Sample analysis method

Abstract

To provide a sample analysis method capable of precisely analyzing an observation spectrum obtained by measurement.SOLUTION: A sample analysis method includes: a step of receiving an acquired observation spectrum by measuring a sample by a mass spectrometer 1; a step of assuming the number of constituents contained in the sample; a step of acquiring a generation spectrum from a parameter regarding the mass of a constituent and the charge of a constituent for the assumed number of constituents; and a step of acquiring information regarding the number of constituents contained in a sample and the respective constituents by comparing a generation spectrum and an observation spectrum.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a method for analyzing a sample. [Background technology]

[0002] MS (mass spectrometry) is a powerful analytical method used for various purposes, such as pharmaceutical development and quality assurance, food inspection, and monitoring of environmental pollutants. In the manufacture of antibodies and nucleic acid drugs, which have been actively developed in recent years, impurities with different modifications are generated. They cause a decrease in the stability, pharmacokinetics, and efficacy of drugs, so it is important to distinguish these impurities. Therefore, it is important in pharmaceutical development and quality assurance to distinguish between the multiple impurities contained in drugs and take measures against them. In addition, if the monoisotopic mass of the component is known, it will be useful information to consider the cause of the impurity generation, and if the ion amount of the component is known, it will be useful information to estimate the degree of the impurity's influence.

[0003] Patent Document 1 discloses a sample analysis method in which an MS spectrum (mass spectrum) obtained by measuring a sample with a mass spectrometer is analyzed by Bayesian estimation to generate a plurality of distributions, and information on the components contained in the sample is obtained from one of the generated distributions that is closest to the result of the MS spectrum. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] U.S. Patent No. 8,604,421 Summary of the Invention [Problem to be solved by the invention]

[0005] Previous attempts have been made to estimate the monoisotopic mass and ion abundance contained in MS spectra, but no prior art has examined a technique for estimating the number of components.

[0006] The present invention has been made to solve the above-mentioned problems, and one object of the present invention is to provide a sample analysis method capable of accurately analyzing an observed spectrum obtained by measurement. [Means for solving the problem]

[0007] A sample analysis method in one aspect of the present invention includes the steps of receiving an observed spectrum obtained by a mass spectrometer, assuming the number of components contained in the sample, obtaining a product spectrum for the assumed number of components from parameters related to the masses of the components and parameters related to the charges of the components, and comparing the product spectrum with the observed spectrum to obtain information regarding the number of components contained in the sample. Effect of the Invention

[0008] The sample analysis method in the above aspect includes the steps of: assuming the number of components contained in the sample; obtaining a product spectrum from parameters related to the masses of the components and parameters related to the charges of the components for the assumed number of components; and comparing the product spectrum with the observed spectrum to obtain information on the number of components contained in the sample. This allows the product spectrum to be generated using the masses and charges of the components contained in the sample as parameters, independent of the observed spectrum, when the number of components is unknown. By comparing the generated product spectrum with the observed spectrum, a suitable product spectrum can be extracted. As a result, the values ​​of parameters including the number of components, masses, and charges when a suitable product spectrum is generated can be obtained with high accuracy. As a result, the masses and charges of the components can be obtained with high accuracy by analyzing the observed spectrum obtained by measuring a sample with an unknown number of components, and the observed spectrum obtained by measurement can be analyzed with high accuracy. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram illustrating an example of a sample analysis system according to one embodiment. [Diagram 2] 1 is a diagram for explaining a sample analysis method according to an embodiment. FIG. [Diagram 3] 1 is a flowchart illustrating a sample analysis process according to one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

[0011] (Configuration of sample analysis system) The configuration of a sample analyzing system 100 according to this embodiment will be described with reference to FIG.

[0012] As shown in Fig. 1, sample analysis system 100 includes a mass spectrometer 1 and a computer 2. Sample analysis system 100 is configured to detect components in a sample and perform analysis. Mass spectrometer 1 is configured to measure the sample. Computer 2 is configured to analyze the measurement results of mass spectrometer 1.

[0013] The mass spectrometer 1 measures and analyzes the masses of components contained in a sample. The mass spectrometer 1 includes a time-of-flight mass spectrometer. In other words, the mass spectrometer 1 has high sensitivity and resolution. The mass spectrometer 1 performs mass analysis of a sample for applications such as pharmaceutical development and quality assurance, food inspection, and monitoring of environmental pollutants. For example, mass analysis is used to identify impurities in the manufacture of antibodies and nucleic acid drugs.

[0014] The computer 2 is configured to acquire the measurement results measured by the mass spectrometer 1, analyze the measurement results, and output information on the components contained in the sample. The computer 2 includes a CPU (Central Processing Unit) 21 and a storage unit 22. The CPU 21 of the computer 2 is configured to perform analysis processing of the sample based on a program 221 stored in the storage unit 22. The computer 2 is also configured to accept operations from an operator via the operation unit 3. The computer 2 is also configured to control the display unit 4 to display a screen for operation, a display of the analysis results, and the like.

