A method for determining differentially expressed genes in RNA-seq and its application

Abstract

The present invention discloses a method and application for determining RNA-seq differentially expressed genes, and relates to the field of single-cell biotechnology. The determination method comprises the steps of: removing random noise interference genes, using t-test to perform differential detection on each gene to find non-differentially expressed genes as negative controls, using PLS to model classification-related components, using the minimum mean square error method to remove classification-irrelevant components, finding outliers based on processed data and non-differentially expressed genes, and judging them as differentially expressed genes; the determination method is applied to the difference of cell expression profile characteristics under different conditions. The selection of the determination criteria in the present invention relies on a data-driven adaptive algorithm, which can be automatically adjusted according to different data, and does not require human intervention, thereby reducing human intervention errors.

Description

Technical Field

[0001] The present invention relates to the field of single-cell biotechnology, and in particular to a method for determining differentially expressed genes by RNA-seq and its application. Background Art

[0002] Single-cell RNA-seq technology can comprehensively describe the expression profile characteristics of a single cell at the time of sampling. Since the current single-cell technology can obtain information from multiple cells at the same time, it provides a powerful tool for better understanding of biological and clinical problems. In many cases, people will obtain samples from multiple conditions, such as disease and normal; then compare the differences in cell expression profile characteristics under different conditions to infer the impact of conditional changes on cells. Among them, the determination of differentially expressed genes is crucial.

[0003] There are currently two approaches to help people find differentially expressed genes, one is based on the magnitude of the difference in expression levels, and the other is based on the significance of the difference. The former most commonly uses the change in gene expression levels. However, due to the lack of widely accepted standards, the threshold for the change in multiples is often determined manually, which can lead to large biases. The latter is based on p-values, which have a widely accepted threshold (0.05). However, since the number of genes to be evaluated is often tens of thousands, the p-value needs to be corrected for multiple tests, such as Bonferroni, FDR correction, etc. However, p-values ​​also have their problems. For example, the selection of correction methods and thresholds also needs to be carefully chosen.

[0004] Since the characteristics of data vary greatly, each set of data needs to select appropriate threshold parameters. Currently, there is no method to select the optimal solution for differentially expressed genes based on the characteristics of different data, and there is a lack of a unified standard for determining differentially expressed genes.

[0005] Therefore, technicians in this field are committed to developing a method and application for determining differentially expressed genes that can be automatically adjusted according to different data and does not require human intervention, thereby reducing human intervention errors. Summary of the invention

[0006] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is the lack of a unified standard in determining differentially expressed genes based on single-cell transcriptome data, that is, how to develop a method for determining differentially expressed genes that can be automatically adjusted according to different data and does not require human intervention, thereby reducing the error of human intervention.

[0007] To achieve the above object, the present invention provides a method for determining differentially expressed genes in single-cell RNA-seq, comprising the following steps:

[0008] Step 1: Compare the gene expression levels of the two groups of single-cell RNA-seq to determine the expected value of the gene variance after a given expression level, where the noise pollution gene is a gene with a variance lower than the expected value; the signal gene is a gene with a variance higher than the expected value; only retain the signal gene;

[0009] Step 2: Use the t-test differentially expressed gene determination method to calculate the p-value of each gene in the signal genes retained in step 1, and select genes with insignificant p-values ​​as non-differentially expressed genes;

[0010] Step 3: Use the PLS model to construct the differential features between the gene expression level data of the two groups of single-cell RNA-seq in step 1;

[0011] Step 4, then use the PLS model to extract and retain the difference characteristic components between the two groups in the data, remove the components irrelevant to the classification, and obtain the processed data;

[0012] Step 5: By calculating the residual value distribution characteristics of the non-differentially expressed genes in the processed data obtained in step 4, the genes whose residual values ​​are significantly higher than the non-differentially expressed genes obtained in step 2 are differentially expressed genes.

[0013] Furthermore, step 1 also includes: using a piecewise linear regression model to determine the expected value of the gene variance after a given expression level.

