JPH07119720B2 - Method of selecting components from standard data in composition analysis of mixtures - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Object of the Invention (Industrial field of application) The present invention relates to standard data of a sample substance (which refers to a chemical substance registered in a database as standard spectral data), to a sample substance mixture (multiple The peak intensity value for the chemical shift value of ^13).
The present invention relates to a method for selecting components from standard data in the compositional product of a mixture by a system for analyzing and identifying a sample substance in a sample substance mixture from measurement of a C-NMR spectrum.

(Prior Art) In general, composition analysis of a mixture is performed by column chromatography, gas chromatography (high performance liquid chromatography, etc.) when each component can be separated, but separation of each component is difficult. In the case of simultaneous analysis of substances with greatly different objects or structures, they cannot be applied, so other methods must be used.

Therefore, recently, researches and developments such as nuclear magnetic resonance (NMR) and electron paramagnetic resonance (ESR), which are effective for analysis of material structure using magnetic resonance, have been advanced in various fields. These are based on the principle that, when the energy levels of particles with a magnetic moment are separated in a static magnetic field, resonance occurs with an oscillating magnetic field or electromagnetic wave having a frequency (frequency) corresponding to the separation interval. It is a thing. However, due to the nature of the detection level, it is nuclear magnetic resonance that is more commonly used as a magnetic sensor, and others are used to standardize more perceptual atomic level magnetic quantities.

Here, ¹³ C-NMR in which nuclear magnetic resonance was applied to chemical analysis
The ¹ H-NMR apparatus will be described based on the Fourier transform nuclear magnetic resonance (FT-NMR) apparatus shown in FIG. This FT
First, the NMR apparatus synchronizes the frequency signal a from the high frequency oscillator 1 and the rectangular pulse signal b from the pulse generator 3 with a predetermined period at the gate 2 in order to obtain the variable synchronizing signal c. The CPU 9 variably controls the frequencies of the high frequency oscillator 1 and the pulse generator 3, and the synchronization signal c is transmitted to the sample W as an amplified synchronization signal d by the power amplifier 4. Sample W
Is set in a solubilized state such that ¹³ C or ¹ H is contained in a static magnetic field by the magnetic poles N and S in a natural abundance ratio, and when an amplified signal d including a predetermined frequency is received to cause resonance, The resonance signal d'is emitted and transmitted to the detector 5.

In the detector 5, the relationship between the chemical shift value with respect to the sample substance and the integral value for each spectral region for obtaining the structural parameter from the distribution of the peak intensity due to nuclear magnetic resonance containing ¹³ C or ¹ H in advance. Obtain information on the spectrum of nuclear magnetic resonance specific to the sample substance, and exchange the sample W with the CPU9.
It can be configured to perform search and identification as measurement of ¹³ C-NMR spectrum and ¹ H-NMR spectrum. However, when ¹ H-NMR is used, the spectrum line becomes complicated and it is difficult to apply it to the composition analysis of the mixture.
On the other hand, in the case of using ¹³ C-NMR, for example, one carbon line in an organic compound corresponds to one spectral line, which facilitates analysis and is effective as a composition analysis of a mixture.
Here, ¹³ C-NMR is used.

Further, since the resonance waveform e detected by the detector 5 is minute, it is amplified by the amplifier 6 to be amplified resonance waveform f.
After removing the noise component through the filter 7, the detected resonance waveform g is obtained. In addition, the detected resonance waveform g is the digital detection resonance waveform h because of the calculation processing by the CPU 9.
Is quantized by the A / D converter 8 so as to be transmitted as.
In this way, the digital detection resonance waveform h can be input to the CPU 9, but the detection resonance waveform g that is the original analog signal
In order to reconstruct the signal, the Fourier transform (F
T) is performed. This Fourier transform (FT) is the sample W
In order to obtain the ¹³ C-NMR spectrum of, the CPU 9 obtains the continuous frequency component of the time function related to the digitally detected resonance waveform h from the nuclear magnetic resonance spectrum corresponding to the chemical shift value. Thus, from the D / A converter 10, the detected resonance waveform g of the sample W based on the predetermined frequency signal a is
It is detected as a peak intensity signal of a chemical shift value based on a spectrum line including the resonance frequency, and this is recorded on the recorder 11 while being displayed on the display 12, so that the components of the sample W (a single chemical substance or a plurality of chemical substances) can be recorded. The distribution of peak intensities peculiar to the mixture of substances) is obtained depending on the chemical shift value.

