CN116705159A - Screening method for methylation markers, method and device for identifying methylation features - Google Patents

Abstract

The application provides a screening method of methylation markers, a method and a device for identifying methylation characteristics, and relates to the technical field of biomarkers. The screening method includes (a) obtaining sequencing reads for each candidate differential methylation region of a feature sample and a control sample; calculating the alpha value of each read; the alpha value is the ratio of the number of methylated cytosines in each read to the number of all methylation sites; (b) And acquiring an alpha distribution model of the candidate differential methylation region, and taking the candidate differential methylation region and/or at least one methylation site contained in the candidate differential methylation region as a methylation marker if the alpha distribution model accords with a set principle. The application relieves the problem that the specificity and the sensitivity of the methylation marker in the prior art are to be improved.

Description

Screening method for methylation markers, method and device for identifying methylation features

technical field

The present invention relates to the technical field of biomarkers, in particular to a screening method for methylation markers, a method and a device for identifying methylation features.

Background technique

A large number of studies in recent years have shown that methylation, as an important aspect of gene epigenetics, is closely related to the occurrence and development of tumors, and can be used as a marker for early diagnosis of tumors. Traditionally, methylation in tumors and normal tissues Methylation-specific research is based on differential sites (DMPs) and differential regions (DMRs). These methods are based on the beta value of methylation: at a CpG site, beta = methylation-modified cytosine Quantity/(number of methylated cytosine + number of unmethylated cytosine), studies have shown that on CpG islands, the beta value of tumor tissue will be higher than that of normal tissue, such as application number CN202010326209. 2. Chinese patent application titled "Breast Cancer Methylation Biomarkers and Screening Method Based on TCGA Database", etc., the methylation markers of tumors can be obtained through the difference of beta value. However, when the content of tumor cells is relatively low, such as samples from blood or urine, the above methods are not sensitive to tumor signals and cannot screen out tumors well. Therefore, how to improve the detection ability of methylated markers in tumors It is a problem that needs to be improved at present.

In view of this, the present invention is proposed.

Contents of the invention

The first object of the present invention is to provide a method for screening methylation markers, and the methylation markers screened by the screening method have higher sensitivity and specificity. Based on the screening method, the present invention also provides bladder cancer methylation markers, a method for establishing a model for identifying methylation features, a method for identifying methylation features, and their applications, so as to increase the number of methylated regions and/or positions. Point-to-tumor signature recognition and prediction capabilities.

In order to solve the problems of the technologies described above, the present invention adopts the following technical solutions:

According to one aspect of the present invention, a method for screening methylation markers is provided, comprising:

(a) Obtain the sequencing reads of each candidate differentially methylated region of the characteristic sample and the control sample; calculate the alpha value of each read; the alpha value is the number of methylated cytosines in each read and all methylated positions The ratio of the number of points;

(b) Obtain the alpha distribution model of the candidate differentially methylated region, if the alpha distribution model meets the following principles, the candidate differentially methylated region, and/or, at least one of the candidate differentially methylated regions Methylation sites as methylation markers:

The alpha value of at least 80% of the reads in the control sample ≤ the hypomethylation threshold;

The alpha value of at least 30% of the reads of the feature sample ≥ the high methylation threshold;

High methylation threshold - low methylation threshold ≥ N, N is at least 0.5;

The high methylation thresholds and/or low methylation thresholds in the principles that the alpha distribution models of the candidate differentially methylated regions conform to are the same or different.

According to one aspect of the present invention, a bladder cancer methylation marker is provided, including at least one of 1-13 in the following table; wherein, the marker corresponding to each item in the following table includes the corresponding Methylated sites and/or regions containing corresponding methylated sites:

Based on the hg19 reference genome sequence.

According to one aspect of the present invention, there is also provided a method for establishing a model for identifying methylation features, comprising: using the differentially methylated region obtained in step (b) of the above screening method as input data, using a one-dimensional volume Build a model using a product neural network, and use a deep learning framework for model training and prediction.

According to one aspect of the present invention, a method for identifying methylation features is also provided, including using the model established by the above establishment method to identify the methylation features of the sequencing reads of the sample to be tested.

According to one aspect of the present invention, a methylation feature recognition device is also provided, the device contains a prediction module, and the prediction module is used to input the methylation pattern of the reads of the differentially methylated region into the model established by the above establishment method , to obtain the methylation feature of the sample to be tested.

