Cancer diagnosis model based on cfDNA transposon fragmentation mode and application

Abstract

The invention belongs to the technical field of cancer diagnosis, and particularly discloses a cancer diagnosis model based on a transposon fragmentation mode in cfDNA and application. The first aspect of the invention provides application of a fragmentation mode of transposon in cfDNA as a marker in preparation of a product. The application of the product is at least one of the following applications: (1) cancer diagnosis; and (2) predicting cancer types. Compared with an existing marker, the cancer diagnosis marker provided by the invention has a better diagnosis effect, has good specificity and sensitivity, can be used for diagnosis of various cancer types, and also has certain tumor tissue traceability. The invention further provides a cancer diagnosis device, a cancer diagnosis model construction method, a cancer diagnosis model and a sample classification method.

Description

Technical Field

[0001] This application relates to the field of cancer diagnostic technology, and in particular to a cancer diagnostic model and its application based on transposon fragmentation patterns in cfDNA. Background Technology

[0002] Circulating cell-free DNA (cfDNA) is a type of DNA fragment produced by the natural degradation of cells after death. Studies have shown that in healthy individuals, cfDNA primarily originates from blood cells and the liver; however, in cancer patients, dead tumor cells also release cfDNA into the peripheral blood. cfDNA fragments from tumors differ significantly from those from non-tumor cells because they are often affected by gene mutations, chromosomal copy number variations, or abnormal methylation levels. Therefore, cancer can be diagnosed by detecting tumor-derived cfDNA signals in peripheral blood, a technique known as liquid biopsy. Since peripheral blood collection is virtually non-invasive and allows for repeated sampling, cancer diagnostic techniques based on peripheral blood cfDNA have high application value. Related research has identified numerous tumor cfDNA biomarkers and developed a large number of cancer diagnostic techniques, such as high-frequency gene mutation kits and cancer-specific methylation biomarker kits.

[0003] In recent years, researchers have discovered that cfDNA fragmentation is not random and varies between tumor and normal cells. Based on this principle, researchers have identified several tumor biomarkers related to cfDNA fragmentation patterns and developed several tumor diagnostic technologies. For example, liver cancer patients have a set of preferred cfDNA end sites distributed across their genomes. These preferred end sites can distinguish liver cancer patients from control groups and can therefore be used as tumor biomarkers. However, currently, there are still relatively few validated pan-cancer, universally applicable cfDNA tumor biomarkers. Related technologies, especially in the detection of early-stage cancer patients, have low sensitivity and accuracy, and their predictive ability for tumor type is also very limited.

[0004] Therefore, it is necessary to improve cfDNA-based cancer diagnostic technology, provide a novel cancer diagnostic technology, enhance the sensitivity and accuracy of cancer diagnosis, and improve the ability to predict tumor types. Summary of the Invention

[0005] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application provides a cancer diagnostic model and application based on transposon fragmentation patterns in free DNA.

[0006] A first aspect of this application provides the use of transposon fragmentation patterns in cfDNA as markers in the preparation of products; said products are used for at least one of the following:

[0007] (1) Diagnosing cancer;

[0008] (2) Predicting cancer type.

[0009] According to the embodiments of this application, the markers provided in this application have at least the following beneficial effects:

[0010] The DNA degradation pattern of tumor cells differs from that of normal cells. Therefore, the fragmentation pattern of cfDNA in cancer patients differs from that in non-cancer patients. Transposons (TEs) are a class of mobile genetic factors found in metazoans. The total length of various types of transposons in the human genome accounts for half of the human genome. Since transposons are usually in a repressed state in normal cells, but often exhibit dysregulation in tumor cells, the fragmentation pattern of transposon DNA in tumors should be more significantly affected during degradation. This application investigated the fragmentation patterns of transposons in cfDNA and found significant differences between cancer patients and non-cancer controls. Using these fragmentation patterns as biomarkers for cancer diagnosis or risk assessment demonstrates superior diagnostic efficacy compared to whole-genome analysis and can be applied to risk assessment and diagnosis of various cancer types. Furthermore, using these fragmentation patterns as biomarkers for cancer diagnosis only requires common whole-genome sequencing technologies, with lower sequencing depth requirements and lower experimental and diagnostic costs. In addition, using these fragmentation patterns as biomarkers for cancer diagnosis also possesses certain tumor tissue tracing capabilities, thus having a wider range of applications.

[0011] According to some embodiments of this application, the cancer diagnosis is to determine whether the sample is a sample of a cancer patient, and the cancer type prediction is to determine whether the sample is a specific cancer type.

[0012] According to some embodiments of this application, the fragmentation pattern of the transposon in cfDNA includes the average coverage of the transposon in cfDNA.

[0013] According to some embodiments of this application, the average coverage is the total number of sequencing reads in the sample and the total length of this type of transposon in the human genome. The denominator is standardized with the total number of cfDNA reads that intersect with this type of transposon, and the formula is:

[0014] The standardization constant can be chosen arbitrarily as needed to prevent the average coverage value from being too small to be easily displayed. In addition, once the transposable is determined, its total length is fixed, so it can also be removed from the formula. The impact may be that different standardization constants need to be used for different transposables, so that multiple transposables can be displayed together more conveniently.

[0015] According to some embodiments of this application, the fragmentation pattern includes at least one of length distribution data in cfDNA, terminal base sequence usage pattern in cfDNA, terminal distribution consistency in cfDNA, average coverage in cfDNA, methylation level in cfDNA, or copy variation data in cfDNA.

[0016] According to some embodiments of this application, the fragmentation pattern includes at least two of the following: length distribution data in cfDNA, terminal base sequence usage pattern in cfDNA, terminal distribution consistency in cfDNA, average coverage in cfDNA, methylation level in cfDNA, or copy variation data in cfDNA.