[0015] Here, MS (mass spectrometry) is a powerful analytical method used for various purposes such as pharmaceutical development and quality assurance, food inspection, and monitoring of environmental pollutants. In the manufacture of antibodies and nucleic acid drugs, which have been actively developed in recent years, impurities with different modifications are generated. These impurities may cause a decrease in the stability, pharmacokinetics, and efficacy of the drug. Therefore, it is important in the development and quality assurance of pharmaceuticals to distinguish between the multiple impurities contained in pharmaceuticals and take measures against them. In addition, if the monoisotopic mass of the component is known, it will be useful information to consider the cause of the impurity generation, and if the ion amount of the component is known, it will be useful information to estimate the degree of the impact of the impurity.

[0016] However, in mass spectrometry, it is generally difficult to directly distinguish impurities with slight modifications of targets of medium or larger molecules and to estimate the correct number of components, their monoisotopic masses, and ion amounts, because the presence of isotopes in the target compounds complicates the spectrum. In particular, the most widely used ESI (electrospray ionization) method produces multiply charged ions, which results in a more complex spectrum, leading to many interpretations and degrees of freedom, making analysis even more difficult. Therefore, a method can be considered to distinguish slight differences between isotopes and modifications by increasing the hardware resolution. For example, FT-ICR is a mass spectrometry method with high resolution, but it is difficult to use it simply because it requires large equipment and a large amount of money. Therefore, it is desirable to be able to analyze using equipment that can be used in a general laboratory, such as Triple-Quadrupole-MS or Q-TOF-MS.

[0017] Therefore, software-based signal analysis is also becoming popular. Various attempts have been made to estimate mass from mass spectrometer data. For example, wavelet transform is a simple method to obtain a list of m / z from a spectrum. In recent years, a peak detection algorithm has been developed that combines continuous wavelet transform and image processing. In this method, image segmentation is applied to the CWT coefficients to create a mask, which is then combined with the CWT maximum value to improve the ROC of peak detection. This method is useful for separating components with similar masses in the measured spectrum when the measurement target is a small molecule with a relatively small isotope distribution spread, or when ionization is performed using methods such as EI and MALDI, which result in a simple charge distribution. However, the spectrum of medium to large molecules with a wide isotope distribution, especially the multicharged ions generated by ESI, has a charge distribution, making it difficult to distinguish the monoisotopic mass that is the subject of interest.

[0018] Therefore, the ReSpect algorithm using the Max Entropy method has long been used for deconvolution to obtain monoisotopic masses from multiply charged ion spectra. This algorithm makes it possible to obtain monoisotopic masses by integrating m / z lists based on the charge distribution constraints. However, ReSpect cannot explicitly estimate the number of components contained in a spectrum, and cannot evaluate discrete states such as the probability that the number of components is k or the probability that the number is k+1. In addition, the entropy term of the objective function becomes larger as the number of peaks in the deconvoluted spectrum increases, so spectra with a large number of peaks are selected.

[0019] In recent years, new methods such as UniDec, which uses Bayesian deconvolution, have emerged. UniDec employs a unique algorithm similar to the Richardson-Lucy method, and is faster than ReSpect. However, the method of iteratively deconvolving to get closer to the observed data does not solve the problem of being unable to evaluate the probability that the number of components is a certain value.

[0020] Therefore, in the present invention, the probability that the number of components is a certain value is compared and evaluated in addition to estimating the monoisotopic mass and ion amount, and the most likely number of components is estimated and obtained. This makes it possible to suggest the presence of impurities in pharmaceuticals. This can contribute to the search for better synthesis conditions for medium to large molecular weight pharmaceuticals and quality assurance in factories.

[0021] 2, the sample analysis method by computer 2 includes the steps of receiving an observed spectrum (m / z distribution) obtained by measuring a sample with a mass spectrometer, assuming the number of components contained in the sample, acquiring a generated spectrum (m / z distribution) for the assumed number of components using the masses and charges of the components as parameters, and comparing the generated spectrum with the observed spectrum to acquire information regarding the number of components contained in the sample. The information regarding the number of components contained in the sample includes at least one of an optimal number of components and a distribution of a suitable number of components.

[0022] In the step of acquiring the generated spectrum, the obtained generated spectrum is compared with the observed spectrum, and parameters are updated so as to approach the observed spectrum, thereby acquiring a new generated spectrum in sequence, thereby acquiring the generated spectrum. Note that a generated spectrum for each parameter may be acquired, and then the generated spectrum may be compared with the observed spectrum to acquire a suitable generated spectrum.

[0023] In the step of acquiring the product spectrum, a charge distribution is assumed from the charges of the components as parameters, and the product spectrum is acquired based on the assumed charge distribution.

