Brain disease diagnosis model based on multiple peripheral tissue DNA methylation data

Abstract

The invention discloses a brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues, and the model obtains advanced feature output vectors corresponding to the DNA methylation data of the peripheral tissues through the preprocessed methylation data composed of multiple peripheral tissues, and further obtains a classification result. According to the method, multiple peripheral tissue DNA methylation data are used, compared with methylation data of a single source, extraction is easier, the multiple peripheral tissue DNA methylation data can be complemented, and the stability and reliability of a classification result are improved by using the DNA methylation data of multiple peripheral tissues of non-sick people and sick people. By combining the advantages of the DNN in the aspect of feature extraction and the high efficiency of the LightGBM classification algorithm in the classification task, the output result of the brain disease diagnosis model based on the multiple peripheral tissue DNA methylation data is more accurate.

Description

Technical Field

[0001] The present invention relates to the field of medical detection and diagnosis technology, and in particular to a brain disease diagnosis model based on multiple peripheral tissue DNA methylation data. Background Art

[0002] With the rapid development of genomics and epigenetics, DNA methylation, as an important epigenetic regulatory mechanism, has shown great potential in the diagnosis of brain diseases. However, the direct extraction of methylation data from brain tissue faces many challenges, including technical complexity and ethical issues of sampling, which limits its actual clinical application. The methylation data of peripheral tissues is not only easy to extract, but also has significant biological correlations between diseases and brain tissues, thus providing an important research avenue for the diagnosis of brain diseases. However, most disease diagnosis models only use methylation data from a single source (such as peripheral blood), ignoring the complementarity and differences in methylation patterns between different biological samples. This limitation may lead to reduced diagnostic accuracy and limited disease monitoring and prediction capabilities. Summary of the invention

[0003] The present invention discloses a brain disease diagnosis model based on multiple peripheral tissue DNA methylation data to overcome the above technical problems.

[0004] In order to achieve the above object, the technical solution of the present invention is:

[0005] A brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues, wherein the method for establishing the model comprises the following steps:

[0006] S1: Obtain DNA methylation data of various peripheral tissues including those without and with the disease, and perform preprocessing;

[0007] S2: Establish a brain disease diagnosis model based on DNA methylation data of various peripheral tissues, including multiple hidden layers and a prediction layer based on the LightGBM classifier;

[0008] S3: according to the preprocessed DNA methylation data of the peripheral tissue, obtaining a high-level feature output vector corresponding to the DNA methylation data of the peripheral tissue based on the multiple hidden layers;

[0009] S4: Obtain classification results based on the high-level feature output vector and the prediction layer based on the LightGBM classifier; and evaluate the classification results of the brain disease diagnosis model based on multiple peripheral tissue DNA methylation data based on the average accuracy index, F1 score index and area under the ROC curve index of the classification results; and complete the establishment of the brain disease diagnosis model based on multiple peripheral tissue DNA methylation data after the evaluation is qualified.

[0010] Furthermore, the DNA methylation data of the peripheral tissues include DNA methylation data of peripheral blood, DNA methylation data of saliva and DNA methylation data of the oral cavity.

[0011] Furthermore, the formula used to obtain the high-level feature output vector is as follows:

[0012]

[0013] Where: represents the high-level feature output vector of the i-th sample; x i Probes representing DNA methylation data of peripheral tissues of the i-th sample; Dense M (x i ) represents the processing result of the dense layer on the DNA methylation data of the peripheral tissue of the i-th sample; Relu(·) represents the activation function; Dropout(·) represents the function used to prevent the model from overfitting; M represents the number of nodes in the hidden layer; i represents the patient sample index value.

[0014] Furthermore, the formula used to obtain the predicted diagnosis result of brain disease is as follows:

[0015]

[0016] Where: Represents the high-level feature output vector of the i-th sample; r is the number of the regression tree model of the LightGBM model; s is the total number of regression trees of the LightGBM model; f r represents the output of the rth independent regression tree model with leaf scores; F represents the space of classification and regression trees;

[0017] Among them, the objective function of the output of the rth independent regression tree model with leaf score is:

[0018]

[0019] Where: L(f r ) represents the objective function of the output of the rth independent regression tree model with leaf scores, that is, the objective function for measuring the prediction performance of the model; i represents the patient sample index value; n represents the total number of patient samples; represents a differentiable loss function; r represents the number of the regression tree of the LightGBM model; s represents the total number of regression trees of the LightGBM model; Ω(f r ) is the regularization term;

[0020]

[0021] Where: γ and λ are constants used to control the degree of regularization; K is the number of leaf nodes in each regression tree model; w r is the value of the score on each leaf.

