Whistle point gene pool data processing method based on clustering analysis

Abstract

The invention belongs to the technical field of data processing, and discloses a sentinel point gene pool data processing method based on clustering analysis, which comprises the following steps: extracting gene expression data of samples in a sentinel point gene pool, and constructing a static feature space and a dynamic time sequence space; performing preliminary clustering on the static feature space to obtain an initial static group; switching to a dynamic time sequence space in each static group, performing secondary clustering on samples in the group based on time sequence similarity, and refining the samples into dynamic evolution subgroups; fitting a dynamic trajectory model of which the gene expression changes along with time for each dynamic evolution subgroup, calculating a deviation degree between a sample and a predicted value of the corresponding dynamic trajectory model in a static feature space, and marking the sample of which the deviation degree exceeds a preset confidence interval as an abnormal sample; and the application scene and value of the sentinel point gene pool in biomedical research are further expanded.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, more particularly, the present application relates to a data processing method for sentinel gene library based on cluster analysis. BACKGROUND

[0002] In biomedical research and clinical application, the sentinel gene library is a core resource for capturing gene expression dynamics and mining disease markers, which contains static information (such as single time point expression) and dynamic time series information (such as expression fluctuation trend). In the process of data processing, both should be taken into account. However, the current sentinel gene library data processing technology has obvious bottleneck, which is difficult to release the data value: Firstly, the existing technology has the bias of "paying attention to static and ignoring dynamic", which is difficult to describe the time evolution law of gene expression. Most methods only rely on single time point static data, or simply splice time series information, ignoring the difference of dynamic trajectory, leading to the misclassification of samples with similar static characteristics but different evolution trend, which covers up the disease subtype heterogeneity and key progress inflection point; Secondly, the cluster analysis adopts a single logic of one clustering to determine the result, and the grouping granularity is rough, which is difficult to realize "static classification and dynamic subdivision", which leads to the covering up of the heterogeneity of samples in the group and the inability to reveal the potential dynamic subgroup; Thirdly, the abnormal sample detection lacks dynamic support and has insufficient accuracy. Traditional methods rely on fixed threshold or static statistical indicators, do not consider the deviation of dynamic trajectory, and lack modelized confidence interval constraint, which is easy to produce false positive and false negative, and is difficult to effectively identify key abnormal samples related to disease phenotype or experimental deviation; Finally, the data processing results are isolated and lack dynamic correlation. The existing method only gives grouping results or abnormal list, cannot quantify and visualize the trajectory similarity between groups, and cannot reveal the relationship between abnormal samples and normal subgroup, which is not conducive to researchers to understand the gene evolution network as a whole.

[0003] In view of this, the present application provides a data processing method for sentinel gene library based on cluster analysis to solve the above problems. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a data processing method for sentinel gene library based on cluster analysis, comprising: S1, extracting gene expression data of samples in the sentinel gene library, and constructing a static feature space and a dynamic time series space; S2, preliminarily clustering the static feature space to obtain an initial static grouping; switching to the dynamic time series space in each static grouping, and performing secondary clustering on the samples in the group based on time series similarity to refine into dynamic evolution subgroups; S3, fitting a dynamic trajectory model of gene expression change over time for each dynamic evolution subgroup, calculating the deviation degree of the sample and the corresponding dynamic trajectory model prediction value in the static feature space, and marking the sample with a deviation degree exceeding the preset confidence interval as an abnormal sample; S4, taking the dynamic evolution subgroup as the core node and the abnormal sample as the independent node, calculating the correlation strength between each node according to the trajectory feature similarity, and constructing the sentinel gene dynamic graph by taking the correlation strength as the edge.

[0005] Specifically, the gene expression data includes transcriptome data and proteome data; the transcriptome data refers to the expression amount of RNA molecules generated by gene transcription, including mRNA expression amount and non-coding RNA expression amount; the proteome data refers to the expression amount of proteins generated by translation of gene transcription products, including protein expression amount and protein modification level.

[0006] Specifically, the method for constructing the static feature space and the dynamic time sequence space comprises: Based on the gene expression data, the gene expression values of the samples at the preset fixed time points are taken as features, which are input after standardization and feature dimension reduction processing, to obtain sample static feature vectors and form a static feature space representing the molecular state of the samples at the static time section. Meanwhile, in the dynamic time sequence space, the gene expression data of the samples at different continuous time points are arranged in time sequence to form a multi-dimensional time sequence; the processed time sequence is mapped to a low-dimensional trajectory vector by principal component analysis method, which is taken as the coordinate point of the sample in the dynamic time sequence space and embedded to constitute the dynamic time sequence space.

