CfDNA fragment characteristic marker combination for diagnosing or predicting abnormal state and application of cfDNA fragment characteristic marker combination

Abstract

The invention belongs to the technical field of molecular diagnosis, and particularly relates to a cfDNA fragment feature marker combination for diagnosing whether a subject has an abnormal state or not or predicting whether the subject has the risk of occurrence of the abnormal state or not and application of the cfDNA fragment feature marker combination. On the basis of the cfDNA fragment abundance spectrum obtained by capillary electrophoresis, a group of cfDNA fragment characteristic markers for diagnosing whether the subject has the abnormal state or predicting whether the subject has the risk of occurrence of the abnormal state are obtained through screening, and selection of the markers for diagnosing or predicting the abnormal state is enriched. The cfDNA fragment feature marker combination can be used for diagnosing or predicting the abnormal state of the digestive system or the lung, has important clinical significance and value, is expected to become a key early screening and auxiliary diagnosis tool in clinical practice, and helps doctors to diagnose or predict the abnormal state earlier and more accurately and improve life quality.

Description

Technical Field

[0001] The present application belongs to the field of molecular diagnostic technology, and specifically, relates to a cfDNA fragment characteristic marker combination and its application for diagnosing or predicting abnormal conditions. Background Art

[0002] Cell-free DNA (cfDNA), also known as circulating free DNA or cell-free DNA, is an important biomarker that plays an increasingly important role in the early diagnosis, treatment monitoring and prognosis assessment of tumors. cfDNA is a DNA fragment released into the blood by apoptotic or necrotic cells. These fragments contain rich tumor gene information and can reflect the presence and characteristics of the tumor. Compared with traditional tissue biopsy, cfDNA testing has the advantages of being non-invasive, easy to operate, and highly sensitive, and has therefore received widespread attention.

[0003] In the blood of patients with abnormal conditions, the concentration of cfDNA is usually higher than that of healthy individuals, and its fragment length distribution shows specific characteristics. Studies have shown that cfDNA fragments associated with abnormal conditions tend to be shorter, with lengths mainly concentrated between 150-200bp. This length distribution feature can be used as a biomarker specific to abnormal conditions, and combined with concentration changes, it can help improve the accuracy of early detection and diagnosis of normal conditions.

[0004] The development of technologies such as high-throughput sequencing (NGS) has significantly improved the accuracy and sensitivity of cfDNA fragment detection. These technologies can efficiently capture and analyze trace amounts of cfDNA fragments in the blood, providing a powerful tool for early diagnosis and dynamic monitoring of abnormal conditions.

[0005] However, the use of NGS sequencing technology requires a high sample volume, a long detection cycle and high detection costs. These problems make it difficult to be widely used in clinical batch testing. Summary of the invention

[0006] In order to solve at least one of the above technical problems, the inventors obtained a cfDNA fragment abundance spectrum based on capillary electrophoresis, and further obtained fragment characteristic markers based on the cfDNA fragment abundance spectrum, which can effectively distinguish abnormal states from non-abnormal states, thereby completing the present application.

[0007] The technical solution of this application is as follows: In a first aspect, the present application provides a method for screening cfDNA fragment characteristic markers for diagnosing whether a subject has an abnormal state or predicting whether a subject has a risk of developing an abnormal state, the method is based on a cfDNA fragment abundance spectrum, and the cfDNA fragment abundance spectrum is obtained by the following method: The cfDNA sample is mixed with at least two reference nucleic acids of different lengths and then subjected to capillary electrophoresis to obtain the migration time and corresponding signal intensity of the cfDNA fragments and the reference nucleic acids in the capillary to the detection position, the migration time is used to characterize the length of the cfDNA fragments, and the signal intensity is used to characterize the abundance of the cfDNA fragments, that is, the cfDNA fragment abundance spectrum; The method for screening markers for diagnosing whether a subject has an abnormal state or predicting whether a subject has a risk of developing an abnormal state comprises the following steps: S1, obtaining a cfDNA fragment abundance profile of a population, wherein the population includes a plurality of patients in abnormal states and a plurality of subjects in non-abnormal states; S2, the time period between the swimming times corresponding to two cfDNA fragments of specified lengths is divided into multiple windows, and the abundance of each swimming time in each window is taken as the representative value as the abundance of the window; S3, screening windows with significant differences in abundance between abnormal state patients and non-abnormal state subjects, that is, obtaining the characteristic markers of the cfDNA fragments.

[0008] In this application, free DNA refers to degraded DNA fragments released into the plasma, and its concentration changes with physiological and pathological processes such as tissue damage, cancer, or inflammatory response. Both the mother and the fetal-placental unit can produce cfDNA. In healthy individuals, cfDNA is mainly derived from cell apoptosis, and its fragment length is relatively uniform, concentrated between 150-200bp. In tumor patients, necrosis or abnormal apoptosis of tumor cells will release more cfDNA, and its fragment length is unevenly distributed, usually shorter and less than 150bp. This fragment length characteristic can be used as an important biomarker for tumor diagnosis.

[0009] In some embodiments of the present application, the abnormal state refers to a state relative to a healthy or normal state, and any state in which the mechanism deviates from or is different from a healthy or normal state can be regarded as an abnormal state.

