Peak width estimation in mass spectrometry

By subdividing the mass spectrometry data and applying mathematical statistical analysis, the specific peak width of the interval is derived, which solves the problem of inaccurate peak width estimation in mass spectrometry data, and achieves rapid and automated peak width estimation and improves the accuracy of mass spectrometry analysis. consistency.

CN118937459APending Publication Date: 2024-11-12BRUKER DALTONIK GMBH & CO KG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410967633.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2021-02-26
Filing Date
2021-07-09
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The prior art is difficult to accurately estimate the peak width in mass spectrometry data, resulting in a degradation in peak picking algorithm performance, and the peak width changes increase with mass. The method of assuming a constant peak width may lead to inaccurate results and poor comparability.

Method used

By subdividing the mass spectrometry data into multiple intervals, applying mathematical statistical analysis such as autocorrelation and interpolation methods, deducing interval-specific peak widths, and using mapping functions to estimate peak widths over the entire mass range, suitable for different types of mass spectra instrument and analyte.

Benefits of technology

It realizes rapid and automated estimation of peak widths in mass spectrometry data, reduces calculation complexity, improves peak picking accuracy and consistency, and is suitable for a large range of MS data and different types of mass spectrometry instruments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118937459A_ABST
    Figure CN118937459A_ABST
Patent Text Reader

Abstract

The present disclosure relates to a method for processing a mass spectrum comprising: (i) providing a mass spectrum comprising a plurality of data pairs, each data pair representing a mass value or mass-related value on a mass scale or mass-related scale and an abundance value or abundance-related value associated with the respective mass value or mass-related value; (ii) selecting a plurality of intervals on a quality scale or a quality-related scale, each interval containing a large number of said data pairs; (iii) for each interval, applying a first mathematical statistical analysis to the data pairs contained in the respective interval to derive an interval-specific peak width; and (iv) using the interval-specific peak width to determine an estimated peak width for each mass value or mass-related value on a mass scale or mass-related scale.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] This application is a divisional application based on the Chinese patent application with the application date of July 9, 2021, application number 2021107779967, and invention name “Peak Width Estimation in Mass Spectrometry”. Technical Field

[0002] The present invention generally relates to a method for processing mass spectra, and in particular, to peak width estimation in mass spectrometry (MS) data. Background Art

[0003] Mass spectrometry In a typical mass spectrometry experiment, molecules are extracted from a sample, ionized in a mass analyzer, and detected. The resulting mass spectrum consists of a series of mass-intensity pairs, where the mass values ​​represent a small mass range (mass bin) and the intensities represent the (absolute or relative) abundance of molecules falling into the corresponding mass range.

[0004] More specifically, depending on the ion source used, singly or multiply charged ions can be produced. The actual physical quantity measured by the mass detector is the mass-to-charge ratio, expressed as m / z. Strictly speaking, therefore, a spectrum consists of pairs of m / z and intensity values. However, for simplicity, the term "mass" is used below as a synonym for mass-to-charge ratio.

[0005] Mass resolution, precision and peak width - A peak in a spectrum results from the detection of a certain number of ions with the same or very similar mass, however, it can also result from a single ion event. Due to the limited measurement resolution of the mass analyser, even ions with the same mass are not necessarily detected in the same single mass bin. Instead, the spectral peak can be spread over several adjacent mass bins, the lower the mass resolution, the broader the peak. Since the mass resolution of an instrument can depend on several factors including the experimental setup and sample preparation, the type of mass analyser instrument and the acquisition parameters, it is usually not possible or at least not easy to predict the expected peak width in a spectrum to a sufficiently accurate degree.

[0006] Furthermore, analysis of MS data is usually performed not on individual spectra but on a summed or averaged spectrum obtained by accumulating a large number of individual spectra. Due to inevitable inaccuracies in measuring ion masses, or in other words, due to mass precision limitations, peaks representing the same ion species in individual spectra may not have their centroids (or maximum or median intensities) in exactly the same mass bin, which can result in broadening of the peaks in the summed or averaged spectrum.

[0007] The width of a peak is usually expressed as the "full width at half maximum" (FWHM). This quantity describes the horizontal distance between the two positions to the left and right of the peak maximum, when the spectral intensity is less than half of the maximum intensity ( Figure 1 ideal value in ).

[0008] Peak width and spectral data analysis As mentioned above, each peak has its own peak width, and common peak picking methods usually produce the width of each individual peak and its position on an abscissa scale (e.g., time-of-flight or mass). However, many peak picking algorithms require an initial estimate of the peak width, and can produce incorrect results when an initial estimate that deviates too much from the actual width (i.e., too small or too large) is used.

[0009] Furthermore, accurate peak picking can be a computationally complex process, so it may be desirable to analyze spectral data without having to perform initial peak picking. This is again greatly simplified if a good estimate of the width of the spectral features in the data is known.

[0010] Peak Picking - Peak picking (or peak detection) algorithms attempt to distinguish actual ion mass peaks of analytical interest from omnipresent or ubiquitous noise in the mass spectrum (such as chemical or electronic noise). Prior knowledge of the peak widths of actual ion signal peaks in the mass spectrum to be processed greatly improves the performance of such peak picking, for example by accelerating the process.

