Cancer methylation marker screening method and device based on cfDNA and electronic equipment

Abstract

The invention discloses a cfDNA-based cancer methylation marker screening method, which comprises the following steps: respectively extracting cfDNA samples of a first population and a second population and methylation values of all CpG sites of the cfDNA samples in a target interval so as to respectively obtain a first population sample set and a second population sample set; selecting training sample sets of the first crowd sample set and the second crowd sample set, combining the training sample sets, averagely dividing the training sample sets into 1 training sample subsets, selecting (l-1) training sample subsets in the training sample subsets for multiple times, and screening different CpG sites of the first crowd sample and the second crowd sample in the training sample subsets, a plurality of candidate methylation marker sets are obtained, a plurality of screening models are constructed by means of the candidate methylation marker sets, and methylation marker sets are extracted. According to the marker between high-heterogeneity cfDNA sequencing samples obtained through cross screening of the method, the accuracy of cancer subtype detection is improved.

Description

Technical Field

[0001] The present application belongs to the field of molecular biomedical technology, and specifically, relates to a method, device and electronic equipment for screening cancer methylation markers based on cfDNA. Background Art

[0002] Early identification of cancer subtypes of different cell origins, especially adenocarcinoma, has an increasingly important positive significance for cancer prevention and treatment. The release of the early Cancer Genome Atlas (TCGA) has promoted basic and clinical research on pan-cancer. Through the integration of multi-dimensional data analysis of 33 types of cancer, researchers found that a variety of molecular features including gene expression and methylation can characterize cancer types from different cell types, such as lung adenocarcinoma, breast cancer, gastric adenocarcinoma, colorectal cancer, esophageal adenocarcinoma, pancreatic cancer and other cancer samples from glandular epithelial cells, which show similar molecular characteristic patterns. However, esophageal squamous cell carcinoma, lung squamous cell carcinoma, cervical cancer, etc. show completely different molecular characteristics.

[0003] Of course, the above research results come from tissue samples of cancer patients and are not applicable to the early detection of cancer. Plasma free DNA (cfDNA) is DNA that is free outside the cells in the peripheral blood. It comes from the body's cells that secrete and release it into the blood through apoptosis and necrosis. Among them, most cfDNA comes from DNA fragments released by blood cells. For cancer patients, such as early-stage patients, there is a very small proportion of DNA fragments in cfDNA that come from cancer cells. Therefore, in recent years, a branch of cfDNA, circulating tumor DNA (ctDNA) detection technology, has become one of the main methods for early cancer screening.

[0004] At present, the technical difficulties of cancer detection based on ctDNA are mainly manifested in the following aspects: 1) Low content and large background noise make it difficult to detect, especially in the early stage of cancer detection. These problems are more significant, which can be overcome by increasing the sequencing depth. In order to balance the sequencing depth and cost, it is usually preferred to determine a group of high-quality target molecules and design probes before performing target molecule detection; 2) Cancer is highly heterogeneous, and the detection sensitivity of markers in cfDNA is usually low. The sensitivity can be improved by marker collection or cancer clustering; 3) Tissue tracing of ctDNA requires markers to be tissue-specific in order to locate the lesion. These problems limit the application of cfDNA markers in cancer subtype detection.

[0005] Taking the above problems into consideration, this application designs a specific marker screening method for tumor cell types of different origins based on the methylation targeted sequencing data of cfDNA, based on the outstanding potential of methylation information of cfDNA in plasma in in vitro detection products for early screening and early diagnosis of cancer subtypes, for early screening and early detection of multiple subtype-specific cancers. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the purpose of the present application is to provide a cancer methylation marker screening method based on cfDNA, which targets cell-source-specific cancer subtypes and uses CpG sites obtained by targeted sequencing of cfDNA extracted from sample plasma to screen highly specific methylation markers, and can provide corresponding methylation markers for early detection of various adenocarcinomas such as lung adenocarcinoma, colorectal adenocarcinoma, gastric adenocarcinoma, thyroid cancer, esophageal adenocarcinoma, and pancreatic cancer; and the present application also provides multiple adenocarcinoma markers based on the cancer methylation marker screening method of cfDNA, which can show excellent cancer sample and healthy sample classification performance in specific cancer subtypes.

[0007] Specifically, this application involves the following aspects:

[0008] 1. A cfDNA-based cancer methylation marker screening method, comprising: extracting multiple cfDNA samples from a first population and a second population, respectively, extracting all CpG sites in a target interval of each cfDNA sample and the methylation value of each CpG site to obtain a first population sample set and a second population sample set, respectively; selecting part or all of the samples from the first population sample set and the second population sample set to form a training sample set, merging the training sample sets from the first population sample set and the second population sample set and evenly dividing them into l training sample subsets, selecting l-1 training sample subsets from the l training sample subsets multiple times, screening the differential CpG sites between the first population and the second population in the l-1 training sample subsets, and obtaining multiple candidate methylation marker sets; constructing multiple screening models through the multiple candidate methylation marker sets, and extracting the methylation marker sets using the multiple screening models.

[0009] 2. The cfDNA-based cancer methylation marker screening method according to item 1, wherein screening the differential CpG sites between the first population and the second population in l-1 training sample subsets comprises: deleting CpG sites with missing values ​​in the l-1 training sample subsets; calculating a first variance for each CpG site belonging to the first population sample set and a second variance for each CpG site belonging to the second population sample set in the l-1 training sample subsets, and deleting the corresponding CpG site in response to the first variance and / or the second variance being greater than a first threshold.

[0010] 3. According to the cfDNA-based cancer methylation marker screening method of item 1, screening the differential CpG sites between the first population and the second population in l-1 training sample subsets includes: using LASSO regression to obtain the contribution value of each CpG site belonging to the first population sample set and / or the second population sample set in the l-1 training sample subsets, and deleting the corresponding CpG site in response to the contribution value being no greater than a second threshold.

[0011] 4. A cfDNA-based cancer methylation marker screening method according to any one of items 1-3, wherein multiple screening models are respectively constructed using multiple candidate methylation marker sets, and extracting methylation marker sets using multiple screening models includes: determining the candidate methylation markers used for each of the multiple screening models; counting the number of recurrences of all candidate methylation markers in the multiple screening models; deleting the corresponding candidate methylation markers in response to the number of recurrences in the multiple screening models being less than a third threshold, and retaining the corresponding candidate methylation markers in response to the number of recurrences in the multiple screening models being greater than the third threshold to form a methylation marker set.

[0012] 5. According to the cfDNA-based cancer methylation marker screening method of item 3, wherein the contribution value of each CpG site belonging to the first population sample set and / or the second population sample set in l-1 training sample subsets is obtained by using LASSO regression, including: setting the regularization parameter of LASSO regression to 1 times the standard deviation of the minimum error to calculate the contribution value of each CpG site belonging to the first population sample set and / or the second population sample set in l-1 training sample subsets.

[0013] 6. According to the cfDNA-based cancer methylation marker screening method of item 1, the first population is a cancer patient population and the second population is a healthy population.