[0022] Furthermore, the classification results of brain diseases include: the providers of DNA methylation data of multiple peripheral tissues are non-diseased people and the providers of DNA methylation data of multiple peripheral tissues are already affected people, thereby being able to obtain the number of true positives, the number of true negatives, the number of false positives and the number of false negatives.

[0023] Beneficial effects: The brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues of the present invention inputs the preprocessed methylation data of multiple peripheral components into the brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues that combines DNN with LightGBM classifier, obtains high-level feature output vectors corresponding to the DNA methylation data of the peripheral tissues, and then obtains classification results. The present invention uses DNA methylation data of multiple peripheral tissues, which is easier to extract than methylation data from a single source, and the DNA methylation data of multiple peripheral tissues can complement each other. By using DNA methylation data of multiple peripheral tissues of non-diseased and ill people, the stability and reliability of the classification results are improved. By combining the advantages of DNN in feature extraction and the high efficiency of the LightGBM classification algorithm in classification tasks, the output results of the brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0025] Figure 1 A flow chart of a method for establishing a brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues according to the present invention;

[0026] Figure 2 A flow chart of a method for establishing a brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues in an embodiment of the present invention;

[0027] Figure 3 This is a structural diagram of a disease diagnosis model based on the DNN-LightGBM algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] This example introduces a brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues, such as Figure 1 and Figure 2 As shown,

[0030] S1: obtaining DNA methylation data of multiple peripheral tissues and performing preprocessing; wherein the DNA methylation data of peripheral tissues includes: DNA methylation data of peripheral blood, DNA methylation data of saliva and DNA methylation data of oral cavity;

[0031] Specifically, this embodiment performs preprocessing operations such as normalization on the methylation data of various peripheral tissues of non-diseased persons and patients to provide high-quality data input for model training.

[0032] S2: Establish a brain disease diagnosis model based on DNA methylation data of various peripheral tissues, including multiple hidden layers and a prediction layer based on the LightGBM classifier; Figure 3 shown.

[0033] S3: according to the preprocessed DNA methylation data of the peripheral tissue, obtaining a high-level feature output vector corresponding to the DNA methylation data of the peripheral tissue based on the multiple hidden layers, so as to learn the high-level feature representation in the DNA methylation data of each peripheral tissue;

[0034] Specifically, given a sample data set D = {(x i ,y i )}, where x i Probes representing DNA methylation data of peripheral tissues, y i is the corresponding binary label, with 0 representing no disease and 1 representing disease. i As the input of the brain disease diagnosis model DNN based on DNA methylation data of various peripheral tissues, the extracted high-level feature output vector is obtained through the input layer and L hidden layers.

[0035] Preferably, the formula used to obtain the high-level feature output vector is as follows:

[0036]

[0037] Where: represents the high-level feature output vector of the i-th sample; x i Probes representing DNA methylation data of peripheral tissues of the i-th sample; Dense M (x i ) represents the processing result of the dense layer on the DNA methylation data of the peripheral tissue of the i-th sample; Relu(·) represents the activation function, which is a rectified linear unit (Rectified LinearUnit), which takes the output of the dense layer, maps all negative values ​​to zero, and keeps the positive values ​​unchanged, thereby introducing nonlinear features; Dropout(·) represents a function used to prevent model overfitting, which is an existing regularization technology used to prevent model overfitting; M represents the number of nodes in the hidden layer; i represents the patient sample index value; in this embodiment, each sample includes DNA methylation data of multiple peripheral tissues. Optionally, a sample can be composed of DNA methylation data of peripheral blood, DNA methylation data of saliva, and DNA methylation data of the oral cavity.

[0038] Specifically, M is the number of nodes in the hidden layer l (l = 1, 2...L), L is the total number of hidden layers, and l is the index number of the hidden layer. The dense layer extracts relevant patterns from the probe sequences of the DNA methylation data of the peripheral tissue of the sample and outputs them as features, where the relevant patterns are the features and structures in the input data, such as the distribution and status of methylation sites. The rectified linear unit Relu (·) takes the output of the dense layer and maps all negative values ​​to zero, thereby introducing nonlinear features. Dropout is a technique used to avoid overfitting, in which some neurons and their connections are deleted (assigned to zero) during training.

[0039] S4: According to the high-level feature output vector and the prediction layer based on the LightGBM classifier, the LightGBM classification algorithm is used to obtain the classification result; and based on the average accuracy index, F1 score index and area under the ROC curve index of the classification result, the brain disease diagnosis model based on multiple peripheral tissue DNA methylation data is evaluated; after the evaluation is qualified, the brain disease diagnosis model based on multiple peripheral tissue DNA methylation data is established.

