JP5679257B2 - Measurement data acquisition and evaluation method - Google Patents

Description

  The present invention relates to a method for acquiring and evaluating measurement data. For example, the present invention relates to quantitative data suitable for use in, for example, metabolome measurement of total metabolites of cells by capillary electrophoresis (CE) -mass spectrometry (MS). The present invention relates to a method for acquiring and evaluating measurement data that can provide an index of sex.

Quantifies the components in the sample (for example, mixed solutions of metabolites extracted from biological samples consisting of solutions containing components extracted from samples such as blood, cerebrospinal fluid, urine, sweat, tears, organs, tissues, cultured cells, and culture media) When analyzing, the operation is generally performed according to the following procedure.
Preparation of original sample → Extraction of target substance → (convert to detectable format) → Detection → Data processing

  Here, “converting to a detectable form” means derivatization, enzymatic reaction, separation operation, and the like. Since each of these procedures includes an error, the numerical value finally obtained is a stack of these errors.

  In the current analysis flow, in order to evaluate the quantitative reliability of these procedures, it is common to use a method of creating a calibration curve by adding a standard substance (Non-patent Document 1). However, the operation is complicated and the case where the matrix effect cannot be eliminated with an actual sample or when the standard substance is difficult to obtain (including the case of an unidentified substance) cannot be applied. Although the sample itself may be diluted and measured, it has no substantial meaning because the matrix effect changes.

  As a method for eliminating the matrix effect, an internal standard method (offset by addition of an internal standard substance that is expected to have a matrix effect equivalent to the measurement target) has been used (Non-patent Document 1). However, since it is impossible to predict the matrix effect of all signals with a small number of internal standard substances, it is not practical in omics that performs analysis using multi-component simultaneous analysis data.

  In addition, a standard substance addition method in which several concentrations of standard substances are added to a sample and obtained by extrapolation is used. However, this method cannot be applied when it is difficult to obtain a standard substance (including the case where it is an unidentified substance). Further, in mass spectrometry, since the signal intensity of the parent ion is affected by the generation rate of the product ions, it cannot be determined whether or not the quantitative reliability is lowered between samples.

  On the other hand, data analyzed by an analytical instrument (for example, a mass spectrometer) includes many noise signals including random noise, spike noise, ringing noise, and the like generated during mass analysis, as illustrated in FIG. ing. When investigating the difference between samples, it is required to correctly associate the peak of the compound. However, in the case of data containing a large amount of noise, a great deal of labor is required to associate them accurately. However, a sufficiently efficient noise removal method has not been devised so far (Non-Patent Documents 2 to 4).

  For example, a technique for reducing noise peaks by performing baseline smoothing is employed in many data processing software. However, there are problems that small peaks become flat, large noise cannot be eliminated, and data itself changes.

  A blank subtraction method has also been proposed in which a blank chromatogram that does not contain a sample is smoothed and subtracted from the original sample data (Patent Document 1). However, there is a problem that large noise and noise derived from the sample cannot be excluded.

  Further, a cut-off method based on an S / N ratio in which a threshold value is set using an S / N ratio as an index and a device having a lower level is excluded as noise is generally used. Therefore, the noise elimination performance is low. In many cases, the basis for setting the threshold is poor.

  In addition, since noise signals are generally low in intensity, there is a cutoff method using a signal intensity threshold that sets a threshold value for signal intensity and deletes the signal intensity below it. However, signals that are low but necessary are also excluded. For this reason, the noise elimination performance is set low (Non-Patent Document 5, Patent Document 2). In many cases, the basis for setting the threshold is poor.

  In addition, substance-derived signal peaks detected in common between samples are expressed as a certain standard (electrophoresis time for CE, elution time for LC (liquid chromatograph) or GC (gas chromatograph), m / for MS. z, an absorption wavelength in the case of an absorptiometer, an excitation wavelength and an emission wavelength in the case of fluorescence analysis, as an index, and the alignment method is repeated as a technique for improving alignment accuracy between measurement data and removing noise. Measurement is often used. This is a method in which the same sample is measured a plurality of times, a signal peak detected in common is left, and the others are processed as noise (Non-Patent Documents 6 to 8). This method has high noise rejection performance, but the number of measurements is greatly increased, and a series of ions derived from the parent ion including multivalent ions, multimer ions, metal addition ions, fragment ions, etc. generated during mass analysis. There is a problem that a product ion which is a group cannot be excluded. In addition, a method has been proposed in which the elution time and the migration time are corrected using a nonlinear function to increase the alignment accuracy (Non-Patent Document 9). However, when the number of signals to be obtained is large, it is not possible to discriminate a signal having a very close elution time or electrophoresis time, so that the alignment accuracy is low.

  In addition, a peak selection method using a reference library is also proposed in which data measured in the past is accumulated to form a library, and the data is compared for each measurement to eliminate noise (Non-patent Documents 10 and 2). ~ 4). Although this method is practical, it cannot cope with differences in platforms, and has the problems that it is necessary to re-create it, and that there is ambiguity in the criteria because of probabilistic judgment.