[0014] 7. A cfDNA-based cancer methylation marker screening method according to item 1, wherein the multiple screening models include one or more generalized linear models.

[0015] 8. A cfDNA-based cancer methylation marker screening device, comprising: a sample extraction unit, extracting multiple cfDNA samples from a first population and a second population, respectively, extracting all CpG sites in the target interval of each cfDNA sample and the methylation value of each CpG site, so as to obtain a first population sample set and a second population sample set, respectively; a marker extraction unit, selecting part or all of the samples from the first population sample set and the second population sample set to form a training sample set, merging the training sample sets from the first population sample set and the second population sample set and evenly dividing them into l training sample subsets, selecting l-1 training sample subsets from the l training sample subsets multiple times, screening the differential CpG sites of the first population and the second population in the l-1 training sample subsets, and obtaining multiple candidate methylation marker sets; a marker determination unit, constructing multiple screening models through multiple candidate methylation marker sets, and extracting methylation marker sets using the multiple screening models.

[0016] 9. A cfDNA-based cancer methylation marker screening device according to item 8, which performs the cfDNA-based cancer methylation marker screening method as described in items 1-7.

[0017] 10. An electronic device comprising: a processor; and a memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes a cfDNA-based cancer methylation marker screening method as described in any one of items 1-7.

[0018] 11. A marker for detecting adenocarcinoma, wherein the marker comprises one or more of the following: a genomic region at positions 142619144 to 142619145 of chromosome 1, a genomic region at positions 149673092 to 149673093 of chromosome 1, a genomic region at positions 133015244 to 133015245 of chromosome 2, a genomic region at positions 234073016 to 234073017 of chromosome 2, a genomic region at positions 242952017 to 242952018 of chromosome 2, a genomic region at positions 75718548 to 75718549 of chromosome 3, a genomic region at positions 4866454 to 4866455 of chromosome 4 the genomic region at position 6455, the genomic region at positions 187880912 to 187880913 of chromosome 4, the genomic region at positions 79946769 to 79946770 of chromosome 5, the genomic region at positions 134259654 to 134259655 of chromosome 5, the genomic region at positions 177388774 to 177388775 of chromosome 5, the genomic region at positions 170530174 to 170530175 of chromosome 6, the genomic region at positions 155143018 to 155143019 of chromosome 7, the genomic region at positions 19319149 to 19319150 of chromosome 8, the genomic region at positions 13 the genomic region from 82966 to 1382967, the genomic region from 91169043 to 91169044 of chromosome 10, the genomic region from 94820635 to 94820636 of chromosome 10, the genomic region from 121578541 to 121578542 of chromosome 10, the genomic region from 93271132 to 93271133 of chromosome 11, the genomic region from 53497146 to 53497147 of chromosome 12, the genomic region from 633914 to 633915 of chromosome 16, the genomic region from 55686381 to 55686382 of chromosome 17 the genomic region from 7254735 to 7254736 of chromosome 17, the genomic region from 21194768 to 21194769 of chromosome 17, the genomic region from 21316068 to 21316069 of chromosome 17, the genomic region from 79945198 to 79945199 of chromosome 17, the genomic region from 57350308 to 57350309 of chromosome 19, the genomic region from 44099811 to 44099812 of chromosome 20, the genomic region from 62809918 to 62809919 of chromosome 20, the genomic region from 9825448 to 9825449 of chromosome 21,The genomic region from 30476506 to 30476507 of chromosome 22 and the genomic region from 13489072 to 13489073 of chromosome Y.

[0019] 12. Use of a marker in preparing a kit for detecting adenocarcinoma, wherein the marker comprises one or two or more of the following: a genomic region at positions 142619144 to 142619145 of chromosome 1, a genomic region at positions 149673092 to 149673093 of chromosome 1, a genomic region at positions 133015244 to 133015245 of chromosome 2, a genomic region at positions 234073016 to 234073017 of chromosome 2, a genomic region at positions 242952017 to 242952018 of chromosome 2, a genomic region at positions 75718548 to 75718549 of chromosome 3, a genomic region at positions 486 The genomic region from 6454 to 4866455, the genomic region from 187880912 to 187880913 of chromosome 4, the genomic region from 79946769 to 79946770 of chromosome 5, the genomic region from 134259654 to 134259655 of chromosome 5, the genomic region from 177388774 to 177388775 of chromosome 5, the genomic region from 170530174 to 170530175 of chromosome 6, the genomic region from 155143018 to 155143019 of chromosome 7, the genomic region from 19319149 to 19319150 of chromosome 10 The genomic region from 1382966 to 1382967 of chromosome 1, the genomic region from 91169043 to 91169044 of chromosome 10, the genomic region from 94820635 to 94820636 of chromosome 10, the genomic region from 121578541 to 121578542 of chromosome 10, the genomic region from 93271132 to 93271133 of chromosome 11, the genomic region from 53497146 to 53497147 of chromosome 12, the genomic region from 633914 to 633915 of chromosome 16, the genomic region from 55686381 to 55686382 of chromosome 1 the genomic region from 7254735 to 7254736 of chromosome 7, the genomic region from 21194768 to 21194769 of chromosome 17, the genomic region from 21316068 to 21316069 of chromosome 17, the genomic region from 79945198 to 79945199 of chromosome 17, the genomic region from 57350308 to 57350309 of chromosome 19, the genomic region from 44099811 to 44099812 of chromosome 20, the genomic region from 62809918 to 62809919 of chromosome 20, the genomic region from 9825448 to 9825449 of chromosome 21,The genomic region from 30476506 to 30476507 of chromosome 22 and the genomic region from 13489072 to 13489073 of chromosome Y.

[0020] 13. A composition for detecting adenocarcinoma, wherein the composition comprises: a nucleic acid for detecting the methylation state of the marker of item 11; wherein the methylation state of the marker is characterized by the methylation of the target sequence of the marker.

[0021] 14. An adenocarcinoma detection kit, comprising a reagent for detecting the marker of item 11.

[0022] Effects of the Invention

[0023] The method described in this application is based on the idea of ​​cross-screening, and utilizes indicators such as variance, contribution value, and number of recurrences to extract differential CpG sites in cfDNA high-depth sequencing samples. As a methylation marker for cancer, especially adenocarcinoma, it has the characteristics of high sensitivity and is a practical method that provides more accurate detection for adenocarcinoma. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The figure illustrates a flow chart of a method for screening cancer methylation markers based on cfDNA according to an embodiment of the present application.

[0025] Figure 2 The figure illustrates a flow chart of screening differential CpG sites according to the cfDNA-based cancer methylation marker screening method according to an embodiment of the present application.

[0026] Figure 3 The figure illustrates a flow chart for further screening a methylation marker set according to the cfDNA-based cancer methylation marker screening method according to an embodiment of the present application.

[0027] Figure 4 The figure shows the performance of the methylation markers of the cfDNA-based cancer methylation marker screening method according to the embodiment of the present application, the markers of the existing method, and the candidate methylation markers.

[0028] Figure 5 The flowchart of the cancer methylation marker screening device based on cfDNA according to an embodiment of the present application is shown.

