JP3270290B2 - Multi-channel chromatogram analysis method and data processing device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to chromatography techniques such as high performance liquid chromatography and gas chromatography, and more particularly to a method for analyzing a multi-channel chromatogram obtained by a diode array detector and the like and a data processing apparatus using the method. .

[0002]

2. Description of the Related Art Factor analysis as a method for separating and resolving overlapping peaks on a multichannel chromatogram is described in, for example, "FACTOR ANALY" by EDMUND R. MALINOWSKI.
SIS IN CHEMISTRY ”, JOHN WILEY & SONS, INC. (199
It is introduced in detail in 1).

[0003] This factor analysis method is one of the multivariate analysis methods. The basic concept is expressed by the following equations (1) and (2).
The data matrix D is modeled as the product of the spectrum matrix X and the elution profile matrix Y as shown in FIG. However, equation (2) is to normalize each component K to a peak area of 1. However, since mathematically it cannot be uniquely decomposed, it is intended to solve the problem by providing various reasonable constraints.

D = XY (1) where Dij: signal intensity XiK: spectrum intensity YKj: elution profile i: channel number K: component number j: time number

[0005]

(Equation 1)

Generally, the eigenvalue problem of the data matrix D is solved, the number n of components is determined by principal component analysis, and the abstract elution profile matrix V having eigenvectors as elements is converted by the matrix T, and physically converted. Obtain a meaningful elution profile matrix Y. Once the matrix Y is obtained, the spectrum matrix X can also be calculated from the data matrix D (the following equations (3) to (1)
1)).

When the eigenvalue problem is solved, it is expressed as a product of an abstract spectrum matrix U and an elution profile matrix V as shown in the following equation (3).

D = UV (3)

[0009]

(Equation 2)

[0010]

(Equation 3)

[0011]

(Equation 4)

Further, as shown in the following equation (7), the matrix V can be converted to the matrix Y by the conversion matrix T. Y = TV (7) Here, the matrix T is an n × n matrix. The matrix T is responsible for oblique rotation and for converting the area of each row vector of the matrix Y to one.

The matrix T can be expressed by the product of the rotation R and the normalization matrix N (T = NR).

[0014]

(Equation 5)

The matrix X is obtained from the matrices D and Y as shown in the following equation (10). X = DY ^T (YY ^T ) ^-1 (10) The matrices X and Y and the matrices U and V are given by the following equation (11).
There is a relationship shown in D = XY = (UT ⁻¹ ) (TV) (11) However, the transformation matrix T cannot be easily determined. Therefore, the following methods have been proposed.

1. Method of known spectrum 1-1. TTFA (Target Transformation Factor Ana
lysis): A method of obtaining a transformation matrix T from an abstract spectrum matrix so as to obtain a known spectrum waveform. 1-2. RAFA (Rank Annihilation Factor Analysi
s): A method in which a known data matrix is obtained from a standard product of each component, and one component is subtracted. 1-3. GRATA (Generalized Rank Annihilation
Factor Analysis): 1-2. RAFA requires a data matrix for each component, but this method uses a data matrix from a standard mixed sample to generate a curve matrix.
Resolution is possible.

2. Modeling method with unknown spectrum 2-1. Gaussian nonlinear least squares: A method in which the elution profile is modeled as Gaussian and fitted by nonlinear least squares. 2-2. The above-mentioned EMG (Expenentially Modified Gau
ssian) nonlinear least squares method: 2-1. Using EMG which can also represent asymmetric peaks instead of Gaussian.

3. Self-modeling method with unknown spectrum 3-1. ITTFA (Iterative Target Transformatio
n Factor Analysis): A method in which a test vector with a pulsed elution profile is initially introduced and adjusted sequentially to approximate the true elution profile. 3-2. EFA (Evolving Factor Analysis): A method in which changes in eigenvalues are plotted over time to find a stable region of the eigenvalues. This stable region is called a window, and the method of fixing the outside of the region to zero and determining the elution profile of each component is particularly called WFA (Window Factor Analysis) (ERMalinowski, J. Chemometrics, 6, 29-40).
(1992), H.C. R. Keller et al., Anal. Chim. Act
a., 246, 379-390 (1991)). 3-3. RAEFA (Rank Annihilation by Evolving
Factor Analysis): 3-2. Is repeated while subtracting the peak resolution by EFA from the data matrix one by one. 3-4. RAFA using information entropy as an index: A method of lowering the rank of a matrix for each component on the basis of minimizing information entropy when obtaining an elution profile (I. Sakuma et al., J. Chromatogr, 506) ,
223-243 (1990)).

