Western blot data classification system based on feature engineering

Abstract

The invention discloses an immunoblotting data classification system based on feature engineering, relates to the technical field of data processing, and solves the problem that whether a protein band exists in an original immunoblotting image is difficult to identify; secondly, feature parameters of an original immunoblotting image and immunoblotting data are difficult to extract; secondly, a feature matrix is difficult to generate according to the feature parameters, and data classification is difficult to carry out after feature subsets are screened out according to a selection algorithm; and finally, the characteristic parameters in the characteristic subsets are difficult to comprehensively analyze. The system comprises the following modules: a data acquisition module, a data processing module, a feature extraction module, a feature processing module, a data classification module and a data analysis module. Through feature engineering and a classification model, classification of immune data is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of data processing, and in particular is an immunoblotting data classification system based on feature engineering. Background Art

[0002] With the development of machine learning technology, immunoblotting is a method to detect a certain protein in complex samples based on antigen-antibody specific binding. Feature engineering is a key step in data mining and machine learning, which involves extracting useful features from raw data in order to better train models and improve prediction performance. In data classification, feature selection and extraction are crucial to the accuracy and efficiency of the model.

[0003] The existing problems are as follows: first, it is difficult to identify whether there are protein bands in the original immunoblot image; second, it is difficult to extract the characteristic parameters of the original immunoblot image and the immunoblot data; then, it is difficult to generate a feature matrix based on the characteristic parameters, and it is difficult to classify the data after screening out the feature subset based on the selection algorithm; finally, it is difficult to comprehensively analyze the feature parameters in the feature subset. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an immunoblot data classification system based on feature engineering, which is used to solve the following technical problems:

[0005] First, it is difficult to identify whether protein bands exist in the original immunoblot image; second, it is difficult to extract the characteristic parameters of the original immunoblot image and the immunoblot data; then, it is difficult to generate a feature matrix based on the characteristic parameters, and difficult to classify the data after screening out the feature subset based on the selection algorithm; finally, it is difficult to comprehensively analyze the feature parameters in the feature subset.

[0006] To solve the above problems, the first aspect of the present invention provides an immunoblot data classification system based on feature engineering, comprising the following modules:

[0007] Data acquisition module: collects original immunoblot images and immunoblot data;

[0008] Data processing module: preprocessing the original immunoblot image to remove noise and enhance the image; preprocessing the immunoblot data includes: cleaning and normalization;

[0009] Feature extraction module: identify whether there are protein bands based on the preprocessed original immunoblot image analysis; extract feature parameters of the original immunoblot image based on the analysis results; analyze and extract the relative expression of proteins based on the extracted feature parameters; perform feature extraction based on the preprocessed immunoblot data;

[0010] Feature processing module: Processing based on the feature parameters extracted from the original immunoblot images and immunoblot data, including: cleaning, noise removal and standardization;

[0011] Data classification module: Generate a feature matrix based on the processed feature parameters; filter out feature subsets through a selection algorithm based on the feature matrix; perform data classification based on the filtered feature subsets;

[0012] Data analysis module: Comprehensively analyze the feature parameters in the feature subset to determine whether anomalies occur.

[0013] As a further solution of the present invention: identifying whether there is a protein band based on the pre-processed original immunoblot image analysis includes the following steps:

[0014] By using edge detection algorithms, the presence of protein band regions is detected and identified in the original immunoblot images;

[0015] When protein bands are identified in the original immunoblot image, the edge detection results are processed through morphological operations including: dilation, erosion, opening operation and closing operation to remove noise points, smooth the band edges and fill small holes in the bands;

[0016] Based on the results of edge detection and morphological processing, the protein bands are separated from the background through image segmentation algorithm, and the characteristic parameters of the protein bands are extracted according to the separation results;

[0017] When it is identified that the original immunoblot image does not contain a protein band, the immunoblot image is re-collected, and the protein band is detected in the re-collected image, and the above steps are repeated.

[0018] As a further solution of the present invention: extracting characteristic parameters of the original immunoblot image according to the analysis results comprises the following steps:

[0019] Separate the original immunoblot image according to the image segmentation algorithm, locate the position of the protein band and detect and count the number of protein bands through the connected domain analysis or contour detection algorithm, and extract the position characteristic parameters and quantity characteristic parameters of the protein band. The position characteristic parameters include: the starting coordinates of the protein band: (x start ,y start ) and the ending coordinates: (x final ,y final ); the quantitative characteristic parameters include: m protein bands;

[0020] According to the located protein bands, the intensity characteristic parameters of the a-th protein band are extracted by calculating the average gray value of the a-th protein band;

[0021] By analyzing the average gray value formula:

[0022] Get the intensity characteristic parameters of the ath protein band Among them, H i Represents the gray value of the i-th pixel, N a represents the total number of pixels in the a-th protein band region, a={1,2,...,m};

[0023] Extract the morphological characteristic parameters of the a-th protein band, including length, width and uniformity; wherein the length of the a-th protein band is obtained by tracking the band outline and calculating the distance between points on the outline, and the width of the a-th protein band is obtained by calculating the number of pixels of the band in the vertical direction; the uniformity of the a-th protein band is evaluated by calculating the coefficient of variation of the gray value;

[0024] By analyzing the coefficient of variation formula:

[0025] Get the uniformity U of the ath protein band a ,in represents the average gray value, i={1,2,...,N};

[0026] The shape characteristic parameters of the a-th protein band are extracted using a shape analysis algorithm, including perimeter and area. The perimeter of the a-th protein band is obtained by using a contour tracking algorithm, and the area of ​​the a-th protein band is obtained by counting the number of pixels inside the a-th protein band.