According to one aspect of the present invention, there is also provided a computer-readable medium, which stores a computer program, and when the computer program is processed and executed, realizes the model established by the above model establishment method.

According to one aspect of the present invention, a system for predicting, diagnosing or assisting in diagnosing cancer is also provided, including at least any one of the following:

(a) Reagents and/or equipment for detecting the methylation markers screened by the above screening method;

(b) reagents and/or equipment for detecting the above-mentioned bladder cancer methylation markers;

(c) the above-mentioned methylation feature recognition device;

(d) the above computer readable medium.

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

The method for screening methylation markers provided by the present invention performs two screenings of differentially methylated regions, the first screening based on conventional methods in the art, such as screening based on chip data; the second screening based on methylation Based on the deep reads of high-throughput targeted sequencing, and using the alpha density distribution model, different parameter models can be established for different differentially methylated regions, reducing the misjudgment rate and improving the accuracy of samples with specific biological states Check out.

The present invention utilizes the convenience of high-throughput sequencing to perform second-generation sequencing on candidate differentially methylated regions on the basis of determining differential sites, and calculates the alpha value of each read at the single-molecule level, that is, all The proportion of methylated C at CpG sites. The alpha value can more sensitively capture the tumor signal than the beta value, while eliminating noise interference. Through mathematical modeling of single-molecule methylation signals in cancer tissues and normal tissues, methylation markers can be identified with higher sensitivity and specificity. Compared with the traditional identification method of methylation marks, cancer prediction based on a single read has higher resolution and can detect smaller tumor signals in blood and urine.

Description of drawings

In order to more clearly illustrate the specific implementation of the present invention or the technical solutions in the prior art, the following will briefly introduce the accompanying drawings that need to be used in the specific implementation or description of the prior art. Obviously, the accompanying drawings in the following description The drawings show some implementations of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

Fig. 1 is the flowchart of the scheme of embodiment 1 of the present invention;

Figure 2 shows the methylation distribution of each reads in the OTX1_200_chr2:63282292-63283121 region of sample RD20230117-Twist15M-1-54525A1, the sequencing depth represented by A in Figure 2, the highest depth is 220×; the X axis of B in Figure 2 Represents the coordinates of chr2, the Y axis is the number of reads, and each read shows different colors at different states of the methylation site: yellow indicates no methylation, black indicates methylation, and the Y axis of C in Figure 2 and B is coincident, indicating a read, and the X axis indicates the alpha value of the read;

Figure 3 shows the alpha distribution of normal samples and cancer samples in different hypermethylated regions. The left side is normal samples, and the right side is cancer samples. The X-axis (horizontal axis) represents the number of reads, and the Y-axis (vertical axis) represents The alpha value of each read;

FIG. 4 is a training evaluation diagram of the convolutional network model established by the embodiment of the present invention. The horizontal axis represents Epoch (referring to the number of times the neural network traverses the entire training data set), and the vertical axis represents Accuracy. The scale of the horizontal axis is 0-38, indicating that the network has trained 38 Epochs; while the scale of the vertical axis is 0.65-1.0, indicating that Accuracy starts from 0.65 and gradually increases to 0.98.

Detailed ways

The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

It should be understood that when used in this specification and the appended claims, the terms "comprising" and "comprises" indicate the presence of described features, integers, steps, operations, elements and/or components, but do not exclude one or Presence or addition of multiple other features, integers, steps, operations, elements, components and/or collections thereof.

It should also be understood that the terminology used in the description of the present invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the present invention. As used in this specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural referents unless the context clearly dictates otherwise.

It should also be further understood that the term "and/or" used in the description of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations .

In this article, "reads" and "read" can be used interchangeably, and "reads" can represent multiple reads, one type of reads, or one read.

As used herein, the terms "cancer," "cancer," and tumor" are used interchangeably to refer to a disease, disorder, or condition in which cells exhibit or exhibit relatively abnormal, uncontrolled, and/or autonomous growth such that they Exhibit or exhibit an abnormally elevated rate of proliferation and/or an abnormal growth phenotype.

