P)=D(X|SP)+λrR(S, P) are used as initial values for making an additional determination for a second optimization which does not include the regularization term. This operation corrects a bias of the residual due to the regularization and enables the selection of spectrum and profile matrix candidates Sr and Pr which are closer to the actual data. Consequently, the spectrum and profile matrices S and P to be ultimately obtained will be more accurate.

[0126] (Clause 9)

[0127] In the chromatograph mass spectrometer according to Clause 9, which is one mode of the chromatograph mass spectrometer according to Clause 7 or 8, the regularization term R(S, P) is L1-norm, or a linear combination of L1-norm and L2-norm, or a function which applies a trace norm, det|S.sup.TS| or log det|S.sup.TS+δI| to the matrix S (where I is a unit matrix, and δ is a hyperparameter for controlling the regularization function) while placing a constraint on the solution so that the total of the values in each column of the matrix P should not exceed 1.

[0128] (Clause 10)

[0129] In the chromatograph mass spectrometer according to Clause 10, which is one mode of the chromatograph mass spectrometer according to one of Clauses 7-9, the cumulative distribution function F(X|Y) is a function selected from a cumulative distribution function calculated from a function expressing a Gaussian distribution, a cumulative distribution function calculated from a function expressing a Poisson distribution, a cumulative distribution function calculated from a function expressing an exponential distribution, and a cumulative distribution function calculated from a function expressing a Tweedie distribution.

[0130] (Clause 11)

[0131] In the chromatograph mass spectrometer according to Clause 11, which is one mode of the chromatograph mass spectrometer according to one of Clauses 7-10, the goodness-of-fit calculator is configured to calculate the goodness of fit by a test selected from a Kolmogorov-Smimov test, a Cramer-von Mises test, and an Anderson-Darling test.

[0132] (Clause 12)

[0133] In the chromatograph mass spectrometer according to Clause 12, which is one mode of the chromatograph mass spectrometer according to one of Clauses 7-10, when the variance σ.sub.nm.sup.2 of the noise in each matrix element X.sub.nm is previously known, the goodness-of-fit calculator defines the probability distribution Pcommon as a standard normal distribution, defines the cumulative distribution function F.sub.nm(X.sub.nm|(SrPr).sub.nm) as (X.sub.nm−(SrPr).sub.nm)/δ.sub.nm, and uses −|σ.sub.y.sup.2−1| as the goodness of fit, where σ.sub.y.sup.2 is given by:

[00003]${\sigma }_y^2=\frac{1}{\mathrm{NM}-1}{.Math.}_{n,m}{y}_{\mathrm{nm}}^2$

which is the unbiased variance of the transformed value y.sub.nm whose mean value is assumed to be zero.

[0134] In the present invention, the various regularization terms R(S, P) mentioned in Clause 9 and the various cumulative functions F(X|Y) mentioned in Clause 10 can be appropriately used. For the calculation of the goodness of fit, the various methods mentioned in Clause 11 or 12, which are commonly known in the area of statistics, can be appropriately used.

REFERENCE SIGNS LIST

[0135] 1 . . . Liquid Chromatograph/Ion Trap Time-of-Flight Mass Spectrometer [0136] 10 . . . Liquid Chromatograph (LC) Unit [0137] 11 . . . Mobile Phase Container [0138] 12 . . . Liquid Supply Pump [0139] 13 . . . Injector [0140] 14 . . . Column [0141] 20 . . . Mass Spectrometry (MS) Unit [0142] 21 . . . Ionization Chamber [0143] 22 . . . ESI Nozzle [0144] 23 . . . Desolvation Tube [0145] 24 . . . First-Stage Intermediate Vacuum Chamber [0146] 25 . . . First Ion Guide [0147] 26 . . . Skimmer [0148] 27 . . . Second-Stage Intermediate Vacuum Chamber [0149] 28 . . . Second Ion Guide [0150] 29 . . . Analysis Chamber [0151] 30 . . . Ion Trap [0152] 31 . . . Time-of-Flight (TOF) Mass Separator [0153] 32 . . . Reflectron Electrode [0154] 33 . . . Ion Detector [0155] 34 . . . Analogue-to-Digital (A/D) Converter [0156] 40 . . . Data Processing Unit [0157] 41 . . . Data Matrix Creator [0158] 42 . . . Matrix Factorization Executer [0159] 421 . . . Regularization Parameter-Regularization Function Preparer [0160] 422 . . . Matrix Candidate Determiner [0161] 423 . . . Probability Distribution Transformer [0162] 424 . . . Goodness-of-Fit Calculator [0163] 425 . . . Matrix Determiner [0164] 43 . . . m/z Detector [0165] 44 . . . Retention Time Detector [0166] 45 . . . MS.sup.2 Analysis Execution Condition Determiner [0167] 50 . . . Analysis Control Unit [0168] 51 . . . LC/MS Analysis Executer [0169] 52 . . . LC/MS.sup.2 Analysis Executer [0170] 61 . . . Storage Unit [0171] 63 . . . Operation Unit [0172] 71 . . . Three-Dimensional Data [0173] 72 . . . Mass Spectrum [0174] 73 . . . Chromatogram