[0027] As a further solution of the present invention: analyzing and extracting the relative expression amount of the protein according to the extracted characteristic parameters, comprising the following steps:

[0028] By analyzing the formula:

[0029] Get the total intensity Q of the ath protein band a ,in represents the average gray value, A a represents the area of ​​the ath protein band;

[0030] By analyzing the formula:

[0031] The relative expression of the protein G is obtained, where m represents the total number of protein bands and Q a represents the total intensity of the ath protein band.

[0032] As a further solution of the present invention: feature extraction is performed based on the pre-processed immunoblot data, comprising the following steps:

[0033] The pre-processed immunoblot data include: electrophoresis separation data, membrane transfer data, blocking and antibody incubation data, color or luminescence signal data, sample data, experimental condition data and protein data;

[0034] The protein sample is separated by polyacrylamide gel electrophoresis, and the characteristic parameters of the electrophoretic separation data of the ath protein in the sample are extracted, including: mobility and relative position;

[0035] Extract the characteristic parameters of the transfer data during the transfer of proteins separated by electrophoresis to the solid support, including: transfer efficiency, time and buffer formula used;

[0036] After transfer, blocking buffer is used to inhibit nonspecific binding, and specific primary and secondary antibodies are added for incubation. The characteristic parameters of the blocking and antibody incubation data are extracted, including: the type and concentration of the blocking agent, the selection of primary and secondary antibodies, the dilution ratio, the incubation time and conditions;

[0037] Detecting the presence of the target protein by chemiluminescence, fluorescence or colorimetry, extracting characteristic parameters of the colorimetric or luminescent signal data including: the intensity, location, duration, type of substrate used and detection equipment parameters of the colorimetric or luminescent signal;

[0038] The characteristic parameters of the extracted sample data include: the source, processing method, and concentration of the sample;

[0039] The characteristic parameters of the experimental condition data extracted include: electrophoresis voltage, current, transfer voltage, temperature, time, batch number and expiration date of reagents;

[0040] The mass of the protein is measured by a mass spectrometer, and the characteristic parameter extracted from the protein data is the molecular weight of the protein.

[0041] As a further solution of the present invention: generating a feature matrix according to the processed feature parameters, comprising the following steps:

[0042] According to the processed characteristic parameters, the characteristic vector is generated by integration; for the a-th protein band, the characteristic vector is expressed as: f a,b =[f a,1 ,f a,2 ,…,f a,n ], where f a,b represents the b-th characteristic parameter value of the a-th protein band, a={1,2,...,m}, b={1,2,...,n}, n represents the number of characteristic parameters, and m represents the number of protein bands;

[0043] Combine the feature vectors of m protein bands into an m×n feature matrix F;

[0044] The feature matrix is ​​expressed as: Where m is the number of rows representing the number of protein bands, and n is the number of columns representing the number of feature parameters.

[0045] As a further solution of the present invention: according to the feature matrix, a feature subset is screened out by a selection algorithm, comprising the following steps:

[0046] Feature selection is performed based on the Lasso regression model. The complexity and penalty term of the Lasso regression model are constrained by introducing the L1 regularization term parameter.

[0047] The feature parameters in the feature matrix are used as input variables and input into the Lasso regression model for training. During the training process, the Lasso regression model automatically penalizes the absolute value of the coefficient of the feature parameter and compresses the coefficient according to the L1 regularization term parameter.

[0048] According to the trained Lasso regression model, the coefficients of the feature parameters in the Lasso regression model are obtained, and the feature parameters are sorted from large to small according to the absolute values ​​of the coefficients; by setting a threshold, the feature parameters of important features whose absolute values ​​of the coefficients are greater than the threshold are screened out; wherein, when the absolute value of the coefficient is equal to 0, the feature parameter is an unimportant feature, and when the absolute value of the coefficient is greater than 0, the feature parameter is an important feature, and the feature parameters of the important features are retained;

[0049] According to the feature parameters retained by the Lasso regression model, the corresponding feature parameters are extracted and screened from the feature matrix to generate a feature subset.

[0050] As a further solution of the present invention: data classification is performed according to the screened feature subset, comprising the following steps:

[0051] Using the support vector machine as a classification model, annotating feature parameters and corresponding classification labels in the feature subset screened by the selection algorithm; the classification labels include: different protein expression categories or subcellular location categories of proteins in immunoblot images;

[0052] Input the labeled data into the support vector machine model for training to learn the mapping relationship between feature parameters and classification labels;

[0053] New immunoblot images and new immunoblot data are collected in real time according to an immunoblot data classification system based on feature engineering. The new immunoblot images and new immune data are subjected to the same feature extraction and preprocessing steps as the original immunoblot images and immune data, and new feature parameters of the new immunoblot images and new immune data are extracted. The processed new feature parameters are input into a trained model. The support vector machine model automatically classifies the newly input new feature parameters according to the learned mapping relationship and outputs the classification results to obtain different protein expression categories or subcellular location categories of proteins in the new immunoblot images.

