Gene detection method and gene detection device based on decorrelation statistical learning

Abstract

The invention discloses a gene detection method and a gene detection device based on decorrelation statistical learning. The method comprises the following steps: obtaining an original data set describing a gene expression level based on an RNA sequencing method; analyzing the original data set, and determining a conversion function in a conversion linear model; performing decorrelation change processing on relevance among genes in the original data set according to the determined transfer function; selecting and gradually increasing gene variables from the processed data set based on a decorrelation forward statistical learning method to obtain a gene set meeting conditions; and performing gene detection according to the gene set, and outputting statistical inference and clinical detection results. According to the invention, a decorrelation forward feature extraction method provided by the invention can be used for processing relevance between genes, so that the genes really related to a research target are extracted, and the accuracy of gene detection is improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology service technology, and in particular to a gene detection method and a gene detection device based on decorrelation statistical learning. Background Art

[0002] Isoprene compounds (also known as terpenoids) are a class of organic compounds widely found in animals, plants and microorganisms. This compound is composed of 5 carbon units (isoprene units) and has a variety of biological functions, such as participating in plant hormone regulation, metabolic processes, cell signal transduction, defense mechanisms, etc. In practical applications, the role of isoprene compounds in different biological processes is revealed by studying the genes of enzymes related to isoprene biosynthesis. Genes encoding enzymes related to isoprene biosynthesis are called isoprene genes.

[0003] In modern biomedical research, the expression levels of isoprenoid genes are studied to investigate the following abilities:

[0004] 1) Plant response to environmental stress: Isoprene compounds play an important role in plant stress resistance and pest and disease defense. Studying their gene expression levels can help us understand how plants respond to environmental stress;

[0005] 2) Metabolic regulation: Gene expression values ​​can be used to study the regulatory mechanisms of key enzymes in the isoprene pathway and their effects on plant metabolism;

[0006] 3) Development and hormone synthesis: The expression of isoprenoid genes is closely related to the biosynthesis of plant growth hormones (such as gibberellins and brassinolide), and gene expression values ​​can reveal the regulatory network of hormone synthesis.

[0007] Isoprene genes are usually reflected by detecting the abundance of mRNA. The commonly used detection method is RNA sequencing: by measuring the transcription abundance of genes under different conditions, the gene expression level is quantitatively analyzed. Through RNA sequencing, a large amount of RNA sequencing data of isoprene genes can be obtained. How to effectively extract features from RNA sequencing data to explain plant response to environmental stress, metabolic regulation, development and hormone synthesis is a hot and difficult issue in modern biological research.

[0008] Commonly used feature extraction methods include the minimum absolute value convergence and selection operator, smoothed edge absolute deviation, measured correlation inference method, uniform independent screening and forward regression method, etc. These methods lose their screening ability when facing strongly correlated variables. In addition, the classic model selection method relies on specific model assumptions, which may be violated in actual situations. To break through this limitation, researchers have proposed some model-free variable selection methods based on correlation metrics. However, these feature extraction methods have difficulty in handling the correlation between genes (for example, co-expression relationships, gene regulatory networks, and gene functional associations), and thus mistakenly extract some unimportant features.

[0009] Therefore, the prior art needs to be improved. Summary of the invention

[0010] The technical problem to be solved by the present invention is that, in view of the defects of the prior art, the present invention provides a gene detection method and a gene detection device based on decorrelation statistical learning to solve the problem that the existing gene feature extraction method is difficult to handle the correlation between genes.

[0011] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0012] In a first aspect, the present invention provides a gene detection method based on decorrelation statistical learning, comprising:

[0013] Based on RNA sequencing method, the original data set describing gene expression levels was obtained;

[0014] Analyzing the original data set to determine a conversion function in a conversion linear model;

[0015] Performing decorrelation change processing on the correlation between genes in the original data set according to the determined conversion function;

[0016] Based on the decorrelation forward statistical learning method, gene variables are selected and gradually added from the processed data set to obtain a gene set that meets the conditions;

[0017] Gene detection is performed based on the gene set, and statistical inference and clinical detection results are output.

[0018] In one implementation, the RNA sequencing method is used to obtain an original data set describing gene expression levels, including:

[0019] For a given number of n research subjects, the RNA sequencing method is used to obtain a raw data set describing the expression level of isoprene genes;

[0020] The original data set is divided into two parts Among them, the first part y i For the purpose of this study, the second part xi For p n The expression value of a gene.

[0021] In one implementation, analyzing the original data set to determine a conversion function in a conversion linear model includes:

[0022] Determine the relationship between the response variable and the predictor variable in the original data set by drawing a scatter plot and a residual plot;

[0023] Selecting a corresponding conversion function according to the nonlinear data features, heteroscedastic data features and multicollinearity data features in the original data set;

[0024] The data of the original data set are transformed according to the selected transformation function, and a linear regression model is applied for fitting, so as to evaluate the effect of the transformation linear model;

[0025] The normality and homoscedasticity of the transformed linear model are detected, and the transformed linear model is optimized according to the residual analysis index and the model goodness of fit index.

