A stability monitoring method and system based on mass spectrometry detection

Abstract

The present invention discloses a stability monitoring method and system based on mass spectrometry detection, belonging to the technical field of mass spectrometry stability monitoring. The specific steps of the monitoring method are as follows: P1, preprocessing mass spectrometry data and extracting data features; P2, displaying each mass spectrometry data through a three-dimensional visualization model; P3, capturing the dependency relationship between different mass spectrometry data and performing nonlinear dimensionality reduction; the present invention can automatically adjust algorithm parameters according to different characteristics of mass spectrometry data, improves the versatility and adaptability of the model on different mass spectrometry data sets, can better capture changes and trends in mass spectrometry data, enhances the stability and reliability of the model in capturing data dependency relationships, can identify abnormal data points at multiple levels, provide more accurate anomaly detection, improve the accuracy and robustness of anomaly detection, can efficiently process a large amount of mass spectrometry data, and is suitable for high-throughput mass spectrometry detection environments.

Description

Technical Field

[0001] The present invention relates to the technical field of mass spectrometry stability monitoring, and in particular to a stability monitoring method and system based on mass spectrometry detection. Background Art

[0002] As a powerful analytical tool, mass spectrometry detection technology is widely used in various scientific research and industrial production. It identifies and quantitatively analyzes compounds by measuring the mass-to-charge ratio of ionized samples. Mass spectrometry detection has the characteristics of high sensitivity, high resolution and high throughput, and can provide rich structural and quantitative information. Although mass spectrometry technology is relatively mature, the complexity and diversity of data are still a major challenge in practical applications. Mass spectrometry data not only contains rich information about the sample, but is also often accompanied by noise, drift and other uncertainties, which put higher requirements on the accuracy and stability of data analysis. Therefore, how to effectively process and analyze these data, especially the stability of monitoring data, has become an important research topic.

[0003] Existing stability monitoring methods and systems based on mass spectrometry detection cannot automatically adjust algorithm parameters according to the different characteristics of mass spectrometry data, which reduces the versatility and adaptability of the model on different mass spectrometry data sets, and the model has poor stability and reliability in capturing data dependency relationships; in addition, existing stability monitoring methods and systems based on mass spectrometry detection cannot identify abnormal data points at multiple levels, which reduces the accuracy and robustness of anomaly detection, and are not suitable for high-throughput mass spectrometry detection environments; therefore, we propose a stability monitoring method and system based on mass spectrometry detection. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose a stability monitoring method and system based on mass spectrometry detection.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A stability monitoring method based on mass spectrometry detection, the specific steps of the monitoring method are as follows:

[0007] P1. Preprocess mass spectrometry data and extract data features.

[0008] As a further solution of the present invention, the specific steps of extracting data features described in P1 are as follows:

[0009] Step 1: Obtain mass spectrometry data and construct a mass spectrometry image, where the horizontal axis of the mass spectrometry image is the mass-to-charge ratio and the vertical axis is the signal intensity. A Gaussian filter is then used to smooth the mass spectrometry data to remove noise, and a polynomial fit is used to correct the baseline.

[0010] Step 2: Construct an N×N window and place the initial position of the window at the upper left corner of the mass spectrometer image, i.e., coordinate Then, the window is slid horizontally or vertically point by point according to the needs until the original image ends;

[0011] Step 3: Each time the window is slid, the coordinates of the window center data point and the coordinates of the other neighboring data points are collected, and the difference between the window center data point and the other points in the window is calculated. Then, the calculated difference is binarized. If the value of the center point is greater than or equal to the neighboring point, the difference is 1, otherwise it is 0.

[0012] Step 4: Combine the binarization results of all neighborhood points into a binary number in clockwise or counterclockwise order to obtain the LBP feature value of each mass spectrum data point, then count the occurrence frequency of all LBP patterns in the image, construct the LBP histogram as the texture feature vector, and then construct the extracted LBP features into a data matrix, where each row represents a mass spectrum sample and each column represents a feature;

[0013] Step 5: Extract the characteristic peaks and their intensity values ​​in the mass spectrometry images of each group of experimental samples, arrange the characteristic intensity values ​​of each sample into a row to form a row of the response matrix, and then combine the characteristic arrangements of all samples together to form a complete response matrix, and then calculate the covariance matrix of the characteristic matrix and the response matrix;

[0014] Step 6: Obtain the eigenvector of the covariance matrix through singular value decomposition and use it as the weight vector. Then project the original data into the new latent variable space to obtain the reduced dimension feature matrix, and use the reduced dimension feature matrix as the data feature of this mass spectrometry data.

[0015] As a further solution of the present invention, the specific calculation formula of the LBP feature value in step 4 is as follows:

[0016] ;

[0017] In the formula, Represents the coordinates The LBP value of the central pixel at ; Represents the center point coordinates of the current N×N window; Represents the size of the current window; represents the binarization function; Representative Points Signal strength at Represents the first The coordinates of the field points; represents the weight of a binary number, where is the number of the neighborhood point.

[0018] P2, display each mass spectrometry data through a three-dimensional visualization model;

[0019] P3, capture the dependencies between different mass spectrometry data and perform nonlinear dimensionality reduction.

[0020] As a further solution of the present invention, the specific steps of capturing the dependency between different mass spectrometry data in P3 are as follows:

[0021] Step 1: Collect knowledge in the field of mass spectrometry through literature research, expert consultation, and historical experimental results, calculate the correlation coefficients between the features, and identify possible dependencies. Then use the chi-square test method to test the conditional independence between the features, cluster the features, obtain the potential relationship between each group of features, and then use the Granger causality test method to identify potential causal relationships.

