Cancer marker screening method and device
By screening and combining the CpG region and DMR region of cancer tissue samples and control samples, combining gene mutation markers, the problem of low sensitivity to early diagnosis of liver cancer is solved, and high-performance early cancer diagnosis and screening is achieved.
Patent Information
- Application Number
- CN202310530237.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-11
- Publication Date
- 2024-11-12
AI Technical Summary
The prior art has low sensitivity in early diagnosis of liver cancer, making it difficult to effectively screen out early cancer markers, resulting in delayed diagnosis and high mortality.
By screening the CpG region and DMR region of the cancer tissue sample and the control sample, combining and deduplication treatment, methylation markers were obtained, combining gene mutation markers, and improving the sensitivity and specificity of cancer diagnosis.
It realizes high-performance, simple and cost-controllable early diagnosis and screening of cancer, significantly improving the early detection rate and specificity, which is far higher than traditional methods.
Smart Images

Figure BDA0004233968030000151 
Figure BDA0004233968030000161 
Figure BDA0004233968030000162
Abstract
Description
Technical Field
[0001] The present invention relates to the field of biomedical technology, and in particular to a method for screening cancer markers and a device for implementing the method. Background Art
[0002] In recent years, cancer early screening technology based on liquid biopsy has gradually emerged, and it has attracted much attention due to its low invasiveness and high accuracy. Among them, circulating tumor DNA (ctDNA) is a DNA molecule released into the circulatory system when tumor cells die. By detecting the molecular information it carries, the occurrence of cancer can be further predicted.
[0003] A variety of high-throughput sequencing technologies have laid the foundation for the detection of tumor molecular signals in ctDNA in plasma. The researchers first explored the trace mutation information in circulating free DNA (cfDNA) and found that although these gene mutation signals have strong marker potential, their proportion in the blood is very low and difficult to detect accurately. To address this problem, the researchers developed cfDNA targeted sequencing to improve the sensitivity of target mutation detection, but there are still limitations such as weak early cancer signals and difficulty in detection.
[0004] More and more studies have shown that abnormal methylation levels of cancer genes often occur in the early stages of cancer, and participate in the occurrence and development of cancer by regulating gene expression and silencing. At the same time, methylation sites in the same region often have co-methylation, which further enhances the identification of cancer signals. Researchers speculate that deep sequencing of methylation marker regions will further improve performance, which provides a new method for applying methylation markers to the field of early cancer screening.
[0005] Liver cancer is one of the malignant tumors with high incidence, low early detection rate and high mortality rate. The situation of liver cancer in my country is even more serious, with new liver cancer cases accounting for about 55% of the world's total. Most liver cancers are already in the late stage once they are discovered, missing the best treatment period. Therefore, improving the early detection rate of liver cancer is one of the important ways to improve the survival rate of liver cancer and reduce the mortality rate of liver cancer. At present, the screening methods for high-risk groups of liver cancer in clinical practice are still relatively simple, and are still mainly based on alpha-fetoprotein (AFP) and ultrasound. In relevant reports and clinical studies, the sensitivity of AFP can only reach 41%-65%. Although the detection sensitivity of ultrasound for nodules larger than 20mm can be as high as 84%, for nodules 5-20mm in the liver, the detection sensitivity of ultrasound is only 33%.
[0006] The current clinical early diagnosis technology for liver cancer (AFP+ultrasound) has the problem of low sensitivity or insufficient specificity. Therefore, there is an urgent need to develop markers that can effectively screen liver cancer, which will be of great significance for early cancer screening. Summary of the invention
[0007] In order to improve the accuracy of cancer diagnosis, especially the detection rate of early cancer, the present invention provides a method for screening markers for cancer diagnosis. The methylation markers obtained by the screening method can achieve high-performance, simple, and cost-controlled early diagnosis and screening of cancer, and further improve the sensitivity and specificity of early screening by simultaneously detecting gene mutations and methylation of target areas.
[0008] In a first aspect, the present invention provides a method for screening a marker for cancer diagnosis, the screening method comprising the following steps:
[0009] (1) Comparing the cancer-related CpG regions of cancer tissue samples with the corresponding regions of control samples to screen out CpG marker combinations;
[0010] (2) comparing cancer-associated DMR regions of cancer tissue samples with corresponding regions of control samples to screen for DMR marker combinations; and
[0011] (3) The CpG marker combination and the DMR marker combination are combined and deduplicated to obtain the methylation marker.
[0012] In some embodiments, in step (1), the control sample includes a paracancerous tissue sample, a benign lesion tissue sample and / or a whole blood sample.
[0013] In some embodiments, in step (2), the control sample includes a paracancerous tissue sample, a benign lesion tissue sample and / or a whole blood sample.
[0014] In some embodiments, step (2) comprises the following steps:
[0015] (a1) clustering the cancer tissue samples according to the average methylation level of DMR to obtain a first cluster set,
[0016] (a2) screening out cancer-specific low methylation markers and cancer-specific high methylation markers of each cluster in the first cluster set according to the statistical test results of contribution ratio and average methylation level,
[0017] (a3) merging, removing duplication, and reducing dimension of the cancer-specific low methylation markers and cancer-specific high methylation markers of each cluster obtained in (a2) to obtain a first DMR marker combination.
[0018] In some embodiments, in step (a1), the first cluster set comprises at least 2 clusters, preferably 2-10 clusters, and more preferably 3-5 clusters. In some embodiments, in step (a1), the clustering is selected from hierarchical clustering.
[0019] In some embodiments, in step (a2), the screening includes screening cancer-specific hypomethylation markers according to a contribution ratio of 95%-100% and satisfying a significant difference; screening cancer-specific hypomethylation markers according to a contribution ratio greater than 85%-90% and satisfying a significant difference.
[0020] In some embodiments, in step (a2), the screening includes screening cancer-specific hypomethylation markers according to a contribution ratio of 95%-100%, and satisfying p-value <0.05, preferably <0.01 and deltaM>0.1; screening cancer-specific hypomethylation markers according to a contribution ratio greater than 85%-90%, and satisfying p-value <0.05, preferably <0.01 and deltaM>0.1.
[0021] In some embodiments, in step (a2), the screening includes screening cancer-specific hypomethylation markers according to a contribution ratio = 100%, and satisfying p-value < 0.01 and deltaM> 0.1; screening cancer-specific hypomethylation markers according to a contribution ratio > 90%, and satisfying p-value < 0.01 and deltaM> 0.1.
[0022] In some embodiments, the contribution ratio is obtained by the following steps: splitting the 95% confidence interval of the methylation level distribution of the para-cancer tissue samples, retaining the samples whose methylation levels are outside the 95% confidence interval of the control group in the cancer tissue samples, and the ratio of the number of retained para-cancer tissue samples to all cancer tissue samples is the contribution ratio. For example, for a cancer-specific hypermethylated marker, the contribution ratio indicator is defined as the proportion of cancer tissue samples with an average methylation rate higher than 95% of the para-cancer tissue samples to all cancer tissue samples; these cancer tissue samples with an average methylation rate higher than 95% of the para-cancer tissue samples are defined as contributing samples.
[0023] In some embodiments, in step (a3), the deduplication process includes deduplication of markers having the same contribution ratio and contributing to the same sample.
[0024] In some embodiments, in step (a3), the dimensionality reduction process includes performing unsupervised clustering on the deduplicated markers, and then selecting the marker with the largest contribution ratio in each cluster. Preferably, the deduplicated markers are unsupervised clustered into 80-120 clusters.