[0026] In one implementation, the conversion function includes: a logarithmic function log(t), a logical function Square root function Polynomial function t 2 One or a combination of .

[0027] In one implementation, the decorrelation change processing of the correlation between genes in the original data set according to the determined conversion function includes:

[0028] The prediction variable matrix X in the original data set is (x1, x2, ..., x n ) T Perform a decorrelation transform:

[0029]

[0030] Among them, 0<λ n <1, I n is the n-order identity matrix;

[0031] Perform decorrelation transformation on the response vector matrix in the original data set:

[0032]

[0033] in, ε=(XX T / p n +λ n I n ) -1 / 2 ε* , ε * =(ε1,ε2,…,ε n ) T is the random error vector;

[0034] Determine the decorrelated residual sum of squares:

[0035]

[0036] in, is the gene set, is the projection matrix, Indicated by The column index falls in the set The sub-matrix formed in ;

[0037] The feature is measured according to the residual sum of squares of the decorrelation The importance of is determined and the data set processed by decorrelation transformation is obtained.

[0038] In one implementation, the decorrelation-based forward statistical learning method selects and gradually adds gene variables from the processed data set to obtain a gene set that meets the conditions, including:

[0039] Initialize the transformation linear model:

[0040] Y * =ε * ;

[0041] Calculate the linear relationship between each predictor variable and the response variable, and select the first gene X1 model that has the strongest correlation with the response variable. The updated conversion linear model is:

[0042] Y * =X1β1+ε * ;

[0043] For each candidate gene, the contribution of the candidate gene and the response variable in the current model is calculated, and the second gene X2 that has the greatest improvement in the model fitting effect is selected to be added to the model. The updated conversion linear model is:

[0044] Y * =X1β1+X2β2+ε * .

[0045] In one implementation, the step of calculating the contribution of the candidate gene and the response variable in the current model, selecting the second gene X2 that has the greatest improvement on the model fitting effect to be added to the model, and then further comprising:

[0046] Each time the remaining predictor variables are evaluated, the variable that contributes most to the improvement of the current transformed linear model is selected, and the selected variable is added to the current transformed linear model;

[0047] when When , we stop selecting candidate genes, and the final transformation linear model is in the form of:

[0048] Y * =X1β1+X2β2+…+X k β k +ε * ;

[0049] Among them, k is the number of variables finally selected.

[0050] In a second aspect, the present invention provides a gene detection device based on decorrelation statistical learning, comprising:

[0051] A raw data acquisition module is used to obtain a raw data set describing gene expression levels based on an RNA sequencing method;

[0052] A conversion function determination module, used to analyze the original data set and determine the conversion function in the conversion linear model;

[0053] A decorrelation change processing module, used for performing decorrelation change processing on the correlation between genes in the original data set according to the determined conversion function;

[0054] A forward statistical learning module is used to select and gradually add gene variables from the processed data set based on a decorrelation forward statistical learning method to obtain a gene set that meets the conditions;

[0055] The gene detection module is used to perform gene detection based on the gene set and output statistical inference and clinical detection results.

[0056] In a third aspect, the present invention provides a terminal comprising: a processor and a memory, wherein the memory stores a gene detection program based on decorrelation statistical learning, and the gene detection program based on decorrelation statistical learning is used to implement the operation of the gene detection method based on decorrelation statistical learning as described in the first aspect when executed by the processor.

[0057] In a fourth aspect, the present invention further provides a medium, which is a computer-readable storage medium, and which stores a gene detection program based on decorrelation statistical learning. When the gene detection program based on decorrelation statistical learning is executed by a processor, it is used to implement the operation of the gene detection method based on decorrelation statistical learning as described in the first aspect.

[0058] The present invention adopts the above technical solution to achieve the following effects:

[0059] The present invention can determine the conversion function in the conversion linear model by pre-analyzing the research target and gene data, thereby converting the relationship between the response variable and the gene covariate under the complex research target into a linear relationship, so that the extracted genes are interpretable. In addition, the present invention takes into account the correlation between genes, and uses the decorrelation transformation method to make the columns of the covariate matrix after the decorrelation transformation asymptotically orthogonal, thereby reducing the interference of the correlation between genes on the feature, and extracting genes that are truly related to the research target based on the decorrelation forward statistical learning method, thereby improving the accuracy of gene detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] 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 the structures shown in these drawings without paying creative work.

[0061] Figure 1 It is a flow chart of the gene detection method based on decorrelation statistical learning in the present invention.

[0062] Figure 2 It is a schematic diagram comparing the model selection results and correlations of an implementation method in the present invention.

[0063] Figure 3 It is a box plot of the prediction errors of eight comparison methods in one implementation of the present invention.

[0064] Figure 4 It is a box plot of the prediction errors of six comparison methods in one implementation of the present invention.

[0065] Figure 5 It is a functional principle diagram of a terminal in one implementation of the present invention.