[0022] Step 2: Take each data feature extracted from the mass spectrometry data as a node, preliminarily connect some significantly dependent features based on expert opinions and literature research results, and then establish initial edge connections based on the results of correlation coefficients and conditional independence tests, check the rationality of the initial probability model structure, and modify the initial probability model structure based on data statistical results and expert opinions;

[0023] Step 3: Calculate the BIC value of the current model structure, that is, the model structure score value. In the current model structure, select a pair of nodes that are not directly connected As a candidate edge, and Add a directed edge between or , calculate the BIC value of the new structure and compare it with the BIC value of the current structure;

[0024] Step 4: In the current model structure, select an existing directed edge as a candidate edge, and simulate deleting the directed edge, then calculate the BIC value of the new structure, and compare it with the BIC value of the current structure. In the current model structure, select an existing directed edge as a candidate edge, reverse the directed edge, calculate the BIC value of the new structure, and compare it with the BIC value of the current structure;

[0025] Step 5: Compare the score changes of adding edges, deleting edges, and reversing edges , select the operation that can maximize the reduction of BIC value, and then update the model structure according to the optimal operation, and initialize it according to the number of nodes and edges in the probability model Group individuals and randomly initialize the position of each group of individuals, where the position represents the model structure, and a binary matrix is ​​used to represent each network structure, and the matrix elements represent the existence or non-existence of edges;

[0026] Step 6: For each group of individuals, calculate the BIC value of the corresponding model structure as the brightness. The smaller the model BIC value, the higher the brightness value of the individual. Compare the brightness values ​​of each individual, calculate the distance between each group of individuals, and divide the individual into groups according to the individual. Towards a brighter individual The update rule of movement adjusts the position of each group of individuals, repeatedly calculates the brightness value of each group of individuals, and updates the individual position information until the preset maximum number of iterations is reached , after randomly perturbing the model structure corresponding to the updated position, the final probability model structure is obtained;

[0027] Step 7: In the final probability model, the dependency between features is confirmed through conditional independence test, and then the Markov Chain Monte Carlo method is used to perform probabilistic inference on each data feature, and the posterior probability of certain features is calculated for prediction or classification. The captured feature dependency is interpreted to obtain the interaction information between features.

[0028] As a further solution of the present invention, the specific calculation formula of the BIC value in step 3 is as follows:

[0029] ;

[0030] In the formula, Represents the model structure score; represents the observed data set, which contains all mass spectrometry data samples; represents the set of parameters of the probability model; Represents directed edge relationships between nodes; Representative data In the parameters and structure Likelihood under; Represents the number of parameters; Represents the number of samples in the dataset;

[0031] The specific calculation formula for updating the individual position in step 6 is as follows:

[0032] ;

[0033] ;

[0034] ;

[0035] In the formula, Representative individual and individuals The distance between Represents the dimension of the adjacency matrix; Represents mass spectrometry data Dimensions; Representative individual The corresponding adjacency matrix elements; Representative individual The corresponding adjacency matrix elements; Representative individual For individuals attractiveness; represents the maximum attraction constant; Represents the light intensity absorption coefficient, which is used to determine the rate at which attraction decays with distance; represents the base of natural logarithms; Represents the updated model structure; Representative individual Current model structure; Representative individual The current model structure.

[0036] P4, identify different stability states and analyze population structure in mass spectrometry data;

[0037] P5. Model the clustering results to handle the multimodal nature of the data;

[0038] P6. Real-time stability monitoring based on mass spectrometry characteristics and clustering results.

[0039] A stability monitoring system based on mass spectrometry detection, comprising an acquisition preprocessing module, an extraction dimension reduction module, a dependency capture module, a structure optimization module, an anomaly recognition module, a report generation module, a capture detection module, an adaptive adjustment module and a visualization module;

[0040] The acquisition and processing module is used to acquire raw mass spectrometry data from the mass spectrometer and perform cleaning and standardization;

[0041] The extraction and dimensionality reduction module is used to extract texture features of mass spectrometry data, capture local patterns in the mass spectrogram, and reduce the dimensionality of the extracted features to reduce redundant information;

[0042] The dependency capture module is used to construct a probability model of mass spectrometry data and capture the dependency relationship between different features;

[0043] The structure optimization module is used to optimize the probability model structure and evaluate the optimized model structure;

[0044] The anomaly identification module is used to detect anomalies in mass spectrometry data and identify potential quality problems.

[0045] As a further solution of the present invention, the specific steps of the abnormality identification module to identify potential quality problems are as follows:

[0046] Step I: Obtaining the preprocessed mass spectrometry dataset ,in Indicates mass spectral data points, including multiple features, then select multiple initial cluster centers, calculate the distance from each data point to the cluster center, assign each group of mass spectral data to the nearest cluster center, and update the center of each cluster to the average value of all points in the cluster;

[0047] Step II: Repeatedly search for cluster centers and assign mass spectrometry data points until the cluster centers converge to the preset threshold. Based on the clustering results, identify points that do not belong to any major cluster or are at the edge of the cluster and mark them as outliers.

