Nasopharynx cancer detection model and construction method thereof

Abstract

The invention discloses a nasopharyngeal carcinoma detection model and a construction method thereof, and relates to the technical field of nasopharyngeal carcinoma detection.The four indexes of promoter activity, DNA fragment fragmentation degree, terminal base combination and EB virus content which can be applied to nasopharyngeal carcinoma detection are excavated by conducting whole genome sequencing on plasma free DNA, and the nasopharyngeal carcinoma detection model is obtained. A classification model of three indexes (promoter activity, DNA fragment fragmentation degree and terminal base combination) is constructed to obtain a prediction score of each index for a sample, and then a logic regression model is used to integrate the scores and the EB virus content to obtain a final classification prediction model and an integrated score. According to the nasopharyngeal carcinoma detection model, the nasopharyngeal carcinoma detection efficiency and accuracy are remarkably improved, the amount of data needed for analysis is small, the detection cost is low, and the nasopharyngeal carcinoma detection model is suitable for analysis and prediction of tumors in all periods and nasopharyngeal carcinoma prognosis evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of nasopharyngeal carcinoma detection, and in particular to a nasopharyngeal carcinoma detection model and a construction method thereof. Background Art

[0002] Nasopharyngeal carcinoma (NPC) is a malignant tumor that affects the nasopharynx and shows a significantly high incidence in East and Southeast Asia. It is highly associated with environmental factors and traditional dietary habits, genetic factors, and Epstein-Barr virus (EBV) infection. In 2018, there were approximately 60,558 newly diagnosed cases of NPC in China, accounting for 47.7% of the total number of cases worldwide. Approximately 75% of patients are already in the advanced stage at the time of initial diagnosis, which leads to recurrence, metastasis, and poor prognosis. In contrast, less than 10% of NPC patients are diagnosed at an early stage, and patients with early NPC can receive curative treatment and have a better prognosis, as evidenced by the 90% or higher 5-year overall survival rate of patients with early NPC. Early symptoms of NPC may include epistaxis, bloody nasal discharge, cervical lymphadenopathy, tinnitus, hearing loss, nasal congestion, and headache. These symptoms may not be obvious in the early stages and are easily overlooked or misdiagnosed as other diseases. Although pathological evidence remains the preferred diagnostic method for NPC according to established guidelines, its invasiveness and high cost limit its application in asymptomatic high-risk individuals, thus resulting in a low detection rate of early NPC.

[0003] The commonly used clinical detection methods for NPC include the following: Epstein-Barr virus serological test: by detecting Epstein-Barr virus-specific antibodies, such as VCA-IgA and EA-IgA, as serological markers for NPC screening and diagnosis; nasopharyngeal endoscopy: using a nasopharyngeal endoscope to directly observe the condition of the nasopharyngeal mucosa, and biopsy the abnormal mucosa to clarify the nature; imaging examination: including CT and MRI, used to observe the thickening of the nasopharyngeal mucosa, the size, range and infiltration depth of the tumor, and whether there is lymph node enlargement or distant metastasis; pathological biopsy: by obtaining tissue samples of the nasopharynx or cervical lymph nodes for pathological examination to confirm NPC; molecular biological detection: using PCR and other technologies to detect Epstein-Barr virus DNA and the expression of Epstein-Barr virus-encoded small RNA (EBER) to assist in pathological diagnosis.

[0004] However, existing detection methods have certain limitations: Although EB virus serological detection has high sensitivity, it has low specificity and may lead to misdiagnosis. For early NPC, EBV reactivation, i.e. early replication, has low cell turnover, which may lead to false negative results in plasma EBV DNA detection. In addition, different laboratories have different detection methods, sample processing and threshold settings, which may affect the detection rate of EBV DNA and the consistency of results; nasopharyngeal endoscopy has high limitations. The nasopharynx is hidden, which may cause discomfort to patients and has high technical requirements for operators; the popularity of imaging examinations is limited. Imaging examination equipment such as CT and MRI is not popular enough in some areas or medical institutions, which limits its application in NPC screening; pathological biopsy itself has certain risks, such as bleeding and infection, and it takes a certain amount of time to obtain results. The detection of EB virus DNA using PCR and other technologies, as well as the expression of EB virus-encoded small RNA (EBER), requires reference materials for standardization, and the lack of quantitative accuracy can be attributed to factors such as DNA separation, amplification bias or cross-contamination, as well as the relatively low accuracy of current diagnostic methods. It is urgent to explore non-invasive detection strategies for NPC patients.