[0025] In some embodiments, step (2) comprises the following steps:
[0026] (b1) dividing the cancer tissue samples into a mutation-positive group sample and a mutation-negative group sample,
[0027] (b2) clustering the mutation-negative group samples according to the methylation level of the DMR, sorting them according to the sample size, and placing the mutation-positive group samples as a separate cluster at the end to obtain a second cluster set,
[0028] (b3) selecting the marker ranked first by the overall contribution ratio of the second cluster set and the priority ranking of the statistical test results, and removing the marker ranked first from the DMR,
[0029] (b4) Including the next marker from the remaining markers of the DMR according to the priority of the additional contribution ratio, the overall contribution ratio and the statistical test result of the second cluster set, and repeating this step until the included markers can cover all samples,
[0030] (b5) Repeat steps (b3) and (b4) to obtain a second DMR marker combination.
[0031] In some embodiments, step (2) includes (b6) additionally prioritizing the DMRs according to their contribution ratios in the mutation-positive group of samples and statistical test results, and selecting top-ranked markers to supplement the second DMR marker combination.
[0032] In some embodiments, in step (b6), the ranking threshold is 10-50, preferably 20-30.
[0033] In some embodiments, the mutation-positive group samples are samples of cancer tissues with a high mutation frequency in which any gene has a mutation; the mutation-negative group samples are samples of the aforementioned defined genes without a mutation. In some specific embodiments, in some embodiments, the mutation-positive group samples are samples of cancer tissues with a high mutation frequency in which any gene has a mutation; the mutation-negative group samples are samples of cancer tissues with a high mutation frequency in which any gene has a mutation.
[0034] In some embodiments, in step (b2), the second cluster set is {cluster-N1…cluster-Nk, cluster-P}, where cluster-N1…cluster-Nk represent mutation-negative groups with sample sizes from large to small, and cluster-P represents mutation-positive groups. In some embodiments, in step (b2), the mutation-negative groups are clustered and sorted according to sample size, and the mutation-positive groups are placed at the end as a separate cluster to obtain a second cluster set, so that markers that perform well in mutation-negative samples can be captured, and the mutation-negative samples are sorted from large to small, and finally the mutation-positive groups are placed at the end, so as to distinguish mutation-negative samples as much as possible.
[0035] In some embodiments, in step (b3), initial priority sorting is performed in sequence based on the overall contribution ratio of each cluster in the second cluster set and the P value, deltaM and variance of the statistical test, and the marker ranked first is selected.
[0036] In some embodiments, the deltaM may also be expressed as ΔM, which is the absolute value of the difference between the average methylation rate of the cancer tissue sample and the average methylation rate of the control sample.
[0037] In some embodiments, the variance is the variance of the average methylation rate of cancer tissue samples.
[0038] In some embodiments, in step (b3), taking the mutation-negative samples clustered into 2 clusters as an example: the sorting rules include performing initial priority sorting in the order of contribution ratio (mutation-negative first cluster), contribution ratio (mutation-negative second cluster), contribution ratio (mutation-positive cluster), P value (mutation-negative first cluster), variance (mutation-negative first cluster), P value (mutation-negative second cluster), deltaM (mutation-negative second cluster), variance (mutation-negative second cluster), P value (mutation-positive cluster), deltaM (mutation-positive cluster), variance (mutation-positive cluster), and selecting the marker with the first initial ranking, and removing the marker with the first ranking from the DMR. In some embodiments, the marker with the highest contribution ratio in the mutation-negative first cluster is selected. If the contribution ratios are the same, the contribution ratios in the negative second cluster are compared. If the index is still the same, the contribution ratios in the mutation-positive cluster are further compared. If the index is still consistent, the ranking is continued in the above order... Finally, the marker with the first ranking is selected as the optimal marker.
[0039] In some embodiments, in step (b4), the next marker is included in the remaining markers of the DMR according to the sorting rules of the additional contribution ratio, overall contribution ratio, P value, deltaM and variance of the statistical test of each cluster in the second cluster set.
[0040] In some specific embodiments, in step (b4), the mutation-negative samples are clustered into 2 clusters as an example: the next marker is included in the order of additional contribution ratio (mutation-negative first largest cluster), additional contribution ratio (mutation-negative second largest cluster), additional contribution ratio (mutation-positive cluster), P value (mutation-negative first largest cluster), deltaM (mutation-negative first largest cluster), variance (mutation-negative first largest cluster), overall contribution ratio (mutation-negative first largest cluster), P value (mutation-negative second largest cluster), deltaM (mutation-negative second largest cluster), variance (mutation-negative second largest cluster), overall contribution ratio (mutation-negative second largest cluster), P value (mutation-positive cluster), deltaM (mutation-positive cluster), variance (mutation-positive cluster), and overall contribution ratio (mutation-positive cluster), and this step is repeated (all included markers are eliminated) until the included markers can cover all samples.
[0041] In some specific embodiments, in step (b6), the markers are sorted in order according to the contribution ratio (mutation-positive cluster), P value (mutation-positive cluster), deltaM (mutation-positive cluster), and variance (mutation-positive cluster), and the top 20 markers are selected.
[0042] In some embodiments, the overall contribution ratio refers to the initial calculation, in which all target samples need to be included in the contribution ratio calculation because there are no markers before, which is equivalent to the contribution ratio calculated when the first marker is selected.
[0043] In some embodiments, the additional contribution ratio refers to the part of the samples that can be covered by the previously selected marker set. These samples are eliminated when calculating the additional contribution ratio, and only the samples that cannot be covered (or cannot be distinguished) by the aforementioned markers are retained to evaluate the contribution ratio.
[0044] In some embodiments, in step (b4), if the remaining markers included cannot cover all samples, step (b4) is terminated.
[0045] In some embodiments, in step (b5), steps N3 and N4 are repeated 0-200 times, preferably 50-150 times.
[0046] In some embodiments, the DMR marker combination includes the first DMR marker combination and / or the second DMR marker combination obtained according to the aforementioned method.
[0047] In some embodiments, the first DMR marker combination and / or the second DMR marker combination are both obtained by screening DMR regions associated with cancer.
[0048] In some embodiments, the cancer-associated DMR region is obtained from previous cancer methylation research information and / or an analysis of a CpG site database on a human reference genome.
[0049] In some embodiments, the CpG marker combination is obtained by a method comprising the following steps:
[0050] (c1) performing a statistical test on the CpG sites in the CpG region to screen out CpG sites with significant differences in methylation levels,
[0051] (c2) performing a statistical test on the average methylation level of the CpG islands in the CpG region to screen out CpG islands with significant differences in average methylation levels,
[0052] (c3) The CpG islands of the CpG site distribution obtained in (c1) and the CpG islands obtained in (c2) are merged and deduplicated to obtain a CpG marker combination.
[0053] In some embodiments, in steps (c1) and (c2), the statistical test includes performing statistical tests on cancer tissue samples and adjacent tissue samples, benign lesion tissue samples, and whole blood samples, respectively, to screen out CpG sites or CpG islands with significant differences in methylation levels in the cancer tissue samples.
[0054] In some embodiments, the cancer-associated CpG region is obtained by analyzing previous cancer methylation research information and / or public databases.
[0055] In some embodiments, in steps (c1) and (c2), the statistical test is selected from the moderated-t test.