[0007] Specifically, the method for performing preliminary clustering on the static feature space comprises: The static feature vectors of all samples in the static feature space are standardized to make different gene expression amounts comparable; based on the standardized static feature vectors, an unsupervised clustering method is used to group the samples to obtain the initial static grouping of the samples in the static feature space.

[0008] Specifically, the method for performing secondary clustering on the samples in the group comprises: For samples belonging to the same initial static group, gene expression data at different continuous time points are extracted to form a time sequence trajectory; the similarity between the time sequences of the samples is measured by a cosine similarity method, and a similarity matrix of the samples in the group is constructed based on the similarity measurement result; Based on the similarity matrix, a time sequence clustering method is used to perform secondary clustering on the samples, and the samples in the same initial static group are refined into different dynamic evolution subgroups, each dynamic evolution subgroup corresponding to a type of samples with similar gene expression evolution trajectories in the time dimension.

[0009] Specifically, the method for obtaining the dynamic trajectory model comprises: For each sample in the dynamic evolution subgroup, gene expression data at consecutive time points are extracted; the gene expression data are preprocessed; a preset function space is selected as the fitting form of the dynamic trajectory model, and the samples in the dynamic evolution subgroup are fitted in the function space by minimizing the fitting error, so as to finally generate a dynamic trajectory model capable of representing the trend of gene expression change over time for each dynamic evolution subgroup.

[0010] Specifically, the method for calculating the deviation degree of the sample from the predicted value of the corresponding dynamic trajectory model comprises: At any time point, the gene expression value of the sample in the static feature space is extracted, and the predicted value of the dynamic trajectory model to which the sample belongs at the corresponding time point is compared, the difference value is calculated as a residual, and the absolute value of the residual is normalized according to the residual standard deviation in the dynamic evolution subgroup to which the corresponding time point belongs, to obtain a normalized deviation value; the normalized deviation values at all time points are averaged to obtain the overall deviation of the sample over the entire time sequence, which is further used as the deviation degree of the sample from the predicted value of the corresponding dynamic trajectory model.

[0011] Specifically, the method for obtaining the abnormal sample comprises: The residuals of all samples in the dynamic evolution subgroup are statistically analyzed, the residual covariance matrix at the time point is obtained by calculating the covariance of the residuals in each dimension; the dynamic trajectory model and the gene expression value of the sample are respectively parameterized and fitted to obtain the summary parameter vector of the dynamic trajectory model and the estimated parameter vector of the sample, and then the cross-space consistency deviation degree is calculated and obtained. When the deviation degree between the sample and the predicted value of the corresponding trajectory model and the cross-space consistency deviation degree both exceed the corresponding preset confidence interval, the sample is determined to be an abnormal sample.

[0012] Specifically, the method for obtaining the correlation strength between nodes comprises: Each dynamic evolution subgroup is taken as a core node, and each abnormal sample is taken as an independent node; for each node, dynamic trajectory features over a time sequence are extracted; the cosine similarity is used to calculate the dynamic trajectory feature similarity between any two nodes, and the correlation strength between the nodes is obtained.

[0013] Specifically, the method for obtaining the dynamic atlas of sentinel genes comprises: The calculated correlation strength is taken as the weight of the connecting edge between nodes in the sentinels gene dynamic atlas; when the correlation strength between two nodes is greater than or equal to a preset correlation strength threshold, a connecting edge is established between the two nodes, and the weight of the connecting edge is set to the correlation strength value between the two nodes; if the correlation strength between the independent node corresponding to the abnormal sample and any core node is less than the preset correlation strength threshold, the independent node is kept isolated; all nodes and their connecting edges form a weighted undirected graph, and then the sentinels gene dynamic atlas is obtained.