[0005] With the development of genome sequencing technology, liquid biopsy technology has gradually matured. Liquid biopsy is non-invasive and can be repeated at different stages of disease treatment. It is not affected by geography and popularity, which has shifted the focus of precision diagnosis and treatment of tumors to liquid biopsy. Among the analytes of liquid biopsy, cfDNA (circulating cell-free DNA, cfDNA) is the most widely used. cfDNA is a DNA fragment from cells that is free in the blood circulation system. It mainly comes from fragmented DNA in the process of cell apoptosis, DNA fragments of necrotic cells, and exosomes secreted by cells. Using liquid biopsy technology to detect circulating tumor DNA (ctDNA) and Epstein-Barr virus DNA content, and using machine learning algorithms to develop new biomarkers have become a new direction for nasopharyngeal carcinoma detection.

[0006] To sum up the above situation, each current research direction has its own advantages and disadvantages for nasopharyngeal carcinoma detection. Overall, there is no particularly ideal result that can meet people's expectations.

[0007] In view of this, the present invention is proposed. Summary of the invention

[0008] The purpose of the present invention is to provide a nasopharyngeal carcinoma detection model and a construction method thereof.

[0009] To achieve the above object, the present invention adopts the following technical solution:

[0010] In the first aspect, an embodiment of the present invention provides a method for constructing a nasopharyngeal carcinoma detection model, which includes: obtaining test data for each classification indicator and Epstein-Barr virus content, wherein the classification indicators include promoter activity, DNA fragmentation degree, and terminal base combination; using the test data of each classification indicator as input data to construct a single indicator classification model to obtain a single indicator prediction score for the sample; using the single indicator prediction scores of all classification indicators and Epstein-Barr virus content as input data, using logistic regression to construct an integrated model for nasopharyngeal carcinoma detection, and obtaining a comprehensive score, called NPC-Score. The test data for promoter activity is the promoter activity difference value;

[0011] The promoter activity difference value = (number of background region fragments / total number of fragments) × background region length - (number of core region fragments / total number of fragments) × core region length; wherein the core region is 100 to 150 bp before the transcription start site TSS to 50 to 150 bp after the transcription start site TSS; the background region is 1500 to 2500 bp extending outward from both sides of the core region;

[0012] The test data of the terminal base combination is the proportion of the number of each terminal base combination;

[0013] The proportion of each terminal base combination number = the sum of the number of each terminal base combination / the number of terminal base arrangement types; the number of terminal base combinations is 4;

[0014] The test data of the degree of DNA fragmentation is the ratio of short fragments to long fragments in each divided area;

[0015] The ratio of short fragments to long fragments = the number of short fragments / the number of long fragments. The divided region refers to the region obtained by dividing the sample genome according to a specific length, and the specific length is 2M.

[0016] The EB virus content refers to the proportion of EB virus genome fragments in the measured DNA fragments; the proportion of EB virus genome fragments = the number of fragments aligned to the EB virus genome / the total number of sequenced fragments.

[0017] The formula of the logistic regression model is as follows: NPC-Score = exp(Z) / (1+exp(Z)), where Z = -B+(a1×promoter activity)+(a2×terminal base combination)+(a3×DNA fragment fragmentation degree)+(a4×Epstein-Barr virus content); wherein NPC-Score is the comprehensive score of the sample, Z is the sum of the sum of each feature item multiplied by the weight and the intercept item; B is the intercept item, a1 is the promoter activity weight, a2 is the terminal base combination weight, a3 is the DNA fragment fragmentation degree weight, and a4 is the Epstein-Barr virus content weight.

[0018] In a second aspect, an embodiment of the present invention provides a nasopharyngeal carcinoma detection model construction device, which includes: a data acquisition module, a prediction score acquisition module, and a nasopharyngeal carcinoma detection model construction module;

[0019] The data acquisition module is used to obtain the classification index of the sample to be tested and the test data of the EB virus content, wherein the classification index includes promoter activity, terminal base combination and DNA fragment fragmentation degree;

[0020] The prediction score acquisition module is used to input the test data of each classification indicator into the single indicator classification model to obtain the single indicator prediction score for the sample, wherein the single indicator classification model is a single indicator classification model constructed by the construction method described in the above embodiment;

[0021] The nasopharyngeal carcinoma detection model building module uses the Epstein-Barr virus content data and the single indicator prediction scores of all classification indicators as input data to build a nasopharyngeal carcinoma detection model.

[0022] In a third aspect, an embodiment of the present invention provides a method for processing test data, which includes: obtaining test data for each classification index and Epstein-Barr virus content of a sample, wherein the classification indexes include promoter activity, terminal base combination, and degree of DNA fragmentation; inputting the test data for each classification index into a single index classification model constructed by the construction method described in the aforementioned embodiment to obtain a single index prediction score for the sample; inputting the Epstein-Barr virus content data and the single index prediction scores of all classification indexes into a nasopharyngeal carcinoma detection model constructed by the construction method described in the aforementioned embodiment.