[0048] Step III: Calculate and record the similarity between the unlabeled mass spectrometry data points and the abnormal data points to construct a standard Laplace matrix, then perform eigenvalue decomposition on the Laplace matrix to obtain multiple sets of eigenvectors corresponding to the minimum eigenvalues ​​to form a feature matrix, and then set the neighborhood radius of each data point , and the minimum number of neighbors required for a mass spectral data point to become a core point ;

[0049] Step IV: Calculate the distance between each pair of data points in the feature matrix and calculate the neighborhood radius of each data point The number of neighbors in, if the number of neighbors is not less than , then this point is the core point. Starting from a core point, all points within the neighborhood radius The points in the cluster are added to the current cluster. If the newly added points are also core points, the cluster will continue to expand, repeatedly selecting core points and creating new clusters until there are no new points to add.

[0050] Step V: Detect each group of mass spectrometry data points after clustering. If a point is neither a core point nor a neighbor of any core point, the point is marked as an outlier. Then, the conditional probability table of each remaining node, the joint probability of each data point in the probability model and the outlier probability are calculated through the constructed probability model. If the outlier probability is higher than the outlier threshold, the point is identified as an outlier.

[0051] Step VI: Explain each group of abnormal points detected, and then use the report generation module to visualize the detection results of each group of mass spectrometry data in the form of charts according to user needs, and inform the user of stability anomalies through text messages, emails and alarms.

[0052] The cluster analysis module is used to detect and analyze mass spectrometry data and identify potential patterns and structures in the data;

[0053] The report generation module is used to automatically generate a mass spectrometry detection result report, providing detailed analysis results and suggestions;

[0054] The capture detection module is used to identify community structures and multi-peak characteristics in capture mass spectrometry data.

[0055] As a further solution of the present invention, the capture detection module identifies the community structure and multi-peak characteristics in the capture mass spectrum data in the following specific steps:

[0056] Step 1: The capture detection module initializes a set of empty network graphs, represents the mass spectrometry data points as nodes in the network graph, and constructs edges between nodes based on the similarity or distance between data points, and calculates the distance between all node pairs through the edges. The number of shortest paths, i.e. the edge betweenness , and remove the edge with the largest betweenness;

[0057] Step 2: After the removal is completed, recalculate the edge betweenness of each edge in the network graph, remove the edge with the largest edge betweenness, repeat the calculation of the edge betweenness of each edge, and remove the largest edge betweenness until the network graph is split into multiple communities, and calculate the internal characteristics of each community;

[0058] Step 3: Build a Gaussian mixture model, define multiple Gaussian components, each component represents a Gaussian distribution, and then use the expectation maximization algorithm to calculate the posterior probability that the data point belongs to each Gaussian component, and update the Gaussian mixture model parameters according to the posterior probability;

[0059] Step 4: Calculate the probability distribution of each mass spectrum data point through the updated Gaussian mixture model, analyze each Gaussian component in the Gaussian mixture model, identify the multi-peak characteristics of the mass spectrum data, and then determine the different peak positions and distribution forms of the mass spectrum data based on the mean and covariance of each Gaussian component.

[0060] The adaptive adjustment module is used to dynamically adjust the model parameters and update the detection algorithm in real time;

[0061] The visualization module is used to construct a virtual reality scene and display a mass spectrum.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. The stability monitoring method based on mass spectrometry detection collects knowledge in the field of mass spectrometry detection through literature research, expert consultation and historical experimental results, calculates the correlation coefficient between each feature, and identifies possible dependencies. The chi-square test method is then used to test the conditional independence between the features, cluster the features, and obtain the potential relationship between each group of features. The Granger causality test method is then used to identify potential causal relationships. Each data feature extracted from the mass spectrometry data is taken as a node, and some significantly dependent features are preliminarily connected. Based on the results of the correlation coefficient and conditional independence test, the initial edge connection is established to check the rationality of the initial probability model structure. Based on the data statistical results and expert opinions, the initial probability model structure is corrected, and the BIC value of the current model structure, that is, the model structure score value, is calculated. In the current model structure, a directed edge is randomly added, and the BIC value of the new structure is calculated and compared with the BIC value of the current structure. In the current model structure, edge addition, edge deletion and edge reversal operations are performed, and the score changes of each operation are compared. , select the operation that can maximize the reduction of BIC value, and then update the model structure according to the optimal operation. Initialize multiple groups of individuals according to the number of nodes and edges in the probability model, and randomly initialize the position of each group of individuals, where the position represents the model structure, and use a binary matrix to represent each network structure. The matrix elements represent the existence or non-existence of edges. For each group of individuals, calculate the BIC value of the corresponding model structure as the brightness, where the smaller the model BIC value, the higher the individual brightness value. Compare the brightness values ​​of each individual, adjust the position of each group of individuals, repeatedly calculate the brightness values ​​of each group of individuals, and update the individual position information until the preset maximum number of iterations is reached, and update the model corresponding to the updated position. After randomly perturbing the mass spectrometry structure, the final probability model structure is obtained. In the final probability model, the dependency relationship between features is confirmed by the conditional independence test. Then, the Markov chain Monte Carlo method is used to perform probabilistic inference on each data feature, and the posterior probability of certain features is calculated for prediction or classification. The captured feature dependency relationship is interpreted to obtain the interaction information between features. The algorithm parameters can be automatically adjusted according to the different characteristics of mass spectrometry data, which improves the versatility and adaptability of the model on different mass spectrometry data sets, can better capture the changes and trends in mass spectrometry data, and enhance the stability and reliability of the model in capturing data dependency relationships.