In addition, regarding the removal of the product ions, the product ions may be adducts of metal ions (Na ⁺ , K ⁺ , Mg ²⁺ , Mn ^2+, etc.) and some known fragment ions (formic acid desorption, water desorption, ammonia desorption). Etc.) is easy to find because m / z can be calculated. However, there is a problem that fragments and adducts (cleavage separation, addition of specific impurities, etc.) peculiar to the compounds and analysis methods are unpredictable.

  In addition, regarding the mixing of materials, in proteomics using two-dimensional electrophoresis, a sample in which all the samples to be measured are mixed in equal amounts is prepared, and the signals of other samples are relatively quantified based on the signal intensity of the mixed sample. A technique is used (Non-Patent Document 11). However, this value was not used for the reliability of measurement data and noise removal, but the purpose was to standardize between gels.

Japanese Patent Laid-Open No. 10-339727 JP 2005-55370 A Japanese Patent Laid-Open No. 2000-131284 Special table 2007-575644 gazette

Analytical Society of Japan Kyushu Chapter Introduction to Instrumental Analysis Revision 3 Nanedo Matsuda Fumio, Oikawa Akira, Kusano Miyako, Kikuchi Satoshi, Saito Kazuki. Current Status and Prospects of Metabolome Analysis Technology. 2. Data Processing Technology. Chemistry and Biology 45: 834-842, 2007. Yoshiaki Ohashi. Using metabolomics well. Bioscience and industry 65: 8-13, 2007 Fiehn, O., Wohlgemuth, G., Scholz, M., Kind, T., Lee, D.-Y., Lu, Y., Moon, S., and Nikolau, B. Quality control for plant metabolomics: reporting MSI-compliant studies.Plant J. 53: 691-704, 2008. Morohashi, M., Shimizu, K., Ohashi, Y., Abe, J., Mori, H., Tomita, M., and Soga, T. P-BOSS: a new filtering method for treasure hunting in metabolomics. Chromatogr. A. 1159 (1-2): 142-148, 2007. Jonsson, P., Johansson, AI, Gullberg, J., Trygg, J., A, J., Grung, B., Marklund, SL, Sjostrom, M., Antti, H., and Moritz, T. High- throughput data analysis for detecting and identifying differences between samples in GC / MS-based metabolomics analyzes.Anal. Chem. 77: 5635-5642, 2005. Jonsson, P., Gullberg, J., Nordstrom, A., Kusano, M., Kowalczyk, M., Sjotrom, M., and Moritz, T. A strategy for identifying differences in large series of metabolomics samples analyzed by GC / MS. Anal. Chem. 76: 1738-1745, 2004. Jonsson, P., Bruce, SJ, Moritz, T., Trygg, J., Sjostrom, M., Plumb, R., Gramger, J., Maibaum, E., Nicholson, JK, Holmes, E., and Antti , H. Extraction, interpretation and validation of information for comparing samples in metabolic LC / MS data sets. Analyst 130: 701-707, 2005. Smith, C., Want, EJ, O'Maille, G., Abagyan, R., and Siuzdak, G. XCMS: processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification. Anal. Chem. 78 : 779-787, 2006. Styczynski, MP, Moxley, JF, Tong, LV, Walther, JL, Jensen, KL, and Stephanopoulos, GN Systematic identification of conserved metabolites in GC / MS data for metabolomics and biomarker discovery. Anal. Chem. 79: 966-973, 2007. David B. Friedman et al.: Proteome analysis of human colon cancer by two-dimensional difference gel electrophoresis and mass spectrometry.Proteomics 2004.4,793-811

  The present invention has been made to solve the above-mentioned conventional problems. It provides quantitative reliability of data, effectively eliminates unnecessary peaks using the index, and applies to statistical analysis. It is an object to provide a method for acquiring and evaluating measurement data that can be applied to analysis.

The present invention creates a mixed sample Blend in which two samples A and B to be measured are mixed in equal amounts, measures measurement event intensities a _i and b _i for the component i of each sample A and B, and measures the measured values. and _{y 1,} y _2, the mixed measurements on samples Blend ingredients i event strength blend _i measured and was used as a measurement value _{y 3,} the regression of these three measurements y _{1, y} 2, _{y 3} as a calibration curve A function is derived, and the above problem is solved by using, as an error index, a measured value relative error that is a deviation of the measured values y ₁ , y ₂ , and y ₃ from the regression function .

Here, the regression function can be a regression line which is a linear function .

  In addition, a trend in which the slope of the regression line is normalized by a ratio to the average value of the measured values can be obtained, and ions generated by ions can be predicted using the trend as an index.