[0023] In a fourth aspect, an embodiment of the present invention provides a test data processing device, which includes: a data acquisition module, an execution module and a prediction module;

[0024] The data acquisition module is used to obtain the classification index of the sample to be tested and the test data of the EB virus content, wherein the classification index includes promoter activity, terminal base combination and DNA fragment fragmentation degree;

[0025] The execution module is used to input the test data of the classification index into the single index classification model constructed by the construction method described in the above embodiment to obtain the single index prediction score for the sample;

[0026] The prediction module is used to input the data of EB virus content and the single index prediction scores of all classification indexes into the nasopharyngeal carcinoma detection model constructed by the construction method described in the above embodiment to obtain the prediction results of the samples.

[0027] In a fifth aspect, an embodiment of the present invention provides an electronic device, comprising: a processor and a memory, wherein the memory is used to store one or more programs, and when the programs are executed by the processor, the processor implements the method for constructing a nasopharyngeal carcinoma detection model as described in the aforementioned embodiment, or the method for processing test data as described in the aforementioned embodiment.

[0028] In a sixth aspect, an embodiment of the present invention provides a computer-readable medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for constructing a nasopharyngeal carcinoma detection model as described in the aforementioned embodiment, or the method for processing test data as described in the aforementioned embodiment, is implemented.

[0029] The advantages of the present invention are:

[0030] The present invention mines multiple dimensions of genomic features (promoter activity, DNA fragmentation degree, terminal base combination, and Epstein-Barr virus content) that can be applied to nasopharyngeal carcinoma detection through whole genome sequencing of plasma free DNA. Classification models are constructed for the three features (promoter activity, DNA fragmentation degree, and terminal base combination) using a machine learning algorithm to obtain a predicted score for each indicator for the sample. These scores are then integrated with the Epstein-Barr virus content data using a logistic regression model to obtain a final classification prediction model.

[0031] The nasopharyngeal carcinoma detection model constructed by the present invention can significantly improve the efficiency and accuracy of nasopharyngeal carcinoma detection, can simultaneously obtain nasopharyngeal carcinoma genome information and Epstein-Barr virus genome information, and the amount of data required for analysis is small, and only the average sequencing depth needs to be 0.1× of the whole genome. The cost and / or effect of the detection exceeds the existing technology, and is suitable for analyzing and predicting nasopharyngeal carcinoma at various stages and prognosis evaluation, especially for early detection of nasopharyngeal carcinoma. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart for constructing a nasopharyngeal carcinoma detection model in Example 1;

[0033] Figure 2 The distribution diagram of promoter activity corresponding to different groups of samples in Example 1;

[0034] Figure 3 The top 10 terminal base combinations with significant differences in proportions between different groups of samples in Example 1;

[0035] Figure 4 The data for comparing the EB virus content in healthy people and nasopharyngeal carcinoma at different stages in Example 1;

[0036] Figure 5The data distribution results of the fragmentation degree between different groups of samples in Example 1;

[0037] Figure 6 The performance results of the nasopharyngeal carcinoma detection model compared with each single index and EB virus content using AUC in Example 2;

[0038] Figure 7 The results of the nasopharyngeal carcinoma detection performance for patients at different tumor stages in Example 3 and the comparison with the clinical EB virus load detection;

[0039] Figure 8 This is an evaluation of the effect of the nasopharyngeal carcinoma detection model in Example 4 on nasopharyngeal carcinoma patients before and after treatment. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the technical scheme in the embodiments of the present invention will be described clearly and completely below. If the specific conditions are not specified in the embodiments, they are carried out according to conventional conditions or conditions recommended by the manufacturer. If the manufacturer of the reagents or instruments used is not specified, they are all conventional products that can be purchased commercially.

[0041] Most of the existing studies only study a single indicator for NPC detection, and rarely conduct joint research from multiple dimensions. Moreover, when multiple tumor detection indicators are known, the detection results that can be achieved by existing technologies still have problems such as low sensitivity, low specificity and / or high detection costs.

[0042] In terms of data analysis, the present invention analyzes data from multiple dimensions, performs dimensionality reduction and feature extraction on the original feature data, defines reasonable model corresponding parameters, uses the grid search method to select the optimal parameter combination, obtains an excellent classification model (integrated model) through iterative training, defines integrated scoring rules, and finally uses the validation set and test set to verify the model effect.