[0017] A second aspect of this application provides an apparatus for diagnosing cancer, the apparatus comprising:

[0018] The acquisition module is used to acquire the terminal position information of cfDNA on the genome of the test sample and the reference sample;

[0019] The interception module is used to intercept cfDNA sequence readings that intersect with the transposon position in the sequencing data of the cfDNA, and obtain the fragmentation mode parameters of the transposon.

[0020] The comparison module is used to compare the fragmentation mode parameters of transposons in the test sample and the reference sample, and to determine the cancer information of the test sample based on the comparison results.

[0021] Since transposons are usually suppressed in normal cells but often dysregulated in tumor cells, the fragmentation pattern of transposon DNA in tumors during degradation should be more significantly affected. Based on this principle, this application uses a comparison module in a device for cancer diagnosis by focusing on the transposon fragmentation pattern in cfDNA. Theoretically, focusing on the transposon fragmentation pattern in cfDNA for cancer diagnosis is more effective than whole-genome analysis, which can improve the accuracy of existing biomarkers in diagnosing cancer, has better specificity and sensitivity, and can also realize pan-cancer diagnosis.

[0022] Furthermore, since there are many types of transposons in the human genome, which account for about half of the total length of the human genome, such as the common ALU, MIR, L2, etc., most of these transposons have a large number of copies in the genome, and multiple copies of the same transposon will be analyzed as one type. Therefore, the device for diagnosing cancer in this application does not have high requirements for the sequencing depth of cfDNA, which can significantly reduce experimental and diagnostic costs.

[0023] Furthermore, the device for diagnosing cancer in this application only requires the collection of peripheral blood to obtain cfDNA sequencing data. Therefore, it is almost non-invasive in the human body, allows for repeated sampling, and has no special requirements for the chemical reagents and sequencing platform used in the experiment. It has strong versatility and broad application prospects.

[0024] According to some embodiments of this application, the reference sample is a non-cancer patient.

[0025] According to some embodiments of this application, the terminal site information includes at least one of upstream terminal site information and downstream terminal site information.

[0026] According to some embodiments of this application, the cfDNA sequencing data is obtained using whole-genome paired-end sequencing technology or single-end sequencing technology.

[0027] According to some embodiments of this application, whole-genome paired-end sequencing or single-end sequencing technology can be used to obtain cfDNA ends. By comparing with a reference genome, the position of one end (for single-end sequencing or paired-end sequencing) or both ends (for paired-end sequencing data) of each cfDNA fragment on the genome can be extracted to obtain the cfDNA end site information. Paired-end sequencing or long-read single-strand sequencing is only required when it is necessary to obtain fragmentation pattern parameters related to length distribution, because short-read single-end sequencing is difficult to accurately obtain length distribution related information.

[0028] According to some embodiments of this application, the cfDNA sequencing data is peripheral blood cfDNA sequencing data.

[0029] According to some embodiments of this application, the transposer includes one transposer, a combination of two transposers, or a combination of multiple transposers.

[0030] According to some embodiments of this application, 16 transposons with a large copy number in the human genome are used, but other transposons with a smaller copy number can also be used; any number (e.g., 1, 5, 10, 50) and any combination of transposons can be used; in addition, for each transposon type, the scheme of this application can use all its copies in the human genome, or a portion of them, such as transposons falling on specific regions or chromosomes, and can further select transposons on specific regions or chromosomes according to the distribution preference of cfDNA terminal sites in the genome of cancer types.

[0031] According to some embodiments of this application, the transposon is a transposon with a copy number greater than 1000 in the genome.

[0032] According to some embodiments of this application, the transposon is a transposon with a copy number greater than 10,000 in the genome.

[0033] According to some embodiments of this application, the fragmentation pattern includes average coverage in cfDNA.

[0034] According to some embodiments of this application, the fragmentation pattern includes at least one of length distribution data in cfDNA, terminal base sequence usage pattern in cfDNA, terminal distribution consistency in cfDNA, average coverage in cfDNA, methylation level in cfDNA, or copy variation data in cfDNA.

[0035] In this application, the fragmentation pattern parameters of transposons in cfDNA can be obtained by extracting cfDNA sequencing reads that intersect with the transposon positions in the existing human reference genome. Analyzing these fragment-selected sequencing reads yields various fragmentation pattern parameters. For example, the cfDNA length distribution includes the proportion of short fragments (e.g., the proportion of cfDNA less than 150 bp, S150, or other length values), and the cfDNA terminal base sequence usage pattern includes the proportion of CCCA terminal sequences (CCCA value, or other sequences such as CCTG).

[0036] In addition, the difference in cfDNA fragmentation pattern parameters before and after fragment screening, i.e., the Δ value, can also be used as the fragmentation pattern parameter of transposons in cfDNA, such as the ΔS150 and ΔM-CCCA parameter values ​​described in the embodiments of this application.

[0037] According to some embodiments of this application, the fragmentation pattern includes at least two of the following: length distribution data in cfDNA, terminal base sequence usage pattern in cfDNA, terminal distribution consistency in cfDNA, average coverage in cfDNA, methylation level in cfDNA, or copy variation data in cfDNA.

[0038] According to some embodiments of this application, obtaining cancer information of the sample to be tested includes determining whether the sample to be tested is a cancer sample or predicting the cancer type of the sample to be tested.

[0039] According to some embodiments of this application, the specific method for determining whether a sample to be tested is a cancer sample is as follows: the fragmentation mode parameter of the transposon of the sample to be tested is compared with the truncation value, and the sample to be tested is determined as a cancer sample based on the comparison result.

[0040] According to some embodiments of this application, the specific method for predicting the cancer type of the test sample is as follows: comparing the fragmentation mode parameter of the transposon of the test sample with the truncation value, and determining whether the cancer of the test sample is a specific cancer type based on the comparison result.

[0041] According to some embodiments of this application, the cutoff value is determined using a method including receiver operating characteristic (ROC) curves.

