Methods, systems, electronic devices and storage media for proteomic biomarker identification

Abstract

This invention discloses a method, system, electronic device, and storage medium for proteomic biomarker identification. The method adds one-hot encoded grouping labels to protein abundance data in preprocessed mass spectrometry quantitative data and encodes the group information corresponding to the protein abundance data into binary vectors. Supervised training of a data augmentation model is performed, clipping the log-variance of the encoder output to the (-10, 10) range. New sample data is generated using the decoder of the trained data augmentation model. Based on the results of principal component analysis, new sample data with similarity reaching a set threshold are grouped into a set. Potential biomarkers are identified using differential expression analysis or random forest methods within these group sets. The potential biomarkers are ranked according to the fold change or the importance parameters of the model output, resulting in multi-group exosome proteomic biomarker identification results. This invention improves the accuracy and reliability of biomarker identification.

Description

Technical Field

[0001] This invention relates to a method and system for identifying proteomic biomarkers, belonging to the field of bioinformatics technology. Background Technology

[0002] Currently, in the field of proteomics analysis in bioinformatics, differential protein analysis (DPA) is a common method used to identify protein biomarkers with significant differences in abundance across multiple sample groups (such as multiple disease sample groups).

[0003] In related technologies, the main steps of differential protein analysis generally include: A1) Data acquisition and preprocessing: obtaining protein abundance data through proteomics sequencing, and performing quality control, missing value imputation, and standardization. A2) Differential protein analysis: using statistical methods (such as t-tests, Wilcoxon tests, etc.) and machine learning methods (such as random forests, support vector machines, etc.) to analyze the preprocessed data to identify proteins with significant differences. A3) Biomarker identification: screening potential protein biomarkers based on statistical significance and fold change.

[0004] Differential protein analysis methods have been widely used in proteomics research, but they have significant limitations when processing exosome samples and comparing samples from multiple groups, mainly in the following aspects:

[0005] B1) Consistency issues of exosome samples:

[0006] High complexity: The preparation and detection of exosome samples are more complex than those of cell samples, resulting in low consistency among duplicate samples in proteomics sequencing results.

[0007] High noise: Due to the high noise background of the data, it is difficult to accurately capture reasonable and robust differences, resulting in insufficient accuracy in differential protein identification.

[0008] B2) Small sample size (i.e., insufficient sample diversity):

[0009] Cost and scarcity: Exosome sample preparation is costly and samples are hard to come by. Typically, only 3-5 samples are needed for omics sequencing of each disease, resulting in a small sample size.

[0010] Limitations of statistical analysis: Small sample size restricts the effectiveness of statistical analysis and the generalization ability of the model, resulting in reduced reliability of marker identification.

[0011] B3) The complexity of multiple comparisons:

[0012] Exacerbated by noise and sample size issues: In the case of multiple group comparisons, the aforementioned problems (B1) and (B2) (data noise and small sample size) have a greater impact on the analysis results because comparisons need to be made between multiple groups, increasing data complexity.

[0013] Limitations of traditional methods: When using traditional statistical tests (such as the Wilcox test) or machine learning methods (such as random forests) to identify potential exosomal protein biomarkers for diseases, insufficient biomarker identification or misjudgment is easily caused by noise and sample size issues.

[0014] In summary, existing technologies have significant limitations in processing exosomal proteome data, especially when comparing multiple groups of samples, and a new method is urgently needed to improve the accuracy and reliability of biomarker identification. Summary of the Invention

[0015] Therefore, this invention provides a method and system for identifying proteomic biomarkers, which solves the problems of poor accuracy and low reliability of traditional techniques in processing exosomal proteomic data, especially when comparing multiple groups of samples.

[0016] To achieve the above objectives, the present invention provides the following technical solution: Firstly, a method for identifying proteomic biomarkers is provided, comprising:

[0017] One-hot encoded grouping labels are added to the protein abundance data in the mass spectrometry quantitative data, and the group information corresponding to the protein abundance data is encoded into a binary vector;

[0018] A data augmentation model based on variational autoencoder is constructed. The protein abundance data and one-hot encoded group labels are used to conduct supervised training on the data augmentation model. The decoder of the trained data augmentation model is then used to generate new sample data.

[0019] Principal component analysis and similarity analysis are performed on the new sample data; based on the results of the principal component analysis, the new sample data with similarity reaching a set threshold are divided into a group set;

[0020] Potential biomarkers are identified using a specified method, and then ranked according to their importance parameters to obtain multi-group exosomal proteomic biomarker identification results.

[0021] As a preferred method for proteomic biomarker identification, the method further includes data preprocessing of the mass spectrometry quantitative data, which includes data quality control, data normalization, and missing value imputation.