[0064] 2. The stability monitoring system based on mass spectrometry detection obtains the preprocessed mass spectrometry data set, then selects multiple initial cluster centers, calculates the distance from each data point to the cluster center, and assigns each group of mass spectrometry data to the nearest cluster center, and updates the center of each cluster to the average value of all points in the cluster, repeatedly searches for cluster centers, and assigns mass spectrometry data points until the cluster centers converge to a preset threshold, identifies points that do not belong to any major cluster or are at the edge of the cluster, and marks them as abnormal points, calculates the similarity between each data point to construct a standard Laplace matrix, and then performs eigenvalue decomposition on the Laplace matrix to obtain eigenvectors corresponding to multiple groups of minimum eigenvalues ​​to form a feature matrix, and then sets the neighborhood radius of each data point , and the minimum number of neighbors required for a mass spectral data point to become a core point , find the core points in the feature matrix, start from a core point, and convert all the points in the neighborhood radius The points in the cluster are added to the current cluster. If the newly added point is also a core point, it will continue to expand, repeatedly select core points and create new clusters until there are no new points to add. After the clustering is completed, each group of mass spectrometry data points is detected. If a point is neither a core point nor a neighbor of any core point, the point is marked as an abnormal point. Then, the conditional probability table of each remaining node, as well as the joint probability and abnormal probability of each data point in the probability model are calculated through the constructed probability model. If the abnormal probability is higher than the abnormal threshold, the point is identified as an abnormal point, and each group of detected abnormal points is explained. Then, the report generation module visualizes the detection results of each group of mass spectrometry data in the form of charts according to user needs, and informs the user of stability anomalies through SMS, email and alarm. It can identify abnormal data points at multiple levels, provide more accurate anomaly detection, improve the accuracy and robustness of anomaly detection, and efficiently process a large amount of mass spectrometry data, which is suitable for high-throughput mass spectrometry detection environments. . BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0066] Figure 1 A flowchart of a stability monitoring method based on mass spectrometry detection proposed by the present invention;

[0067] Figure 2 This is a system block diagram of a stability monitoring system based on mass spectrometry detection proposed by the present invention. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Example 1

[0069] Reference Figure 1 This embodiment discloses a stability monitoring method based on mass spectrometry detection, and the specific steps of the monitoring method are as follows:

[0070] Preprocess mass spectrometry data and extract data features.

[0071] Specifically, mass spectrometry data is acquired and a mass spectrometry image is constructed, wherein the horizontal axis of the mass spectrometry image is the mass-to-charge ratio and the vertical axis is the signal intensity. A Gaussian filter is then used to smooth the mass spectrometry data to remove noise, and a polynomial fitting is used to correct the baseline. An N×N window is constructed, and the initial position of the window is placed at the upper left corner of the mass spectrometry image, i.e., the coordinate At the point, the window is slid horizontally or vertically point by point according to the needs until it slides to the end of the pristine image. Each time the window is slid, the coordinates of the window center data point and the coordinates of the other neighborhood data points are collected, and the difference between the window center data point and the other points in the window is calculated. Then, according to the calculated difference, binarization is performed. If the value of the center point is greater than or equal to the neighborhood point, the difference is 1, otherwise it is 0. The binarization results of all neighborhood points are combined into a binary number in clockwise or counterclockwise order to obtain the LBP eigenvalue of each mass spectrum data point. Then, the occurrence frequency of all LBP patterns in the image is counted to construct an LBP histogram as the texture feature vector. The extracted LBP features are then constructed into a data matrix, where each row represents a mass spectrometry sample and each column represents a feature. The characteristic peaks and their intensity values ​​in the mass spectrometry images of each group of experimental samples are extracted, and the characteristic intensity values ​​of each sample are arranged in a row to form a row of the response matrix. The features of all samples are then arranged and combined together to form a complete response matrix. The covariance matrix of the feature matrix and the response matrix is ​​then calculated, and the eigenvector of the covariance matrix is ​​obtained through singular value decomposition, and is used as a weight vector. The original data is then projected into the new latent variable space to obtain the feature matrix after dimensionality reduction, and the feature matrix after dimensionality reduction is used as the data feature of this mass spectrometry data.

[0072] In this embodiment, the specific calculation formula of the LBP characteristic value is as follows:

[0073] ;

[0074] In the formula, Represents the coordinates The LBP value of the central pixel at ; Represents the center point coordinates of the current N×N window; Represents the size of the current window; represents the binarization function; Representative Points Signal strength at Represents the first The coordinates of the field points; represents the weight of a binary number, where is the number of the neighborhood point.

[0075] In addition, it needs to be further explained that the response matrix needs to select a group of representative samples according to the purpose of the study, and set experimental conditions, such as temperature, pressure, reagent concentration, etc., to ensure that the response data obtained under different conditions are comparable, and then adjust the parameters of the mass spectrometer, such as ion source type, mass-to-charge ratio range, scanning speed, etc., to obtain high-quality data, and perform mass spectrometry on each sample, record the signal intensity at each mass-to-charge ratio, and then pre-process the acquired mass spectrometry data, extract the characteristic peaks and their intensity values ​​in the mass spectrum, arrange the characteristic intensity values ​​of each sample into a row to form a row of the response matrix, and then combine the characteristics of all samples together to form a complete response matrix.

[0076] The mass spectrometry data are displayed through a three-dimensional visualization model.

[0077] Capture dependencies between different mass spectrometry data and perform nonlinear dimensionality reduction.