[0014] Compared with the prior art, the present application has the following beneficial effects: By simultaneously constructing a static feature space and a dynamic time sequence space, and combining two-level clustering to realize "static grouping-dynamic subdivision", the present application can accurately depict the evolution trajectory of gene expression on the basis of global layering, effectively avoid the misclassification problem of samples that are similar in static state but different in dynamic path in the prior art, and thus reveal the key subtypes and inflection points of disease progression. Secondly, by fitting a dynamic trajectory model in the dynamic evolution subgroup and calculating double deviation degrees based on model residuals and cross-space consistency, a statistical discrimination mechanism for abnormal samples is established; compared with the traditional empirical threshold method, this mechanism is more objective and robust, can significantly reduce false positives and false negatives, improve the accuracy of abnormal sample detection, and facilitate researchers to capture important information related to disease phenotypes, drug reactions or experimental bias. Thirdly, a dynamic atlas construction method is introduced, taking the dynamic evolution subgroup as the core node and the abnormal sample as the independent node, and quantifying the node correlation through trajectory feature similarity, to form a sentinels gene dynamic atlas with structured expression capability; this atlas can not only visually display typical evolution patterns, but also reveal the branching paths and attribution tendency of abnormal samples, providing a visual tool for global understanding of gene networks. In summary, the present application breaks through the limitations of traditional methods, and can provide more refined, dynamic and visual data processing means for sentinel gene library research, and improve the exploration of disease mechanisms and clinical application value. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The figure is a flowchart of the present application of data processing method for sentinel gene library based on clustering analysis. Figure 2 The figure is a structural diagram of the present application of data processing system for sentinel gene library based on clustering analysis. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Example 1 See also Figure 1 As shown, this embodiment 1 further illustrates the sentinel gene library data processing method based on cluster analysis proposed by the present invention, including: In biomedical research and clinical applications, sentinel gene libraries serve as core data resources for capturing dynamic changes in gene expression, identifying disease diagnostic markers, and analyzing drug mechanisms of action. The accuracy and effectiveness of their data processing directly determine the reliability of subsequent research conclusions. With the rapid development of high-throughput sequencing technology, the number of samples accumulated in sentinel gene libraries continues to grow. Gene expression data not only contains static information reflecting the basic biological characteristics of the samples (such as gene expression levels and differential expression folds at specific time points), but also includes dynamic time series information reflecting changes in gene expression with physiological status, disease progression, drug intervention, and other factors (such as fluctuation trends in gene expression levels at different time points and inflection point characteristics). This places a dual requirement on data processing methods to "balance the integrity of static features with the regularity of dynamic time series."

[0018] However, current data processing technologies for sentinel gene libraries still have many shortcomings: First, existing technologies generally have a processing bias of "focusing on statics over dynamics," failing to characterize the temporal evolution of gene expression. Most methods still focus on static data from a single time point, or simply concatenate time series data from multiple time points into static vectors, or even ignore temporal information. This can lead to the misclassification of samples with similar static features but different dynamic pathways. For example, in cancer research, some samples may have consistent initial sentinel gene expression but show increasing or decreasing trends over the course of the disease. Existing technologies are unable to distinguish between these, masking the heterogeneity of disease subtypes and missing key turning points in disease progression.

[0019] Second, cluster analysis, the core method for grouping sentinel gene bank samples, often relies on a single, single-step clustering logic, resulting in a coarse grouping granularity that makes it difficult to refine sample heterogeneity. Due to individual sample differences, traditional single-step clustering cannot achieve "homogeneity within groups and heterogeneity between groups," and it cannot meet the requirements of "statically grouping into broad categories and dynamically subdividing into subgroups."

[0020] Third, the abnormal sample detection accuracy is low, and lacks dynamic support. The abnormal samples in the sentinel gene library (such as abnormal fluctuations in gene expression and samples deviating from the overall evolutionary trend) often correspond to key information such as special disease phenotypes, experimental operation errors or drug intervention failures, and are the focus of research. However, the existing technology relies on static indicators such as "gene expression exceeding the mean ± n times the standard deviation" and "differential expression fold greater than a fixed threshold" to judge abnormalities, without considering the relative deviation of samples in the dynamic evolution process, and the abnormality judgment lacks statistical model support and relies only on subjective experience threshold, resulting in high false positive (normal sample misjudged as abnormal) and false negative (abnormal sample missed) rates.

[0021] Fourth, the data processing results are isolated and lack dynamic correlation. Traditional methods only output grouping lists or abnormal sample lists, and cannot quantify and visualize group and sample correlations, which cannot show subgroup trajectory similarity or clearly show the correlation tendency of abnormal samples and normal subgroups, which is not conducive to researchers' overall understanding of gene evolution networks and affects in-depth exploration.