[0029] Figure 6 A block diagram of an electronic device according to an embodiment of the present application is illustrated. DETAILED DESCRIPTION

[0030] Methylation level

[0031] The ratio of 5-methylcytosine at a specific CpG site to the total CpG sites in genomic DNA. This level is often used to assess the regulation of gene expression and function, because high DNA methylation usually leads to gene silencing, while low or no methylation is associated with gene activation. The methylation level of a specific position or region can usually be determined based on sequencing data by calculating the ratio of the number of methylated bases in the specific position or region to the total number of covered bases.

[0032] Methylation value

[0033] An indicator used to measure DNA methylation levels, commonly used in biostatistics and genomics research. The methylation value reflects the proportion of methylated cytosine at a specific CpG site, ranging from 0 to 1, where 0 indicates complete unmethylation and 1 indicates complete methylation. In practical applications, the methylation value can be used to compare methylation differences between different samples and different cell types, and is of great significance in disease diagnosis and gene expression regulation research.

[0034] The present application is further described below in conjunction with examples. It should be understood that the examples are only used to further describe and illustrate the present application and are not used to limit the present application.

[0035] Unless otherwise defined, the technical and scientific terms in this specification have the same meaning as those generally understood by those skilled in the art. Although similar or identical methods and materials as described herein may be used in experiments or practical applications, materials and methods are described herein below. In the event of a conflict, the present specification including the definitions therein shall prevail. In addition, materials, methods and examples are provided for illustration only and are not restrictive. The present application is further described below in conjunction with specific embodiments, but is not intended to limit the scope of the present application.

[0036] Application Overview

[0037] Adenocarcinoma, as a subtype of cancer, especially pancreatic cancer and gastric adenocarcinoma, generally shows high heterogeneity. There may be significant genetic and epigenetic differences between different patients or even different tumor areas of the same patient. This heterogeneity increases the difficulty of extracting markers from cfDNA. The scheme of detecting tumor development through tissue tracing requires markers with high tissue specificity. For ctDNA, which has a low content, the screening effect of multiple types of high-specific markers is poor and the quality of the obtained markers is low. For the direct extraction of methylation markers, it is necessary to identify the methylation pattern associated with the specific tumor type or tumor development stage; in addition, the content of ctDNA in cfDNA is low, and factors such as cfDNA degradation and background noise during marker extraction will affect marker extraction and even the accuracy of detection.

[0038] Nevertheless, multi-cancer markers screened using the methylation level variation characteristics of cfDNA still have unique advantages over tissue tracing markers. They do not need to detect cancer by distinguishing the tissue origin in ctDNA, but rather screen markers for one or more types of cancer phenotypes, which can more directly and effectively consider the heterogeneity between cancer cfDNA samples and between cancer and healthy cfDNA samples, thereby providing higher classification performance. Specifically, by performing high-depth targeted sequencing on cfDNA in plasma, the methylation intervals of multiple methylated CpG sites obtained by sequencing can be used through multiple validations, cross-validations, etc. to consider more methylation heterogeneity between different cfDNA samples and reduce the occurrence of false positive and false negative intervals, thereby ensuring that the selected intervals as methylation markers can be truly highly correlated with the selected cancer, so as to distinguish the changes in methylation levels of various adenocarcinomas and / or the same adenocarcinoma at different stages.

[0039] Based on the above ideas, the present application provides a cfDNA-based cancer methylation marker screening method, screening device and electronic device, which obtains the methylation spectrum of all CpG sites in the target interval of the cfDNA samples of the selected population, so as to obtain a highly specific methylation marker set from a large-scale methylation interval through multiple screening methods, and can further use the methylation marker set to build a high-accuracy adenocarcinoma classification model. Specifically, the cfDNA-based cancer methylation marker screening method, screening device and electronic device of the present application respectively extract cfDNA from the plasma samples of the first population and the second population, and use the bioinformatics analysis process to analyze and extract all CpG sites and the methylation value of each CpG site in the specified target interval of the cfDNA of each plasma sample, so as to obtain the first population sample set and the second population sample set containing the CpG site methylation spectrum, respectively, the methylation spectrum covers all possible specific marker sites, and the first population sample set and the second population sample set contain multiple methylated CpG sites and their intervals.

[0040] Furthermore, the training sample sets of the first population sample set and the second population sample set are divided respectively, the training sample sets are merged and evenly divided into l training sample subsets, l-1 training sample subsets are selected in turn from the l training sample subsets, and the differential CpG sites between the first population and the second population in the l-1 training sample subsets are screened to obtain multiple candidate methylation marker sets; multiple screening models are constructed respectively through the multiple candidate methylation marker sets, and the methylation marker sets are extracted using the multiple screening models. In this way, multiple effective screenings of cfDNA methylation markers are achieved, and the methylation marker sets can be used to train multi-cancer classification models, such as multiple adenocarcinoma classification models.

[0041] Through the cfDNA-based cancer methylation marker screening method, screening device and electronic device of this application, multiple cross- and repeated screening of potential methylation sites in the target interval are used to discover and pay more attention to marker intervals with higher cfDNA sample sensitivity and specificity, effectively overcoming the problem that a variety of adenocarcinoma methylation markers are low in content and difficult to extract in cfDNA, while also shielding noise interference. The methylation markers screened have higher detection sensitivity for the source population of cfDNA. The screening method provided in this application also obtains effective markers for single adenocarcinoma detection / adenocarcinoma pan-cancer detection, as shown below.

[0042] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0043] Exemplary Methods

[0044] Figure 1 The figure illustrates a flow chart of a method for screening cancer methylation markers based on cfDNA according to an embodiment of the present application.

[0045] like Figure 1 As shown, the cfDNA-based cancer methylation marker screening method according to an embodiment of the present application includes the following steps.

[0046] Step S110, extract multiple cfDNA samples of the first population and the second population, extract all CpG sites and methylation values ​​of each CpG site in the target interval of each cfDNA sample, so as to obtain the first population sample set and the second population sample set respectively. It can be understood that the first population and the second population differ in cancer conditions. The first population can be cancer patients with glandular epithelial carcinoma, such as lung adenocarcinoma, colorectal adenocarcinoma, gastric adenocarcinoma, thyroid cancer, esophageal adenocarcinoma or pancreatic cancer, especially adenocarcinoma patients, while the second population is a healthy population or a population that does not suffer from any of the above adenocarcinomas. In addition to the difference in cancer, the first population and the second population can also have clinical baseline differences, such as gender, age, height, weight and other medical history. The cfDNA sample can be extracted from the collected plasma of the first population and the second population. The extraction method can include centrifugation, kit extraction, magnetic bead method or extraction using organic solvents in combination with AND precipitants, and then the CpG site methylation spectrum of each cfDNA sample in the target interval is obtained using a bioinformatics analysis process.

