JP4355281B2 - Peak extraction method and peak extraction apparatus - Google Patents

Description

  The present invention relates to a technique for extracting a peak in spectral characteristics by information processing using an electronic computer. The present invention also relates to a diagnostic apparatus using a peak extraction technique. Various biological samples such as plasma, serum, urine, tissue extract, and cerebrospinal fluid using mass spectrometer such as time-of-flight mass spectrometer (TOF-MS) and liquid chromatograph From the waveform data obtained by measurement (hereinafter referred to as “waveform data” or “waveform”), a “peak” portion corresponding to a biological molecule such as a protein contained in the sample is obtained by information processing using a computer. The present invention relates to an extraction method and apparatus.

  Various techniques have been proposed as techniques for extracting peaks in spectral characteristics. In a general peak extraction method, for example, a method of extracting a peak from a single waveform based on a condition such that the SN ratio is a certain value or more is used (referred to as a first step in this column).

  For example, when comparing single waveforms having different origins, it is often necessary to group and compare peaks extracted from different single waveforms in the first step as the second step. At this time, all the peaks extracted in the first step are all handled equally in the peak grouping process in the second step (see Patent Document 1).

  When peaks obtained from a single waveform are overlapped between multiple waveforms and the peaks between adjacent waveforms in the superimposed state are closer than a certain distance, they are grouped with the same peak group. Assuming that the peaks are identified as the same origin. For a plurality of peaks that belong to the same peak group, if some peaks are sufficiently different from other peaks in terms of their intensity, specificity will be seen with respect to waveforms having such peaks, for example By comparing with a known substance, a substance corresponding to the specific peak can be identified, and if the specificity is specific to a specific disease patient, it can be used for diagnosis.

JP 2003-315341 A

However, when the above-described peak extraction method is used, the peak extraction accuracy may be reduced in the following cases.
(1) Quality degradation due to noise Peaks obtained from a single waveform, whether peak-like or noise-like, are all treated equally in the next peak grouping process. FIG. 15 is a diagram illustrating an example of a procedure for performing peak extraction by performing the above general processing and then performing peak grouping. As shown in the middle diagram on the left side of FIG. 15, when there is noise, the peak due to this noise is also regarded as one peak group in the peak grouping process. As a result, it is easy to generate a peak group consisting of only noise or mostly noise, and there is a problem that it is impossible to distinguish between a peak group containing a lot of noise and a peak group containing no noise.

(2) Degradation of quality in peak dense section Another problem, the degradation of peak extraction quality in a peak dense section, will be described with reference to FIG. FIG. 16 is a diagram illustrating an example of a procedure for performing peak extraction by performing the above-described general processing and then performing grouping of peaks.

  In the section where fine peaks are dense as shown in FIG. 16, the interval between adjacent peaks is narrowed. Therefore, in the conventional simple peak grouping method that groups all peaks within a certain distance, as shown in the right figure, many peaks are assigned to the same group as a result. The assigned peaks are not correctly matched. In particular, in a polymer region where a broad peak, that is, a state in which the peak top is not clear or adjacent peaks are often overlapped, the result obtained by peak extraction shows that “fine peaks are concentrated as shown in FIG. In many cases, this problem is more likely to occur.

  The present invention has been made in view of the above problems, and an object thereof is to improve the accuracy of peak extraction and grouped peak association.

  In the present invention, reliability is given to each peak, and a total value of the reliability of peaks existing at the same position in a plurality of waveforms is obtained as a peak density, and the reliability of the peak group is determined based on this peak density. The method of granting is adopted. As a result, the reliability of a group consisting mainly of noise peak groups or exceptional peaks that can be seen in only a small part of the waveform is set low. As a result, the problem (1) of quality degradation due to noise can be dealt with by removing peak groups with low reliability. For the problem (2) above, instead of simply grouping neighboring peaks, the peak density is calculated for each minute section, and grouping is performed in consideration of the increase and decrease in peak density. Only can be grouped (see FIG. 17).

  Furthermore, giving a confidence level to a peak group corresponds to ordering the peak groups. Therefore, when a large number of peak groups are obtained, subsequent work can be performed only from the peak group with high reliability or from the peak group with high reliability, and the work efficiency can be improved.

  In the present invention, reliability is given to individual peaks, and the total value of the reliability of peaks existing at the same position in a plurality of waveforms is obtained as a peak density, and the reliability of the peak group is determined based on the peak density. The method of giving is adopted. As a result, the reliability of a group consisting mainly of noise peak groups or exceptional peaks that can be seen in only a small part of the waveform is set low. As a result, the problem (1) of quality degradation due to noise can be dealt with by removing peak groups with low reliability. For the problem (2) above, instead of simply grouping neighboring peaks, the peak density is calculated for each minute section, and grouping is performed in consideration of the increase and decrease in peak density. It is possible to group only.

  Furthermore, giving a confidence level to a peak group corresponds to ordering the peak groups. Therefore, when a large number of peak groups are obtained, subsequent work can be performed only from the peak group with high reliability or from the peak group with high reliability, and the work efficiency can be improved.

  In this specification, peak extraction processing refers to processing for extracting peaks from a single waveform and grouping the extracted peaks into a plurality of waveforms.