As a general method for analyzing a two-dimensional chromatogram, a method utilizing deconvolution processing has been proposed (US Pat. No. 4941101; Paul Benjamine Cr).
illy, IEEE Transaction on Instrumentation and Meas
urement, 40 (1991) 558-562; Paul Benj
amine Crilly, Journal of Chemometrics, 5 (1991) 8
5-95).

Several techniques have been proposed for the deconvolution processing algorithm to speed up convergence. Including Gaussian elimination, which performs the inverse matrix operation,
acobi method, Gauss-Seidel method, Fourier transform method, Van Citt
ert method, Constrained Iterative method, Jansson method, Gold's
It is devised mainly by iterative methods such as the ratio method (edited by Shigeo Minami, "Waveform Data Processing for Scientific Measurement", CQ Publishing Company (1986) pp. 122-139; PA Jansson, De.
convolution With Applications in Spectroscopy, New
York, Academic (1984)).

[0021]

However, the method for separating and resolving overlapping peaks in a multi-channel chromatogram by the above-described factor analysis has the following problems.

1. When using a method with a known spectrum, it is necessary to know the spectrum of the standard product in advance,
Not applicable if you do not know the spectrum. 2. Modeling methods with unknown spectra have to introduce model functions and cannot be said to reproduce actual elution profiles. 3. Each of the self-modeling methods with unknown spectra has its own problems. That is, 3-
1. ITTFA and 3-4. If the information entropy is used as an index, RAFA requires high resolution for overlapping peaks in the measurement data matrix (JK STRASTERS et al., J.
OURNAL OF LIQUID CHROMATOGRAPHY, 12 (1 & 2), 3-
22 (1989); Sakuma et al. ). When the resolution is low, the calculation processing does not converge, and a meaningful solution cannot be obtained.

Also, 3-2. EFA and 3-3. R
AEFA is a method that assumes a region where the elution profile is zero outside the window. Strictly speaking, however, there are no regions in the elution profile that are close to zero but close to zero, so the decomposed results will include hypothetical errors. In practice, it is difficult to clearly define the boundary between zero and non-zero. In addition, the deconvolution processing is susceptible to noise, has a large amplification after the processing, and has a problem that pseudo peaks often appear.

An object of the present invention is to realize a multi-channel chromatogram analysis method and a data device capable of analyzing an overlapping peak of a multi-channel chromatogram with high precision and analyzing the same.

Here, a deconvolution factor analysis (hereinafter abbreviated as DFA) is proposed as a method of making use of the advantages of factor analysis and deconvolution and compensating for each other's disadvantages. The purpose of DFA is (1) unknown spectrum,
(2) The model function is free, and (3) The spectrum matrix X and the elution profile matrix Y are determined from the data matrix D under the condition that no zero region is assumed in the elution profile.

In the DFA, the calculation process converges even on an overlapping peak having a low resolution such as a shoulder peak. In addition, it is possible to overcome the disadvantage that the deconvolution processing has a weakness against noise, which is used to decompose the information from multi-channel information.

[0027]

The present invention is configured as follows to achieve the above object. In the method of analyzing overlapping peaks on a multi-channel chromatogram,
A step of performing deconvolution processing on the multi-channel chromatogram using a predetermined spread function, a step of performing multivariate analysis from the processed multi-channel chromatogram, and a width of the predetermined spread function on the basis that the peak is isolated. A step of determining a corresponding multi-channel chromatogram after processing and a step of obtaining a spectrum waveform from the processed multi-channel chromatogram are provided.

Also, in the method of analyzing overlapping peaks on a multi-channel chromatogram, a step of deconvolution processing the multi-channel chromatogram using a predetermined spread function, and a step of factor-analyzing the processed multi-channel chromatogram. Determining from the relationship between the width of the predetermined spread function and the corresponding factor analysis solution whether the predetermined variable of the multi-channel chromatogram is stable; and the width of the spread function based on the isolation of the peak. And a step of determining a processed multi-channel chromatogram corresponding thereto and a step of obtaining a spectrum waveform from the processed multi-channel chromatogram.

Also, in the method of analyzing overlapping peaks on a multi-channel chromatogram, a step of deconvolution processing the multi-channel chromatogram using a predetermined spread function, and a step of factor-analyzing the processed multi-channel chromatogram. Determining whether the factor analysis corresponding to the width of the predetermined spread function has a unique solution; and, based on the fact that the factor analysis has a unique solution, the width of the spread function and the corresponding processed Determining a multi-channel chromatogram; and acquiring a spectral waveform from the processed multi-channel chromatogram.