[0047] Specifically, the cfDNA sample can be obtained by the above-mentioned cfDNA extraction method and library construction and targeted sequencing can be performed. The base quality of the original sequencing data can be checked by quality control software, such as fastp, and low-quality bases and reads can be filtered. The filtered clean reads can be aligned to the reference genome using Bismark bowtie2 software; deduplication_bismark can be used to deduplicate the reads on the paired genome; and then the methylation value of the CpG site in the target interval can be extracted using bismark_methylation_extractor software. The reference genome can be selected from the human genome hg19 (GRCh37), and the target interval can be determined based on existing data such as public research / reports / products. For example, for the possible target interval of lung adenocarcinoma marker extraction, relevant published papers on lung adenocarcinoma and methylation marker kits listed by the company can be referred to and determined; in some possible embodiments, the current interval can also be set to the whole genome, which requires more CpG site extraction time and screening time, but can extract high-sensitivity markers to the maximum extent through more times of put-back screening.

[0048] Step S120, select part or all of the samples from the first population sample set and the second population sample set to form a training sample set, merge the training sample sets from the first population sample set and the second population sample set and evenly divide them into l training sample subsets, select l-1 training sample subsets from the l training sample subsets in turn, screen the differential CpG sites between the first population and the second population in the l-1 training sample subsets, and obtain multiple candidate methylation marker sets. It can be understood that the above-mentioned part or all of the samples refer to part or all of the cfDNA samples in the first population sample set and the second population sample set, and the selected samples all contain all of their CpG sites and methylation values ​​of the sites.

[0049] The purpose of selecting some samples from the first population sample set and the second population sample set to form a training sample subset is that the screening method according to the embodiment of the present application not only finally obtains methylation markers of cancer, such as adenocarcinoma, but also can further construct a classification model for the first population and the second population based on these markers; therefore, the unselected part of the sites can be used to form a test sample set to evaluate the performance of the classification model or help the classification model to perform incremental learning, etc. For example, the first population sample set and the second population sample set are divided into a training sample set and a test sample set respectively in a ratio of 2:1, and the division process takes into account differences in gender, age or other personal factors to reflect the robustness of the methylation markers when evaluating the classification model.

[0050] When the value of l is too small, the sample size of CpG sites used to screen the candidate methylation marker set is small, and the performance of multiple screening models constructed with multiple candidate methylation marker sets may fluctuate greatly due to the heterogeneity of sample division, causing some candidate methylation marker sets to be overfitted and the classification stability between multiple screening models to be low. When the value of l is too large, the sample size of CpG sites used to screen the candidate methylation marker set is large, and the classification performance evaluation results of the candidate methylation marker set are more stable, improving the sample sensitivity of the candidate methylation markers retained after each screening model training, and making the candidate methylation markers that are reproduced multiple times have a higher classification credibility; however, as the value of l increases, the computational complexity of screening the candidate methylation marker set and updating the screening model increases, and when the amount of sequencing sample data is large, the marker screening cost increases significantly.

[0051] Preferably, according to the cfDNA-based cancer methylation marker screening method of the embodiment of the present application, the training sample set is evenly divided into 5 training sample subsets, and 4 training sample subsets of the 5 training sample subsets are selected in turn to screen the differential CpG sites between the first population and the second population. This balances the screening accuracy of the candidate methylation marker set and the computational complexity of the screening model construction, making the described method more practical.

[0052] Taking into account the population heterogeneity of adenocarcinoma methylation markers, the screening method according to the embodiment of the present application performs multiple group comparisons on the currently obtained training sample subsets containing all valid CpG sites, each time randomly extracting l-1 training sample subsets from l training sample subsets, and repeating the operation of extracting l-1 training sample subsets multiple times, and screening out multiple differential CpG sites based on the CpG sites in the l-1 training sample subsets extracted each time and using the intervals where they are located as candidate methylation markers that may recur.

[0053] In particular, a preferred embodiment of this method is to extract l-1 training sample subsets from l training sample subsets in sequence, that is, to extract all possible combinations of any l-1 training sample subsets. For example, 4 training sample subsets are extracted in sequence from 5 training sample subsets, namely, the first subset, the second subset, the third subset, the fourth subset and the fifth subset. The first time, the first subset, the second subset, the third subset and the fourth subset are extracted and the differential CpG sites therein are screened; the second time, the first subset, the second subset, the third subset and the fifth subset are extracted and the differential CpG sites therein are screened; the third time, the first subset, the second subset, the fourth subset and the fifth subset are extracted and the differential CpG sites therein are screened; the fourth time, the first subset, the third subset, the fourth subset and the fifth subset are extracted and the differential CpG sites therein are screened; the fifth time, the second subset, the third subset, the fourth subset and the fifth subset are extracted and the differential CpG sites therein are screened. In this way, the intervals of differential CpG sites obtained by each screening of differential CpG sites are used as candidate methylation markers, which are obtained through multiple cross-screening in the training sample subset; after invalid markers and obviously inefficient markers are screened out by other feasible methods, the methylation markers with high sensitivity for cfDNA samples are more likely to reappear after multiple screenings.

[0054] That is, the candidate methylation marker set formed by the above-mentioned preferred multiple replacement screening fully includes all the effective differential CpG sites in the training sample set. In addition, it is also possible to choose to randomly extract a subset of training samples, which has the risk of missing candidate markers, but ensures the randomness of cross-extraction and can be better used for the reproduction of high-sensitivity markers. Then, based on the number of reproductions of all candidate methylation markers in the candidate methylation marker set, the methylation marker set with the best ability to distinguish between the cfDNA samples of the first population and the second population is determined, thereby achieving the purpose of extracting targeted markers based on heterogeneous target disease groups.

[0055] In addition, different screening methods can have a significant impact on the construction of methylation marker sets, so, for example, Figure 2 As shown, screening the differential CpG sites between the first population and the second population in l-1 training sample subsets may further include the following screening steps.

[0056] Step S1201, delete CpG sites with missing values ​​in l-1 training sample subsets, because methylation modification usually occurs at these sites, if the CpG sites have missing values, their methylation levels may not be accurately measured. In particular, it is necessary to delete CpG sites with missing values ​​in promoter regions and genomic functional elements, which are enriched with a large number of CpG sites and may affect gene expression and become model noise.

[0057] Step S1202, calculate the first variance of each CpG site belonging to the first population and the second variance of each CpG site belonging to the second population in the l-1 training sample subsets, and delete the corresponding CpG site in response to the first variance and / or the second variance being greater than the first threshold. The degree of dispersion of each site can be determined according to the distribution of the methylation values ​​of the CpG sites of the first population sample and the second population sample. In order to prevent the candidate methylation markers obtained by screening from being extreme values ​​in the target interval, a first threshold can be set so that the CpG sites with a methylation value variance greater than the first threshold are excluded; it can be understood that the same first threshold is preferably used to exclude extreme values ​​in the two populations in order to maintain the consistency of sample distribution, but different variance screening thresholds can also be set for the Cp sites of the first population and the second population. For example, in the case where there are obvious population differences in part of the target interval, CpG sites with heterogeneous high variance values ​​in methylation values ​​can be key sites. In this case, these sites can be retained, and the present application preferably provides a more general solution that can ensure the universal adenocarcinoma classification quality of methylation markers.