[0011] The peak width of the ion signal is an important factor in the definition of resolution and therefore an important factor in the performance of the mass spectrometer used to acquire the mass spectrum. The peak width is often a characteristic of the particular detection scheme employed, such as a multichannel plate or a secondary electron cascade in a series of dynodes. Many peak picking algorithms process mass spectra with the a priori assumption that the peak width is essentially constant over the entire mass range processed, which can spread over thousands of atomic mass units or Daltons. However, this is often not true. Instead, the peak width can vary significantly over the entire mass range, typically increasing with increasing mass. Therefore, using unsupervised peak picking methods that assume constant peak width over the entire mass range may result in significant performance impairments, such as processing slowdowns; or may even prompt the user to manually intervene, which further delays processing; and may introduce subjectivity, which in turn affects the comparability of the results.

[0012] While it is possible to identify varying peak widths by regularly performing calibration measurements on the mass spectrometer used to acquire the mass spectra, the corresponding results represent only the state of the mass spectrometer at the time of calibration and do not take into account any time-related instrument or parameter shifts (drifts) that may have occurred since then.

[0013] The following is a brief description of the prior art that may be relevant to the present disclosure:

[0014] David PAKilgour et al. (J.Am.Soc.Mass Spectrom.(2017)28,253-262:"Autopiquer-a Robust and Reliable Peak Detection Algorithm for Mass Spectrometry") propose a simple algorithm that achieves robust and unsupervised peak detection by determining the noise threshold in isotope-resolved mass spectrometry data. It is said that solving this problem greatly reduces the subjective and time-consuming manual picking of mass spectrometry peaks, which proves to be beneficial in many research applications. The method uses autocorrelation to test the presence of (isotope) structure in overlapping windows on the spectrum. Within each window, the noise threshold is optimized to remove the most unstructured data while retaining as much (isotope) structure as possible. The algorithm has been demonstrated for both peak detection and spectral compression for data from many different categories of mass spectrometers and different sample types, and it is said that the method can be extended to other types of data containing regularly spaced discrete peaks.

[0015] U.S. Patent No. 5,073,713 discloses a method and apparatus for detecting and analyzing the separation of multiply charged ions by charge separation tandem mass spectrometry. Analyte molecules are ionized to form multiply charged parent ions. In the first stage mass spectrometer, a specific charge parent ion state is selected and its mass-to-charge ratio (m / z) is detected to determine its mass and charge. The selected parent ion is then separated, each of which is separated into a plurality of fragments including a group of daughter ions, each daughter ion having a mass of at least one molecular weight and a charge of at least one molecular weight. The number of daughter ion groups (sibling ions, daughterions) produced by the separation of a parent ion is different, but generally includes two to four ions, one or more multiply charged ions. The second stage mass spectrometer detects the mass-to-charge ratio (m / z) of the daughter ions and the time or time-space relationship between them. This relationship is used to associate the daughter ions to determine which (m / z) ratios belong to a group of sibling ions. The value of the mass and charge of each sibling ion is determined simultaneously by their corresponding (m / z) ratios, so that the sibling ion charge is an integer and the sum is equal to the parent ion charge.

[0016] U.S. Patent No. 5,900,628 discloses a method for extracting information about the mass of a substance from a complex mass spectrum containing peaks generated by multivalent ions. These multivalent ions possess charges that are integer multiples of the basic charge. Ions including these multivalent ions are detected and a mass spectrum is obtained. On a mass spectrum, the ion intensity I is expressed as a function of a variable x corresponding to the mass-to-charge ratio. The mass spectrum given by I(x) is converted to a mass spectrum I(t) of variable t using the relationship t=1 / (xH), where H is the unit mass of the ions added to the multivalent ions. Peaks appear on the mass spectrum I(t). Among these peaks, at least a given number of peaks appear at regular intervals. The spacing between these regularly spaced peaks is found.

[0017] U.S. Patent No. 6,104,027 proposes a method for identifying low-charge, low-weight ions with multiple charge states in mass spectrometry. Ions are matched to charge series by testing peak widths, isotope spacings, and isotope ratios using instrument resolution and information derived from the instrument spectrum. By following these tests, incorrect assignment of ions to charge series is avoided.

[0018] U.S. Publication No. 2005 / 0206363A1 discloses a method for processing a spectrum, which includes obtaining a plurality of spectra, each of which includes a signal including a signal intensity as a function of a time of flight, a mass-to-charge ratio, or a value derived from the time of flight or the mass-to-charge ratio. Then, a signal cluster is formed by clustering signals from the plurality of spectra with a time of flight, a mass-to-charge ratio, or a value derived from the time of flight or the mass-to-charge ratio within a window defined using an expected signal width value.

[0019] U.S. Publication No. 2014 / 0372048A1 proposes a time-of-flight (TOF) mass spectrometer for analyzing samples, which generates a time series of data points representing the amount of ions detected per unit time. The spectrometer resolution, the spectrometer digitization time period, and the minimum number of points per peak required to maintain the information content of the peak are received. The peak width value of each point is calculated from the resolution and the time of each point. The calculated peak width value of each point is divided by the minimum number of points per peak. The maximum time difference between the points of each point is generated. Points are selected based on the digitization time period. Neighboring points of the selected point are found. If the difference between adjacent points does not exceed the sum of the maximum time differences of the adjacent points, the selected point is deleted to compress the time series.

