DNA methylation prediction method integrating multi-omics features

Abstract

The application discloses a DNA methylation prediction method integrating multi-omics features, comprising determining a to-be-predicted CpG site, acquiring a multi-omics feature set of the CpG site in para-cancer tissue, acquiring a methylation feature set of the CpG site in cancer tissue, respectively calculating correlation coefficients between the methylation features of the cancer tissue CpG site and the miRNA, mRNA and methylation features of the para-cancer tissue based on Pearson correlation coefficients, respectively selecting K miRNAs, Q mRNAs and L methylation features according to the values of the correlation coefficients, and constructing a multi-omics correlation feature set, constructing a DNA methylation prediction model based on a deep neural network, training the DNA methylation prediction model according to the multi-omics correlation feature set, calculating evaluation indexes of the trained DNA methylation prediction model, acquiring an evaluated DNA methylation prediction model when the evaluation indexes meet threshold values, and predicting the methylation of the cancer tissue according to the evaluated DNA methylation prediction model. The DNA methylation prediction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of DNA methylation prediction, and particularly relates to a DNA methylation prediction method integrating multi-omics features. BACKGROUND

[0002] DNA methylation of cancer tissues and paracancer tissues is closely related to the occurrence and development of cancer, and analyzing the variation of DNA methylation helps to reveal the molecular biological mechanism of cancer. It is difficult to predict DNA methylation by using only a single omics, and with the rapid development of sequencing technology, researchers have obtained a large amount of multi-omics data, and there is a significant correlation between DNA methylation of cancer tissues and paracancer tissues and multi-omics data, so it is necessary to predict DNA methylation by integrating multi-omics features. In summary, DNA methylation data is very important for cancer research, and existing studies only use a single omics to predict DNA methylation, and a single omics feature provides relatively single biological information, and the prediction effect of DNA methylation data is not good. SUMMARY

[0003] The present application provides a DNA methylation prediction method integrating multi-omics features to overcome the above technical problems.

[0004] A DNA methylation prediction method integrating multi-omics features, comprising,

[0005] Step one, determining a to-be-predicted CpG site, obtaining a multi-omics feature set of the CpG site in the paracancer tissue, the multi-omics feature set comprising a miRNA feature set, an mRNA feature set, and a methylation feature set, and obtaining a methylation feature set of the CpG site in the cancer tissue,

[0006] Step two, calculating the correlation coefficients between the methylation features of the cancer tissue CpG site and the miRNA features, mRNA features, and methylation features of the paracancer tissue based on Pearson correlation coefficients, selecting K miRNA features from the miRNA feature set according to the values of the correlation coefficients between the methylation features of the cancer tissue CpG site and the miRNA features of the paracancer tissue, selecting Q mRNA features from the mRNA feature set according to the values of the correlation coefficients between the methylation features of the cancer tissue CpG site and the mRNA features of the paracancer tissue, and selecting L methylation features from the methylation feature set according to the values of the correlation coefficients between the methylation features of the cancer tissue CpG site and the methylation features of the paracancer tissue,

[0007] Step three, constructing a multi-omics correlation feature set according to the K miRNA features, the Q mRNA features, and the L methylation features, constructing a DNA methylation prediction model based on a deep neural network, training the DNA methylation prediction model based on the multi-omics correlation feature set, calculating an evaluation index of the trained DNA methylation prediction model, and obtaining an evaluated DNA methylation prediction model when the evaluation index meets a threshold, and predicting the methylation of the cancer tissue based on the evaluated DNA methylation prediction model.

[0008] Preferably, the calculating the correlation coefficients between the methylation features of the CpG sites of the cancer tissues and the miRNA features, the mRNA features, and the methylation features of the para-cancer tissues based on the Pearson correlation coefficients comprises calculating the correlation coefficients according to formula (1),

[0009]

[0010] wherein x i represents the CpG site methylation feature value, the mRNA feature value, or the miRNA feature value of the para-cancer tissue of the i-th sample in the multi-omics feature set, represents the feature mean of all samples of the CpG site; y i represents the methylation feature value of the corresponding CpG site in the cancer tissue of the i-th sample in the multi-omics feature set, represents the methylation feature mean of all samples of the CpG site, and n is the number of samples.

[0011] Preferably, the constructing the DNA methylation prediction model based on the deep neural network comprises that the DNA methylation prediction model comprises v input neurons, k hidden layer neurons, and h output layer neurons, and the input received by the q-th neuron in the hidden layer is:

[0012]

[0013] The output of the q-th neuron in the hidden layer is:

[0014]

[0015] wherein the weight between the p-th neuron in the input layer and the q-th neuron in the hidden layer is w pq , x i is an input vector, the input received by the r-th neuron in the output layer is a feature b j of the multi-omics correlation feature set, n is the number of samples, and the weight between the h-th neuron in the hidden layer and the r-th neuron in the output layer is e hr .