[0058] Step S1203, using LASSO regression to calculate the contribution value of each CpG site belonging to the first population and / or the second population in the l-1 training sample subsets, and deleting the corresponding CpG site in response to the contribution value being no greater than the second threshold. Specifically, LASSO regression is based on ordinary least squares method, and adds L1 regularization term to perform marker selection and simplify the classification model. The L1 regularization term constrains the coefficients of the classification model, and can screen out markers with non-zero coefficients. These markers can be considered as markers that contribute more to the classification model and have higher sensitivity. The process of determining the regularization parameter of the marker coefficient by LASSO regression is shown in the following formula:

[0059]

[0060]

[0061] Among them, y i is the observed value of the sample classification result, x iis the methylation value of the CpG site, β is the coefficient vector of the CpG site, and λ is the regularization parameter. The minimum value of LASSO is calculated by setting the regularization parameter to penalize the L1 norm of the coefficient of each CpG site so that the coefficient value of the unimportant CpG site is zeroed. The calculation method can be a gradient descent method, a coordinate descent method, etc. Therefore, in some possible implementation schemes, the second threshold can be set as the minimum coefficient value of LASSO regression, so that the CpG site with a coefficient value of 0 is deleted to effectively extract markers and control the number of markers, reduce the construction complexity of the classification model and the calculation and time cost of adenocarcinoma detection; the regularization parameter can be preferably set to 1 times the standard deviation of the minimum error to simplify the LASSO regression model.

[0062] Step S130, construct multiple screening models respectively through multiple candidate methylation marker sets, and extract methylation marker sets using the multiple screening models. It can be understood that this step is a process of further screening methylation markers with high sensitivity using the number of repetitions of each candidate methylation marker in the multiple candidate methylation marker sets. Figure 3 As shown, the following specific steps may be included.

[0063] Step S1301, determine the candidate methylation markers used for each screening model in multiple screening models. It can be seen from the above extraction of l-1 training sample subsets from l training sample subsets that l candidate methylation marker sets can be obtained by replacing l training sample subsets, and l screening models are constructed using these candidate methylation marker sets; in another possible embodiment, more than or less than l candidate methylation marker sets can be obtained by randomly replacing l training sample subsets, and corresponding number of screening models are constructed using these candidate methylation marker sets. The screening model can be a generalized linear model (GLM) or any available classification model, such as a decision tree, random forest, support vector machine, logistic regression, or other statistical models, etc. Those skilled in the art can determine the required screening model based on the overall performance of the candidate methylation marker set under different models or partial performance such as specificity and sensitivity.

[0064] Preferably, the screening model is set to GLM, because the methylation sequencing data of cfDNA samples is huge, and GLM has high calculation and fitting efficiency for this type of continuous data. In addition, GLM usually includes clear mathematical expressions and statistical significance, which makes it very interpretable. Differential CpG sites that have a significant impact on the classification results can be directly obtained through GLM as candidate methylation markers. Those skilled in the art can intuitively understand which candidate methylation markers and which types of cancer, especially which types of adenocarcinomas, have high correlation or discrimination; while other classifiers either have difficulty in interpreting markers, or need to obtain the candidate methylation marker set again through inverse transformation or reconstruction to determine the reproducible candidate methylation markers, and information will be lost and data quality will be reduced in the process.

[0065] Step S1302, counting the number of recurrences of all candidate methylation markers in multiple screening models. For example, in 5 screening models constructed from 5 candidate methylation marker sets, counting the candidate methylation markers that appear 5 times, 4 times, 3 times, and 2 times respectively. It can be understood that this step can be implemented by any statistical algorithm applied to multiple trained screening models.

[0066] Step S1303, in response to the number of recurrences in multiple screening models being less than the third threshold, the corresponding candidate methylation marker is deleted, and in response to the number of recurrences in multiple screening models being greater than the third threshold, the corresponding candidate methylation marker is retained and a methylation marker set is formed. For example, the third threshold can be set to 2, 3, 4, 5 or more. In this way, a set of multiple candidate methylation marker sets is obtained, and the sensitivity of all candidate methylation markers is determined by retrieving each candidate methylation marker in the set, that is, the number of repetitions of each screened CpG site in multiple screening models. The more recurrences a CpG site has, the more likely it is to be a more important methylation marker.

[0067] Therefore, according to the cfDNA-based cancer methylation marker screening method of the embodiment of the present application, a third threshold for judging the sensitivity of the candidate methylation marker can be determined based on the number of candidate methylation marker sets or the number of screening models. In one example, 4 of the 5 training sample subsets are repeatedly extracted in sequence, and the differential CpG sites between the first population and the second population in the 4 training sample subsets are screened and 5 screening models are finally obtained. The third threshold is set to 5, and the intervals where multiple CpG sites appear in all screening models are obtained, which show high classification performance in all random division situations, and therefore have general sensitivity to sample data, avoiding the randomness of data representation, which makes the classification model constructed using the intervals where these CpG sites are located as methylation markers, such as the GLM classification model, more stable.

[0068] In some preferred embodiments, the method further comprises weighting the methylation markers in the finally obtained methylation marker set to obtain a weighted marker. Specifically, for any obtained methylation marker j, its weight w can be calculated using the following formula:

[0069]

[0070]

[0071] Among them, i is the recurring methylation marker j, that is, the type of differential CpG site, that is, its recurrence number, which can be based on the value of l, including {2, 3, 4, 5, ..., l}, and β is the methylation level of the recurring differential CpG site, that is, the methylation value. The weighted methylation marker is obtained by integrating the weights w of some or all recurring differential CpG sites, which can further optimize the classification performance of the marker set according to the importance of the marker, and can also significantly reduce the data dimension of the marker set and further reduce the complexity of the population classification model based on marker training.

[0072] Finally, the cancer methylation marker screening method based on cfDNA according to the embodiment of the present application includes step S140: constructing a cancer classification model through a methylation marker set, and using the cancer classification model to predict the source population of the cfDNA sample. Constructing a cancer classification model in the methylation marker set, such as a GLM cancer classification model, can be used to determine the discrimination threshold between cancer populations, especially adenocarcinoma populations and healthy populations. Finally, the performance of these methylation markers and the constructed cancer classification model can also be evaluated in the test sample set.

[0073] In particular, in order to verify the effectiveness of the cfDNA-based cancer methylation marker screening method according to the embodiment of the present application, steps S120-S140 can be repeated multiple times, for example 100 times, using the above steps S1201-S1203 to screen methylation markers, steps S1301-S1303 to further screen methylation markers and cancer classification models, calculate the classification sensitivity, specificity and AUC of the test sample set, and calculate the average value. In some possible embodiments, the methylation marker set obtained by the screening method of the embodiment of the present application and the markers obtained by only extracting a training sample subset once for screening, and the classification performance of the markers screened in the conventional single-use training set for the first population and the second population can be compared to evaluate the effectiveness of the methylation markers provided by the present application. The specific process and results can be seen in the embodiments.