[0066] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with implementation examples. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0068] Exemplary Methods

[0069] When effectively extracting features from RNA sequencing data, commonly used feature extraction methods include the minimum absolute value convergence and selection operator, smoothed edge absolute deviation, measured correlation inference method, uniform independent screening, and forward regression methods. These methods lose their screening capabilities when faced with strongly correlated variables. In addition, classic model selection methods rely on specific model assumptions that may be violated in actual situations. To break through this limitation, researchers have proposed some model-free variable selection methods based on correlation metrics. However, these feature extraction methods have difficulty handling the correlation between genes (e.g., co-expression relationships, gene regulatory networks, and gene functional associations), thereby incorrectly extracting some unimportant features.

[0070] In view of the above technical problems, a gene detection method based on decorrelation statistical learning is provided in an embodiment of the present invention. The method can determine the conversion function in the conversion linear model by pre-analyzing the research target and gene data, thereby converting the relationship between the response variable and the gene covariate under the complex research target into a linear relationship, so that the extracted genes are interpretable. Moreover, the method takes into account the correlation between genes, and makes the columns of the covariate matrix after the decorrelation transformation asymptotically orthogonal through the decorrelation transformation method, thereby reducing the interference of the correlation between genes on the features, and extracting genes that are truly related to the research target based on the decorrelation forward statistical learning method, thereby improving the accuracy of gene detection.

[0071] How to characterize the relationship between the response variable y and multiple predictor variables x under complex research objectives is a hot and difficult issue in gene extraction. Generally speaking, linear models are often used to describe the relationship between response variables or predictor variables. However, it is difficult for linear models to describe the relationship between response variables or predictor variables under complex research objectives. For example, when exploring pathogenic genes, the response variable is whether the biological individual is sick, that is, the response variable is a 0-1 variable. At this time, the relationship between the response variable or the predictor variable is not linear. Complex models have been proposed in existing studies to characterize the relationship between response variables or predictor variables. However, complex models do not have model interpretability. In a linear model, the regression coefficient represents the impact of a 1% change in the predictor variable on the response variable, and this interpretation has higher practical significance in many fields. Therefore, in an embodiment of the present invention, a conversion linear model is proposed to handle the complex relationship between the response variable and the predictor variable, while maintaining the model interpretability of the linear model.

[0072] like Figure 1 As shown, an embodiment of the present invention provides a gene detection method based on decorrelation statistical learning, comprising the following steps:

[0073] Step S100, obtaining an original data set describing gene expression levels based on an RNA sequencing method.

[0074] Specifically, in an implementation of this embodiment, step S100 includes the following steps:

[0075] Step S101, for given n research subjects, using the RNA sequencing method to obtain an original data set describing the expression level of isoprene genes;

[0076] Step S102: Divide the original data set into two parts The first part yi is the research objective, and the second part x i For p n The expression value of a gene.

[0077] In this example, RNA sequencing is performed on n research objects (e.g., plants), and the original data set obtained is recorded as where y i is the research goal, x i Yes n The research objectives in this embodiment are very broad and may specifically include the following aspects:

[0078] 1) Find the disease-causing gene;

[0079] 2) Establish the genetic regulatory mechanisms of plant responses to environmental stress;

[0080] 3) Find genes related to specific metabolic regulation;

[0081] 4) Search for genes related to specific development and hormone synthesis.

[0082] In this example, after obtaining the original data set describing the expression levels of isoprene genes, a conversion function in the conversion linear model is determined to perform decorrelation processing on the correlation between genes in the original data set through the conversion function in the conversion linear model.

[0083] like Figure 1 As shown, an embodiment of the present invention provides a gene detection method based on decorrelation statistical learning, comprising the following steps:

[0084] Step S200: Analyze the original data set to determine the conversion function in the conversion linear model.

[0085] In this embodiment, a new transformation linear model is designed based on the traditional linear model; the traditional linear model is a commonly used statistical model used to describe the relationship between the response variable y and multiple predictor variables x:

[0086] y=x T β+ε;

[0087] Where ε is a random error. Given that existing model selection methods are very sensitive to the correlation between genes. From a data perspective, the high correlation between the components of x reflects the correlation between genes. In order to deal with the correlation between genes, this embodiment decorrelates the prediction vector x. Let Y = (y1, y2, ..., y n ) T , X=(x1,x2,…,x n ) T , A=(XX T / p n +λ n I n ) -1 is the matrix for performing decorrelation transformation, where I n is the n-th order identity matrix.

[0088] Since it is difficult for traditional linear models to describe the relationship between the response variable y and multiple predictor variables x under complex research objectives, this embodiment considers using a transformation linear model to describe the relationship between the response variable y and multiple predictor variables x:

[0089] g(y)=x T β+ε;

[0090] Wherein, g is a commonly used conversion function, including a logarithmic function, a logical function, a square root function or a polynomial function.