[0056] In some embodiments, the candidate set of CpG regions associated with cancer is a preliminary screening result, which comes from CpG regions associated with cancer proposed in relevant literature research. In some embodiments, the candidate set of CpG regions associated with cancer includes CpG islands and CpG sites with significant differences between cancer samples and control samples obtained by screening public literature related to cancer.
[0057] In some embodiments, the cancer-associated differentially methylated region (DMR) is a preliminary screening result, which is derived from a DMR proposed in a relevant literature study and / or a DMR initially screened based on a public database.
[0058] In some embodiments, the step of screening for DMRs using a public database comprises:
[0059] (1) Performing moderated-t test on the global sample groups to screen for DMRs with significant differences;
[0060] (2) then pairing the DMRs with large differences in methylation levels between the global sample groups;
[0061] (3) The DMR obtained in (1) and the DMR obtained in (2) are combined and deduplicated.
[0062] In some embodiments, the global sample grouping includes a group of cancer tissue samples and adjacent adjacent tissue samples and a group of cancer tissue samples and whole blood tissue samples.
[0063] In some embodiments, 450K methylation chip data of cancer tissue, paracancerous tissue, and whole blood in a public database are used for screening.
[0064] In some embodiments, the method further comprises the step of screening for gene mutation markers.
[0065] In some embodiments, the gene mutation marker is screened through a public database and related retrospective cohort data. In some embodiments, the public database includes one or more of the MSK database and the COSMIC database.
[0066] In some specific embodiments, based on the cancer mutation frequencies of the MSK database, the COSMIC database, and the Gene+ liver cancer retrospective clinical cohort data, the three data are combined to find hotspot mutation genes with high rankings in the frequency of cancer mutation to be tested as markers of gene mutation.
[0067] In some embodiments, the cancer comprises liver cancer, breast cancer, ovarian cancer, uterine cancer, cervical cancer, brain tumor, thyroid cancer, esophageal cancer, lung cancer, gastric cancer, pancreatic cancer, kidney cancer, colorectal cancer, bladder cancer, lymphoma, melanoma, prostate cancer, testicular cancer, penile cancer.
[0068] In a second aspect, the present invention provides a marker obtained according to the method of the first aspect.
[0069] In some embodiments, the markers include the methylation markers shown in Table 2.
[0070] In some embodiments, the markers further include gene mutation markers, and the gene mutation markers include TP53, TERT, CTNNB1, ARID1A, AXIN1, KRAS, IDH1, PTEN, RB1 and BAP1.
[0071] In a third aspect, the present invention provides a device for the method of the first aspect, comprising a methylation marker module for screening methylation markers;
[0072] Wherein, the methylation marker module includes the following submodules:
[0073] The first submodule is used to compare the cancer-related CpG regions of the cancer tissue sample with the corresponding regions of the control sample to screen out the CpG marker combination;
[0074] A second submodule is used to compare the cancer-related DMR region of the cancer tissue sample with the corresponding region of the control sample to screen out a DMR marker combination; and
[0075] The third submodule is used to merge and remove duplicates of the CpG marker combination and the DMR marker combination to obtain the methylation marker.
[0076] In some embodiments, the device further comprises a gene mutation marker module for screening gene mutation markers.
[0077] In a fourth aspect, the present invention provides a device for predicting cancer, comprising the device described in the third aspect, and a prediction module, wherein the prediction module is used to predict whether the organism to which the sample to be tested belongs is a cancer-positive individual or a cancer-negative individual based on the cfDNA sequencing data of the sample to be tested and the marker.
[0078] In some embodiments, the prediction module uses a logistic regression model to predict whether the organism to which the sample to be tested belongs is a cancer-positive individual or a cancer-negative individual.
[0079] In some embodiments, the prediction module predicts whether the organism to which the sample to be tested belongs is a liver cancer positive individual or a liver cancer negative individual based on the cfDNA test data of the sample to be tested and the filtered single nucleotide mutations.
[0080] In some embodiments, the sample to be tested is a body fluid sample.
[0081] In some embodiments, the body fluid sample includes at least one of blood, plasma, urine, and saliva.
[0082] In some embodiments, the cfDNA sequencing data is methylation sequencing data.
[0083] In a fifth aspect, the present invention provides an apparatus comprising:
[0084] Memory, used to store programs;
[0085] A processor is used to implement the method as described in the first aspect by executing the program stored in the memory.
[0086] In some embodiments, the processor is also used to implement a prediction step by executing a program stored in the memory, wherein the prediction step includes predicting whether the organism to which the sample to be tested belongs is a cancer-positive individual or a cancer-negative individual based on the cfDNA sequencing data of the sample to be tested and the marker.
[0087] In some embodiments, the prediction module uses a logistic regression model to predict whether the organism to which the sample to be tested belongs is a cancer-positive individual or a cancer-negative individual.
[0088] In some embodiments, the prediction module predicts whether the organism to which the sample to be tested belongs is a liver cancer positive individual or a liver cancer negative individual based on the cfDNA test data of the sample to be tested and the filtered single nucleotide mutations.
[0089] In some embodiments, the sample to be tested is a body fluid sample.
[0090] In some embodiments, the body fluid sample includes at least one of blood, plasma, urine, and saliva.
[0091] In some embodiments, the cfDNA sequencing data is methylation sequencing data.
[0092] In a sixth aspect, the present invention provides a computer-readable storage medium having a program stored thereon, wherein the program can be executed by a processor to implement the method as described in the first aspect.
[0093] In some embodiments, the program can also be executed by a processor to implement a prediction step, which includes predicting whether the organism to which the sample to be tested belongs is a cancer-positive individual or a cancer-negative individual based on the cfDNA sequencing data of the sample to be tested and the cancer methylation marker.
[0094] The screening method of cancer markers provided by the present invention combines gene mutations with multi-omics joint markers of methylation to further improve the diagnostic sensitivity and specificity of cancer. The present invention first constructs an initial methylation marker set for screening, which includes CpG regions related to cancer proposed in relevant literature research and differential methylation regions (DMRs) screened from public databases and referenced to the method of CancerDetector, and then screens cancer-specific methylation markers from the constructed set. At the same time, based on the liver cancer mutation frequency of the MSK database and the COSMIC database, the two public databases and the Gene+ retrospective study cohort data are combined to find hot mutation genes with the highest mutation frequency as markers of gene mutations. The screening method provided by the present invention is used to combine gene mutations with methylation markers, and the sensitivity and specificity of cancer diagnosis screening are further improved from the perspective of combined multi-omics, which is much higher than the screening methods currently commonly used in clinical practice.
[0095] Examples and drawings are provided below to help understand the present invention. However, it should be understood that these examples and drawings are only used to illustrate the present invention, but do not constitute any limitation. The actual protection scope of the present invention is set forth in the claims. It should be understood that any modifications and changes can be made without departing from the spirit of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0096] Figure 1 The screening process of methylation markers in Example 1 of the present application is shown.
[0097] Figure 2 A heat map of the average methylation rates of markers screened by method 1 in Example 1 of the present application is shown.
[0098] Figure 3 A heat map of the average methylation rates of markers screened by method 2 in Example 1 of the present application is shown.
[0099] Figure 4 The performance of diagnosing liver cancer using a single example methylation marker in Example 2 of the present application is shown.
[0100] Figure 5 The performance of diagnosing liver cancer using a single example gene mutation marker in Example 2 of the present application is shown.