[0043] Compared with the existing technology, it has the advantages of high sensitivity and good specificity, and the detection cost is low, so it is suitable for promotion.

[0044] Specifically, a method for constructing a nasopharyngeal carcinoma detection model provided in an embodiment of the present invention comprises the following steps:

[0045] Obtaining test data of each classification index and EB virus content, wherein the classification index includes promoter activity, DNA fragmentation degree, and terminal base combination;

[0046] The test data of each classification indicator is used as input data to build a single indicator classification model to obtain the single indicator prediction score for the sample;

[0047] The data of EB virus content and the single indicator prediction scores of all classification indicators were used as input data, and logistic regression was used to construct an integrated model for NPC detection.

[0048] Promoter activity: Through whole genome sequencing of cfDNA isolated from plasma, it can be found that the distribution of nucleosomes in cfDNA is closely related to the nuclear structure, gene structure and promoter activity observed in cells. In the promoter region of transcriptionally active genes, the chromosome structure is relatively loose and the nucleosome distribution is relatively sparse, while this phenomenon does not exist in silent genes, the chromosome structure is relatively tight, and the nucleosome distribution is relatively uniform and dense. Therefore, the promoter activity can be inferred based on the difference in nucleosome distribution, because in the process of tissue cell DNA degradation into the blood, DNA that is not protected by nucleosomes is more easily degraded and less likely to be detected. Therefore, when sequencing cfDNA, the closer the active gene is to the transcription start site, the fewer DNA molecules will be detected, which is reflected in the sequencing data as the sequencing depth is lower. Therefore, the promoter activity can be quantified based on this principle. The same tumor cells should also have the same unique characteristics, so they can be used as indicators for nasopharyngeal carcinoma detection.

[0049] The core region and background region are defined according to the transcription start site. The core region is 100 to 150 bp before the transcription start site (TSS) to 50 to 150 bp after the transcription start site (TSS); the background region is 1500 to 2500 bp extending outward from the edges of both sides of the core region.

[0050] The sequencing depth detected for each region is counted and used to calculate the test data of the promoter activity. Preferably, the test data of the promoter activity is a promoter activity difference value, and the promoter activity difference value = (number of background region fragments / total number of fragments (million)) × background region length (Kb) - (number of core region fragments / total number of fragments) × core region length (Kb). This score represents the degree of promoter activity and is used as an input feature for model construction.

[0051] Terminal base combination: The breakage of cfDNA is not a random event, especially for ctDNA released from tumor cells, which has a preferred breakage site, that is, it will form specific break ends. Deep sequencing can find these tumor-specific breakage end sites, but for low-depth sequencing data, it is generally difficult to see fixed sites, but low-depth whole genome sequencing can find different terminal base combinations. The present invention found that statistically analyzing the proportion of base combination types at the end of DNA can more effectively enrich the differences between nasopharyngeal carcinoma patients and healthy groups. Find the types of terminal base combinations with significant differences, and use these base combination types with different proportions as input features for model construction.

[0052] Preferably, the test data of the terminal base combination is the proportion of differential terminal base combinations, and the proportion of differential terminal base combinations = (type of base arrangement with significant difference compared with the base combination at the end of cfDNA fragments of a healthy population / sum of terminal base arrangement types).

[0053] The "type of base combination with significant difference" can be specifically confirmed in the following way: based on the base distribution at the end of each DNA molecule in the healthy sample and the sample to be tested, the quantity proportion of each terminal base combination is counted; when the quantity proportion of a terminal combination is significantly different between the healthy group and the sample to be tested, the terminal base arrangement is marked as the type of terminal base combination with significant difference.

[0054] In this article, "significant difference" may refer to a situation where P < 0.05 or P < 0.01, that is, the difference is statistically significant.

[0055] Preferably, the terminal base arrangement refers to the arrangement of the last 2 to 6 bases at the end of the cfDNA fragment, and specifically can be an arrangement of 2, 3, 4, 5 or 6 bases.

[0056] More preferably, it is 4. The difference in the proportion of the terminal 4 base combinations can more effectively reflect the difference between nasopharyngeal carcinoma samples and healthy groups, thereby more accurately and effectively detecting nasopharyngeal carcinoma samples. When the terminal base combination is a combination of the terminal 4 bases, the sum of the terminal base combinations is 4×4×4×4=256. The same applies to other base numbers and will not be repeated.

[0057] Degree of DNA fragmentation: After entering the blood, DNA will be degraded to form cfDNA. The fragment length of cfDNA in NPC patients is shorter than that in healthy people. Compared with non-tumor individuals, the ratio of short to long fragments in cfDNA of NPC patients is more disordered, the degree of fragmentation is higher, and there are differences in different regions of the genome.