[0091] In this embodiment, the linear model g(y)=x is transformed T β+ε, to characterize the relationship between the response variable y and multiple predictor variables x under complex research objectives, and use the DF method (i.e., decorrelated forward regression method, which mainly decorrelates the independent variables, starts from zero independent variables, and gradually adds independent variables to the model) to extract key genes. Theoretically, the genes extracted by the DF method under the transformed linear model in this embodiment are all key genes that have an important impact on the research objectives, and an upper bound on the number of key genes that have an important impact on the research objectives is also provided.

[0092] Specifically, in an implementation of this embodiment, step S200 further includes the following steps:

[0093] Step S201, determining the relationship between the response variable and the prediction variable in the original data set by drawing a scatter plot and a residual plot;

[0094] Step S202, selecting a corresponding conversion function according to the nonlinear data features, heteroscedastic data features and multicollinearity data features in the original data set;

[0095] Step S203, transforming the data of the original data set according to the selected transformation function, applying a linear regression model for fitting, and evaluating the effect of the transformation linear model;

[0096] Step S204, detecting the normality and homoscedasticity of the transformed linear model, and optimizing the transformed linear model according to the residual analysis index and the model fit goodness index.

[0097] In this embodiment, by converting the linear model g(y)=x T β+ε, to characterize the relationship between the response variable y and multiple predictor variables x under complex research objectives, the following steps are required to determine the conversion function g in the conversion linear model:

[0098] 1. Check the relationship between the data: Preliminarily observe the relationship between the response variable and the predictor variable by drawing scatter plots, residual plots, etc.

[0099] 2. Select the transformation method: Select the appropriate model transformation method based on the data characteristics (for example, nonlinearity, heteroscedasticity, multicollinearity, etc.).

[0100] 3. Apply the transformed model: After transforming the data, apply the linear regression model to fit and evaluate the effect of the transformed linear model.

[0101] 4. Verify model assumptions: Check the assumptions of the transformation linear model, such as normality, homoscedasticity, etc., and further optimize the transformation linear model based on indicators such as residual analysis and model goodness of fit.

[0102] Commonly used conversion functions include: logarithmic function log(t), logical function Square root function Polynomial function t 2 Etc. In this embodiment, the appropriate conversion function g in the conversion linear model can be determined through the above steps.

[0103] like Figure 1 As shown, an embodiment of the present invention provides a gene detection method based on decorrelation statistical learning, comprising the following steps:

[0104] Step S300 , decorrelation transformation processing is performed on the correlation between genes in the original data set according to the determined conversion function.

[0105] In this embodiment, Thus, we can get the conversion linear model g(y)=x T Sample representation of β+ε:

[0106] Y * =Xβ+ε * ;

[0107] here is the response vector, X = (x1, x2, ..., x n ) T is the covariate matrix, ε * =(ε1,ε2,…,ε n ) T is the random error vector. Definition is a set of key genes. That is, a set of genes that are important for the research objectives of this embodiment. According to the definition of the transformation linear model,

[0108] Next, a new gene extraction method is proposed in this example.

[0109] Specifically, in an implementation of this embodiment, step S300 includes the following steps:

[0110] Step S301: Calculate the prediction variable matrix X in the original data set. n ) T Performing decorrelation transformation;

[0111] Step S302: performing decorrelation transformation on the response vector matrix in the original data set.

[0112] In this embodiment, when decorrelation processing is performed on the correlation between genes, the correlation between genes is first preprocessed, and the correlation between genes is reflected in the prediction variable x. i In order to deal with the predictor variable x i There is a strong correlation between them. In this embodiment, the data feature matrix X is first decorrelated:

[0113]

[0114] Among them, 0<λ n <1, I n is the n-order identity matrix;

[0115] Under the general assumption, Different features are asymptotically orthogonal. This inspires this embodiment to consider Y * =Xβ+ε * Perform decorrelation transformation, thus we have:

[0116]

[0117] in, ε=(XX T / p n +λ n I n) -1 / 2 ε * , ε * =(ε1,ε2,…,ε n ) T is the random error vector.

[0118] Specifically, in an implementation of this embodiment, step S300 further includes the following steps:

[0119] Step S303, determining the decorrelated residual sum of squares;

[0120] Step S304, measuring the feature according to the decorrelated residual sum of squares The importance of is determined and the data set processed by decorrelation transformation is obtained.

[0121] In this embodiment, after preprocessing the correlation between genes, the genes after decorrelation transformation are screened according to the gene importance measurement criterion; in order to screen out important genes, the following residual sum of squares is introduced in this embodiment to measure the feature For a given set of genes The decorrelated residual sum of squares is defined as:

[0122]

[0123] in, is the gene set, is the projection matrix, Indicated by The column index falls in the set The sub-matrix formed in .

[0124] This embodiment performs decorrelation transformation processing on the correlation between genes in the original data set through the above steps, and then uses the decorrelation forward statistical learning method to extract key genes that have an important impact on the research target from the data set after the decorrelation transformation processing.

[0125] like Figure 1 As shown, an embodiment of the present invention provides a gene detection method based on decorrelation statistical learning, comprising the following steps:

[0126] Step S400 , selecting and gradually adding gene variables from the processed data set based on a decorrelation forward statistical learning method to obtain a gene set that meets the conditions.