[0054] As a further solution of the present invention: comprehensively analyzing the characteristic parameters in the characteristic subset to determine whether an abnormality occurs includes the following steps:

[0055] According to the feature subset screened by the selection algorithm, the feature subset is expressed as: Where x is the number of rows representing the number of protein bands, and y is the number of columns representing the number of important feature parameters retained;

[0056] By analyzing the formula:

[0057] Get the statistic S of the feature parameters in the feature subset; where f i,j represents the characteristic parameter value of the i-th row and j-th column, x is the row number representing the number of protein bands, y is the column number representing the number of important characteristic parameters retained, i = {1, 2, ..., x}, j = {1, 2, ..., y};

[0058] By analyzing the formula: β=|f i,j -S|

[0059]

[0060] when When , it means that the feature parameters in the feature subset have abnormal values, otherwise, the feature parameters in the feature subset have no abnormal values.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] The present invention realizes the automated analysis of immunoblotting data by analyzing the pre-processed immunoblotting images, identifying protein bands, and extracting relevant characteristic parameters, thereby improving the efficiency and accuracy of the analysis.

[0063] The present invention realizes the automatic and intelligent classification of immunoblotting data by using a feature matrix and screening feature subsets for data classification;

[0064] The present invention optimizes classification performance through feature engineering, enhances the generalization ability of the model, and improves the accuracy and efficiency of classification;

[0065] The present invention comprehensively analyzes the characteristic parameters in the characteristic subset, timely discovers and determines whether there are abnormal values, and provides timely warning and intervention opportunities for scientific researchers. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0067] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0068] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] See also Figure 1 In a first aspect, an embodiment of the present invention provides an immunoblotting data classification system based on feature engineering, comprising the following modules:

[0070] Data acquisition module: collects original immunoblot images and immunoblot data;

[0071] Data processing module: preprocessing the original immunoblot image to remove noise and enhance the image; preprocessing the immunoblot data includes: cleaning and normalization;

[0072] Feature extraction module: identify whether there are protein bands based on the preprocessed original immunoblot image analysis; extract feature parameters of the original immunoblot image based on the analysis results; analyze and extract the relative expression of proteins based on the extracted feature parameters; perform feature extraction based on the preprocessed immunoblot data;

[0073] Feature processing module: Processing based on the feature parameters extracted from the original immunoblot images and immunoblot data, including: cleaning, noise removal and standardization;

[0074] Data classification module: Generate a feature matrix based on the processed feature parameters; filter out feature subsets through a selection algorithm based on the feature matrix; perform data classification based on the filtered feature subsets;

[0075] Data analysis module: Comprehensively analyze the feature parameters in the feature subset to determine whether anomalies occur.

[0076] Specifically, use a high-resolution camera or scanner to capture the immunoblot image to ensure clear image quality and uniform background. Obtain numerical data related to immunoblotting from experimental equipment or databases to ensure the accuracy and completeness of the data. Use filters to remove noise in the image and adjust image parameters such as contrast and brightness to make the protein bands clearer. Remove duplicate, invalid or missing data and scale the data to the same scale for subsequent analysis. Use edge detection algorithms to identify protein bands in the image. Extract parameters such as the width, height, area, and grayscale value of the band, and calculate the relative expression of the protein based on the grayscale value or area of ​​the band. Extract relevant features from the preprocessed immunoblot data. Cleaning feature parameters includes: removing redundant, irrelevant or noise features. Use mean filtering to further remove noise in the features and scale the feature parameters to the same scale for subsequent analysis. Combine the processed feature parameters into a feature matrix, and use an algorithm to screen out the most representative feature subsets. Use a classification algorithm to classify the feature subsets. Analyze the feature parameters in the feature subsets, and use statistical methods to further mine the information in the data to determine whether there are any abnormalities.

[0077] In one embodiment of the present invention, the analysis of the original immunoblot image after pre-processing to identify whether a protein band exists comprises the following steps:

[0078] By using edge detection algorithms, the presence of protein band regions is detected and identified in the original immunoblot images;

[0079] When protein bands are identified in the original immunoblot image, the edge detection results are processed through morphological operations including: dilation, erosion, opening operation and closing operation to remove noise points, smooth the band edges and fill small holes in the bands;

[0080] Based on the results of edge detection and morphological processing, the protein bands are separated from the background through image segmentation algorithm, and the characteristic parameters of the protein bands are extracted according to the separation results;

[0081] When it is identified that the original immunoblot image does not contain a protein band, the immunoblot image is re-collected, and the protein band is detected in the re-collected image, and the above steps are repeated.