[0020] U.S. Publication No. 2014 / 0138526A1 shows a method for determining the mass-to-charge ratio of ions in a sample, which includes: determining a data acquisition time, wherein the data acquisition time is a predetermined fraction of the maximum flight time; providing a continuous beam of ions from the sample to a time-of-flight mass analyzer at a pulse interval having a duration equal to the predetermined fraction of the maximum flight time; and after adding data acquired during several pulse intervals, measuring the peak width and flight time values ​​of each of the ion species in the sample; and correcting the measured flight time value based on the correlation of the measured peak width value with calibration data of peak width versus flight time.

[0021] In view of the above, there remains a need for additional mass spectrometry pre-processing steps aimed at identifying and estimating in a more accurate manner the closely watched varying peak widths across the mass range before subsequent analysis of the mass spectra begins. Summary of the invention

[0022] A method for estimating the peak width of spectral features occurring in mass spectrometry (MS) data is described, which method is suitable for full automation.

[0023] Peak width is usually considered as a parameter of a single individual peak in a spectrum. On the other hand, spectral peaks that appear at similar masses within the same spectrum usually have very similar peak widths. In addition, the variation of peak width as a function of mass can usually be approximated by a low-degree polynomial (i.e., a constant, linear function, or quadratic function).

[0024] The method presented here provides an estimate of the functional relationship between peak width and mass. It is fast and does not rely on manual interaction by the user (and can therefore be fully automated), a priori assumptions about operating parameters, a previous peak picking step (although it can be followed by one), and is applicable to a wide range of MS data, including different analytes (e.g., lipids, metabolites, glycans, peptides, proteins) and different types of MS instrumentation (e.g., time of flight (either linear), reflectron and / or orthogonal TOF, Fourier transform ion cyclotron resonance, FT ICR, or Kingdon type analyzers such as Thermo Fisher Scientific's instruments, the latter two having inherently varying peak widths over the observed mass range).

[0025] In a first aspect, the present disclosure relates to a method for processing a mass spectrum, comprising: (i) providing a mass spectrum comprising a plurality of data pairs, each data pair representing a mass value or a mass-related value on a mass scale or a mass-related scale and an abundance value or an abundance-related value associated with the corresponding mass value or mass-related value, (ii) selecting a plurality of intervals on the mass scale or the mass-related scale, each interval comprising a large number of said data pairs, (iii) for each interval, applying a first mathematical statistical analysis to said data pairs contained in the corresponding interval to derive a (average) interval-specific peak width, and (iv) using said (average) interval-specific peak width to determine an estimated peak width for each (relevant) mass value or mass-related value on the mass scale or the mass-related scale, or using said (average) interval-specific peak width to derive a mapping function, by which an estimated peak width for each (relevant) mass value or mass-related value on the mass scale or the mass-related scale can be determined.

[0026] For the sake of clarity, it is emphasized here that in the context of the present disclosure, raw data from which actual physical parameters (such as mass, mass-to-charge ratio m / z or intensity) are derived, or in other words, raw data related to these physical parameters (related values), can in principle also be used to put the technical teachings of the present disclosure into practice, for example, the flight time from a time-of-flight mass spectrometer and the voltage amplitude from an ion detector before digitization, because they represent agents of actual physical parameters (such as mass and intensity that cannot be directly observed (e.g., flight time=>mass m or m / z; voltage=>intensity)), and they can and usually are converted into such actual physical parameters by precisely specified calculation rules. The method can also include, for example, measuring the raw data by calibration (e.g., measured time or frequency to mass) and / or by transformation (e.g., measured time signal to Fourier transform of spectrum) and processing the measured raw data. Preferably, the mass spectrum is densely filled with peaks, wherein substantially all biologically possible masses are present in the selected interval. The mapping function can be used to determine the estimated peak width of multiple selected mass values ​​or mass-related values ​​on a mass scale or a mass-related scale.

[0027] The first mathematical statistical analysis preferably comprises only a direct and non-iterative process on said data pairs (eg using autocorrelation of Fourier transforms) in order to provide a fast method.Preferably, the first mathematical statistical analysis of said data pairs applied to selected intervals is performed in parallel for different intervals.

[0028] In various embodiments, the derivation of the interval-specific peak widths may include calculating interval-specific reliabilities for each (selected) interval, and wherein the determination of the estimated peak width may use the interval-specific peak widths weighted as a function of the respective reliabilities.

[0029] In various embodiments, the estimated peak width can be determined by at least one of interpolating between interval-specific peak widths, extrapolating from interval-specific peak widths, fitting a curve to interval-specific peak widths, and performing regression analysis to derive a mathematical relationship that maps mass values ​​or mass-related values ​​to estimated peak widths. The estimated peak width can be determined by linear or spline interpolation. Alternatively (or in addition), the estimated peak width can be determined by fitting a (piecewise) constant, linear, or higher order polynomial.