  Hereinafter, a peak extraction technique according to an embodiment of the present invention will be described with reference to the drawings. First, before describing a specific embodiment, referring to FIGS. 1A and 1B, an example of mass spectrometry using Surface-Enhanced Laser Desorption / Ionization TOF MS (SELDI TOF-MS) will be described. However, an explanation will be given.

  FIG. 1A is a diagram illustrating the positioning of the proteome analyzer. As shown in FIG. 1A, the SELDI TOF-MS is a system in which, for example, a metal chip and a time-of-flight mass spectrometer are combined. Various proteins can be analyzed in a short time by using a small amount of sample obtained from the living body directly or roughly. The surface of the chip is chemically modified such as ion exchange and IMAC (metal ions such as Ni and Cu). Using the chemical properties, proteins are captured on the surface of the chip and measured by TOF-MS. To do. It is possible to analyze protein expression in biological samples, identify proteins, analyze post-translational modifications, and the like. The proteome waveform data measured by SELDI TOF-MS is a spectrum in which the horizontal axis represents the molecular weight or the mass-to-charge ratio (m / z) as the equivalent amount, and the vertical axis represents the number of molecules or the intensity as the equivalent amount. As obtained. In this case, the peak corresponding to each m / z is a peak corresponding to a specific molecule such as a protein. Furthermore, from a waveform data obtained by measuring a biological sample such as plasma using this SELDI TOF-MS, a “peak” portion corresponding to a biological derived molecule such as a protein contained in the sample is obtained using a computer. A peak extraction device is provided for extraction by processing. For the waveform data obtained from each of a plurality of samples using SELDI TOF-MS, the peaks estimated to originate from the same molecule are associated with each other between the plurality of waveform data, and the peaks are grouped. The quality of peak grouping has a great influence on the accuracy of subsequent data analysis processing, and consequently has a great influence on the accuracy of diagnosis and the like based on the data analysis.

  FIG. 1B is a diagram illustrating a schematic configuration example of the peak extraction apparatus according to the present embodiment. As shown in FIG. 1B, the peak extraction apparatus according to the present embodiment performs grouping of peaks, and a waveform data acquisition unit 1 that acquires waveform data, a peak extraction unit 3 that extracts peaks in single waveform data, and the like. A grouping processing unit 5 and a data output unit 7 for outputting data are provided. Furthermore, it has the difference detection part 11 which detects the difference between single waveform data based on the grouped peak. Each processing unit actually includes a ROM for storing a program for causing a computer to execute the method described below, a RAM for expanding data in the ROM, a CPU for executing a program expanded in the RAM, The processing is performed by the following.

  The peak extraction apparatus according to the present embodiment extracts a peak from single waveform data and gives a reliability to the peak. After that, grouping of peaks across multiple waveforms and imparting reliability are performed.

  That is, first, regarding the waveform data, a parameter that can change the width of the peak region to be described later is set, and by arbitrarily changing this parameter, various peaks from a fine peak to a coarse peak can be changed into “regions” (hereinafter referred to as “region”). This is referred to as “peak area”. Further, the degree of overlap of the peak areas (hereinafter referred to as “inclusion level”) is obtained from the inclusion relation of these peak areas. The “inclusion level” can be used as an index representing “peakness”, that is, the reliability of the extracted peak.

  In this case, when associating peaks across a plurality of waveform data, the reliability of the peak group itself is quantified using the “inclusion level” described above. That is, at a position where a peak with a high inclusion level (= high reliability) is seen in many waveform data, a peak group with high reliability is obtained, whereas in the opposite case, the obtained peak group has low reliability. It can be judged.

  As described above, by determining the quality of peak grouping, it is possible to know the accuracy of subsequent data analysis processing, and hence the accuracy of diagnosis and the like based on final data analysis.

  In the following, description will be given in the order of peak extraction processing from single waveform data and peak grouping processing from a plurality of waveform data. FIG. 2 is a diagram showing an overview of peak extraction processing (step 1) and peak grouping processing (step 2) from single waveform data according to the present embodiment. As shown in FIG. 2, peak extraction from single waveform data from waveform 1 to waveform N is performed in step 1 from single waveform 1 to waveform N obtained by measuring a plurality of objects. Then, in step 2, peak grouping is performed. Hereinafter, the peak extraction process from the single waveform data in step 1 and the peak grouping process in step 2 will be described in this order.

(1) Step 1: Peak Extraction Processing from Single Waveform Data The peak extraction apparatus according to this embodiment detects a portion where a peak exists as a peak region. In the process of detecting as a peak region, peak regions having various widths can be obtained by setting various parameters used internally. In other words, processing is performed to select a parameter and detect a peak region so that peak regions having various widths can be obtained. When the value of this parameter is large, a wide region including a high peak is obtained. Conversely, when the parameter value is small, a narrow region including a low peak is obtained. In general, when the parameter value is small, many peaks are obtained, but a lot of noise is included. On the other hand, if the parameter value is large, the probability of omission of peak detection increases while noise is small.