[0058] Preferably, the test data of the degree of DNA fragmentation is the difference distribution ratio of DNA fragments of different lengths in each region with a difference in the degree of DNA fragmentation, and the difference distribution ratio of DNA fragments of different lengths = the number of short fragments / the number of long fragments, wherein the region with a difference in the degree of DNA fragmentation refers to a divided region where the ratio of short fragments to long fragments is significantly different compared with a healthy sample, and the divided region refers to a region obtained by dividing the sample genome according to a specific length, the length of the long fragment is 151bp to 220bp, and the length of the short fragment is 90bp to 150bp.

[0059] Preferably, the specific length is 0.1-5M, more preferably 2M.

[0060] EB virus content: EB virus DNA will also be released into the blood along with the fragmentation and apoptosis of nasopharyngeal carcinoma cells. Through whole genome sequencing technology, EB virus genome DNA can also be detected synchronously with the human genome. The higher the EB virus content in the blood, the greater the possibility of nasopharyngeal carcinoma. The proportion of EB virus genome fragments = the number of fragments aligned to the EB virus genome / the total number of sequenced fragments.

[0061] The “prediction score” in this article can be selected from the classification result of the probability that the sample suffers from nasopharyngeal carcinoma output by the classification model.

[0062] Preferably, the construction method uses any one of the classification algorithms of support vector machine (SVM), random forest, multi-layer perceptron, elastic network, decision tree and deep neural network to construct a classification model for each classification indicator, preferably SVM.

[0063] Preferably, the present application also includes using a feature dimensionality reduction method to reduce the dimension and extract features of the original data of each classification indicator, defining reasonable corresponding parameters of the classification model, using a network search method to select the optimal parameter combination, and obtaining an excellent classification model through iterative training.

[0064] Optionally, the feature dimension reduction method is selected from any one of LASSO, PCA (principal component analysis), elastic network, random forest and recursive elimination, preferably LASSO.

[0065] In the construction method provided in an embodiment of the present invention, the test data of the classification index is obtained based on the sequencing data of the sample, and the sequencing can be high-depth sequencing, medium-depth sequencing or low-depth sequencing, preferably low-depth sequencing, preferably 0.1-5× sequencing, preferably 0.1×.

[0066] The logistic regression model plots the single indicator model and the integrated ROC curve in the same coordinates to intuitively distinguish the pros and cons. The model represented by the ROC curve close to the upper left corner has the most accurate prediction. The AUC is used to quantify the performance of the model, and the model performance can also be compared by directly comparing the size of the AUC.

[0067] Preferably, the construction method includes using data from a training set and a test set to evaluate and optimize the performance of the classification model.

[0068] During optimization, the predicted score and EB virus content corresponding to a single classification indicator are multiplied by the weight corresponding to each input feature, and the feature item is calculated multiplied by the weight and added to the intercept item. During training, the weight value is further adjusted according to the size of the error until the number of iterations reaches the specified number or the error is less than a certain threshold, then the iteration is stopped.

[0069] Preferably, the construction method further comprises mapping the summed result to [0, 1] using a sigmoid function, that is, converting the result into a percentage.

[0070] Preferably, the construction method includes integrating the single indicator prediction scores corresponding to all classification indicators to obtain a single indicator score, which is then used together with the EB virus content data as input data for the construction of the nasopharyngeal carcinoma detection model. That is, the nasopharyngeal carcinoma detection model score includes three parts: single indicator score, EB virus content, and logistic regression score.

[0071] As long as the technical solution of outputting sample analysis results based on single indicator scoring, EB virus content and logistic regression scoring is achieved through machine learning methods, it falls within the protection scope of this application.

[0072] The present invention does not impose any specific limitation on the specific type of experimental reagents, as long as the reagents can be used to obtain test data of sample classification indicators and EB virus content.

[0073] The embodiment of the present invention further provides a nasopharyngeal carcinoma detection model construction device, which includes: a data acquisition module, a prediction score acquisition module, and a nasopharyngeal carcinoma detection model construction module.

[0074] The data acquisition module is used to obtain the classification index of the sample to be tested and the test data of the EB virus content, wherein the classification index includes promoter activity, DNA fragment fragmentation degree, and terminal base combination;

[0075] The prediction score acquisition module is used to input the test data of each classification indicator into the classification model to obtain the single indicator prediction score of the sample, and the classification module is a classification model constructed by the construction method described in any of the above embodiments;

[0076] The NPC detection model building module is used to take the data of EB virus content and the single indicator prediction scores of all classification indicators as input data, and use logistic regression to build a NPC detection model.