[0022] Therefore, the present application proposes a sentinel gene library data processing method based on clustering analysis, comprising: S1, extracting gene expression data of samples in the sentinel gene library, constructing a static feature space and a dynamic time sequence space; S2, performing preliminary clustering on the static feature space to obtain initial static groups; switching to the dynamic time sequence space within each static group, performing secondary clustering on the samples in the group based on time sequence similarity to refine into dynamic evolution subgroups; S3, fitting a dynamic trajectory model of gene expression change over time for each dynamic evolution subgroup, calculating the deviation degree of the sample and the predicted value of the corresponding dynamic trajectory model in the static feature space, and marking the sample with a deviation degree exceeding a pre-set confidence interval as an abnormal sample; S4, taking the dynamic evolution subgroup as the core node and the abnormal sample as the independent node, calculating the correlation strength between each node according to the trajectory feature similarity, and constructing a sentinel gene dynamic map with the correlation strength as the edge.

[0023] Gene expression data includes transcriptome data and proteome data; transcriptome data refers to the expression of RNA molecules produced by gene transcription, including mRNA (messenger RNA) expression and non-coding RNA expression, and is the most basic and commonly used gene expression data type, which is used to reflect the expression activity of "transcription level" of genes (i.e. whether the gene is "activated" and how much it is activated); proteome data refers to the expression of proteins generated by the translation of gene transcription products, including protein expression and protein modification level, which is the final product reflecting the "translation level" of genes (RNA needs to be translated into protein to directly participate in physiological / pathological processes); Note: Gene expression refers to the process of converting genetic information in genes into molecules with biological functions through transcription (DNA→RNA) and translation (RNA→protein), so gene expression data is essentially a quantitative record of key molecules in this process, including transcriptome data and proteome data, which correspond to different detection dimensions: Among them, the mRNA expression amount directly reflects the degree of gene activation transcription, and is the template for protein synthesis. Commonly used standardized indicators include FPKM (transcripts per million mapped reads per kilobase) and TPM (transcripts per million). Non-coding RNA expression includes lncRNA (long non-coding RNA), miRNA (microRNA), circRNA (circular RNA), etc. Although they do not directly encode proteins, they are involved in gene expression regulation, and changes in their expression can indirectly reflect the functional state of genes. Protein expression: Protein abundance detected by mass spectrometry, protein chip and other technologies can directly reflect the functional effects of gene expression (avoiding the problem of “transcription-translation inconsistency”); Protein modification level: Post-translational modification states such as phosphorylation, acetylation, methylation, etc. These modifications can significantly affect protein function, and their quantitative data (such as the abundance of a protein phosphorylation site) are important supplements to the functional state of gene expression.

[0024] The method for constructing a static feature space and a dynamic time sequence space includes: Based on gene expression data, the gene expression values of samples at preset fixed time points are input as features, and after standardization and feature dimension reduction processing, the sample static feature vectors are obtained, forming a static feature space representing the molecular state of samples at a static time section. At the same time, in the dynamic time sequence space, the gene expression data of samples at different continuous time points is arranged in time sequence to form a multi-dimensional time sequence. The processed time sequence is mapped to a low-dimensional trajectory vector by principal component analysis method, which is used as the coordinate point of the sample in the dynamic time sequence space, and is embedded to form a dynamic time sequence space.

[0025] The method for preliminary clustering of the static feature space includes: Standardize all sample static feature vectors in the static feature space to make different gene expression amounts comparable; based on the standardized static feature vectors, use an unsupervised clustering method to group the samples; the unsupervised clustering method can be any one of K-means clustering, hierarchical clustering or density-based clustering method, to obtain the initial static grouping of samples in the static feature space.

[0026] The method for performing secondary clustering on the samples in the group includes: For samples belonging to the same static group, gene expression data at different consecutive time points are extracted to form time series trajectories; the similarity between the time series of samples is measured by the cosine similarity method, and a similarity matrix of samples within the group is constructed based on the similarity measurement results; Based on the similarity matrix, a time series clustering method is used for secondary clustering of samples, such as K-means clustering based on dynamic time warping, time series clustering method based on hierarchical clustering, or trajectory clustering based on Gaussian process mixture model, to refine the samples in the same static group into different dynamic evolution subgroups, each dynamic evolution subgroup corresponding to a class of samples with similar gene expression evolution trajectories in the time dimension.