[0022] As a preferred method for proteomic biomarker identification, during the data quality control process, protein abundance data with abundance values ​​of missing values ​​are removed from more than a specified number of samples.

[0023] During the data normalization process, a specified normalization method is used to normalize the protein abundance data;

[0024] During the missing value imputation process, a specified data imputation model is used to imput missing values ​​in the protein abundance data after data normalization.

[0025] As a preferred method for proteomic biomarker identification, the loss function of the data augmentation model includes a reconstruction error function and a KL divergence function; an optimizer is used to minimize the loss function of the data augmentation model to optimize its parameters. For example, the Adam (Adaptive Moment Estimation) optimizer is an efficient optimization algorithm used in deep learning, which has proven effective in practice and has become one of the default optimizers for many deep learning models.

[0026] As a preferred method for proteomic biomarker identification, the expression for the reconstruction error function in the loss function of the data augmentation model is as follows:

[0027]

[0028] In the formula, recon_batch represents the reconstructed protein abundance data; data represents the original input protein abundance data; and recon_batch i This represents the reconstructed protein abundance data for the i-th protein; data i This represents the i-th protein abundance data from the original input.

[0029] As a preferred method for proteomic biomarker identification, the expression for the KL divergence function in the loss function of the data augmentation model is as follows:

[0030]

[0031] In the formula, μ represents the mean of the latent space; D represents the dimension of the latent space; μ i σ represents the mean of the i-th dimension of the latent space. 2 i Let represent the variance of the i-th dimension of the latent space.

[0032] As a preferred method for proteomic biomarker identification, during the supervised training of the data augmentation model, the logarithmic variance of the encoder output of the data augmentation model is clipped to the range of (-10, 10).

[0033] As a preferred method for proteomic biomarker identification, during principal component analysis of the new sample data:

[0034] Calculate the covariance matrix of protein abundance data, solve for the eigenvalues ​​and eigenvectors of the covariance matrix, select the eigenvectors corresponding to the covariance matrix that reach the set eigenvalues ​​as principal components, and project the protein abundance numbers onto the determined principal components.

[0035] In the process of similarity analysis of new sample data after principal component analysis: based on the distribution of data points after dimensionality reduction by principal component analysis, the similarity between samples of a specified group is evaluated.

[0036] As a preferred method for proteomics biomarker identification, potential biomarkers are identified using differential expression analysis or random forest methods through the group set. The potential biomarkers are then ranked according to the fold change or the importance parameters output by the model filled with specified data to obtain multi-group exosomal proteomics biomarker identification results.

[0037] In a second aspect, the present invention also provides a proteomic biomarker identification system, comprising:

[0038] The data preparation module is used to add uniquely coded group labels to the protein abundance data in the mass spectrometry quantitative data and encode the group information corresponding to the protein abundance data into a binary vector.

[0039] The data augmentation model processing module is used to construct a data augmentation model based on variational autoencoder, and to perform supervised training on the data augmentation model using the protein abundance data and one-hot encoded group labels, and to generate new sample data using the decoder of the trained data augmentation model.

[0040] The group set generation module is used to perform principal component analysis and similarity analysis on the new sample data; based on the results of the principal component analysis, the new sample data with similarity reaching a set threshold are divided into a group set;

[0041] The biomarker identification module is used to identify potential biomarkers using a specified method, sort the potential biomarkers according to their importance parameters, and obtain multi-group exosomal proteomic biomarker identification results.

[0042] As a preferred embodiment of the proteome biomarker identification system, it also includes a data preprocessing module for preprocessing the mass spectrometry quantitative data, wherein the data preprocessing includes data quality control, data normalization and missing value imputation.

[0043] During the data quality control process, protein abundance data with missing values ​​are removed from more than a specified number of samples.

[0044] During the data normalization process, a specified normalization method is used to normalize the protein abundance data;

[0045] During the missing value imputation process, a specified data imputation model is used to imput missing values ​​in the protein abundance data after data normalization.

[0046] As a preferred embodiment of the proteomic biomarker identification system, the data augmentation model processing module includes:

[0047] The loss function of the data augmentation model includes a reconstruction error function and a KL divergence function; the Adam optimizer is used to minimize the loss function of the data augmentation model in order to optimize the parameters of the data augmentation model;

[0048] The expression for the reconstruction error function in the loss function of the data augmentation model is as follows:

[0049]

[0050] In the formula, recon_batch represents the reconstructed protein abundance data; data represents the original input protein abundance data; and recon_batch i This represents the reconstructed protein abundance data for the i-th protein; data i This represents the i-th protein abundance data from the original input;

[0051] The expression for the KL divergence function in the loss function of the data augmentation model is as follows:

[0052]

[0053] In the formula, μ represents the mean of the latent space; D represents the dimension of the latent space; μ i σ represents the mean of the i-th dimension of the latent space. 2 i Let represent the variance of the i-th dimension of the latent space;

[0054] In the data augmentation model processing module, during the supervised training of the data augmentation model, the logarithmic variance of the encoder output of the data augmentation model is clipped to the range of (-10, 10).