[0074] In some preferred embodiments, the cfDNA-based cancer methylation marker screening method according to the embodiment of the present application obtains an adenocarcinoma methylation marker set consisting of 32 adenocarcinoma methylation markers that are repeated twice, which can be used for multiple cancer detections such as lung adenocarcinoma, colorectal cancer, gastric cancer, thyroid cancer and esophageal adenocarcinoma. These adenocarcinoma methylation markers include the following intervals where the CpG sites are repeated twice:

[0075] The genomic region from 142619144 to 142619145 of chromosome 1, the genomic region from 149673092 to 149673093 of chromosome 1, the genomic region from 133015244 to 133015245 of chromosome 2, the genomic region from 234073016 to 234073017 of chromosome 2, the genomic region from 242952017 to 242952018 of chromosome 2, the genomic region from 75718548 to 75718549 of chromosome 3, the genomic region from 4866454 to 4866455 of chromosome 4, and the genomic region from 187880912 to 187880 913 genomic region, 79946769 to 79946770 genomic region of chromosome 5, 134259654 to 134259655 genomic region of chromosome 5, 177388774 to 177388775 genomic region of chromosome 6, 170530174 to 170530175 genomic region of chromosome 7, 155143018 to 155143019 genomic region of chromosome 7, 19319149 to 19319150 genomic region of chromosome 8, 1382966 to 1382967 genomic region of chromosome 10, 91169 to 1382968 genomic region of chromosome 11 043 to 91169044 genomic region, 94820635 to 94820636 genomic region of chromosome 10, 121578541 to 121578542 genomic region of chromosome 10, 93271132 to 93271133 genomic region of chromosome 11, 53497146 to 53497147 genomic region of chromosome 12, 633914 to 633915 genomic region of chromosome 16, 55686381 to 55686382 genomic region of chromosome 17, 7254735 to 7254736 genomic region of chromosome 17, 21194768 to 21194769 genomic region, 21316068 to 21316069 genomic region of chromosome 17, 79945198 to 79945199 genomic region of chromosome 17, 57350308 to 57350309 genomic region of chromosome 19, 44099811 to 44099812 genomic region of chromosome 20, 62809918 to 62809919 genomic region of chromosome 20, 9825448 to 9825449 genomic region of chromosome 21, 30476506 to 30476507 genomic region of chromosome 22,The genomic region from position 13489072 to position 13489073 of chromosome Y. As described above, the weight values ​​w of these differential CpG sites can be further calculated and weighted methylation markers can be calculated. For example, it can be understood that multiple differential CpG sites with similar weight values ​​w are weighted and merged to significantly reduce the data dimension of the methylation marker set and improve the practicality of the cancer classification model.

[0076] In this way, the use of the above-mentioned markers in preparing a kit for adenocarcinoma detection can be obtained. It can be understood that the marker is also one or two or more of the above-mentioned adenocarcinoma methylation markers, namely:

[0077] The genomic region from 142619144 to 142619145 of chromosome 1, the genomic region from 149673092 to 149673093 of chromosome 1, the genomic region from 133015244 to 133015245 of chromosome 2, the genomic region from 234073016 to 234073017 of chromosome 2, the genomic region from 242952017 to 242952018 of chromosome 2, the genomic region from 75718548 to 75718549 of chromosome 3, the genomic region from 4866454 to 4866455 of chromosome 4, and the genomic region from 187880912 to 187880 913 genomic region, 79946769 to 79946770 genomic region of chromosome 5, 134259654 to 134259655 genomic region of chromosome 5, 177388774 to 177388775 genomic region of chromosome 6, 170530174 to 170530175 genomic region of chromosome 7, 155143018 to 155143019 genomic region of chromosome 7, 19319149 to 19319150 genomic region of chromosome 8, 1382966 to 1382967 genomic region of chromosome 10, 91169 to 1382968 genomic region of chromosome 11 043 to 91169044 genomic region, 94820635 to 94820636 genomic region of chromosome 10, 121578541 to 121578542 genomic region of chromosome 10, 93271132 to 93271133 genomic region of chromosome 11, 53497146 to 53497147 genomic region of chromosome 12, 633914 to 633915 genomic region of chromosome 16, 55686381 to 55686382 genomic region of chromosome 17, 7254735 to 7254736 genomic region of chromosome 17, 21194768 to 21194769 genomic region, 21316068 to 21316069 genomic region of chromosome 17, 79945198 to 79945199 genomic region of chromosome 17, 57350308 to 57350309 genomic region of chromosome 19, 44099811 to 44099812 genomic region of chromosome 20, 62809918 to 62809919 genomic region of chromosome 20, 9825448 to 9825449 genomic region of chromosome 21, 30476506 to 30476507 genomic region of chromosome 22,The genomic region from position 13489072 to 13489073 of the Y chromosome.

[0078] Furthermore, a composition for detecting adenocarcinoma can be obtained, wherein the composition comprises: a nucleic acid for detecting the methylation status of the above-mentioned adenocarcinoma marker; wherein the methylation status of the above-mentioned adenocarcinoma marker is characterized by the methylation of its target sequence.

[0079] Furthermore, an adenocarcinoma detection kit can be obtained, which contains reagents for detecting the above-mentioned adenocarcinoma markers.

[0080] The methylation marker set obtained by the cfDNA-based cancer methylation marker screening method provided in the present application has high sensitivity and specificity in detecting a variety of cancers, especially adenocarcinomas, such as lung adenocarcinoma, colorectal adenocarcinoma, gastric adenocarcinoma, thyroid cancer, esophageal adenocarcinoma or pancreatic cancer, and can be used in medical scenarios such as early screening and detection of adenocarcinoma and prognosis review of cancer patients. It also provides a reference for an improved solution for the cancer marker extraction method based on cfDNA methylation.

[0081] Exemplary Devices

[0082] Figure 5 The figure illustrates a block diagram of a cfDNA-based cancer methylation marker screening device according to an embodiment of the present application.

[0083] like Figure 5 As shown, according to the embodiment of the present application, the cfDNA-based cancer methylation marker screening device 200 includes: a sample extraction unit 210, respectively extracting multiple cfDNA samples of a first population and a second population, extracting all CpG sites of each cfDNA sample in the target interval and the methylation value of each CpG site to obtain a first population sample set and a second population sample set respectively; a marker extraction unit 220, selecting part or all of the samples from the first population sample set and the second population sample set to form a training sample set, merging the training sample sets from the first population sample set and the second population sample set and evenly dividing them into l training sample subsets, selecting l-1 training sample subsets from the l training sample subsets multiple times, screening the different CpG sites of the first population and the second population in the l-1 training sample subsets, and obtaining multiple candidate methylation marker sets; a marker determination unit 230, respectively constructing multiple screening models through multiple candidate methylation marker sets, and extracting methylation marker sets using the multiple screening models.

[0084] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the above-mentioned cfDNA-based cancer methylation marker screening device 200 have been referred to above. Figures 1 to 3The description of the cfDNA-based cancer methylation marker screening method has been described in detail, and therefore, its repeated description will be omitted.