[0078] Specifically, we collected knowledge in the field of mass spectrometry detection through literature research, expert consultation, and historical experimental results, calculated the correlation coefficients between the features, and identified possible dependencies. We then used the chi-square test method to test the conditional independence between the features, clustered the features, and obtained the potential relationships between the groups of features. We then used the Granger causality test method to identify potential causal relationships. We took each data feature extracted from the mass spectrometry data as a node, and preliminarily connected some significantly dependent features based on expert opinions and the results of literature research. We then established initial edge connections based on the results of the correlation coefficient and conditional independence tests, checked the rationality of the initial probability model structure, modified the initial probability model structure based on data statistical results and expert opinions, and calculated the BIC value of the current model structure. In the current model structure, we selected a pair of nodes that were not directly connected. As a candidate edge, and Add a directed edge between or , calculate the BIC value of the new structure and compare it with the BIC value of the current structure. In the current model structure, select an existing directed edge as a candidate edge and simulate deleting the directed edge. Then calculate the BIC value of the new structure and compare it with the BIC value of the current structure. In the current model structure, select an existing directed edge as a candidate edge, reverse the directed edge, calculate the BIC value of the new structure, that is, the model structure score value, and compare it with the BIC value of the current structure. Compare the score changes of adding edges, deleting edges, and reversing edges , select the operation that can maximize the reduction of BIC value, and then update the model structure according to the optimal operation, and initialize it according to the number of nodes and edges in the probability model Group individuals and randomly initialize the position of each group of individuals, where the position represents the model structure. A binary matrix is ​​used to represent each network structure, and the matrix elements represent the existence or non-existence of edges. For each group of individuals, the BIC value of the corresponding model structure is calculated as the brightness. The smaller the model BIC value, the higher the individual brightness value. The brightness values ​​of each individual are compared, and then the distance between each group of individuals is calculated, and the distance between the individuals is calculated according to the individual. Towards a brighter individual The update rule of movement adjusts the position of each group of individuals, repeatedly calculates the brightness value of each group of individuals, and updates the individual position information until the preset maximum number of iterations is reached After randomly perturbing the model structure corresponding to the updated position, the final probability model structure is obtained. The dependency between features is confirmed in the final probability model through the conditional independence test. Then, the Markov chain Monte Carlo method is used to perform probabilistic inference on each data feature, and the posterior probability of certain features is calculated for prediction or classification. The captured feature dependency is interpreted to obtain the interaction information between features.

[0079] It should be further explained that the specific calculation formula of BIC value is as follows:

[0080] ;

[0081] In the formula, Represents the model structure score; represents the observed data set, which contains all mass spectrometry data samples; represents the set of parameters of the probability model; Represents directed edge relationships between nodes; Representative data In the parameters and structure Likelihood under; Represents the number of parameters; Represents the number of samples in the dataset;

[0082] The specific calculation formula for individual position update is as follows:

[0083] ;

[0084] ;

[0085] ;

[0086] In the formula, Representative individual and individuals The distance between Represents the dimension of the adjacency matrix; Represents mass spectrometry data Dimensions; Representative individual The corresponding adjacency matrix elements; Representative individual The corresponding adjacency matrix elements; Representative individual For individuals attractiveness; represents the maximum attraction constant; Represents the light intensity absorption coefficient, which is used to determine the rate at which attraction decays with distance; represents the base of natural logarithms; Represents the updated model structure; Representative individual Current model structure; Representative individual The current model structure.

[0087] Identify distinct stability states and analyze population structure in mass spectrometry data.

[0088] The clustering results are modeled to handle the multimodal nature of the data.

[0089] Real-time stability monitoring was performed based on mass spectrometry features and clustering results. Example 2

[0090] Reference Figure 2 , the present embodiment discloses a stability monitoring system based on mass spectrometry detection, including an acquisition preprocessing module, an extraction dimension reduction module, a dependency capture module, a structure optimization module, an anomaly recognition module, a report generation module, a capture detection module, an adaptive adjustment module and a visualization module;

[0091] The acquisition and processing module is used to collect raw mass spectrometry data from the mass spectrometer and perform cleaning and standardization; the extraction and dimensionality reduction module is used to extract texture features of mass spectrometry data, capture local patterns in the mass spectrum, and reduce the dimensionality of the extracted features to reduce redundant information; the dependency capture module is used to construct a probabilistic model of mass spectrometry data and capture the dependency relationship between different features; the structure optimization module is used to optimize the probabilistic model structure and evaluate the optimized model structure; the anomaly recognition module is used to detect anomalies in mass spectrometry data and identify potential quality problems.

[0092] Specifically, obtain the preprocessed mass spectrometry data set ,in Indicates mass spectral data points, including multiple features, then select multiple initial cluster centers, calculate the distance from each data point to the cluster center, assign each group of mass spectral data to the nearest cluster center, and update the center of each cluster to the average value of all points in the cluster, repeatedly search for cluster centers, and assign mass spectral data points until the cluster centers converge to a preset threshold, identify points that do not belong to any major cluster or are at the edge of the cluster according to the clustering results, and mark them as abnormal points, calculate and record the similarity between unlabeled mass spectral data points and abnormal data points to construct a standard Laplace matrix, then perform eigenvalue decomposition on the Laplace matrix, obtain eigenvectors corresponding to multiple groups of minimum eigenvalues, form a feature matrix, and then set the neighborhood radius of each data point , and the minimum number of neighbors required for a mass spectral data point to become a core point , calculate the distance between each pair of data points in the feature matrix, and calculate the neighborhood radius of each data point The number of neighbors in, if the number of neighbors is not less than , then this point is the core point. Starting from a core point, all points within the neighborhood radius The points inside are added to the current cluster. If the newly added point is also a core point, the cluster will continue to expand, repeatedly select core points and create new clusters until there are no new points to add. After the clustering is completed, each group of mass spectrometry data points will be detected. If a point is neither a core point nor a neighbor of any core point, the point will be marked as an anomaly. The conditional probability table of each remaining node, the joint probability of each data point in the probability model and the anomaly probability are calculated through the constructed probability model. If the anomaly probability is higher than the anomaly threshold, the point is identified as an anomaly, and the detected groups of anomalies are explained. Then, the report generation module is used to visualize the detection results of each group of mass spectrometry data in the form of charts according to user needs, and the user is informed of stability anomalies through text messages, emails and alarms.