[0055] As a preferred embodiment of the proteomic biomarker identification system, the group set generation module includes:

[0056] Calculate the covariance matrix of protein abundance data, solve for the eigenvalues ​​and eigenvectors of the covariance matrix, select the eigenvectors corresponding to the covariance matrix that reach the set eigenvalues ​​as principal components, and project the protein abundance numbers onto the determined principal components.

[0057] In the process of similarity analysis of new sample data after principal component analysis: based on the distribution of data points after dimensionality reduction by principal component analysis, the similarity between samples of a specified group is evaluated.

[0058] As a preferred embodiment of the proteomic biomarker identification system, the biomarker identification module includes:

[0059] Potential biomarkers are identified using differential expression analysis or random forest methods within the group set. The potential biomarkers are then ranked according to the fold change or the importance parameters output by the model, resulting in the identification results of exosomal proteomic biomarkers across multiple groups.

[0060] Thirdly, the present invention provides an electronic device, comprising: one or more processors;

[0061] A storage device having stored one or more programs thereon, which, when executed by one or more processors, cause the one or more processors to implement the proteomic marker identification method described in any of the implementations of the first aspect above.

[0062] Fourthly, the present invention provides a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the proteomic biomarker identification method described in any of the implementations of the first aspect above.

[0063] This invention has the following advantages: It adds one-hot encoded grouping labels to protein abundance data in mass spectrometry quantitative data and encodes the group information corresponding to the protein abundance data into binary vectors; it constructs a data augmentation model based on variational autoencoders, and uses the protein abundance data and one-hot encoded grouping labels to perform supervised training on the data augmentation model, using the decoder of the trained data augmentation model to generate new sample data; it performs principal component analysis and similarity analysis on the new sample data; based on the results of principal component analysis, it divides the new sample data with similarity reaching a set threshold into a group set; it identifies potential biomarkers using a specified method, and ranks the potential biomarkers according to their importance parameters to obtain multi-group exosomal proteomic biomarker identification results.

[0064] This invention is more suitable for data augmentation in exosome proteomics, avoiding gradient explosion during model training; it can retain more information, enabling the data augmentation model to capture the complex features of the data more comprehensively and improve overall performance; it can obtain a large number of new samples that match the potential features of the original data (similarity) and have a certain degree of diversity (difference), increasing the sample size and diversity, which greatly reduces the difficulty of downstream analysis; when processing exosome proteomics data and comparing multiple groups of samples, it significantly improves the robustness of the data and the accuracy of biomarker identification, overcoming the limitations of existing technologies and providing a more effective solution for the field of bioinformatics. Attached Figure Description

[0065] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0066] Figure 1 This is a schematic diagram of the proteomic biomarker identification method provided in the embodiments of the present invention;

[0067] Figure 2 This diagram illustrates the similarity analysis in the proteomic biomarker identification method provided in this embodiment of the invention. (a) shows the dimensionality reduction result of the real sample; (b) shows the dimensionality reduction result of the simulated sample generated during the 10th iteration of the data augmentation model training; (c) shows the dimensionality reduction result of the simulated sample generated during the 90th iteration of the data augmentation model training; and (d) shows the dimensionality reduction result of the simulated sample generated during the 180th iteration of the data augmentation model training.

[0068] Figure 3 This is a schematic diagram of the results of tear exosome protein markers for Parkinson's disease (compared to Alzheimer's disease) identified by the proteomic marker identification method provided in this embodiment of the invention.

[0069] Figure 4 The receiver operating characteristic curve of the tear exosomal protein marker results for Parkinson's disease (compared to Alzheimer's disease) identified by the proteomic marker identification method provided in the embodiments of the present invention;

[0070] Figure 5 This is a diagram of the proteomic biomarker identification system architecture provided in this embodiment of the invention;

[0071] Figure 6 This is an electronic device architecture diagram provided in an embodiment of the present invention. Detailed Implementation

[0072] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example 1

[0074] See Figure 1 Embodiment 1 of the present invention provides a method for identifying proteomic biomarkers, comprising the following steps:

[0075] S1. Obtain mass spectrometry quantitative data of exosome proteomics, the mass spectrometry quantitative data including protein abundance data of preset groups; perform data preprocessing on the mass spectrometry quantitative data, the data preprocessing including data quality control, data normalization and missing value imputation.