[0042] The cut-off value is the threshold value for judging a positive or negative result in a test, in order to distinguish between normal and abnormal. The cut-off value in this application can be determined by any method well known in the art, such as using the receiver operating characteristic curve.

[0043] ROC curves are used to evaluate the accuracy, sensitivity, and specificity of cancer diagnosis at different cutoff values. AUC (Area Under the Curve) values ​​are used to assess the overall effectiveness of the diagnostic method: a higher AUC indicates better performance. The cutoff value is determined based on actual needs; for example, community screening may require specificity >99%, while high-risk groups may require sensitivity >95%. This determines the cutoff value needed to meet specific requirements.

[0044] According to some embodiments of this application, the fragmentation mode parameters include fragmentation mode parameters that integrate two or more fragmentation mode parameters using artificial intelligence technology.

[0045] According to some embodiments of this application, the artificial intelligence technology is machine learning.

[0046] This application proposes a comparison module in a cancer diagnostic device that integrates fragmentation pattern parameters of two or more transposons in cfDNA. Specifically, it uses machine learning models (such as neural networks, gradient boosting decision trees, etc.) to optimize and train the fragmentation pattern parameters, and uses the generated prediction results to determine diagnostic parameters.

[0047] According to some embodiments of this application, the specific method for integrating two or more fragmentation mode parameters includes the following steps:

[0048] S10. Calculate the fragmentation mode parameters of each type of transposon in each sample;

[0049] S20. Divide the sample into equal parts randomly, and take a portion as the training set each time, and the remaining portion as the test set.

[0050] S30. Using two or more fragmented mode parameters of transposons, perform parameter optimization and training on the training set using parameter optimization and training tools, and use the trained model to predict samples in the test set.

[0051] S40. Repeat step S30 for training and testing, so that each equally divided sample appears as a test set in one experiment, record the prediction results, and integrate the prediction results of the test set.

[0052] The embodiments of this application verify that the integration of fragmentation pattern parameters of multiple transposons using machine learning for cancer diagnosis is effective. ROC analysis shows that the AUC value is further improved compared to individual fragmentation pattern parameters, achieving progress in diagnostic sensitivity and specificity for multiple cancer types and different stages of cancer.

[0053] According to some embodiments of this application, when obtaining the cancer information of the test sample is for predicting the cancer type of the test sample, the specific method for integrating two or more fragmented mode parameters includes the following steps:

[0054] S100, Calculate the fragmentation mode parameters of each type of transposon in each sample;

[0055] S200. Divide the sample into equal parts randomly, and take a portion as the training set each time, and the remaining portion as the test set.

[0056] S300. Train the model separately for each cancer type: Use fragmented mode parameters of two or more transposons, take the samples corresponding to the target cancer type as positive samples and the rest as negative samples on the training set, use parameter optimization and training tools to optimize parameters and train, and use the trained model to predict samples on the test set.

[0057] S400. Repeat step S300 for training and testing, so that each equally divided sample appears as a test set in one experiment, record the prediction results, obtain the prediction results of each sample in the model corresponding to different cancer types, and take the maximum value to predict the tumor type.

[0058] Since there are many types of transposons in the human genome, and the activation of different transposons varies in different cancers, the embodiments of this application have found that the average coverage of different transposons varies significantly, and there are also differences between cancer patients and non-cancer patients. Based on this principle, the fragmentation pattern parameters of transposons in cfDNA are used to predict cancer type. The device for diagnosing cancer in this application can be used for tumor tissue tracing, thereby expanding the scope of application.

[0059] According to some embodiments of this application, the cancer includes at least one of bile duct cancer, breast cancer, colorectal cancer, gastric cancer, pancreatic cancer, lung cancer, ovarian cancer, esophageal cancer, liver cancer, brain cancer, head and neck cancer, cervical cancer, prostate cancer, thyroid cancer, lymphoma, skin cancer, leukemia, small intestine cancer, bladder cancer, testicular cancer, ovarian cancer, uterine cancer, or oropharyngeal cancer.

[0060] According to some embodiments of this application, the cancer includes at least one of bile duct cancer, breast cancer, colorectal cancer, gastric cancer, pancreatic cancer, lung cancer, ovarian cancer, esophageal cancer, or liver cancer.

[0061] A third aspect of the present invention provides a method for constructing a cancer diagnostic model, comprising the following steps:

[0062] Sequencing data of cfDNA from cancer patients and non-cancer individuals were obtained and compared with a reference genome to obtain the terminal site information of the cfDNA.

[0063] Based on the terminal site information of the cfDNA, the cfDNA sequence readings that intersect with the transposon position are extracted to obtain the fragmentation mode parameters of the transposon;

[0064] The cancer diagnostic model is constructed based on the fragmentation pattern parameters of the transposon.

[0065] According to some embodiments of this application, the method for constructing the cancer diagnostic model based on the fragmentation pattern parameters of the transposon includes:

[0066] By comparing the transposon fragmentation patterns of cancer patients and non-cancer individuals, a transposon fragmentation parameter model representing cancer patients and non-cancer individuals is constructed, which is the cancer diagnostic model.

[0067] According to some embodiments of this application, the method for constructing the cancer diagnostic model based on the fragmentation pattern parameters of the transposon includes:

[0068] According to some embodiments of this application, the method for constructing the cancer diagnostic model based on the fragmentation pattern parameters of the transposon includes:

[0069] By comparing the transposon fragmentation patterns of cancer patients and non-cancer individuals with different cancer types, a transposon fragmentation parameter model representing different cancer types and non-cancer individuals is constructed, which is the cancer diagnostic model.

[0070] A fourth aspect of this application provides a cancer diagnostic model, which is constructed using the construction method described in the third aspect of this application.