[0027] The method for obtaining the dynamic trajectory model comprises: For samples in each dynamic evolution subgroup, gene expression data at consecutive time points are extracted; the gene expression data is preprocessed, including standardization, smoothing and noise suppression operations; a pre-set function space is selected as the fitting form of the dynamic trajectory model, and the samples in the dynamic evolution subgroup are fitted in the function space by minimizing the fitting error, finally generating a dynamic trajectory model for each dynamic evolution subgroup that can represent the trend of gene expression change over time.

[0028] The method for calculating the deviation degree of the sample from the predicted value of the corresponding dynamic trajectory model comprises: At any time point, the gene expression value of the sample in the static feature space is extracted, and the predicted value of the dynamic trajectory model to which the sample belongs at the corresponding time point is compared, the difference between the two is calculated as a residual, and the absolute value of the residual is normalized according to the residual standard deviation (plus a zero constant) within the dynamic evolution subgroup to which the corresponding time point belongs, to obtain a normalized deviation value; the normalized deviation values of all time points are averaged to obtain the overall deviation of the sample over the entire time series, which is further used as the deviation of the sample from the predicted value of the corresponding dynamic trajectory model.

[0029] In addition, if the calculation process of the deviation degree is expressed by a mathematical formula: deviation ; wherein, represents the deviation between the sample and the corresponding trajectory model; represents the total number of time points, i.e. the number of time nodes at which the gene expression value of the sample is collected; represents the time point; represents the actual observed gene expression value of the sample at the time point ; represents the dynamic evolution subgroup corresponding to the initial static group to which the sample belongs The established dynamic trajectory model, at the time point For samples The predicted value of Indicates the dynamic evolution subgroup to which the sample belongs at the time point The residual standard deviation is used to characterize the uncertainty of model prediction; The index representing the time point; Represents a non-zero constant introduced to prevent the denominator from being zero; Indicates the number of the initial static group; Indicates the number of the dynamically evolving subgroup.

[0030] Methods for obtaining abnormal samples include: Statistical analysis was performed on the residuals of all samples within the dynamic evolution subgroup. The residual covariance matrix at that time point was obtained by calculating the covariance of the residuals in each dimension. Parametric fitting was performed on the dynamic trajectory model and the gene expression values ​​of the samples, respectively, to obtain the summary parameter vector of the dynamic trajectory model and the estimated parameter vector of the samples, and then the cross-space consistency deviation was calculated. Cross-space consistency deviation: ;in, Representation sample The cross-space consistency deviation; represents the estimated parameter vector of the sample; Indicates the The first static group Summary parameter vector of the dynamic trajectory model for each dynamically evolving subgroup; Representation sample The inverse of the residual covariance matrix; It should be noted that the cross-space consistency deviation here is achieved by parametrically fitting the sample and trajectory models and comparing the differences between the two in the parameter space. It reflects the consistency between the dynamic parameters predicted by static features and the actual trajectory parameters. It is an evaluation indicator at the overall parameter level and focuses on the consistency between static features and dynamic evolution.

[0031] In a certain dynamic evolution subgroup, the gene expression of many samples will show similar changing trends over time. We hope to use a "mathematical formula" to characterize this trend; for example, the gene expression of the dynamic evolution subgroup may first rise and then fall, so we use a suitable function (such as a polynomial, exponential function or spline curve) to fit this curve, and we get a dynamic trajectory model; among them, this function contains several parameters (such as slope, inflection point position, and degree of curve curvature), which together constitute the summary parameter vector of this dynamic evolution subgroup, which is equivalent to the "typical trajectory identity card" of this dynamic evolution subgroup, used to represent the typical pattern of the group; For a single sample, we can also use the same "mathematical formula" to fit its observed data, although the data points are less than the dynamic evolution subgroup and the noise is larger; the fitting result will obtain a set of parameters (such as its expression level, change trend direction), which is the estimated parameter vector of the sample; it can be understood as: the sample "in the static feature space, the simplified representation of its own corresponding trajectory form", used for individual and group comparison and reference; When the deviation degree between the sample and the corresponding trajectory model predicted value and the cross-space consistency deviation degree exceed the corresponding preset confidence interval, the sample is determined as an abnormal sample; Solved the following problems existing in the prior art: the traditional analysis method based on static characteristics or a single time point is difficult to capture the dynamic change characteristics of gene expression over time, resulting in lack of time sequence sensitivity in identifying abnormal samples; the existing method usually ignores the natural fluctuation of samples in the group and measurement errors, and is prone to false positives or false negatives when determining abnormal samples. The traditional method usually only analyzes a single dimension of dynamic or static characteristics, and cannot verify the consistency of the sample between static characteristics and dynamic trajectory, so that abnormal samples may be missed or normal samples may be misjudged. The existing technology relies on a single index to identify abnormal samples, which may lead to incomplete and unstable abnormal sample detection results, and cannot consider dynamic deviation, residual uncertainty and cross-space consistency at the same time.