[0030] In various embodiments, the first mathematical statistical analysis may include calculating the autocorrelation of the mass spectrum in the corresponding interval (ie, the correlation of the mass spectrum with a shifted version of itself). Preferably, the maximum value of the autocorrelation (R max ) and minimum value (R min ) and find the autocorrelation and its arithmetic mean (R max +R min ) / 2, the interval-specific peak width is determined as the full width at half maximum (FWHM). Width representation metrics other than FWHM may also be used in the methods according to the present disclosure. The calculation of interval-specific reliability may include calculating the total change in signal power and autocorrelation of the mass spectrum for each interval.

[0031] In various embodiments, the values ​​on the mass scale or mass-related scale can be expressed in atomic mass units or Daltons. It is expected that the estimated peak width of each mass value or mass-related value on the mass scale or mass-related scale is determined to be applicable to the mass range or mass-related range under consideration, that is, the range where the ion signal of interest usually occurs such as from the lower limit of the first selected interval on the scale to the upper limit of the last selected interval on the scale. Preferably, the length of each interval can be between 50 and 1000 atomic mass units or Daltons. The number of selected intervals is preferably between 2 and 50.

[0032] In various embodiments, the autocorrelation may be calculated for a hysteresis range between 0 and 1 atomic mass unit or Dalton on a mass scale or a mass-related scale.

[0033] In various embodiments, the selected intervals preferably have the same width on the mass scale (or mass-related scale), but this is not necessary. The selected intervals can cover the entire mass scale (or mass-related scale) of the mass spectrum in adjacent and continuous intervals. If the mass range contains data singularities caused by temporary acquisition problems, or if the mass range contains almost no identifiable peaks, making the data pool insufficient, it is also possible to select discontinuous mass intervals, for example, by omitting certain narrow mass ranges.

[0034] In various embodiments, the mass spectrum within each interval may be resampled to equally spaced values ​​on a mass scale or mass-related scale prior to the first mathematical statistical analysis. A constant spacing between equally spaced mass values ​​or mass-related values ​​can be selected for each interval so that the resampled mass spectrum within the corresponding interval consists of the same number of data pairs as the original mass spectrum. Preferably, the constant spacing between equally spaced mass values ​​or mass-related values ​​may be greater than or equal to one thousandth of an atomic mass unit or one millidalton.

[0035] In various embodiments, the mass spectrum supplemented with the estimated peak width can undergo a second mathematical statistical analysis. Preferably, the second mathematical statistical analysis uses the estimated peak width or a mapping function. Preferably, the second mathematical statistical analysis can be a peak picking algorithm for distinguishing signal peaks of different analytical interests in the mass spectrum. Many peak picking algorithms require an initial estimate of the peak width, and can produce incorrect results when using an initial estimate that deviates too much from the actual width (i.e., too small or too large). The peak picking algorithm preferably uses the estimated peak width as an initial estimate, which reduces computational complexity, speeds up peak picking and improves the accuracy achieved.

[0036] In various embodiments, providing a mass spectrum can include aggregating multiple individual mass spectra acquired by the same mass spectrometer within a limited time period. Multiple individual mass spectra can be acquired from a sample comprising one of a tissue, a tissue section, and an extract from a tissue. Preferably, aggregation can include averaging. The mass spectrum can, for example, be the sum or average of multiple individual mass spectra acquired at different locations on a tissue section. The estimated peak width can, for example, be used in a peak picking algorithm that is applied to the total or average spectrum itself and / or any one individual mass spectrum.

[0037] In a second aspect, the present disclosure relates to a mass spectrometer having an operating system adapted and configured to perform the method as described above.

[0038] In a third aspect, the present disclosure relates to a method for processing a mass spectrum, comprising: (i) providing a mass spectrum comprising a plurality of data pairs, each data pair representing a mass value or a mass-related value on a mass scale or a mass-related scale and an abundance value or an abundance-related value associated with the corresponding mass value or mass-related value, (ii) selecting a plurality of intervals on the mass scale or the mass-related scale, each interval comprising a large number of the data pairs, (iii) for each interval, applying a first mathematical statistical analysis to the data pairs contained in the corresponding interval to derive an interval-specific peak width, and (iv) using the interval-specific peak width to determine an estimated peak width of a plurality of selected mass values ​​or mass-related values ​​on the mass scale or the mass-related scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present invention may be better understood by reference to the following drawings. The elements in the drawings are not necessarily drawn to scale, emphasis instead being placed on illustrating (usually schematically) the principles of the invention:

[0040] Figure 1 A close-up of an example spectrum with a peak at mass 100.0 Daltons (dashed line) is schematically depicted. Individual spectral intensities are shown as vertical lines and black dots, with a bin width of approximately 0.087 Daltons. Using the FWHM criterion, the peak width is 0.4 Daltons (dashed line between open circles).

[0041] Figure 2 Schematic diagrams of embodiments of the proposed method are shown (from left to right). A mass spectrum is provided (step 1), the complete (or a portion of the complete) mass range or mass-related range is subdivided into a plurality of (continuous or discontinuous) intervals (step 2), an estimate (and optionally, an accuracy score) of the peak width is calculated for each interval (step 3), the estimate (and optionally, the score) is combined into a function defined over a selected mass range or mass-related range (step 4), and optionally, the obtained peak width estimate is used in further processing of the spectrum (step 5).