  In the peak extraction apparatus according to the present embodiment, parameter values are changed in several ways, and peak regions having different widths are obtained for the respective parameter values. The inclusion relation is constructed from the degree of overlap of these peak areas, and the depth of the inclusion relation is quantified as the “inclusion level”. By treating this inclusion level as an index of “peakness”, it is possible to give a confidence level to the peak and to detect a peak region with a confidence level as a peak.

  FIG. 3 is a diagram illustrating a process of extracting a peak from single waveform data and obtaining an inclusion level. In the single waveform shown in FIG. 3, the horizontal axis is m / z and the vertical axis is intensity. However, m is mass and z is an electric charge. As shown in FIG. 3, four peaks are extracted from the right side (in order, peaks 1 to 4). The peak area is detected when the internal parameters are 0.5, 1, 5, and 10. When the internal parameter is set to 0.5, a peak region is detected in any peak from peak 1 to peak 4. As for the peak region with the internal parameter = 1, the peak region is detected at peak 2 and peak 4. As for the peak region at the internal parameters = 5 and 10, the peak region is detected at the peak 4.

  In this way, peak areas at four different internal parameter values are detected, and the inclusion level is determined for each of the peaks 1 to 4 based on the structure of the detected peak area.

  Hereinafter, Step 1 will be described in more detail. Step 1 is divided into the following two substeps. Hereinafter, each sub-step will be described in detail.

(1-1) Peak Area Detection In the peak area detection step according to the present embodiment, as shown in FIG. 3, the value of the internal parameter P is changed in several ways, and a single waveform is obtained for each value P. The peak area from is detected. A specific value of the parameter P to be changed is given in advance, and the number and value are common in all waveform data. For example, if N parameter values are given in advance, N peak areas are detected for a single waveform, and a separate peak area detection result is obtained for each parameter value.

  The peak area detection process is performed by estimating whether or not each point in the waveform is near the peak. A set of consecutive points estimated to be in the vicinity of the peak becomes one peak region. The estimation as to whether or not each point is in the vicinity of the peak is performed by sequentially searching the left and right sides of the point and determining whether or not the point has increased or decreased beyond a preset threshold value. A specific procedure for estimating the point will be specifically described with reference to FIGS. 4 (a) to 4 (d). FIG. 4A to FIG. 4D are four cases in which the search is terminated.

  (1) First, let the point be point A. (2) Next, the left side of the point A, that is, the direction in which the m / z value decreases is searched. At this time, (a) B is assumed to be A. (B) The point on the left of B is changed to B again. (C) If B corresponds to one of the following search end cases, the process proceeds to the following (3), and otherwise, the process proceeds to (a). (3) Perform a right search for point A. That is, the same processing as in (2) is performed on the right side, that is, in the direction in which the m / z value increases. (4) Based on the search end cases of the left search and right search of point A, it is estimated whether or not point A is in the vicinity of the peak based on Table 1.

Here, the search end cases are as follows. That is, as shown in FIG. 4A, as search end case 1, the intensity at point B exceeds the upper threshold value T1. In this case, since it is estimated that a higher peak exists on the left side, it can be estimated that the point A is not near the peak. In case 2 shown in FIG. 4B, the intensity at point B is below the lower threshold. In this case, since the point B is lower than the self (point A), the point A may be near the peak, but cannot be estimated as the vicinity of the peak until the right side is searched. . Case 3 shown in FIG. 4C is a case where the vehicle has reached the predetermined distance D from the point A without entering any of the cases 1 and 2. Case 4 shown in FIG. 4D is a case where the end of the waveform is reached without matching any of the above. Here, D in case 3 of FIG. 4C may be a variable value depending on A or B.
Note that FIG. 4C can be omitted.

  Table 1 is an estimation table for estimating the vicinity of the peak. “◯” means near the peak. “X” means not near the peak. “Left” means to make the same as the estimation result of the point adjacent to the left of the point A (substitute the previous estimation result).

  By searching the left side and the right side and determining which of the cases 1 to 4 corresponds to each search result, it can be estimated whether or not it is in the vicinity of the peak. Based on this estimation result, the peak region can be determined. Needless to say, Table 1 and the following calculation method are merely examples and can be changed or modified.

  The “upper threshold value T1” and the “lower threshold value T2” in the search end case are values that depend on the parameter P. Although there are various methods for determining the threshold value, the following three methods are shown in Table 2 below.

  Where Y (A) is the Y coordinate or intensity of A, and max {Y (A),..., Y (B)} is the maximum intensity of all points from A to B. , N (A) indicates the intensity of noise at the position A. As shown in Table 2, T1 and T2 may be variable values depending on A or B.

As shown in Table 2, in method 1, T1 is defined as a value that is increased by a parameter P from the intensity at point A, and T2 is defined as a value that is decreased by P by a parameter P from the intensity at point A. In the method 2, T1 is defined as a value increased by P by the parameter P from the intensity at the point A, and T2 is defined as a value decreased by the parameter P from the maximum value of the intensity between the positions A and B. In method 3, T1 is a large value only function n (A) × 2 ^P represented by the strength and the parameter P of the noise than the intensity of the point A, the maximum intensity of between T2 from position A to B a small value by the function n (a) × 2 ^P represented by the strength and the parameter P of the noise from the value. Here, if the function n (A) × ^2P is P = 1, twice the noise intensity n (A) is the above function. Below, the case where the method 3 is employ | adopted is demonstrated.
The peak region in a single waveform can be detected by the method as described above.