[0024] In addition, in the step of acquiring the generated spectrum, the number of components is changed and a generated spectrum is acquired for each of the changed numbers of components. In addition, in the step of acquiring the generated spectrum, the number of components is changed while increasing by 1, and a generated spectrum is acquired for each of the changed numbers of components.

[0025] In addition, the step of acquiring the produced spectrum and the step of acquiring information on the number of components contained in the sample by comparing the produced spectrum with the observed spectrum are repeatedly performed while changing the number of components within a predetermined range, i.e., the process is repeated while changing the number of components by one from the number of components set at the lower limit to the number of components set at the upper limit.

[0026] Furthermore, in the step of acquiring the product spectrum, based on the mass of the first component contained in the sample obtained using a first number of components (e.g., k), the range of change in the mass of the first component as a parameter to be changed is set smaller for a second number of components (e.g., k+1) greater than the first number of components, and a product spectrum is acquired.

[0027] In addition, in the step of acquiring information regarding the number of components contained in the sample, the generated spectrum and the observed spectrum are compared for a number of different component numbers, and information regarding the number of components contained in the sample is acquired based on the generated spectrum that is closest to the observed spectrum.

[0028] Specifically, in the step of acquiring information regarding the number of components contained in the sample, the generated spectrum and the observed spectrum are compared for a number of different numbers of components, and information regarding the number of components contained in the sample is acquired based on the generated spectrum that is closest to the observed spectrum.

[0029] The sample analysis method by computer 2 also includes a step of acquiring an analyte including the mass and charge of a component contained in the sample based on the generated spectrum approximating the observed spectrum. That is, the mass and charge of the component are acquired as the analyte, as parameters used in generating the generated spectrum approximating the observed spectrum.

[0030] In addition, in the step of acquiring the product spectrum, for each of the components contained in the sample, which is one or more components, a mass distribution of the mass of the component as a parameter to be changed is calculated, and the product spectrum is acquired based on the calculated mass distribution of each component. The mass distribution is calculated based on the monoisotopic mass of the component, the abundance ratio of the isotopes, and the number of atoms. For example, the mass distribution is calculated by approximating it using a binomial distribution.

[0031] In addition, in the step of acquiring the product spectrum, a charge distribution of charges as a parameter to be changed is calculated for each of the components contained in the sample as one or more components, and the product spectrum is acquired based on the calculated charge distribution of each component. The charge distribution is calculated based on the charge rate of the functional group of the component and the number of functional groups. For example, the charge distribution is calculated by approximating it with a binomial distribution.

[0032] In addition, in the step of acquiring information regarding the number of components contained in the sample, a plurality of generated spectra and an observed spectrum are compared based on posterior probability to acquire information regarding the number of components contained in the sample. In other words, a generated spectrum having a high likelihood (small difference) is extracted based on the difference between the generated spectrum and the observed spectrum.

[0033] In the sample analysis method using computer 2, first, a mass spectrometry system is modeled using parameters including the mass and charge of each component for each number of components assumed to be contained in the sample. Then, MAP estimation of these parameters is performed from the measured observation spectrum. Then, the maximum posterior probability of the model for each number of components is compared and evaluated to determine the model with the most suitable number of components. Note that this model has a large number of parameters. In addition, the posterior probability of the monoisotopic mass, which is one of the parameters, is flat in most of the search area, with multiple locally steep peaks. Therefore, the gradient method is not suitable for this case because the gradient is expected to disappear during the search process. Therefore, the means for estimating the parameters is a combination of NUTS, a type of MCMC that probabilistically searches the parameter space, and Simulated Annealing.

[0034] The purpose of using Simulated Annealing is to introduce a temperature parameter and select a high-temperature search parameter distribution to perform a sufficiently active parameter search even in areas where the posterior probability is flat or has a steep peak. This makes it possible to widely search the entire parameter space and to reduce the likelihood of falling into a local solution, thereby reducing the probability of missing a global solution.

[0035] In addition, NUTS can search parameters broadly and thinly in areas with small gradients, and search parameters in detail and with focus in areas with large gradients. Therefore, by introducing NUTS, it is possible to efficiently search a wide, high-dimensional parameter space.

[0036] On the other hand, while MCMC is good at searching for global solutions, it does not necessarily arrive at the optimal solution within a certain number of search steps. Therefore, parameters with large posterior probabilities obtained by NUTS and Simulated Annealing are used as initial values, and probabilistic variational inference is applied to search for the optimal parameter solution with the largest posterior probability in the vicinity, thereby improving the accuracy of parameter estimation. However, as the number of components increases, a kind of curse of dimensionality occurs, in which the parameter space to be searched expands explosively, and there is a possibility that the search efficiency and search accuracy will decrease. Therefore, we set the number of components k=1 to the maximum possible number of components k=k max We use a method in which the number of components is successively increased up to and the optimal parameters and posterior probability distribution are calculated each time.

