A cancer primary site tracing method, device, system and storage medium

Abstract

The application provides a cancer primary focus tracing method, device, system and storage medium, and relates to the technical field of cancer tracing. The cancer primary focus tracing method comprises the following steps: acquiring sample tissue data, giving each cancer sample a primary focus label through a methylation chip; determining candidate CpG sites; and constructing a cancer tracing classifier. The application realizes the technology of tracing multiple cancers by using DNA methylation biological characteristics and machine learning methods, and has the following characteristics: due to the high stability of DNA molecules, cancer information is relatively completely retained, so that sample storage and transportation are more convenient; the cancers targeted by the application have a wide coverage range and include general common tumors; cancer-specific CpG sites have a wide coverage; through analysis of the cancer-specific sites, the interference of normal tissues is excluded, so that the accuracy and application range of the classifier are improved; and the high-precision classifier makes the cancer diagnosis process simpler, more accurate and more efficient.

Description

Technical Field

[0001] This invention relates to the field of cancer tracing technology, and more specifically, to a method, apparatus, system, and storage medium for tracing the primary lesion of cancer. Background Technology

[0002] Traditional tumor pathological diagnosis mainly includes cellular and tissue morphological examination, typically involving microscopic observation of cell morphology at the lesion site to qualitatively confirm the benign or malignant nature and cellular origin of the tumor tissue. This process generally includes tissue preparation, sectioning, staining, and microscopic examination. Morphological examination is highly accurate and has long been the standard method for tumor diagnosis. However, the limitations of cellular morphology characteristics lead to unclear diagnoses for some tumors, failing to meet clinical treatment needs, especially poorly differentiated tumors. Furthermore, significant biological differences exist between cancers with similar cellular morphological characteristics. Specialized staining and immunohistochemistry techniques have greatly promoted the development of cell morphology-based pathological diagnostic techniques. Combinations of multiple monoclonal antibodies can assist in identifying the origin of various cancer cells. However, from a technical operation perspective, the process from specimen preparation to microscopic observation is both time-consuming and labor-intensive; the results of microscopic examination are subject to a certain degree of subjectivity; and the training of professional personnel is lengthy, with the discovery of clinically effective immunomarkers reaching a bottleneck. This results in only about 200 commonly used immunohistochemical antibodies, leaving a large number of malignant tumors in clinical practice that cannot be accurately classified or whose origin is unknown. In recent years, against the backdrop of the rapid development of genomics and epigenetics based on high-throughput sequencing technology and the continuous breakthroughs in artificial intelligence algorithms, artificial intelligence molecular classifier technology based on omics data, especially methylation biomarkers, has the potential to become the next generation of tumor pathology auxiliary diagnostic technology.

[0003] The progression from normal cells to tumors involves changes in biomolecules, including mRNA expression and epigenetic modifications. The National Comprehensive Cancer Network (NCCN) guidelines highlight the importance of tissue-of-origin (TO) assays in cancer screening. To date, the 2000-gene expression microarray method has received FDA approval, achieving an overall concordance rate of 88.5% between predictions and clinical diagnoses. Against this backdrop of omics development, Shumei Kato et al. believe that systematic tissue-of-origin assays will become an indispensable part of future cancer treatment (Kato et al. 2021). Clinical studies have shown that in some cancers of unknown primary site (CUP), identifying the primary organ and guiding specific medication through assays can lead to significant survival improvements, such as colorectal cancer, breast cancer, ovarian cancer, kidney cancer, prostate cancer, bladder cancer, and non-small cell lung cancer.

[0004] Currently, domestic molecular detection methods for identifying the origin of cancer cells mainly focus on the RNA expression level (Yang Jialiang et al. 2019; Luo Qibin et al. 2021), but there are few technologies developed and used to assist in the diagnosis of benign and malignant tumors and the origin of tumor cells. Summary of the Invention

[0005] In view of the above, and to address the aforementioned technical problems, the present invention provides a method for tracing the primary tumor of cancer, comprising:

[0006] Acquire sample tissue data, and assign a corresponding primary lesion label to each cancer sample in the sample tissue data using methylation data;

[0007] Candidate CpG sites were determined based on the sample tissue data bearing the primary lesion label;

[0008] Based on the sample tissue data with the primary lesion label and the candidate CpG sites, a cancer source tracing classifier is constructed to facilitate the tracing of cancer samples through the cancer source tracing classifier.