[0010] In some specific embodiments of the present application, the abnormal state includes diseases of the digestive system or lungs, including but not limited to polyps, nodules, inflammation, abnormal hyperplasia, cysts and tumors. In some preferred embodiments of the present invention, the digestive system includes the stomach, intestines, liver, pancreas, gallbladder, pharynx and esophagus. Further, benign diseases or precancerous lesions of the stomach include but are not limited to gastric ulcers, gastric polyps, chronic gastritis, atrophic gastritis, giant gastric mucosal folds (MD) and intestinal metaplasia; benign diseases or precancerous lesions of the intestine include but are not limited to intestinal polyps, intestinal inflammation, intestinal metaplasia, intestinal abnormal hyperplasia, non-progressive adenomas and progressive adenomas; benign diseases or precancerous lesions of the liver include but are not limited to fatty liver, cirrhosis, liver fibrosis, hepatic atypical hyperplasia, hepatic hemangioma, focal nodular hyperplasia and hepatocellular adenoma; benign diseases or precancerous lesions of the pancreas include but are not limited to pancreatic cysts, pancreatitis, pancreatic endocrine tumors, pancreatic ductal epithelial atypical benign diseases or precancerous lesions of the gallbladder include but are not limited to gallstones, cholecystitis, gallbladder polyps, gallbladder adenoma, gallbladder adenomyosis, bile duct papilloma, atypical hyperplasia of the bile duct epithelium and bile duct cyst; benign diseases or precancerous lesions of the pharynx include but are not limited to pharyngitis, tonsillitis, pharyngeal papilloma, leukoplakia, pharyngeal erythema and ectopic squamous epithelium of the pharynx; benign diseases or precancerous lesions of the esophagus include but are not limited to esophagitis, benign esophageal tumors, esophageal polyps, esophageal varices, iron deficiency dysphagia, esophageal squamous cell dysplasia and gastroesophageal junction dysplasia.

[0011] In some embodiments of the present application, the abnormal state refers to a state relative to a benign state. For example, for patients with lung nodules, benign nodules are considered to be a non-abnormal state, while malignant nodules are considered to be an abnormal state.

[0012] In the present application, the cfDNA sample is extracted from a biological sample of a patient with an abnormal state or a subject with a non-abnormal state, and the biological sample includes but is not limited to blood, urine, feces, saliva, pleural effusion and ascites.

[0013] In the present application, capillary electrophoresis is a method that uses a capillary as a separation channel and a high-voltage direct current electric field as a driving force to achieve separation based on the differences in mobility and distribution behavior between cfDNA of different lengths.

[0014] In some embodiments of the present application, the reference nucleic acid is used as a marker to locate the positions of different fragment sizes of cfDNA.

[0015] In some embodiments of the present application, the reason for using the swimming time to characterize the length of the cfDNA fragment is that the swimming time of cfDNA fragments of different lengths is inconsistent. The longer the length, the slower the swimming and the longer the swimming time; the shorter the length, the faster the swimming and the shorter the swimming time. However, it should be noted that the length of the cfDNA fragment is linearly related to the swimming time only within a specific range. In some embodiments of the present application, the inventors found that in the range of 20bp~260bp, the length of the cfDNA fragment and the swimming time have a certain linear relationship. When the length of the cfDNA fragment exceeds 260bp, the two show an exponential relationship, and the swimming time of cfDNA fragments of adjacent sizes will be very close. In some embodiments of the present application, those skilled in the art can fit the length of other cfDNA fragments according to the length of the reference nucleic acid and its corresponding swimming time.

[0016] In some embodiments of the present application, the reason for using the signal intensity to characterize the abundance of cfDNA fragments is that the higher the abundance of cfDNA fragments, the stronger the detection signal intensity. In some embodiments of the present application, those skilled in the art can obtain the concentration of cfDNA fragments corresponding to other signal confidences based on the concentration of the reference substance and its corresponding detection signal intensity, and further obtain the abundance of cfDNA fragments. Furthermore, the abundance of cfDNA fragments is standardized using the total cfDNA concentration of the biological sample.

[0017] In some embodiments of the present application, the lengths of two reference nucleic acids are used as designated lengths.

[0018] In some embodiments of the present application, the swimming time corresponding to the cfDNA fragment of a specified length is obtained by fitting based on the length of the reference nucleic acid and its corresponding swimming time. For example, the length of the first reference nucleic acid is recorded as L1, and the swimming time is recorded as T1; the length of the second reference nucleic acid is recorded as L2, and the swimming time is recorded as T2, then the swimming time T corresponding to the cfDNA of length L can be obtained by fitting based on the nucleic acid electrophoresis kinetic model.

[0019] In some embodiments of the present application, the number of the reference nucleic acids is 2, the length of the smaller reference nucleic acid is 5-80 bp, and the length of the larger reference nucleic acid is 500-2000 bp. In some embodiments of the present application, the length of the smaller reference nucleic acid is 20 bp, and the length of the larger reference nucleic acid is 1000 bp. In some preferred embodiments of the present application, the lengths of the two reference nucleic acids are used as the specified lengths.

[0020] In some embodiments of the present application, in step S2, the windows are of equal length or of unequal length.

[0021] In some embodiments of the present application, the number of windows is determined according to the difference between two specified lengths, and it should be neither too large nor too small. If the length is too large, important cfDNA fragment information will be missed. Therefore, in some specific embodiments of the present application, it is appropriate that the time corresponding to each window does not exceed 0.8 seconds; if the length is too small, there will be too many windows with the same abundance (for example, multiple windows correspond to cfDNA fragments of the same length). In some specific embodiments of the present application, the cfDNA fragment corresponding to each window is not less than 1bp.

[0022] In some embodiments of the present application, the window division method is: standardize the length of the time period between two specified swimming times to 0-1, and use a sliding window segmentation algorithm to cut the interval into 20-125 equally divided windows at intervals of 0.008-0.05.

[0023] In some preferred embodiments of the present application, the inventors found that when the specified lengths are 20 bp and 1000 bp, respectively, the window is divided into 111 windows, and the performance of the markers screened is optimal. The specific division method of the 111 windows is: using a sliding window segmentation algorithm to cut the interval into 111 equally divided windows at an interval of 0.009, numbered bin_0.009, bin_0.018, bin_0.027, ..., bin_0.981, bin_0.990 and bin_0.999, corresponding to the 1st, 2nd, 3rd, ..., 109th, 110th and 111th windows.

[0024] In some embodiments of the present application, in step S2, the representative value is selected from one of a maximum value, a minimum value, a mean value, a median value, a quarter quantile value, and a third quarter quantile value.

[0025] In some embodiments of the present application, the method further comprises a step of normalizing the abundance.

[0026] In some embodiments of the present application, in step S3, windows with significant differences in abundance between abnormal state patients and non-abnormal state subjects are screened by statistical methods or machine learning methods.

[0027] In some embodiments of the present application, in step S3, the statistical method includes variance selection, chi-square test, correlation system, mutual information, T test, rank sum test. Further, the rank sum test includes Wilcoxon rank sum test and KW test.