[0037] Therefore, in order to utilize prior knowledge regarding the balance between the complexity of the model (number of components) and the suitability of the model (loss to data), a term including the number of components is added to the prior distribution for each parameter. In addition, a term including the distance between multiple monoisotopic masses is also incorporated into the prior distribution so that the monoisotopic masses of multiple components do not take the same value in the model. Then, a model with the number of components k=1 is created, and the optimal parameters and maximum posterior probability are obtained based on the above prior distribution. Next, a model with the number of components k=2 is created. At this time, the prior distribution of the parameters of the first component of k=2 is narrowed down to the optimal parameters already estimated with k=1. This prevents a significant increase in the parameter space to be searched. Based on this new prior distribution, the optimal parameters are estimated and the maximum posterior probability is obtained.

[0038] Hereafter, the upper limit of the number of components assumed is k maxThe method efficiently searches for optimal parameters for each model with each number of components, and obtains the maximum a posteriori probability. Finally, the maximum a posteriori probabilities corresponding to each model with each number of components are compared, the model with the highest probability is selected, and the associated monoisotopic mass and ion abundance are estimated.

[0039] (Physical model of a mass spectrometer) In a time-of-flight mass spectrometer, the flight time t tof The relationship between mass m and charge z is ideally expressed as equation (1).

number

[0040] The actual signal obtained is the delta function δ(t tof ) convolved with the detector response waveform R(t). tof is a log-normal distribution Λ tof It is known that the dispersion equation (2) follows:

number

[0041] In reality, the response S(t) when a single ion of mass m and charge z enters the detector is expressed as shown in equation (3), taking into account the variation in flight time and the variation in detector response intensity.

number

[0042] The mass m and charge z of component j physically follow the distributions shown in the following formulas (4) and (5), respectively. Since the natural abundance ratio of isotopes differs depending on the element, and the charge rate differs for each functional group, the probability mass function of the distribution to which the mass and charge belong is originally expressed as a multinomial distribution as shown below.

number

[0043] Since this model has a large number of parameters, when executing MCMC, a simple binomial distribution approximated by a representative value is used as follows: In this approximation, assuming that the probability of isotope substitution for all atoms is equal, the neutron increment is 1, and that the probability of charging for all functional groups is equal, formulas (6) and (7) are derived.

number

[0044] From the above, the number of ions contained in a certain component j is i jmax If we assume that there are 1, the spectrum can be expressed as shown in equation (8-1).

number

number

[0045] The noise of a time-of-flight mass spectrometer has been found to be stationary and normally distributed in a prior experiment. Therefore, the noise to be added to the entire spectrum is r(t)~N(μ noise ,σ noise ) It is known that the thermal noise of detection circuits such as those found in mass spectrometers follows a normal distribution, and in this case, it is presumed that thermal noise is the dominant factor in the overall noise. From this, the spectrum of one component combined with multiple components can be expressed as in equation (9).

number

[0046] (Estimation algorithm) <Probability model> Observation Data S obs When (t) is obtained, the parameters are θ:[(m' 1 ,I 1 , n 1 , u 1 , l 1 , v 1 ),···(m' j ,I j , n j , u j , l j , v j )], the posterior distribution P(θ|S obs (t)) is defined by Bayes' theorem as follows:

number

[0047] Here, the prior distribution of each parameter is basically uniform over the domain. However, as shown in equation (10), the prior distribution includes a regularization term w ex and a regularization term w to strike an appropriate balance between the model's complexity (number of components) and its fit (loss to the data). bic In other words, we introduce a logarithmic prior distribution as shown in Eq. (11).

number

[0048] To select an appropriate number of components k, we use the Bayesian Information Criterion (BIC) term that represents the model complexity, and we use the regularization term w bic is defined as in equation (12). BIC is a statistical index that balances the fit to the data and the complexity of the model when evaluating the quality of the model. n is the number of observations S obs It represents the dimension of (t), which is the number of data points in the time direction.

number

[0049] In addition, in this case, one component is defined as having a certain monoisotopic mass. Therefore, if the estimated values ​​of the monoisotopic mass parameters of multiple components in the algorithm end up being the same value, the number of components cannot be counted correctly. For this reason, an extremely large penalty is imposed when the estimated values ​​of two components approach a certain interval or less, without affecting the estimates of components with different monoisotopic masses. Here, a penalty that increases exponentially when the difference between the estimated values ​​of the monoisotopic masses is small is used, and the logarithmic prior distribution (regularization term) w exis defined as equation (13). In addition, the integral of the spectrum is multiplied as a coefficient so that the magnitude of the effect of the penalty does not change depending on the scale of the observed data.

number

[0050] Here, the spectrum obtained by substituting the parameter θ generated by MCMC etc. into the model of equation (9) is called S prop For (t), we assume that the noise has a normal distribution. Note that σ, which represents the standard deviation of the noise, is set to, for example, 2000. The logarithm of the posterior probability distribution can be expressed as in Equation (14).