[0076] S2. Add uniquely coded group labels to the protein abundance data in the preprocessed mass spectrometry quantitative data, and encode the group information corresponding to the protein abundance data into a binary vector;

[0077] S3. Construct a data augmentation model based on a variational autoencoder, wherein the loss function of the data augmentation model includes a reconstruction error function and a KL divergence function; use the Adam optimizer to minimize the loss function of the data augmentation model in order to optimize the parameters of the data augmentation model;

[0078] S4. Using the protein abundance data after data preprocessing and the group labels of one-hot encoding, supervised training is performed on the data augmentation model so that the logarithmic variance of the encoder output of the data augmentation model is clipped to the range of (-10, 10), and new sample data is generated using the decoder of the trained data augmentation model.

[0079] S5. Perform principal component analysis on the new sample data generated by the decoder of the data augmentation model, and perform similarity analysis on the new sample data after principal component analysis;

[0080] S6. Based on the results of principal component analysis, new sample data with similarity reaching a set threshold are divided into a group set, and potential biomarkers are identified through the group set using differential expression analysis or random forest method. The potential biomarkers are ranked according to the fold difference or the importance parameters output by the specified data to obtain the identification results of exosomal proteomic biomarkers in multiple groups.

[0081] In this embodiment, in step S1, exosome proteomics mass spectrometry quantitative data is acquired, including protein abundance data from multiple groups (such as different disease sample groups and healthy groups). During data quality control of the mass spectrometry quantitative data, protein abundance data with missing values ​​in more than 2 / 3 of the samples are removed to ensure data reliability and stability. During data normalization of the mass spectrometry quantitative data, a specified normalization method is used to normalize the protein abundance data to eliminate systematic bias between samples. During missing value imputation of the mass spectrometry quantitative data, a specified data imputation model is used to impute missing values ​​in the normalized protein abundance data. Random forests can infer and impute missing values ​​based on data characteristics, improving data integrity.

[0082] The specified normalization method can be at least one of Cycloess, log2, median, mean, vsn, quantum, rlr, and gi. The Cycloess method is preferred.

[0083] Specifically, Cycloess (Cyclic Locally Weighted Scatterplot Smoothing) normalization is a technique for normalizing gene expression data to correct systematic biases. This method smooths the data through local regression (Loess) to remove technical variation while preserving biological variation. The basic principle is based on the MA plot, where M represents the log ratio and A represents the average log intensity. Iterative application of locally weighted regression (Loess) eliminates the dependency between M and A.

[0084] The following are the detailed steps and formulas for the Cycloess normalization method:

[0085] 1) For each pair of samples i and j, calculate the M and A values:

[0086] M = log2(X i ) - log2(X j )

[0087] A = (log2(X i ) + log2(X j )) / 2

[0088] Among them, X i and X j These represent the expression values ​​of samples i and j, respectively;

[0089] 2) Apply Loess regression to the MA plot to obtain the fitted curve f(A); the Loess regression formula is as follows: Loess regression uses locally weighted multinomial regression. Loess regression fits a local multinomial to each point (x0, y0) in the dataset. During the fitting process, points closer to (x0, y0) receive larger weights, while points farther away receive smaller weights. The fitting formula is:

[0090] y = β0 + β1(x - x0) + β2(x - x0) 2 + ... + β p (x - x0) p

[0091] Where p typically takes the value 1 or 2; β0, β1, β2 ... β p These are regression coefficients, estimated using weighted least squares. A cubic function is typically used as the weighting function.

[0092] w(x) = (1 - |x| 3 ) 3 When |x| < 1;

[0093] w(x) = 0 when |x| ≥ 1;

[0094] Here, x is the standardized distance, which is usually obtained by dividing the actual distance by a certain bandwidth parameter.

[0095] 3) Calculate the adjusted expression value:

[0096] X i_adjusted = X i * 2 (-f(A) / 2)

[0097] X j_adjusted = X j * 2 (f(A) / 2)

[0098] 4) Iterate in a loop, repeating steps 1)-3) until the trend in the MA chart disappears or the preset number of iterations is reached. The preset number of iterations is set to 3 in this application.

[0099] Thus, in this way, Loess regression can capture nonlinear relationships in the data, enabling the Cycloess method to effectively handle nonlinear biases in protein mass spectrometry data and enhance data consistency.

[0100] In this embodiment, in step S2, one-hot encoded group labels are added to the end of the protein abundance data after normalization and missing value filling, and the group information is encoded into a binary vector to facilitate subsequent data augmentation model training.

[0101] Because exosome proteomics data suffers from low consistency among duplicate samples and high noise levels, and the data itself is too complex to provide sufficient effective information, conventional unsupervised training methods cannot be used to directly train the model. Additional human guidance is necessary. Adding one-hot encoded grouping labels to the end of the normalized and missing value-impregnated protein abundance data can significantly improve the learning performance of the data augmentation model.