[0009] Preferably, the primary tumor marker includes thymoma, germ cell tumor, serous cystadenocarcinoma, pancreatic cancer, sarcoma, endometrial cancer, liver cancer, lung adenocarcinoma, prostate cancer, B-cell lymphoma, malignant melanoma, glioma, thyroid cancer, neurotumor, gastrointestinal tumor, breast cancer, kidney cancer, other squamous cell carcinomas, and cervical squamous cell carcinoma.

[0010] Preferably, determining candidate CpG sites based on the sample tissue data bearing the primary lesion label includes:

[0011] Significantly differentially expressed CpG sites were identified based on the tissue data from the samples.

[0012] Calculate the absolute median difference of the significantly different CpG sites, and determine the candidate CpG sites based on the absolute median difference.

[0013] Preferably, determining the significantly differentially expressed CpG sites based on the sample tissue data includes:

[0014] The sample tissue data was analyzed using ChAMP software to obtain analysis results; wherein, the analysis results are the significantly different CpG sites between normal tissue or adjacent non-cancerous tissue and the corresponding cancer tissue in the sample tissue data;

[0015] The significance of the significantly different CpG sites in the analysis results was P < 0.05;

[0016] The fold change in the significantly different CpG sites in the analysis results

[0017] Preferably, calculating the absolute median difference of the significantly differentiating CpG sites and determining the candidate CpG sites based on the absolute median difference includes:

[0018] Calculate the median of the significantly differentially expressed CpG sites in each cancer.

[0019] Based on the median, calculate the absolute median difference of the significantly different CpG sites in each cancer.

[0020] Preferably, in calculating the absolute median difference of the significantly differentiating CpG sites in each cancer, the formula for calculating the absolute median difference is:

[0021] MAD = 1.4826 × median(|X i -X m |);

[0022] Where MAD is the absolute median difference; i is the i-th sample in the sample tissue data; X m The median;

[0023] Preferably, the absolute median difference (MAD) is ≥ 0.45.

[0024] Preferably, the step of constructing a cancer origination classifier based on the sample tissue data with the primary lesion label and the candidate CpG sites includes:

[0025] A cancer dataset is established; the cancer dataset includes a training set and a validation set; in the cancer dataset, the ratio of the training set to the validation set is 7:3;

[0026] A random forest model is trained on the training set to obtain the trained cancer source tracing classifier.

[0027] The samples in the sample tissue data are listed as data in the validation set of the probe and input into the cancer source tracing classifier to obtain the probability of each sample having each cancer.

[0028] For the probability, and the primary lesion label of each sample, calculate the overall precision, recall, and AUC value, and use the overall precision, recall, and AUC value as evaluation metrics for the cancer source tracing classifier;

[0029] If the overall accuracy, recall, and AUC value of the cancer source tracing classifier all reach the corresponding preset thresholds, then the cancer source tracing classifier is determined to have passed the evaluation, i.e., the evaluated cancer source tracing classifier is obtained.

[0030] Preferably, training the random forest model on the training set to obtain the trained cancer source tracing classifier includes:

[0031] In the random forest model, a decision tree is built to determine the target number.

[0032] The data in the training set are classified using the target number of decision trees to obtain classification results; the classification results are the percentage scores of each sample in the training set for each cancer, and the percentage scores are used as the probability of each sample having each cancer;

[0033] Each classification is used as the training for the random forest model, thus obtaining the trained cancer source tracing classifier.

[0034] Preferably, after constructing the cancer origination classifier based on the sample tissue data with the primary lesion label and the candidate CpG sites, the method further includes:

[0035] The random forest model was evaluated using 10x cross-validation and target ratio stepwise feature elimination to obtain the number of non-redundant features L corresponding to the random forest model.

[0036] For the target number of decision trees, the distribution of preset features in the decision trees is randomly shuffled;

[0037] Calculate the average reduction in accuracy before and after shuffling;

[0038] Arrange the averages from largest to smallest, and take the top L features as the DNA methylation set corresponding to the cancer source tracing classifier;

[0039] Preferably, the formula for calculating the average of the reduction in accuracy before and after scrambling is as follows:

[0040]

[0041] Wherein, MDA is the average; v1 is the accuracy before shuffling; v2 is the accuracy after shuffling; and the target quantity is 1000.

[0042] Furthermore, to address the aforementioned problems, the present invention also provides a device for tracing the primary cancer lesion, comprising:

[0043] The labeling module is used to acquire sample tissue data and assign a corresponding primary lesion label to each cancer sample in the sample tissue data through methylation data.