[0127] In this embodiment, the prediction vector x is decorrelated. Let Y = (y1, y2, ..., y n ) T , X=(x1,x2,…,x n ) T , A=(XXT / p n +λ n I n ) -1 is the matrix for performing decorrelation transformation, where I n is an n-order unit matrix. Inspired by the fact that the column vectors in the matrix AX are asymptotically orthogonal to each other, in this embodiment, (Y, X) is mapped to (AY, AX), and forward regression is performed on (AY, AX). This method is called the decorrelated forward (DF) gene extraction method. Different from the classical forward regression method, the DF method uses a decorrelated linear transformation matrix A to process the correlation between genes.

[0128] Specifically, in an implementation of this embodiment, step S400 includes the following steps:

[0129] Step S401, initializing the conversion linear model;

[0130] Step S402, calculating the linear relationship between each predictor variable and the response variable, and selecting the first gene X1 model that has the strongest correlation with the response variable;

[0131] Step S403, for each candidate gene, calculate the contribution of the candidate gene and the response variable in the current model, and select the second gene X2 that has the greatest improvement on the model fitting effect to be added to the model;

[0132] Step S404, each time the remaining prediction variables are evaluated, the variable that contributes most to the improvement of the current conversion linear model is selected, and the selected variable is added to the current conversion linear model;

[0133] Step S405: when the conditions are met, stop selecting candidate genes and obtain the final conversion linear model.

[0134] In this embodiment, the important gene set is extracted by decorrelation forward regression (DF). Decorrelation forward regression is a gene extraction method that gradually selects genes to build a regression model. Its core idea is to start from an empty set and gradually add gene variables to the model until the fitting effect of the model cannot be significantly improved. The steps of decorrelation forward regression are as follows:

[0135] 1) Initialize the model: Starting from an empty model with no predictors, transform the linear model into:

[0136] Y * =ε * ;

[0137] 2) Select the first gene: Calculate the linear relationship between each predictor variable and the response variable, and select a gene X1 model that has the strongest correlation with the response variable (i.e., significantly reduces the error or improves the model fit), that is, gene X1 satisfies in is an empty collection. The updated model is:

[0138] Y * =X1β1+ε * ;

[0139] 3) Select the second gene: After the first gene is introduced into the model, evaluate the remaining genes again. For each candidate gene, calculate its contribution to the response variable in the current model. Select the second gene X2 that has the greatest improvement in model fitting effect to be added to the model, that is, gene X2 satisfies in The updated model is:

[0140] Y * =X1β1+X2β2+ε * ;

[0141] 4) Repeat the process: Repeat the process again and again, evaluating the remaining predictor variables each time, selecting the variable that contributes the most to the model improvement, and adding it to the model.

[0142] 5) Stop condition: If Stop selecting candidate genes, that is, the remaining genes can no longer significantly improve the model fitting effect. Here c n，k is a constant whose specific value is given in Algorithm 1. The final model may be in the form of:

[0143] Y * =X1β1+X2β2+…+X k β k +ε * ;

[0144] Where k is the number of variables finally selected.

[0145] In this embodiment, the above gene extraction method is called a decorrelated forward (DF) statistical learning method. The specific implementation steps of the DF learning method are shown in Algorithm 1:

[0146] Algorithm 1: DF learning algorithm

[0147] Initialization: Input c, n, λ n and p n .

[0148] i) Decorrelation transformation

[0149] 1: Calculate Ψ=(XXT / p n +λ n I n ) -1 / 2 , and

[0150] ii) Forward selection

[0151] 2: Loop from k=1, 2, ..., n

[0152] 2.1 Calculation and

[0153] 2.2 Update

[0154] 2.3 Order

[0155] 2.4 Calculation

[0156] 2.5: If l k ≤c n，k , then stop the loop and update

[0157] iii) Return

[0158] The decorrelation forward statistical learning method proposed in this embodiment has the following advantages:

[0159] (1) Simple and intuitive: The decorrelation forward selection method is easy to understand and implement, and is suitable for dealing with gene extraction problems in large-scale gene datasets;

[0160] (2) It can handle the correlation between genes: the correlation between genes is reflected in the response variable x i The proposed decorrelation transformation can handle the correlation between genes well, thus ensuring that the proposed genes are all important genes for the research objectives;

[0161] (3) Reduce model complexity: By gradually adding genes, it can control the complexity of the model, avoid using unnecessary variables, and prevent model overfitting;

[0162] (4) Strong interpretability: Only one gene is added at a time, and the constructed model is easy to interpret, making it easy to understand the contribution of each gene to the response variable.

[0163] like Figure 1 As shown, an embodiment of the present invention provides a gene detection method based on decorrelation statistical learning, comprising the following steps:

[0164] Step S500, performing gene testing according to the gene set, and outputting statistical inference and clinical test results.