[0102] In this embodiment, in step S3, the architecture of the data augmentation model is as follows:

[0103] The encoder receives preprocessed protein abundance data and one-hot encoded group labels, using this high-dimensional data as input. The encoder maps the high-dimensional input data to a latent space, representing the distribution of latent variables in the data.

[0104] Latent space: The generation of latent variables is achieved by using the mean and variance of the encoder output and employing reparameterization techniques.

[0105] Decoder: Similar to the inverse process of encoder, it reconstructs the dataset from vectors in the latent space.

[0106] Since the purpose is to perform data augmentation rather than to directly perform a discrimination task, there is no need to test the data augmentation model's discriminative ability on new samples, and therefore there is no need to use unsupervised training. Essentially, this invention maximizes the data augmentation capability by sacrificing the model's discriminative ability.

[0107] Specifically, data augmentation models can learn the true characteristics of exosome proteomics data and reduce its dimensionality to a latent variable space. Samples in this space are represented as lower-dimensional vectors, minimizing noise interference. Based on this latent variable space, data augmentation models can generate a large number of new samples with high intra-group consistency, thereby improving the accuracy and robustness of differentially expressed protein identification.

[0108] The loss function of the data augmentation model includes the reconstruction error function and the Kullback-Leibler divergence function. The reconstruction error measures the difference between the data generated by the decoder and the original input data, using the mean squared error (MSE). The Kullback-Leibler divergence measures the difference between the distribution of the latent variables and the standard normal distribution.

[0109] Specifically, the expression for the reconstruction error function in the loss function of the data augmentation model is as follows:

[0110]

[0111] In the formula, recon_batch represents the reconstructed protein abundance data; data represents the original input protein abundance data; and recon_batch i This represents the reconstructed protein abundance data for the i-th protein; data i This represents the i-th protein abundance data from the original input.

[0112] Specifically, the expression for the KL divergence function in the loss function of the data augmentation model is as follows:

[0113]

[0114] In the formula, μ represents the mean of the latent space; D represents the dimension of the latent space; μ i σ represents the mean of the i-th dimension of the latent space. 2 i Let represent the variance of the i-th dimension of the latent space.

[0115] Adding the reconstruction error and the KL divergence loss together yields the complete ELBO (Evidence Lower Bound) loss:

[0116]

[0117] By minimizing the loss function using the Adam optimizer, the parameters of the data augmentation model are optimized, enabling the data augmentation model to effectively generate samples from the latent space that are similar to the distribution of the input data.

[0118] In this embodiment, in step S4, protein abundance data that has undergone quality control, missing value imputation and standardization are horizontally spliced ​​with one-hot encoded group labels to perform supervised training of the data augmentation model, and the logarithmic variance of the encoder output is clipped to the range of (-10, 10).

[0119] Specifically, due to the complex characteristics of exosome proteomics data, there are a large number of noise or outliers. These values ​​may generate unstable gradients during training, especially when the magnitude of noise or outliers is much larger than that of normal data. They can cause a significant increase in gradients during backpropagation, ultimately leading to gradient explosion. Gradient explosion refers to the gradual increase in gradients during backpropagation, even approaching infinity. This results in excessively large weight updates, making the model unstable during training. Gradient explosion can cause model weights to diverge rapidly after a few updates, preventing the model from learning effective feature representations. By taking the nearest boundary value for any log-variance values ​​outside the range (-10, 10), while leaving values ​​within the range unchanged, the magnitude of the gradient can be effectively controlled. This helps stabilize the training process of data augmentation models and prevents them from becoming too large and causing gradient explosion.

[0120] In many bioinformatics studies, the encoder of a data augmentation model is directly used to reduce the dimensionality of the data and infer the features (proteins) that contribute significantly. This usually requires the latent space dimension output by the encoder to be consistent with the number of groups. The data augmentation model in this invention does not use an encoder directly, but rather a decoder, allowing the encoder to output a higher latent space dimension, unrestricted by the number of groups. This design retains more information, enabling the data augmentation model to more comprehensively capture the complex features of the data and improve overall performance. Furthermore, when the data augmentation model's learning performance is poor, the dimension of the latent space can be dynamically adjusted, providing greater flexibility for downstream applications.

[0121] Specifically, using a pre-trained data augmentation model's decoder to generate a large number of new samples makes downstream analytical methods requiring a large sample size possible. Because the data augmentation model learns the features of the data and maps them to a latent variable space, the encoder can generate new samples that match the latent features of the original data based on this space. During generation, latent variables are randomly sampled from a prior distribution that follows a standard normal distribution, introducing diversity into the newly generated samples. Therefore, by using a pre-trained data augmentation model's decoder to generate a large number of new samples, we can obtain a large number of new samples that match the latent features of the original data (similarity) and possess a certain degree of diversity (difference). The increased sample size and diversity significantly reduce the difficulty of downstream analysis.