[0032] The beneficial effects of the prior art are: by fitting the dynamic trajectory model of gene expression over time in each dynamic evolution subgroup, and calculating the deviation degree between the sample and the model predicted value, the dynamic trajectory deviation abnormal sample which shows significant deviation from the trajectory pattern in the time sequence can be accurately identified, and the limitations of the prior art in dynamic anomaly detection are solved.

[0033] By statistically analyzing the residual of each time point sample and the fitted trajectory, the covariance matrix of the residual in the group is calculated, so as to model the uncertainty of the residual, which can identify abnormal samples that are outside the scope explained by the residual covariance, and enhance the accuracy and reliability of the anomaly detection.

[0034] The cross-space consistency deviation degree is introduced, which compares the dynamic summary parameters predicted by the static characteristics of the sample with the summary parameters of the actual dynamic trajectory model, so as to identify samples with too large difference between static and dynamic characteristics, so as to ensure that the determination of abnormal samples considers multi-dimensional space information, and improves the comprehensiveness of identification; by considering dynamic trajectory deviation, residual uncertainty and cross-space consistency at the same time, the present application can identify comprehensive abnormal samples which are abnormal in three dimensions. This method makes up for the defects of the prior art which only relies on a single index and is prone to one-sided judgment, so that the abnormal sample detection is more comprehensive and stable.

[0035] The method for obtaining the correlation strength between nodes comprises: Each dynamic evolution subgroup is taken as a core node, and each abnormal sample is taken as an independent node; for each node, dynamic trajectory features on a time sequence are extracted; between any two nodes, cosine similarity is used to calculate dynamic trajectory feature similarity, and then the correlation strength between the nodes is obtained.

[0036] The method for obtaining the dynamic atlas of sentinel genes comprises: The calculated correlation strength is taken as the weight of the connection edge between nodes in the dynamic atlas of sentinel genes; when the correlation strength between two nodes is greater than or equal to a preset correlation strength threshold, a connection edge is established between the two nodes, and the weight of the connection edge is set to the correlation strength value between the two nodes; if the correlation strength between the independent node corresponding to the abnormal sample and any core node is less than the preset correlation strength threshold, the independent node remains isolated, which is used to reveal a potential unique abnormal pattern; all nodes and their connection edges form a weighted undirected graph, and then the dynamic atlas of sentinel genes is obtained; in the dynamic atlas of sentinel genes, the connection edges between core nodes reveal the similarity of typical evolution patterns, and the connection edges between core nodes and independent nodes reveal the source and degree of abnormal deviation, so that the typical evolution patterns and abnormal branch paths are simultaneously revealed in the atlas.

[0037] The preset signal interval is set by the staff based on the analysis results of historical data, and the historical analysis process comprises collecting the deviation degrees of multiple data points by the system and calculating the average value as a reference, which can be adjusted by the staff during the system operation; similarly, the preset correlation strength threshold is set.

[0038] In this embodiment, by simultaneously constructing a static feature space and a dynamic time sequence space, and combining two-level clustering to realize “static grouping-dynamic subdivision”, the evolution trajectory of gene expression can be accurately described on the basis of global layering, the misclassification problem of samples with static similarity but different dynamic paths in the prior art can be effectively avoided, and the key subtypes and turning points of disease progression can be revealed; Secondly, by fitting a dynamic trajectory model in the dynamic evolution subgroup and calculating double deviation degrees based on model residuals and cross-space consistency, a statistical discrimination mechanism of abnormal samples is established; compared with the traditional empirical threshold method, this mechanism is more objective and robust, can significantly reduce false positives and false negatives, improve the accuracy of abnormal sample detection, and facilitate researchers to capture important information related to disease phenotypes, drug reactions or experimental bias; Thirdly, a dynamic graph construction method is introduced, taking the dynamic evolution subgroups as the core nodes and the abnormal samples as the independent nodes, and quantifying the node association through the trajectory feature similarity, to form a sentinel gene dynamic graph with structured expression ability; the graph can not only intuitively show the typical evolution mode, but also reveal the branch path and attribution tendency of the abnormal samples, providing an intuitive tool for global understanding of the gene network; In summary, the application breaks through the limitations of traditional methods, and can provide more refined, dynamic and visual data processing means for sentinel gene library research, and improve the disease mechanism exploration and clinical application value.