According to one aspect of the present invention, a screening method for methylation markers is provided. The methylation markers screened by the screening method provided in the present invention refer to the methylation level of the markers and the specific characteristics of the tested individual. Biological state (such as disease), the present invention is not limited to a specific specific biological state, in an optional embodiment, the specific biological state includes cancer.

The screening methods include:

(a) Obtain the sequencing reads of each candidate differentially methylated region of the characteristic sample and the control sample; calculate the alpha value of each read; the alpha value is the number of methylated cytosines in each read and all methylated positions The ratio of the number of points.

There is at least one known methylation site in each candidate differentially methylated region, and each candidate differentially methylated region can be obtained according to conventional methods in the art, for example, by referring to published textbooks, references, process manuals, The differentially methylated regions recorded in product descriptions, databases, and standard documents, or the differentially methylated regions recorded in the screening method for differentially methylated regions, such as the differentiated regions screened out based on beta values, are not limited by the present invention .

A characteristic sample refers to a sample with the above-mentioned specific biological state, and a control sample refers to a sample without the above-mentioned specific biological state, or a sample with the above-mentioned specific biological state below a threshold. Samples can be conventional biological samples in the field, including but not limited to tissues, cells, body fluids, and nucleotides.

(b) Obtain the alpha distribution model of the candidate differentially methylated region, if the alpha distribution model meets the following principles, the candidate differentially methylated region, and/or, at least one of the candidate differentially methylated regions Methylation sites as methylation markers:

(I) The alpha value of at least 80% of the reads in the control sample ≤ the hypomethylation threshold, such as but not limited to the alpha value of at least 80%, 82%, 85%, 87%, 90%, 92% or 95% of the reads ≤ low methylation threshold; the more the alpha value of the reads in the control sample is lower than the low methylation threshold, the higher the specificity of the corresponding differentially methylated region as a marker.

(II) The alpha value of at least 30% of the reads in the feature sample ≥ the high methylation threshold, for example, it can be but not limited to at least 30%, 35%, 40%, 45% or 50% of the alpha value ≥ the high methylation threshold, the more in the feature sample The alpha value of the reads is higher than the high methylation threshold, and the corresponding differentially methylated region is more sensitive as a marker.

(III) High methylation threshold-low methylation threshold ≥ N, N is at least 0.5, for example, but not limited to 0.5, 0.6, 0.7 or 0.8.

Moreover, the high methylation threshold and/or the low methylation threshold in the principle that the alpha distribution models of each candidate differentially methylated region conform to are the same or different. For example, the alpha value of 30% of the reads in a characteristic sample of a candidate differentially methylated region is greater than 0.9, then the high methylation threshold is 0.9, and at the same time, the alpha value of 80% of the reads in the control sample in this region is less than 0.3, then For this methylated region, its set of model parameters is 0.3/0.9. For another candidate differentially methylated region characteristic sample, the alpha value of 30% of the reads is greater than 0.85, and at the same time, the alpha value of 80% of the reads of the control sample in this region is less than 0.2, then a set of model parameters for this region It is 0.2/0.85; different candidate differentially methylated regions according to the methylation characteristics of their respective regions, the parameters of the alpha distribution model can be the same or different, or partly the same, for example, the parameters of at least two candidate differentially methylated regions The hypomethylation threshold is the same but the hypermethylation threshold is different. Different candidate differentially methylated regions can have a different set of alpha distribution models to evaluate whether the region can be used as a methylation marker. Setting parameters according to the methylation characteristics of respective regions can reduce the misjudgment rate and improve the detection of samples with specific biological states.

The low methylation threshold and the high methylation threshold can be divided according to the general rules in this field, and the value range of the low methylation threshold and the high methylation threshold is 0-1.

In an optional embodiment, in step (b), if the differentially methylated region exists in at least three pairs of characteristic samples and control samples, then the differentially methylated region and/or the methylation sites therein are determined Can be used as a methylation marker.

In an optional embodiment, the methylation markers include cancer methylation markers, the characteristic sample is cancer tissue, and the control sample is para-cancerous normal tissue. The cancer optionally includes bladder cancer.

In an optional embodiment, the candidate differentially methylated region is obtained in the following manner: obtain tumor sample data and normal sample data in the TCGA database, perform differential analysis according to the beta value to obtain the candidate differentially methylated region, and obtain the candidate differentially methylated region according to the differential methylation Probes are designed for regions to capture nucleic acid samples of cancer tissues and normal adjacent tissues, and sequence reads of candidate differentially methylated regions are obtained after sequencing.