[0122] In this embodiment, during the principal component analysis of the new sample data generated by the decoder of the data augmentation model in step S5: the covariance matrix of the protein abundance data is calculated, the eigenvalues ​​and eigenvectors of the covariance matrix are solved, the eigenvectors corresponding to the covariance matrix that reach the set eigenvalues ​​are selected as principal components, and the protein abundance numbers are projected onto the determined principal components; during the similarity analysis of the new sample data after principal component analysis: the similarity between samples of the specified groups is evaluated based on the data point distribution after dimensionality reduction by principal component analysis.

[0123] Specifically, observe the distribution of data points after dimensionality reduction via principal component analysis to assess the similarity between samples from different groups. The closer the data points are on the plane, the higher the similarity. See also... Figure 2 This example demonstrates the application of data augmentation models to tear exosome proteomics samples. Principal component analysis was used to reduce the dimensionality of the omics data. Figure 2 Part (a) shows the dimensionality reduction results of the real samples. The samples from different groups are mixed together and cannot be distinguished, indicating that the data pattern is complex and noisy. Figure 2 Part (b) shows the dimensionality reduction result of the simulated samples generated by the data augmentation model when it has been trained to the 10th iteration. At this point, the model is still unable to distinguish between the samples well. Figure 2 Part (c) shows the dimensionality reduction results of the simulated samples generated by the data augmentation model when it has been trained to the 90th iteration. At this point, the model gradually shows a tendency to distinguish between different groups of samples. Figure 2 Part (d) shows the dimensionality reduction results of the simulated samples generated by the data augmentation model when it is trained to the 180th iteration. At this point, the relative proximity between the samples in each group is consistent with experience, indicating that the data augmentation model can learn the latent representation of the samples and output new samples with a certain degree of diversity after augmentation.

[0124] In this embodiment, in step S6, based on the results of principal component analysis, several groups with high similarity can be defined as a group set. First, the biomarkers of several groups with low similarity are identified, and then the biomarkers of each group within the set are identified. Differential expression analysis is performed on the generated new samples, or a specified data imputation model is used to identify important proteins, in order to obtain proteins with large inter-group differences or proteins that contribute significantly to phenotypic classification. The potential biomarkers are then sorted according to their fold change or the importance output by the specified data imputation model to obtain a list of potential biomarkers for further consideration and validation.

[0125] The specified data imputation model can be at least one of the following: random forest model, fixed value imputation, mean / median / mode imputation, K nearest neighbor (KNN) method, regression-based imputation (such as linear regression, random forest regression), genotype imputation (specifically for genotype data), interpolation method (such as linear interpolation, polynomial interpolation), and multiple imputation.

[0126] See Figure 3 Additional samples were tested using an enzyme-linked immunosorbent assay (ELISA) to identify tear exosome protein markers for Parkinson's disease (as opposed to Alzheimer's disease) identified by this method. Figure 3 The p-value represents the confidence level of the T-tailed test, and HSGP2 is the gene name of this protein marker. The results showed that HSGP2 in tear exosomes differed significantly between Parkinson's disease and Alzheimer's disease samples.

[0127] See Figure 4 , used Figure 3 The test results were used to plot the receiver operating characteristic (ROC) curve, which illustrates the biomarker's ability to distinguish between the two diseases. AUC represents the area under the ROC curve, with a value closer to 1 indicating better distinguishing performance.

[0128] The effectiveness of the method was verified by applying it to tear exosome proteomic data from five neurological diseases and healthy individuals. The performance of the new method was evaluated by comparing biomarkers identified from new samples generated based on augmented simulated datasets with those identified directly from real samples. The results show that the data augmentation model can effectively learn the latent representation of mass spectrometry data and reflects the similarities between diseases through the generated new samples. The theoretical performance of biomarkers generated based on the data augmented simulated dataset is superior to that of biomarkers identified directly from real samples, demonstrating the superiority of the data augmentation module.