[0014] Furthermore, the calculation formula of the piecewise linear regression model is:

[0015] σ 2 / μ 2 =a0+a1 / μ,

[0016] Among them, μ is the average expression level of each gene, σ 2 is the variance of the average expression level of each gene. The expected value of the gene variance after a given expression level is the expected value of the variance of each gene calculated according to the formula σ1 2 , a0 and a1 are regression coefficients, and a linear regression model can be used to get μ and σ 2 estimate.

[0017] Furthermore, the true variance of the gene σ 2 Less than the expected value σ1 2 , then the gene is considered to be a noise pollution gene.

[0018] Furthermore, step 2 also includes: genes with insignificant p-values ​​are genes with the largest p-values ​​in the top 20% and greater than 0.05.

[0019] Furthermore, step 4 also includes: using the minimum mean square error method to extract and retain the difference feature components between the two groups in the data.

[0020] Furthermore, step 5 also includes: using the non-differentially expressed genes obtained in step 2 as a baseline, determining the genes whose average expression levels in each group of the two groups of single-cell RNA-seq gene expression levels exceed the 99% confidence interval of the baseline distribution as the differentially expressed genes that are upregulated in the group.

[0021] Furthermore, for the processed data, since the non-differentially expressed genes basically do not contain differential feature components, their residual values ​​are close to zero.

[0022] Furthermore, genes whose mean expression levels exceeded the 99% confidence interval of the baseline distribution were considered outliers of gene expression.

[0023] The present invention also provides an application of a single-cell RNA-seq differentially expressed gene determination method in comparing differences in cell expression profile characteristics under different conditions.

[0024] Furthermore, the application is suitable for comparing the differences in cell expression profile characteristics between two groups under different conditions.

[0025] In a preferred embodiment of the present invention, the process of the single-cell transcriptome data dynamic modeling algorithm is described in detail.

[0026] The technical effects of the present invention are as follows:

[0027] The present invention solves the problem of lack of unified standards in determining differentially expressed genes based on single-cell transcriptome data. The selection of the determination criteria in the present invention relies on a data-driven adaptive algorithm, which can be automatically adjusted according to different data without the need for human intervention, thereby reducing human intervention errors.

[0028] The traditional determination method needs to consider both the variation range and significance of gene expression. The present invention reflects these two factors with one metric, thereby reducing errors.

[0029] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flowchart of the algorithm operation of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following describes several preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0032] like Figure 1 The algorithm operation flow chart shows the process of the single-cell transcriptome data dynamic modeling algorithm in detail, as follows:

[0033] Step 1: Remove random noise interference genes: Merge cells from two categories, X and Y, together. Based on the average expression level μ and variance σ of each gene 2 Relationship: σ 2 / μ 2 =a0+al / μ, calculate the expected value of each gene variance σ1 2 If the true variance value σ of the gene 2 <σ1 2 , then the gene is considered to be a noise pollution gene.

[0034] Step 2: Use t-test to test the difference of each gene. The top 20% genes with the largest p-value greater than 0.05 are selected as non-differentially expressed genes.

[0035] Step 3: Use PLS to model the classification-related components: Use the PLS method again to extract the distinguishing characteristic components between the two types of samples. The solution of the PLS model is obtained by the NIPALS algorithm. Specifically, (1) the original data X and Y are centered to X0, Y0. (2) The covariance matrix cov(X, Y) = X T Y performs eigenvalue decomposition cov(X, Y) = w∑u. Take the first column of w as w1 and calculate t1 = X0w1. (3) Use t1 to calculate (4) Remove the modeled components from X0 and Y0 (5) Repeat steps 2-4 above N times. N is usually set to 4. (6) Construct a transformation matrix W = [w1, w2, w3, w4].

[0036] Step 4: Use the minimum mean square error method to remove classification-irrelevant components: remove the components in X and Y that are irrelevant to the classification.

[0037] X′=XW PLS1 *pinv(W PLS1 )*X

[0038] Step 5: Find outliers based on the processed data and non-differentially expressed genes and judge them as differentially expressed genes: Use the non-differentially expressed genes determined in step 1 as the baseline, and determine the genes whose average expression level SS in X' exceeds the 99% confidence interval of the baseline SS distribution (outliers) as differentially expressed genes that are up-regulated in class X. Use a similar method to find differentially expressed genes that are up-regulated in class Y.