[0042] Figure 3 An example of the autocorrelation R used to calculate a spectrum is presented. Top: Input spectrum (1) and shifted copies of the same spectrum (2 to 5), corresponding to lag values ​​of 0.1, 0.25, 0.5 and 1.0 Dalton. For illustrative reasons, the data are restricted to the mass range between 1090 and 1100 Dalton. Bottom left: xy plot of the shift versus the original intensity values ​​for different lag values. Bottom right:

[0043] Autocorrelation as a function of the lagged value.

[0044] Figure 4 An example of applying a peak width estimation algorithm to the average spectrum of a mass spectrometry imaging data set is presented, resulting in a constant peak width estimate of 0.3702 Daltons over the entire mass range examined.

[0045] Figure 5 Another example is shown where the constant value (0.292 Daltons) is closest to the peak width over the entire mass range studied.

[0046] Figure 6 An example of applying a peak width estimation algorithm to the average spectrum of a mass spectrometry imaging data set is shown, resulting in a relative peak width estimate of 190.79 ppm, ie, the peak width increases linearly with mass.

[0047] Figure 7 Another example is depicted where the proportional approximation model for peak widths over a wide mass range produces the best result (65.80 ppm). DETAILED DESCRIPTION

[0048] While the invention has been shown and described with reference to several different embodiments thereof, those skilled in the art will recognize that various changes in form and details may be made therein without departing from the scope of the invention as defined in the following claims.

[0049] An embodiment of the proposed method may basically include the following steps (eg Figure 2 (schematically shown in ): (i) providing a spectrum. (ii) subdividing the (preferably complete) mass range (or mass correlation range) into two or more (continuous or discontinuous) mass intervals (or mass correlation intervals). (iii) for each mass interval (or mass correlation interval), calculating an estimate of the (average) peak width in that interval. (iii-a) optionally, calculating an accuracy score for each mass interval (or mass correlation interval) that describes the reliability of the corresponding peak width estimate. (iv) combining the peak width estimates (and optionally, the accuracy scores) to generate a mathematical function that maps any mass value (or mass correlation value) in the (preferably complete) mass range (or mass correlation range) to a corresponding peak width estimate. (v) optionally, performing further mathematical statistical analysis on the processed spectrum using the estimated peak width (e.g., peak picking). These steps are described in more detail below:

[0050] Provide spectrum

[0051] The spectrum relates ion abundance information (such as intensity) to ion mass that cannot be observed directly but must be inferred from proxy data, such as the flight time of the ions in the field-free drift region (such as Bruker's TOF mass spectrometers), the orbital frequency of ions in a magnetic field (ion cyclotron resonance mass spectrometers, ICR, such as Bruker's ) or the oscillation frequency of the ion in the secondary trapping potential of an electrostatic ion trap (Kingdon-type mass spectrometers, such as Thermo Fisher ), where the latter two are obtained by Fourier transform (FT) of the image current picked up during orbit and oscillation, respectively.

[0052] (ii) Subdividing the mass range (which may be a finite range on a theoretically infinite mass scale)

[0053] By subdividing the mass range (or mass-related range) into several intervals and estimating the (average) peak width in each interval, it is possible to derive the mathematical relationship between mass (or mass-related value) and peak width. However, the appropriate choice of the number of intervals is a trade-off between the accuracy of describing this relationship and the accuracy of estimating the (average) peak width in each interval. With regard to estimating the above-mentioned relationship, the finer the subdivision of the mass scale or mass-related scale, the lower the algorithmic complexity of estimating the peak width of each mass value or mass-related value using the derived interval-specific peak width may become. In an extreme but still feasible embodiment, for the purpose of subsequent data processing and / or further statistical analysis, the derived interval-specific (average) peak width can be assigned to all mass values ​​or mass-related values ​​in the (narrow) interval. Mathematically, this will be equivalent to piecewise constant interpolation.

[0054] As a rule of thumb, it is recommended to select equally sized, adjacent, continuous intervals (at least two intervals and no more than 50 intervals) with a length close to 200 Daltons to ensure sufficient data pooling for each interval. In some (rare) cases, if the mass range contains data singularities caused by temporary acquisition problems, or if the mass range contains few identifiable peaks making the data pool insufficient, even discontinuous mass intervals can be selected, for example by omitting certain narrow mass ranges.

[0055] In the following, the number of mass intervals (or mass-related intervals) will be denoted by K and the centers of these intervals by express.

[0056] (iii) Estimation of peak width

[0057] Within each mass bin (or mass correlation bin), the peak width estimation can be based on calculating the central autocorrelation of the corresponding spectral data points. Autocorrelation is a well-known concept that is used to analyze and describe certain types of regularities in signals. Roughly speaking, it measures the similarity between a signal and a shifted copy of itself expressed as a function of the shift lag. Here, the input signal is contained in a portion of the input spectrum within a given mass bin (or mass correlation bin). Figure 3 An illustrative example is shown in .