[0093] The cluster analysis module is used to detect and analyze mass spectrometry data and identify potential patterns and structures in the data; the report generation module is used to automatically generate mass spectrometry detection result reports, providing detailed analysis results and suggestions; the capture detection module is used to identify and capture community structures and multi-peak characteristics in mass spectrometry data.

[0094] Specifically, the capture detection module initializes a set of empty network graphs, represents the mass spectrometry data points as nodes in the network graph, and constructs edges between nodes based on the similarity or distance between data points, and calculates the distance between all node pairs through the edges. The number of shortest paths, i.e. the edge betweenness , and remove the edge with the largest edge betweenness. After the removal is completed, recalculate the edge betweenness of each edge in the network graph and remove the edge with the largest edge betweenness. Repeat the calculation of the edge betweenness of each edge and remove the maximum edge betweenness until the network graph is split into multiple communities, calculate the internal characteristics of each community, construct a Gaussian mixture model, define multiple Gaussian components, each component represents a Gaussian distribution, and then use the expectation-maximization algorithm to calculate the posterior probability that the data point belongs to each Gaussian component, and update the Gaussian mixture model parameters according to the posterior probability. The probability distribution of each mass spectrometry data point is calculated through the updated Gaussian mixture model, and then each Gaussian component in the Gaussian mixture model is analyzed to identify the multi-peak characteristics of the mass spectrometry data. Then, according to the mean and covariance of each Gaussian component, the different peak positions and distribution forms of the mass spectrometry data are determined.

[0095] The adaptive adjustment module is used to dynamically adjust the model parameters and update the detection algorithm in real time; the visualization module is used to build a virtual reality scene and display the mass spectrum.