[0028] In some embodiments of the present application, in step S3, the machine learning method is selected from any one of the following: (1) Wrapping method By training a machine learning model to evaluate the quality of feature subsets, the feature selection process is performed through the model's predictive performance. Common methods: Recursive feature elimination: Start with all features, recursively remove the least important features, and evaluate the performance of each feature subset through cross-validation.

[0029] Forward selection: Start with an empty feature set and gradually add to the model the features that most improve model performance.

[0030] Backward elimination: Start with the full feature set and progressively remove the least important features.

[0031] Genetic Algorithm: Use genetic algorithm to explore the optimal feature subset and evaluate the adaptability of feature combinations through natural selection.

[0032] (2) Embedding method The embedding method embeds the feature selection process into the model training process, and the model automatically selects features during the training process. These methods usually rely on the capabilities of the model itself, such as feature weights, to make selections. Common methods: L1 regularization (Lasso regression): By adding the L1 regularization term to the loss function, the coefficients of some features are forced to become zero, thereby selecting important features. Applicable to linear regression and classification tasks.

[0033] Decision Tree / Random Forest / Gradient Boosted Tree: Tree models (such as decision tree, random forest, gradient boosted tree, etc.) select features based on the importance of each feature, usually measured by information gain or Gini impurity when splitting nodes. Among them, Extreme Gradient Boosted Tree (XGBoost), as an algorithm based on gradient boosted tree, also provides feature importance scores, and features can be selected based on the scores.

[0034] L2 regularization: Through L2 regularization, the model penalizes all features, but does not completely compress the weights of irrelevant features to zero like L1 regularization. It is suitable when there is a linear correlation between features.

[0035] (3) Deep Learning Deep learning models, especially autoencoders, can extract and select important features through unsupervised learning methods. Autoencoders learn low-dimensional representations of data by compressing and reconstructing input data. These low-dimensional representations can be regarded as selected features. Common methods: Autoencoder: An unsupervised neural network that selects the most informative features by mapping the input data into a low-dimensional latent space and then reconstructing it.

[0036] Convolutional Neural Network (CNN): In data types such as images, convolutional neural networks can automatically extract the most characteristic information.

[0037] Deep Neural Networks (DNNs): Deep neural networks can also select important features by learning high-dimensional representations of the data.

[0038] (4) Dimensionality reduction method Dimensionality reduction techniques perform feature selection by projecting features into a low-dimensional space. Although dimensionality reduction methods are not strictly speaking feature selection methods in the traditional sense, they can help retain the most important information in the data and remove redundant features.

[0039] Common methods: Principal Component Analysis (PCA): Selects the most important features by maximizing the variance of the data. It is suitable for numerical data.

[0040] Linear Discriminant Analysis (LDA): It selects features by maximizing the between-class variance and minimizing the within-class variance, and is usually used for classification tasks.

[0041] Factor analysis: used to reveal underlying factors or structures in the data in order to reduce the dimensionality of the data.

[0042] In some specific embodiments of the present application, when the specified lengths are 20 bp and 1000 bp respectively, and the window is divided into 111 windows, the windows corresponding to the cfDNA fragment characteristic markers obtained by screening are selected from bin_0.612, bin_0.630, bin_0.621, bin_0.639, bin_0.603, bin_0.594, bin_0.648, bin_0.657, bin_0.585, bin_0.666, bin_0.135, bin_0. .432, bin_0.099, bin_0.081, bin_0.423, bin_0.171, bin_0.675, bin_0.126, bin_0.072, bin_0.153, bin_0.027, At least one of bin_0.09, bin_0.117, bin_0.576, bin_0.207, bin_0.144, bin_0.414, bin_0.189, bin_0.441 and bin_0.405. window bin_0.612, bin_0.630, bin_0.621, bin_0.639, bin_0.603, bin_0.594, bin_0.648, bin_0.657, bin_0.585, bin_0.666, bin_0.135, b in_0.432, bin_0.099, bin_0.081, bin_0.423, bin_0.171, bin_0.675, bin_0.126, bin_0.072, bin_0.153, bin_0.027, bin_0.09, bin_0 .117, bin_0.576, bin_0.207, bin_0.144, bin_0.414, bin_0.189, bin_0.441, and bin_0.405 correspond to the 68th, 70th, 69th, 71st, 67th, 66th, 72nd, 73rd, 65th, 74th, 15th, 48th, 11th, 9th, 47th, 19th, 75th, 14th, 8th, 17th, 3rd, 10th, 13th, 64th, 23rd, 16th, 46th, 21st, 49th, and 45th windows, respectively.

[0043] In some embodiments of the present application, the window used as a marker is fitted according to the fragment size of the reference nucleic acid and its corresponding separation time to obtain the cfDNA fragment feature. The cfDNA fragment feature refers to dividing the cfDNA fragments into different fragment intervals according to different lengths, and all cfDNA fragments in each fragment interval are a cfDNA fragment feature. For example, the cfDNA fragment feature is: 25bp, including cfDNA fragments with a fragment length of 25bp; for another example, the cfDNA fragment feature is: 61-65bp, including cfDNA fragments with fragment lengths of 61bp, 62bp, 63bp, 64bp and 65bp. In the present application, the cfDNA fragment feature is characterized by a window, and each window may include cfDNA of multiple lengths, such as cfDNA of 25~30bp, or cfDNA of 700~900bp. Of course, each window can also include only one length of cfDNA fragment.

[0044] In the present application, the cfDNA fragment feature is characterized by a representative value of the length of the cfDNA fragments it contains, and the representative value can be one of an average value, a median value, a maximum value, and a minimum value. In some embodiments of the present application, the cfDNA fragment feature 25-29 bp can be expressed as 25 bp (minimum value), 30 bp (maximum value), or 27 bp (average value or median value).