[0077] The embodiment of the present invention also provides a method for processing test data, which includes:

[0078] Obtaining test data of each classification index and EB virus content of the sample, wherein the classification index includes promoter activity, DNA fragment fragmentation degree, and terminal base combination;

[0079] Input the test data of each classification index into the classification model constructed by the construction method described in any of the above embodiments to obtain the prediction score of the sample;

[0080] The data of EB virus content and the single index prediction scores of all classification indexes are input into the nasopharyngeal carcinoma detection model constructed by the construction method described in any of the above embodiments.

[0081] The data processing method is not aimed at the diagnosis or treatment of nasopharyngeal carcinoma directly, but is mechanically processed according to the test data.

[0082] The embodiment of the present invention further provides a test data processing device, which includes: a data acquisition module, an execution module and a prediction module.

[0083] The data acquisition module is used to obtain the classification index of the sample to be tested and the test data of the EB virus content, wherein the classification index includes promoter activity, DNA fragment fragmentation degree, and terminal base combination;

[0084] An execution module, used for inputting the test data of the classification index into the classification model constructed by the construction method described in any of the above embodiments, to obtain a single index prediction score for the sample;

[0085] The prediction module is used to input the data of EB virus content and the single indicator prediction scores of all classification indicators into the nasopharyngeal carcinoma detection model constructed by the construction method described in any of the above embodiments to obtain the prediction results of the samples.

[0086] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.

[0087] Example 1

[0088] Refer to the attached Figure 1 , this embodiment provides a method for constructing a nasopharyngeal carcinoma detection model, which includes the following steps.

[0089] 1. Extraction of cfDNA and low-depth WGS sequencing.

[0090] Plasma separation: Peripheral blood is collected in non-invasive anticoagulation tubes, and the sample information and sample status are recorded and verified. The blood collection tube is placed in a centrifuge, the supernatant is aspirated and centrifuged again, and the supernatant is aspirated again to obtain a plasma sample.

[0091] Extraction of cell-free DNA (cfDNA) in blood: Use plasma free DNA extraction kit to extract cell-free DNA in plasma. After the plasma is completely thawed, centrifuge at high speed, transfer the centrifuged plasma, add lysis buffer, magnetic beads and other reagents, shake and mix, react for a period of time, then use cleaning solution to wash, and use elution solution to recover DNA.

[0092] DNA quality detection: DNA concentration is detected using a Qubit nucleic acid / protein quantitative fluorometer or microplate reader.

[0093] WGS library construction and sequencing: The whole genome sequencing library construction reagent is used for library construction, which mainly includes two steps: end repair (end repair buffer, end repair enzyme) and adapter connection (adapter connection buffer, adapter ligase, index adapter). The library is quantitatively analyzed using the qPCR method to check whether various parameters such as the melting curve meet the requirements and calculate the library concentration. After the library is quantified, it is sequenced on the machine.

[0094] 2. Each individual classification index and EB virus content come from the low-depth whole-genome sequencing data obtained by the above experimental method.

[0095] Obtain sample data. The data used in this embodiment is low-depth whole genome second-generation sequencing (WGS) of 234 nasopharyngeal carcinoma patients and 319 healthy control individuals recruited by Fujian Cancer Hospital; the sequencing data is then quality controlled, compared, and data mining of classification indicators is performed.

[0096] 3. Perform quality control and comparison on sequencing data.

[0097] Fastp software was used to filter the data from the sequencing machine, including subtracting the sequencing adapter sequence, removing DNA fragments with a sequencing read length of less than 30, and removing DNA fragments with low sequencing quality; BWA was used to compare the filtered data with the Hg19 reference genome to obtain the specific location information corresponding to each DNA fragment on the genome.

[0098] 4. According to the result file after comparison, obtain the data of promoter activity, DNA fragment fragmentation degree, terminal base combination and EB virus content.

[0099] The test data of promoter activity is the promoter activity difference value, which is (number of background region fragments / total number of fragments (million)) × length of background region (Kb) - (number of core region fragments / total number of fragments) × length of core region (Kb); wherein the core region is 150 bp before the transcription start site to 150 bp after the transcription start site; the background region is 1500 bp extending outward from both sides of the central region;

[0100] The test data of the terminal base combination is the proportion of the number of different terminal base combinations, and the proportion of the number of different terminal base combinations = (the number of types of base arrangements with significant differences compared with the healthy group / the sum of the types of terminal base arrangements);

[0101] The test data of the degree of DNA fragmentation is the difference distribution ratio of DNA fragments of different lengths in each region with a difference in the degree of DNA fragmentation, and the difference distribution ratio of DNA fragments of different lengths = the number of short fragments / the number of long fragments, wherein the region with a difference in the degree of DNA fragmentation refers to a divided region where the ratio of short fragments to long fragments is significantly different compared with a healthy sample, and the divided region refers to a region obtained by dividing the sample genome according to a specific length, the length of the long fragment is 151bp to 220bp, and the length of the short fragment is 90bp to 150bp.