When the number of samples is an even number of 4 or more, the average of the measured values Z _{1 to} Z _n of the half samples _{1 to n is} the measured value y _1, and the measured values Z of the remaining half samples (n + 1) to 2n are used. the average of _n + 1 _{to Z 2n} and the measured value _{y 2,} the measured values _{Z 2n + 1} of the sample mixed all samples at the same volume can be the measured value _{y 3.}

When the number of samples is an odd number of 3 or more, the average of the measured values Z _{1 to} Z _{n + 1} of samples 1 to (n + 1), which is one more than approximately half, is defined as the measured value y ₁ and is one less than the remaining approximately half. samples (n + 2) and ~ (2n + 1) and the measured values _{Z n + 2 ~Z 2n + 1} , Z 2n + the measured value _{y 2} average of ₂ samples mixed with all samples equal amounts, measurement of the sample obtained by mixing all samples equal amount The value Z _{2n + 2} can be the measured value y ₃ .

The present invention also includes a first mixed sample C in which two samples A and B to be measured are mixed at a predetermined ratio p: (1-p) (mixing ratio p ≠ 0.5), and a predetermined ratio A second mixed sample D mixed at a ratio q: (1−q) (mixing ratio q ≠ 0.5) (where q−p> 0) was prepared, and three measurements of the samples A, B, and C were performed. A regression function f (c _i ) as a calibration curve is derived from the values, and a regression function f (d _i ) as a calibration curve is derived from the three measured values of the samples A, B, and D , and the first mixed sample C Quantified value f (c _i ) converted by the relationship represented by the regression function f (c _i ) of the measured event intensity c _i for the component i of the sample i and the measured event intensity d _i for the component i of the second mixed sample D the regression function f _{(d i)} by converting the relationship represented by the quantitative value f _{(d i)} is the following formula _{f (c i) (1-} q) / (1-p) ≦ the _{(D i) ≦ f (c} i) q / p
When the above relationship is not satisfied, the data c _i and d _i are rejected to solve the above problem.

Here, the rejection area can be adjusted by adjusting the mixing ratios p and q in the range of 0 <p <q <0.5, respectively.

  Further, the sum of the mixing ratios p and q can be 1.

In the present invention, the following information is obtained by measuring a sample obtained by mixing samples to be compared at a predetermined ratio to obtain measured event intensity (also referred to as quantitative data) .
1. 1. Quantitative data and quantitative data group for multiple samples for which measurement values are to be obtained 2. Quantitative data and quantitative data group regarding the mixture of measurement sample pairs 3. Measurement error index and measurement error index group based on measurement data of a mixture of multiple samples and measurement sample pairs for which measurement values are to be obtained Measurement value trend and measurement value trend group based on measurement data of a mixture of multiple samples and measurement sample pairs for which measurement values are to be obtained (in the case of claim 3)

According to the present invention, the quantitative data relating to a plurality samples it is desired to obtain a measurement value, it is possible to obtain quantitative data for components in each sample. Furthermore, the following problems are solved.

  (1) By displaying an error indicator of data, a reliability indicator can be given to quantitative data.

The reliability of the measurement data of two or more samples was obtained using the linearity or function suitability of the quantitative data for multiple samples and the quantitative data for the mixture of measurement sample pairs as an index. Can be evaluated. At this time, the error index becomes the scale. As an error index, for example, the square or the sum of errors can be used in addition to a quantitative relative error (RSE: Relative Standard Error). The closer RSE is to zero, the smaller the error and the higher the quantitative reliability. This index can be evaluated by linearity in the standard substance addition method (Non-patent Document 1) when a pure standard substance is obtained, but when the number of samples to be measured is large and it is difficult to obtain the standard substance Not applicable. According to the present invention, it is possible to directly know the reliability of measurement data regardless of whether the analysis target is known or unknown. Thereby, even when the signal intensity is close to noise, significant data can be obtained without providing a threshold (threshold setting based on the S / N ratio is common, but the basis is poor). Further, even when the signal strength is high, a decrease in reliability due to a saturation phenomenon or the like can be detected. However, this index is limited to the actually measured concentration range of a plurality of samples for which measurement values are to be obtained. In the invention of claim 6, only the rejection area of the measured value is determined.

  (2) It is possible to identify a process that reduces data reliability in the quantitative analysis procedure.

  In quantitative analysis, a sample is prepared through several procedures, and final measurement data is obtained. Until now, researchers have sought out where in the procedure there are elements that reduce reliability. In the present invention, by changing the procedure for mixing the samples, as shown in FIG. 2, it is possible to find the problem procedure by using as an index where the error from the regression function decreases. For example, if each sample is mixed before and after derivatization of the sample for gas chromatography, it can be judged how much the reliability of the data is reduced by derivatization.

  (3) A noise signal with low quantitativeness can be eliminated.

  If the error indicator from the regression function takes a large value, the data is considered unreliable. On the other hand, in such a case, when the signal is random noise, there is a possibility that the sample has changed due to mixing. In the case of a general analysis that can eliminate the latter possibility, an error from the regression function is an index for detecting a noise signal.