[0045] In some specific embodiments of the present application, the windows used as markers are bin_0.612, bin_0.630, bin_0.621, bin_0.639, bin_0.603, bin_0.594, bin_0.648, bin_0.657, bin_0.585, bin_0.666, bin_0.135, bin_0.432, bin_0.099, bin_0.081, bin_0.423, bin_0.171, bin_0.675, bin_0.126, bin_0.072, bin_0. The cfDNA fragment characteristics corresponding to 153, bin_0.027, bin_0.09, bin_0.117, bin_0.576, bin_0.207, bin_0.144, bin_0.414, bin_0.189, bin_0.441 and bin_0.405 are: 235bp~238bp, 242bp~245bp, 239bp~241bp, 246bp~249bp, 231bp~234bp, 227bp~230bp, 250bp~253bp, 254bp~257bp, 223bp ~226bp, 258bp~260bp, 62bp~64bp, 163bp~166bp, 51bp~52bp, 45bp~47bp, 160bp~162bp, 74bp~76bp, 261bp~263bp, 59bp~61bp, 42bp ~44bp, 68bp~70bp, 27bp~28bp, 48bp~50bp, 56bp~58bp, 219bp~222bp, 85bp~87bp, 65bp~67bp, 157bp~159bp, 79bp~81bp, 167bp~169b p and 154bp~156bp, and each cfDNA fragment feature is represented by the average length of each cfDNA fragment included therein, which are: 237bp, 244bp, 240bp, 248bp, 233bp, 229bp, 252bp, 256bp, 225bp, 259bp, 63bp, 165bp, 51bp, 46bp, 161bp, 75bp, 262bp, 60bp, 43bp, 69bp, 28bp, 49bp, 57bp, 221bp, 86bp, 66bp, 158bp, 80bp, 168bp and 155bp. It should be noted that the markers screened in one embodiment of the present application are only screened from a group of samples used as examples to demonstrate the feasibility of the present application of screening cfDNA fragment feature markers for diagnosing whether a subject has an abnormal state or predicting whether a subject has a risk of developing an abnormal state.When technicians in this field use different samples, the characteristic markers of cfDNA fragments obtained may be different.

[0046] After obtaining the characteristic markers of cfDNA fragments, when applied, obtaining the abundance of the characteristic markers of cfDNA fragments is not limited to the method of the present invention, and the abundance of the characteristic markers of cfDNA fragments can also be obtained by sequencing and other methods.

[0047] The second aspect of the present application provides a system for screening cfDNA fragment characteristic markers for diagnosing whether a subject has an abnormal state or predicting whether a subject has a risk of developing an abnormal state, comprising: The cfDNA fragment abundance spectrum input module is used to obtain the cfDNA fragment abundance spectrum of a population, wherein the population includes a plurality of abnormal state patients and a plurality of non-abnormal state subjects, and the cfDNA fragment abundance spectrum is obtained using the following method: The cfDNA sample is mixed with at least two reference nucleic acids of different lengths and then separated by capillary electrophoresis to obtain the migration time and corresponding signal intensity of the cfDNA fragments and the reference nucleic acids in the capillary to the detection position, the migration time is used to characterize the length of the cfDNA fragments, and the signal intensity is used to characterize the abundance of the cfDNA fragments, that is, the cfDNA fragment abundance spectrum; A window division module is connected to the cfDNA fragment abundance spectrum input module and is used for dividing the time period between the separation times corresponding to two cfDNA fragments of specified lengths into multiple windows for the cfDNA fragment abundance spectrum of each individual, and the abundance of each swimming time in each window is taken as a representative value as the abundance of the window; The screening module is connected to the window division module and is used to screen windows with significant differences in signal intensity abundance between abnormal state patients and non-abnormal state subjects to obtain the cfDNA fragment characteristic markers.

[0048] The third aspect of the present application provides a computer device, comprising: a memory for storing a computer program; and a processor for implementing the steps of any method described in the first aspect of the present application when executing the computer program.

[0049] A fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements any method described in the first aspect of the present application.

[0050] The fifth aspect of the present application provides a combination of characteristic markers of cfDNA fragments screened using the method described in the first aspect of the present application.

[0051] In some embodiments of the present application, the cfDNA fragment characteristic marker combination includes 235bp~238bp, 242bp~245bp, 239bp~241bp, 246bp~249bp, 231bp~234bp, 227bp~230bp, 250bp~253bp, 254bp~257bp, 223bp~226bp, 258bp~260bp, 62bp~64bp, 163bp~166bp, 51bp~52bp, 4 At least one of 5bp~47bp, 160bp~162bp, 74bp~76bp, 261bp~263bp, 59bp~61bp, 42bp~44bp, 68bp~70bp, 27bp~28bp, 48bp~50bp, 56bp~58bp, 219bp~222bp, 85bp~87bp, 65bp~67bp, 157bp~159bp, 79bp~81bp, 167bp~169bp and 154bp~156bp.

[0052] In some embodiments of the present application, the cfDNA fragment characteristic marker combination includes at least one selected from 237bp, 244bp, 240bp, 248bp, 233bp, 229bp, 252bp, 256bp, 225bp, 259bp, 63bp, 165bp, 51bp, 46bp, 161bp, 75bp, 262bp, 60bp, 43bp, 69bp, 28bp, 49bp, 57bp, 221bp, 86bp, 66bp, 158bp, 80bp, 168bp, 155bp, for example, 1, 2, 3, ..., 30. When selecting from these cfDNA fragment characteristics, the cfDNA fragment characteristics with the top order are preferably selected, for example, when 3 types are selected, 237bp, 244bp and 240bp are preferably selected.

[0053] The sixth aspect of the present application provides the use of a detection reagent or tool for the characteristic marker of cfDNA fragments described in the fifth aspect of the present application in the preparation of a kit for diagnosing whether a subject has an abnormal state or is at risk of developing an abnormal state.

[0054] In the present application, if the abundance of characteristic markers of cfDNA fragments in the subject's biological sample is abnormal, the subject is diagnosed as having an abnormal state. If further clinical diagnosis does not reveal a clinically recognized abnormal state, the subject is predicted to be at risk of developing an abnormal state.

[0055] In some embodiments of the present application, the detection reagent includes a cfDNA extraction reagent and / or an enrichment reagent.