[0039] Embodiment 2 Please refer to Figure 2 As shown in the figure, some parts of this embodiment are not described in detail in the description of embodiment 1, and a sentinel gene library data processing system based on cluster analysis is provided, which includes: The feature mapping module extracts the gene expression data of the samples in the sentinel gene library, and constructs a static feature space and a dynamic time sequence space; The hierarchical time sequence clustering module performs preliminary clustering on the static feature space to obtain an initial static grouping; in each static grouping, it switches to the dynamic time sequence space, and performs secondary clustering on the samples in the group based on the time sequence similarity to refine into dynamic evolution subgroups; The trajectory deviation detection module fits a dynamic trajectory model of gene expression change over time for each dynamic evolution subgroup, calculates the deviation degree of the sample and the predicted value of the corresponding dynamic trajectory model in the static feature space, and marks the sample whose deviation degree exceeds the preset confidence interval as an abnormal sample; The gene association construction module takes the dynamic evolution subgroup as the core node and the abnormal sample as the independent node, calculates the association strength between each node according to the trajectory feature similarity, and constructs a sentinel gene dynamic graph with the association strength as the edge.

[0040] Since the electronic device introduced in this embodiment is the electronic device used to implement the sentinel gene library data processing method based on cluster analysis in the embodiments of the application, the specific implementation of the electronic device and its various forms of change can be understood by those skilled in the art based on the sentinel gene library data processing method based on cluster analysis introduced in the embodiments of the application, so the method of how to implement the method in the embodiments of the application will not be described in detail. As long as the electronic device used to implement the sentinel gene library data processing method based on cluster analysis in the embodiments of the application is implemented by those skilled in the art, it belongs to the scope of protection of the application.

[0041] The above formulas are dimensionless numerical calculations, the formula is a formula obtained by software simulation of a large amount of data to obtain the most recent real situation, and the preset parameters and threshold values in the formula are set by those skilled in the art according to the actual situation.

[0042] The above merely describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-described embodiments. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary technical operators in the technical field, some improvements and refinements without departing from the principles of the present application shall also be considered as falling within the protection scope of the present application.

Claims

1. A sentinel gene library data processing method based on cluster analysis, characterized in that: include: S1. Extract gene expression data of samples in the sentinel gene library and construct static feature space and dynamic time series space; S2, perform preliminary clustering on the static feature space to obtain initial static grouping; Switch to the dynamic time series space within each static group, perform secondary clustering on the samples within the group based on time series similarity, and refine them into dynamically evolving subgroups; S3. Fit a dynamic trajectory model of gene expression changes over time for each dynamically evolving subgroup, calculate the deviation between the sample and the corresponding dynamic trajectory model prediction value in the static feature space, and mark the sample whose deviation exceeds the preset confidence interval as an abnormal sample; S4. Taking the dynamically evolving subgroups as core nodes and abnormal samples as independent nodes, the correlation strength between nodes is calculated based on the similarity of trajectory features, and the sentinel gene dynamic map is constructed with the correlation strength as the edge.

2. The sentinel gene library data processing method based on cluster analysis according to claim 1, characterized in that: The gene expression data includes transcriptome data and proteome data; transcriptome data refers to the expression level of RNA molecules produced by gene transcription, including mRNA expression level and non-coding RNA expression level; proteome data refers to the expression level of proteins generated by translation of gene transcription products, including protein expression level and protein modification level.

3. The sentinel gene library data processing method based on cluster analysis according to claim 2, characterized in that: The method for constructing the static feature space and the dynamic time series space includes: Based on gene expression data, the gene expression value of the sample at a preset fixed time point is used as the feature input. After standardization and feature dimensionality reduction, the static feature vector of the sample is obtained, forming a static feature space that characterizes the molecular state of the sample at the static time section; At the same time, in the dynamic time series space, the gene expression data of the samples at different continuous time points are arranged in chronological order and organized into a multidimensional time series; the processed time series are mapped into low-dimensional trajectory vectors through principal component analysis, which serve as the coordinate points of the samples in the dynamic time series space and are embedded into the dynamic time series space.