(1-2) Selection of peak areas and setting of inclusion levels In sub-step (1-2), the peak area groups based on several parameter values obtained in the above-mentioned sub-step (1-1) Based on the inclusion relation, the peak area is selected and the inclusion level is set. FIG. 5 is a diagram illustrating an example of the peak region obtained in the substep (1-1). Hereinafter, the process of the sub-step (1-2) will be described with reference to FIG. Peak regions obtained by variously changing the parameter P, that is, all peak regions indicated by bold lines in FIG. 5 obtained when the parameter P is P ₁ , P ₂ , P ₃ , P ₄ , Check the inclusive relation to any two sets of. FIG. 6 is a diagram illustrating a state in which the inclusion relation is inspected. As shown in FIG. 6, the inclusion relationship is indicated by an arrow from the included side (for example, indicated by reference symbol A) to the included side (for example, indicated by reference symbol B). It should be noted that, when the inclusion relationship is inspected, an “inclusion relationship establishment condition” to be described later is set so that the inclusion relationship is determined to be established if one of the two does not completely include the other. As a result, it is possible to construct an inclusion relationship that allows some protrusion. Note that the inclusion relationship does not necessarily match the size of the parameter P.

  FIG. 7 is a diagram illustrating an example of the inclusion relationship establishment condition that can be considered to be included although it is an incomplete inclusion relationship. Even if a complete inclusion relationship that satisfies the following equation (1) is not satisfied, the peak region A is included in the peak region B if all the conditions shown in the equations (2) and (3) are satisfied. Shall be.

The complete inclusion relationship is the following relationship:
c <0 (1)
However, the condition is that both the left end and the right end of the peak region are satisfied.

Although incomplete, the relationship that is considered to be an inclusion relationship in the present embodiment is the following relationship.
c <ka (2)

However, the condition is that both the left end and the right end of the peak region are satisfied.
pb ≦ a <qb (3)

  Where a: width of the peak area A b: width of the peak area B c: length of protrusion of the peak area A, k, p, q: constants determined in advance (eg, k = 0.2, p = 0) .2, q = 0.95). A value of 1 or less close to 1 is suitable for k between 0.2 and 0.3. By adjusting according to the situation, it is possible to accurately grasp the inclusion relationship.

  Next, as shown in FIG. 8, the inclusion level (CL) of each peak area is set. As shown in FIG. 8, as long as there is a peak region including the peak region from each peak region, they are sequentially traced. That is, the process is followed until the arrow cannot be attached because the inclusion relation is not taken in the direction of the arrow. The number of times of tracing is defined as the inclusion level CL. However, when there are a plurality of routes, the maximum value of the number of traces is set as the inclusion level CL.

  Next, as shown in FIG. 9, a peak region including any other peak region and a peak region that does not have an inclusive relationship with another peak region are removed (on the side indicated by the arrow). Remove the corresponding area). Thereby, it is possible to leave only a peak region that is included in another peak region and does not include another peak region.

Finally, as shown in FIG. 10, when there is an overlap between two peak areas, only one of them is removed. There are various ways of determining which one to remove, but FIG. 10 shows an example of the case where “removes the smaller parameter P” is specified.
By the above procedure, the peak region can be extracted with high accuracy. Further, the extracted peak area is given an inclusion level, that is, the reliability of the peak area. Here, the point having the highest intensity in the peak region is defined as the apex of the peak region.

(2) Step 2: Grouping of peaks from a plurality of waveform data In step 2, a grouping of peaks obtained from a plurality of waveform data is performed. When performing peak extraction in each of a plurality of waveform data and comparing the respective peaks, it is necessary to determine which peak in one waveform corresponds to the peak in another waveform.

However, it is rare that the m / z values of the corresponding plural peaks completely match, and it is difficult to match the peaks when the m / z values completely match. Therefore, even if the m / z value is slightly shifted, it is necessary to appropriately group relatively close peaks as belonging to the same group. In the peak extraction apparatus according to the present embodiment, grouping of peaks with added reliability is realized by using the inclusion level of each peak region obtained in step 1. The peak grouping process consists of the following sub-steps.
(2-1) Overlapping vertex coordinates of peak areas obtained from a plurality of waveforms.
(2-2) Divide the m / z axis into minute sections.
(2-3) The sum of the inclusion levels is calculated for each section to obtain a peak density curve over the entire section.
(2-4) A peak grouping range is determined based on the peak density curve.
(2-5) Peaks belonging to the same grouping range are grouped.

In the following, each of the above sub-steps will be described. Before that, the outline of the flow of step 2 will be described with reference to FIG. First, waveform data that is a peak from each waveform data and has the peak region extracted in step 1 and the peak coordinate of the peak is prepared. The peak coordinates of the peaks in these individual waveforms are displayed in an overlapping manner, and in this state, the m / z axis is divided into minute sections. Next, the sum of the inclusion levels is calculated for each minute section, and is curved. This is called a peak density curve. For this peak density curve, a value exceeding this with reference to a certain threshold value L is defined as a grouping range. Thereby, the reliability can be given to the grouping range.
Below, each sub-step is demonstrated in detail.