[0165] In this embodiment, after processing the correlation between genes and extracting key genes that have a significant impact on the research objectives based on the decorrelation forward statistical learning method, these key genes that have a significant impact on the research objectives can be used to perform statistical inference and clinical testing tasks, thereby achieving accurate gene testing tasks and outputting statistical inference and clinical test results.

[0166] As an experimental result of this embodiment, this embodiment compares the performance of the DF method and the FARMS method (i.e., the factor adjustment regularized model selection method, a regularized model selection method based on the adjustment of the covariate factor structure) when screening features affected by correlation. Consider a special linear model i=1,2,…,n, where sample size n=200 and dimension p n =1000,ε i ~N(0,1), Factor structure of the covariate vector x i =Bf i +u i ,in u i ~U(0,0.5),

[0167] In this structure, the number of factors K is equal to the dimension p n . Let the correlation of variables be ρ, and let ρ increase from 0 to 0.8 in steps of 0.1. For each given ρ, this embodiment repeats the simulation calculation 600 times, and calculates the average number of real variables included in the selected model, the average number of unimportant variables selected by mistake, and the ratio of model screening consistency. Figure 2 As shown, Figure 2 The left figure in the middle shows the average number of real variables included in the selected model (a total of 3 real variables). Figure 2 The middle graph shows the average number of incorrectly selected insignificant variables. Figure 2 The right figure in the middle shows the proportion of model screening consistency. Obviously, as ρ increases, correlation has a great impact on the model selection results of FARMS. In contrast, the DF method proposed in this example is very robust to the high correlation between the predictor variables.

[0168] As an example, the performance of the method of this embodiment in analyzing Arabidopsis gene data is as follows:

[0169] Arabidopsis is a good material for genetic research and is praised by scientists as the fruit fly in plants. Arabidopsis is widely used in molecular genetics research due to its own advantages (small genome, few repetitive sequences; short life cycle, obvious flower characteristics, etc.). Next, this embodiment uses Arabidopsis gene data to demonstrate the screening performance of the T-DF algorithm. This data set can be obtained in the supplementary materials of the document Wille (2004), which contains 118 samples, each of which contains the expression values ​​of 834 genes in 58 different metabolic pathways.

[0170] From a biological point of view, in Arabidopsis, some key compounds such as carotenoids, chlorophyll and gibberellins use geranyl diphosphate (GGPP) as a precursor. The production of GGPP mainly depends on the gene GGPPS11. The goal of this example is to explore how the expression level of GGPPS11 is affected by the other 833 genes. Among these 833 genes, there is a strong correlation between the expression levels of most genes.

[0171] In this example, the gene data set is first standardized, and after data pre-analysis, a linear model is determined for fitting:

[0172]

[0173] Among them, Y i represents the expression value of gene GGPPS11, x ij represents the expression value of the jth gene in the ith individual.

[0174] Next, this example applies 8 feature selection methods to this data set, including the T-DF learning method proposed in this project. The detailed gene selection results are shown in Table 1. Here, the model size refers to the number of gene variables in the model.

[0175] Table 1: Model sizes selected by eight compared methods

[0176]

[0177] In order to evaluate the performance of these methods, this embodiment implemented cross validation on the data set and compared the prediction errors. This process was repeated 600 times. Figure 3 As shown, Figure 3 The box plot in the figure shows the prediction error performance of different methods. It is worth noting that the T-DF learning algorithm proposed in this embodiment always tends to select a medium-sized model, that is, the number of parameters of the model is medium-sized and shows the lowest prediction error. This shows that other methods may have overfitting and underfitting when facing strong correlation feature selection, resulting in an increase in prediction error.

[0178] As an example, the performance of the method of this embodiment in detecting breast cancer associated factor data is as follows:

[0179] This example demonstrates the performance of the proposed T-DF learning algorithm by analyzing breast cancer data. Breast cancer is the second most common cancer in the world and poses a major threat to human health, especially women. Accurate prediction of tumor metastasis is essential for effective breast cancer treatment. Here, this example uses the gene expression data originally studied by van't Veer et al. (2002) to explore the effect of gene probes on breast cancer tumor metastasis.

[0180] The data set includes 97 patients with breast cancer aged 55 or below with negative lymph nodes. Among them, 46 patients had distant metastasis within five years (coded as 1), and 51 patients had no metastasis for at least five years (coded as 0). This embodiment analyzes the expression levels of 24,188 gene probes, and there are no missing values ​​in these data. This embodiment focuses on the top 2,000 genes with the highest variance and generates a heat map. Through their estimated correlation matrix, it is found that there is a strong correlation between these genes, which further emphasizes the necessity of the decorrelation learning method proposed in this embodiment.

[0181] According to the model g(y)=x T β+ε, this example uses a large number of highly correlated covariates to perform logistic regression to predict binary outcomes. The logistic regression model is expressed as:

[0182]

[0183] Among them, π i Represents Y i The marginal mean, x ij is the expression value of the jth gene probe in the ith patient.

[0184] The binary result data were analyzed using the T-DF learning method proposed in this embodiment and five other methods. The following table (Table 2) shows the scale of the model selected by each method.