[0058] As a first step in calculating the autocorrelation, the spectral data within a given interval can be resampled to a constant mass bin spacing (or mass-dependent bin spacing). The bin spacing can be chosen so that the number of data points after resampling is the same as before resampling, subject to a minimum and maximum number of data points. A good choice for the minimum number of data points is six, and the maximum can be chosen so that the bin width is no less than 1 millidalton.

[0059] Therefore, we obtain the values ​​m with equal spacing.i and the corresponding intensity s i of the resampled data points (m i , s i ). i=1…N . The central autocorrelation R = (R j ) j=0…M is given by:

[0060]

[0061] where denotes the average value of (s i ), and the window width M is chosen such that m i+M - m i ≥ 1 Dalton, but ensuring 5 ≤ M < N. For typical mass spectrometry data, the plot of R is assumed to have a maximum at j = 0, decrease to a minimum at some j * and then increase again ( Figure 3 , bottom right). For the purpose of estimating the peak width, the absolute value assumed by R is immaterial. Instead, it examines how quickly R decreases to the minimum. This can be achieved by interpolating R with a piecewise linear function u(t) for 0 ≤ t ≤ j * and finding the point (t * , u * ) where u(t) assumes the intermediate level between the maximum (at t = 0) and the minimum (at t = j * ), thus solving

[0062]

[0063] The peak width estimate Δm for the considered mass interval is then given by:

[0064]

[0065] (iii-a) Calculate the accuracy score (optional)

[0066] If the spectral data within the considered mass interval (or mass-related interval) contains a sufficient number of peaks at a sufficiently high signal-to-noise ratio, the typical shape of R as shown in Figure 3 can be expected. If this is not the case, the shape of R may be more irregular, resulting in a less accurate peak width estimate. In the final step of finding the mathematical relationship between mass and peak width, it is desirable to reduce the dependence on those peak width estimates with low accuracy. Therefore, the accuracy score of each peak width estimate can be used as a weight factor in subsequent model approximation steps and can be calculated as follows:

[0067] For each mass bin, two factors can be considered to represent the total signal power of the spectral data in that bin and the rule of the autocorrelation function R. The power factor p can be obtained by taking the square root of R0 The regularity factor q can be obtained by dividing the total change of R(tv(R)) by the range of R(rg(R)),

[0068] q=tv(R) / rg(R),

[0069]

[0070] rg(R)=max(R j )-min(R j ).

[0071] In the case where rg(R)=0 or tv(R)≤rg(R), q is set to 1.

[0072] Given a total of K mass intervals (or mass correlation intervals), these factors p are calculated for each interval k=1...K k and q k . Factor q k Further converted to factor q′ k ,

[0073]

[0074] Where σ represents (q k ) k=1…K Therefore, p k and q′ k All assume higher values ​​of the mass interval (or mass correlation interval) where the peak width estimate is considered more reliable, i.e. obtained from stronger signal peaks and more regular autocorrelation functions, whereas small values ​​close to 0 are associated with less reliable peak width estimates.

[0075] To obtain the estimated peak width Δm for each k The accuracy score / weight factor w k , we can calculate p k and q′ k The geometric mean of , and the resulting weights are normalized to unity sum,

[0076]

[0077] (iv) Calculation of total peak width

[0078] From the previous step, the peak width estimate Δm has been obtained for the individual mass bins (or mass-related bins). k (and optionally, a weight factor w k), k = 1 ... K. If the more elaborate calculations above have not been performed, the weighting factors can be set to w k = 1 / K, thereby giving equal weight to each bin-specific peak width. The final step consists in finding a function F that maps the mass value m to the corresponding peak width Δm, thereby approximating the point in the following way When the corresponding weight factor w k When is large, the approximation error is small.

[0079] There are many different possible methods for finding such a function F, including linear or spline interpolation, various forms of regression, or any other type of curve fitting. The most common method is to find the function F by weighted least squares approximation, assuming that F belongs to a class of model functions (e.g., linear functions or quadratic functions) parameterized by some parameter vector θ. Therefore, F(m)=F(m; θ * ), and finding F is equivalent to finding the optimal parameter vector θ * , so that the weighted sum of squared errors E(θ) is minimized,

[0080]

[0081] As mentioned above, the actual peak widths that appear in the spectral data depend on the type of mass analyzer used and the details of the data acquisition process. It is well known that in many cases, a more or less constant peak width or a simple linear or quadratic relationship between mass (or mass-related value) and peak width can be assumed. For these cases, the solution of the above minimization problem is straightforward and can be obtained using the weighted sum M defined as μv To express:

[0082]

[0083] Model (1)

[0084] Constant peak width: The model function is a constant function F(m;θ0)=θ0 with a single parameter θ0. Optimal By Δm k The weighted average of

[0085] Model (2)

[0086] Peak width proportional to mass (or mass-dependent value): The model function is a linear function with a single parameter θ1 and no constant offset F(m;θ1)=θ1m. Optimal Given by:

[0087]

[0088] Model (3)

[0089] Linear peak width correlation: The model function is a linear function with two parameters, θ = (θ0, θ1) and (m; θ) = θ0 + θ1m. Optimal θ * Given by weighted linear regression,