number

[0051] <Markov Chain Monte Carlo Method> Posterior distribution P(θ|S obsFrom the set of parameters θ, we sample the parameters θ and select the parameter θ that maximizes the posterior probability. The sampling method we use is NUTS, which is one of the Markov chain Monte Carlo methods (MCMC) that has been widely used in recent years. NUTS (No-U-Turn Sampler) is a type of Markov chain Monte Carlo method (MCMC). In particular, it is a derivative of the Hamiltonian Monte Carlo method (HMC). The Hamiltonian Monte Carlo method (HMC) is a method for efficiently sampling from high-dimensional probability distributions using concepts from physics. However, it is difficult to use HMC in order to appropriately select the number of leapfrog steps (the number of steps the parameters move during the simulation). In other words, if the number of steps is too small, the sampler cannot move effectively through the search space, and if the number of steps is too large, the sampler will make a U-turn back to the point where the search began.

[0052] NUTS dynamically chooses an appropriate number of steps to explore the energy surface of a Hamiltonian based on the principle that the step should be terminated before the sampler begins to make a U-turn, which allows NUTS to solve the leapfrog step number adjustment problem of HMC and efficiently sample from high-dimensional probability distributions.

[0053] The Hamiltonian is defined as H(q,p)=U(q)+K(p), where q is the variable we want to sample from the desired probability distribution (position) and p is an auxiliary variable (momentum), U(q) is the potential energy and K(p) is the kinetic energy.

[0054] First, we randomly initialize the momentum p for the current position q, and use the leapfrog algorithm to calculate a new (q,p) pair. In this process, we first update the momentum p by a half step, p(t+ε / 2)=p(t)-ε / 2×∂U / ∂q(t), where ε is the step size and t is the current time step. Next, we update the position q by a full step, p(t+ε)=q(t)+ε×∂K / ∂p(t+ε / 2). Finally, we update the momentum p by another half step, p(t+ε)=p(t+ε / 2)-ε / 2×∂U / ∂q(t+ε).

[0055] If p·p'≦0, it is determined that the sampler has made a U-turn and the search is terminated. After all steps have been completed, the Metropolis method performs an acceptance / rejection step. In this step, the difference in Hamiltonian energy is calculated to obtain ΔH=H(q',p')-H(q,p), where (q,p) is the current sample and (q',p') is the proposed new sample. If the Hamiltonian energy of the new sample is lower or equal to that of the current sample (ΔH≦0), the new sample is accepted. Conversely, if the Hamiltonian energy of the new sample is higher than that of the current sample (ΔH>0), the new sample is accepted with probability min(1,exp(-ΔH)).

[0056] <Probabilistic Variational Inference> After performing MCMC, the MAP is inherited as the initial value, and better parameters are found by Stochastic Variational Inference (SVI). A delta function is used to approximate the posterior distribution. In this case, Adam (Adaptive Moment Estimation), a type of stochastic gradient descent method, is used to maximize the ELBO (Evidence Lower Bound). Adam is a type of Stochastic Gradient Descent (SGD) method that is widely used in machine learning. By individually adjusting the learning rate α for each parameter, parameters with steeper gradients receive smaller updates, and parameters with gentler gradients receive larger updates. This allows the problem to be scaled automatically. In addition, Adam takes into account the first and second moments of the past gradients, making it possible to suppress the oscillations that were a problem with SGD. At this time, the parameter w is updated as follows: v t =β 1 v t-1 +(1-β 1 ) G, s t =β 3 v t-1 +(1-β 2 )G 2 , w t+1 =w t -α v t / (√(s t +∈)), where α=0.001, β 1 = 0.9, β 2 =0.999,∈=10 -8 , G is the gradient of the loss function. The number of iterations is 20,000, and the step size is 0.0005.

[0057] (Sample analysis processing) The sample analysis process by the computer 2 will now be described with reference to FIG.

[0058] In step 201 of FIG. 3, the analysis results from the mass spectrometer 1 are obtained.

[0059] The number of components k contained in the sample is set in step 202. First, the number of components is set to 1.

[0060] In step 203, the inverse temperature is set. First, the inverse temperature is set to 0.25.

[0061] In step 204, MCMC is used to sample from the posterior distribution a specified number of times. Specifically, 1000×4 samples are taken.

[0062] In step 205, the MAP obtained by MCMC is set as the initial value for the next MCMC.

[0063] In step 206, the inverse temperature is increased by five times, and the processes in steps 203 to 206 are repeated.

[0064] In step 207, the MAP obtained by MCMC is set as the initial value of the SVI.

[0065] In step 208, the parameters are optimized by SVI.

[0066] In step 209, the MAP obtained by SVI is set as the initial value for the next MCMC.

[0067] In step 210, the number of components is incremented by 1, and the processes in steps 202 to 210 are repeated.

[0068] In step 211, the number of components k=1 to the number of components k max The maximum posterior probability of the model is compared to evaluate the model for each number of components.

[0069] In step 212, the model with the largest number of components that results in the maximum posterior probability is selected, and the optimal parameters at that time are obtained.