Claims

1. A stability monitoring method based on mass spectrometry detection, characterized in that: The specific steps of the monitoring method are as follows: P1, preprocess mass spectrometry data and extract data features; P2, display each mass spectrometry data through a three-dimensional visualization model; P3, capturing the dependencies between different mass spectrometry data and performing nonlinear dimensionality reduction; P4, identify different stability states and analyze population structure in mass spectrometry data; P5. Model the clustering results to handle the multimodal nature of the data; P6, real-time stability monitoring based on mass spectrometry characteristics and clustering results; The specific steps of extracting data features are as follows: Step 1: Obtain mass spectrometry data and construct a mass spectrometry image, where the horizontal axis of the mass spectrometry image is the mass-to-charge ratio and the vertical axis is the signal intensity. A Gaussian filter is then used to smooth the mass spectrometry data to remove noise, and a polynomial fit is used to correct the baseline. Step 2: Construct an N×N window and place the initial position of the window at the upper left corner of the mass spectrometer image, i.e., coordinate Then, the window is slid horizontally or vertically point by point according to the needs until the original image ends; Step 3: Each time the window is slid, the coordinates of the window center data point and the coordinates of the other neighboring data points are collected, and the difference between the window center data point and the other points in the window is calculated. Then, the calculated difference is binarized. If the value of the center point is greater than or equal to the neighboring point, the difference is 1, otherwise it is 0. Step 4: Combine the binarization results of all neighborhood points into a binary number in clockwise or counterclockwise order to obtain the LBP feature value of each mass spectrum data point, then count the occurrence frequency of all LBP patterns in the image, construct the LBP histogram as the texture feature vector, and then construct the extracted LBP features into a data matrix, where each row represents a mass spectrum sample and each column represents a feature; Step 5: Extract the characteristic peaks and their intensity values ​​in the mass spectrometry images of each group of experimental samples, arrange the characteristic intensity values ​​of each sample into a row to form a row of the response matrix, and then combine the characteristic arrangements of all samples together to form a complete response matrix, and then calculate the covariance matrix of the characteristic matrix and the response matrix; Step 6: Obtain the eigenvector of the covariance matrix through singular value decomposition and use it as the weight vector. Then project the original data into the new latent variable space to obtain the feature matrix after dimensionality reduction, and use the feature matrix after dimensionality reduction as the data feature of this mass spectrometry data. The specific steps of capturing the dependency between different mass spectrometry data are as follows: Step 1: Collect knowledge in the field of mass spectrometry through literature research, expert consultation, and historical experimental results, calculate the correlation coefficients between the features, and identify possible dependencies. Then use the chi-square test method to test the conditional independence between the features, cluster the features, obtain the potential relationship between each group of features, and then use the Granger causality test method to identify potential causal relationships. Step 2: Take each data feature extracted from the mass spectrometry data as a node, preliminarily connect some significantly dependent features based on expert opinions and literature research results, and then establish initial edge connections based on the results of correlation coefficients and conditional independence tests, check the rationality of the initial probability model structure, and modify the initial probability model structure based on data statistical results and expert opinions; Step 3: Calculate the BIC value of the current model structure, that is, the model structure score value. In the current model structure, select a pair of nodes that are not directly connected As a candidate edge, and Add a directed edge between or , calculate the BIC value of the new structure and compare it with the BIC value of the current structure; Step 4: In the current model structure, select an existing directed edge as a candidate edge, and simulate deleting the directed edge, then calculate the BIC value of the new structure, and compare it with the BIC value of the current structure. In the current model structure, select an existing directed edge as a candidate edge, reverse the directed edge, calculate the BIC value of the new structure, and compare it with the BIC value of the current structure; Step 5: Compare the score changes of adding edges, deleting edges, and reversing edges , select the operation that can maximize the reduction of BIC value, and then update the model structure according to the optimal operation, and initialize it according to the number of nodes and edges in the probability model Group individuals and randomly initialize the positions of each group of individuals, where the positions represent the model structure, and a binary matrix is ​​used to represent each network structure, where the matrix elements represent the existence or non-existence of an edge; Step 6: For each group of individuals, calculate the BIC value of the corresponding model structure as the brightness. The smaller the model BIC value, the higher the brightness value of the individual. Compare the brightness values ​​of each individual, calculate the distance between each group of individuals, and divide the individual into groups according to the individual. Towards a brighter individual The update rule of movement adjusts the position of each group of individuals, repeatedly calculates the brightness value of each group of individuals, and updates the individual position information until the preset maximum number of iterations is reached , after randomly perturbing the model structure corresponding to the updated position, the final probability model structure is obtained; Step 7: In the final probability model, the dependency between features is confirmed by conditional independence test, and then the Markov chain Monte Carlo method is used to perform probabilistic reasoning on each data feature, and the posterior probability of certain features is calculated for prediction or classification, and the captured feature dependency is interpreted to obtain the interaction information between features; The specific steps of identifying different stability states are as follows: Step I: Obtaining the preprocessed mass spectrometry dataset ,in Indicates mass spectral data points, including multiple features, then select multiple initial cluster centers, calculate the distance from each data point to the cluster center, assign each group of mass spectral data to the nearest cluster center, and update the center of each cluster to the average value of all points in the cluster; Step II: Repeatedly search for cluster centers and assign mass spectrometry data points until the cluster centers converge to the preset threshold. Based on the clustering results, identify points that do not belong to any major cluster or are at the edge of the cluster and mark them as outliers. Step III: Calculate and record the similarity between the unlabeled mass spectrometry data points and the abnormal data points to construct a standard Laplace matrix, then perform eigenvalue decomposition on the Laplace matrix to obtain multiple sets of eigenvectors corresponding to the minimum eigenvalues ​​to form a feature matrix, and then set the neighborhood radius of each data point , and the minimum number of neighbors required for a mass spectral data point to become a core point ; Step IV: Calculate the distance between each pair of data points in the feature matrix and calculate the neighborhood radius of each data point The number of neighbors in, if the number of neighbors is not less than , then this point is the core point. Starting from a core point, all points within the neighborhood radius The points in the cluster are added to the current cluster. If the newly added points are also core points, the cluster will continue to expand, repeatedly selecting core points and creating new clusters until there are no new points to add. Step V: Detect each group of mass spectrometry data points after clustering. If a point is neither a core point nor a neighbor of any core point, the point is marked as an outlier. Then, the conditional probability table of each remaining node, the joint probability of each data point in the probability model and the outlier probability are calculated through the constructed probability model. If the outlier probability is higher than the outlier threshold, the point is identified as an outlier. Step VI: Explain each group of abnormal points detected, and then use the report generation module to visualize the detection results of each group of mass spectrometry data in the form of charts according to user needs, and inform the user of stability abnormalities through SMS, email and alarm; The specific steps of modeling the clustering results to handle the multi-modal characteristics of the data are as follows: Step 1: The capture detection module initializes a set of empty network graphs, represents the mass spectrometry data points as nodes in the network graph, and constructs edges between nodes based on the similarity or distance between data points, and calculates the distance between all node pairs through the edges. The number of shortest paths, i.e. the edge betweenness , and remove the edge with the largest betweenness; Step 2: After the removal is completed, recalculate the edge betweenness of each edge in the network graph, remove the edge with the largest edge betweenness, repeat the calculation of the edge betweenness of each edge, and remove the largest edge betweenness until the network graph is split into multiple communities, and calculate the internal characteristics of each community; Step 3: Build a Gaussian mixture model, define multiple Gaussian components, each component represents a Gaussian distribution, and then use the expectation maximization algorithm to calculate the posterior probability that the data point belongs to each Gaussian component, and update the Gaussian mixture model parameters according to the posterior probability; Step 4: Calculate the probability distribution of each mass spectrum data point through the updated Gaussian mixture model, analyze each Gaussian component in the Gaussian mixture model, identify the multi-peak characteristics of the mass spectrum data, and then determine the different peak positions and distribution forms of the mass spectrum data based on the mean and covariance of each Gaussian component.

2. The stability monitoring method based on mass spectrometry detection according to claim 1, characterized in that: The specific calculation formula of the LBP feature value in step 4 is as follows: ； In the formula, Represents the coordinates The LBP value of the central pixel at ; Represents the center point coordinates of the current N×N window; Represents the size of the current window; represents the binarization function; Representative Points Signal strength at Represents the first The coordinates of the field points; represents the weight of a binary number, where is the number of the neighborhood point.