In the method of analyzing overlapping peaks on a multi-channel chromatogram, a step of deconvolution processing the multi-channel chromatogram using a predetermined spread function and a step of solving an eigenvalue problem from the processed multi-channel chromatogram And determining whether the peak is isolated from the relationship between the width of the spread function and the solution of the corresponding eigenvalue problem; and, based on the isolation of the peak, the width of the spread function and the corresponding multi-channel after processing. Determining a chromatogram; and obtaining an elution profile from the processed multi-channel chromatogram.

Preferably, in the method for analyzing a multi-channel chromatogram, the step of determining the multi-channel chromatogram further specifies the number of components from the solution of the eigenvalue problem and the width of the spread function.

In a data processing apparatus for analyzing overlapping peaks on a multi-channel chromatogram, a storage unit for a spread function width, a deconvolution processing unit for deconvolution processing the spread function stored in the storage unit, Using a factor analysis unit that performs factor analysis on the multichannel chromatogram that has been subjected to the factorization, a storage unit that stores the solution of the eigenvalue problem of the multichannel chromatogram that has undergone the factor analysis, and the width of the spread function determined based on the solution of the eigenvalue problem And an output unit for outputting a spectrum waveform obtained from the deconvolution-processed multi-channel chromatogram.

Preferably, in the data processing device, the output unit further specifies the number of components from the solution of the eigenvalue problem and the width of the spread function.

[0034]

Next, the basic procedure of the DFA will be described. 1. A deconvolution process is performed on the data matrix D. The deconvolution processing is to deconvolve the spread function h from the data matrix D to obtain a deconvoluted data matrix d, as represented by the following equation (12).

Deconvolution processing Dij = hi (σ) * dij (σ) (12) Here, D is obtained by convolving h and d.
The following equation (13) expresses the deconvolution in an integral form by replacing the discrete time j with a continuous variable t. However, the convolved matrix is expressed in lower case.

[0036]

(Equation 6)

Since the deconvolution process is executed using the time t as a variable, each channel number i relating to the spectrum is processed separately. That is, deconvolution processing is performed on each row vector of the matrix D using the same spread function h.

Here, in the case of a chromatogram, a Gaussian such as the following equation (14) is generally used for the function h. It is well known that Gaussian is convolved as a spreading function in peak formation.

[0039]

(Equation 7)

Regarding the spreading mechanism, see “DYNAMI
CS OF CHROMATOGRAPY ”, by JC Giddings, Marcel De
This is described in detail in kker, New York, 1965.

2. Prior to the deconvolution processing, a standard deviation σ serving as a measure of Gaussian spread must be determined. Simply, a standard deviation σ such that all overlapping peaks are separated and isolated may be used. In fact, if all peaks are isolated on any channel,
The eigenvector shown in equation (3) corresponds to each component, and the elution profile matrix y can be determined.

Further, a spectrum matrix X is obtained according to the following equation. d = Xy (15) X = dy ^T (yy ^T ) ⁻¹ (16) where y is an elution profile matrix isolated by the deconvolution processing. y is obtained by rotating and normalizing the eigenvector at the time of isolation.

Also, the elution profile matrix Y can be determined according to the following equation (17), that is, the matrix Y is
And X, the decomposition of the overlapping peak is completed. Y = (X ^T X) ⁻¹ X ^T D (17) Alternatively, when the matrix y can be determined, the function h at the time of the deconvolution process is used to isolate the isolated peak. By performing convolution again, the original elution profile matrix Y can be restored (hereinafter, this process is referred to as reconvolution). Subsequently, according to equation (10),
The spectrum matrix X can also be obtained.

3. It is ideal that all peaks on the chromatogram are isolated by one deconvolution process as described above, but in reality (1) the number of components of overlapping peaks is unknown, and ( 2) In general, when the deconvolution process is performed using the spread function h having a standard deviation σ larger than necessary, there is a problem that noise is extremely amplified. To solve this, RAFA
Can be applied.

Further, the matrix D is deconvolved while gradually expanding σ to obtain a matrix d (σ). A factor analysis is performed on each matrix d (σ). As described above, the transformation matrix T is generally undefined. When the peak is isolated, there is a property that a physically significant elution profile matrix Y can be obtained by orthogonal rotation conversion. To obtain a proper orthogonal rotation transformation matrix T, varimax or quart
imax and so on. For example, each matrix d (σ) is converted to σ by varimax.
When one column vector in the obtained spectrum matrix X has a constant spectrum waveform, it can be determined that the peak has been isolated. This fixed column vector is determined as the spectrum of the first component, and the rank of the first component is reduced by one. The same rank annihilation process is repeated to complete the factor analysis of peak resolution.

Similarly, a method of determining oblique rotation instead of orthogonal rotation (obliquemax, quartimin, biquartimin, covari
min, binormamin, maxplane, promax, etc.). Also, a criterion that makes the eigenvector as positive as possible and maximizes the value of the following equation (18) can be adopted.