(2-1): Peak vertex superposition step First, the peak vertices obtained from each single waveform are superposed such that both the m / z axis and the intensity axis are aligned.

(2-2): Division of m / z axis The m / z axis is divided into minute sections. As a method of dividing a section, a) a method of dividing with a fixed section width, b) a method of dividing by varying the section width, c) a dividing method that allows overlapping of adjacent sections, d) a partition that does not allow There are methods. Hereinafter, a typical division method will be described with reference to Table 3 and FIG. FIG. 12 is a diagram illustrating a specific division example of the division methods 1 to 4.

  Of the methods shown in Table 3, the dividing method 1 is a method of dividing the waveform at equal intervals from the left end (no overlapping of sections), and is a basic method of dividing at equal intervals with respect to m / z. On the other hand, the dividing method 2 is a method of dividing (no overlapping of sections) into sections whose width varies according to the value of m / z from the left end of the waveform. For example, in SELDI TOF-MS, as m / z increases, the resolution (resolution) becomes coarse (decreases). Therefore, the m / z axis can be effectively made in a direction close to an equal interval by such a method. it can. Dividing method 3 is a method of dividing a section with a fixed width (for example, 1.2) around a position where the decimal point is 0 (with overlapping sections), and the center positions of the divided ranges are equally spaced. become. The method of the division method 4 is a method of dividing into sections (with overlapping sections) having a width that varies according to the value of m / z, centering on the position where the decimal point is 0 or less. In this embodiment, the division method 4 that is optimized in consideration of applying to the SELDI TOF-MS is adopted.

(2-3) Derivation of peak density curve For each of the sections obtained in the substep (2-2) regarding the division of the m / z axis, list all peak areas whose vertices are included in the section, and include them A process of regarding the total value of the levels as the peak density for the section is performed. The higher the inclusion level and the higher the peak density, the higher the peak. By calculating the peak density for each section and plotting it for all sections, a peak density curve along this plot is obtained. The peak density calculation method includes various alternative methods in addition to the basic method (basic method) for summing up the above-described inclusion levels. Table 4 shows typical examples of peak density calculation methods.

In the peak density calculation method 1 shown in Table 4, for a certain monotonously increasing function F (x), the peak density is obtained by the following equation.
Peak density = Σ F (CLi) i = 1,, n. (4)

However, in Formula (4), P1 to Pn are peaks belonging to the section, and CLi is the inclusion level of Pi. Various peak density calculation methods can be realized by specifically determining the monotonically increasing function F (x). Some examples are shown below.
F (x) = x (5)

When the monotonically increasing function F (x) is expressed by the equation (5), the peak density calculation method 1 matches the basic method. That is, the peak density corresponds to the sum of the inclusion levels.
Otherwise, the following formula can be used.
F (x) = x · x (6)

Using equation (6) as F (x), the peak density corresponds to the sum of the squares of the inclusion levels. Equation (6) is weighted considering that the higher the inclusion level, the greater the influence on the peak density.
F (x) = 0 (x <3)
1 (x ≧ 3) (7)
Equation (7) is a method in which a peak with a low inclusion level is estimated as a peak due to noise and ignored, and only peaks with an inclusion level of 3 or more are considered.

  Next, peak density calculation method 2 will be described. In the peak density calculation method 2, a section including the section as a center and having a width S times wider than the section is defined as a reference section. For the section and the reference section, the peak density is calculated by any one of the above-described peak density calculation methods 1 and is set to D1 and D2, respectively, and the numerical value S · D1 / D2 is changed to the actual peak density. FIG. 13 is a conceptual diagram of a peak density calculation method for obtaining the peak density by this method. Represents the relative height of the numerical value of the previous method compared to the surrounding area. According to this method, a more accurate peak density can be obtained by referring to a wide range before and after and comparing it with the front and back. In the present embodiment, the calculation method 2 is used. In the lower peak density calculation method used in calculation method 2, equation (6) is used as the monotonically increasing function F (x).

(2-4) Determination of grouping range Based on the peak density curve obtained in substep (2-3), the grouping range is determined by the following procedure. FIG. 14 is a diagram showing an outline of a grouping range determination method. First, if there is no section having a peak density larger than the threshold L ₀ in the peak density curve, the process is terminated. Second, if there is a section having a peak density larger than the threshold L ₀ in the peak density curve, the section having the highest peak density is set as the grouping center C. The peak density is L (a). Third, a predetermined range including the center C is set as a grouping range (a ′) by a method designated in advance as described later. Further, the reliability of the grouping range is L (a). Fourth, the peak density of all sections within the grouping range is set to “0” (b). Further, the left and right sides of the grouping range are sequentially inspected, and as long as the peak density decreases, the peak density in that section is set to “0” (b). Here, returning to the first, the same operation is performed for the remaining peaks (in this case, L is changed each time (c, d, e)). If there is no longer a section having a peak density greater than the threshold L ₀ , the process is terminated and the final grouping range is determined (e). Various methods can be used as a method for setting a certain range including C as a grouping range. A method for determining a representative grouping range is shown in Table 5.