[0085] As described above, the cfDNA-based cancer methylation marker screening device 200 according to the embodiment of the present application can be implemented in various terminal devices, such as a server for training screening models and cancer classification models. In one example, the cfDNA-based cancer methylation marker screening device 200 according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the cfDNA-based cancer methylation marker screening device 200 can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the cfDNA-based cancer methylation marker screening device 200 can also be one of the many hardware modules of the terminal device.

[0086] Alternatively, in another example, the cfDNA-based cancer methylation marker screening device 200 and the terminal device may also be separate devices, and the cfDNA-based cancer methylation marker screening device 200 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0087] Exemplary Electronic Devices

[0088] Below, reference Figure 6 To describe an electronic device according to an embodiment of the present application.

[0089] Figure 6 A block diagram of an electronic device according to an embodiment of the present application is illustrated.

[0090] like Figure 6 As shown, the electronic device 10 includes one or more processors 11 and a memory 12 .

[0091] The processor 13 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0092] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on a computer-readable storage medium, and the processor 11 may run the program instructions to implement the cfDNA-based cancer methylation marker screening method and / or other desired functions of the various embodiments of the present application described above. Various contents such as training sample subsets, candidate methylation marker sets, etc. may also be stored in the computer-readable storage medium.

[0093] In one example, the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0094] The input device 13 may include, for example, a keyboard, a mouse, etc.

[0095] The output device 14 can output various information to the outside, including the trained GLM model, etc. The output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0096] Of course, to simplify, Figure 6 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 10 may also include any other appropriate components.

[0097] Example

[0098] This application provides a general and / or specific description of the materials and test methods used in the experiment. The raw materials or instruments used without indicating the manufacturer are all conventional raw materials or instruments that can be obtained through commercial purchase.

[0099] Example 1 Methylation marker screening

[0100] Data collection: The methylation detection kit of Boercheng (Beijing) Technology Co., Ltd. was used to collect methylation targeted sequencing data of 129 cfDNA samples of adenocarcinoma patients and 93 cfDNA samples of healthy controls. The adenocarcinoma population included 35 patients with lung adenocarcinoma, 36 patients with colorectal cancer, 33 patients with gastric cancer, 22 patients with thyroid cancer, and 3 patients with esophageal adenocarcinoma. First, the base quality of the original targeted sequencing data was checked using quality control software such as fastp, and low-quality bases and sequencing reads were filtered. The filtered clean reads were aligned to the reference genome hg19 (GRCh37) using Bismark bowtie2 software; deduplication_bismark was used to deduplicate the reads aligned to the reference genome; then bismark_methylation_extractor software was used to extract CpG sites and their methylation values ​​in the target interval to obtain the sample set of adenocarcinoma patients and the sample set of control patients.

[0101] Sample grouping: The two sample sets are divided into training sample set S(train) and test sample set S(test) in a ratio of 2:1, and gender and age differences are considered in the grouping process. The training sample set contains 148 target interval samples, and the test sample set contains 74 target interval samples.

[0102] Feature screening: The methylation markers were screened using the 5-group comparison method, that is, the samples belonging to the adenocarcinoma population and the samples belonging to the control population in the training sample set S(train) were equally divided into 5 subsets, and any 4 subsets were extracted and merged to obtain 4 training sample subsets S(sub), which contained 118 target interval samples: 69 samples belonging to the adenocarcinoma population and 49 samples belonging to the control population. For each training sample subset S(sub), the differential CpG sites were screened by the following steps:

[0103] 1) Delete CpG sites with missing values ​​in the training sample subset S(sub);

[0104] 2) Calculate the variance S of each CpG site in all samples of the training sample subset S(sub) to obtain the intra-group variance S(T) of the adenocarcinoma population and the intra-group variance S(C) of the control population. Eliminate CpG sites with S(T) or S(C)>S;

[0105] 3) The LASSO regression method was used to screen CpG sites with large contribution values, where the parameter lambda was selected as 1 times the standard deviation of the minimum error;

[0106] 4) Using the intervals of the differential CpG sites obtained by screening to build a GLM screening model in the 4 training sample subsets S(sub) obtained each time, and using the remaining 1 training sample subset S(sub) to evaluate the performance of the screening model, 5 candidate methylation marker sets were obtained;

[0107] 5) Extract candidate methylation markers that appear repeatedly in the five GLM screening models as methylation markers. The results of repeated marker extraction are: 32 sites that appear at least twice, 16 sites that appear at least three times, 8 sites that appear at least four times, and 4 sites that appear at least five times.

[0108] Model building: GLM cancer classification models were constructed for methylation markers with different recurrence times in the training sample set, and the thresholds for five adenocarcinoma populations and healthy control populations were determined. In the test sample set, the classification performance of the methylation markers obtained through the above steps and the GLM cancer classification model constructed by one word was evaluated. The results are shown in the following table:

[0109] Table 1 Comparison of sample classification performance of models constructed using different methylation markers

[0110]

[0111] Among them, the locations of the methylation markers that reappeared twice on the chromosome are as follows:

[0112] chromosome1:142619144-142619145

[0113] chromosome1:149673092-149673093

[0114] chromosome2:133015244-133015245

[0115] chromosome2:234073016-234073017

[0116] chromosome2:242952017-242952018

[0117] Chromosome3:75718548-75718549

[0118] Chromosome4:4866454-4866455

[0119] chromosome4:187880912-187880913

[0120] chromosome5:79946769-79946770

[0121] chromosome5:134259654-134259655

[0122] chromosome5:177388774-177388775

[0123] chromosome6:170530174-170530175

[0124] chromosome7:155143018-155143019

[0125] chromosome8:19319149-19319150

[0126] chromosome10:1382966-1382967

[0127] chromosome10:91169043-91169044

[0128] chromosome10:94820635-94820636

[0129] chromosome10:121578541-121578542

[0130] chromosome11:93271132-93271133

[0131] chromosome12:53497146-53497147

[0132] chromosome16:633914-633915

[0133] chromosome16:55686381-55686382

[0134] chromosome17:7254735-7254736

[0135] chromosome17:21194768-21194769

[0136] chromosome17:21316068-21316069

[0137] chromosome17:79945198-79945199

[0138] chromosome19:57350308-57350309

[0139] chromosome20:44099811-44099812

[0140] chromosome20:62809918-62809919

[0141] chromosome21:9825448-9825449

[0142] chromosome22:30476506-30476507

[0143] chromosomeY:13489072-13489073

[0144] Method stability assessment: Repeat the above marker screening process and cancer classification model construction process 100 times, and calculate the classification sensitivity, specificity and AUC of the model in the test sample set. The average number of features and the average value of model indicators are shown in the following table:

[0145] Table 2 Classification performance of the model constructed with different numbers of methylation markers after repeated grouping

[0146]

[0147]

[0148] From the above, we can see that the sample classification sensitivity of the methylation markers that reappeared 3 times was the best, and the sample classification specificity of the methylation markers that reappeared 5 times was the best; the sensitivity of the markers that reappeared 2-4 times was above 0.9, and all showed high cancer detection performance. It can be seen that the GLM cancer classification model constructed by methylation markers has a stable screening effect on the above adenocarcinoma.