[0047]

(Equation 8)

In this case, a spectrum can be obtained even if the peak is not completely isolated, and there is an advantage that rank annihilation can be sequentially performed by small-scale deconvolution.

In general factor analysis, the spectral matrix X
And the elution profile matrix Y cannot be uniquely determined. In this DFA, by performing deconvolution processing,
Isolate the elution profile of each component, matrix X and matrix Y
Can be determined. This is important in that the matrix D is the product of the matrix X and the matrix Y, and when the deconvolution is performed on the matrix D, the deconvolution is actually performed only on the matrix Y. Therefore, when the elution profile is isolated, the matrix X is preserved, so that the spectrum waveform can be immediately obtained. The matrix Y can be calculated from the matrix D using the matrix X, or can be obtained by reconvolution from the isolated elution profile.

On the other hand, when the deconvolution process is performed on the two-dimensional chromatogram, there is a problem that a pseudo peak is amplified. Since the DFA is expanded to multi-channel data, noise cancels out each other and generation of a pseudo peak is suppressed.

Further, by performing the deconvolution processing while performing the factor analysis, the validity of the processing itself can be checked. This check is important because the present process is repeatedly performed in the present invention. For example, matrix D
From the matrix V to the matrix v
(Abstract elution profile after processing). On the other hand, the matrix U remains unchanged by this processing. As shown in the left half of FIG. 9, v obtained after solving the eigenvalue problem from D and passing through V after deconvolution processing should be equal to v which can be deconvolved from D and solving the eigenvalue problem via d. It is.

The validity of the deconvolution processing can be checked by comparing these two v. If the data exceeds a certain allowable range, this processing is reviewed again, and the processing can be corrected to a processing that is almost permissible from any route, thereby ensuring consistency. The same can be said for the right half of FIG. By comparing the two paths from V to y, the validity of the deconvolution process is checked, and it is possible to modify this process to ensure its consistency. However, in this case, since the transformation T is involved, it is necessary to rotate the two paths at exactly the same angle. Normally, the rotation R determined after the deconvolution processing is used. By solving the eigenvalue problem by the method described here, the deconvolution processing can be corrected to a more appropriate one, and the generation of a pseudo peak due to noise amplification, which is a disadvantage thereof, can be suppressed as much as possible.

Further, as shown in FIG. 10, the validity of the entire DFA is also determined by deconvolving Y calculated from X using X by Y (upper path) and the same X from d. It can be checked that y (the lower path) obtained using

[0054]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments according to the present invention will be described with reference to the accompanying drawings. FIG. 1 is an operation flowchart of a multi-channel chromatogram analysis method according to one embodiment of the present invention, and FIG. 2 is a schematic configuration diagram of a multi-channel chromatogram analyzer according to one embodiment of the present invention. The example of FIG. 2 is an example in which the present invention is applied to a drug monitoring HPLC system.

In FIG. 2, the pump 40 sends an eluent 49 according to a command from the control unit 44. A sampler 43 having a movable needle is used for a standard sample 52 contained in a sample container.
Alternatively, an unknown sample (measurement target) 51 is injected and sent to the flow path to the column 41. The sample 51 or 52 is sent to the column 41 together with the eluent 49, the components are separated and developed, and detected by the diode array detector 42.
The multi-channel chromatogram, which is the detection data, is supplied to the data analyzer 45.

Further, components and the like to be described later are input from the input unit 54 to the data analysis unit 45. The result analyzed by the data analysis unit 45 is displayed on the CRT 46 and the printer 47.

The data analysis unit 45 includes a data matrix determination unit (data setting unit) 451, a deconvolution unit 458, and an elution profile matrix calculation unit (factor analysis unit).
462, a spectrum matrix calculation unit 463, and a component identification quantitative calculation unit 461.

FIG. 3 is a multi-channel chromatogram of the unknown sample 51 obtained by the present system. When a biological sample is measured as in the present system, unknown contaminant peaks appear and often overlap. Because the components 2 and 3 overlap, it is difficult to identify not only the components but also the components. Therefore, it is necessary to separate and decompose the overlapping peaks, accurately determine the retention time, and accurately determine the spectrum, and identify the components. In addition, by accurately determining the size of the peak by separation and decomposition,
From the comparison with the standard sample 52, quantitative analysis can also be performed.

The operation, calculation flowchart, and result output after the multi-channel chromatogram is obtained will be described in order. The flow starts in step 111 of FIG. In step 112, an overlap peak to be separated / decomposed is specified. A contour map of the three-dimensional chromatogram as shown in FIG. 3 is displayed on the CRT 46. Components 2 and 3 are input by arrows using an input unit 54 such as a mouse or a pen. Or input the coordinates directly by numerical value or line movement on the display.