[0102] The EB virus content, that is, the proportion of EB virus genome fragments = the number of fragments aligned to the EB virus genome / the total number of sequenced fragments. The sequencing data that can be perfectly aligned to the human genome is first removed, and then the remaining data is re-aligned with the EB virus genome, and the number of DNA fragments that can perfectly match the EB virus genome is counted, and finally the EB virus content is calculated.

[0103] like Figure 2-5 As shown, promoter activity differed between groups ( Figure 2 ); the proportion of different types of terminal base combinations will also be significantly different between different groups of samples ( Figure 3 ); Low-depth WGS analysis of plasma from patients with NPC found that the EB virus content was significantly different between healthy people and patients with NPC, and there were also differences between different stages of NPC ( Figure 4 ); The fragment length of ctDNA in plasma is shorter than that of normal cfDNA, and the degree of fragmentation is higher ( Figure 5 ). The above features are used as indicators for building a NPC detection model.

[0104] 5. Randomly divide the recruited samples into training sets and test sets. During the training process, the LASSO (least absolute shrinkage and selection operator) regression algorithm is used on the original data of the three classification indicators mentioned above to reduce the dimension of the data and extract relatively important features. For the extracted features, the SVM is used to build a classification model through the cross-validation method to obtain the prediction score of a single indicator.

[0105] 6. The single indicator prediction scores of all classification indicators of the samples and the EB virus content of the three classification models are used as input features, and the data of multiple indicators are integrated using the logistic regression model to obtain the sample prediction score (NPC-Score).

[0106] The logistic regression formula is: NPC-Score = exp(Z) / (1+exp(Z)), where Z = -B + (a1×promoter activity) + (a2×terminal base combination) + (a3×DNA fragmentation degree) + (a4×Epstein-Barr virus content); where NPC-Score is the comprehensive score of the sample, Z is the sum of the sum of each feature item multiplied by the weight and the intercept term; B is the intercept term, a1 is the promoter activity weight, a2 is the terminal base combination weight, a3 is the DNA fragmentation degree weight, and a4 is the Epstein-Barr virus content weight.

[0107] The performance of the final model was evaluated using the data from the training set and the test set. The predicted score and EB virus content of the single-index classification model are input items, and each input feature corresponds to a corresponding weight. The sum of each feature item multiplied by the weight and the intercept term is the final output. During training, the weight value will be further adjusted according to the size of the error until the number of iterations reaches the specified number or the error is less than a certain threshold. The iteration will stop, and the sum result is finally mapped to [0,1] using the sigmoid function. The corresponding promoter activity weight a1 is 1.77, the terminal base combination weight a2 is 1.30, the DNA fragment fragmentation weight a3 is 2.35, the EB virus content weight a4 is 1.55, and the intercept term B is -2.89.

[0108] Example 2

[0109] The construction method of Example 1 was used to construct a nasopharyngeal carcinoma detection model. The sensitivity and specificity of the nasopharyngeal carcinoma detection model (NPC-Score) in the training set of the present invention were 98.33% and 99.18%, and the AUC was 0.999. The sensitivity and specificity in the test set were 98.15% and 100%, and the AUC was 0.999 (the final result was subject to the test set effect), which greatly improved the detection accuracy.

[0110] The sensitivity, specificity and AUC were calculated using the nasopharyngeal carcinoma detection model provided in Example 1 and the model constructed by a single classification index obtained by data mining, and the performance results were compared. Figure 6 and Table 1). According to the AUC of the model, it can be judged that the NPC-Score obtained by the NPC detection model is significantly greater than the AUC of other classification indicators. From the results, it can be seen that the sensitivity and specificity of the NPC detection model are also better than the performance of the model constructed by a single indicator AUC, greatly improving the accuracy of detection.

[0111] Table 1

[0112]

[0113] Example 3

[0114] For patients with tumors at different clinical TNM stages, the detection efficiency of the NPC detection model is basically the same, and it is better than the clinical EB virus serological test in each stage. Therefore, it can be concluded that it is applicable to each stage of the tumor ( Figure 7 ). This result shows that the nasopharyngeal carcinoma detection model scoring method used in the present invention is significantly better than the current clinical indicators.