[0185] Table 2: Model sizes selected by the six compared methods

[0186]

[0187] Similarly, this embodiment implemented cross validation on this data set, compared the prediction errors, and repeated this process 600 times, as shown in FIG. Figure 4 As shown, Figure 4 The box plots in the figure show the prediction errors of different methods. Once again, the T-DF learning algorithm proposed in this embodiment always tends to choose a medium-sized model and shows the lowest prediction error.

[0188] It is understandable that after solving the effective selection of gene sets, based on the idea of ​​data (etc.) segmentation cross-fitting, the estimation of the sub-models returned by the DF learning algorithm is aggregated in this embodiment, and the asymptotic variance of the estimated data after aggregation is minimized to determine the selection of the weight matrix during aggregation. On this basis, the present invention constructs a test statistic that can realize the consistent most advantageous test in the sense of data segmentation, greatly improving the detection capability of data features under dimensional data.

[0189] This embodiment achieves the following technical effects through the above technical solution:

[0190] This embodiment can determine the conversion function in the conversion linear model by pre-analyzing the research target and gene data, thereby converting the relationship between the response variable and the gene covariate under the complex research target into a linear relationship, so that the extracted genes are interpretable. Moreover, in this embodiment, the correlation between genes is taken into account, and the columns of the covariate matrix after the decorrelation transformation are made asymptotically orthogonal through the decorrelation transformation method, thereby reducing the interference of the correlation between genes on the features, and extracting genes that are truly related to the research target based on the decorrelation forward statistical learning method, thereby improving the accuracy of gene detection.

[0191] Exemplary Devices

[0192] Based on the above embodiments, the present invention further provides a gene detection device based on decorrelation statistical learning, comprising:

[0193] A raw data acquisition module is used to obtain a raw data set describing gene expression levels based on an RNA sequencing method;

[0194] A conversion function determination module, used to analyze the original data set and determine a conversion function in a conversion linear model;

[0195] A decorrelation change processing module, used for performing decorrelation change processing on the correlation between genes in the original data set according to the determined conversion function;

[0196] A forward statistical learning module is used to select and gradually increase gene variables from the processed data set based on a decorrelation forward statistical learning method to obtain a gene set that meets the conditions;

[0197] The gene detection module is used to perform gene detection based on the gene set and output statistical inference and clinical detection results.

[0198] This embodiment achieves the following technical effects through the above technical solution:

[0199] This embodiment can determine the conversion function in the conversion linear model by pre-analyzing the research target and gene data, thereby converting the relationship between the response variable and the gene covariate under the complex research target into a linear relationship, so that the extracted genes are interpretable. Moreover, in this embodiment, the correlation between genes is taken into account, and the columns of the covariate matrix after the decorrelation transformation are made asymptotically orthogonal through the decorrelation transformation method, thereby reducing the interference of the correlation between genes on the features, and extracting genes that are truly related to the research target based on the decorrelation forward statistical learning method, thereby improving the accuracy of gene detection.

[0200] Based on the above embodiment, the present invention further provides a terminal, whose principle block diagram can be as follows: Figure 5 shown.

[0201] The terminal includes: a processor, a memory, an interface, a display screen and a communication module connected through a system bus; wherein the processor of the terminal is used to provide computing and control capabilities; the memory of the terminal includes a storage medium and an internal memory; the storage medium stores an operating system and a computer program; the internal memory provides an environment for the operation of the operating system and the computer program in the storage medium; the interface is used to connect to external devices; the display screen is used to display corresponding information; and the communication module is used to communicate with a cloud server or other devices.

[0202] When the computer program is executed by a processor, it is used to implement the operation of a gene detection method based on decorrelation statistical learning.

[0203] It can be understood by those skilled in the art that Figure 5 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal to which the scheme of the present invention is applied. The specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0204] In one embodiment, a terminal is provided, which includes: a processor and a memory, wherein the memory stores a gene detection program based on decorrelation statistical learning, and the gene detection program based on decorrelation statistical learning is used to implement the operation of the above-mentioned gene detection method based on decorrelation statistical learning when executed by the processor.

[0205] In one embodiment, a storage medium is provided, wherein the storage medium stores a gene detection program based on decorrelation statistical learning, and when the gene detection program based on decorrelation statistical learning is executed by a processor, it is used to implement the operation of the above-mentioned gene detection method based on decorrelation statistical learning.

[0206] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and volatile memory.

[0207] In summary, the present invention provides a gene detection method and a gene detection device based on decorrelation statistical learning, including: obtaining an original data set describing gene expression levels based on an RNA sequencing method; analyzing the original data set to determine a conversion function in a conversion linear model; decorrelation-changing the correlation between genes in the original data set according to the determined conversion function; selecting and gradually adding gene variables from the processed data set based on a decorrelation forward statistical learning method to obtain a gene set that meets the conditions; performing gene detection based on the gene set, and outputting statistical inference and clinical detection results. The present invention proposes a decorrelation forward feature extraction method that can process the correlation between genes, thereby extracting genes that are truly relevant to the research target, and improving the accuracy of gene detection.