[0044] A determination module is used to determine candidate CpG sites based on the sample tissue data with the primary lesion label;

[0045] A construction module is used to construct a cancer source tracing classifier based on the sample tissue data with the primary lesion label and the candidate CpG sites, so as to facilitate the source tracing of cancer samples through the cancer source tracing classifier.

[0046] In addition, to solve the above problems, the present invention also provides a cancer primary lesion tracing system, including a memory and a processor. The memory stores a cancer primary lesion tracing program, and the processor runs the cancer primary lesion tracing program to enable the cancer primary lesion tracing system to perform the cancer primary lesion tracing method as described above.

[0047] In addition, to solve the above problems, the present invention also provides a computer-readable storage medium storing a primary cancer lesion tracing program, which, when executed by a processor, implements the primary cancer lesion tracing method as described above.

[0048] This invention provides a method for tracing the origin of primary cancer lesions, comprising: acquiring sample tissue data; assigning a corresponding primary lesion label to each cancer sample in the sample tissue data using a methylation chip; determining candidate CpG sites based on the sample tissue data with the primary lesion label; and constructing a cancer tracing classifier based on the sample tissue data with the primary lesion label and the candidate CpG sites, so as to facilitate tracing the origin of cancer samples through the cancer tracing classifier. This invention provides a method for tracing the primary source of cancer based on DNA methylation. It combines a random forest algorithm to establish a cancer diagnostic classifier. Compared with RNA sequencing-based cancer screening methods, PCR methylation detection methods, or traditional pathological examinations, this method utilizes DNA methylation biomarkers and machine learning to trace the source of multiple cancers. It has the following characteristics: due to the high stability of DNA molecules, cancer information is preserved more completely, making sample storage and transportation more convenient; the cancer coverage of this invention is broad, including common tumors; the coverage of cancer-specific sites is comprehensive; through the analysis of cancer-specific sites, interference from normal tissues is eliminated, improving the accuracy and applicability of the classifier; the invention of a high-precision classifier makes the cancer screening process simpler, more accurate, and more efficient. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the hardware operating environment involved in an embodiment of the cancer primary lesion tracing method of the present invention;

[0050] Figure 2 This is a flowchart illustrating the first embodiment of the cancer primary lesion tracing method of the present invention;

[0051] Figure 3 This is a detailed flowchart of step S200 in the second embodiment of the cancer primary lesion tracing method of the present invention.

[0052] Figure 4 This is a detailed flowchart of step S220 in the second embodiment of the cancer primary lesion tracing method of the present invention;

[0053] Figure 5 This is a detailed flowchart of step S300 in the third embodiment of the cancer primary lesion tracing method of the present invention;

[0054] Figure 6 This is a detailed flowchart of step S320 in the third embodiment of the cancer primary lesion tracing method of the present invention.

[0055] Figure 7 This is a flowchart illustrating the steps following step S300 (steps S400-S700) in the third embodiment of the cancer primary lesion tracing method of the present invention.

[0056] Figure 8This is a schematic diagram of the overall workflow in the third embodiment of the cancer primary lesion tracing method of the present invention;

[0057] Figure 9 This is a schematic diagram showing the functional modular connection of the cancer primary lesion tracing method of the present invention.

[0058] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0059] The embodiments of the present invention are described in detail below, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0060] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0061] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0062] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0063] like Figure 1 The diagram shown is a structural schematic of the hardware operating environment of the terminal involved in an embodiment of the present invention.

[0064] The cancer primary lesion tracing system of this invention can be a PC, or a mobile terminal device such as a smartphone, tablet, or portable computer. This cancer primary lesion tracing system may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen, an input unit such as a keyboard, or a remote control; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory or a stable memory, such as a disk storage device. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001. Optionally, the cancer primary lesion tracing system may also include RF (Radio Frequency) circuitry, audio circuitry, a Wi-Fi module, etc. In addition, the cancer primary lesion tracing system can also be equipped with other sensors such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, which will not be elaborated here.

[0065] Those skilled in the art will understand that Figure 1 The cancer primary lesion tracing system shown is not intended to limit it and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. Figure 1 As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a data interface control program, a network connection program, and a cancer primary lesion tracing program.

[0066] In summary, the cancer primary lesion tracing method characterized by DNA methylation provided by this invention, combined with the random forest algorithm to establish a cancer diagnostic classifier, is superior to cancer screening methods based on RNA sequencing, PCR methylation detection methods, or traditional pathological tests. It has the following characteristics: due to the high stability of DNA molecules, cancer information is preserved more completely, making sample storage and transportation more convenient; it covers a wide range of cancers, including common tumors; it has comprehensive coverage of cancer-specific sites; through the analysis of cancer-specific sites, interference from normal tissues is eliminated, improving the accuracy and applicability of the classifier; the invention of a high-precision classifier makes the cancer diagnosis process simpler, more accurate, and more efficient.