[0040] The classification results of brain diseases include: the providers of DNA methylation data of multiple peripheral tissues are non-diseased persons and the providers of DNA methylation data of multiple peripheral tissues are already-diseased persons, thereby being able to obtain the number of true positives, the number of true negatives, the number of false positives and the number of false negatives;

[0041] Specifically, the features and labels extracted by DNN are input into the LightGBM classifier. As an efficient classification and regression tree ensemble learning algorithm, LightGBM generates a basic regression tree model in each step and adds it to the entire model, integrating their prediction results to improve classification accuracy. Assuming that the LightGBM model consists of s regression trees, the optimization objective function is given by equation (2):

[0042]

[0043] Where: Represents the high-level feature output vector of the i-th sample; r is the number of the regression tree model of the LightGBM model; s is the total number of regression trees of the LightGBM model; f r represents the output of the rth independent regression tree model with leaf scores; F represents the space of classification and regression trees;

[0044] Among them, r = 1 represents the first tree, f r corresponds to an independent regression tree with leaf fractions. The objective function is defined by the equation:

[0045]

[0046] Where: L(f r ) represents the objective function used to measure the output of the r-th independent regression tree model with leaf scores, that is, the objective function for measuring the prediction performance of the model; i represents the patient sample index value; n represents the total number of patient samples; represents a differentiable loss function; r represents the number of the regression tree of the LightGBM model; s represents the total number of regression trees of the LightGBM model; Ω(f r ) is the regularization term;

[0047] Among them, the differentiable loss function Used to measure the true label y i and predicted labels The difference between the regularization term Ω(f r ), which can constrain the complexity of the model to avoid overfitting, is expressed as:

[0048]

[0049] Where: γ and λ are constants used to control the degree of regularization; K is the number of leaf nodes in each regression tree model; w r is the value of the score on each leaf;

[0050] Specifically, this embodiment uses 10-fold cross validation to evaluate the performance of the classification algorithm. This process divides the data set into 10 folds of approximately equal size. One fold is used as a test data set alone, and the remaining 9 folds are used as training data sets. In order to optimize the best hyperparameters of the classification model, a grid search method and a 3-fold internal cross validation are used. This internal cross validation is only performed on the training set. Finally, the model is fitted on the training set with the parameters that achieve the best performance and evaluated on the test set.

[0051] Specifically, this embodiment evaluates the performance of the classification model based on the average accuracy index (ACC), F1 score index (F1) and ROC curve area index (AUC) of the classification results. In this embodiment, when all three indicators can meet the requirements, it means that the model can meet the usage requirements.

[0052] Among them, the calculation formula of the average accuracy index of the classification result is as follows:

[0053]

[0054] The calculation formula of F1 score indicator is as follows:

[0055]

[0056] Wherein: ACC represents the average accuracy of the classification results; TP represents the number of true positives; TN represents the number of true negatives; FP represents the number of false positives; FN represents the number of false negatives; F1-score represents the result of F1 score.

[0057] The calculation method of the area under the ROC curve indicator is as follows:

[0058] First, calculate the true positive rate and false positive rate. The formula is as follows:

[0059]

[0060] Where: TPR represents the true positive rate; FPR represents the false positive rate;

[0061] Then, the FPR and TPR values ​​corresponding to each threshold are used as coordinate points, and these points are plotted in the coordinate system and connected to form the ROC curve. Then, the area under the ROC curve is calculated, that is, the area under the curve between the true positive rate and the false positive rate.

[0062] Specifically, deep learning is a promising technology in machine learning that attempts to create and simulate neural networks in the human brain to learn and interpret data. It is possible to form an abstraction of high-level representation by appropriately combining low-level features with the objective function. High-level features are more abstract features further extracted from low-level features. These features are more relevant to the semantic content of the disease, so that the rules of the data can be found. Therefore, feature selection is important. Among them, low-level features are closer to the original data and are the methylation levels of single CpG sites extracted directly from the methylation data. These are the most basic methylation data, reflecting the methylation status of specific locations in the genome. The distinction between low-level features and high-level features lies in their level of abstraction and the richness of semantic information. Low-level features are closer to the original data, while high-level features contain more abstract and semantic information, which can better represent and distinguish different disease states.

[0063] In this example, we studied DNA methylation data from peripheral blood, saliva, and oral cavity, using methylation information from a variety of biological samples. A deep neural network (DNN) was used as a feature extractor to preprocess and extract features from multi-tissue methylation data. Through multiple layers of nonlinear transformations, DNN can automatically learn high-level feature representations in the data and improve the generalization ability of the model. Subsequently, the features and labels extracted by DNN were input into the LightGBM classifier. Finally, a model for diagnosing brain diseases using multi-tissue DNA methylation data was established.