[0101] Figure 6 The performance of diagnosing liver cancer using a combination of exemplary methylation markers and gene mutation markers in Example 2 of the present application is shown. DETAILED DESCRIPTION
[0102] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with embodiments. The specific embodiments described herein are only used to explain the present invention and are not intended to constitute any limitation of the present invention. In addition, in the following description, the description of known structures and technologies is omitted to avoid unnecessary confusion of the concepts of the present disclosure. Such structures and technologies are also described in many publications.
[0103] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0104] The serial numbers assigned to the components in this article, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning.
[0105] The term "significant difference" or "significant difference" used herein is a statistical term. It is a statistical evaluation of the difference in data. Generally, only when the experimental results reach the 0.05 level or the 0.01 level can it be said that the difference between the data is significant or extremely significant. In some specific embodiments of the present invention, the moderated-t test is used to judge the significant difference. When the p value <0.05, preferably the p value <0.01, it is judged to be a significant difference.
[0106] In this article, DNA methylation is an epigenetic modification, which is catalyzed by DNA methyltransferase (DNMT) using S-adenosylmethionine (SAM) as a methyl donor to selectively add methyl groups to the cytosine of the two nucleotides CG of DNA, mainly forming 5-methylcytosine (5-mC) (commonly found in the 5'-CG-3' sequence of genes) and a small amount of N6-methylpurine (N6-mA) and 7-methylguanine (7-mG).
[0107] The term "CpG region" used herein is mainly located in the promoter and exon regions of genes, and is a region rich in CpG dinucleotides with a length of 300-3000bp. Here, CpG is the abbreviation of cytosine (C)-phosphate (p)-guanine (G). The CpG sequence density in the 5' untranslated region and the first exon region of the promoter region of the gene is very high, exceeding the average by more than 5 times, becoming a guanine and cytosine-rich region, with a GC content of more than 55%.
[0108] The term "CpG region" used herein is mainly located in the promoter and exon regions of genes, and is a region rich in CpG dinucleotides with a length of 300-3000bp. Here, CpG is the abbreviation of cytosine (C)-phosphate (p)-guanine (G). The CpG sequence density in the 5' untranslated region and the first exon region of the promoter region of a gene is very high, exceeding the mean by more than 5 times, becoming a guanine and cytosine enriched region with a GC content of more than 55%. Herein, "CpG region" includes "CpG island" and "CpG site".
[0109] The term "differentially methylated region (DMR)" used herein refers to a region where the methylation modification of the same region of a DNA molecule between samples is significantly different.
[0110] The present invention provides a method for screening cancer markers, wherein the markers include gene mutation markers and methylation markers, wherein the screening of the methylation markers includes the following steps:
[0111] (1) Comparing the cancer-related CpG regions of cancer tissue samples with the corresponding regions of control samples to screen out CpG marker combinations;
[0112] (2) comparing cancer-associated DMR regions of cancer tissue samples with corresponding regions of control samples to screen for DMR marker combinations; and
[0113] (3) The CpG marker combination and the DMR marker combination are combined and deduplicated to obtain the methylation marker.
[0114] In some embodiments, the CpG region is derived from a CpG region associated with cancer proposed in relevant literature research; the DMR is derived from a differentially methylated region (DMR) screened from a public database and a CpG cluster is screened as a DMR using a method based on CancerDetector.
[0115] In some embodiments, using methylation chip data in a public database, firstly, statistical tests are performed through global sample grouping to screen for significant DMRs, and then the DMRs with large differences in methylation levels between groups are screened, and the two parts of DMRs are merged and deduplicated.
[0116] In the present invention, the above-mentioned CpG regions and DMRs are first used to construct a candidate set of initial methylation markers, and then the CpG regions and DMRs are further screened for cancer-specific markers.
[0117] In some embodiments, the CpG regions from the literature are screened using 450K methylation chip data from a public database, including:
[0118] (1) Analyze CpG sites and perform grouped moderated-t test to screen out CpG sites that meet the requirements of significant differences in methylation levels in the three groups of tests.
[0119] (2) Analyze the average methylation level of CpG islands, perform the same group moderated-t test, and screen out CpG islands that have significant differences in average methylation levels.
[0120] (3) All CpG sites involved in (1) and (2) are combined and removed to obtain a CpG marker combination.
[0121] In some embodiments, the methylation chip data of cancer tissue samples and adjacent tissue samples in a public database are used for screening, and the screening method includes the following two methods:
[0122] Method 1: All cancer tissue samples were clustered according to the average DMR methylation level to obtain the first cluster set, and the Hypo markers (cancer-specific low methylation markers) and Hyper markers (cancer-specific high methylation markers) were screened in different clusters in the first cluster set through the statistical test results of contribution ratio and average methylation; the markers of different clusters were further merged and the markers with the same contribution ratio and the same contributing samples were deduplicated; the dimension of the markers was further reduced by unsupervised clustering and selecting the marker with the largest contribution in each cluster (such as clustering into 100 clusters, selecting the marker with the largest contribution in each cluster, and finally reducing the dimension of the markers to 100), and the first DMR marker combination was obtained.
[0123] Method 2: The cancer tissue samples are divided into mutation-positive group samples and mutation-negative group samples for methylation marker screening, where the mutation-positive group samples refer to any gene mutation in the top five genes with the highest mutation frequency in the cancer tissue samples. In order to better identify effective methylation markers (complementary to the mutation dimension) in mutation-negative group samples and supplementary methylation markers in mutation-positive group samples, the present invention constructs an iterative algorithm:
[0124] (1) Unsupervised clustering of cancer tissue samples in the mutation-negative group was performed based on the methylation level of DMR to obtain the second cluster set, which was sorted according to the sample size. The mutation-positive group was placed at the end as a separate cluster, thus obtaining the cluster set {cluster-N1…cluster-Nk, cluster-P}, where cluster-N1…cluster-Nk represent mutation-negative clusters with decreasing sample size, and cluster-P represents the mutation-positive group.
[0125] (2) Perform initial priority sorting according to the overall contribution ratio of the cluster set (the order of the cluster set), statistical test P value, deltaM, and variance, select the current optimal DMR, and remove this DMR from the DMR set.
[0126] (3) Further traverse the remaining markers, and include the next marker according to the priority sorting rules of the additional contribution ratio of the cluster set (the order of the cluster set), the P value of the cluster set, deltaM, variance, and the overall contribution ratio (the order of the cluster set). Repeat this step until the included marker can cover all samples. This is one screening. If the remaining markers cannot cover all samples, terminate the screening.
[0127] (4) repeating the processes (2) and (3) to screen and obtain a second DMR marker combination;
[0128] (5) The DMRs are additionally prioritized according to their contribution ratios in the mutation-positive group and statistical test results, and the top-ranked markers are selected and added to the second DMR marker combination; and the DMR marker combination B is deduplicated.
[0129] The CpG marker combination obtained by the above screening and the first DMR marker combination and the second DMR marker combination are combined and deduplicated to form the final cancer methylation diagnostic marker composition.
[0130] In some embodiments, gene mutation markers are screened through public databases and related retrospective cohort data. In some embodiments, the public database includes one or more of the MSK database and the COSMIC database.
[0131] In some specific embodiments, based on the cancer mutation frequencies of the MSK database, the COSMIC database, and the Gene+ liver cancer retrospective clinical cohort data, the three data are combined to find hotspot mutation genes with high rankings in the frequency of cancer mutation to be tested as markers of gene mutation.
[0132] The above-screened methylation markers are combined with gene mutation markers to form combined markers, which further improve the sensitivity and specificity of cancer diagnosis and screening from the perspective of combined multi-omics.