[0070] (Example) An example in which a sample was analyzed using the sample analysis method according to this embodiment will be described.

[0071] In the example, four components with altered base sequences were added to Fomivirsen (ID:A), and a total of five component spectra were generated by simulation, as shown in Table 1. The C→U mutation is called deamination, and is one of the impurities that can be generated during the synthesis process due to solvent conditions, heat stress, etc. [Table 1]

[0072] Data was generated by sampling 20,000 times (total amount of incoming ions) from a multinomial distribution based on the method. The combinations are shown in Table 2 below. [Table 2]

[0073] The results of estimating the number of components using the algorithm of the sample analysis method according to the present embodiment are shown in Table 3 below. [Table 3]

[0074] The true number of components could be estimated for Mixture No. 1, 2, 3, 5, 7, 8, 9, 11, 12, 13, 14, and 15, with a probability of 80%. Even for those for which the true number of components was not estimated, the difference from the true number of components was 1, and the difference in likelihood from the true number of components was small. In addition, the presence or absence of impurities (distinguishing between k=1 and others) could be determined 100% of the time.

[0075] The estimated monoisotopic masses when k is the true value are as shown in Table 4 below. [Table 4]

[0076] (Comparative Example) Deconvolution was performed on the same generated data as in the above example using the existing method UniDec (Comparative Example). As a result, the accuracy rate of the number of components was 13%.

[0077] Therefore, it was shown that the sample analysis method according to this embodiment can analyze samples with higher accuracy. In other words, the sample analysis method according to this embodiment uses NUTS, Simulated Annealing, and probabilistic variational inference to estimate parameters such as monoisotopic mass from measurement results, and can increase the accuracy rate of the number of components compared to existing methods. This is thought to be because creating a model for each number of components makes it possible to compare and evaluate and select models with multiple numbers of components. This makes it possible to suggest the presence of impurities in pharmaceuticals, which can contribute to the search for better synthesis conditions for medium to high molecular weight pharmaceuticals and quality assurance in factories.

[0078] (Effects of this embodiment) In this embodiment, the following effects can be obtained.

[0079] In this embodiment, as described above, the sample analysis method includes the steps of: assuming the number of components contained in the sample; acquiring a generated spectrum for the assumed number of components using the mass and charge of the components as parameters; and comparing the generated spectrum with the observed spectrum to acquire information on the number of components contained in the sample. In this way, a generated spectrum can be generated using the mass and charge of the components contained in the sample as parameters, independent of the observed spectrum, when the number of components is unknown. Therefore, a suitable generated spectrum can be extracted by comparing the generated generated spectrum with the observed spectrum. As a result, the values ​​of parameters including the number of components, mass, and charge when a suitable generated spectrum is generated can be obtained with high accuracy. As a result, the mass and charge of the components can be obtained with high accuracy by analyzing the observed spectrum obtained by measuring a sample with an unknown number of components, and the observed spectrum obtained by measurement can be analyzed with high accuracy.

[0080] In addition, in the above embodiment, the following additional effects can be obtained by configuring as follows.

[0081] That is, in the present embodiment, as described above, the step of acquiring the generated spectrum assumes a charge distribution from the component charges as parameters, and acquires the generated spectrum based on the assumed charge distribution. This makes it possible to easily acquire the generated spectrum based on the assumed charge distribution from the component charges.

[0082] In addition, in this embodiment, as described above, the step of acquiring a produced spectrum and the step of acquiring information on the number of components contained in the sample by comparing the produced spectrum with the observed spectrum are repeatedly performed while changing the number of components within a predetermined range. As a result, when acquiring a produced spectrum, the same process is repeated while changing the number of components, so that produced spectra with different numbers of components can be easily acquired.

[0083] In addition, in the present embodiment, as described above, the step of acquiring the generated spectrum acquires the generated spectrum by making the range of change in the mass of the first component as a parameter to be changed smaller for the second number of components larger than the first number of components based on the mass of the first component contained in the sample acquired by the first number of components than for the first number of components. This makes it possible to suppress the range of change in the parameter from becoming large when acquiring the generated spectrum while changing the number of components to increase, and therefore makes it possible to suppress an excessive increase in the number of trials even when the number of parameters increases as the number of components increases. As a result, it is possible to improve the efficiency of the analysis process.

[0084] In addition, in the present embodiment, as described above, the step of acquiring information regarding the number of components contained in the sample compares the generated spectrum and the observed spectrum for a plurality of different numbers of components, and acquires information regarding the number of components contained in the sample based on the generated spectrum that is approximate to the observed spectrum. This makes it easier to acquire information regarding the number of components in the sample based on the number of components when generating the generated spectrum that is approximate to the observed spectrum, which is the measurement result.