[0016] Preferably, the evaluation index of the calculation of the trained DNA methylation prediction model comprises the absolute value of the Pearson correlation coefficient calculated according to formula (4), the mean absolute error of the Pearson correlation coefficient calculated according to formula (5),

[0017]

[0018] In the formula y i. respectively represent the predicted DNA methylation value and the actual DNA methylation feature value of the i th sample, respectively represent the predicted mean value and the actual mean value, respectively represent the standard deviation; y ij respectively represent the predicted DNA methylation value and the actual DNA methylation value of the j th feature of the i th sample.

[0019] Preferably, the DNA methylation prediction model can also optimize the parameters of the DNA methylation prediction model through ten-fold cross-validation.

[0020] The present application provides a DNA methylation prediction method integrating multi-omics features, based on a feature selection method, extracts multi-omics features related to the target CpG site, and then establishes a model for predicting the DNA methylation level of cancer tissue integrating multi-omics features. By comparing performance indicators such as mean absolute error, analyzing the influence of key parameters such as neural network structure and feature number on the performance of the DNA methylation prediction model, optimizing the model parameters, and improving the accuracy of DNA methylation prediction. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0022] Figure 1 is a flow chart of the method of the present application;

[0023] Figure 2 is a process diagram of the present application. DETAILED DESCRIPTION

[0024] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0025] Figure 1 For the method flowchart of the present application, as shown in Figure 1 , the method of the present embodiment can include:

[0026] Step one, determining a CpG site to be predicted, obtaining a multi-omics feature set of the CpG site in the para-cancer tissue, the multi-omics feature set including an miRNA feature set, an mRNA feature set, a methylation feature set, obtaining a methylation feature set of the CpG site in the cancer tissue,

[0027] Step two, calculating the correlation coefficients between the methylation feature of the cancer tissue CpG site and the miRNA feature, mRNA feature and methylation feature of the para-cancer tissue based on the Pearson correlation coefficient, selecting K miRNA features from the miRNA feature set according to the value of the correlation coefficient between the methylation feature of the cancer tissue CpG site and the miRNA feature of the para-cancer tissue, selecting Q mRNA features from the mRNA feature set according to the value of the correlation coefficient between the methylation feature of the cancer tissue CpG site and the mRNA feature of the para-cancer tissue, and selecting L methylation features from the methylation feature set according to the value of the correlation coefficient between the methylation feature of the cancer tissue CpG site and the methylation feature of the para-cancer tissue,

[0028] Step three, constructing a multi-omics correlation feature set according to the K miRNA features, Q mRNA features and L methylation features, constructing a DNA methylation prediction model based on a deep neural network, training the DNA methylation prediction model according to the multi-omics correlation feature set, calculating the evaluation index of the trained DNA methylation prediction model, obtaining the evaluated DNA methylation prediction model when the evaluation index meets the threshold, and predicting the methylation of the cancer tissue according to the evaluated DNA methylation prediction model.

[0029] Based on the above scheme, the multi-omics features related to the target CpG site are extracted by the feature selection method, and then a model integrating multi-omics features is established to predict the DNA methylation level of the cancer tissue. By comparing performance indicators such as mean absolute error, the influence of key parameters such as neural network structure and feature quantity on the performance of the DNA methylation prediction model is analyzed, the model parameters are optimized, and the accuracy of DNA methylation prediction is improved.

[0030] Step one, determine the CpG site to be predicted, obtain the multi-omics feature set of the CpG site in the para-cancer tissue, the multi-omics feature set includes the miRNA feature set, the mRNA feature set, the methylation feature set, obtain the methylation feature set of the CpG site in the cancer tissue,

[0031] Specifically, first, the multi-omics features are extracted by using a feature selection method:

[0032] The miRNA data matrix is defined as I i =(miRNA i1 , miRNA i2 ,..., miRNA ia ),

[0033] The mRNA data matrix is N i =(mRNA i1 , mRNA i2 ,..., mRNA ib ),

[0034] The methylation data matrix is M i =(CpG i1 , CPG i2 ,..., CpG ic ),

[0035] Where i represents a sample (a total of n samples), and j represents a feature (a total of a miRNA data, a total of b mRNA data, and a total of c methylation data).