  The grouping range determination method 1 in Table 5 is a method in which a section of ± w around the point C is set as the grouping range. w may be a fixed value or a value that varies depending on the value of m / z of the point C. The grouping range determination method 2 is a method in which the left side and the right side of the point C are sequentially inspected, and the point where the peak density has decreased to L ′ is set as the grouping range. Here, L ′ is a value determined by the following equation, for example, using the peak density L at the point C. As a formula, first, L ′ = L · α. However, α is a constant satisfying 0 ≦ α <1, for example, about 0.1. Alternatively, it can be determined based on L ′ = L−β. However, β is a constant satisfying 0 <β ≦ L. The grouping range determination method 3 is a method in which the above 1 and 2 are used in combination. In the present embodiment, the grouping range determination method 3 is used.

  In the above method 1, since a predetermined range is used as a grouping range without considering changes in peak density, adjacent peaks are assigned to the same grouping range in a portion where fine peaks are concentrated. There's a problem. On the other hand, since the method 2 does not limit the width of the obtained grouping range, there may be a case where a very wide grouping range that cannot be actually obtained is obtained. Method 3 can avoid the occurrence of these problems by using both methods together.

(2-5) Peak Grouping Returning to the original, each peak vertex is observed, and it is inspected to which grouping range the peak vertex belongs in the sub-step (2-4). Then, peaks belonging to the same grouping range are grouped. Peaks that do not belong to any grouping range are not grouped, i.e. considered to be noise.

  As described above, in the embodiment of the present invention, the reliability is given to each peak, and the total value of the reliability of peaks existing at the same position in a plurality of waveforms is obtained as the peak density. Gives peak group reliability based on density. As a result, the reliability of a group consisting mainly of noise peak groups or exceptional peaks that can be seen in only a small part of the waveform is set low. Therefore, while preventing quality degradation due to noise and not simply grouping nearby peaks, the peak density is calculated for each minute section, and grouping is performed in consideration of changes in peak density. It is possible to group only.

  Furthermore, if a large number of peak groups are obtained by assigning reliability to the peak group, the subsequent operations can be performed only from the peak group with high reliability or from the peak group with high reliability. There is an advantage that work efficiency can be improved.

  In the following, an example using the plasma of 40 rats is described. In FIG. 18, it is assumed that the 40 rats used are divided into several groups such as “normal rats” and “rats subjected to surgical treatment” depending on their properties. FIG. 18 is a flowchart showing the flow of processing according to this embodiment. As shown in FIG. 18, in step S1, measurement is performed by SELDI TOF-MS. The collected plasma of 40 rats is measured twice using SELDI TOF-MS, and the measurement results are output as waveform data that can be loaded into a computer. Waveform data is obtained for each measurement (step S1).

  Next, a computer process (step S2) described later is performed on each waveform data fetched into the computer. In step S2, first, waveform data is prepared, baseline removal (step S2-1) and smoothing processing are performed (step S2-2), and peak extraction processing is performed in step S2-3. In the peak extraction process in step S2-3, peak extraction from a single waveform in step S2-3-1 (step 1 above) and grouping of peaks from a plurality of waveforms in step S2-3-2 (above described above). Step 2) is performed. Next, in step S2-4, a difference between groups is detected based on the result of step S2-1 or the result of step S2-2 and the result of step S2-3.

  FIG. 19 shows 10 samples of waveform data (corresponding to point (a) in the figure) obtained by removing the baseline (step S2-1) and smoothing (step S2-2). When the waveform data of samples 1 to 10 shown in FIG. 19 are input to the peak extraction device, the peak extraction processing from a single waveform is performed individually. As a result, a peak for each waveform data and an inclusion level for each peak are obtained (corresponding to point (b) in FIG. 18). The peak extraction result for sample 1 in FIG. 19 is shown in FIG. FIG. 21 shows data obtained by enlarging a part of FIG. The waveforms in these figures are the data after the smoothing process. Moreover, the point in a figure shows a peak vertex.

  After performing peak extraction and inclusion level calculation for each waveform data, “step 2 grouping of peaks from multiple waveforms” is performed. The result of grouping is shown in FIG. FIG. 23 shows data obtained by enlarging a part of the data. In the upper part of the figure, the waveform data of samples 1 to 10 and the extracted peak vertices are shown. In the middle of the figure, the obtained peak density curve is drawn. In the lower part of the figure, peaks grouped based on the peak density curve are shown. The peak vertices indicated by the same mark belong to the same group.