[0129] In summary, this invention acquires exosome proteomic mass spectrometry quantitative data, which includes protein abundance data for preset groups; performs data preprocessing on the mass spectrometry quantitative data, including data quality control, data normalization, and missing value imputation; adds one-hot encoded group labels to the protein abundance data in the preprocessed mass spectrometry quantitative data, and encodes the group information corresponding to the protein abundance data into binary vectors; constructs a data augmentation model based on variational autoencoder, the loss function of which includes a reconstruction error function and a KL divergence function; and uses Adam... The optimizer minimizes the loss function of the data augmentation model to optimize its parameters. Supervised training is performed on the data augmentation model using the preprocessed protein abundance data and one-hot encoded group labels, clipping the log-variance of the encoder output to the range (-10, 10). New sample data is then generated using the trained decoder. Principal component analysis (PCA) is performed on the new sample data generated by the decoder, followed by similarity analysis. Based on the PCA results, new sample data with similarity reaching a set threshold are grouped into a set of groups. Potential biomarkers are identified using differential expression analysis or random forest methods within these groups. The potential biomarkers are ranked according to fold change or the importance parameter of the model output, resulting in multi-group exosomal proteomic biomarker identification results. This invention is more suitable for data augmentation in exosome proteomics, avoiding gradient explosion during model training; it can retain more information, enabling the data augmentation model to capture the complex features of the data more comprehensively and improve overall performance; it can obtain a large number of new samples that match the potential features of the original data (similarity) and have a certain degree of diversity (difference), increasing the sample size and diversity, which greatly reduces the difficulty of downstream analysis; when processing exosome proteomics data and comparing multiple groups of samples, it significantly improves the robustness of the data and the accuracy of biomarker identification, overcoming the limitations of existing technologies and providing a more effective solution for the field of bioinformatics.

[0130] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0131] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0132] Example 2

[0133] See Figure 5 Embodiment 2 of the present invention also provides a proteomic biomarker identification system, comprising:

[0134] The data preparation module 100 is used to add uniquely coded group labels to the protein abundance data in the mass spectrometry quantitative data and encode the group information corresponding to the protein abundance data into a binary vector.

[0135] The data augmentation model processing module 200 is used to construct a data augmentation model based on variational autoencoder, and to perform supervised training on the data augmentation model using the protein abundance data and one-hot encoded group labels, and to generate new sample data using the decoder of the trained data augmentation model.

[0136] The group set generation module 300 is used to perform principal component analysis and similarity analysis on the new sample data; based on the results of the principal component analysis, the new sample data with similarity reaching a set threshold are divided into a group set;

[0137] The biomarker identification module 400 is used to identify potential biomarkers using a specified method, sort the potential biomarkers according to their importance parameters, and obtain multi-group exosomal proteomic biomarker identification results.

[0138] In this embodiment, a data preprocessing module 500 is also included, which is used to preprocess the mass spectrometry quantitative data. The data preprocessing includes data quality control, data normalization and missing value imputation.

[0139] During the data quality control process, protein abundance data with missing values ​​are removed from more than a specified number of samples.

[0140] During the data normalization process, a specified normalization method is used to normalize the protein abundance data;

[0141] During the missing value imputation process, a specified data imputation model is used to imput missing values ​​in the protein abundance data after data normalization.

[0142] In this embodiment, the data augmentation model processing module 200 includes:

[0143] The loss function of the data augmentation model includes a reconstruction error function and a KL divergence function; the Adam optimizer is used to minimize the loss function of the data augmentation model in order to optimize the parameters of the data augmentation model;

[0144] The expression for the reconstruction error function in the loss function of the data augmentation model is as follows:

[0145]

[0146] In the formula, recon_batch represents the reconstructed protein abundance data; data represents the original input protein abundance data; and recon_batch i This represents the reconstructed protein abundance data for the i-th protein; data i This represents the i-th protein abundance data from the original input.

[0147] The expression for the KL divergence function in the loss function of the data augmentation model is as follows:

[0148]

[0149] In the formula, μ represents the mean of the latent space; D represents the dimension of the latent space; μ i σ represents the mean of the i-th dimension of the latent space. 2 i Let represent the variance of the i-th dimension of the latent space.

[0150] In the data augmentation model processing module 200, during the supervised training of the data augmentation model, the logarithmic variance of the encoder output of the data augmentation model is clipped to the range of (-10, 10).

[0151] In this embodiment, the group set generation module 300 includes:

[0152] Calculate the covariance matrix of protein abundance data, solve for the eigenvalues ​​and eigenvectors of the covariance matrix, select the eigenvectors corresponding to the covariance matrix that reach the set eigenvalues ​​as principal components, and project the protein abundance numbers onto the determined principal components.

[0153] In the process of similarity analysis of new sample data after principal component analysis: based on the distribution of data points after dimensionality reduction by principal component analysis, the similarity between samples of a specified group is evaluated.

[0154] In this embodiment, the marker recognition module 400 includes:

[0155] Potential biomarkers are identified using differential expression analysis or random forest methods within the group set. The potential biomarkers are then ranked according to the fold change or the importance parameters output by the model, resulting in the identification results of exosomal proteomic biomarkers across multiple groups.

[0156] It should be noted that the information interaction and execution process between the modules of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0157] Example 3

[0158] refer to Figure 6 This document illustrates a structural schematic of an electronic device (e.g., a computing device) 600 suitable for implementing some embodiments of the present disclosure. The electronic devices in some embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0159] like Figure 6 As shown, electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 601, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 602 or a program loaded from storage device 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of electronic device 600. The processing device 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0160] Typically, the following devices can be connected to I / O interface 605: input devices 603 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic device 600 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6An electronic device 600 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 6 Each box shown can represent a device or multiple devices as needed.