  (4) Alignment errors can be reduced by utilizing association between samples.

  When it is desired to compare a quantitative difference between samples, generally, identification is performed based on parameters of a specific substance. For example, in the case of LC-MS, identification and alignment are performed based on the elution time and m / z value. However, in the case of multi-component simultaneous analysis such as omics, it is often difficult to reliably identify and align only with such existing parameters, which is problematic. Therefore, when the error index from the regression function is used as a capture parameter, the identification and alignment accuracy are remarkably improved. In the case of LC-MS, in addition to the elution time and m / z value, data with a large error from the regression function is excluded because it is judged that the noise is large, and a signal pair with a small error from the regression function is excluded. It can be solved by searching.

  (5) By using the trend as an index, isotope and product ion peaks can be found in mass spectrometry (claim 3).

In mass spectrometry data, a plurality of signals are detected according to the isotope ( ¹³ C, ³⁴ S, etc.) content even in an object of the same composition formula. Further, in the ionization process, product ions such as multivalent ion generation, fragmentation, adduct, and multimerization are generated, and the number of signals is increased. Of these, fragmentation is difficult to predict theoretically. In addition, there are many combinations of product ion production, which makes it difficult to predict. The trend of the mixed sample is a value specific to the component in the sample pair, and the trend of the product ions can be expected to be equal to the trend of the parent ions. Therefore, it is possible to predict the parent-child relationship of ions using the trend as an index. For example, in the case of LC-MS, it can be determined that signals having the same trend in the homogeneous elution period are in the relationship of product ions.

  (6) Give a logical index for multivariate analysis of a quantitative data group to be compared.

  When performing multivariate analysis (for example, principal component analysis) using statistical methods from multivariate data obtained by simultaneous analysis of multiple components, the criteria for evaluating differences between sample groups are often ambiguous, There is little ground for judgment. For example, when discriminant analysis for dividing a group is performed, a normal vector of a discriminant function that minimizes the variance of each group and maximizes the distance between groups is obtained, but a boundary line cannot be obtained because an intercept is not determined. Therefore, a straight line passing through the center of gravity of all data is generally adopted, but this is not applicable when the number of data and the variance of each group are different. On the other hand, a more rational discriminant function can be obtained as illustrated in FIG. 3 by using the quantitative data group relating to the mixture of measurement sample pairs according to the present invention. That is, by using the center of gravity of the mixed sample data group, a basis for the intercept of the discriminant function can be given.

  (7) It is possible to perform preprocessing in consideration of the confirmation of scale and reliability when performing statistical analysis using a quantitative data group.

  In each statistical method, since the scale of the quantitative value of the target data is determined, it is necessary to confirm the scale of the data (interval scale or order scale) when performing statistical analysis. Statistical processing cannot be performed on data that is not quantitatively understood. However, if the error from the regression function is small, a scale of at least the interval scale can be guaranteed. Further, it is possible to give a reliability weight to each numerical value in the quantitative data group based on the error index from the regression function. Usually, even if data is mixed with good or bad quantitativeness, the test is performed with the same weight for each data. However, in the present invention, the quality of the quantitativeness can be judged from the error index from the regression function. Therefore, the smaller the error index from the regression function, the larger the weight is adjusted. Applied tests can be performed.

Now, w1, w2,..., Wl,..., Wm are weights for the pairs to be mixed.

By using this, the following weighted t statistic considering an error index (RSE or the like) from the regression function can be considered.
However, Ux and Uy are the unbiased dispersion of each group, respectively.

  In addition, when w1 = w2 =... = Wm = 1 / m, it is a conventional Student's t statistic.

  (8) The basis for the difference between the two quantitative values can be given.

  When comparing two quantitative values, indicators such as differences and ratios are used, but in order to determine that there is a difference, it is necessary to set a threshold for each indicator. A logical basis cannot be given. However, by using a mixed sample, it is possible to estimate an error range that each quantitative value can have. Based on the error range of each quantitative value, if it can be confirmed that there is no difference, that is, the probability of an event that results in a slope of zero is very small, the rationale is that each quantitative value is different. Can be shown.

  By using the above characteristics, especially in analysis with multi-component simultaneous analysis data such as omics, the process from design of data acquisition method to data processing, higher-order statistical analysis has been made efficient and highly accurate, In addition, logical consistency of each process can be given.

  Table 1 shows a comparison between the conventional method and the present invention.