[0056] In some embodiments of the present application, the detection reagent is used to detect the abundance of each cfDNA fragment feature in the cfDNA fragment feature marker combination. In some specific embodiments of the present application, as described above, the abundance of the cfDNA fragment feature is characterized by the signal intensity detected by capillary electrophoresis.

[0057] The seventh aspect of the present application provides a method for constructing a model for diagnosing whether a subject has an abnormal state or has a risk of developing an abnormal state, comprising the following steps: Obtaining cfDNA data of biological samples of a population, the population comprising subjects in abnormal states and subjects in non-abnormal states, the cfDNA data comprising abundance data of a combination of characteristic markers of cfDNA fragments as described in any one of the fifth aspects of the present application; A prediction model was constructed and validated using the cfDNA data of the biological samples of the population.

[0058] In some preferred embodiments of the present application, the cfDNA data also includes cfDNA concentration data.

[0059] In some embodiments of the present application, the prediction model is constructed using a statistical method or a machine learning method. Further, the statistical method is selected from any one of the following: Linear regression, logistic regression, linear discriminant analysis, and Naive Bayes.

[0060] The machine learning method is selected from any one of the following: Decision Trees, Random Forests, Gradient Boosting Machines, XGBoost, Lightweight Gradient Boosting Machines (LightGBM), Categorical Feature Boosting Machines (CatBoost), Support Vector Machines, K-Nearest Neighbors, and Neural Networks.

[0061] It should be noted that although Naive Bayes is based on statistical principles, in machine learning, Naive Bayes is usually used as a supervised learning method for classification tasks. It builds a classification model by learning the probabilistic relationship between features and categories.

[0062] In some specific embodiments of the present application, the abnormal state is an abnormal state of the digestive system or the lungs, and the prediction model is constructed using linear regression.

[0063] In an eighth aspect, the present application provides a system for diagnosing whether a subject has an abnormal state or has a risk of developing an abnormal state, comprising the following modules: A data input module, used to input cfDNA data of the obtained biological sample of the subject, wherein the cfDNA data includes the abundance data of the cfDNA fragment characteristic marker combination described in the fifth aspect of the present application; A database storage module, for storing the cfDNA data of biological samples of a group, wherein the group includes subjects in abnormal state and subjects in non-abnormal state; The prediction module is connected to the data input module and the database storage module respectively, and is used to build a prediction model using the cfDNA data of the biological samples of the group, and diagnose whether the subject has an abnormal state or has a risk of developing an abnormal state based on the cfDNA data of the subject obtained from the data input module.

[0064] The ninth aspect of the present application also provides another computer device, including: a memory for storing a computer program; and a processor for implementing the steps of any method described in the seventh aspect of the present application when executing the computer program.

[0065] In the tenth aspect of the present application, another computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it implements any method described in the seventh aspect of the present application.

[0066] Compared with the prior art, this application has the following beneficial effects: The present application is based on the abundance spectrum of cfDNA fragments obtained by capillary electrophoresis, combined with statistical algorithms to screen markers for diagnosing whether a subject has an abnormal state or predicting whether a subject has the risk of developing an abnormal state, further enriching the selection of characteristic markers of cfDNA fragments for clinical diagnosis or prediction of abnormal tissue states.

[0067] The combination of characteristic markers of cfDNA fragments obtained by screening in this application can be used to diagnose or predict abnormal tissue states, which has important clinical significance and value, and is expected to become a key early screening and auxiliary diagnosis tool in clinical practice, helping doctors to diagnose or predict abnormal tissue states earlier and more accurately, thereby improving patients' survival rate and quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The cfDNA electrophoresis curve of a sample in Example 1 of the present application is shown.

[0069] Figure 2 A curve of the standardized swimming time of cfDNA and the size of cfDNA fragments of a sample in Example 1 of the present application is shown.

[0070] Figure 3 The classification performance of different second models further constructed by using window cutting screening at different intervals to obtain cfDNA fragment features in Example 1 of the present application is shown.

[0071] Figure 4The risk scores of abnormal groups and healthy control pairs obtained in the training set by the classification model constructed in Example 1 of the present application are shown.

[0072] Figure 5 The risk scores of the abnormal group and the healthy control group obtained by the classification model constructed in Example 1 of the present application in the test set are shown.

[0073] Figure 6 The ROC curve of the classification model constructed in Example 1 of the present application for distinguishing training set samples is shown.

[0074] Figure 7 The ROC curve of the classification model constructed in Example 1 of the present application for distinguishing test set samples is shown.

[0075] Figure 8 The risk scores of the abnormal group and the healthy control group obtained by the classification model constructed in Example 5 of the present application in the validation set 1 are shown.

[0076] Fig. 9 The risk scores of the abnormal group and the healthy control group obtained by the classification model constructed in Example 5 of the present application in the validation set 2 are shown.

[0077] Fig.10 The ROC curve of the classification model constructed in Example 5 of the present application for distinguishing samples of validation set 1 is shown.

[0078] Fig.11 The ROC curve of the classification model constructed in Example 5 of the present application for distinguishing samples of validation set 2 is shown. DETAILED DESCRIPTION

[0079] Unless otherwise indicated, implied from the context, or customary in the prior art, all parts and percentages in this application are based on weight, and the tests and characterization methods used are all synchronized with the filing date of this application. Where applicable, the contents of any patent, patent application or disclosure involved in this application are fully incorporated herein by reference, and their equivalent patent families are also introduced as references, especially the definitions of relevant terms in the art disclosed in these documents. If the definition of a specific term disclosed in the prior art is inconsistent with any definition provided in this application, the definition of the term provided in this application shall prevail.

[0080] In order to make the technical problems, technical solutions and beneficial effects solved by the present application more clearly understood, the present application is further described in detail below in conjunction with embodiments.

[0081] The following examples are used to demonstrate the preferred embodiments of the present application. It will be understood by those skilled in the art that the techniques disclosed in the following examples represent techniques discovered by the inventors that can be used to implement the present application, and therefore can be considered as preferred embodiments of the present application. However, it should be understood by those skilled in the art based on this specification that many modifications can be made to the specific embodiments disclosed herein, and still the same or similar results can be obtained without departing from the spirit or scope of the present application.