[0090] and

[0091]

[0092] Model (4)

[0093] Peak width proportional to the square of the mass (or mass-related value): The model function is a quadratic function with a single parameter θ2: F(m;θ2) = θ2m 2 Optimal Given by:

[0094]

[0095] For any of these models, the above process produces a mapping function F(m) that can be used to calculate a peak width estimate Δm=F(m) for any mass m (or mass-related value) in the mass range (or mass-related range) covered by the spectral data. Since it may not be known in advance which of the above models best fits the data, it may be necessary to calculate solutions to multiple different models and select the model that produces the minimum sum of squared errors E(θ) or optimizes other appropriate quality parameters. Note that the first two models above (constant peak width and proportional peak width) are special cases of linear models. Therefore, the errors of models (1) and (2) may not be less than the error of model (3). A typical approach would be to calculate and compare solutions for models (1) and (2), solutions for models (1), (2) and (4), or solutions for models (3) and (4).

[0096] Example

[0097] Example 1 - Constant Peak Width

[0098] Figure 4 An example of applying the above peak width estimation algorithm to MS imaging experimental data is shown in FIG. In an MS imaging experiment, a tissue sample is scanned pixel by pixel ( Figure 4 , top left), and the mass spectrum of each pixel is acquired in the range of about 750 to 3000 m / z on a mass scale or mass-related scale. The peak width estimate is then applied to the average spectrum of all pixels (top row center). The MALDI reflectron time-of-flight analyzer from Bruker Get the data.

[0099] The results of the peak width estimation algorithm are as follows Figure 4Upper right: Estimated peak width Δm considering a total of K = 11 mass bins k The size of the circle marks represents the relative weight value w. k .

[0100] The peak width and weight were parameterized according to the above four models F(m;θ), namely (1) constant (solid horizontal line), (2) proportional (dashed line), (3) linear (dashed line) and (4) square model (dashed line). Among the three models (1), (2) and (4), the constant model produced the smallest model error in this case, resulting in a constant peak width estimate of Δm = 0.3702 Daltons.

[0101] The intermediate algorithm results are Figure 4 In the first step, the autocorrelation R is calculated for different lag values ​​and for each mass interval (middle row, left). In all mass ranges, the autocorrelation curves show Figure 3 The same characteristics as seen in the illustrative example in . In addition, a slight increase in the FWHM hysteresis value is observed.

[0102] To calculate the weight, calculate the signal power factor p k ( Figure 4 , middle panel), and the signal power factor and the rule factor q′ k Combine to get the weight value w l (middle row, right). Note that the autocorrelation curves (middle row, left) and weight factor curves (middle row, right) are shown as normalized for illustration purposes.

[0103] Figure 4 A comparison of the peak width estimate Δm=F(m) with the actual peaks present in the average spectrum is shown in the bottom row. Five major peaks in five different sub-ranges of the total mass range are shown from left to right, with the vertical dashed lines indicating the corresponding peak widths Δm as produced by the peak width estimation algorithm using the model that produced the lowest error deviation.

[0104] Figure 5 An example of estimating peak widths over a wide mass range using constant width (approximation model (1)) on MS imaging data that yields the best results is shown in Figure . The location and content of the figure correspond to Figure 4 The location and content in.

[0105] Example 2 - Ratio Peak Width

[0106] Figure 6A second example of applying the peak width estimation algorithm to an MS imaging data set of a tissue sample is shown in . The description of the figures given in the previous examples applies analogously to this figure. Again, the peak width estimation algorithm was used with a MALDI reflectron TOF mass spectrometer from Bruker. to get the data.

[0107] It can be observed that in this example, the individual peak width estimates Δm vary between 0.21 and 0.42 Daltons. k shows a stronger linear dependence on mass (grey circles in the upper right panel, FWHM markers in the middle row, left). Therefore, the proportional model (2) produces the smallest model error, yielding a peak width estimate of Δm = 190.79 × 10 -6 m, which corresponds to a relative peak width of 190.79 ppm (parts per million). This is consistent with the actual peak observed in the spectrum, such as Figure 6 shown in the bottom row of .

[0108] Figure 7 Another example of using the proportional relationship between mass and peak width (approximate model (2)) to estimate peak widths over a wide mass range on MS imaging data that yields the best results is shown in Figure 1. TM to get the data. As before, the location and content of the graph corresponds to Figure 6 The location and content in.

[0109] The present invention has been shown and described above with reference to a number of different embodiments of the present invention. However, it will be appreciated by those skilled in the art that various aspects or details of the present invention may be varied, or various aspects or details of different embodiments may be arbitrarily combined, without departing from the scope of the present invention, if practicable. In general, the foregoing description is intended only for illustrative purposes and not for limiting purposes of the present invention, which is limited only by the appended claims, which may include any equivalent embodiments as appropriate.