The starting point and the ending point of the entire overlapping peak specified in the step 113 are searched, and the time point of the data matrix D is determined from the starting point to the ending point. The number of measurement points from the start point to the end point is preferably 20 to 30, but if the number of points exceeds this, it is desirable to reduce the number of points by averaging or the like. Regarding the number of points in the wave number direction, the number of measurement channels is simply adopted. Alternatively, reduce the number of points by averaging or thinning.

In step 114, the eigenvalue problem is solved, and eigenvalues and eigenvectors are calculated. As shown in the above equation (3), a matrix Z is obtained from the matrix D, and an eigenvalue ζ and an eigenvector v of the matrix Z are calculated (FIG. 4A). Let の も の max be the largest eigenvalue. In order to determine the number n of components in step 115, for example, components having eigenvalues equal to or larger than (ζmax / 100) are counted for convenience. In this embodiment, n = 2.

Next, a matrix T for converting a matrix V having the eigenvectors v as row components into an elution profile matrix Y is estimated. Here, varimax is used as the estimation method. After conversion, varimax is ± 1 for each row vector.
Or approaching 0 (Chuichi Okuno et al., “Multivariate Analysis Method (Revised Edition)”, Nikka Giren Publishing Company (1
992) pp. 323-372), so that the obtained matrix Y
Can be assumed to largely reflect the individual pulsed elution profile. Actually, the matrix T also has a role of normalizing the matrix Y to the area 1. Next, the processing shifts to deconvolution processing. In step 116, an apparatus function h (σ) used for the deconvolution processing is set. The initial value of the standard deviation σ is, for example, 1/100 of the time width of the matrix D, a fixed value, or an input value. Also, it is one method to set the initial value to 0 first and see through the determination in step 119. In step 117, the matrix D is subjected to deconvolution processing using h (σ) as in the equation (12). Any method such as the Jansson method can be used. Here, a data matrix d (σ) subjected to the deconvolution processing is obtained (FIG. 6).

In step 118, eigenvalues and eigenvectors are calculated for the matrix d in the same manner as for the matrix D, and then converted to a matrix y by the matrix T according to the varimax criterion. The spectrum matrix X corresponding to the matrix y is calculated by the above-described equations d = Xy and X = dy ^T (yy ^T ) ⁻¹ . In step 119, it is determined that the peaks of the matrix d are sufficiently separated and isolated. The criterion is that the matrix X obtained here matches the matrix X obtained immediately before. At first, the matrix D is compared with the matrix X obtained based on the matrix D, and after the second time, the comparison is performed with the matrices d sequentially subjected to the deconvolution processing. Generally, the degree of coincidence is determined by a correlation coefficient (FIG. 7).

In essence, comparing the degree of coincidence of the matrix X is equivalent to comparing the rotation angle of the transformation. In this embodiment, the matrix T has a role of rotation R and normalization N, but comparison of only the rotating transformation matrix R is equivalent.

Further, the number of components described above can be automatically determined using a ζ-σ plot (FIG. 8) showing the relationship between the eigenvalue ζ and the standard deviation σ. gradually increase σ,
Measure the change in ζ. The two large eigenvalues increase monotonically as σ increases. This is because the peak waveform is sharpened by the deconvolution. By utilizing this property, the number of components can be determined by counting the eigenvalues that monotonically increase as σ increases. If this criterion is satisfied, the process proceeds to step 121. If so, the standard deviation σ is increased in step 120, and the deconvolution process is attempted again. For the increment, for example, the above-described initial value of the standard deviation is used. Here, if an abnormally large pseudo peak or the like occurs, it is determined that the processing is defective, an error occurs, and the flow ends.

In step 121, the matrix d (σ) finally determined when the judgment in step 119 is passed is registered. In step 122, each eigenvector is orthogonally rotated to obtain an area (sum)
Is normalized to 1 and is set as a row vector of the elution profile matrix y subjected to the deconvolution processing.

In step 123, a spectrum matrix X is obtained according to equations (15) and (16) (FIG. 5). In step 124, the elution profile matrix Y is calculated according to equation (17). Component identification is performed based on the spectrum waveform of the matrix X and the retention time of the matrix Y. The spectrum of the drug is stored in advance, and a search is performed from the library. Step 1
In step 25, the magnitude of the matrix X is used to refer to the spectrum matrix of the standard sample 52, and quantitative calculation is performed in step 126.

In step 127, the results as shown in FIGS. 3 to 5 are displayed on the CRT 46 and can be printed out by the operator's selection.

At step 128, the flow ends.