[0082] Specifically, according to the characteristics of the image, a suitable edge detection algorithm is selected, including: Canny edge detection, Sobel edge detection or Laplacian edge detection, etc. The selected edge detection algorithm is applied to the preprocessed original immunoblot image to detect the edge information in the image. Identify the protein band region: According to the results of edge detection, the protein band region that may exist in the image is identified. This usually involves a connectivity analysis of the detected edges to determine which edges belong to the same protein band. The identified protein band region is expanded to expand the range of the band to make it clearer. Then, an erosion operation is performed to remove the small noise points on the edge of the band. An opening operation is applied, first erosion and then expansion to further smooth the edge of the band and remove small isolated areas. A closing operation is applied, first expansion and then erosion to fill the small holes in the band to make it more complete. According to the characteristics of the image after morphological processing, the image segmentation algorithm is selected, including: threshold segmentation, region growing or level set method, etc. The selected image segmentation algorithm is applied to the image after morphological processing to separate the protein band from the background. Determine whether there are protein bands: Based on the results of image segmentation and feature extraction, determine whether there are protein bands in the original immunoblot image. If it is identified that there are no protein bands in the original image or the bands are unclear, re-acquire the immunoblot image. Repeat the above steps for the re-acquired image until clear protein bands are successfully identified and valid feature parameters are extracted.

[0083] In one embodiment of the present invention, extracting characteristic parameters of the original immunoblot image according to the analysis result comprises the following steps:

[0084] Separate the original immunoblot image according to the image segmentation algorithm, locate the position of the protein band and detect and count the number of protein bands through the connected domain analysis or contour detection algorithm, and extract the position characteristic parameters and quantity characteristic parameters of the protein band. The position characteristic parameters include: the starting coordinates of the protein band: (x start ,y start ) and the ending coordinates: (x final ,y final ); the quantitative characteristic parameters include: m protein bands;

[0085] According to the located protein bands, the intensity characteristic parameters of the a-th protein band are extracted by calculating the average gray value of the a-th protein band;

[0086] By analyzing the average gray value formula:

[0087] Get the intensity characteristic parameters of the ath protein band Among them, H iRepresents the gray value of the i-th pixel, N a represents the total number of pixels in the a-th protein band region, a={1,2,...,m};

[0088] Extract the morphological characteristic parameters of the a-th protein band, including length, width and uniformity; wherein the length of the a-th protein band is obtained by tracking the band outline and calculating the distance between points on the outline, and the width of the a-th protein band is obtained by calculating the number of pixels of the band in the vertical direction; the uniformity of the a-th protein band is evaluated by calculating the coefficient of variation of the gray value;

[0089] By analyzing the coefficient of variation formula:

[0090] Get the uniformity U of the ath protein band a ,in represents the average gray value, i={1,2,...,N};

[0091] The shape characteristic parameters of the a-th protein band are extracted using a shape analysis algorithm, including perimeter and area. The perimeter of the a-th protein band is obtained by using a contour tracking algorithm, and the area of ​​the a-th protein band is obtained by counting the number of pixels inside the a-th protein band.

[0092] Specifically, first, the target area in the original immunoblot image, i.e., the area where the protein bands are located, is separated from the background area using an image segmentation algorithm. Next, the specific position of each protein band is located by connected domain analysis or contour detection algorithm. This process can accurately identify the starting and ending coordinates of the bands, thereby extracting the position characteristic parameters. After all protein bands are located, the number of bands is counted, recorded as m protein bands, as the quantitative characteristic parameters. For each located protein band, its average grayscale value is calculated. This step is completed by adding the grayscale values ​​of all pixels in the band area and then dividing it by the total number of pixels. The obtained average grayscale value is used as the intensity characteristic parameter of the protein band, reflecting the overall brightness or signal intensity of the band. The length of the band can be obtained by tracing the outline of the band and calculating the distance between each point on the outline. In the direction perpendicular to the long axis of the band, the number of pixels occupied by the band is counted to obtain the width of the band. The uniformity of the band is evaluated by calculating the coefficient of variation of the grayscale value. The coefficient of variation is the ratio of the standard deviation to the average grayscale value, which reflects the degree of dispersion of the grayscale value of the strip. The smaller the coefficient of variation, the more uniform the grayscale value of the strip. Use the contour tracking algorithm to track the border of the strip to calculate the perimeter of the strip. By counting the number of pixels inside the strip, the area of ​​the strip can be obtained.

[0093] In one embodiment of the present invention, analyzing and extracting the relative expression amount of the protein according to the extracted characteristic parameters comprises the following steps:

[0094] By analyzing the formula:

[0095] Get the total intensity Q of the ath protein band a ,in represents the average gray value, A a represents the area of ​​the ath protein band;

[0096] By analyzing the formula:

[0097] The relative expression of the protein G is obtained, where m represents the total number of protein bands and Q a represents the total intensity of the ath protein band.

[0098] Specifically, the average gray value and area of ​​the a-th protein band are obtained from the extracted characteristic parameters. For each protein band, the total intensity of the a-th protein band is calculated using the product of its average gray value and area. The total number of protein bands is obtained from the extracted characteristic parameters. All protein bands are traversed, the total intensity of each band is calculated, and they are added together to obtain the sum of the total intensities. The relative expression amount G of the a-th protein band is calculated using the sum of the total intensities of all protein bands as the denominator and the total intensity of each protein band as the numerator.