[0082] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs, and the disclosure and materials cited therein are hereby incorporated by reference.

[0083] Those skilled in the art will recognize, or be able to ascertain using no more than routine experimentation, many technical equivalents to the specific embodiments of the invention described herein. Such equivalents are intended to be encompassed by the claims.

[0084] The experimental methods in the following examples are conventional methods unless otherwise specified. The instruments and equipment used in the following examples are conventional laboratory instruments and equipment unless otherwise specified; the experimental materials used in the following examples are purchased from conventional biochemical reagent stores unless otherwise specified.

[0085] Example 1 Classification model construction 1. Cohort and clinical information A total of 1043 clinical samples were included in this example, including 87 intestinal abnormality samples, 74 liver abnormality samples, 122 stomach abnormality samples, 56 pancreatic abnormality samples, 153 lung abnormality samples, 76 esophageal abnormality samples, and 551 healthy control samples. The subsequent training set was used for model construction, and the test set was used for threshold determination. The detailed sample information is shown in Table 1.

[0086] Table 1. Grouping of sample for model construction

[0087] Blood samples from abnormal state subjects and healthy control subjects were collected before surgery. Each abnormal state subject enrolled was accurately classified according to the pathological examination results after surgery.

[0088] 2. Collecting, separating and storing blood The specific steps include: 1) Collect blood sample to the 10mL mark; 2) After removing the blood collection needle, immediately and gently invert the blood collection tube 10 times to fully mix the blood sample and the protective agent. The whole blood is stored at room temperature; the plasma is separated by centrifuge and needs to be frozen at -20℃. It should be avoided from freezing and thawing during storage and transportation.

[0089] 3. Extraction of plasma cfDNA 3.1 Pre-experimental treatment Thaw plasma, centrifuge at 4°C, 10,000 rpm (about 16,000 g) for 10 min, and use a pipette to slowly aspirate the upper supernatant, which is plasma, for automated cfDNA extraction of plasma samples.

[0090] 3.2 Nucleic acid extraction cfDNA was extracted using a nucleic acid extraction kit (manufacturer: Angkai, catalog number: OC70404V-24) according to the instructions.

[0091] 4. cfDNA Quantification The cfDNA concentration was determined using a rapid double-stranded DNA (dsDNA) fluorescence quantitative detection kit (manufacturer: Yisheng Biotechnology (Shanghai) Co., Ltd., catalog number: 12642ES60), and the concentration was detected using a Qubit 4.0 fluorescence quantifier.

[0092] 5. Detection of cfDNA using capillary electrophoresis Here we take the Qsep400 instrument as an example to detect cfDNA. The method is as follows: (1) Quality inspection concentration: Samples above 0.4ng / µL: 1µL sample + 9µL water; 0.2ng / µL~0.4ng / µL sample: 2µL sample + 8µL water; Samples below 0.2ng / µL: stock solution quality inspection.

[0093] (2) Place the N1 card holder at room temperature for 15 minutes and allow the glue to pass through for 3-5 minutes.

[0094] (3) Connect the card clip and check the number of times the card clip is used and its validity period.

[0095] (4) Place the sample, buffer, size marker (C109200-100), and alignment marker (N1:20-1K). The size marker is used to calculate the size and concentration of cfDNA fragments, i.e., the reference nucleic acid. The alignment marker is added during each run to equalize the differences in the swimming time of all channels.

[0096] (5) Correctly input the sample number and the diluted product and place it into the corresponding well of the quality inspection instrument.

[0097] (6) Set up the running program: Click on the blank space and select (a) sample position; (b) electrophoresis method; (c) injection time; (d) number of detections; (e) separation time; (f) result file name; (g) parameter settings.

[0098] (7) After the settings are completed, click the card holder to recalibrate (20-1k), and click OK until "Card holder calibration successful" appears.

[0099] (8) After the cartridge is calibrated successfully, click Run to start running the reference nucleic acid and samples.

[0100] (9) After the sample test is completed, select the target result file and open it. In the fragment distribution analysis (Smear) state, check the average fragment size and fragment distribution ratio.

[0101] (10) Export the raw data detected by the instrument.

[0102] 6. Fitting calculation of cfDNA fragment length and abundance Here, we take the data exported by the Qsep400 instrument as an example to perform the fragment length and abundance fitting calculation. The specific method is as follows: (1) Correcting data batches In the raw data derived from instrument detection, cfDNA fragments of different lengths appear in enrichment peaks at different time points, and these peaks can reflect the abundance of cfDNA in the sample.

[0103] According to the corresponding information of time and signal strength (RFU) in the offline data, data batch correction is performed: First, the inventors captured the time intervals when the signal intensity peaks of the reference nucleic acids of 20 bp (LM) and 1000 bp (UM) appeared in each sample and used them as the subsequent calculation range.

[0104] Then, LOESS nonlinear regression was used to fit the corresponding relationship between the length of different fragments and the time when the signal intensity peak appeared (mobility time), and the mobility curve was drawn using the mobility time of different cfDNA fragments and the corresponding signal intensity (RFU). The cfDNA mobility curve of one sample is shown in the figure below: Figure 1 As shown, the horizontal axis is characterized by the length of the cfDNA fragment. Further, the maximum and minimum value algorithm is used to standardize the length of the swimming time interval corresponding to the reference nucleic acid to 0-1, as shown in Figure 2 shown.

[0105] In this embodiment, the inventors used a sliding window segmentation algorithm to cut the interval into 20 to 120 equally divided intervals (bins) at different intervals, which are called cfDNA fragment windows.

[0106] (2) Standardization Since the total amount of cfDNA in the samples is different, the signal abundance under different cfDNA fragment windows obtained by (1) cannot be directly compared between samples. The "sum normalization" normalization method is used to obtain the standardized abundance of the cfDNA fragment window. Based on this, the cfDNA fragment window abundance value matrix is ​​constructed.

[0107] 7. Build a classification model (1) Using the normal distribution model, the threshold for distinguishing abnormal samples from healthy control samples is determined based on the quantitative concentration of cfDNA. This is the first model.