With such an FT-NMR apparatus, it is easy to measure ¹³ C-NMR spectra by introducing a chemical shift value into the resonance spectrum analysis result of chemical substances (substances that can be measured by ¹³ C-NMR containing carbon nuclei). A standard data of a sample substance is created in advance for a part of the sample substance mixture, thereby realizing a method of performing a composition analysis from a search based on the standard data using a computer (CPU9). still,
When quantitative analysis is performed from the measurement of the ¹³ C-NMR spectrum, a method of mixing an internal standard substance as a measurement solvent is used.

(Problems to be solved by the invention) Based on the standard data created in advance as described above, ¹³
When a mixture component is searched and identified from the measurement of C-NMR spectrum, selection from the standard data for the sample substance as the component becomes complicated. The standard data basically stores the measurement results of ¹³ C-NMR spectra for various sample substances, and has peak intensity values and chemical shift values of each sample substance.

However, if each sample substance that makes up the sample substance mixture exhibits a unique single peak intensity distribution, it is not so difficult to identify it, but in reality, the peak intensity of multiple sample substances is Since the value is contained in each peak intensity value of the sample substance mixture, and a plurality of candidates are selected for attribution from the standard data, identification of the composition for it becomes a problem. When targeting a sample substance mixture that contains sample substances that have similar structures, rather than an arbitrary sample substance, assigning all peaks and performing an accurate composition calculation by the least squares method results in a huge calculation. It will be impossible to calculate at the personal computer level. Therefore, a selection method for composition analysis that accurately and quickly calculates sample substance mixtures consisting of more general types of sample substances and sample substance mixtures consisting of complex sample substances with different structures Has not been realized.

The present invention has been made under such circumstances, and an object of the present invention is also to provide a sample substance mixture containing various types of sample substances and a sample substance mixture containing various sample substances consisting of complex chemical substances having different structures. ^In order to facilitate the assignment of ¹³ C-NMR spectrum measurement results from standard data, preprocessing is performed only on the portions with reliable peak assignments before the final composition calculation, and subsequent final composition calculations are performed. It is an object of the present invention to provide a method for selecting components from standard data in composition analysis of a mixture, which enables the calculation of the composition ratio of the sample substance to be carried out with high certainty.

Configuration of the invention; (Means for Solving the Problems) The present invention, in advance to create a standard data by measuring the ¹³ C-NMR spectrum of the sample material, for sample material mixture ¹³
A method for selecting components from standard data by a system for searching and identifying the composition analysis of the sample substance mixture based on the standard data from the measurement of C-NMR spectrum,
The above object of the present invention is based on the chemical shift value specific to each sample substance presented by the ¹³ C-NMR spectrum with respect to the distribution of the peak intensity of the sample substance mixture, and based on the standard data, the sample substance mixture. The multiple regression method (identification of the components of the sample mixture from the multiple regression model by least-squares calculation) based on the value of the composition ratio that temporarily sets the model formula formed from the assigned components The method of performing the composition calculation for performing the calculation without limiting the calculation so that the obtained composition ratio can take any of positive, negative and zero values) or the least squares method (the least squares calculation from the multiple regression model) The composition calculation for identifying the components of the mixture is performed by the method of limiting the calculation so that the obtained composition ratio can be either positive or 0), and the result is obtained. When the value of each composition ratio is not a positive value in the multiple regression method and is a component close to 0 in the least squares method, the component substance corresponding to the corresponding composition ratio is deleted,
Based on the model formula after this deletion, the values of the respective composition ratios are sequentially changed until the corresponding component disappears, and repeated calculation is performed by the multiple regression method or the least squares method. This is achieved by obtaining the composition ratio when the human being disappears.