[0085] In addition, in the present embodiment, as described above, the step of acquiring information regarding the number of components contained in the sample compares the generated spectrum and the observed spectrum for a plurality of different numbers of components, and acquires information regarding the number of components contained in the sample based on the generated spectrum that is most similar to the observed spectrum. This makes it possible to more easily acquire information regarding the number of components of the sample based on the number of components when the generated spectrum that is most similar to the observed spectrum, which is the measurement result, is generated.

[0086] In addition, as described above, the present embodiment includes a step of acquiring an analyte amount including mass and charge of a component contained in a sample based on a generated spectrum approximating the observed spectrum, which is a measurement result, so that the mass and charge of the component, which are parameters when generating the generated spectrum approximating the observed spectrum, can be easily acquired as the mass and charge of the component of the analysis result.

[0087] In addition, in the present embodiment, as described above, the step of acquiring the product spectrum includes calculating a mass distribution of the mass of each of the components contained in the sample as a parameter to be changed, which is one or more components, and acquiring the product spectrum based on the calculated mass distribution of each of the components. This allows the mass distribution of the component to be calculated in consideration of the abundance ratio of the isotope from the monoisotopic mass of the component, and the product spectrum to be easily acquired based on the calculated mass distribution.

[0088] In the present embodiment, as described above, the step of acquiring the generated spectrum includes calculating the charge distribution of the charge as a parameter to be changed for each of the components contained in the sample, which may be one or more components, and acquiring the generated spectrum based on the calculated charge distribution of each component. This allows the charge distribution to be calculated from the charges of the components, and the generated spectrum to be easily acquired based on the calculated charge distribution.

[0089] In the present embodiment, as described above, the step of acquiring information regarding the number of components contained in the sample compares the multiple generated spectra with the observed spectrum using posterior probabilities to acquire information regarding the number of components contained in the sample. This allows the multiple generated spectra to be easily compared with the observed spectrum using posterior probabilities, making it possible to easily extract a suitable generated spectrum from the multiple generated spectra.

[0090] In addition, in this embodiment, as described above, the information on the number of components contained in the sample includes at least one of an optimal number of components and a suitable distribution of the number of components. As a result, by acquiring the optimal number of components as the information on the number of components, the number of components contained in the sample can be easily understood. In addition, by acquiring the suitable distribution of the number of components as the information on the number of components, the distribution of the number of components that may be contained in the sample can be easily understood.

[0091] [Variations] The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present invention is indicated by the claims, not by the description of the embodiments above, and further includes all modifications (variations) within the meaning and scope of the claims.

[0092] For example, in the above embodiment, an example of a configuration in which parameters are searched for by evaluating the posterior probability when acquiring a generated spectrum has been shown, but the present invention is not limited to this. For example, when acquiring a generated spectrum, parameters may be changed sequentially without evaluating the posterior probability.

[0093] In the above embodiment, the observed spectrum of the measurement result and the generated spectrum are compared using the posterior probability, but the present invention is not limited to this. For example, the observed spectrum of the measurement result and the generated spectrum may be compared by a matching process.

[0094] In the above embodiment, the number of components is sequentially changed within a set range to generate a plurality of generated spectra, and a generated spectrum that is similar to the observed spectrum is extracted from the generated plurality of generated spectra. However, the present invention is not limited to this. For example, the number of components may be changed until a generated spectrum that is similar to the observed spectrum is generated. In other words, the number of components may be changed and the process may be repeated until the degree of approximation of the generated spectrum converges.

[0095] In the above embodiment, an example of a configuration in which the number of components is changed from 1 to an increased number to generate a generated spectrum has been shown, but the present invention is not limited to this. For example, the number of components may be changed from 2 or more to an increased number to generate a generated spectrum. Alternatively, the number of components may be changed to a smaller number to generate a generated spectrum.

[0096] In the above embodiment, when the number of components is changed and incremented by 1, the range in which the mass of the component as a parameter is changed is narrowed based on the mass of the component analyzed using the previous number of components, but the present invention is not limited to this. For example, when the number of components is changed and incremented by 1, the range in which the mass of the component as a parameter is changed may be the same.

[0097] In the above embodiment, a configuration example was shown in which a plurality of generated spectra and an observed spectrum were compared to obtain information on the number of components contained in a sample and the masses and charges of the components for that number of components, but the present invention is not limited to this. For example, as long as information on at least the number of components contained in a sample is obtained, it is not necessary to obtain the masses and charges of the components.

[0098] [Aspects] It will be appreciated by those skilled in the art that the exemplary embodiments described above are examples of the following aspects.

[0099] (Item 1) receiving an observed spectrum obtained by measuring a sample with a mass spectrometer (1); Assuming the number of components contained in the sample; A step of obtaining a product spectrum from parameters related to masses of components and parameters related to charges of components for the assumed number of components; comparing the generated spectrum with the observed spectrum to obtain information regarding the number of components contained in the sample.

[0100] (Item 2) 2. The sample analysis method according to item 1, wherein the step of acquiring the product spectrum includes assuming a charge distribution from parameters related to the charges of the components, and acquiring the product spectrum based on the assumed charge distribution.