[0064] This embodiment integrates DNN with the LightGBM algorithm to establish a model that uses multi-tissue methylation data and adopts a lightweight gradient boosting classifier, that is, a two-classification brain disease diagnosis model that combines the LightGBM classifier with the DNN architecture, that is, a brain disease diagnosis model based on multiple peripheral tissue DNA methylation data. First, the model uses DNN to automatically extract high-level abstract features. Then, the high-level abstract features extracted by DNN are used together with the class label as the input of the LightGBM classifier. In the DNN architecture, the LightGBM predictor is used instead of the neural network-based predictor to improve the classification accuracy. This model combines the advantages of DNN in automatic feature extraction and complex pattern recognition, and the excellent performance of LightGBM in efficient classification and regression tasks, and uses multi-tissue methylation data from peripheral blood, saliva and oral cavity. By comprehensively analyzing multiple biological samples, the limitations of single data in complex disease diagnosis are overcome, and the stability and accuracy of diagnosis are significantly improved. This embodiment can provide a more efficient and accurate new way for early detection and treatment of diseases. Taking epilepsy as an example, the model can use multi-tissue DNA methylation data of epilepsy patients and their healthy control populations for efficient and accurate disease diagnosis. Through predictive analysis, doctors can identify potential epilepsy patients at an early stage, providing strong support for subsequent disease progression monitoring and personalized treatment plan formulation. In addition, the model also opens up new avenues for the diagnosis and research of other complex brain diseases (such as Alzheimer's disease), heralding broad application prospects in the field of precision medicine in the future.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A brain disease diagnosis model based on DNA methylation data of multiple peripheral tissues, characterized in that: The model building method includes the following steps: S1: Obtain DNA methylation data of various peripheral tissues including those without and with the disease, and perform preprocessing; S2: Establish a brain disease diagnosis model based on DNA methylation data of various peripheral tissues, including multiple hidden layers and a prediction layer based on the LightGBM classifier; S3: according to the preprocessed DNA methylation data of the peripheral tissue, obtaining a high-level feature output vector corresponding to the DNA methylation data of the peripheral tissue based on the multiple hidden layers; S4: Obtain classification results based on the high-level feature output vector and the prediction layer based on the LightGBM classifier; and evaluate the classification results of the brain disease diagnosis model based on multiple peripheral tissue DNA methylation data based on the average accuracy index, F1 score index and area under the ROC curve index of the classification results; and complete the establishment of the brain disease diagnosis model based on multiple peripheral tissue DNA methylation data after the evaluation is qualified.

2. A brain disease diagnosis model based on multiple peripheral tissue DNA methylation data according to claim 1, characterized in that: The DNA methylation data of the peripheral tissues include DNA methylation data of peripheral blood, DNA methylation data of saliva and DNA methylation data of the oral cavity.

3. A brain disease diagnosis model based on multiple peripheral tissue DNA methylation data according to claim 1, characterized in that: The formula used to obtain the high-level feature output vector is as follows: Where: represents the high-level feature output vector of the i-th sample; x i Probes representing DNA methylation data of peripheral tissues of the i-th sample; Dense M (x i ) represents the processing result of the dense layer on the DNA methylation data of the peripheral tissue of the i-th sample; Relu(·) represents the activation function; Dropout(·) represents the function used to prevent the model from overfitting; M represents the number of nodes in the hidden layer; i represents the patient sample index value.

4. A brain disease diagnosis model based on multiple peripheral tissue DNA methylation data according to claim 1, characterized in that: The formula used to obtain the predicted diagnosis of brain diseases is as follows: Where: Represents the high-level feature output vector of the i-th sample; r is the number of the regression tree model of the LightGBM model; s is the total number of regression trees of the LightGBM model; f r represents the output of the rth independent regression tree model with leaf scores; F represents the space of classification and regression trees; Among them, the objective function of the output of the rth independent regression tree model with leaf score is: Where: L(f r ) represents the objective function of the output of the rth independent regression tree model with leaf scores, that is, the objective function for measuring the prediction performance of the model; i represents the patient sample index value; n represents the total number of patient samples; represents a differentiable loss function; r represents the number of the regression tree of the LightGBM model; s represents the total number of regression trees of the LightGBM model; Ω(f r ) is the regularization term; Where: γ and λ are constants used to control the degree of regularization; K is the number of leaf nodes in each regression tree model; w r is the value of the score on each leaf.

5. A brain disease diagnosis model based on multiple peripheral tissue DNA methylation data according to claim 1, characterized in that: The classification results include: the providers of DNA methylation data of multiple peripheral tissues are non-diseased persons and the providers of DNA methylation data of multiple peripheral tissues are already infected persons. According to the classification results, the number of true positives, the number of true negatives, the number of false positives and the number of false negatives can be obtained.

Images