(Function) The present invention sets a multiple regression model by assigning the sample substance from the unique chemical shift value of each sample substance constituting the mixture based on the measurement result of the ¹³ C-NMR spectrum of the sample substance mixture. , A method of identifying the composition of the sample substance mixture by selecting the sample substances constituting the sample substance mixture from the standard data based on the peak intensity value of each sample substance. Here, first, paying attention to the distribution of peak intensities peculiar to each sample substance from the measurement results of the ¹³ C-NMR spectrum of the sample substance mixture, the sample substance was determined from the standard data based on the chemical shift value peculiar to each sample substance. Attribute uniquely. In the standard data, peak intensity values and various chemical shift values thereof for various sample substances are stored, and here, they are uniquely assigned based on the chemical shift values. This assignment does not have a large effect on the component selection even if it is rough. Therefore, the assignment is performed only for the sample substance mixture having a relatively easy interval between chemical shift values, and not for the portion having a narrow chemical shift interval. Therefore, the peak group in the portion where the chemical shift interval is narrow is ignored in the composition calculation according to the present invention.

Thereafter, in the present invention, for example, Gauss-elimination (Gaus
s-Newton) multiple regression calculation or least squares calculation is introduced to calculate the composition of the attribution component uniquely assigned from the standard data as described above. That is, as far as the multiple regression model, which is the attribution component of the sample substance mixture set based on the chemical shift value of each sample substance, it is the peak intensity value of each sample substance that is considered to be the most probable without error, the standard The composition of the sample substance mixture is calculated based on the peak intensity values of the plurality of sample substances assigned from the data. The composition ratio can be calculated by determining the correlation coefficient of regression. That is, since the present invention is a composition analysis targeting a sample substance mixture, confirmation processing regarding presence / absence of overlapping components and redesignated components obtained by recombining component substances is performed by changing the setting of the multiple regression model. This also calculates the regression coefficient by the least-squares calculation. That is, due to the nature of composition ratio calculation by multiple regression calculation or least square calculation, 0 or a negative component may actually be calculated. In such a case, it may be regarded as a non-corresponding component and automatically deleted, but that alone is insufficient for analysis. Therefore, if a sample substance that is not a positive value is included, check the redundancy with a multiple regression model that targets the selected sample substance, and if a sample substance that is not a positive value is included, see Check the respecified components in the regression model.

That is, the multiple regression model is set as follows. When the intensity of a certain peak of the sample substance mixture is Mi, the composition ratio of the sample substance that includes the peaks attributed to the above-mentioned peaks of the sample substance mixture among the assigned sample substances (here, the three components are assumed) ) Are variables A, B, C, and the peak intensities (known values) of the respective sample substances are ai, bi, ci, the following equation Mi = A · ai + B · bi + C · ci for the peak of an arbitrary sample substance mixture. ... (1) is obtained. Usually, the number of peaks of the sample substance mixture is larger than the number of assigned sample substances, and a multiple regression model is set. Such a multiple regression model can be set for all peaks in the sample substance mixture, except for the peaks which are ignored due to narrow shift intervals. The composition ratio (A, B, C) of the sample substance mixture can be obtained by solving the equation (1). By the way, there is a case where 0 or a negative component is present in the composition ratio obtained from the equation (1), and it is highly possible that this component is not actually contained in the mixture. However, in the present invention, for confirmation, the composition ratio of the negative component in the obtained composition ratio is set to 0, and the peak intensity of each peak of the sample substance mixture is stored in the standard data as the component. The simulation is performed using each peak intensity, and the composition ratio is further varied, that is, the component material having a composition ratio of 0 is also varied to the positive side for the simulation. The simulation is performed until the difference between the simulation result and the actually measured peak intensity is minimized. This is judged by the sum of squares of all peak intensity differences.