The method using oblique rotation has the advantage that rank annihilation can be performed in order by smaller-scale deconvolution. For example cov instead of varimax
Can rotate obliquely with respect to arimin. If the spectral matrix can be estimated using a smaller σ before deconvolution using the standard deviation σ that completely isolates the data, rank anihilation can be performed faster and without amplifying noise.

In the operation procedure, DFA analysis can be performed immediately after data collection using preset parameters. Here, an operation procedure as secondary processing of data analysis will be described.
First, a three-dimensional chromatogram is read out on a CRT (FIG. 3). Here, in order to set a data matrix D for resolving the overlapping peaks, the time domain, the wavelength domain, and the numbers of rows and columns are input. Solve the eigenvalue problem and display the eigenvalues in descending order.

The user estimates the number of significant components in descending order and inputs the number n of components. If this is set in advance as a rule, the number n of components can be automatically estimated. If rank annihilation is set, the process can be advanced as two components for the time being. Next, the type of rotation conversion is selected. Generally vari
Select max and set covarimin to allow up to oblique rotation. Next, an increase Δσ of the standard deviation σ of the spread function is set. For example, every second.

With respect to the above-described DFA method according to the embodiment of the present invention, a method of proceeding with the deconvolution processing until a peak is isolated or a chromatogram region consisting of purely one component appears. Until this purely single component chromatogram region appears,
It can be said that the method of proceeding with the deconvolution processing is based on purity by sharpening the peak to the extent that a spectrum of one component is obtained (see FIGS. 11 and 12).

On the other hand, the DFA method can be based on uniqueness. In other words, even if the peaks are not isolated or a pure one-component chromatogram region does not exist, if the factor analysis can uniquely have a solution, the deconvolution processing is performed at that time. The process ends.

For example, FIG. 14 shows that the rotation angle of the rotation matrix R is indefinite in an area where the standard deviation σ is small. Here, it is only required that both the spectrum matrix X and the elution profile matrix Y are positive. For example, after σ = 6 s, the indeterminate region of the component 1 disappears, and the rotation angle is uniquely obtained. Based on this uniqueness, the deconvolution process can be completed and rank annihilation performed.

As described above, the DFA method can be completed with a small-scale deconvolution process until a chromatogram region consisting of one component appears, as compared with the method of proceeding with the deconvolution process. Then, the overlapping peaks can be resolved without relatively deforming the original data matrix. This means that noise and distortion can be suppressed in the finally obtained spectrum X and elution profile Y (see FIG. 18).

Referring to FIG. 13, it is shown that in the small-scale deconvolution processing in which σ is 6 s or less, the factor analysis is still undefined. Component 1 has a rotation angle of 180
At any transformation angle around °, one can see a difference with a positive (non-negative) spectrum and a positive elution profile. When σ is 6 s, a positive spectrum X and a positive elution profile Y can only be obtained at about 160 °. When σ is less than 6 s, the rotation angle of the component 1 is undefined, but the angle θ is uniquely determined with σ = 6 s as a boundary.

The deconvolution process ends at this point, and the spectrum of component 1 and the elution profile are determined. Once reached, the spectrum and elution profile of component 2 can also be determined by rank annihilation.

Further, the outline will be described.
If σ of 3 is less than about 10 s, component 2 is a region where a solution can be obtained by oblique rotation. Component 2 has no uniqueness yet. At 10 s or longer, the region is a region where a solution is obtained by orthogonal rotation. The difference between the angles θ of the components 1 and 2 is exactly 90 °.

Here, both the spectrum X and the elution profile Y have positive conditions. As described above, even if it is positive, the degree to which the value fluctuates negatively due to noise must be allowed.
Accepted.

When both the spectrum X and the elution profile Y are positive only, when there are two components having the same spectrum, an elution profile having two maximum values as shown in FIG. 14 is obtained. To prevent this, unimodalit
y (only one peak per profi-le) is conditioned on the peak shape (Anal. chem. 1993, 65, 2040-2043).

Next, an operation procedure for causing the apparatus to execute the DFA by the operator will be described with reference to FIG. First, in step 200, DFA is selected from the disassembly method menu (displayed on the display) shown in FIG. Next, in step 201, when a diagonal point is specified (drag) on the screen while viewing the data matrix D for analysis by the user while looking at the contour diagram (FIG. 3), the number of data points in rows and columns is automatically set. Decide about 20 rows and 30 columns. Here, the user can also change the number of data points.

Subsequently, in steps 202 and 203, the deconvolution process proceeds while increasing the standard deviation σ at intervals of about 1 second, and the ζ-σ plot (FIG. 8)
And automatically estimate the number n of components. Here, the user can also change the component n.