According to one aspect of the present invention, a bladder cancer methylation marker is provided. The bladder cancer methylation marker provided in the present invention is obtained by screening the above methylation marker screening method, including the following table At least one of 1-13; wherein, the markers corresponding to each item in the following table include the corresponding methylation site and/or the region containing the corresponding methylation site:

marker serial number chromosome starting point end position 1 chr1 50886724 50886843 2 chr1 63785486 63785605 3 chr2 63282955 63283074 4 chr2 66667375 66667494 5 chr6 108485949 108486068 6 chr7 27205056 27205284 7 chr7 27205323 27205442 8 chr7 27205600 27205719 9 chr7 19157943 19158062 10 chr10 22518137 22518594 11 chr11 43602804 43602923 12 Chr17 17685349 17685468 13 chr18 22929349 22929468

Based on the hg19 reference genome sequence.

According to one aspect of the present invention, a method for establishing a model for identifying methylation features is also provided. The method is based on the candidate differentially methylated regions screened in step (b) of the above screening method, including differentially methylated Using regionalized objects as input data, a one-dimensional convolutional neural network is used to build a model, and a deep learning framework is used for model training and prediction.

In an optional embodiment, according to the differentially methylated region selected in step (b), the methylation pattern of the reads is encoded and input into a convolutional neural network to establish a model, the model is constructed based on the Python language, and the Tensorflow deep learning framework is used for Model training and prediction.

In an optional embodiment, the methylation pattern of reads is encoded by one-hot encoding, and then a sequential model (Sequential model) is created, using a Conv1D (one-dimensional convolutional neural network) layer To build a model, perform convolution processing on the input data, extract the feature information of the data, and then add other types of layers to increase the complexity and accuracy of the model, and finally perform the processing of the fully connected layer, and map the extracted feature information to Output the result, and compile the loss function and optimization algorithm for the established convolutional neural network model.

In an optional implementation, the parameters of the Conv1D layer are:

filters=32, kernel_size=3, strides=1, padding="same", activation="relu";

In an optional implementation manner, the other types of layers include a pooling layer and/or a Dropout layer;

In an optional embodiment, the parameters of the pooling layer are: pool_size=2, strides=2;

In an optional embodiment, the Dropout layer parameter is: 0.5.

In an optional implementation manner, the loss function is selected from a cross-entropy loss function.

In an optional embodiment, the optimization algorithm is selected from the stochastic gradient descent method; the parameters of the stochastic gradient descent method are preferably lr=0.01, momentum=0.9, and weight_decay=0.0001.

In an optional embodiment, at least 4 pairs of feature samples and control samples are then used to train the model.

In an alternative embodiment, the model is used to identify methylation signatures of cancerous tissues, where the cancer optionally includes bladder cancer.

According to one aspect of the present invention, a method for identifying methylation features is also provided, including using the model established by the above establishment method to identify the methylation features of the sequencing reads of the sample to be tested.

In an optional embodiment, the methylation pattern of the reads of the differentially methylated region of the sample to be tested is input into the model, and the differentially methylated region is the differentially methylated region used to establish the model, namely For the differentially methylated regions obtained in step (b) of the above screening method, the model outputs the methylation features of the reads of the sample to be tested.

According to one aspect of the present invention, a methylation feature recognition device is also provided, the device includes a prediction module, and the prediction module is used to input the obtained methylation pattern of the marker reads into the model established by the above model establishment method , to obtain the methylation feature of the sample to be tested.

The prediction module can be stored in the memory in the form of software or firmware (Firmware), or be fixed in the operating system (Operating System, OS) of the methylation feature recognition device. The above-mentioned methylation feature recognition device may also optionally be pre-installed with a storage module, in which data and program codes required for executing the above-mentioned modules are stored.

The methylation feature recognition device may include a memory, a processor, a bus and a communication interface, and the memory, the processor and the communication interface are electrically connected to each other directly or indirectly to realize data transmission or interaction. For example, these components may be electrically connected to each other through one or more buses or signal lines. A processor may process information and/or data related to object recognition to perform one or more functions described herein.