[0161] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0162] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0163] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0164] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0165] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0166] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor can be described as including: a first acquisition unit, a determination unit, a training unit, a second acquisition unit, a fine-tuning unit, and an input unit. The names of these units do not necessarily limit the specific unit; for example, the input unit can also be described as "a unit that inputs the synthetic information to be detected into the aforementioned target intelligent synthetic information recognition model to obtain the synthetic information detection result."

[0167] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0168] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for identifying proteomic biomarkers, characterized in that, include: One-hot encoded grouping labels are added to the protein abundance data in the mass spectrometry quantitative data, and the group information corresponding to the protein abundance data is encoded into a binary vector; A data augmentation model based on variational autoencoder is constructed. The protein abundance data and one-hot encoded group labels are used to conduct supervised training on the data augmentation model. The decoder of the trained data augmentation model is then used to generate new sample data. Principal component analysis and similarity analysis are performed on the new sample data; based on the results of the principal component analysis, the new sample data with similarity reaching a set threshold are divided into a group set; Potential biomarkers are identified using differential expression analysis or random forest methods within the group set. The potential biomarkers are then ranked according to the fold change or the importance parameters output by the model, resulting in the identification results of exosomal proteomic biomarkers across multiple groups.

2. The method for identifying proteomic biomarkers according to claim 1, characterized in that, It also includes data preprocessing of the mass spectrometry quantitative data, including data quality control, data normalization, and missing value imputation.

3. The method for identifying proteomic biomarkers according to claim 2, characterized in that, During the data quality control process, protein abundance data with missing values ​​are removed from more than a specified number of samples. During the data normalization process, a specified normalization method is used to normalize the protein abundance data; During the missing value imputation process, a specified data imputation model is used to imput missing values ​​in the protein abundance data after data normalization.

4. The method for identifying proteomic biomarkers according to claim 1, characterized in that, The loss function of the data augmentation model includes a reconstruction error function and a KL divergence function; the optimizer is used to minimize the loss function of the data augmentation model in order to optimize the parameters of the data augmentation model; The expression for the reconstruction error function in the loss function of the data augmentation model is as follows: ； In the formula, recon_batch represents the reconstructed protein abundance data; data represents the original input protein abundance data; and recon_batch i This represents the reconstructed protein abundance data for the i-th protein; data i This represents the i-th protein abundance data from the original input; The expression for the KL divergence function in the loss function of the data augmentation model is as follows: ； In the formula, μ represents the mean of the latent space; D represents the dimension of the latent space; μ i σ represents the mean of the i-th dimension of the latent space. 2 i Let represent the variance of the i-th dimension of the latent space.

5. The method for identifying proteomic biomarkers according to claim 1, characterized in that, During the supervised training of the data augmentation model, the logarithmic variance of the encoder output of the data augmentation model is clipped to the range of (-10, 10).

6. The method for identifying proteomic biomarkers according to claim 1, characterized in that, During the principal component analysis of the new sample data: Calculate the covariance matrix of protein abundance data, solve for the eigenvalues ​​and eigenvectors of the covariance matrix, select the eigenvectors corresponding to the covariance matrix that reach the set eigenvalues ​​as principal components, and project the protein abundance numbers onto the determined principal components. In the process of similarity analysis of new sample data after principal component analysis: based on the distribution of data points after dimensionality reduction by principal component analysis, the similarity between samples of a specified group is evaluated.

7. A proteomic biomarker identification system, employing the proteomic biomarker identification method according to any one of claims 1 to 6, characterized in that, include: The data preparation module is used to add uniquely coded group labels to the protein abundance data in the mass spectrometry quantitative data and encode the group information corresponding to the protein abundance data into a binary vector. The data augmentation model processing module is used to construct a data augmentation model based on variational autoencoder, and to perform supervised training on the data augmentation model using the protein abundance data and one-hot encoded group labels, and to generate new sample data using the decoder of the trained data augmentation model. The group set generation module is used to perform principal component analysis and similarity analysis on the new sample data; based on the results of the principal component analysis, the new sample data with similarity reaching a set threshold are divided into a group set; The biomarker identification module is used to identify potential biomarkers through the group set using differential expression analysis or random forest methods, and to rank the potential biomarkers according to the fold change or the importance parameters output by the model, so as to obtain the identification results of exosomal proteomics biomarkers in multiple groups.

8. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the proteomic biomarker identification method as described in any one of claims 1-6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the proteomic biomarker identification method as described in any one of claims 1-6.

Images