  Next, based on the peak group obtained from the peak extractor and the grouping range thereof, a difference between a normal rat group (hereinafter referred to as group A) and a rat group subjected to surgical treatment (hereinafter referred to as group B) is detected. FIG. 24 is a diagram showing the waveform data by dividing the A group and the B group vertically in the vicinity of a typical grouping range where a difference is seen between the A group and the B group. FIG. 25 is a diagram showing peak vertices or representative points described later around the same grouping range. Hereinafter, the difference detection will be described with reference to FIGS. First, the intensity of each peak vertex is obtained for all peak vertices included in a certain peak group. Among the obtained intensities, those related to rats in Group A (black circle in FIG. 25) or Group B (marked in FIG. 25) are taken out (strength acquisition processing), and appropriate statistical tests such as Mann-Whitney U test are performed. The significance probability with respect to the difference in intensity between group A and group B is calculated by the method. Similar significance is calculated for all peak groups obtained from the peak extractor, and a peak group with sufficiently small significance is detected as a peak group in which a difference is recognized between the A group and the B group. The points to be noted in the “strength acquisition process” will be described below.

  (i) When there are multiple peaks on the same waveform data related to rats in Group A or Group B, select one peak from them and use it as a representative point. Take out. As a method for selecting a representative point, for example, there is a method in which a representative point having the maximum intensity is used.

  (ii) If none of the peaks in the same waveform data relating to the rats in Group A or Group B are included in the peak group, the representative points on the waveform data described above are set within the grouping range of the peak group. Choose one and take out its strength. As a method for selecting a representative point, for example, there is a method of using a point having the maximum intensity as a representative point.

(iii) When selecting representative points in (i) and (ii), the target waveform data may be either before smoothing or after smoothing ((e) and (f) in FIG. 18 respectively). However, it is necessary to unify which is the target throughout the data subject to statistical tests.
-Strength normalization may be performed as necessary.

  Although rat plasma was used in this example, the present invention can be similarly applied to various biological samples such as plasma, serum, urine, tissue extract, and spinal fluid of other animals or humans. Further, when the group to be compared is “healthy person”, “patient of disease S”, etc., it becomes possible to find a grouping range in which the healthy person and the disease patient can be distinguished. Furthermore, a biological sample such as plasma obtained separately is measured by SELIDIOF-MS to obtain waveform data, and by examining the intensity of the waveform data in a different grouping range, It is possible to estimate the presence or absence of a disease related to the provided human, that is, to make a diagnosis.

  In this embodiment, the statistical test method is used to detect the difference between the rat groups, but another method may be used as long as the method can detect the difference. In this example, the difference between the rat groups was detected for each single peak group, but the difference can also be detected by combining a plurality of peak groups. Further, in this example, the difference between the two groups of the A group and the B group was detected, but by using an appropriate technique such as the Kruskal-Wallis test, a difference between three or more groups is detected. Is also possible.

  Note that the method according to the present embodiment is applicable not only to SELDI TOF-MS but also to all devices that can obtain output in a waveform, such as liquid chromatograph and mass spectrometry. This is particularly effective when the peak is broad, that is, when the peak top is unclear or adjacent peaks overlap.

  Various biological samples such as plasma, serum, urine, tissue extract, and cerebrospinal fluid using mass spectrometer such as time-of-flight mass spectrometer (TOF-MS) and liquid chromatograph The present invention relates to a technique for extracting a waveform peak corresponding to biologically derived molecules such as proteins contained in a biological sample from waveform data of spectral characteristics obtained by measurement by information processing using an electronic computer.

It is a figure which shows the positioning of the proteome analysis apparatus by one embodiment of this invention. It is a figure which shows the schematic structural example of the peak extraction apparatus by this Embodiment. It is a figure which shows the outline | summary of the peak extraction process (step 1) and the peak grouping process (step 2) from single waveform data by this Embodiment. It is a figure which shows the process which extracts a peak from the single waveform data by this Embodiment, and calculates | requires an inclusion level. FIG. 4A to FIG. 4D show four cases where the peak search is finished. It is a figure which shows the peak area group by the several parameter value obtained in substep (1-1). Parameter P is a diagram showing a state of inspecting an inclusion relationship for the case of _{P 1, P 2, P 3} , P 4. It is a figure which shows the example which can be considered that it is included although it is incomplete inclusion relation among the inclusion relation formation conditions. It is a figure which shows a mode that the inclusion level (CL) of each peak area | region is set. It is a figure which shows the procedure which removes the peak area | region which includes any other peak area | region, and the peak area | region which does not have the inclusion relation with another peak area | region. It is a figure which shows the procedure which removes only one when there exists overlap between two peak areas. It is a figure which shows the grouping process of the peak obtained from several waveform data. It is a figure which shows the specific division example to the division | segmentation method methods 1-4 of the area. It is a peak density calculation method 2 and is a conceptual diagram of a peak density calculation method for obtaining a peak density by a method that defines a section including the section as a center and having a width S times wider than the section as a reference section. It is a figure which shows the outline of the grouping range determination method. It is a figure which shows the example of the procedure which performs peak extraction from single waveform data, and then performs grouping of peaks. It is a figure which shows the example of the procedure which performs said general process, performs peak extraction, and then performs grouping of peaks. It is a figure which shows the method of calculating the peak density for every micro area, and performing the grouping which considered the increase / decrease in peak density. It is a flowchart figure which shows the flow of the process by the Example of this invention. It is a figure which shows the sample 10 example of the waveform data (equivalent to the (a) point in FIG. 18) obtained by the removal and smoothing process of a baseline. It is a figure which shows the peak extraction result with respect to the sample 1 of FIG. It is a figure which shows the data which expanded a part in FIG. It is a figure which shows the result of having grouped the peak from several waveforms, after performing peak extraction and calculation of an inclusion level with respect to each waveform data. It is a figure which shows the data which expanded a part of FIG. In the vicinity of a typical grouping range where there is a difference between a normal rat group (hereinafter referred to as group A) and a surgically treated rat group (hereinafter referred to as group B), the waveforms are divided vertically into groups A and B. It is the figure which showed data. In the vicinity of a typical grouping range where there is a difference between a normal rat group (hereinafter referred to as Group A) and a surgically treated rat group (hereinafter referred to as Group B), representative points of each waveform within each grouped range FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Waveform data acquisition part, 3 ... Peak extraction part, 5 ... Grouping process part, 7 ... Data output part, 11 ... Difference detection part