In practical applications, the methylation feature recognition device can be a server, cloud platform, mobile phone, tablet computer, notebook computer, ultra mobile personal computer (ultra mobile personal computer, UMPC), handheld computer, netbook, personal digital assistant (personal digital assistant) , PDA), wearable electronic equipment, virtual reality equipment and other equipment, so the embodiment of the present application does not limit the types of the above-mentioned devices.

According to one aspect of the present invention, there is also provided a computer-readable medium, which stores a computer program, and when the computer program is processed and executed, realizes the model established by the above model establishment method. Computer-readable media include, but are not limited to, various media that can store program codes, such as USB flash drives, removable hard drives, read-only memories, random access memories, magnetic disks, or optical disks.

According to one aspect of the present invention, a system for predicting, diagnosing or assisting in diagnosing cancer is also provided, including at least any one of the following:

(a) Reagents and/or equipment for detecting the methylation markers screened by the above screening method;

(b) reagents and/or equipment for detecting the above-mentioned bladder cancer methylation markers;

(c) the above-mentioned methylation feature recognition device;

(d) the above computer readable medium.

The present invention will be further described below through specific examples, however, it should be understood that these examples are only used for more detailed description, and should not be construed as limiting the present invention in any form.

Example 1

This embodiment provides a method for identifying the methylation characteristics of bladder cancer, the concept of which is shown in Figure 1, including the following steps:

1. Select differentially methylated sites according to the beta value and design probes:

There are 433 samples of bladder cancer (Bladder Urothelial Carcinoma, BLCA) in the Tumor Genome Atlas Database (TCGA), including 412 tumor samples (Primary Tumor) data, 21 normal samples (Solid Tissue Normal) data, and a total of 450k downloaded data. 485,577 cg (methylation) sites were differentially analyzed using the Champ package, logFC_t>0.4; P.Value_t<0.001, 217 differential cg sites were obtained, and probes for corresponding positions were designed.

2. Methylation library construction and probe capture:

Extract 1 pair of bladder cancer and adjacent normal tissue DNA, carry out enzyme digestion, end repair and dA tail addition; then use the methylation linker in the Aijitaikang double-strand library construction kit to link, and use the zymo kit after purification of magnetic beads Carry out A conversion, amplify after A conversion, and finally capture according to the probe designed in the first step.

3. Quality control and comparison:

The data after next-generation sequencing off-machine is first used for quality control with trim-glora, parameters: --illumina--paired--fastqc--length20-q20--clip_r110--clip_r210--three_prime_clip_r110--three_prime_clip_r210; using comparison software bismark, reference genome hg19, other parameters are default values, use bismark_methylation_extractor to extract methylation position and information after deduplicate_bismark deduplication, next step use script to merge OT/OB CpG information.

4. Identify the methylation pattern of reads:

Each captured region identifies the methylation pattern of CpG, 1 indicates methylation, 0 indicates unmethylation, and calculates the number of CpG and alpha value, as shown in Table 1, where the methylation site Recorded as "1" and unmethylated sites as "0". Taking sample RD20230117-Twist15M-1-54525A1 in the OTX1_200_chr2:63282292-63283121 region as an example, the methylation distribution of each reads is shown in Figure 2.

Table 1

5. Alpha distribution model of tumor and normal samples

With the increase of tumor cells, the reads with high aphla value will also increase. Extract the methylation position with significant difference in the distribution of alpha as the candidate region, and the distribution of alpha model in the differential methylation region Follow the principles below:

(1) To improve the specificity of the model, the alpha value of 80% reads of normal samples ≤ the hypomethylation threshold;

(2) To improve the sensitivity of the model, the alpha value of 30% reads of the tumor sample ≥ the high methylation threshold.

(3) In the candidate differential methylation region, the high methylation threshold - low methylation threshold ≥ 0.5, the greater the difference between the two, the stronger the model's robustness.

Based on the alpha distribution model above, the methylated regions were further screened, and the results are shown in Figure 3.

6. Specific differentially methylated regions:

Using 4 pairs of bladder cancer and paracancerous tissues, the corresponding differential regions were found using the method described above.