Claims

1. A method of processing a mass spectrum, the mass spectrum comprising a first set of signals and a second set of signals, the first set of signals comprising actual ion mass peaks of analytical interest, the second set of signals comprising omnipresent or ubiquitous noise, such as chemical or electronic noise, the method being used to improve the distinction between the first set of signals and the second set of signals in the mass spectrum, the method comprising: providing a mass spectrum comprising a plurality of data pairs, each data pair representing a mass value or a mass-related value on a mass scale or a mass-related scale and an abundance value or an abundance-related value associated with the corresponding mass value or mass-related value, selecting a plurality of intervals on the mass scale or mass-related scale, each interval containing a plurality of the data pairs, For each bin, applying a first mathematical statistical analysis to said data pairs contained in the corresponding bin to derive a bin-specific peak width, using the bin-specific peak widths to determine an estimated peak width for each mass value or mass-related value on the mass scale or mass-related scale, and The mass spectrum is supplemented with the estimated peak widths and subjected to a second mathematical statistical analysis, wherein the second mathematical statistical analysis is a peak picking or peak detection algorithm for distinguishing the first group of signals from the second group of signals in the mass spectrum.

2. The method according to claim 1, wherein: Derivation of the bin-specific peak widths comprises calculating, for each bin, a bin-specific reliability, and wherein determination of the estimated peak widths uses the bin-specific peak widths weighted as a function of the respective reliability.

3. The method according to claim 1, wherein: The estimated peak width is determined by at least one of interpolating between the interval-specific peak widths, extrapolating from the interval-specific peak widths, fitting a curve to the interval-specific peak widths, and performing a regression analysis to derive a mathematical relationship that maps mass values ​​or mass-related values ​​to the estimated peak width.

4. The method according to claim 3, wherein: The estimated peak width is determined by linear interpolation or spline interpolation.

5. The method according to claim 3, wherein: The estimated peak width is determined by fitting a constant, linear or higher order polynomial.

6. The method according to claim 1, wherein: The first mathematical statistical analysis includes calculating an autocorrelation of the mass spectrum within a corresponding interval, the autocorrelation of the mass spectrum indicating a correlation of the mass spectrum with a shifted version of itself.

7. The method according to claim 6, wherein: By calculating the maximum value of the autocorrelation (R max ) and minimum value (R min ), and find the autocorrelation and arithmetic mean (R max +R min ) / 2, the peak width specific to that interval was determined as the full width at half maximum (FWHM).

8. The method according to claim 6, wherein: Calculating the interval-specific reliability includes calculating the signal power of the mass spectrum and the total change of the autocorrelation for each interval.

9. The method according to claim 1, wherein: Values ​​on such mass scales or mass-related scales are expressed in atomic mass units or in Daltons.

10. The method according to claim 9, wherein: The length of each interval is between 50 and 1000 atomic mass units or Daltons.

11. The method according to claim 6, wherein: The autocorrelations are calculated for hysteresis ranging between 0 and 1 atomic mass unit or Dalton on the mass scale or mass related scale.

12. The method according to claim 1, wherein: Prior to the first mathematical statistical analysis, the mass spectrum within each bin is resampled to equally spaced values ​​on the mass scale or mass-related scale.

13. The method according to claim 12, wherein: Equidistant mass values ​​or constant intervals between mass-related values ​​were chosen for each bin so that the resampled mass spectra in the corresponding bins consisted of the same number of data pairs as the original mass spectra.

14. The method according to claim 12, wherein: The constant spacing between equidistant mass values ​​or mass-related values ​​is greater than or equal to one thousandth of an atomic mass unit or one millidalton.

15. The method according to claim 1, wherein: The plurality of intervals are selected to be continuous or discontinuous on the mass scale or mass-related scale.

16. The method according to claim 1, wherein: The method is performed during a mass spectrometry imaging experiment.

17. The method according to claim 1, wherein: Providing the mass spectrum includes aggregating a plurality of individual mass spectra acquired by the same mass spectrometer within a finite period of time.

18. The method according to claim 17, wherein: The plurality of individual mass spectra are acquired from a sample selected from the group consisting of a tissue, a tissue section, and an extract from a tissue.

19. The method according to claim 17, wherein: The aggregating includes averaging.

20. A mass spectrometer having an operating system adapted and configured to perform the method according to claim 1.

21. A method for processing a mass spectrum, the mass spectrum comprising a first set of signals and a second set of signals, the first set of signals comprising actual ion mass peaks of analytical interest, the second set of signals comprising omnipresent or ubiquitous noise, such as chemical or electronic noise, the method being used to improve the distinction between the first set of signals and the second set of signals in the mass spectrum, the method comprising: providing said mass spectrum comprising a plurality of data pairs, each data pair representing a mass value or a mass-related value on a mass scale or a mass-related scale and an abundance value or an abundance-related value associated with the corresponding mass value or mass-related value, selecting a plurality of intervals on the mass scale or mass-related scale, each interval containing a plurality of the data pairs, For each bin, applying a first mathematical statistical analysis to said data pairs contained in the corresponding bin to derive a bin-specific peak width, using the interval-specific peak widths to determine estimated peak widths for a plurality of selected mass values ​​or mass-related values ​​on the mass scale or mass-related scale, and The mass spectrum is supplemented with the estimated peak widths and subjected to a second mathematical statistical analysis, wherein the second mathematical statistical analysis is a peak picking or peak detection algorithm for distinguishing the first group of signals from the second group of signals in the mass spectrum.