[0133] Example 1 Screening of liver cancer diagnostic marker combinations
[0134] 1. Construction of candidate sets of methylation markers
[0135] This embodiment pre-constructs a candidate set of methylation markers with a wide range of sources, which mainly includes two parts.
[0136] The first part includes literature on liver cancer methylation (e.g., John B Kisiel, et al., Hepatocellular Carcinoma Detection by Plasma Methylated DNA: Discovery, Phase I Pilot, and Phase II Clinical Validation.. Hepatology, 2019, Mar; 69(3): 1180-1192, and Xu Rui-Hua, et al., Circulating tumour DNA methylation markers for diagnosis and prognosis of hepatocellular carcinoma. Nature materials, 2017Nov;16(11):1155-1161), including ACP1, AK055957, CCNJ_3707, CCNJ_3124, CLEC11A, DAB2IP, ECE1, EFNB2, EMX1, HOXA1, PFKP, SPINT2, SCRN1, TSPYL5, BDH1, COTL1, LRRC4, ARHGAP25, ATAD2, ATXN1, BMPR1A, Chr6:170, Chr6:3, Chr8:20, KLF3, PLAC8, CpG island regions in PSD, and cg13601799 site, totaling 126 CpG islands and 1 CpG site.
[0137] The second part includes differentially methylated regions (DMRs) associated with liver cancer, including, for example, 3284 DMRs provided in the Cancer Detector document (Li W, et al. Cancer Detector: ultrasensitive and non-invasive cancer detection at the resolution of individual reads using cell-free DNA methylation sequencing data. Nucleic Acids Res 2018 Sep6; 46(15): e89), as the first DMR set. In addition, the DMRs obtained by screening are also obtained by the following method, and the screening method comprises the following steps:
[0138] Cluster splitting is performed based on 26752698 CpG sites on the human reference genome:
[0139] (1) First, we took the 100 bp intervals upstream and downstream of each site and merged them if they overlapped, resulting in 3,941,197 intervals.
[0140] (2) Filtering step (1) to retain intervals covering ≥ 3 CpG sites, resulting in 2,063,765 intervals; then, further filtering using the 450K chip (Illumina), requiring the interval to contain at least one 450K site, resulting in 147,438 intervals;
[0141] (3) CpG clusters with a length of >1 kb were further split; a sliding window split was performed with a window size of 1000 bp and a step size of 500 bp, retaining a total of 312,973 CpG clusters; after removing blacklist regions such as multi-region alignment and SNP-related regions, 277,418 CpG clusters were finally obtained;
[0142] (4) A moderated-t test was performed on the CpG clusters obtained in (3) using 582 liver cancer tissue samples from the public database (derived from the TCGA database, and GSE60753, GSE54503, GSE75041, GSE99036, GSE89852, and GSE77269 from the GEO database) and 291 adjacent paracancerous tissue samples (derived from the TCGA database, and GSE60753, GSE54503, GSE48325, GSE61258, GSE89852, GSE107038, and GSE77269 from the GEO database). CpG clusters with a P value less than 0.01 and an average methylation rate of liver cancer tissue samples higher than 0.1 or lower than 0.2 of the average methylation rate of adjacent paracancerous tissue samples were screened out;
[0143] (5) A moderated-t test was performed on the CpG clusters obtained in (4) using 582 liver cancer tissue samples (the same as in (4)) and 732 white blood cell (WBC) samples (from GSE87571) in the public database, and CpG clusters whose average methylation rate in liver cancer tissue samples was 0.1 higher than the average methylation rate in WBC samples or 0.2 lower than the average methylation rate in WBC samples were screened out. A total of 8,618 CpG clusters were recorded as the second DMR set.
[0144] The 126 CpG islands and 1 CpG site obtained above, the first DMR set containing 3284 DMRs, and the second DMR set containing 8618 CpG clusters together constitute a candidate set of methylation markers.
[0145] 2. Screening of methylation markers
[0146] The above pre-constructed candidate set of methylation markers was further screened. The screening process was mainly divided into two parts. One part was to screen 126 CpG islands and 1 CpG site from the literature, and the other part was to screen the first and second DMR sets (i.e., 3284+8618 DMRs). Figure 1 shown.
[0147] (1) Screening of CpG islands and CpG sites
[0148] In the screening process of CpG islands and CpG sites, this example included 582 liver cancer tissue samples (same as above), 291 adjacent cancer tissue samples (same as above), 94 liver cirrhosis samples (from GSE60753, GSE99036), and 732 WBC samples (same as above) for statistical screening.
[0149] ①First, three groups of moderated-t tests were performed on all CpG sites and CpG sites on CpG islands (liver cancer vs. adjacent tissues, liver cancer vs. cirrhosis, liver cancer vs. whole blood) to screen CpG sites that simultaneously met the requirements of methylation level difference P value < 0.01 and deltaM difference above 0.1. A total of 193 CpG sites were found, distributed on 48 CpG islands.
[0150] ②Analyze the average methylation level of CpG islands. Similarly, three groups of moderated-t tests were performed (liver cancer vs. adjacent tissues, liver cancer vs. cirrhosis, liver cancer vs. whole blood). CpG islands that simultaneously satisfied the conditions of deltaM greater than 0.1 (liver cancer vs. adjacent tissues), 0.05 (liver cancer vs. cirrhosis), and 0.048 (liver cancer vs. whole blood) and P values less than 0.01 were screened, resulting in 35 CpG islands.
[0151] ③ The CpG islands obtained in ① and ② were merged and deduplicated, resulting in a total of 50 CpG islands, covering 436 CpG sites on the 450K chip.
[0152] (2) Screening of DMRs
[0153] In the process of DMR screening, this example included 375 liver cancer tissue samples and 50 adjacent tissue samples from TCGA for screening. Two methods were used to further screen DMRs and finally merge them.
[0154] Method 1: According to the methylation levels of all samples, the liver cancer group samples were hierarchically clustered and divided into three clusters: cluster 1, cluster 2, and cluster 3. The frequencies of core mutations in the three clusters (TERT, TP53, AXIN1, ARID1A, and CTNNB1) were counted. The results showed that fewer samples had mutations in cluster 3, while more samples had mutations in cluster 1 and cluster 2. The clusters were divided to screen for Hypo markers (regional methylation levels of liver cancer tissues were lower than those of adjacent tissues) and Hyper markers (regional methylation levels of liver cancer tissues were higher than those of adjacent tissues):
[0155] ① Screening of Hypo markers: Cluster 1 obtained 378 markers based on contribution ratio = 100%; Cluster 2 obtained 1748 markers based on contribution ratio = 100%; Cluster 3 had no markers with contribution ratio = 100%; further screening of three groups of comparisons that simultaneously satisfied deltaM>0.1 and p-value <0.01 resulted in 158 markers. The three groups of markers were merged to remove duplicates, and markers with exactly the same contribution sample set were removed, and finally 128 markers were retained.
[0156] ② Screening of Hyper markers: Cluster 1 obtained 210 markers based on contribution ratio>90%; Cluster 2 obtained 361 markers based on contribution ratio>90%; similarly, Cluster 3 had no markers whose contribution ratio did not meet the requirements, and further screening of three groups of comparisons that simultaneously met deltaM>0.1 and p-value<0.01 resulted in 117 markers. The three groups of markers were merged to remove duplicates, and markers with exactly the same contribution sample set were removed, and finally 105 markers were retained.