[0099] In one embodiment of the present invention, feature extraction is performed based on the preprocessed immunoblot data, comprising the following steps:

[0100] The pre-processed immunoblot data include: electrophoresis separation data, membrane transfer data, blocking and antibody incubation data, color or luminescence signal data, sample data, experimental condition data and protein data;

[0101] The protein sample is separated by polyacrylamide gel electrophoresis, and the characteristic parameters of the electrophoretic separation data of the ath protein in the sample are extracted, including: mobility and relative position;

[0102] Extract the characteristic parameters of the transfer data during the transfer of proteins separated by electrophoresis to the solid support, including: transfer efficiency, time and buffer formula used;

[0103] After transfer, blocking buffer is used to inhibit nonspecific binding, and specific primary and secondary antibodies are added for incubation. The characteristic parameters of the blocking and antibody incubation data are extracted, including: the type and concentration of the blocking agent, the selection of primary and secondary antibodies, the dilution ratio, the incubation time and conditions;

[0104] Detecting the presence of the target protein by chemiluminescence, fluorescence or colorimetry, extracting characteristic parameters of the colorimetric or luminescent signal data including: the intensity, location, duration, type of substrate used and detection equipment parameters of the colorimetric or luminescent signal;

[0105] The characteristic parameters of the extracted sample data include: the source, processing method, and concentration of the sample;

[0106] The characteristic parameters of the experimental condition data extracted include: electrophoresis voltage, current, transfer voltage, temperature, time, batch number and expiration date of reagents;

[0107] The mass of the protein is measured by a mass spectrometer, and the characteristic parameter extracted from the protein data is the molecular weight of the protein.

[0108] Specifically, the protein samples were separated using polyacrylamide gel electrophoresis. The electrophoretic separation data were extracted, focusing on the electrophoretic mobility and relative position of the ath protein. The mobility reflects the speed of protein movement in the electric field and is related to factors such as the molecular weight and charge of the protein. The relative position provides specific position information of the protein on the gel. The transfer data of the protein after electrophoretic separation was extracted and transferred to the solid support PVDF membrane. The characteristic parameters include transfer efficiency, transfer time and the buffer formula used. The transfer efficiency reflects the efficiency of protein transfer from gel to membrane and is an important indicator for evaluating the transfer effect; the transfer time and buffer formula may affect the transfer efficiency and protein stability. After transfer, a blocking buffer was used to inhibit nonspecific binding, and specific primary and secondary antibodies were added for incubation. The characteristic parameters of the blocking and antibody incubation data were extracted, including the type and concentration of the blocking agent, the selection of primary and secondary antibodies, the dilution ratio, the incubation time and conditions. These parameters directly affect the binding efficiency and specificity of the antibody and protein. The presence of the target protein is detected by chemiluminescence, fluorescence or colorimetry, and the characteristic parameters of the color or luminescence signal data are extracted, including the intensity, position, duration, substrate type used and detection equipment parameters of the color or luminescence signal. The signal intensity reflects the expression level of the target protein, and the position provides the specific location information of the target protein on the membrane. The duration and substrate type may affect the stability and sensitivity of the signal. The detection equipment parameters may affect the acquisition and analysis of the signal. The characteristic parameters of the sample data are extracted, including the source, treatment method, concentration, etc. of the sample. The characteristic parameters of the experimental condition data are extracted, including electrophoresis voltage, current, transfer voltage, temperature, time, batch number and expiration date of the reagent. These parameters may affect the accuracy and repeatability of the experimental results. The mass spectrometer is used for detection, and the actual spectrum is compared with the theoretical primary mass spectrum peak and secondary mass spectrum peak generated after the protein is digested by protease, and the protein identification and molecular weight measurement are performed. The characteristic parameters of the extracted protein data are the molecular weight of the protein.

[0109] In one embodiment of the present invention, generating a feature matrix according to the processed feature parameters includes the following steps:

[0110] According to the processed characteristic parameters, the characteristic vector is generated by integration; for the a-th protein band, the characteristic vector is expressed as: f a,b =[f a,1 ,f a,2 ,…,f a,n ], where f a,b represents the b-th characteristic parameter value of the a-th protein band, a={1,2,...,m}, b={1,2,...,n}, n represents the number of characteristic parameters, and m represents the number of protein bands;

[0111] Combine the feature vectors of m protein bands into an m×n feature matrix F;

[0112] The feature matrix is ​​expressed as: Where m is the number of rows representing the number of protein bands, and n is the number of columns representing the number of feature parameters.

[0113] Specifically, the processed characteristic parameters are collected and sorted, including but not limited to electrophoretic mobility, relative position, transfer efficiency, blocking and antibody incubation conditions, color development or luminescence signal intensity, sample information, experimental conditions and protein molecular weight. For the a-th protein band, a characteristic vector is constructed based on all its characteristic parameter values. In this vector, the value range of a is 1 to m, indicating the number of protein bands; the value range of b is 1 to n, indicating the number of characteristic parameters. The characteristic vectors of all m protein bands are combined in order into a matrix with a dimension of m×n. Each row of this matrix represents the characteristic vector of a protein band, and each column represents the value of a characteristic parameter in all protein bands, where m is the number of rows, indicating the number of protein bands; n is the number of columns, indicating the number of characteristic parameters.