[0071] A fifth aspect of this application provides a method for classifying samples, comprising the following steps:

[0072] S1. Obtain the terminal position information of the cfDNA on the genome of the sample, extract the cfDNA sequence reading that intersects with the transposon position, and obtain the fragmentation mode parameters of the transposon.

[0073] S2. Compare the fragmentation mode parameters of transposons in the sample and the reference sample;

[0074] S3. Determine the classification of the sample based on the comparison results.

[0075] A sixth aspect of this application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the construction method described in the third aspect of this application or the classification method described in the fifth aspect.

[0076] A seventh aspect of this application provides an electronic device, including a processor and a memory;

[0077] The memory stores a computer program that can run on the processor, and the processor, when running the computer program, implements the construction method described in the third aspect of this application or the classification method described in the fifth aspect.

[0078] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing this application. Attached Figure Description

[0079] The present application will be further described below with reference to the accompanying drawings and embodiments, wherein:

[0080] Figure 1 This is a schematic diagram illustrating the overall process of using transposon fragmentation patterns in cfDNA as cancer diagnostic biomarkers in an embodiment of this application.

[0081] Figure 2 The calculation results are for the E-index values ​​of various transposons and whole genomes of different types of samples in Dataset 1 of Embodiment 2 of this application;

[0082] Figure 3 ROC analysis curves for cancer diagnosis using whole-genome E-index values ​​and E-index values ​​of different types of transposons as an example in Example 2 of this application;

[0083] Figure 4 This refers to the average coverage calculation results of various transposons for different types of samples in Dataset 1 of Embodiment 2 of this application;

[0084] Figure 5 The ROC analysis curve for cancer diagnosis using the average coverage of transposons in Embodiment 2 of this application;

[0085] Figure 6 ROC analysis curve for cancer diagnosis using the cancer diagnosis machine learning model of Embodiment 2 of this application;

[0086] Figure 7 The results show the calculated methylation levels of various transposons and the whole genome for different types of samples in Dataset 2 of Example 3 of this application;

[0087] Figure 8 ROC analysis curves for cancer diagnosis using whole-genome methylation levels and methylation levels of different types of transposons as an example in Example 3 of this application;

[0088] Figure 9 ROC analysis curve for cancer diagnosis using the cancer diagnosis machine learning model of Embodiment 3 of this application.

[0089] Figure 10 This refers to the calculated average coverage of various transposons in the clinical data of Example 4 of this application;

[0090] Figure 11 The ROC analysis curve for cancer diagnosis using the average coverage of different types of transposons in Example 4 of this application;

[0091] Figure 12 The results of E-index value calculation for various transposons and the whole genome in the clinical data of Example 4 of this application;

[0092] Figure 13 The ROC analysis curves for cancer diagnosis using the E-index values ​​of the whole genome and the E-index values ​​of different types of transposons in Example 4 of this application are shown. Detailed Implementation

[0093] The following will clearly and completely describe the concept and technical effects of this application in conjunction with embodiments, so as to fully understand the purpose, features and effects of this application. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are all within the scope of protection of this application.

[0094] Unless otherwise specified in the examples, the procedures should be performed under standard conditions or conditions recommended by the manufacturer. Reagents or instruments whose manufacturers are not specified are all commercially available products.

[0095] The term "sequence readout" refers to a string of nucleotides sequenced from any part or all of a nucleic acid molecule. For example, a sequence readout can be a short nucleotide string (e.g., 20-150 nucleotides) sequenced from a nucleic acid fragment, a short nucleotide string at one or both ends of a nucleic acid fragment, or sequencing of an entire nucleic acid fragment present in a biological sample. Sequence readouts can be obtained in various ways, such as using sequencing technologies or using probes, such as hybridization arrays or capture probes, or amplification technologies, such as polymerase chain reaction (PCR) or linear or isothermal amplification using single primers. As part of a biological sample analysis, at least 1,000 sequence readouts can be analyzed. As other examples, at least 10,000, or 50,000, or 100,000, or 500,000, or 1,000,000, or 5,000,000 sequence readouts can be analyzed. Sequence readouts may include "terminal sequences" associated with the ends of a fragment. Terminal sequences may correspond to the outermost N bases of the fragment, for example, the last 2-30 bases of the fragment. If a sequence read corresponds to an entire fragment, that read may contain two end sequences. When paired-end sequencing provides two sequence reads corresponding to the ends of a fragment, each read may contain one end sequence.

[0096] The term "cutoff value" refers to a predetermined quantity used in operation. For example, a cutoff size can refer to the size beyond which no fragment is included. A cutoff value can be a "reference value" or can be derived from reference values ​​representing a particular category or distinguishing two or more categories. As those skilled in the art will understand, such reference values ​​can be determined in various ways. For example, a metric can be determined for two distinct subjects with different known classifications, and the reference value can be chosen as a representative of one classification (e.g., the mean) or a value between two clusters of the metric (e.g., chosen to obtain desired sensitivity and specificity). As another example, a reference value can be determined based on statistical simulation of a sample.

[0097] The term "about" may mean within an acceptable range of deviation from a specific value as determined by a person skilled in the art, which will depend in part on how the value is measured or determined, i.e., the limitations of the measurement system. For example, according to practice in the art, "about" may mean within one or more standard deviations. Alternatively, "about" may mean a range of up to 20%, up to 10%, up to 5%, or up to 1% of a given value. Alternatively, particularly for biological systems or processes, the term "about" or "approximately" may mean within orders of magnitude of the value, within five times the value, and more preferably, within two times the value. When a particular value is described in this application and claims, unless otherwise indicated, it should be assumed that the term "about" means that the particular value is within an acceptable range of error. The term "about" may have the meaning commonly understood by a person skilled in the art. The term "about" may mean ±10%. The term "about" may mean ±5%.