[0157] ③Evaluate the coverage of the markers selected in ① and ② for cluster 3 samples, and reduce the dimension of the markers through unsupervised clustering. Specifically, cluster them into 100 clusters through k-means clustering, select the markers with the largest contribution ratio of each cluster to cluster 3, and finally select 100 markers with a size of 73K. The average methylation rate heat map of the selected markers is shown in Figure 2 As shown, the sampletype label is the label of cancer and adjacent to cancer.
[0158] The contribution ratio is obtained by the following steps: splitting the 95% confidence interval of the methylation level distribution of para-cancerous tissue samples, retaining the samples whose methylation levels are outside the 95% confidence interval of the control group, and the ratio of the number of retained para-cancerous tissue samples to all cancer tissue samples is the contribution ratio. For example, for a cancer-specific hypermethylated marker, the contribution ratio indicator is defined as the proportion of cancer tissue samples with an average methylation rate higher than 95% of para-cancerous tissue samples to all cancer tissue samples. These cancer tissue samples with an average methylation rate higher than 95% of para-cancerous tissue samples are defined as contributing samples.
[0159] Method 2: Samples with both mutation information and methylation in TCGA (196 cases of liver cancer tissues and 50 cases of adjacent cancer tissues) were selected, and the cancer tissue samples were divided into a mutation-positive group and a mutation-negative group (among which, the mutation-negative group samples refer to the top five genes with the highest mutation frequency in cancer tissue samples, TERT, TP53, AXIN1, ARID1A, and CTNNB1, without mutations, and the mutation-positive group refers to the five genes, TERT, TP53, AXIN1, ARID1A, and CTNNB1, with any mutation). The samples in the mutation-negative group were clustered into three clusters (cluster 1, cluster 2, and cluster 3) according to the methylation level, and the initial priority was sorted according to the P value, deltaM, variance, and overall contribution ratio of Cluster 2, Cluster 3, Cluster 1, and Cluster 4 (sorted by sample size, among which cluster 4 was the mutation-positive cluster). Specifically, according to the order of Cluster 2 contribution ratio, Cluster 3 contribution ratio, Cluster 1 contribution ratio, Cluster 4 contribution ratio, Cluster 2 P value, Cluster 2 deltaM, Cluster 2 variance, Cluster 3 P value, Cluster 3 deltaM, Cluster 3 variance, Cluster 1 P value, Cluster 1 deltaM, Cluster 1 variance, Cluster 4 P value, Cluster 4 deltaM, and Cluster 4 variance (where contribution ratio and deltaM are sorted from large to small, and P value and variance are sorted from small to large), the marker ranked first is selected as the first marker. In other words, first look at the marker with the largest contribution ratio of Cluster 2. If they are equally high, look at the contribution ratio of Cluster 3. If they are still equally high, look at the next set of data, and finally select the marker ranked first as the first marker.
[0160] Then traverse the remaining markers, and include the next marker according to the additional contribution ratio of Cluster 2, the additional contribution ratio of Cluster 3, the additional contribution ratio of Cluster 1, the additional contribution ratio of Cluster 4, the P value of Cluster 2, the deltaM of Cluster 2, the variance of Cluster 2, the overall contribution ratio of Cluster 2, the P value of Cluster 3, the deltaM of Cluster 3, the variance of Cluster 3, the overall contribution ratio of Cluster 3, the P value of Cluster 1, the deltaM of Cluster 1, the variance of Cluster 1, the overall contribution ratio of Cluster 1, the P value of Cluster 4, the deltaM of Cluster 4, the variance of Cluster 4, the overall contribution ratio of Cluster 4, and the sorting rules. Repeat this step until the included markers can cover all samples. This is one screening. This process was repeated 100 times, and 54 groups of markers were finally selected. The DMRs were prioritized according to their contribution ratio in Cluster 4 and statistical test results, and the top 20 markers were selected. The 54 groups of markers were combined with the 20 positive mutation group markers for deduplication. The average methylation rate heat map of the screening results is shown in Figure 2. Figure 3 As shown; after removing the regions overlapping with the 100 DMRs and 50 CpG islands screened by the aforementioned method, a total of 193 methylation markers remained.
[0161] Among them, deltaM is the absolute value of the difference between the average methylation rate of cancer tissue samples and the average methylation rate of control samples. The overall contribution ratio refers to the initial calculation, because there are no markers before, so all target samples need to be included in the contribution ratio calculation, which is equivalent to the contribution ratio calculated when the first marker is selected. The additional contribution ratio refers to the part of the samples that can be covered by the previously selected marker set. These samples are eliminated when calculating the additional contribution ratio, and only the samples that cannot be covered (or cannot be distinguished) by the aforementioned markers are retained to evaluate the contribution ratio.
[0162] (3) Methylation marker combination
[0163] The 50 CpG islands screened in step (1), the 100 markers screened in method 1 in step (2), and the 193 markers screened in method 2 were included in the final methylation marker combination, for a total of 343 methylation markers.
[0164] 3. Screening of mutation markers
[0165] Based on the cancer mutation frequency of MSK database, COSMIC database, and Gene+ liver cancer retrospective clinical cohort data, the three databases were combined to find hotspot mutation genes with high cancer mutation frequencies to be tested, and 10 mutation markers were screened, namely TP53, TERT, CTNNB1, ARID1A, AXIN1, KRAS, IDH1, PTEN, RB1, and BAP1. Among them, TP53, TERT, CTNNB1, ARID1A and AXIN1 are hotspot genes for hepatocellular carcinoma, KRAS, IDH1 and PTEN are hotspot genes for intrahepatic ductal carcinoma, and RB1 and BAP1 are hotspot genes for cancer drug use.
[0166] Example 2 Performance Verification of Liver Cancer Diagnostic Marker Combination
[0167] 1. Targeted capture sequencing for co-detection of mutations and methylation
[0168] (1) Plasma separation and DNA extraction
[0169] For whole blood, plasma / blood cell separation should be performed promptly (EDTA anticoagulant tube, within 4 hours; Streck tube within 72 hours). The separation steps are as follows:
[0170] ① Centrifuge at 1600g for 10 minutes at 4°C. After centrifugation, divide the upper layer of plasma into multiple 1.5mL or 2.0mL centrifuge tubes. Be careful not to aspirate the white blood cells in the middle layer during the plasma aspiration process.
[0171] ② Centrifuge at 16000g for 10 min at 4°C to remove residual cells, transfer the supernatant into a new 1.5mL or 2.0mL centrifuge tube (be careful not to aspirate the white blood cells at the bottom of the tube) to obtain the required plasma.
[0172] Plasma cfDNA was extracted according to the QIAamp Circulating Nucleic Acid Kit (Qiagen) extraction reagent instructions.
[0173] (2) Library construction and hybridization capture
[0174] Using Hieff The Ultima Pro DNA Library Prep Kit for Illumina is used to extract cfDNA from plasma samples of liver cancer, healthy patients, cirrhosis patients, and liver nodules. The fragmented cfDNA is end-repaired, "A" is added, and adapters are added. It is then oxidized by TET enzyme and reduced by pyridine borane. It is then amplified by PCR to construct a sequencing-recognizable library and complete hybridization capture.
[0175] (3) Sequencing
[0176] Gene+seq sequencer and other sequencers with the same principle were used for sequencing. Sequencing experiments were performed according to the instructions provided by the manufacturer.
[0177] 2. Sequencing results analysis method
[0178] The present invention collected sequencing data of methylation and mutation co-detection of 165 samples by performing targeted methylation high-throughput sequencing on a total of 165 clinical plasma samples from 61 liver cancer patients (30 of whom were BCLC A stage patients), 25 healthy people, 43 patients with benign liver nodules, and 36 patients with cirrhosis.