[0114] In one embodiment of the present invention, according to the feature matrix, a feature subset is screened out by a selection algorithm, comprising the following steps:

[0115] Feature selection is performed based on the Lasso regression model. The complexity and penalty term of the Lasso regression model are constrained by introducing the L1 regularization term parameter.

[0116] The feature parameters in the feature matrix are used as input variables and input into the Lasso regression model for training. During the training process, the Lasso regression model automatically penalizes the absolute value of the coefficient of the feature parameter and compresses the coefficient according to the L1 regularization term parameter.

[0117] According to the trained Lasso regression model, the coefficients of the feature parameters in the Lasso regression model are obtained, and the feature parameters are sorted from large to small according to the absolute values ​​of the coefficients; by setting a threshold, the feature parameters of important features whose absolute values ​​of the coefficients are greater than the threshold are screened out; wherein, when the absolute value of the coefficient is equal to 0, the feature parameter is an unimportant feature, and when the absolute value of the coefficient is greater than 0, the feature parameter is an important feature, and the feature parameters of the important features are retained;

[0118] According to the feature parameters retained by the Lasso regression model, the corresponding feature parameters are extracted and screened from the feature matrix to generate a feature subset.

[0119] Specifically, according to the generated feature matrix, the Lasso regression model is selected as the feature selection method. Lasso regression can realize feature selection and coefficient sparseness by introducing L1 regularization terms to constrain the coefficients of the model. The feature parameters in the feature matrix are used as input variables, and the input variables are trained using the Lasso regression model. During the training process, Lasso regression minimizes the objective function, including: square loss and L1 regularization terms. Setting the regularization parameter to control the strength of the penalty term has an important impact on the result of feature selection. The optimal regularization parameter value can be selected through cross-validation. During the training process, the Lasso regression model automatically penalizes the coefficients of the feature parameters, compresses the coefficients of the feature parameters so that the coefficients of some features become zero. After the training is completed, check the coefficients of the feature parameters of the Lasso regression model. The coefficients reflect the contribution of each feature parameter. Sort the coefficients from large to small according to the absolute value of the coefficients, set the threshold to 0, and filter out important feature parameters whose absolute values ​​of the coefficients are greater than the threshold. When the absolute value of the coefficient is equal to 0, it is considered that the contribution of the feature parameter to the target variable is small and it is an unimportant feature and is eliminated. When the absolute value of the coefficient is greater than 0, it is considered that the feature parameter has a greater contribution to the target variable and is an important feature and is retained. According to all feature parameters whose coefficients are not 0 retained by the Lasso regression model, the corresponding feature parameters are extracted from the original feature matrix to form a feature subset.

[0120] In one embodiment of the present invention, data classification is performed according to the screened feature subset, including the following steps:

[0121] Using the support vector machine as a classification model, annotating feature parameters and corresponding classification labels in the feature subset screened by the selection algorithm; the classification labels include: different protein expression categories or subcellular location categories of proteins in immunoblot images;

[0122] Input the labeled data into the support vector machine model for training to learn the mapping relationship between feature parameters and classification labels;

[0123] New immunoblot images and new immunoblot data are collected in real time according to an immunoblot data classification system based on feature engineering. The new immunoblot images and new immune data are subjected to the same feature extraction and preprocessing steps as the original immunoblot images and immune data, and new feature parameters of the new immunoblot images and new immune data are extracted. The processed new feature parameters are input into a trained model. The support vector machine model automatically classifies the newly input new feature parameters according to the learned mapping relationship and outputs the classification results to obtain different protein expression categories or subcellular location categories of proteins in the new immunoblot images.

[0124] Specifically, the feature parameters in the selected feature subset are annotated, that is, a corresponding classification label is assigned to each feature parameter. The classification labels include but are not limited to different protein expression categories or subcellular location categories of proteins in immunoblot images. The annotated data is input into the support vector machine model for training. The support vector machine model learns the mapping relationship between feature parameters and classification labels, and divides the data into two parts by finding the hyperplane with the maximum margin, each of which is a class of its binary classification. For the case of linear inseparability, the support vector machine model maps the input from the input space to the feature space by introducing the kernel technique to achieve linear separability. During the training process, the support vector machine model optimizes the objective function to maximize the classification interval while satisfying the constraints. According to the training results, the parameters of the support vector machine model are adjusted, including: penalty parameters, kernel function parameters, etc., to optimize the model performance. The feature engineering-based immunoblot data classification system collects new immunoblot images and immunoblot data in real time. The new immunoblot images and new immunoblot data are subjected to the same feature extraction and preprocessing steps as the original immunoblot images and immunoblot data. The processed new feature parameters are input into the trained support vector machine model. The new feature parameters of the new input are automatically classified according to the learned mapping relationship. The model outputs the classification results, different protein expression categories or subcellular location categories of proteins in new immunoblot images.