[0039] SS=X′ 2

[0040]

[0041] Among them, SSgenes represents the mean square expression level SS, CI 99％ (SS nonDEgenes ) represents the 99% confidence interval of the mean square expression distribution of non-differentially expressed genes, and DEgenes represents genes judged to be differentially expressed.

[0042] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for determining differentially expressed genes in single-cell RNA-seq, characterized in that: The method comprises the following steps: Step 1, comparing the gene expression levels of the two groups of single-cell RNA-seq, and determining the expected value of the gene variance after a given expression level, wherein the noise contaminated gene is a gene whose variance is lower than the expected value; The signal gene is a gene whose variance is higher than the expected value; only the signal gene is retained; Step 2, using the t-test differentially expressed gene determination method to calculate the p-value of each gene in the signal gene retained in step 1, and selecting the genes with insignificant p-values ​​as non-differentially expressed genes; Step 3, using the PLS model to construct the differential features between the gene expression level data of the two groups of single-cell RNA-seq in step 1; Step 4, then using the PLS model to extract and retain the difference characteristic components between the two groups in the data, removing components irrelevant to the classification, and obtaining processed data; Step 5, calculating the residual value distribution characteristics of the non-differentially expressed genes in the processed data obtained in step 4, and the gene whose residual value is significantly higher than the non-differentially expressed gene obtained in step 2 is the differentially expressed gene; Among them, the steps of using the PLS to model the classification-related components in step 3 are: specifically: (1) The original data X and Y of the gene expression level data of the two groups of single-cell RNA-seq are centered as X0 and Y0; (2) For the covariance matrix cov(X,Y)=X T Y performs matrix eigenvalue decomposition cov(X,Y)=wΣu, Take the first column of w as w1, and calculate t1=X0w1; (3) Calculate using t1: (4) Remove the modeled components from X0 and Y0 (5) Repeat steps (2) to (4) 4 times; (6) Construct the transformation matrix W = [w1, w2, w3, w4]; In step 4, the removal of classification-irrelevant components uses the minimum mean square error method: removing the classification-irrelevant components in X and Y, X'=X-W PLS1 *pinv(W PLS1 )*X。 2. The method according to claim 1, characterized in that The step 1 also includes: using a piecewise linear regression model to determine the expected value of the gene variance given the expression level.

3. The method according to claim 2, characterized in that The calculation formula of the piecewise linear regression model is: 2 / μ 2 =a0+a1 / μ, Wherein, μ is the average expression level of each gene, and σ 2 is the variance of the average expression level of each gene, and the expected value of the gene variance after the given expression level is the expected value of each gene variance calculated according to the calculation formula of the piecewise linear regression model σ1 2 , the a0 and the a1 are regression coefficients, which can be obtained from the μ and the σ using a linear regression model. 2 estimate.

4. The method according to claim 3, characterized in that The true variance value σ of the gene 2 Less than the estimated value σ1 2 , then the gene is considered to be the noise pollution gene.

5. The method according to claim 1, characterized in that The step 2 also includes: the genes with insignificant p-values ​​are the genes with the largest p-values ​​in the top 20% and greater than 0.

05.

6. The method according to claim 1, characterized in that The step 4 also includes: using the minimum mean square error method to extract and retain the difference feature components between the two groups in the data.

7. The method according to claim 1, characterized in that The step 5 also includes: using the non-differentially expressed genes obtained in step 2 as a baseline, determining the genes whose average expression levels in each group of the two groups of single-cell RNA-seq gene expression levels exceed the 99% confidence interval of the baseline distribution as the differentially expressed genes that are upregulated in the group.

8. The method according to claim 7, characterized in that Genes whose mean expression levels exceeded the 99% confidence interval of the baseline distribution were considered outliers in gene expression.

9. Application of a single-cell RNA-seq differentially expressed gene determination method as described in any one of claims 1 to 4 in comparing differences in cell expression profile features under different conditions.

10. The use according to claim 9, characterized in that The application is suitable for comparing the differences in cell expression profile characteristics between two groups under different conditions.

Images