[0179] (1) Methylation model
[0180] The methylation model refers to the formula in the CancerDetector literature (Li W, Li Q, Kang S, et al. CancerDetector: ultrasensitive and non-invasive cancer detection at the resolution of individual reads using cell-free DNA methylation sequencing data. Nucleic Acids Res 2018; 46: e89).
[0181] (1-1) Extract all reads in the methylation marker region and calculate the probability of coming from the tumor
[0182] All reads in the methylation marker region are extracted from the sequencing results, and the CpG site status in the methylation marker region where each read is located is counted. Based on the TCGA liver cancer tissue and adjacent tissue 450K chip data, the methylation rate distribution of each methylation marker can be obtained (in accordance with the beta distribution). Therefore, based on the prior distribution of each methylation marker (TCGA data), the probability of each read coming from a tumor or non-tumor can be calculated by the joint probability density. The formula is as follows:
[0183]
[0184]
[0185] r represents the methylation signal of a read, r = (r1, r2, ...), where r j=1 or 0 represents the methylation status of the jth CpG site on the r read as methylated or unmethylated; m = Beta (η, ρ) represents the prior probability distribution (beta distribution) of the methylation rate in the region; r j Each CpG site on the network conforms to the Bernoulli distribution Bernoulli(p), where p~Beta(η,ρ); m c Represents the number of methylated CpG sites on the read, um c Represents the number of unmethylated CpG sites in the read.
[0186] (1-2) Evaluate the proportion of ctDNA at each methylation marker level using the maximum likelihood method
[0187] The possible ratios are pre-set, such as dividing 0-1 into 1000 equal parts (i.e., from 0, 0.001, 0.002…, 1), and each ratio is applied to the reads probability obtained above by the maximum likelihood method to evaluate the tumor component of the plasma sample. The formula is as follows:
[0188]
[0189] Where θ represents the preset ratio, P(R|θ,M) is the maximum likelihood probability when the tumor component of the plasma sample is estimated to be θ; r i Represents the methylation signal of the i-th read in the k-th marker region; represents the a priori distribution pattern of methylation of the k-th marker in the tumor sample, represents the a priori distribution pattern of methylation of the kth marker in non-tumor samples; finally, θ when the maximum likelihood probability is the largest is selected as the final tumor component estimate (TF=θ), that is, each marker obtains a TF prediction value.
[0190] (1-3) Logistic regression model to evaluate the liver cancer risk value of samples
[0191] The TF prediction values of all markers were modeled using a 10-fold cross-validation based on the logistic regression model. In each of the 10 cross-validations, 90% of the samples were used as training sets (for modeling) and 10% as validation sets (for model validation). That is, each sample had a prediction value (Meythlation score) as the validation set, and the diagnostic performance was evaluated using the prediction value.
[0192] Meythlation score = f(TF majer1 ,TF maker ,…TF maker30 )
[0193] (2) Mutation diagnosis
[0194] ① Identification and annotation of mutations
[0195] The information analysis process of the plasma ctDNA low-frequency mutation and methylation co-detection technology disclosed in CN112634984A (RealSeq2 Pipeline) completes the quality control, filtering, initial alignment, clustering error correction and deduplication of reads, and extracts the deduplicated reads including natural repeat fragments; uses GATK's RealignerTargetCreator to perform local realignment of sequences near INDELs to reduce the alignment error rate near INDELs; uses GATK's BaseRecalibrator and PrintReads to recalibrate the base quality values of reads in the bam file, so that the quality values of the bases in the reads in the final output bam file can be closer to the actual probability of mismatch with the reference genome, and re-outputs the quality-corrected reads; uses the mutation and methylation signal co-recognition software realDecaller2 and Mutect2 disclosed in CN115410649A to detect somatic mutations and hematopoietic clone mutations, and uses the annotation software SomVAS to annotate the original variant detection mutations, including mutation information and external databases ESP, GAD, EXAC, 1000Genomes and GenomesAD.
[0196] ② Filtering of mutations
[0197] Single nucleotide mutations that passed the following filters were retained:
[0198] Preliminary filtering of mutations was performed using mutation information, methylation information, annotation information, and a healthy human baseline constructed using plasma samples from healthy humans;
[0199] A single nucleotide mutation source prediction model was constructed using a machine learning model, and a logistic regression model was constructed using the following 12 mutation features. The training dataset included healthy cfDNA and a defined positive set (high-quality mutations detected in both liver cancer tissue and plasma). Single nucleotide mutations with mutation prediction values higher than the threshold were retained.
[0200] SNV score=f(Ds,Mbfz,Vpir,Pq,Mq,Trans,Sfs1,Sfs2,Mnonb,Nbfs,Ndd,Hot)
[0201] SNV score is the predicted value of the single nucleotide mutation source prediction model, and f is the logistic regression model; Ds represents the number of double-stranded support mutations; Mbfz represents the average cluster size of support mutations, Vpir represents the average position of mutations on reads, Pq represents the quality value of mutations, Mq represents the proportion of alignment quality ≥ 30 in reads supporting mutations, Trans represents whether the mutation is a transversion mutation, and Sfs1 and Sfs2 represent the fragment features associated with mutation reads (for details, refer to Lung-CLiP literature [1] ), Mnonb represents the average number of non-reference bases in the reads supporting the mutation, Nbfs represents the normalized average cluster size supporting the mutation, Ndd represents the normalized effective depth at the mutation site, and Hot represents whether the mutation is located in the 28 liver cancer hotspot mutation sets.
[0202] ③ Mutation for liver cancer risk assessment
[0203] After mutation identification, annotation, and filtering, if one or more single nucleotide mutations are detected, the sample is predicted to be positive for liver cancer; if none are detected, the sample is predicted to be negative.
[0204] 3. Diagnostic Performance Evaluation
[0205] (1) Methylation performance evaluation
[0206] 30 markers were randomly selected from the 343 methylation marker sets screened in Example 1 for model construction and performance evaluation was performed using a 10-fold cross-validation validation set, and the above operation was repeated 10 times to evaluate the predictive performance of the screened methylation markers. The results are shown in Table 1. When the specificity is about 97%, the predictive sensitivity of the methylation marker can reach 82%-89%, and the marker sets randomly selected 10 times all have good predictive performance for hepatocellular carcinoma HCC.
[0207] Table 1
[0208]
[0209]
[0210] From the 10 randomly selected combinations, the set with the highest specificity and sensitivity (Run 7) was selected as a specific example to demonstrate the composition and performance. Its marker composition is shown in Table 2 below; Figure 4 The performance of Run7's methylation model in predicting liver cancer was demonstrated, with an AUC of 0.957, a specificity of 0.971, and a sensitivity of 0.885. Its prediction sensitivity for patients in the BCLCA stage reached 0.833.
[0211] Table 2
[0212]
[0213]
[0214] (2) Mutation performance evaluation
[0215] Figure 5 is the AUC of the mutation marker in the same sample set. The results showed that AUC = 0.82, specificity was 100%, and sensitivity was 63.9%; among them, the prediction sensitivity of BCLC A stage samples was 60%.