Normal (adjacent tissue):

50765B1_BLCA-N/58590C1_BLCA-N/54525D1_BLCA-N/56577C1_BLCA-N

Tumor (bladder cancer tissue):

50765C1_BLCA-T/58590A1_BLCA-T/54525A2_BLCA-T/56577A1_BLCA-T

Afterwards, the differentially methylated regions that exist in at least 3 pairs of samples are screened out as the differentially methylated regions for subsequent data training. The results are as follows:

Table 2

In the feature description column in the above table: A|B|C|D represent respectively, A: methylation site localization gene; B: length of methylation region; C: distance from transcription start site (Transcription Start Site, TSS) distance; D indicates the region in the gene where the methylated region is located. For example, "HOXA9|120|-451|Promoter" in the above table indicates the gene name HOXA9, the region length is 120, the distance from TSS is -451, and the promoter region of the promoter. For example, "EBLN1|458|-19225|Exon" indicates the gene name EBLN1, the region length is 458, the distance from TSS is -19225, and the exon region.

7. Deep learning model:

According to the selected alpha-based differential methylation region, the methylation pattern of the reads is encoded, ATGC is encoded as 0, 1, 2, and 3 methylated C is encoded as 4, and the encoded matrix is used as input, and then The convolutional neural network (Conv1D) in deep learning is used to build the model based on the Python language, and the Tensorflow deep learning framework is used for model training and prediction. Including the following steps:

(1) Extract the reads of the differentially methylated regions of each sample, and the corresponding A/T/unmethlC/G/methylC codes, so that subsequent models can understand and process the reads data obtained by next-generation sequencing.

(2) Using the one-hot encoding method, the reads correspond to 5 variables (A/T/unmethlC/G/methylC), and the one-hot encoding converts it into a 5-dimensional one-hot vector, of which only one element is 1, and the remaining elements are 0. Then, this vector is fed into Conv1D as part of the input tensor for training.

(3) Create a Sequential model, use the Conv1D layer to build the model, perform convolution processing on the above data, and extract the feature information of the data. Parameters: filters=32, kernel_size=3, strides=1, padding="same", activation="relu", add other types of layers (pooling layer/Dropout layer) to increase the complexity and accuracy of the model. Pooling layer parameters: pool_size=2, strides=2; Dropout layer parameters: 0.5.

Dense: activation='relu', kernel_regularizer=None, bias_regularizer=None

Finally, Dense (relu) is used to process the fully connected layer, and the extracted feature information is mapped to the output result.

(4) For the Conv1D model established above, select the loss function (Cross-entropy) and optimization algorithm (SGD) to compile, compare and select the parameters of the SGD optimization algorithm through multiple experiments: lr=0.01 , momentum=0.9, weight_decay=0.0001.

(5) Training model: use 4 pairs of bladder cancer and paracancerous tissues to train the model

Normal: 50765B1_BLCA-N/58590C1_BLCA-N/54525D1_BLCA-N/56577C1_BLCA-N;

Tumor: 50765C1_BLCA-T/58590A1_BLCA-T/54525A2_BLCA-T/56577A1_BLCA-T.

The training evaluation diagram of the convolutional network model is shown in Figure 4.

8. Model prediction and performance evaluation:

Use 15 bladder samples, including 10 bladder cancer samples and 5 adjacent normal samples, for model verification. After training and testing the data of the training set (60%) and the test set (20%), the model has obtained a comparative Good classification prediction results. After comparing the results of actual labels and predicted labels, it is found that the accuracy rate of this model on the test set is as high as 97.8%, which can better meet the needs of classification prediction.

The model established in this embodiment has better classification and prediction ability, and has certain learning and self-adjustment ability. Then predict the 16 blood samples, the prediction results are as follows, only one sample (result_RD20230421-CFD-nova-1598) has 7% reads containing Tumor components, the results are shown in the following table:

table 3

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: It is still possible to modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the various embodiments of the present invention. scope.

Claims (10)