[0149] Example 2 Comparison of screening methods

[0150] (1) Construction of classification model using markers screened by conventional methods

[0151] Using the training sample set of Example 1, a conventional marker screening method was used to obtain markers. Specifically, LASSO regression was used to screen candidate markers in the entire training sample set, and then a GLM cancer classification model was constructed using the candidate markers in the training sample set, the discrimination threshold was determined based on the Youden index maximization principle, and the GLM cancer classification model was used to calculate the probability of adenocarcinoma in the test sample set, and the sensitivity and specificity of the model classification were obtained.

[0152] (2) Constructing a classification model using candidate markers from a subset of training samples

[0153] The method of Example 1 was used to extract the intervals where the differential CpG sites in the four training sample subsets were located as candidate methylation marker sets. The extraction process was repeated 100 times to obtain a corresponding number of methylation marker sets consisting of methylation markers that were repeated three times. A corresponding number of GLM cancer classification models were constructed in the training sample sets, and the classification performance of each methylation marker set was evaluated in the test sample set and the average was taken.

[0154] (3) Screening markers based on the method of Example 1 and constructing a classification model

[0155] The method of Example 1 was used to extract the intervals where the differential CpG sites in the four training sample subsets were located as candidate methylation marker sets. The extraction process was repeated 100 times to obtain a corresponding number of methylation marker sets consisting of methylation markers that were repeated three times. The methylation markers in each methylation marker set were weighted integrated and a corresponding number of GLM cancer classification models were constructed in the training sample set. The classification performance of each methylation marker set was evaluated in the test sample set and the average was taken.

[0156] (4) Comparison of cancer classification model performance among three methods

[0157] By comparing the performance of the GLM cancer classification model constructed based on markers screened by conventional methods, markers extracted three times from a training sample subset, and markers extracted three times from a training sample subset and weighted, it was found that the weighted markers that recurred three times in Example 1 can achieve marker dimensionality reduction at the data processing level while maintaining the excellence of the model performance. The specific results are as follows Figure 4 shown.

[0158] Depend on Figure 4 It can be seen that A is the number of methylation markers screened by the three methods, and B and C are the sensitivity and specificity of the GLM cancer classification model constructed using the methylation markers screened by the three methods. The number of 3 recurring methylation markers obtained by the cancer methylation marker screening method based on cfDNA according to the embodiment of the present application is significantly reduced compared with the conventional method and the method that does not further extract markers based on the number of recurrences, which will significantly improve the fitting speed of classification models such as GLM; in addition, its classification performance is also improved to a certain extent compared with the commonly used methods, especially in terms of the detection of healthy control samples, which is significantly better than the markers screened by the existing methods.

[0159] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0160] In this application, words such as "including", "comprising", "having", etc. are open words, meaning "including but not limited to", and can be used interchangeably therewith. The words "or" and "and" used herein mean the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein means the phrase "such as but not limited to", and can be used interchangeably therewith.

[0161] It should also be noted that in the method, system and device of the present application, each step or each module can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0162] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. cfDNA-based cancer methylation marker screening method, including: Extracting multiple cfDNA samples from the first population and the second population respectively, extracting all CpG sites in the target interval of each cfDNA sample and the methylation value of each CpG site, so as to obtain the first population sample set and the second population sample set respectively; Selecting part or all of the samples from the first population sample set and the second population sample set to form a training sample set, merging the training sample sets from the first population sample set and the second population sample set and evenly dividing them into l training sample subsets, selecting l-1 training sample subsets from the l training sample subsets multiple times, screening the differential CpG sites between the first population and the second population in the l-1 training sample subsets, and obtaining multiple candidate methylation marker sets; A plurality of screening models are respectively constructed using the plurality of candidate methylation marker sets, and the methylation marker sets are extracted using the plurality of screening models.

2. The method for screening cancer methylation markers based on cfDNA according to claim 1, in, The screening of the differential CpG sites between the first population and the second population in the l-1 training sample subsets comprises: Deleting CpG sites with missing values ​​in the l-1 training sample subsets; Calculate a first variance of each CpG site belonging to the first population sample set and a second variance of each CpG site belonging to the second population sample set in the l-1 training sample subsets, and delete the corresponding CpG site in response to the first variance and / or the second variance being greater than a first threshold.

3. The method for screening cancer methylation markers based on cfDNA according to claim 1, in, The screening of the differential CpG sites between the first population and the second population in the l-1 training sample subsets comprises: LASSO regression is used to obtain the contribution value of each CpG site belonging to the first population sample set and / or the second population sample set in the l-1 training sample subsets, and the corresponding CpG site is deleted in response to the contribution value being no greater than the second threshold.

4. The method for screening cancer methylation markers based on cfDNA according to any one of claims 1 to 3, in, Constructing a plurality of screening models respectively through the plurality of candidate methylation marker sets, and extracting the methylation marker sets using the plurality of screening models comprises: determining a candidate methylation marker for each screening model in the plurality of screening models; Counting the number of recurrences of all the candidate methylation markers in the multiple screening models; In response to the number of recurrences in the multiple screening models being less than a third threshold, the corresponding candidate methylation markers are deleted; in response to the number of recurrences in the multiple screening models being greater than the third threshold, the corresponding candidate methylation markers are retained and a methylation marker set is formed.

5. The method for screening cancer methylation markers based on cfDNA according to claim 3, in, The contribution value of each CpG site belonging to the first population sample set and / or the second population sample set in the l-1 training sample subsets obtained by LASSO regression includes: The regularization parameter of LASSO regression is set to 1 times the standard deviation of the minimum error to calculate the contribution value of each CpG site belonging to the first population sample set and / or the second population sample set in the l-1 training sample subsets.

6. The method for screening cancer methylation markers based on cfDNA according to claim 1, in, The first group is cancer patients, and the second group is healthy people.

7. The method for screening cancer methylation markers based on cfDNA according to claim 1, in, The plurality of screening models includes one or more generalized linear models.

8. A cfDNA-based cancer methylation marker screening device, comprising: A sample extraction unit extracts a plurality of cfDNA samples from the first population and the second population, respectively, extracts all CpG sites in the target interval of each cfDNA sample and the methylation value of each CpG site, so as to obtain a sample set of the first population and a sample set of the second population, respectively; a marker extraction unit, selecting part or all of the samples from the first population sample set and the second population sample set to form a training sample set, merging the training sample sets from the first population sample set and the second population sample set and evenly dividing them into l training sample subsets, selecting l-1 training sample subsets from the l training sample subsets multiple times, screening the differential CpG sites between the first population and the second population in the l-1 training sample subsets, and obtaining a plurality of candidate methylation marker sets; The marker determination unit constructs a plurality of screening models respectively through the plurality of candidate methylation marker sets, and extracts a methylation marker set using the plurality of screening models.

9. The cfDNA-based cancer methylation marker screening device according to claim 8, which performs the cfDNA-based cancer methylation marker screening method according to claims 1-7.

10. An electronic device, comprising: processor; as well as A memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes the cfDNA-based cancer methylation marker screening method as described in any one of claims 1-7.

Images