[0216] (3) Integration performance of methylation and mutation
[0217] If any of the methylation and mutation indicators is positive, the sample is predicted to be positive. Table 3 below shows the performance of the integration of different randomly selected methylation marker combinations and mutations. From the results, the prediction model constructed by the methylation marker set of 10 random combinations can improve the sensitivity by 1.7%-6.6% (an average increase of 4.11%) through the complementation of mutations; that is, when the methylation prediction sensitivity is already high, adding mutations still has a significant complementary improvement, which also reflects that the methylation marker screening strategy can provide a better methylation marker component for the multi-omics marker combination.
[0218] Table 3
[0219]
[0220] Also take Run7 as a specific example, Figure 6 This is the performance result of the integration of Run7's methylation markers and mutation markers, with a specificity of 0.971, a sensitivity of 0.918, and a prediction sensitivity of 0.900 for BCLC stage A patients. Compared with the methylation dimension alone, the prediction sensitivity of the multi-omics prediction model for BCLC stage A patients increased by 6.67%.
[0221] The technical solution of the present invention is not limited to the above-mentioned specific embodiments. All technical variations made according to the technical solution of the present invention fall within the protection scope of the present invention.
Claims
1. A method for screening cancer markers, the screening method comprising the following steps: (1) Comparing the cancer-related CpG regions of cancer tissue samples with the corresponding regions of control samples to screen out CpG marker combinations; (2) comparing cancer-related DMR regions of cancer tissue samples with corresponding regions of control samples to screen for DMR marker combinations; and (3) The CpG marker combination and the DMR marker combination are combined and deduplicated to obtain the methylation marker.
2. The method according to claim 1, characterized in that In step (1), the control sample includes a paracancerous tissue sample, a benign lesion tissue sample and / or a whole blood sample; and / or In step (2), the control sample includes a paracancerous tissue sample, a benign lesion tissue sample and / or a whole blood sample.
3. The method according to claim 1, characterized in that Step (2) comprises the following steps: (a1) clustering the cancer tissue samples according to the average methylation level of DMR to obtain a first cluster set, (a2) screening out cancer-specific low methylation markers and cancer-specific high methylation markers of each cluster in the first cluster set according to the statistical test results of contribution ratio and average methylation level, (a3) merging, removing duplications and reducing the dimension of the cancer-specific low methylation markers and cancer-specific high methylation markers of each cluster obtained in (a2) to obtain a first DMR marker combination.
4. The method according to claim 3, characterized in that: In step (a1), the first cluster set comprises at least 2 clusters, preferably 2-10 clusters, and more preferably 3-5 clusters.
5. The method according to claim 3, characterized in that: In step (a2), the screening includes: screening cancer-specific hypomethylation markers according to a contribution ratio of 95%-100% and satisfying a significant difference; screening cancer-specific hypomethylation markers according to a contribution ratio greater than 85%-90% and satisfying a significant difference.
6. The method according to claim 3, characterized in that In step (a3), the deduplication process includes deduplication of markers with the same contribution ratio and the same contributing sample, and / or The dimensionality reduction process includes performing unsupervised clustering on the markers after deduplication, and then selecting the marker with the largest contribution ratio in each cluster. Preferably, the markers after deduplication are clustered unsupervisedly into 80-120 clusters.
7. The method according to claim 1, characterized in that Step (2) comprises the following steps: (b1) dividing the cancer tissue samples into a mutation-positive group sample and a mutation-negative group sample, (b2) clustering the mutation-negative group samples according to the methylation level of the DMR, sorting them according to the sample size, and placing the mutation-positive group samples as a separate cluster at the end to obtain a second cluster set, (b3) selecting the first-ranked marker by ranking the overall contribution ratio of the second cluster set and the priority of the statistical test results, and removing the first-ranked marker from the DMR, (b4) Including the next marker from the remaining markers of the DMR according to the priority of the additional contribution ratio, the overall contribution ratio and the statistical test result of the second cluster set, and repeating this step until the included markers can cover all samples, (b5) repeating steps (b3) and (b4) to obtain a second DMR marker combination; Preferably, the method further comprises (b6) prioritizing the DMRs according to the contribution ratios of the mutation-positive group samples and the statistical test results, and selecting markers with higher rankings to supplement the second DMR marker combination; Preferably, in step (b4), if the remaining markers included cannot cover all samples, step (b4) is terminated; Preferably, in step (b5), steps N3 and N4 are repeated 0-200 times, preferably 50-150 times; Preferably, in step (b6), the ranking threshold is 10-50, preferably 20-30.
8. The method according to claim 7, characterized in that In step (b3), initial priority ranking is performed in turn according to the overall contribution ratio of each cluster in the second cluster set and the P value, deltaM and variance of the statistical test, and the marker ranked first is selected; and / or In step (b4), the next marker is included in the remaining markers of the DMR according to the priority ranking of the additional contribution ratio, overall contribution ratio, P value, deltaM and variance of each cluster in the second cluster set.
9. The method according to claim 1, characterized in that: The DMR marker combination includes a first DMR marker combination obtained according to the method of any one of claims 3 to 6 and / or a second DMR marker combination obtained according to the method of claim 7 or 8.
10. The method according to claim 1, characterized in that The CpG marker combination is obtained by a method comprising the following steps: (c1) performing a statistical test on the CpG sites in the CpG region to screen out CpG sites with significant differences in methylation levels, (c2) performing a statistical test on the average methylation level of the CpG islands in the CpG region to screen out CpG islands with significant differences in average methylation levels, (c3) merging and removing duplicates of the CpG islands of the CpG site distribution obtained in (c1) and the CpG islands obtained in (c2) to obtain a CpG marker combination; Preferably, in steps (c1) and (c2), the statistical test includes performing statistical tests on cancer tissue samples and adjacent tissue samples, benign lesion tissue samples and whole blood samples, respectively, to screen out CpG sites or CpG islands with significant differences in methylation levels in the cancer tissue samples.
11. The method according to claim 1, characterized in that: The method also includes the step of screening for gene mutation markers.
12. The method according to claim 1, characterized in that The cancers include liver cancer, breast cancer, ovarian cancer, uterine cancer, cervical cancer, brain tumor, thyroid cancer, esophageal cancer, lung cancer, stomach cancer, pancreatic cancer, kidney cancer, colorectal cancer, bladder cancer, lymphoma, melanoma, prostate cancer, testicular cancer, and penile cancer.
13. A marker obtained according to the method of any one of claims 1 to 12.
14. A device for use in the method according to any one of claims 1 to 12, comprising a methylation marker module for screening methylation markers; in, The methylation marker module includes the following submodules: The first submodule is used to compare the cancer-related CpG regions of the cancer tissue sample with the corresponding regions of the control sample to screen out the CpG marker combination; The second submodule is used to compare the cancer-related DMR regions of the cancer tissue samples with the corresponding regions of the control samples to screen out the DMR marker combination; and The third submodule is used to merge and remove duplicates of the CpG marker combination and the DMR marker combination to obtain the methylation marker.
15. The device according to claim 14, characterized in that The device also includes a gene mutation marker module for screening gene mutation markers.
16. A device for predicting cancer, comprising the device according to claim 14 or 15, and a prediction module, wherein the prediction module is used to predict whether the organism to which the sample to be tested belongs is a cancer-positive individual or a cancer-negative individual based on the cfDNA sequencing data of the sample to be tested and the marker.
17. An apparatus comprising: Memory, used to store programs; A processor, configured to implement the method according to any one of claims 1 to 12 by executing the program stored in the memory.
18. A computer-readable storage medium, wherein a program is stored on the medium, and the program can be executed by a processor to implement the method according to any one of claims 1 to 12.