[0098] The embodiments of this application are described in detail below. The described embodiments are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0099] In the description of this application, "several" means one or more, "multiple" means two or more, and "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" is for distinguishing technical features only and should not be construed as indicating or implying relative importance, the number of indicated technical features, or the order of the indicated technical features. If a logical order is described in the flowchart, but in some cases the described or shown steps may be performed in a different order than that shown in the flowchart, this is not mandatory.

[0100] In the description of this application, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0101] Example 1

[0102] Figure 1 This is a schematic diagram illustrating the overall process of using transposon fragmentation patterns in cfDNA as cancer diagnostic biomarkers for cancer diagnosis. (Example) Figure 1 As shown:

[0103] 1. Collection and processing of peripheral blood from subjects:

[0104] Peripheral blood (6 mL) was collected from a vein using a blood collection tube containing an anticoagulant (EDTA). The tube was centrifuged at 4°C and 1600g for 15 min. The supernatant was then aspirated and transferred to a new centrifuge tube. The tube was then centrifuged again at 4°C and 16000g for 15 min. The supernatant (i.e., plasma) was then aspirated and transferred to a cryovial. The tube was stored at -80°C until use.

[0105] 2. Extraction, sequencing, and basic analysis of cfDNA:

[0106] cfDNA was extracted from plasma using commercially available kits and equipment, a sequencing library was constructed, and high-throughput sequencing (paired-end sequencing, single-end sequencing can also be used) was performed. The raw sequencing data was processed by removing sequencing adapters, low-quality cycles, and repetitive sequences, and then aligned to the whole human genome. Repetitive sequences and low-quality alignment results were removed to obtain the final alignment results (considering the influence of sex on the genome, data from the M, X, and Y chromosomes can be removed). Based on these alignment results, the positions of both ends of each cfDNA fragment on the genome were extracted, which are the cfDNA fragment end sites.

[0107] 3. Obtain the location information of transposons in the human genome:

[0108] Since the sequences of transposons are known, they can be identified in the human genome using algorithmic software (such as RepeatMasker or RepeatModeler). In this embodiment, the location information of transposons in the NCBI GRCh38 human reference genome identified by RepeatMasker software was downloaded from the UCSC genome browser. Because different transposons have different copy numbers in the human genome, this embodiment extracted all transposons with copy numbers greater than 10,000, totaling 16 families, as shown in Table 1.

[0109] Table 1 Information on transposons with a copy number of 10,000 or more

[0110] Class Family Copy number Total length (bp) Percentage of the human genome (%) LTR ERVK 11,903 8,942,031 0.30 LINE RTE-X 15,944 3,490,643 0.12 DNA TcMar-Mariner 16,871 2,793,571 0.10 LTR Gypsy 17,427 3,580,535 0.12 DNA hAT-Blackjack 20,062 3,354,287 0.11 DNA hAT-Tip100 47,824 8,471,679 0.29 LINE CR1 69,603 11,981,017 0.41 DNA TcMar-Tigger 123,237 36,232,217 1.23 LTR ERVL 171,950 56，571,848 1.93 LTR ERV1 187，201 84,389,053 2.87 DNA hAT-Charlie 270,942 45,888,115 1.56 LTR ERVL-MaLR 367,470 110,990,630 3.78 LINE L2 486,431 103,563,083 3.53 SINE MIR 616,589 84,408,192 2.87 LINE L1 1,031,524 521,402,241 17.75 SINE Alu 1,282,821 307,193,764 10.46

[0111] 4. Obtain the fragmentation pattern of transposons in cfDNA:

[0112] For each transposon in a sample (such as one of the 16 types of transposons in Table 1), the sequence readings of cfDNA that intersect with its position are extracted from the sequencing data. These sequence readings can then be analyzed to obtain various fragmentation mode parameters of cfDNA fragments of various transposons. The fragmentation mode parameters can be selected from parameters such as length distribution (e.g., the proportion of readings below 150bp commonly used), end sequence distribution and complexity, end distribution consistency (E-index), N-index, ΔS150, or ΔM-CCCA. The calculation method of the E-index parameter refers to the Chinese invention patent with publication number CN115019952A (the principle and calculation method are described in An et al. DNA methylation analysis explores the theolecular basis of plasma cell-free DNA fragmentation. Nat Commun. 2023 Jan 18; 14(1): 28). The calculation methods of the N-index, ΔS150, and ΔM-CCCA parameters refer to the Chinese invention patent with application number 202410378081.2 (the principle and calculation method are described in Ju). et al. Cell-free DNAend characteristics enable accurate and sensitiive cancer diagnosis. Cell Rep Methods 2024 Oct 21;4(10):100877). If the sequencing data is methylation sequencing data (such as BS-seq, EM-seq or TAPS-seq, etc.), the corresponding methylation level can also be analyzed at the same time.

[0113] In addition, for all types of transposons, a new transposon fragmentation mode parameter can be selected—the average coverage of transposons in the sample. In this embodiment, the total number of sequencing reads in the sample and the total length of this type of transposon in the human genome are used as the denominator to normalize the total number of cfDNA reads that intersect with this type of transposon. The formula is as follows: The standardization constant set in the formula is to prevent the average coverage value from being too small to be easily displayed. It can be chosen arbitrarily as needed, but is typically set to 1,000,000. Since the total length of each type of transposon is determined after selecting a reference genome (e.g., hg38), this value can be ignored during standardization. Furthermore, there are multiple ways to define "cfDNA readings falling on this type of transposon." For example, it could be considered a fall if one base of cfDNA overlaps with the transposon, or it could be considered a fall if a certain proportion (e.g., 50%) or even if all cfDNA bases are present in the transposon. However, a consistent definition should be used when comparing different samples.