[0125] In one embodiment of the present invention, comprehensively analyzing the feature parameters in the feature subset to determine whether an abnormality occurs includes the following steps:

[0126] According to the feature subset screened by the selection algorithm, the feature subset is expressed as: Where x is the number of rows representing the number of protein bands, and y is the number of columns representing the number of important feature parameters retained;

[0127] By analyzing the formula:

[0128] Get the statistic S of the feature parameters in the feature subset; where f i,jrepresents the characteristic parameter value of the i-th row and j-th column, x is the row number representing the number of protein bands, y is the column number representing the number of important characteristic parameters retained, i = {1, 2, ..., x}, j = {1, 2, ..., y};

[0129] By analyzing the formula: β=|f i,j -S|

[0130]

[0131] when When , it means that the feature parameters in the feature subset have abnormal values, otherwise, the feature parameters in the feature subset have no abnormal values.

[0132] Specifically, the feature subset is represented in matrix form, where x is the number of rows, indicating the number of protein bands; y is the number of columns, indicating the number of important feature parameters retained. The elements in the matrix represent the feature parameter values ​​of the i-th row and j-th column. Statistics are used to quantify the distribution or variation of the feature parameters in the feature subset. For each feature parameter value in the feature subset, the mean and standard deviation of the feature parameter value are calculated. Each feature parameter value in the feature subset is traversed to detect whether there is an outlier based on the judgment condition. If there is an outlier, the outlier information is output; otherwise, it means that there is no outlier in the feature parameter in the feature subset.

[0133] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A feature engineering-based immunoblotting data classification system, characterized in that: Includes the following modules: Data acquisition module: collects original immunoblot images and immunoblot data; Data processing module: preprocessing the original immunoblot image to remove noise and enhance the image; The immunoblot data were preprocessed including: cleaning and normalization; Feature extraction module: identify whether there are protein bands based on the preprocessed original immunoblot image analysis; extract feature parameters of the original immunoblot image based on the analysis results; analyze and extract the relative expression of proteins based on the extracted feature parameters; perform feature extraction based on the preprocessed immunoblot data; Feature processing module: Processing based on the feature parameters extracted from the original immunoblot images and immunoblot data, including: cleaning, noise removal and standardization; Data classification module: Generate a feature matrix based on the processed feature parameters; filter out feature subsets through a selection algorithm based on the feature matrix; perform data classification based on the filtered feature subsets; Data analysis module: Comprehensively analyze the feature parameters in the feature subset to determine whether anomalies occur.

2. The immunoblot data classification system based on feature engineering according to claim 1, characterized in that: The presence or absence of protein bands can be identified based on the pre-processed original immunoblot image analysis, including the following steps: By using edge detection algorithms, the presence of protein band regions is detected and identified in the original immunoblot images; When protein bands are identified in the original immunoblot image, the edge detection results are processed through morphological operations including: dilation, erosion, opening operation and closing operation to remove noise points, smooth the band edges and fill small holes in the bands; Based on the results of edge detection and morphological processing, the protein bands are separated from the background through image segmentation algorithm, and the characteristic parameters of the protein bands are extracted according to the separation results; When it is identified that the original immunoblot image does not contain a protein band, the immunoblot image is re-collected, and the protein band is detected in the re-collected image, and the above steps are repeated.

3. The immunoblot data classification system based on feature engineering according to claim 2, characterized in that: Extracting characteristic parameters of the original immunoblot image according to the analysis results includes the following steps: Separate the original immunoblot image according to the image segmentation algorithm, locate the position of the protein band and detect and count the number of protein bands through the connected domain analysis or contour detection algorithm, and extract the position characteristic parameters and quantity characteristic parameters of the protein band. The position characteristic parameters include: the starting coordinates of the protein band: (x start ,y start ) and the ending coordinates: (x final ,y final ); the quantitative characteristic parameters include: m protein bands; According to the located protein bands, the intensity characteristic parameters of the a-th protein band are extracted by calculating the average gray value of the a-th protein band; By analyzing the average gray value formula: Get the intensity characteristic parameters of the ath protein band Among them, H i Represents the gray value of the i-th pixel, N a represents the total number of pixels in the a-th protein band region, a={1,2,...,m}; Extract the morphological characteristic parameters of the a-th protein band, including length, width and uniformity; wherein the length of the a-th protein band is obtained by tracking the band outline and calculating the distance between points on the outline, and the width of the a-th protein band is obtained by calculating the number of pixels of the band in the vertical direction; the uniformity of the a-th protein band is evaluated by calculating the coefficient of variation of the gray value; By analyzing the coefficient of variation formula: Get the uniformity U of the ath protein band a ,in represents the average gray value, i={1,2,...,N}; The shape characteristic parameters of the a-th protein band are extracted using a shape analysis algorithm, including perimeter and area. The perimeter of the a-th protein band is obtained by using a contour tracking algorithm, and the area of ​​the a-th protein band is obtained by counting the number of pixels inside the a-th protein band.

4. The immunoblot data classification system based on feature engineering according to claim 3, characterized in that: Analyzing and extracting the relative expression of proteins according to the extracted characteristic parameters includes the following steps: By analyzing the formula: Get the total intensity Q of the ath protein band a ,in represents the average gray value, A a represents the area of ​​the ath protein band; By analyzing the formula: The relative expression of the protein G is obtained, where m represents the total number of protein bands and Q a represents the total intensity of the ath protein band.