1. A method for screening methylation markers, comprising: (a) Obtain the sequencing reads of each candidate differentially methylated region of the characteristic sample and the control sample; calculate the alpha value of each read; the alpha value is the number of methylated cytosines in each read and all methylated positions The ratio of the number of points; (b) Obtain the alpha distribution model of the candidate differentially methylated region, if the alpha distribution model meets the following principles, the candidate differentially methylated region, and/or, at least one of the candidate differentially methylated regions Methylation sites as methylation markers: The alpha value of at least 80% of the reads in the control sample ≤ the hypomethylation threshold; The alpha value of at least 30% of the reads of the feature sample ≥ the high methylation threshold; High methylation threshold - low methylation threshold ≥ N, N is at least 0.5; The high methylation thresholds and/or low methylation thresholds in the principles that the alpha distribution models of the candidate differentially methylated regions conform to are the same or different. 2. The screening method according to claim 1, wherein in step (b), the differentially methylated region exists in at least three pairs of characteristic samples and control samples; Preferably, the methylation markers include cancer methylation markers, the characteristic sample is cancer tissue, and the control sample is paracancerous normal tissue; Preferably, the cancer comprises bladder cancer; Preferably, tumor sample data and normal sample data are obtained in the TCGA database, differential analysis is performed according to beta values to obtain candidate differentially methylated regions, probes are designed according to differentially methylated regions, and nucleic acid samples of cancer tissues and adjacent normal tissues are captured After sequencing, the sequencing reads of the candidate differentially methylated regions are obtained. 3. Bladder cancer methylation marker, characterized in that it includes at least one of 1-13 in the following table; wherein, the marker corresponding to each item in the following table includes the corresponding methylation site and/or regions containing corresponding methylated sites: marker serial number chromosome starting point end position 1 chr1 50886724 50886843 2 chr1 63785486 63785605 3 chr2 63282955 63283074 4 chr2 66667375 66667494 5 chr6 108485949 108486068 6 chr7 27205056 27205284 7 chr7 27205323 27205442 8 chr7 27205600 27205719 9 chr7 19157943 19158062 10 chr10 22518137 22518594 11 chr11 43602804 43602923 12 Chr17 17685349 17685468 13 chr18 22929349 22929468 Based on the hg19 reference genome sequence. 4. A method for establishing a model for identifying methylation features, comprising: using the differentially methylated region obtained in the step (b) of the screening method according to claim 1 or 2 as input data, using a one-dimensional Convolutional neural networks build models, and use deep learning frameworks for model training and prediction. 5. The model building method according to claim 4, characterized in that, according to the differential methylation region selected in step (b), the methylation pattern of reads is encoded and input into a convolutional neural network to establish a model, the model is based on Python language construction, using Tensorflow deep learning framework for model training and prediction; Preferably, the methylation pattern of the reads is encoded by using a one-bit efficient encoding method, and then a sequential model is created, and a one-dimensional convolutional neural network layer is used to construct the model, and the input data is convolutionally processed to extract the data. Feature information, and then add other types of layers to increase the complexity and accuracy of the model, and finally process the fully connected layer, map the extracted feature information to the output result, and select the loss function for the established convolutional neural network model Compile with optimized algorithm; Preferably, at least 4 pairs of feature samples and control samples are used to train the model; Preferably, the other types of layers include pooling layers and/or Dropout layers; Preferably, the loss function is selected from a cross-entropy loss function; Preferably, the optimization algorithm is selected from the stochastic gradient descent method; the parameters of the stochastic gradient descent method are preferably lr=0.01, momentum=0.9, and weight_decay=0.0001. 6. The method for establishing a model according to claim 4 or 5, wherein the model is used to identify the methylation characteristics of cancer tissue; Preferably, the cancer comprises bladder cancer. 7. The method for identifying methylation features, characterized in that, using the model established by the establishment method according to any one of claims 4 to 6 to identify the methylation features of the sequencing reads of the sample to be tested; Preferably, the methylation pattern of the reads of the differentially methylated region of the sample to be tested is input into the model, the differentially methylated region is the differentially methylated region used to establish the model, and the model outputs the Measure the methylation characteristics of the reads of the sample. 8. The methylation feature recognition device, characterized in that it contains a prediction module, and the prediction module is used to input the methylation pattern of the reads in the differentially methylated region into the establishment method described in any one of claims 4 to 6 In the established model, the methylation characteristics of the sample to be tested are obtained. 9. A computer-readable medium, characterized in that a computer program is stored, and when the computer program is processed and executed, the model described in any one of claims 4-6 is realized. 10. A system for predicting, diagnosing or assisting in diagnosing cancer, characterized in that it includes at least any one of the following: (a) reagents and/or equipment for detecting the methylation markers screened by the screening method according to claim 1 or 2; (b) reagents and/or equipment for detecting the bladder cancer methylation marker according to claim 3; (c) the methylation feature recognition device according to claim 8; (d) The computer readable medium of claim 9.