[0114] 5. Constructing cancer diagnostic models:

[0115] cfDNA is produced during chromatin degradation when cells die. This process is regulated by many factors (such as DNA methylation) and involves various nucleases (such as DNASE1L3). Because the DNA degradation pattern in tumor cells differs from that in normal cells, the cfDNA fragmentation pattern in cancer patients differs from that in non-cancer patients. Furthermore, since most transposons are in a repressed state in normal cells but often exhibit dysregulation in tumors, the fragmentation pattern of transposon DNA during degradation in tumors is theoretically more significantly affected, resulting in a greater difference from the fragmentation pattern of transposons in cfDNA in non-cancer patients.

[0116] Therefore, by comparing and analyzing the fragmentation patterns of different transposons in the cfDNA sequencing data of samples with those of the control group (non-cancer patients) (and further analyzing methylation levels in methylation sequencing data), a diagnostic model can be developed and corresponding parameters determined. Since the fragmentation patterns of transposons differ between cancer patient and non-cancer patient samples, a cutoff value can be used for differentiation, enabling cancer diagnosis. Furthermore, artificial intelligence technology can be combined to integrate the fragmentation pattern parameters of all or some transposons, developing a diagnostic model with higher accuracy.

[0117] In addition, the activation of different transposon types varies in different cancers. Therefore, analyzing the cfDNA fragmentation patterns of multiple transposons can be used to predict tumor type.

[0118] Receiver operating characteristic (ROC) curves can be used to evaluate the accuracy, sensitivity, and specificity of cancer diagnosis using different cutoff values ​​as diagnostic parameters. The cutoff value can be determined based on specific needs; for example, community screening might require specificity >99%, while high-risk groups might require sensitivity >95%. The AUC (Area Under the Curve) value is then used to assess diagnostic effectiveness: a higher AUC value indicates better results.

[0119] The following Examples 2 and 3 use Dataset 1 and Dataset 2 to validate the cancer diagnosis model of this example. The two datasets have different sources and different experimental schemes, but both contain at least one type of cancer patient and non-cancer patient (control group).

[0120] Example 2

[0121] This embodiment uses dataset 1 to validate a cancer diagnostic model based on transposon fragmentation patterns in cfDNA:

[0122] Dataset 1 contains cfDNA sequencing data from 231 non-cancer patients and 208 cancer patients (cholangiocarcinoma, breast cancer, colorectal cancer, gastric cancer, pancreatic cancer, lung cancer, and ovarian cancer) (data source: Cristiano et al. Genome-wide cell-free DNA fragmentation in patients with cancer. Nature 2019; 570:(385-389)).

[0123] The E-index values ​​of these samples were calculated according to the method provided in Example 1 (the principle and calculation method are described in Chinese Invention Patent Publication No. CN115019952A), and the results are as follows. Figure 2 As shown, the diagnosis was performed using overall E-index values ​​and E-index values ​​of different transposon types. Colorectal cancer was used as an example, and its ROC curve was presented as follows: Figure 3 As shown.

[0124] Depend on Figure 2 and Figure 3 The results showed that, compared with the whole genome, the E-index of the 16 transposons differed more significantly between cancer patients and non-cancer patients, leading to more accurate diagnosis. For example, the AUC value of cancer diagnosis using the whole genome E-index was 0.718, while the AUC value of cancer diagnosis using the E-index values ​​of different types of transposons ranged from 0.79 to 0.88 (all P values ​​were less than 0.0001), which was superior to the whole genome E-index.

[0125] The sequencing depth (i.e., average transposon coverage, the average sequencing reads falling within each type of transposon, normalized using the total sequencing reads for each sample) in different samples was calculated using the method in Example 1. The results are as follows: Figure 4 As shown, significant differences in sequencing depth were found between different transposons within the same sample, and these differences also existed between cancer patient samples and non-cancer patient samples. For example, cancer patients showed higher sequencing depths in SINE / MIR type transposons, while patients with most cancer types showed lower sequencing depths in LINE / L1 type transposons. ROC analysis is as follows: Figure 5 As shown, the average coverage of multiple transposons was found to be useful for cancer diagnosis.

[0126] Furthermore, multiple fragmented mode parameters of transposons are combined with machine learning (such as neural networks, gradient boosting trees, etc.) to construct a machine learning model for cancer diagnosis. The specific method (refer to Cristiano et al. Nature 2019; 570:(385-389)) is as follows:

[0127] 1) For all samples in the dataset, calculate the various fragmentation mode parameters for each type of transposon, including E-index, N-index, coverage, ΔS150, ΔM-CCCA, and other fragmentation mode parameters;

[0128] 2) Randomly divide the sample into 10 parts, each part containing approximately 10% of the total sample. Each time, take 9 parts as the training set and the remaining part as the test set.

[0129] 3) Using one or more of the cell-free DNA fragmentation parameters in the above transposons, perform parameter optimization and training on the training set using the caret and GBM packages in R software, and use the trained model to predict samples in the test set.

[0130] 4) Repeat the above training and testing 10 times. Then any one of the 10 equal parts will appear as the test set in one experiment, and its prediction result will be recorded.

[0131] 5) Integrate the prediction results of all the above samples as the test set, and use ROC analysis to examine the performance of this parameter (or parameter group) in cancer diagnosis.

[0132] The results are as follows Figure 6 As shown, by Figure 6 The results show that the AUC values ​​for diagnosing cancer using only the E-index or coverage parameter were 0.876 and 0.924, respectively. However, by further combining parameters such as N-index, ΔS150, and ΔM-CCCA, the AUC value could be increased to 0.957 (all P values ​​less than 0.01), achieving excellent diagnostic results. Further analysis using the cancer diagnosis machine learning model to diagnose seven cancer types yielded AUC values ​​between 0.922 and 0.989. The AUC values ​​for diagnosing stage I, II, III, and IV cancers were 0.924, 0.960, 0.973, and 0.975, respectively (all P values ​​less than 0.01) (Stage X in the figure represents samples with unknown tumor stages). Notably, at 95% specificity, the sensitivity among various cancer types and stages ranged from 75% to 100%, as shown in Table 2.