5. The immunoblot data classification system based on feature engineering according to claim 1, characterized in that: Feature extraction is performed based on the preprocessed immunoblot data, including the following steps: The pre-processed immunoblot data include: electrophoresis separation data, membrane transfer data, blocking and antibody incubation data, color or luminescence signal data, sample data, experimental condition data and protein data; The protein sample is separated by polyacrylamide gel electrophoresis, and the characteristic parameters of the electrophoretic separation data of the ath protein in the sample are extracted, including: mobility and relative position; Extract the characteristic parameters of the transfer data during the transfer of proteins separated by electrophoresis to the solid support, including: transfer efficiency, time and buffer formula used; After transfer, blocking buffer is used to inhibit nonspecific binding, and specific primary and secondary antibodies are added for incubation. The characteristic parameters of the blocking and antibody incubation data are extracted, including: the type and concentration of the blocking agent, the selection of primary and secondary antibodies, the dilution ratio, the incubation time and conditions; Detecting the presence of the target protein by chemiluminescence, fluorescence or colorimetry, extracting characteristic parameters of the colorimetric or luminescent signal data including: the intensity, location, duration, type of substrate used and detection equipment parameters of the colorimetric or luminescent signal; The characteristic parameters of the extracted sample data include: the source, processing method, and concentration of the sample; The characteristic parameters of the experimental condition data extracted include: electrophoresis voltage, current, transfer voltage, temperature, time, batch number and expiration date of reagents; The mass of the protein is measured by a mass spectrometer, and the characteristic parameter extracted from the protein data is the molecular weight of the protein.

6. The immunoblot data classification system based on feature engineering according to claim 1, characterized in that: According to the processed feature parameters, a feature matrix is ​​generated, including the following steps: According to the processed characteristic parameters, the characteristic vector is generated by integration; for the a-th protein band, the characteristic vector is expressed as: f a,b =[f a,1 ,f a,2 ,…,f a,n ], where f a,b represents the b-th characteristic parameter value of the a-th protein band, a={1,2,...,m}, b={1,2,...,n}, n represents the number of characteristic parameters, and m represents the number of protein bands; Combine the feature vectors of m protein bands into an m×n feature matrix F; The feature matrix is ​​expressed as: Where m is the number of rows representing the number of protein bands, and n is the number of columns representing the number of feature parameters.

7. The immunoblot data classification system based on feature engineering according to claim 6, characterized in that: According to the feature matrix, a feature subset is screened out through a selection algorithm, including the following steps: Feature selection is performed based on the Lasso regression model. The complexity and penalty term of the Lasso regression model are constrained by introducing the L1 regularization term parameter. The feature parameters in the feature matrix are used as input variables and input into the Lasso regression model for training. During the training process, the Lasso regression model automatically penalizes the absolute value of the coefficient of the feature parameter and compresses the coefficient according to the L1 regularization term parameter. According to the trained Lasso regression model, the coefficients of the feature parameters in the Lasso regression model are obtained, and the feature parameters are sorted from large to small according to the absolute values ​​of the coefficients; by setting a threshold, the feature parameters of important features whose absolute values ​​of the coefficients are greater than the threshold are screened out; wherein, when the absolute value of the coefficient is equal to 0, the feature parameter is an unimportant feature, and when the absolute value of the coefficient is greater than 0, the feature parameter is an important feature, and the feature parameters of the important features are retained; According to the feature parameters retained by the Lasso regression model, the corresponding feature parameters are extracted and screened from the feature matrix to generate a feature subset.

8. The immunoblot data classification system based on feature engineering according to claim 7, characterized in that: Data classification based on the filtered feature subset includes the following steps: Using the support vector machine as a classification model, annotating feature parameters and corresponding classification labels in the feature subset screened by the selection algorithm; the classification labels include: different protein expression categories or subcellular location categories of proteins in immunoblot images; Input the labeled data into the support vector machine model for training to learn the mapping relationship between feature parameters and classification labels; New immunoblot images and new immunoblot data are collected in real time according to an immunoblot data classification system based on feature engineering. The new immunoblot images and new immune data are subjected to the same feature extraction and preprocessing steps as the original immunoblot images and immune data, and new feature parameters of the new immunoblot images and new immune data are extracted. The processed new feature parameters are input into a trained model. The support vector machine model automatically classifies the newly input new feature parameters according to the learned mapping relationship and outputs the classification results to obtain different protein expression categories or subcellular location categories of proteins in the new immunoblot images.

9. The immunoblot data classification system based on feature engineering according to claim 1, characterized in that: Comprehensively analyze the feature parameters in the feature subset to determine whether an abnormality occurs, including the following steps: According to the feature subset screened by the selection algorithm, the feature subset is expressed as: Where x is the number of rows representing the number of protein bands, and y is the number of columns representing the number of important feature parameters retained; By analyzing the formula: Get the statistic S of the feature parameters in the feature subset; where f i,j represents the characteristic parameter value of the i-th row and j-th column, x is the row number representing the number of protein bands, y is the column number representing the number of important characteristic parameters retained, i = {1, 2, ..., x}, j = {1, 2, ..., y}; By analyzing the formula: β=|f i,j -S| when When , it means that the feature parameters in the feature subset have abnormal values, otherwise, the feature parameters in the feature subset have no abnormal values.

Images