4. The sentinel gene library data processing method based on cluster analysis according to claim 3, characterized in that: The method for performing preliminary clustering on the static feature space comprises: The static feature vectors of all samples in the static feature space are standardized to make the expression levels of different genes within a comparable range; based on the standardized static feature vectors, the samples are grouped using an unsupervised clustering method to obtain the initial static grouping of the samples in the static feature space.

5. The sentinel gene library data processing method based on cluster analysis according to claim 4, characterized in that: The method for performing secondary clustering on samples within a group includes: For samples belonging to the same initial static group, gene expression data at different consecutive time points are extracted to form time series trajectories. The similarity between the time series of samples is measured using the cosine similarity method, and a similarity matrix of samples within the group is constructed based on the similarity measurement results. Based on the similarity matrix, the time series clustering method is used to perform secondary clustering on the samples, and the samples in the same initial static grouping are refined into different dynamic evolution subgroups. Each dynamic evolution subgroup corresponds to a class of samples with similar gene expression evolution trajectories in the time dimension.

6. The sentinel gene library data processing method based on cluster analysis according to claim 5, characterized in that: The method for obtaining the dynamic trajectory model includes: For samples within each dynamically evolving subgroup, gene expression data at consecutive time points are extracted; the gene expression data are preprocessed; a preset function space is selected as the fitting form of the dynamic trajectory model, and the samples within the dynamically evolving subgroup are fitted in the function space by minimizing the fitting error. Finally, a dynamic trajectory model is generated for each dynamically evolving subgroup that can characterize the temporal trend of its gene expression.

7. The sentinel gene library data processing method based on cluster analysis according to claim 6, characterized in that: The method for calculating the deviation between a sample and a corresponding dynamic trajectory model prediction value includes: At any time point, the gene expression value of the sample in the static feature space is extracted and compared with the predicted value of the dynamic trajectory model to which the sample belongs at the corresponding time point. The difference between the two is calculated as the residual, and the absolute value of the residual is normalized according to the standard deviation of the residual in the dynamic evolution subgroup to which the corresponding time point belongs to obtain the normalized deviation value; the normalized deviation values ​​of all time points are averaged to obtain the overall deviation of the sample in the entire time series, which is then used as the deviation of the sample from the predicted value of the corresponding dynamic trajectory model.

8. The sentinel gene library data processing method based on cluster analysis according to claim 7, characterized in that: The method for obtaining the abnormal sample includes: Statistical analysis was performed on the residuals of all samples within the dynamic evolution subgroup. The residual covariance matrix at that time point was obtained by calculating the covariance of the residuals in each dimension. Parametric fitting was performed on the dynamic trajectory model and the gene expression values ​​of the samples, respectively, to obtain the summary parameter vector of the dynamic trajectory model and the estimated parameter vector of the samples, and then the cross-space consistency deviation was calculated. When the deviation and cross-space consistency deviation between a sample and the corresponding trajectory model prediction value both exceed the corresponding preset confidence interval, the sample is judged to be an abnormal sample.

9. The sentinel gene library data processing method based on cluster analysis according to claim 8, characterized in that: The method for obtaining the association strength between the nodes includes: Each dynamic evolution subgroup is regarded as a core node, and each abnormal sample is regarded as an independent node. For each node, the dynamic trajectory features in the time series are extracted. Between any two nodes, the cosine similarity is used to calculate the similarity of the dynamic trajectory features, and then the correlation strength between the nodes is obtained.

10. The sentinel gene library data processing method based on cluster analysis according to claim 9, characterized in that: The method for obtaining the sentinel gene dynamic map includes: The calculated association strength is used as the weight of the connecting edge between nodes in the sentinel gene dynamic map; when the association strength between two nodes is greater than or equal to the preset association strength threshold, a connecting edge is established between the two nodes, and the weight of the connecting edge is set to the association strength value between the two nodes; if the association strength between the independent node corresponding to the abnormal sample and any core node is less than the preset association strength threshold, it remains isolated; all nodes and their connecting edges are formed into a weighted undirected graph to obtain the sentinel gene dynamic map.

Images