Next, in step 204, DFA processing is executed. In other words, based on the uniqueness of factor analysis,
A deconvolution process is performed using the optimal standard deviation σ. Decomposition is optionally terminated using rank annihilation. Then, the process proceeds to step 205
And outputs the spectrum matrix X ((3) in FIG. 4) and the elution profile matrix Y (FIG. 5). Here, quantitative qualitative analysis becomes possible.

Next, in steps 206 and 207, the user can change the number n of components and the data matrix D as necessary. When changing the number n of components or the data matrix D, the process returns to step 203. Step 20
In steps 6 and 207, when the number of components n and the like are not changed, the processing is terminated.

Note that, based on the uniqueness of the factor analysis,
When the uniqueness is not directly used as a reference, the degree of coincidence of spectra (FIG. 7) and the degree of coincidence of rotation angles are used as references. In practice, the condition is that the degree of coincidence is within 1% or the rotation angle is within 1 degree. In this case, the user needs to tentatively select a method serving as a reference for orthogonal or oblique rotation from the rotation references shown in FIG. FIG.
The oblique rotation 8 in 7 is referred to as the positive max for convenience.

FIG. 18 is a diagram showing an example of a three-dimensional chromatogram obtained by performing deconvolution processing without performing factor analysis processing, and is a comparative example between the present invention and the prior art. As shown in FIG. 18, when the three-dimensional chromatogram is simply deconvolved,
Noise and spurious peaks appear. FIG. 18B shows a cross section taken along line AA ′ of FIG. Thus, simply deconvolution processing of a three-dimensional chromatogram is equivalent to the existence of a two-dimensional chromatogram having several different wavelengths. In this case, the information of the entire three-dimensional chromatogram is not used and processed, and the disadvantage that the noise and the pseudo peaks that were present in the case of the two-dimensional chromatogram are amplified remains. .

On the other hand, according to the embodiment of the present invention, the spectrum and elution profile of the unknown component can be obtained by utilizing the large amount of information of the multi-channel chromatogram. As a result, overlapping peaks can be separated and resolved, and component identification and quantitative calculation can be performed with high accuracy.

In the above-described embodiment of the present invention, the deconvolution processing and the factor analysis are combined. However, the present invention is not limited to the factor analysis, and other multivariate analysis methods may be combined with the deconvolution processing. It is also possible.

That is, for example, regression analysis, which is one of the multivariate analysis methods, can be combined with deconvolution processing to improve the method for resolving overlapping peaks in a multi-channel chromatogram. In the above-described conventional nonlinear least square method, the regression coefficient is more difficult to determine as the degree of overlapping of the peaks is larger.

For this reason, if a regression analysis is performed on the data matrix d that has been subjected to the deconvolution processing, the peak overlap is weaker than that of the original data matrix D, so that the regression coefficient can be obtained relatively easily. In order to restore the original peak waveforms, the determined regression function such as Gaussian is subjected to a reconvolution process. Alternatively, a method of estimating the initial value of the regression coefficient for the matrix D based on the regression function obtained by the reconvolution processing and performing regression analysis again is also effective.

Although the above-described example is an example in which the present invention is applied to a drug monitoring HPLC system, the present invention is not limited to the above-mentioned system, but is applicable to other chromatogram analysis systems.

Further, in the above example, the characteristic component of the object to be measured is the absorbance. However, the present invention can be applied to the case of the fluorescence intensity.

The multi-channel chromatogram has three intensity components: quantitative intensity, qualitative intensity, and retention intensity. The qualitative component includes a wavelength component, a fluorescence wavelength, a mass unit (m / z) of a mass spectrometer, an electrochemical detector (E
CD) oxidation-reduction potential. And the mass mentioned above
The present invention is also applicable to units and the like.

[0095]

The multi-channel chromatogram is configured to be analyzed using multivariate analysis and deconvolution processing, so that the spectrum and elution profile of the unknown component are obtained by small-scale deconvolution processing, and overlapping peaks are separated.・ Decompose, and component identification and quantitative calculation can be performed accurately.

[Brief description of the drawings]

FIG. 1 is an operation flowchart of a multi-channel chromatogram analysis method according to an embodiment of the present invention.

FIG. 2 is a schematic configuration diagram of a multi-channel chromatogram analyzer according to one embodiment of the present invention.

FIG. 3 is a diagram showing a data display example of a multi-channel chromatogram.

FIG. 4 is a waveform diagram showing a matrix of components subjected to deconvolution processing.

FIG. 5 is a waveform chart showing a spectrum intensity matrix of a component k.

FIG. 6 is a waveform diagram of a data matrix subjected to deconvolution processing.

FIG. 7 is a graph showing a change in the degree of coincidence according to the standard deviation σ.