3. The stability monitoring method based on mass spectrometry detection according to claim 1, characterized in that: The specific calculation formula of the BIC value in step 3 is as follows: ； In the formula, Represents the model structure score; represents the observed data set, which contains all mass spectrometry data samples; represents the set of parameters of the probability model; Represents directed edge relationships between nodes; Representative data In the parameters and structure Likelihood under; Represents the number of parameters; Represents the number of samples in the data set; The specific calculation formula for updating the individual position in step 6 is as follows: ； ； ； In the formula, Representing an individual and individuals The distance between Represents the dimension of the adjacency matrix; Represents mass spectrometry data Dimensions; Representing an individual The corresponding adjacency matrix elements; Representing an individual The corresponding adjacency matrix elements; Representing an individual For individuals attractiveness; represents the maximum attraction constant; Represents the light intensity absorption coefficient, which is used to determine the rate at which attraction decays with distance; represents the base of natural logarithms; Represents the updated model structure; Representing an individual Current model structure; Representing an individual The current model structure.

4. A stability monitoring system based on mass spectrometry detection, used to implement the monitoring method according to any one of claims 1 to 3, characterized in that: It includes acquisition preprocessing module, extraction and dimensionality reduction module, dependency capture module, structure optimization module, anomaly recognition module, report generation module, capture detection module, adaptive adjustment module and visualization module; The acquisition and processing module is used to acquire raw mass spectrometry data from the mass spectrometer and perform cleaning and standardization; The extraction and dimensionality reduction module is used to extract texture features of mass spectrometry data, capture local patterns in the mass spectrogram, and reduce the dimensionality of the extracted features to reduce redundant information; The dependency capture module is used to construct a probability model of mass spectrometry data and capture the dependency relationship between different features; The structure optimization module is used to optimize the probability model structure and evaluate the optimized model structure; The anomaly identification module is used to detect anomalies in mass spectrometry data and identify potential quality problems; The cluster analysis module is used to detect and analyze mass spectrometry data and identify potential patterns and structures in the data; The report generation module is used to automatically generate a mass spectrometry detection result report, providing detailed analysis results and suggestions; The capture detection module is used to identify community structures and multi-peak characteristics in capture mass spectrometry data; The adaptive adjustment module is used to dynamically adjust the model parameters and update the detection algorithm in real time; The visualization module is used to construct a virtual reality scene and display a mass spectrum.

5. A stability monitoring system based on mass spectrometry detection according to claim 4, characterized in that: The specific steps of the abnormality identification module to identify potential quality problems are as follows: Step I: Obtaining the preprocessed mass spectrometry dataset ,in Indicates mass spectral data points, including multiple features, then select multiple initial cluster centers, calculate the distance from each data point to the cluster center, assign each group of mass spectral data to the nearest cluster center, and update the center of each cluster to the average value of all points in the cluster; Step II: Repeatedly search for cluster centers and assign mass spectrometry data points until the cluster centers converge to the preset threshold. Based on the clustering results, identify points that do not belong to any major cluster or are at the edge of the cluster and mark them as outliers. Step III: Calculate and record the similarity between the unlabeled mass spectrometry data points and the abnormal data points to construct a standard Laplace matrix, then perform eigenvalue decomposition on the Laplace matrix to obtain multiple sets of eigenvectors corresponding to the minimum eigenvalues ​​to form a feature matrix, and then set the neighborhood radius of each data point , and the minimum number of neighbors required for a mass spectral data point to become a core point ; Step IV: Calculate the distance between each pair of data points in the feature matrix and calculate the neighborhood radius of each data point The number of neighbors in, if the number of neighbors is not less than , then this point is the core point. Starting from a core point, all points within the neighborhood radius The points in the cluster are added to the current cluster. If the newly added points are also core points, the cluster will continue to expand, repeatedly selecting core points and creating new clusters until there are no new points to add. Step V: Detect each group of mass spectrometry data points after clustering. If a point is neither a core point nor a neighbor of any core point, the point is marked as an outlier. Then, the conditional probability table of each remaining node, the joint probability of each data point in the probability model and the outlier probability are calculated through the constructed probability model. If the outlier probability is higher than the outlier threshold, the point is identified as an outlier. Step VI: Explain each group of abnormal points detected, and then use the report generation module to visualize the detection results of each group of mass spectrometry data in the form of charts according to user needs, and inform the user of stability anomalies through text messages, emails and alarms.

6. A stability monitoring system based on mass spectrometry detection according to claim 4, characterized in that: The specific steps of the capture detection module to identify the community structure and multi-peak characteristics in the capture mass spectrum data are as follows: Step 1: The capture detection module initializes a set of empty network graphs, represents the mass spectrometry data points as nodes in the network graph, and constructs edges between nodes based on the similarity or distance between data points, and calculates the distance between all node pairs through the edges. The number of shortest paths, i.e. the edge betweenness , and remove the edge with the largest betweenness; Step 2: After the removal is completed, recalculate the edge betweenness of each edge in the network graph, remove the edge with the largest edge betweenness, repeat the calculation of the edge betweenness of each edge, and remove the largest edge betweenness until the network graph is split into multiple communities, and calculate the internal characteristics of each community; Step 3: Build a Gaussian mixture model, define multiple Gaussian components, each component represents a Gaussian distribution, and then use the expectation maximization algorithm to calculate the posterior probability that the data point belongs to each Gaussian component, and update the Gaussian mixture model parameters according to the posterior probability; Step 4: Calculate the probability distribution of each mass spectrum data point through the updated Gaussian mixture model, analyze each Gaussian component in the Gaussian mixture model, identify the multi-peak characteristics of the mass spectrum data, and then determine the different peak positions and distribution forms of the mass spectrum data based on the mean and covariance of each Gaussian component.

Images