The figure which shows the example of the noise signal contained in the data analyzed with the analytical instrument The figure which shows a mode that the process which reduces quantitative reliability by this invention is identified. The figure which shows the example which has given the logical parameter | index at the time of carrying out multivariate analysis of quantitative data group by this invention The figure which shows the sample mixing method of 1st Embodiment of this invention. Figure showing the rejection area The figure which shows the sample mixing method of 2nd Embodiment of this invention. The figure which similarly shows the definition of measured value relative error (RSE) The figure which shows the sample preparation method of 3rd Embodiment of this invention. The figure which similarly shows the sample preparation method of 4th Embodiment The figure which shows the analysis result of the serum lipoprotein data of the hyperlipidemia patient by the Example of this invention Figure showing the results of simultaneous analysis of a mixture of ionic standards The figure which similarly shows capillary electrophoresis mass spectrometry data of a mouse liver extract Figure showing the results of simultaneous analysis using the same general quantitative evaluation criteria

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  The present invention is configured based on the theory of quantitative evaluation criteria based on mixed samples. Furthermore, the quantitative evaluation criteria are classified into general quantitative evaluation criteria, special quantitative evaluation criteria, and extended special quantitative evaluation criteria according to the handling of the mixing ratio.

  The first embodiment of the present invention using the general quantitative evaluation criteria will be described below.

  In a measurement method that does not use a calibration technique, such as a DNA microarray, the reliability of the data cannot be guaranteed, and even a measurement value that cannot normally occur is used for subsequent data analysis. According to the present invention, it is possible to obtain an index for eliminating such an unreliable measurement value. It is obtained by preparing a sample in which two samples to be measured are mixed at a specified ratio, and a rejection area for those measured values is used.

First, as shown in FIG. 4, the sample to be measured was mixed at a certain ratio (p: 1-p, q: 1-q) (where the mixing ratio p ≠ 0.5, q ≠ 0.5). Prepare two samples. The measurement event intensity (signal intensity or the like) of the specific component i in the sample A is a _i , and the intensity in the sample B is b _i . Assuming that these intensities have a relationship represented by a function f (x) ( regression function as a calibration curve; generally a linear function), the components in the samples (sample C and sample D) generated by mixing them The intensity of i is obtained by the following equation.
f (c _i ) = f (a _i ) p + f (b _i ) (1−p) (3)
f (d _i ) = f (a _i ) q + f (b _i ) (1−q) (4)

From equations (3) and (4), a _i and b _i are obtained by the following equations.
f (a _i ) = {f (d _i ) (1-p) −f ( c _i ) (1-q)} / (qp) (5)
f (b _i ) = {f (c _i ) q−f (d _i ) p} / (q−p) (6)

Since each intensity is zero or a positive real number (a _i ≧ 0, b _i ≧ 0), if qp> 0 , the following equation is obtained by modifying equations (5) and (6).
f (c _i ) (1-q) / (1-p) ≦ f (d _i ) ≦ f (c _i ) q / p (7)

That is, the rejection range of c _i and d _i is as follows:
f (d _i ) <f (c _i ) (1-q) / (1-p),
f (c _i ) q / p <f (d _i ) (8)

That is, if c _i and d _i are in a relationship that does not satisfy this condition, their values can be rejected as unreliable. Further, by adjusting the mixing ratio parameters p and q in the range of 0 <p <q <0.5, respectively, the rejection area can be adjusted as shown in FIG. For example, when p = 0.4, data having f (c _i ) / f (d _i ) of 3/2 times or more and 2/3 times or less can be rejected.

If p = 0,
Since c _i and d _i can take all values of zero or positive real numbers, the rejection region cannot be defined.

Further, when the samples C and D are prepared at opposite ratios and the sum of the mixing ratios p and q is 1, that is, q = 1−p, the equations (4) to (9) become as follows: .
f (d _i ) = f (a _i ) (1−p) + f (b _i ) p (4 ′)
f (a _i ) = {f (d _i ) (1-p) −f (d _i ) p} / (1-2p) (5 ′)
f (b _i ) = {f (c _i ) (1-p) −f (d _i ) p} / (1-2p) (6 ′)
f (c _i ) p / (1-p) ≦ f (d _i ) ≦ f (c _i ) (1-p) / p (7 ′)
f (d _i ) <f (c _i ) p / (1-p),
{F (c _i ) (1-p)} / p <f (d _i ) (8 ′)

  Next, a second embodiment of the present invention using a special quantitative evaluation standard will be described.

In general quantitative evaluation criteria used in the first embodiment, when the other singularities p = 0.5, c _i and d _i by mixing the sample A and B in equal amounts is equal, a Obviously, _i and b _i cannot be calculated. Therefore, by measuring the measurement event intensity a _i and b _i related component i of the sample desired to obtain a measurement value, further, as shown in FIG. 6, measurements on components i of mixing a measurement sample at the same volume sample event strength blend _i Measure.

Here, the following equation holds.

Therefore, even if the concentration of either sample component i at the mixing source is zero, the mixed sample always has a signal i, and the measured event intensity is always the average of the intensity in the mixing source sample. The measured value relative error (RSE) is defined by evaluating the consistency between these three measured values and the regression function (generally a linear function) derived therefrom.