[0133] Table 2. Sensitivity of the diagnostic model at 95% specificity across various cancer types and stages.

[0134]

[0135] Furthermore, by combining the various parameters from the transposons mentioned above with machine learning, a cancer type prediction model can be constructed. The specific method is as follows:

[0136] 1) Extract all cancer samples from the dataset and calculate various parameters for each type of transposon, including E-index, N-index, coverage, ΔS150, ΔM-CCCA, etc.

[0137] 2) Randomly divide all samples into 10 parts, each part containing approximately 10% of the total samples. Each time, take 9 parts as the training set and the remaining part as the test set.

[0138] 3) Train a separate model for each cancer type: Using one or more of the parameters mentioned above, use the samples corresponding to the target cancer type as positive samples and the rest as negative samples in the training set. Use the caret package in R software to call the GBM package for parameter optimization and training. Use the trained model to predict the samples in the test set. Repeat the above training and testing 10 times. Then, any one of the 10 equal parts will appear as the test set in one experiment, and record its prediction result for that cancer type.

[0139] 4) Complete the above training and prediction for all cancer types. At this point, for each sample, you can obtain its prediction results in the model corresponding to different cancer types, and take the maximum value as the predicted cancer type;

[0140] 5) Compare and statistically analyze the predicted cancer type of all samples with their actual types to determine the accuracy of the tumor tracing method in the embodiments.

[0141] The results are shown in Table 3. The overall prediction accuracy of the cancer type prediction model for the seven types of cancer reached 52.4%. Based on the results in Table 3, it can be seen that the cancer type prediction model has a certain ability to trace the origin of tumor tissue.

[0142] Table 3 Results of Tumor Type Prediction

[0143]

[0144] Example 3

[0145] This embodiment uses dataset 2 to validate a cancer diagnostic model based on transposon fragmentation patterns in cfDNA:

[0146] Dataset 2 contains cfDNA sequencing data from 497 non-cancer patients and 780 cancer patients (breast cancer, colorectal cancer, esophageal cancer, liver cancer, lung cancer, pancreatic cancer, and gastric cancer) (data source: Bie et al. Multimodal analysis of cell-free DNA whole-methylome sequencing for cancer detection and localization. Nature Communications 2023; 14:6042.). All samples in this dataset are methylated data. Analysis revealed that DNA methylation levels on transposons were significantly decreased in cancer patients. Figure 7 As shown.

[0147] Further analysis revealed that using DNA methylation levels in transposons for cancer diagnosis outperformed using whole-genome DNA methylation levels, such as... Figure 8 As shown, the ROC curve results of different transposon and whole-genome DNA methylation levels in the diagnosis of gastric cancer show that the AUC value of diagnosis using whole-genome DNA methylation level is 0.71, while the AUC value of diagnosis using 16 types of transposon DNA methylation level is between 0.737 and 0.783 (P values ​​are all less than 0.05), achieving better diagnostic results.

[0148] Next, the fragmentation patterns and DNA methylation levels of each type of transposon were calculated, and a cancer diagnosis model was constructed using machine learning (the method is the same as in Example 2, with all parameter features including the DNA methylation levels of 16 transposons). The results for cancer diagnosis are as follows: Figure 9 As shown in the results, the AUC values ​​when using only E-index, DNA methylation, or coverage parameters were 0.812, 0.861, and 0.922, respectively (all P values ​​less than 0.05). However, the AUC value when combining parameters such as E-index, coverage, N-index, ΔS150, and ΔM-CCCA could be increased to 0.95. Further analysis of seven cancer types showed AUC values ​​between 0.908 and 0.98. The AUC values ​​for diagnosing stage I, II, III, and IV cancers were 0.898, 0.946, 0.947, and 0.978, respectively (all P values ​​less than 0.05). This indicates that fragmentation patterns in transposons can also be applied to cancer diagnosis using DNA methylation data, and integrating fragmentation patterns with DNA methylation levels can achieve even better diagnostic results.

[0149] Example 4

[0150] This embodiment uses a set of clinical data to validate a cancer diagnostic model based on transposon fragmentation patterns in cfDNA:

[0151] The clinical data included cfDNA sequencing of 56 patients with hepatocellular carcinoma and 24 controls (all of whom were hepatitis B carriers or had non-tumor liver disease). The data came from the applicant’s previous work (Ju et al. Cell-free DNA end characteristics enable accurate and sensitive cancer diagnosis. Cell Reports Methods 2024; 4(10):100877).

[0152] The average coverage of 16 transposon classes in these samples was calculated according to the methods in Examples 1 and 2, and the results are as follows: Figure 10 As shown, among the 16 transposons, 12 showed significantly upregulated mean coverage in liver cancer. The mean coverage of different transposon types was used for diagnosis, and their ROC curves are presented as follows. Figure 11 As shown, it was found that it had a certain ability to distinguish between liver cancer patients and control groups (AUC values ​​were between 0.66 and 0.91, and P values ​​were all less than 0.05).

[0153] Similarly, the E-index of the 16 transposon classes in the sample is calculated, such as Figure 12 As shown, the global E-index was decreased in liver cancer patients, but this did not reach statistical significance. The E-index decreased in all transposons, and was significantly lower in liver cancer patients than in the control group, indicating a greater decrease in transposons. Using the E-index values ​​of transposons can significantly improve diagnostic accuracy, as shown by the ROC curve. Figure 13 As shown, its AUC value increased from 0.58 at the global level to 0.60–0.69 (all P values ​​were less than 0.05).

[0154] Example 5

[0155] This embodiment provides a device for diagnosing cancer, the device comprising:

[0156] The acquisition module obtains the end-site information of cfDNA on the genome from the test sample and non-cancer patient samples.

[0157] The extraction module extracts cfDNA sequence readings that intersect with transposon locations from the cfDNA sequencing data to obtain the fragmentation mode parameters of the transposon.