[0108] (2) Feature screening based on the cfDNA fragment window abundance value matrix. In the training set, the rank sum test was used to compare abnormal samples and healthy control samples to identify cfDNA fragment windows with significant differences.

[0109] (3) Feature optimization: Based on the cfDNA fragment window features obtained in (2), a logistic linear regression model was constructed.

[0110] (4) The logistic linear regression algorithm is used again to construct a model based on the model scores of the samples in (3) above. This is the second model.

[0111] By cutting the interval into 20 to 120 equal intervals according to different interval sizes, different cfDNA fragment characteristics can be screened to construct different second models. The performance of the second classification model obtained using different cutting methods is shown in Figure 2. Figure 3 As shown, when the interval is cut into 111 equal intervals at an interval of 0.009, the second model further constructed by screening the cfDNA fragment features has the best performance.

[0112] The above-mentioned first model and the second model constructed using the constructed cfDNA fragment features screened from 111 equally divided intervals are combined together to obtain an integrated classification model.

[0113] 8. Model performance evaluation According to the predicted value of the classification model of the sample and the pathological test results, the receiver operating characteristic curve (ROC curve) of the training set and the test set was drawn. Then, the threshold for distinguishing abnormal patients from healthy people was set according to the optimal threshold in the test set. Based on the results predicted by the model, the pathological test results were taken as the true value to evaluate the prediction efficiency of the model. The model evaluation methods include the area under the ROC curve (AUC), specificity (value range 0-1) and sensitivity (value range 0-1). The higher the value, the better the effect. The AUC of the above preferred cfDNA fragment features in the training set and the test set are shown in Table 2.

[0114] Table 2 AUC of cfDNA fragment features in training and test sets

[0115] It can be seen that each cfDNA fragment feature has a certain classification performance in both the training set and the test set. Relatively speaking, the performance of the 160-170bp fragment, which is commonly used to distinguish tumors from healthy controls, is not optimal.

[0116] Furthermore, the risk scores of the samples in the training set and the test set were calculated using the classification model (concentration + cfDNA fragment features, i.e., the first model and the second model constructed using the constructed cfDNA fragment features screened from 111 equally divided intervals). The results are as follows: Figure 4 and Figure 5 As shown. Figure 4 and Figure 5 It can be seen that using the risk scores obtained by the classification model, abnormal patients are significantly distinguished from healthy controls.

[0117] The risk score and the optimal threshold are used to evaluate the classification performance of the classification model for the training set and the test set. The ROC curves are as follows: Figure 6 and Figure 7 The performance evaluation results are shown in Table 3.

[0118] Table 3 Classification performance of classification models in training set and test set

[0119] It can be seen that the AUC of the classification model in the training set reached 0.908, and the specificity and sensitivity were 0.890 and 0.793 respectively; the AUC in the test set was also as high as 0.903, and the specificity and sensitivity were 0.869 and 0.853 respectively, with excellent performance.

[0120] Example 2: Ability of the classification model to identify and diagnose benign and malignant lung nodules The inventors found that the classification model of Example 1 has a good ability to identify benign and malignant pulmonary nodules. To verify this, the inventors collected 92 subjects with benign pulmonary nodules and 32 patients with malignant pulmonary nodules.

[0121] The classification model was used to calculate the risk scores of subjects with benign pulmonary nodules and patients with malignant pulmonary nodules. The risk scores and optimal thresholds were further used to evaluate the classification performance of the classification model for benign and malignant pulmonary nodules. The performance evaluation results are shown in Table 4.

[0122] Table 4 Classification performance of the classification model for benign and malignant pulmonary nodules

[0123] As shown in Table 4, the classification model showed strong diagnostic ability in the identification of benign and malignant lung nodules, with a sensitivity of 78.1% and a specificity of 86.9%. This shows that the classification model can effectively distinguish between malignant and benign lung nodules and has high potential in accurately identifying the nature of lung nodules. A high specificity means that the classification model can reduce the misdiagnosis rate and provide reliable auxiliary decision support for clinicians, especially in the early screening of lung cancer, which has important application value. Its application prospects in clinical practice are broader, and it is expected to become an important tool for judging lung nodules, helping to improve diagnostic efficiency, reduce misdiagnosis and missed diagnosis rates, and provide patients with more accurate diagnosis and treatment services.

[0124] Example 3: Ability of the classification model to identify and diagnose benign intestinal diseases and precancerous lesions In order to verify whether the classification model of Example 1 can identify benign intestinal diseases and precancerous lesions, the inventors collected 12 patients with intestinal polyps, 23 patients with enteritis, 10 patients with non-progressive adenomas, and 14 patients with progressive adenomas.

[0125] The classification model of Example 1 was used to calculate the risk score of each patient, and the risk score and the optimal threshold were further used to evaluate the performance of the classification model in identifying benign intestinal diseases and precancerous lesions. The results are shown in Table 5.

[0126] Table 5 Performance of classification models in identifying benign intestinal diseases and precancerous lesions

[0127] As shown in Table 5, the sensitivity of the classification model in identifying benign intestinal diseases and precancerous lesions related to colorectal cancer is 57.6%. Although there is still room for improvement in this sensitivity, it already has good clinical value, especially in the early detection of colorectal cancer and its precancerous lesions, and has certain potential. With the continuous accumulation of data and continuous optimization of the model, its recognition ability is expected to be further improved in the future, so as to play a greater role in clinical screening and early intervention. At present, the classification model can provide strong support for clinical practice, helping to achieve early detection and timely intervention of precancerous lesions. Through further optimization, the sensitivity of the model can be significantly improved, and its application effect in the detection of early cancer and precancerous lesions can be enhanced, providing more accurate and reliable auxiliary decision support for clinical diagnosis and treatment.

[0128] Example 4: Ability of the classification model to identify and diagnose benign gastric diseases and precancerous lesions In order to verify whether the classification model of Example 1 can identify benign gastric diseases and precancerous lesions, the inventors collected 8 patients with gastric ulcer, 13 patients with chronic gastric inflammation (such as superficial gastritis, etc.), 9 patients with gastric adenoma (benign), and 10 patients with intestinal metaplasia.