Here, as illustrated in FIG. 7, a case will be described in which three measurement values are applied to the most general regression line (linear function). The straight line is determined by regression and uses an error with respect to the model straight line as an index. Here, an example in which a relative error (RSE: Relative Standard Error) from the model is employed and RSE is calculated by the following equation will be described.

Here, n indicates the number of points (n = 3) when creating a straight line model, that is, the number of elements of x and y. x represents an arbitrary numerical value indicating an average relationship (x3 may be an average of x1 and x2), and y represents a measured value of each sample. RSE indicates that the closer to zero, the closer to an ideal straight line. The trend is the slope of this regression line normalized by the ratio to the average value of y. The final a _i and b _i are correctly obtained from the regression line obtained here. Also, if any of a _i and b _i is zero, the ratio of a _i and b _i cannot be calculated. However, when it is obtained from a regression line, many do not become zero, so this problem can also be avoided. .

  Next, the third and fourth embodiments of the present invention using the extended special quantitative evaluation criteria will be described.

  The special quantitative evaluation criteria can be extended and applied even when three or more samples are mixed. FIG. 8 shows the third embodiment when the number of samples is an even number of 4 or more, and FIG. 9 shows the fourth embodiment when the number of samples is an odd number of 3 or more (the case where the number of samples is 2 is the second embodiment). Special quantitative evaluation criteria used in the form).

When the number of samples is an even number of 4 or more, as shown in FIG. 8, the average of the measured values Z _{1 to} Z _n of half of the samples _{1 to n} is taken as the measured value y ₁ and the remaining half of the samples (n + 1) to 2n The average of the measured values Z _{n + 1 to} Z _2n of the sample is defined as the measured value y ₂ , the measured value Z _{2n + 1} of the sample Blend obtained by mixing all the samples in equal amounts is defined as the measured value y ₃ , and the relative error (RSE _{2n, i} ) is It is obtained by the following formula.
Here, the intensity of component i in the mixed sample is Z _{2n + 1} .

On the other hand, when the number of samples is an odd number of 3 or more, as shown in FIG. 9, the average of measured values Z _{1 to} Z _{n + 1} of samples 1 to (n + 1), which is one more than approximately half, is the measured value y ₁ and the rest Measured values Z _{n + 2 to} Z _{2n + 1} and Z _{2n + 2} of samples Blend obtained by mixing equal amounts of samples (n + 2) to (2n + 1), which are one less than about half of the samples, are the measured values y ₂ , Relative error (RSE _{2 + 1n, i} ) is obtained by the following equation with the measured value Z _{2n + 2} of the sample Blend mixed in equal amounts as the measured value y ₃ .

Example 1. Serum lipoprotein data from patients with hyperlipidemia As an example of experiments in which it is difficult to confirm the quantitativeness, protein quantification by gel electrophoresis was examined. In general, when the amount of protein is quantitatively compared by gel electrophoresis, signal intensity is read from a photograph of a gel stained with protein, and the signal area is compared. However, gel electrophoresis has low resolution, and when band separation is poor, quantification often depends on the researcher's senses. In many cases, a standard protein is not obtained, and it is difficult to prepare a strict calibration curve using a standard substance, rather than a standard addition method. Therefore, we applied quantitative evaluation criteria to protein gel electrophoresis to examine quantitative reliability.

  Specifically, serum lipoproteins of healthy human subjects and type II hyperlipidemia patients undergoing medication were analyzed by polyacrylamide gel disc electrophoresis (lipophore). The band intensity of the gel was detected by software, and the area value of the signal peak corresponding to each lipoprotein was calculated.

The results are shown in FIG. Hyperlipidemic patients are characterized by elevated serum VLDL, which is bad cholesterol, and the appearance of IDL. In particular, the appearance of IDL is considered to reflect the severity. In patients with hyperlipidemia being treated, VLDL decreased, IDL increased, and HDL decreased compared to healthy subjects. Although HDL is a slight difference, it can be seen that the difference is high when RSE is taken into consideration. VLDL also has high quantitative reliability and is significantly reduced, indicating that the effects of medication are well exhibited. However, the IDL midband looks like the shoulder of the LDL signal peak. When this shoulder portion is quantified using the quantitative evaluation criteria, a certain amount of quantitative reliability is obtained. Therefore, it can be seen that the presence of IDL is definite and further treatment needs to be continued. In this way, highly reliable data can be provided by evaluating the shoulder of a signal peak whose presence is difficult to determine based on a quantitative index.

Example 2 Simultaneous analysis of ionic standard substance mixture In order to evaluate the function of special quantitative evaluation criteria, separation analysis was performed using a mixed solution of standard compounds as an ideal system, and data processing was performed.

Specifically, a mixed solution containing 47 types of ionic compounds prepared at a concentration of 20 μM and 100 μM was prepared, and a sample obtained by mixing the two and each solution were subjected to capillary electrophoresis-time-of-flight mass spectrometer (CE-). (TOFMS). The obtained signal was analyzed by the following procedure.
Data acquisition → Signal peak detection → STEP1 → STEP2

  In STEP 1, the components in the sample solution were aligned, and random noise components were excluded by calculating RSE (RSE = 0.2 or less was excluded). In STEP 2, the trend value was calculated and the generated ions were excluded.