[0158] The comparison module compares the fragmentation pattern parameters of transposons in the test sample and non-cancer patient samples, and obtains cancer information of the test sample based on the comparison results, such as whether it is a cancer sample, cancer stage, and cancer type.

[0159] In summary, this application proposes a novel cancer diagnostic biomarker and related methods and apparatus based on the fragmentation patterns of peripheral blood cfDNA in transposons. The biomarker proposed in this application can improve the diagnostic accuracy of existing biomarkers. The cancer diagnostic method of this application can utilize common whole-genome sequencing for obtaining cfDNA sequencing data, which is simple to operate and has low requirements for initial materials. Furthermore, since transposons in the human genome have many copies, multiple copies of the same transposon are analyzed as a single type, thus reducing the requirement for sequencing depth and lowering experimental and diagnostic costs. In addition, this technology has the following characteristics:

[0160] (1) Only free DNA sequencing data is required, and DNA methylation sequencing data can be used in combination with DNA methylation to improve accuracy;

[0161] (2) Diagnostic models and devices can be used in a variety of cancer types (i.e., pan-cancer diagnosis can be achieved);

[0162] (3) The diagnostic model and device have no specific requirements for the chemical reagents and sequencing platform used in the experiment (high versatility);

[0163] (4) Combining artificial intelligence technology can further improve the accuracy of cancer diagnosis of biomarkers and obtain high-accuracy cancer diagnosis models and tumor origin prediction models.

[0164] The embodiments of this application have been described in detail above. However, this application is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of this application. Furthermore, unless otherwise specified, the embodiments and features described in the embodiments of this application can be combined with each other.

Claims

1. The application of transposon fragmentation patterns in cfDNA as markers in product preparation; the product is used for at least one of the following purposes: (1) Diagnosing cancer; (2) Predicting cancer type.

2. The application according to claim 1, characterized in that, The fragmentation pattern of transposons in cfDNA includes the average coverage of transposons in cfDNA; Preferably, the formula for calculating the average coverage of the transposon in cfDNA is:

3. A device for diagnosing cancer, characterized in that, The device includes: The acquisition module is used to acquire the terminal position information of cfDNA on the genome of the test sample and the reference sample; The interception module is used to intercept cfDNA sequence readings that intersect with the transposon position in the sequencing data of the cfDNA, and obtain the fragmentation mode parameters of the transposon. The comparison module is used to compare the fragmentation mode parameters of transposons in the test sample and the reference sample, and to obtain the cancer information of the test sample based on the comparison results.

4. The device for diagnosing cancer according to claim 3, characterized in that, The reference sample consisted of non-cancer patients; Preferably, the transposer includes one type of transposer or a combination of two or more transposers; Preferably, the transposon is a transposon with a copy number greater than 10,000 in the genome.

5. The device for diagnosing cancer according to claim 3, characterized in that, The fragmentation pattern includes at least one of the following: length distribution data in cfDNA, terminal base sequence usage pattern in cfDNA, terminal distribution consistency in cfDNA, average coverage in cfDNA, methylation level in cfDNA, or copy variation data in cfDNA. Preferably, the fragmentation pattern includes at least two of the following: length distribution data in cfDNA, terminal base sequence usage pattern in cfDNA, terminal distribution consistency in cfDNA, average coverage in cfDNA, methylation level in cfDNA, or copy variation data in cfDNA.

6. The device for diagnosing cancer according to claim 3, characterized in that, The process of obtaining cancer information of the sample to be tested includes determining whether the sample to be tested is a cancer sample or predicting the type of cancer in the sample to be tested. Preferably, the cancer includes at least one of the following: bile duct cancer, breast cancer, colorectal cancer, gastric cancer, pancreatic cancer, lung cancer, ovarian cancer, esophageal cancer, liver cancer, brain cancer, head and neck cancer, cervical cancer, prostate cancer, thyroid cancer, lymphoma, skin cancer, leukemia, small intestine cancer, bladder cancer, testicular cancer, ovarian cancer, uterine cancer, or oropharyngeal cancer. Preferably, the cancer includes at least one of bile duct cancer, breast cancer, colorectal cancer, gastric cancer, pancreatic cancer, lung cancer, ovarian cancer, esophageal cancer, or liver cancer.

7. A method for constructing a cancer diagnostic model, characterized in that, Includes the following steps: Sequencing data of cfDNA from cancer patients and non-cancer individuals were obtained and compared with a reference genome to obtain the terminal site information of the cfDNA. Based on the terminal site information of the cfDNA, the cfDNA sequence readings that intersect with the transposon position are extracted to obtain the fragmentation mode parameters of the transposon; The cancer diagnostic model is constructed based on the fragmentation pattern parameters of the transposon.

8. The method for constructing a cancer diagnostic model according to claim 7, characterized in that, The method for constructing the cancer diagnostic model based on the fragmentation pattern parameters of the transposon includes: By comparing the fragmentation patterns of transposons in cancer patients and non-cancer individuals, a fragmentation parameter model representing transposons in cancer patients and non-cancer individuals is constructed, which is the cancer diagnostic model.

9. A cancer diagnostic model, characterized in that, The cancer diagnostic model is constructed using the construction method described in claim 7 or 8.

10. A method for classifying samples, characterized in that, Includes the following steps: S1. Obtain the terminal position information of the cfDNA on the genome of the sample, extract the cfDNA sequence reading that intersects with the transposon position, and obtain the fragmentation mode parameters of the transposon. S2. Compare the fragmentation mode parameters of transposons in the sample and the reference sample; S3. Determine the classification of the sample based on the comparison results.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the construction method of claim 7 or 8, or the classification method of claim 10.

12. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program that can run on the processor, the processor implementing the construction method of claim 7 or 8, or the classification method of claim 12, when running the computer program.

Images