[0036] Step two, based on the Pearson correlation coefficient, the correlation coefficients between the methylation features of the cancer tissue CpG site and the miRNA features, mRNA features, and methylation features of the para-cancer tissue are calculated,

[0037] The correlation coefficients between the methylation features of the cancer tissue CpG site and the miRNA features, mRNA features, and methylation features of the para-cancer tissue are calculated based on the Pearson correlation coefficient, which includes calculating the correlation coefficient according to formula (1),

[0038]

[0039] Where x i represents the CpG site methylation feature value, mRNA feature value, or miRNA feature value of the para-cancer tissue of the i-th sample in the multi-omics feature set, represents the feature mean of all samples of the CpG site; y ia methylation feature value of a corresponding CpG site in a cancer tissue of an i-th sample in a multi-omics feature set, represents a methylation feature mean of all samples of the CpG site, and n is the number of samples.

[0040] According to the correlation coefficient values of the methylation features of the CpG sites in the cancer tissues and the miRNA features of the para-cancer tissues, K miRNA features are selected from the miRNA feature set, according to the correlation coefficient values of the methylation features of the CpG sites in the cancer tissues and the mRNA features of the para-cancer tissues, Q mRNA features are selected from the mRNA feature set, and according to the correlation coefficient values of the methylation features of the CpG sites in the cancer tissues and the methylation features of the para-cancer tissues, L methylation features are selected from the methylation feature set, specifically,

[0041] After calculating the correlation coefficients between the CpG sites in the cancer tissues, i.e., target sites, and the three kinds of omics features, for each target site, the top K miRNA (top miRNA) with high correlation coefficient values are selected according to formula (2), the top Q mRNA (top mRNA) are selected according to formula (3), and the top L methylation (top Methy) features are selected according to formula (4):

[0042]

[0043]

[0044]

[0045] wherein CpG target j represents a j-th predicted target site, m represents the number of target sites, miRNA1…miRNA K represents the top K miRNA features most correlated with the CpG target j site, mRNA1…mRNA Q represents the top Q mRNA features most correlated with the CpG target j site, and CpG1…CpG L represents the top L methylation features most correlated with the CpG target j site.

[0046] Step three, constructing a multi-omics correlation feature set according to the K miRNA features, the Q mRNA features, and the L methylation features, constructing a DNA methylation prediction model based on a deep neural network, training the DNA methylation prediction model based on the multi-omics correlation feature set, and establishing the DNA methylation prediction model: taking the multi-omics correlation features (miRNA1…miRNA K , mRNA1…mRNA QCpG1…CpG L ) as input data, the dimension of the input vector is v, and the dimension of the output vector is h.

[0047] The DNA methylation prediction model based on the deep neural network comprises v input neurons, k hidden layer neurons, and h output layer neurons, and the input received by the qth neuron in the hidden layer is:

[0048]

[0049] The output of the qth neuron in the hidden layer is:

[0050]

[0051] wherein the weight between the pth neuron in the input layer and the qth neuron in the hidden layer is w pq , x i is the input vector, and the input received by the rth neuron in the output layer is the features b j of the multi-omics related feature set, n is the number of samples, and the weight between the hth neuron in the hidden layer and the rth neuron in the output layer is e hr .

[0052] The evaluation index of the trained DNA methylation prediction model is calculated,

[0053] The evaluation index of the trained DNA methylation prediction model comprises the absolute value of the Pearson correlation coefficient calculated according to formula (7) and the mean absolute error of the Pearson correlation coefficient calculated according to formula (8),

[0054]

[0055]

[0056] wherein y i. respectively represent the predicted DNA methylation value and the actual DNA methylation feature value of the ith sample, respectively represent the predicted mean value and the actual mean value, respectively represent the standard deviation; y ij respectively represent the predicted DNA methylation value and the actual DNA methylation value of the jth feature of the ith sample.

[0057] When the evaluation index meets the threshold, the evaluated DNA methylation prediction model is obtained, and the methylation of the cancer tissue is predicted according to the evaluated DNA methylation prediction model. The DNA methylation prediction model can also optimize the parameters of the DNA methylation prediction model through ten-fold cross-validation, specifically, the data set is divided into 10 equal and mutually exclusive subsets, and the consistency of the data is maintained during the division, the union of 9 subsets is used as the training set each time, and the remaining is used as the test set, 10 times of training are performed, and finally the average value of 10 results is taken.

[0058] In the mathematical model studied in this embodiment, the multi-omics data in the substitute tissue is used to predict the DNA methylation data in the target tissue, which can be divided into the following technical steps, and the flowchart is as shown in Figure 2

[0059] The miRNA, mRNA and DNA methylation data of the cancer tissue and the para-cancer tissue are matched and preprocessed.

[0060] The multi-omics data is extracted and fused using a feature selection strategy based on correlation.

[0061] A mathematical model of DNA methylation between the para-cancer tissue and the cancer tissue is established.