  The results are shown in FIG. In the raw data obtained by CE-TOFMS analysis, 867 signals were detected in a 20 μM solution and 1,665 signals were detected in a 100 μM solution. Here, since four types of internal standard substances are added for the purpose of movement time correction and the like, a total of 51 components are included. That is, the calculation results in 17 to 33 signals per average component. From this point, the operation of STEP 1 was performed to eliminate random noise, and the signal was reduced to 292 signals with a 20 μM solution and 379 signals with a 100 μM solution. Further, STEP2 was performed, and the signal derived from the generated ions was excluded. Finally, both 20 μM and 100 μM solutions had 88 signals. This final signal contained 37 components in addition to the 51 components contained in the solution, and these were derived from impurities contained in the standard reagent. The RSE of the standard substance in this measurement was 0.063 maximum for Methionine, which was considered to be affected by Methionine instability at low concentrations. The trend value is 1.33 theoretically, but for many substances, a value close to this was obtained. The materials that showed extremely low RSE were Spermine (trend = 1.02), Spermidine (trend = 1.28) and Uracil (trend = 1.10). It was found that Spermine and Spermidine are affected by ion suppression because they have a short migration time and overlap with the metal ion signal. Uracil, on the other hand, had a long migration time, and overlapped with the signal of neutral or anionic substances, and was thought to have been affected by ion suppression. From the above results, it has been shown that the quantitative reliability index based on the quantitative evaluation criteria is in good agreement with the cause of the decrease, and is effective for eliminating random noise and product ions.

Example 3 Capillary electrophoresis-mass spectrometry data of mouse liver extract We applied quantitative evaluation criteria to biological samples, and examined whether they can be used effectively not only in standard solutions but also in actual samples. Actually, simultaneous component analysis of mouse liver extract was performed, and data processing was performed in the same manner as in Example 1.

Specifically, prepare 3 mice each with whole body perfusion with buffer (no effect of blood components) and 3 mice with no blood components (with effects of blood components), and extract and extract liver tissue A product was prepared. Immediately before the measurement by CE-TOFMS, both samples were mixed in equal amounts, and each sample and the mixed sample were separated and analyzed. Mixing was performed in pairs of perfused and non-perfused mice.

  The results are shown in FIG. In the raw data obtained by CE-TOFMS analysis, 8,354 signals, 7,471 signals, and 6,762 signals were detected in each pair. From this point, the operation of STEP 1 was performed to eliminate random noise, and the signals were 1,917, 1,641, and 1,778, respectively. Furthermore, when STEP2 was performed and signals derived from the generated ions were excluded, 506 signals, 430 signals, and 474 signals were obtained, respectively. When these final signals are realigned and the data are integrated, 818 signals are obtained. 26% are detected in all 6 samples, 20% are detected in 4 samples, and only 2 samples are detected. Was 54%. Substance signals obtained with only two samples are low in reliability. Therefore, when these substances were excluded and substances were identified, 17% were identified in the applicant's (HMT) database, and the Kyoto Genome Encyclopedia In the encyclopedia KEGG (Kyoto Encyclopedia of Genes Genomes), 34% were predicted substance names, and 49% were not identified. This result was almost the same as the result of normal manual peak processing, and it was shown that the data processing performance based on the quantitative evaluation criteria was high. In addition, the time required for data processing based on the quantitative evaluation standard was several minutes when an Excel (registered trademark) -based program was executed using a normal desktop computer. Since manual peak processing so far has required more than two weeks, it has become clear that the present invention provides a rational explanation for reducing work time, improving accuracy, and selecting signals.

Example 4 Simultaneous analysis using general quantitative evaluation criteria In order to evaluate the function of general quantitative evaluation criteria, separation analysis is performed using a mixed solution of two types of plasma samples of normal and hyperlipidemic rabbits, and data processing is performed. I did it. The mixing ratio was 0.45: 0.55 (p = 0.45).

Specifically, after two types of plasma samples (Sample A and Sample B) are pretreated, Sample C in which each sample is mixed at a ratio of 0.45: 0.55, and 0.55: 0.45. Sample D mixed at the ratio was prepared and measured using CE-MS.
Data acquisition → Signal peak detection → STEP1 → STEP2

In STEP 1, the components in the sample solution were aligned, and random noise components were excluded by calculating the rejection area. The rejection area in the specific component i when p = 0.45 is f (d _i ) <(9/11) f (c _i ), (11/9) f (c _i ) <f (d _i ) Become. Here, assuming that the function f is a linear expression passing through the origin, the rejection area of each signal is d _i <(9/11) c _i , (11/9) c _i <d _i . In STEP2, the trend value was calculated and the generated ions were excluded.