FIG. 8 is a graph plotting ζ and σ.

FIG. 9 is a schematic diagram of a validity check of a deconvolution process.

FIG. 10 is a schematic diagram of a validity check of the entire DFA process.

FIG. 11 is an explanatory diagram of an overlapping peak of two components.

FIG. 12 is an explanatory diagram of an overlapping peak of three components.

FIG. 13 is a graph for explaining indeterminacy of a rotation angle.

FIG. 14 is a graph showing an example of an elution profile having two maxima.

FIG. 15 is an operation flowchart of execution of a DFA process by an operator.

FIG. 16 is a diagram showing a display example of a menu of a disassembly method.

FIG. 17 is a diagram illustrating a display example of a rotation reference.

FIG. 18 is a waveform chart for comparing the present invention with a conventional technique.

[Explanation of symbols]

 Reference Signs List 40 pump 41 column 42 diode array detector 43 sampler 44 control unit 45 data analysis unit 46 CRT 47 printer 451 data matrix determination unit 458 deconvolution unit 461 component identification quantitative calculation unit 462 elution profile matrix calculation unit 463 spectrum matrix calculation unit

Continuation of the front page (56) References JP-A-7-270387 (JP, A) JP-A-2-120662 (JP, A) JP-A-60-104238 (JP, A) JP-A-63-308560 (JP) , A) (58) Field surveyed (Int. Cl. ⁷ , DB name) G01N 30/86

Claims (7)

(57) [Claims] 1. A method for analyzing overlapping peaks on a multi-channel chromatogram, comprising: a step of deconvolution processing the multi-channel chromatogram using a predetermined spread function; and performing a multivariate analysis from the processed multi-channel chromatogram. Determining the width of the predetermined spread function and the corresponding processed multi-channel chromatogram based on the isolation of the peak; and obtaining a spectrum waveform from the processed multi-channel chromatogram. A method for analyzing a multi-channel chromatogram, comprising: 2. A method for analyzing overlapping peaks on a multi-channel chromatogram, comprising: a step of performing deconvolution processing on the multi-channel chromatogram using a predetermined spread function; and a step of performing factor analysis on the processed multi-channel chromatogram. Determining from the relationship between the width of the predetermined spread function and the corresponding factor analysis solution whether or not a predetermined variable of the multi-channel chromatogram is stable; and Multi-channel chromatogram analysis, comprising: determining a width of a function and a corresponding multi-channel chromatogram after processing; and obtaining a spectrum waveform from the multi-channel chromatogram after processing. Method. 3. A method of analyzing overlapping peaks on a multi-channel chromatogram, wherein a step of deconvolution processing the multi-channel chromatogram using a predetermined spread function, and a step of factor-analyzing the processed multi-channel chromatogram. Determining whether the factor analysis corresponding to the width of the predetermined spread function has a unique solution; and determining the width of the spread function and the width of the spread function based on the fact that the factor analysis has a unique solution. A method of analyzing a multi-channel chromatogram, comprising: a step of determining a multi-channel chromatogram after processing; and a step of acquiring a spectrum waveform from the multi-channel chromatogram after processing. 4. A method for analyzing overlapping peaks on a multi-channel chromatogram, comprising the steps of: deconvolution processing the multi-channel chromatogram using a predetermined spread function; and solving an eigenvalue problem from the processed multi-channel chromatogram. A step of determining whether a peak is isolated from the relationship between the width of the spread function and the solution of the corresponding eigenvalue problem; and, after the processing of the width of the spread function and the corresponding processing, based on the fact that the peak is isolated. A method for analyzing a multi-channel chromatogram, comprising: a step of determining a multi-channel chromatogram of: and a step of obtaining an elution profile from the multi-channel chromatogram after the above processing. 5. The method of analyzing a multi-channel chromatogram according to claim 4, wherein the step of determining the multi-channel chromatogram further comprises specifying the number of components from the solution of the eigenvalue problem and the width of the spread function. A method for analyzing a multi-channel chromatogram. 6. A data processing apparatus for analyzing overlapping peaks on a multi-channel chromatogram, comprising: a storage unit for a spread function width; a deconvolution processing unit for deconvolution processing the spread function stored in the storage unit; A factor analysis unit that performs factor analysis on the deconvoluted multi-channel chromatogram, a storage unit that stores the solution of the eigenvalue problem of the multi-channel chromatogram that is subjected to the factor analysis, and a spread function determined based on the solution of the eigenvalue problem An output unit for outputting a spectrum waveform obtained from a multi-channel chromatogram subjected to deconvolution processing using the output unit. 7. The data processing apparatus according to claim 6, wherein the output unit further specifies the number of components from the solution of the eigenvalue problem and the width of the spread function.