When the composition ratio 0 or a component close to 0 obtained from the above simulation matches 0 and the negative component of the composition ratio initially obtained, this component was detected by attribution, but in reality, the sample substance mixture It is determined that the component is not contained in and deleted from the composition. After that, the multiple regression model is set and the simulation is performed again using the remaining composition components, and the components that are not actually contained are deleted again. The above process is repeated several times to complete the component selection. Judgment of termination is made when all composition ratio values to be compared are either positive values of multiple regression method or least squares method, or one value of multiple regression method or least squares method Is positive, the comparison is performed when the other value becomes 0. Next, the peaks of the attribution components are assigned to almost all the peaks in the sample substance mixture using the above components, and the composition is calculated to obtain the final result.

(Examples) Hereinafter, the present invention will be described in detail with reference to Examples.

FIG. 1 is a flow chart showing an operational processing example of the method for composition analysis of a sample substance mixture based on the sample substance assigned from the standard data with respect to the measurement result of ¹³ C-NMR spectrum of the present invention. First, when ¹³ C-NMR spectrum measurement is performed on a sample substance mixture, the peak intensity value for each sample substance appears uniquely on the chemical shift value, so after performing the pretreatment necessary for signal processing, the composition calculation The component A is designated (step S1). Next, the sample substance is uniquely assigned as the attributed component from the standard data based on the chemical shift value specific to each sample substance (step S2). The composition calculation for identifying the components of the sample mixture by the least squares calculation from the multiple regression model for this assigned sample substance is performed so that the underlying composition ratio can take any value of positive, negative and zero. A simple composition calculation by the multiple regression method (step S3), which is a method without any restriction on the calculation, and a composition calculation for identifying the components of the sample mixture by the least squares calculation from the multiple regression model, are obtained. A simple composition calculation by the least-squares method, which is a method to limit the calculation so that it can take only one value or 0 (step S4)
And are executed in parallel. Then, the sample substance B having a composition ratio of 0 or negative by the multiple regression method and the sample substance C having a composition ratio of close to 0 by the least squares method are combined (step
S5), the duplicate components are checked (step S6), and the duplicate sample substance D is deleted (step S7). This sample substance D is judged not to be involved in the composition of the sample substance mixture even if it is retrieved from the standard data. After that, the remaining sample substance (A-B) by the multiple regression method, the remaining sample substance (B + C-D) after the overlapping component check, and the remaining sample substance (A-C) by the least squares method are used. To respecify the sample substance A '(= A-D) (step S
8). D = 0 for this redesignated sample material A ′,
That is, it is checked whether or not there are overlapping components (step S
9) If there is an overlapping component (D ≠ 0), the process returns to the first steps S3 and S4, the composition ratio is successively changed, and the above-described arithmetic processing is repeated. Then, when there is no overlapping component, that is, when D = 0, when all the composition ratio values to be contrasted are either positive values of the multiple regression method or the least squares method, Alternatively, if one of the values of the multiple regression method or the least squares method is positive and the other value to be compared becomes 0, the routine proceeds to the final composition calculation routine by reassigning peaks.