[0101] (Item 3) 3. The sample analysis method according to item 1 or 2, wherein the step of acquiring the product spectrum and the step of acquiring information regarding the number of components contained in the sample by comparing the product spectrum with the observed spectrum are repeatedly performed while changing the number of components within a predetermined range.

[0102] (Item 4) 4. The sample analysis method according to item 3, wherein the step of acquiring the product spectrum acquires the product spectrum by setting a range of change in the mass of the first component as a parameter to be changed in a second number of components greater than the first number of components based on the mass of the first component contained in the sample obtained from a first number of components, to be smaller than in the case of the first number of components.

[0103] (Item 5) 5. The sample analysis method according to item 3 or 4, wherein the step of acquiring information regarding the number of components contained in the sample comprises comparing the generated spectrum and the observed spectrum for a plurality of numbers of components that are changed, and acquiring information regarding the number of components contained in the sample based on the generated spectrum that is closest to the observed spectrum.

[0104] (Item 6) 6. The sample analysis method according to item 5, wherein the step of acquiring information regarding the number of components contained in the sample comprises comparing the generated spectrum and the observed spectrum for a plurality of numbers of components that are changed, and acquiring information regarding the number of components contained in the sample based on the generated spectrum that is closest to the observed spectrum.

[0105] (Item 7) 7. The method for analyzing a sample according to any one of claims 1 to 6, further comprising a step of acquiring an analyte amount including a mass and a charge of the component contained in the sample based on the generated spectrum that is approximate to the observed spectrum.

[0106] (Item 8) 8. The sample analysis method according to any one of items 1 to 7, wherein the step of acquiring the product spectrum comprises calculating a mass distribution of each of one or more components contained in the sample from a parameter related to the mass of the component, and acquiring the product spectrum based on the calculated mass distribution of each of the components.

[0107] (Item 9) 9. The sample analysis method according to any one of items 1 to 8, wherein the step of acquiring information regarding the number of components contained in the sample comprises comparing a plurality of the generated spectra with the observed spectrum based on a posterior probability to acquire information regarding the number of components contained in the sample.

[0108] (Item 10) 10. The sample analysis method according to any one of items 1 to 9, wherein the information on the number of components contained in the sample includes at least one of an optimal number of components and a distribution of suitable numbers of components. [Explanation of symbols]

[0109] 1 Mass spectrometer 2. Computer 100 Sample Analysis System

Claims

1. receiving an observed spectrum obtained by measuring a sample with a mass spectrometer; Assuming the number of components contained in the sample; A step of obtaining a product spectrum from parameters related to masses of components and parameters related to charges of components for the assumed number of components; comparing the generated spectrum with the observed spectrum to obtain information regarding the number of components contained in the sample.

2. 2. The sample analysis method according to claim 1, wherein the step of acquiring the product spectrum comprises assuming a charge distribution from parameters related to charges of components, and acquiring the product spectrum based on the assumed charge distribution.

3. 2. The sample analysis method according to claim 1, wherein the step of acquiring the product spectrum and the step of acquiring information regarding the number of components contained in the sample by comparing the product spectrum with the observed spectrum are repeatedly performed while changing the number of components within a predetermined range.

4. 4. The sample analysis method according to claim 3, wherein the step of acquiring the product spectrum acquires the product spectrum by making a range of change in the mass of the first component as a parameter to be changed smaller for a second number of components greater than the first number of components based on the mass of the first component contained in the sample obtained from a first number of components than for the first number of components.

5. 4. The sample analysis method according to claim 3, wherein the step of acquiring information regarding the number of components contained in the sample comprises comparing the generated spectrum and the observed spectrum for a plurality of different numbers of components, and acquiring information regarding the number of components contained in the sample based on the generated spectrum that is closest to the observed spectrum.

6. 6. The sample analysis method according to claim 5, wherein the step of acquiring information regarding the number of components contained in the sample comprises comparing the generated spectrum and the observed spectrum for a plurality of different numbers of components, and acquiring information regarding the number of components contained in the sample based on the generated spectrum that is closest to the observed spectrum.

7. The method of claim 1 , further comprising obtaining analytes including masses and charges of the components contained in the sample based on the product spectrum approximating the observed spectrum.

8. 2. The sample analysis method according to claim 1, wherein the step of acquiring the product spectrum comprises calculating a mass distribution of each of one or more components contained in the sample from parameters related to the mass of the component, and acquiring the product spectrum based on the calculated mass distribution of each of the components.

9. 2. The sample analysis method according to claim 1, wherein the step of acquiring information regarding the number of components contained in the sample comprises comparing a plurality of the generated spectra with the observed spectrum using posterior probability to acquire information regarding the number of components contained in the sample.

10. The sample analysis method according to claim 1 , wherein the information regarding the number of components contained in the sample includes at least one of an optimal number of components and a distribution of suitable numbers of components.

Images