[0062] The multi-omics integrated data of the substitute tissue is used to predict the DNA methylation data of the target tissue.

[0063] The influence of the number of layers, the number of neurons and the feature dimension of the deep learning model on the performance of the model is analyzed, and the model parameters are optimized through ten-fold cross-validation, that is, the data set is divided into 10 equal and mutually exclusive subsets, and the consistency of the data is maintained during the division, the union of 9 subsets is used as the training set each time, and the remaining is used as the test set, 10 times of training are performed, and finally the average value of 10 results is taken.

[0064] The absolute value (R) of the Pearson correlation coefficient and the mean absolute error (MAE, mean absolute error) are used to evaluate the prediction performance of the model.

[0065] The overall beneficial effects are:

[0066] The application provides a DNA methylation prediction method integrating multi-omics features, based on a feature selection method, multi-omics features related to a target CpG site are extracted, and then a model integrating multi-omics features for predicting the DNA methylation level of a cancer tissue is established, the influence of key parameters such as the neural network structure and the feature quantity on the performance of the DNA methylation prediction model is analyzed by comparing performance indicators such as the mean absolute error, the model parameters are optimized, and the accuracy of the DNA methylation prediction is improved. ​

[0067] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can still be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting DNA methylation integrating multiple omics features, characterized in that, include, Step 1: Identify the CpG sites to be predicted and obtain the multi-omics feature set of CpG sites in adjacent normal tissue. The multi-omics feature set includes miRNA feature set, mRNA feature set, and methylation feature set. Obtain the methylation feature set of CpG sites in cancer tissue. Step 2: Based on the Pearson correlation coefficient, calculate the correlation coefficients between the methylation characteristics of CpG sites in cancer tissue and the miRNA, mRNA, and methylation characteristics of adjacent normal tissue. Based on the correlation coefficients between the methylation characteristics of CpG sites in cancer tissue and the miRNA characteristics of adjacent normal tissue, select K miRNA features from the miRNA feature set. Based on the correlation coefficients between the methylation characteristics of CpG sites in cancer tissue and the mRNA characteristics of adjacent normal tissue, select Q mRNA features from the mRNA feature set. Based on the correlation coefficients between the methylation characteristics of CpG sites in cancer tissue and the methylation characteristics of adjacent normal tissue, select L methylation features from the methylation feature set. The calculation of the correlation coefficients between the methylation characteristics of CpG sites in cancerous tissue and the miRNA characteristics, mRNA characteristics, and methylation characteristics of adjacent normal tissue based on the Pearson correlation coefficient includes calculating the correlation coefficients according to formula (1). （1） in The first representative of multi-omics feature set The methylation signature of CpG sites, mRNA signature, or miRNA signature of adjacent non-cancerous tissue in a sample. This represents the characteristic mean of all samples at this CpG site; The first representative of multi-omics feature set Methylation signatures of CpG sites in individual cancer tissue samples This represents the mean methylation signature of all samples at this CpG site. n The number of samples; Step 3: Construct a multi-omics related feature set based on K miRNA features, Q mRNA features, and L methylation features. Build a DNA methylation prediction model based on a deep neural network. Train the DNA methylation prediction model using the multi-omics related feature set. Calculate the evaluation index of the trained DNA methylation prediction model. When the evaluation index meets the threshold, obtain the evaluated DNA methylation prediction model. Predict the methylation of cancer tissue based on the evaluated DNA methylation prediction model. The DNA methylation prediction model constructed based on deep neural networks includes the following: v One input neuron, k Hidden layer neurons, h The output layer neuron, the hidden layer neuron q The input received by each neuron is: (2) Hidden layer q The output of each neuron is: (3) Among them, the input layer is the first p The first neuron and the hidden layer q The weights between the neurons are: w pq , Given the input vector, the output layer is the [missing information]. r The input received by each neuron is a feature set related to multiple omics. b j , n Let be the number of samples, and let be the number of neurons in the hidden layer h and the number of neurons in the output layer h. r The weights between the neurons are: ; The evaluation metrics for the computationally trained DNA methylation prediction model include the absolute value of the Pearson correlation coefficient calculated according to formula (4) and the mean absolute error of the Pearson correlation coefficient calculated according to formula (5). （4） （5） In the formula , Representing the first i Predicted DNA methylation values ​​and actual DNA methylation characteristics for each sample. These represent the predicted mean and the actual mean, respectively. These represent the predicted standard deviation and the actual standard deviation, respectively. , These represent the predicted DNA methylation value and the actual DNA methylation value of the j-th feature of the i-th sample, respectively.

2. The DNA methylation prediction method integrating multi-omics features according to claim 1, characterized in that, The DNA methylation prediction model was optimized by using 10-fold cross-validation.

Images