FIG. 2 (A) shows the peak intensity value and chemical shift value which are the measurement results of ¹³ C-NMR spectrum for the sample substance mixture, and FIGS. 2 (B) to (F) correspond to this chemical shift value. Substances stored in standard data α ₀ , β ₀ , α ₃ ,
The example of the spectrum data of (lambda) ₂ , (alpha) ₁ is shown. here,
If the sample substance mixture is composed of four kinds of sample substances A, B, C, and D from the measurement results of ¹³ C-NMR spectrum shown in FIG. 2 (A), according to the measurement of ¹³ C-NMR spectrum, The sample substances A, B, C and D exhibit independent peak intensity values Y on their respective chemical shift values X [ppm]. Therefore, the intensity value indicated by each sample material of a sample substance mixtures _{A p, B p, C p} , D p from its respective unique chemical shift values _{A s, B s, C s} , based on the D _s, Fig. 3 It is possible to easily assign the sample substance from the standard data prepared in advance as shown in. Sample materials (α ₀ , α ₁ , ...), (β ₀ , β ₁ , ...), (λ ₀ , λ ₁ ,
...) are just a few candidate examples, and actually, chemical shift values and peak intensity values for many kinds of substances are also stored. Thus, each sample material A shown in FIG. 2 (A), B, C, each chemical shift value A _s of D, B _s, C _s, with respect to the assignment of sample material from the standard data based on the D _s, for example Sample substance
The chemical shift values A _s and D _s of A and D, respectively (E),
Substances λ ₂ and α ₀ as shown in (B) are assigned to correspond to a single substance, and the chemical shift value B _s of the sample substance B is shown in (C), (F) of FIG. For the chemical shift value C _s , the sample substance β ₁ as shown in (D) and (F) of FIG.
α ₁ and α ₃ are redundantly assigned.

Due to such attribution of sample substances, when the ¹³ C-NMR measurement result for the sample substance mixture M is composed of four kinds of sample substances as shown in FIG. 2 (A), first, the respective peak intensities of the respective sample substances are Each peak M _pi (known) of the sample substance mixture M based on the values (A _pi , B _pi , C _pi , D _pi ) (known) and the composition ratio of the sample substance (b, c, d, e) (unknown) The multiple regression model for M _pi = a + b * _Api + c * _Bpi + d * Cpi + e * _Dpi ………
(2) Set like this. However, although the error term a is provided here, the setting of the multiple regression model can be arbitrarily performed and is not limited to this. Basically, there are a plurality of attribution components for each sample substance, and a least squares operation is performed to calculate each regression coefficient a, b, c, d, e based on this attribution component. For example, if the error of the peak intensity value M _pi of the sample mixture M is Z _pi , then Z _pi = M _pi − (a + b · A _pi + c · B _pi + d · C _pi + e ·
It is expressed as D _pi ) ... (3), and the sum of squares S of the error is assumed in order to obtain each regression coefficient which is these unknown constants. The composition ratio is calculated by obtaining each regression coefficient a, b, c, d, e that minimizes. As an initial condition, (∂S / ∂a)
= 0, (∂S / ∂b) = 0, (∂S / ∂c) = 0, (∂S / ∂d) =
Solving the five normal equations satisfying 0, (∂S / ∂e) = 0 gives a unique solution, but in reality it is complicated, so its explanation is omitted here. An electronic calculator may be used to calculate the regression coefficient of such a multiple regression model. By such a procedure, the composition of the sample substance mixture M can be identified. But,
The composition ratio of the sample substance mixture calculated from the attribution component of each sample substance may be determined to be 0 or a negative component that does not correspond to the composition ratio according to the calculation result of each regression coefficient by the least-squares calculation. In that case, it is considered that there is a high possibility that the selected component with respect to the attribution component is not included in the composition of the sample substance mixture M, but in the present invention, the following confirmation processing is further performed by the least squares calculation. For example, as an example of the operation processing, the negatively determined attribution component is set to 0 on the multiple regression model to make a modified multiple regression model, and the remaining attribution component is included in the overlapping components for each sample substance. It is checked based on the regression coefficient of the modified multiple regression model. Next, the negatively determined attributed component is regarded as positive, a modified multiple regression model is set while varying the composition for that component substance, and the redesignated component is checked. The least-squares calculation at this time is sequentially performed as calculation of each regression coefficient for each modified multiple regression model, separately from each regression coefficient calculated by the least-squares calculation for the initial multiple regression model. Furthermore, if 0 or a negative judgment is also made by the confirmation processing, it is considered that there is no room to be included in the sample substance of the sample substance mixture M even if it is attributed and selected from the standard data,
The relevant sample substance is deleted from the components of the composition.