Claims (9)

A peak extraction method for extracting peaks from waveform data in a plurality of spectral characteristics having a variable axis and an intensity axis,
For each point in the single waveform data of each spectral characteristic, it is estimated whether the point is in the vicinity of the peak on the variable axis, and consecutive points estimated to be in the vicinity of the peak For each of the parameters whose values are changed in a plurality of ways by performing the process of setting the set of peaks as a peak region on the basis of a parameter which is a parameter and the parameter value is changed in a plurality of ways. Determining different peak areas for,
Obtaining an inclusive relation of the peak area determined by the parameter, and assigning a first reliability of the peak area based on the number of the inclusive relation;
Preparing a plurality of the single waveform data, and determining a second reliability on the variable axis based on the plurality of first reliability by the plurality of waveform data;
A grouping range is determined based on the second reliability, and exists within the grouping range among a group of peak vertices obtained as the maximum point on the intensity axis in the peak region in the single waveform data And a step of grouping a group of peak vertices into one group. Estimating whether it is in the vicinity of the peak defines a point A in the waveform data;
Sequentially searching for the point B existing in the search direction on the variable axis with respect to the point A in the search direction and the decreasing direction of the variable on the variable axis of the point A; Sequentially searching in the increasing direction of the variable in the variable axis,
The searching step is terminated when the point B in the search direction corresponding to the point A corresponds to any of the following, and the case is regarded as a search end case, and the search in the decreasing direction and the increasing direction are performed. The peak extraction method according to claim 1, wherein it is estimated whether or not the point A is in the vicinity of a peak based on a result of the search end case relating to a search for a search.
Case 1: The intensity axis coordinate of the point B exceeds the upper threshold.
Case 2: The intensity axis coordinate of the point B is below the lower threshold.
Case 3: Does not apply to any of the above cases, and reaches the end of the waveform data.
However, the upper threshold and the lower threshold are values determined based on the parameters. 2. The step of assigning the first reliability includes sequentially tracing the peak region from the peak region as long as a peak region including the first region exists, and setting the number of traces as the reliability. 2. The peak extraction method according to 2.
However, when there are a plurality of routes when tracing, the value when the number of times of tracing is the maximum is taken as the reliability. The step of obtaining the second confidence level includes the step of dividing the variable axis into a plurality of sections and generating a peak density curve based on the first confidence level for each section,
4. The step of determining the grouping range includes a step of determining a grouping range according to a level of reliability based on the peak density curve and a certain threshold value. The peak extraction method according to any one of the above.   The peak extraction method according to any one of claims 1 to 4, wherein the spectral characteristic is a spectral characteristic resulting from a biological molecule. The program for making a computer perform the peak extraction method of any one of Claim 1-5 . A waveform data acquisition unit for acquiring waveform data in a plurality of spectral characteristics having a variable axis and an intensity axis;
A peak extraction unit for extracting a peak in single waveform data with the reliability of the peak; and
In a peak extraction device having a grouping processing unit that groups the peaks based on the reliability of the peaks,
The peak extraction unit estimates whether or not the point is in the vicinity of the peak on the variable axis with respect to the point in the single waveform data of each of the spectral characteristics. The process of setting the estimated set of consecutive points as the peak area is a parameter, and the value of the parameter is changed in multiple ways by changing the value of the parameter in multiple ways, thereby changing the value in multiple ways. A different peak area is obtained for each of the parameters, an inclusion relation of the peak areas obtained by the parameters is obtained, and a first reliability of the peak area is given based on the number of the inclusion relations. A plurality of the single waveform data, and a second signal on the variable axis based on the plurality of first reliability values obtained by the plurality of waveform data. Determine the degree,
The grouping processing unit determines a grouping range based on the second reliability, and includes a peak vertex group obtained as a maximum point on the intensity axis in a peak region in the single waveform data. The peak extraction apparatus characterized in that groups of peak vertices existing in the grouping range are grouped as one group. Further, one or more of the single peaks based on at least one of a difference in intensity of each peak included in the grouped peak or a difference in intensity in the grouping range of each of the single waveform data. The peak extraction apparatus according to claim 7 , further comprising a difference detection unit that detects a difference between a plurality of different waveform data groups, which is a waveform data group including waveform data. The peak extraction apparatus according to claim 7 or 8 , wherein a peak having a spectral characteristic caused by a biological molecule is extracted.

Images