[0115] Example 4

[0116] This example provides an application of the NPC detection model in the evaluation of the prognosis of NPC. The EB virus content in plasma is a valuable biomarker for evaluating the disease progression of NPC patients after treatment. In order to demonstrate the applicability of NPC-Score in NPC monitoring, plasma samples of 24 NPC patients were selected after treatment, and the NPC-Score was calculated using the NPC detection model. At the end of treatment, the EB virus content decreased significantly ( Figure 8 ). In addition, NPC-Score also decreased significantly after chemoradiotherapy ( Figure 8 ). The above findings indicate that the NPC detection model has potential for monitoring the disease status of NPC.

[0117] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A method for constructing a nasopharyngeal carcinoma detection model, characterized in that: It includes the following steps: The test data of each classification index and Epstein-Barr virus content are obtained, and the classification indexes include promoter activity, DNA fragment fragmentation degree, and terminal base combination; the test data of each classification index is used as input data to construct a single index classification model to obtain a single index prediction score for the sample; the single index prediction scores of all classification indexes and the Epstein-Barr virus content data are integrated using a logistic regression model to obtain a comprehensive score for the sample, which is called NPC-Score.

2. The method for constructing a nasopharyngeal carcinoma detection model according to claim 1, characterized in that: The test data of the promoter activity is the promoter activity difference value; The promoter activity difference value = (number of background region fragments / total number of fragments) × background region length - (number of core region fragments / total number of fragments) × core region length; wherein the core region is 100 to 150 bp before the transcription start site TSS to 50 to 150 bp after the transcription start site TSS; the background region is 1500 to 2500 bp extending outward from both sides of the core region; The test data of the terminal base combination is the proportion of the number of each terminal base combination; The proportion of each terminal base combination number = the sum of the number of each terminal base combination / the number of terminal base arrangement types; the number of terminal base combinations is 4; The test data of the degree of DNA fragmentation is the ratio of short fragments to long fragments in each divided area; The ratio of short fragments to long fragments = the number of short fragments / the number of long fragments. The divided region refers to the region obtained by dividing the sample genome according to a specific length, and the specific length is 2M.

3. The method for constructing a nasopharyngeal carcinoma detection model according to claim 1, characterized in that: The EB virus content refers to the proportion of EB virus genome fragments in the measured DNA fragments; The proportion of the EB virus genome fragments = the number of fragments aligned to the EB virus genome / the total number of sequenced fragments.

4. The method for constructing a nasopharyngeal carcinoma detection model according to claim 1, characterized in that: The formula of the logistic regression model is as follows: NPC-Score = exp(Z) / (1+exp(Z)), where Z = -B+(a1×promoter activity)+(a2×terminal base combination)+(a3×DNA fragment fragmentation degree)+(a4×Epstein-Barr virus content); wherein NPC-Score is the comprehensive score of the sample, Z is the sum of the sum of each feature item multiplied by the weight and the intercept item; B is the intercept item, a1 is the promoter activity weight, a2 is the terminal base combination weight, a3 is the DNA fragment fragmentation degree weight, and a4 is the Epstein-Barr virus content weight.

5. A method for processing test data, characterized in that: It includes: Obtaining test data of each classification index and EB virus content of the sample, wherein the classification index includes promoter activity, DNA fragment fragmentation degree, and terminal base combination; Input the test data of each classification index into the single index classification model constructed by the construction method according to any one of claims 1 to 3 to obtain the single index prediction score of the sample; The single indicator prediction scores of all classification indicators and the EB virus content were input into the logistic regression model to obtain the comprehensive score of the sample.

6. A device for constructing a nasopharyngeal carcinoma detection model, characterized in that: It includes: A data acquisition module is used to obtain classification indicators of the sample to be tested and test data of the EB virus content, wherein the classification indicators include promoter activity, DNA fragment fragmentation degree, and terminal base combination; A prediction score acquisition module, used to input the test data of each classification indicator into a single indicator classification model to obtain a single indicator prediction score for the sample, wherein the single indicator classification model is a single indicator classification model constructed by the construction method according to any one of claims 1 to 3; The nasopharyngeal carcinoma detection integrated model building module is used to take the Epstein-Barr virus content data and the single indicator prediction scores of all classification indicators as input data to build a nasopharyngeal carcinoma detection model.

7. An electronic device, characterized in that: It includes: A processor and a memory, wherein the memory is used to store one or more programs, and when the programs are executed by the processor, the processor implements the method for constructing a nasopharyngeal carcinoma detection model as described in any one of claims 1 to 4, or the method for processing test data as described in claim 5.

8. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for constructing a nasopharyngeal carcinoma detection model according to any one of claims 1 to 4 is implemented, or the method for processing test data according to claim 5 is implemented.

Images