The multiple regression model is set again for each sample substance corresponding to each of the remaining constituents, and all the sample substances constituting the sample substance mixture M are selected. Figure 2 (A)
In the case of the sample substance mixture M in (1), the multiple regression model is sequentially changed and settings are iterated so that a, b, c, d are given positive values. That is, the selection of the sample substance is completed by obtaining the set multiple regression model so that the regression coefficients for all the component substances of the sample substance mixture M can be obtained with positive values, and the composition of the sample substance mixture can be identified. . This is because, in the multiple regression model in which the regression coefficient for each sample substance is obtained as a positive value, the regression coefficient shows the composition ratio and shows the composition identification of the attribution component. FIG. 4 shows an example of the calculation result by the multiple regression, and FIG. 5 shows an example of the calculation result by the least square method.

Although the Gauss-elimination multiple regression method has been described above, other methods such as the Damping Gauss-Newton method and the Marquardt method may be used.

EFFECTS OF THE INVENTION As described above, according to the present invention, the sample substance which is each constituent component can be easily attributed from the standard data to the measurement result of the ¹³ C-NMR spectrum of the sample substance mixture, and the mathematical component A method has been realized which introduces statistical least squares operation to obtain composition identification with high certainty. ¹³ C-NMR
When a spectrum was measured for a mixture, it was a problem in the past because the constituents of various kinds have different structures or the structures are different from each other, and the selectivity from standard data becomes complicated. By using the composition identification method based on the components, rapid composition analysis can be performed in a personal computer level system even for various kinds of mixtures, and application to qualitative / quantitative analysis can be realized.

[Brief description of drawings]

FIG. 1 is a flow chart showing a method for analyzing the composition of a sample substance mixture based on the attribution component of the sample substance assigned from the standard data for the measurement result of ¹³ C-NMR spectrum of the present invention, FIG. 2 (A). Is a diagram showing, as an example, the distribution of peak intensities for each sample substance that is a constituent component by measurement of the ¹³ C-NMR spectrum of the sample substance mixture.
(F) is a diagram showing chemical shift values and peak intensities corresponding to respective sample substances, FIG. 3 is a diagram showing an example of standard data, FIG. 4 is a diagram showing an example of calculation results by multiple regression, and FIG. 5 is FIG. 6 is a diagram showing an example of a calculation result by the least-squares method, and FIG. 6 is a diagram showing an example of an apparatus for Fourier transform nuclear magnetic resonance. 1 ... High-frequency oscillator, 2 ... Gate, 3 ... Pulse generator, 4 ... Power amplifier, 5 ... Detector, 6 ... Amplifier,
7 ... Filter, 8 ... A / D converter, 9 ... CPU, 10 ...
D / A converter, 11 …… Recorder, 12 …… Display, X
... chemical shift value, Y ... peak intensity value, M ... sample substance mixture.

Claims (1)

[Claims] 1. A standard substance is prepared by measuring a ¹³ C-NMR spectrum of a sample substance, and the ¹³
In the system for searching and identifying the composition analysis of the sample substance mixture based on the standard data from the C-NMR spectrum, the measurement of the ¹³ C-NMR spectrum exhibits the distribution of the peak intensities of the sample substance mixture. Based on the chemical shift value specific to the sample substance, the sample substance constituting the sample substance mixture is uniquely assigned from the standard data, and the value of the composition ratio tentatively set the model formula formed from the assigned sample substance When the value of each composition ratio is not a positive value in the multiple regression method and is a component close to 0 in the least squares method, the corresponding composition is calculated by the multiple regression method or the least squares method. The sample substance corresponding to the ratio is deleted, and the values of the composition ratios are sequentially changed based on the model formula after the deletion until the corresponding components are eliminated by the multiple regression method or the least squares method. A composition analysis of a mixture characterized in that a composition is identified by performing a recalculation and obtaining a composition ratio when all the composition ratio values to be contrasted do not satisfy the above conditions. Method for selecting components from standard data in.