[0208] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A gene detection method based on decorrelation statistical learning, characterized in that: include: Based on RNA sequencing method, the original data set describing gene expression levels was obtained; Analyzing the original data set to determine a conversion function in a conversion linear model; Performing decorrelation change processing on the correlation between genes in the original data set according to the determined conversion function; Based on the decorrelation forward statistical learning method, gene variables are selected and gradually added from the processed data set to obtain a gene set that meets the conditions; Gene detection is performed based on the gene set, and statistical inference and clinical detection results are output.

2. The gene detection method based on decorrelation statistical learning according to claim 1, characterized in that: The original data set describing gene expression levels obtained based on the RNA sequencing method includes: For a given number of n research subjects, the RNA sequencing method is used to obtain a raw data set describing the expression level of isoprene genes; The original dataset consists of two parts Among them, the first part y i For the purpose of this study, the second part x i For p n The expression value of a gene.

3. The gene detection method based on decorrelation statistical learning according to claim 1, characterized in that: The analyzing the original data set to determine the conversion function in the conversion linear model includes: Determine the relationship between the response variable and the predictor variable in the original data set by drawing a scatter plot and a residual plot; Selecting a corresponding conversion function according to the nonlinear data features, heteroscedastic data features and multicollinearity data features in the original data set; The data of the original data set are transformed according to the selected transformation function, and a linear regression model is applied for fitting, so as to evaluate the effect of the transformation linear model; The normality and homoscedasticity of the transformed linear model are detected, and the transformed linear model is optimized according to the residual analysis index and the model goodness of fit index.

4. The gene detection method based on decorrelation statistical learning according to claim 1 or 3, characterized in that: The conversion function includes: logarithmic function log(t), logical function Square root function Polynomial function t 2 One or a combination of .

5. The gene detection method based on decorrelation statistical learning according to claim 1, characterized in that: The decorrelation change processing of the correlation between the genes in the original data set according to the determined conversion function includes: The prediction variable matrix X in the original data set is (x1, x2, ..., x n ) T Perform a decorrelation transform: Among them, 0<λ n <1,I n is the n-order identity matrix; Perform decorrelation transformation on the response vector matrix in the original data set: in, ε=(XX T / p n +λ n I n ) -1 / 2 ε * , ε * =(ε1,ε2,…,ε n ) T is the random error vector; Determine the decorrelated residual sum of squares: in, is the gene set, is the projection matrix, Indicated by The column index falls into the set The submatrix formed in ; The features are measured based on the residual sum of squares of the decorrelation The importance of is determined and the data set processed by decorrelation transformation is obtained.

6. The gene detection method based on decorrelation statistical learning according to claim 1, characterized in that: The decorrelation-based forward statistical learning method selects and gradually adds gene variables from the processed data set to obtain a gene set that meets the conditions, including: Initialize the transformed linear model: Y * =e * ; Calculate the linear relationship between each predictor variable and the response variable, and select the first gene X1 model that has the strongest correlation with the response variable. The updated conversion linear model is: Y * =X1β1+ε * ; For each candidate gene, the contribution of the candidate gene and the response variable in the current model is calculated, and the second gene X2 that has the greatest improvement in the model fitting effect is selected to be added to the model. The updated conversion linear model is: Y * =X1β1+X2β2+ε * 。 7. The gene detection method based on decorrelation statistical learning according to claim 6, characterized in that: The step of calculating the contribution of the candidate gene and the response variable in the current model, selecting the second gene X2 that has the greatest improvement on the model fitting effect and adding it to the model, further includes: Each time the remaining predictor variables are evaluated, the variable that contributes most to the improvement of the current transformed linear model is selected, and the selected variable is added to the current transformed linear model; when When , we stop selecting candidate genes, and the final transformation linear model is in the form of: Y * =X1β1+X2β2+…+X k b k +e * ; Among them, k is the number of variables finally selected.

8. A gene detection device based on decorrelation statistical learning, characterized in that: include: A raw data acquisition module is used to obtain a raw data set describing gene expression levels based on an RNA sequencing method; A conversion function determination module, used to analyze the original data set and determine a conversion function in a conversion linear model; A decorrelation change processing module, used for performing decorrelation change processing on the correlation between genes in the original data set according to the determined conversion function; A forward statistical learning module is used to select and gradually add gene variables from the processed data set based on a decorrelation forward statistical learning method to obtain a gene set that meets the conditions; The gene detection module is used to perform gene detection based on the gene set and output statistical inference and clinical detection results.

9. A terminal, characterized in that: include: A processor and a memory, wherein the memory stores a gene detection program based on decorrelation statistical learning, and when the gene detection program based on decorrelation statistical learning is executed by the processor, it is used to implement the operation of the gene detection method based on decorrelation statistical learning as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a gene detection program based on decorrelation statistical learning, and when the gene detection program based on decorrelation statistical learning is executed by a processor, it is used to implement the operation of the gene detection method based on decorrelation statistical learning as described in any one of claims 1-7.

Images