[0129] The classification model of Example 1 was used to calculate the risk score of each patient, and the risk score and the optimal threshold were further used to evaluate the performance of the classification model in identifying benign gastric diseases and precancerous lesions. The results are shown in Table 6.

[0130] Table 6 Performance of classification models in identifying benign gastric diseases and precancerous lesions

[0131] As shown in Table 6, the sensitivity of the classification model in identifying benign intestinal diseases and precancerous lesions associated with gastric cancer is 59.0%. Therefore, it also has future clinical application value.

[0132] It should be noted that although cases of diseases related to the liver, pancreas, gallbladder, pharynx and esophagus have not yet been collected, the inventors believe that the classification model also has the potential for identification and application.

[0133] Example 5 Classification model's ability to identify and diagnose malignant diseases In order to verify the recognition ability of Example 1 for malignant diseases, the inventors collected 30 intestinal abnormalities, 14 liver abnormalities, 15 pancreatic abnormalities, 48 ​​stomach abnormalities, 19 esophageal abnormalities, 33 lung abnormalities, and 203 healthy controls to generate two validation sets (validation set 1 and validation set 2). The sample grouping is shown in Table 7.

[0134] Table 7 Validation set sample grouping

[0135] The classification model is used to calculate the risk scores of validation set 1 and validation set 2 respectively. The results are as follows: Figure 8 and Fig. 9 As shown. Figure 8 and Fig. 9 It can be seen that the risk scores obtained by the classification model show that the abnormal groups in both validation set 1 and validation set 2 are significantly different from the healthy control group.

[0136] The risk score and the optimal threshold are used to evaluate the classification performance of the classification model for validation set 1 and validation set 2. The ROC curves are as follows: Fig.10 and Fig.11 The performance evaluation results are shown in Table 8.

[0137] Table 8 Classification performance of the classification model in the independent validation set

[0138] It can be seen that the AUC of the classification model in validation set 1 reached 0.852, the specificity and sensitivity reached 0.889 and 0.765 respectively; the AUC in validation set 2 reached 0.851, the specificity reached 0.923 and the sensitivity reached 0.696, and the performance was very excellent. In this embodiment, all abnormalities were cancer, indicating that the classification model can be used to diagnose malignant diseases such as cancer.

[0139] All documents mentioned in this application are cited as references in this application, just as each document is cited as reference individually. In addition, it should be understood that after reading the above teaching content of this application, those skilled in the art can make various changes or modifications to this application, and these equivalent forms also fall within the scope defined by the claims attached to this application.

Claims

1. Use of a detection reagent or tool for a combination of cfDNA fragment characteristic markers in the preparation of a kit for diagnosing whether a subject has an abnormal state or predicting whether a subject has a risk of developing an abnormal state, characterized in that: The cfDNA fragment feature combination is obtained using the following method: The cfDNA sample was mixed with reference nucleic acids of 20 bp and 1000 bp in size, and then capillary electrophoresis was performed to obtain the migration time of the cfDNA fragments and the reference nucleic acids in the capillary to the detection position. The time period between the swimming times corresponding to the two reference nucleic acids is divided into 111 windows of equal length, and the combination of cfDNA fragment characteristics corresponding to the 3rd, 8th, 9th, 10th, 11th, 13th, 14th, 15th, 16th, 17th, 19th, 21st, 23rd, 45th, 46th, 47th, 48th, 49th, 64th, 65th, 66th, 67th, 68th, 69th, 70th, 71st, 72nd, 73rd, 74th and 75th windows in order of swimming time from small to large is the cfDNA fragment characteristic marker combination.

2. The use according to claim 1, characterized in that: The characteristic combinations of cfDNA fragments include: 27bp~28bp, 42bp~44bp, 45bp~47bp, 48bp~50bp, 51bp~52bp, 56bp~58bp, 59bp~61bp, 62bp~64bp, 65bp~67bp, 68bp~70bp, 74bp~76bp, 79bp~81bp, 85bp~87bp, 154bp~156bp, 157bp~159bp, 160bp ~162bp, 163bp~166bp, 167bp~169bp, 219bp~222bp, 223bp~226bp, 227bp~230bp, 231bp~234bp, 235bp~238bp, 239bp~241bp, 242bp~245bp, 246bp~249bp, 250bp~253bp, 254bp~257bp, 258bp~260bp and 261bp~263bp.

3. The use according to claim 1 or 2, characterized in that: The abnormal state is an abnormal state of the digestive system or the lungs.

4. The use according to claim 3, characterized in that: Such abnormal conditions include inflammation, nodules, polyps, dysplasia, cysts and tumors.

5. A method for constructing a model for diagnosing whether a subject has an abnormal state or predicting whether a subject has a risk of developing an abnormal state, comprising the following steps: Obtaining cfDNA data of biological samples of a population, the population comprising subjects in abnormal states and subjects in non-abnormal states, the cfDNA data comprising abundance data of the combination of characteristic markers of cfDNA fragments as described in claim 1 or 2; Using cfDNA data, a prediction model was constructed and validated using statistical methods.

6. The method according to claim 5, characterized in that The cfDNA data also includes cfDNA concentration data.

7. The method according to claim 5 or 6, characterized in that: The statistical method is selected from any one of the following: linear regression, logistic regression, linear discriminant analysis and naive Bayes.

8. A computer device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the method according to any one of claims 5 to 7 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 5 to 7 is implemented.

10. A system for diagnosing whether a subject has an abnormal state or predicting whether a subject has a risk of developing an abnormal state, comprising the following modules: A data input module, used to input cfDNA data of a biological sample of a subject obtained, wherein the cfDNA data includes abundance data of the cfDNA fragment characteristic marker combination described in claim 1 or 2; A database storage module, for storing the cfDNA data of biological samples of a group, wherein the group includes subjects in abnormal state and subjects in non-abnormal state; The prediction module is connected to the data input module and the database storage module respectively, and is used to construct a prediction model based on the cfDNA data in the biological samples of the group using statistical methods, and to diagnose whether the subject has an abnormal state or predict whether the subject has a risk of developing an abnormal state based on the cfDNA data of the subject obtained from the data input module.

Images