The results are shown in FIG. In raw data obtained by CE-TOFMS analysis, 4,569 signals were detected in sample C and 4,910 signals were detected in sample D. STEP 1 was operated from here, and random noise was eliminated. As a result, both Sample C and Sample D were 1,501 signals. Since both samples had a data size of 1/3 or less in STEP 1, the high noise content of each measurement data can be seen. In STEP 2, the slope is calculated from the signal intensities of the sample C and the sample D (in this case, a simple signal intensity ratio may be used), and those having a close slope are excluded as product ions. As a result, 412 signals were finally obtained for each sample. The signal intensity of each specific component i in sample A and sample B is calculated as follows.
a _i = (− 9/2) c _i + (11/2) d _i (14)
b _i = (11/2) c _i − (9/2) d _i (15)

  These operations are easy to automate, and can be realized with much fewer man-hours than manual noise removal.

  The present invention is a method for quantitative analysis of a sample containing a large number of chemical substances, metabolomics, transcriptomics (DNA chip and microarray and DNA sequencer), proteomics, genomics, capillary electrophoresis CE), liquid chromatography (LC), gas chromatography (GC), spectrophotometric analysis (diode array), fluorescence intensity analysis, mass spectrometry (MS), NMR, one-dimensional and two-dimensional gel electrophoresis (Western blotting, Southern blotting) Method, including Northern blotting method), real-time PCR, substance quantification by enzyme method, all other quantitative analysis methods, and quantitative data in general when samples can be mixed (not limited to chemical analysis).

A, B ... Sample Blend, C, D ... Mixed sample

Claims (8)

A mixed sample Blend in which two samples A and B to be measured are mixed in equal amounts is created, and measurement event intensities a _i and b _i relating to component i of each sample A and B are measured to obtain measured values y ₁ , y ₂ , the measured event intensity blend _i for component i of the mixed sample Blend is measured to obtain a measured value y ₃ ,
From these three measured values y ₁ , y ₂ and y ₃ , a regression function as a calibration curve is derived,
A measurement data acquisition / evaluation method, wherein a measurement value relative error, which is a deviation of the measurement values y ₁ , y ₂ , and y ₃ from the regression function , is used as an error index . The measurement data acquisition / evaluation method according to claim 1, wherein the regression function is a regression line that is a linear function .   The measurement data acquisition / acquisition according to claim 2, wherein a trend obtained by normalizing a slope of the regression line with a ratio of an average value of measurement values is obtained, and ions generated are predicted using the trend as an index. Evaluation method. When the number of samples is an even number of 4 or more, the average of the measured values Z _{1 to} Z _n of the half samples ₁ to _{n is} defined as the measured value y _1, and the measured values Z _{n + 1} to 2n of the remaining half samples (n + 1) to 2n are used. The average value of Z _{2n is} the measured value y _2, and the measured value Z _{2n + 1} of a sample obtained by mixing all samples in equal amounts is the measured value y ₃ . Measurement data acquisition and evaluation method. When the number of samples is an odd number of 3 or more, the average of the measured values Z _{1 to} Z _{n + 1} of samples 1 to (n + 1), which is one more than approximately half, is defined as the measured value y _1, and one sample less than the remaining approximately half ( n + 2) ~ (2n + 1) and all samples were a mean of an equal amount by measurement of the mixed sample _{Z n + 2 ~Z 2n + 1} , Z 2n + 2 and the measured value _{y 2,} the measured value Z of sample mixed all samples equal amount _The method for acquiring and evaluating measurement data according to any one of claims 1 to ₃ , wherein _{2n + 2} is set as the measurement value y3. A first mixed sample C obtained by mixing two samples A and B to be measured at a predetermined ratio p: (1-p) (mixing ratio p ≠ 0.5), and a predetermined ratio q: (1- q) (mixing ratio q ≠ 0.5) A second mixed sample D mixed at (where q−p> 0) is prepared,
A regression function f (c _i ) as a calibration curve is derived from the three measurement values of the samples A, B, and C, and a regression function f (d ) as a calibration curve from the three measurement values of the samples A, B, and D. _i )
A first measurement relating to component i of the mixed sample C event intensity c _i of the regression function f (c _i) in terms the relationship represented by the quantitative value f (c _i), the component of the second mixed sample D i The quantitative value f (d _i ) converted by the relationship represented by the regression function f (d _i ) of the measured event intensity d _i with respect to the following expression f (c _i ) (1-q) / (1-p) ≦ f (d _i ) ≦ f (c _i ) q / p
When the relationship is not satisfied, the data c _i and d _i are rejected. The measurement data acquisition / evaluation method according to claim 6, wherein the rejection area is adjusted by adjusting the mixing ratios p and q within a range of 0 <p <q <0.5, respectively.   The method for acquiring and evaluating measurement data according to claim 6, wherein the sum of the mixing ratios p and q is 1.