[0067] Example 1:

[0068] Reference Figure 2 The first embodiment of the present invention provides a method for tracing the primary tumor of cancer, comprising:

[0069] Step S100: Obtain sample tissue data, and assign a corresponding primary lesion label to each cancer sample in the sample tissue data through methylation data;

[0070] As described above, prior to step S100, the process may include collecting primary solid tissue and blood type samples, performing diagnoses on the tissues and samples, and thereby obtaining sample tissue data.

[0071] The sample tissue data mentioned above may include information on the cancer-originating tissue and diagnostic information corresponding to the primary solid tissue and blood type samples.

[0072] It should be noted that the methylation data refers to a collection of data in which C (cytosine) in DNA is modified into mC (cytosine binds a methyl group). This data can be used for epigenetic research and may include, but is not limited to, methylation microarrays, NGS (next-generation sequencing) data for the purpose of methylation sequencing, and nanopore sequencing.

[0073] The methylation chip mentioned above can be Illumina's Infinium Methylation EPIC Bead Chip. Illumina's earliest methylation chip data was 27K (K represents 1000, indicating the approximate number of CpG sites that can be measured), which was later increased to 450K (the mainstream methylation chip). Currently, Illumina has released a new generation product, EPIC (850K). The core technology is now mature at 450K, so 450K is also the mainstream methylation chip data in current databases.

[0074] Furthermore, the primary tumor markers include 19 types of cancer markers, namely: thymoma, germ cell tumor, serous cystadenocarcinoma, pancreatic cancer, sarcoma, endometrial cancer, liver cancer, lung adenocarcinoma, prostate cancer, B-cell lymphoma, malignant melanoma, glioma, thyroid cancer, neurotumor, gastrointestinal tumor, breast cancer, kidney cancer, other squamous cell carcinomas, and cervical squamous cell carcinoma.

[0075] As described above, based on the sample tissue data corresponding to each cancer sample (primary solid tissue and blood sample), and using methylation chips, each cancer sample is assigned a corresponding confirmed primary lesion label, thus obtaining sample tissue data with primary lesion labels.

[0076] Step S200: Determine candidate CpG sites based on the sample tissue data with the primary lesion tag;

[0077] The above-described candidate CpG sites represent the steps of identifying and screening characteristics. Specific candidate CpG sites are then screened from the methylation chip.

[0078] Step S300: Based on the sample tissue data with the primary lesion label and the candidate CpG sites, a cancer source tracing classifier is constructed to facilitate the source tracing of the sample through the cancer source tracing classifier.

[0079] By using a random forest model and training with the obtained candidate CpG sites, a cancer source tracing classifier based on the trained model is constructed. When tracing cancer sources for samples with unknown primary lesions, the data of the cancer samples with unknown primary lesions are input into this cancer source tracing classifier for analysis, thereby obtaining the corresponding cancer source tracing classification information. Specifically, it can be determined that the sample corresponds to a specific cancer among 19 types of cancer.

[0080] In summary, the cancer primary lesion tracing method characterized by DNA methylation provided in this embodiment, combined with the random forest algorithm to establish a cancer diagnostic classifier, has the following characteristics compared with cancer screening methods based on RNA sequencing, PCR methylation detection methods, or traditional pathological tests: due to the high stability of DNA molecules, cancer information is preserved more completely, making sample storage and transportation more convenient; the scope of cancers it targets is broad, including common tumors; the coverage of cancer-specific sites is relatively complete; through the analysis of cancer-specific sites, interference from normal tissues is eliminated, improving the accuracy and applicability of the classifier of this invention; the invention of a high-precision classifier makes the cancer diagnosis process simpler, more accurate, and more efficient.

[0081] Example 2:

[0082] Reference Figure 3 The second embodiment of the present invention provides a method for tracing the primary cancer lesion, based on the above embodiment 1. Step S200, determining candidate CpG sites based on the sample tissue data with the primary lesion tag, includes:

[0083] Step S210: Determine significantly differentially expressed CpG sites based on the sample tissue data;

[0084] Further, step S210, determining significantly differentially expressed CpG sites based on the sample tissue data, includes:

[0085] Step S211: Analyze the sample tissue data using ChAMP software to obtain analysis results; wherein, the analysis results are the significantly different CpG sites between normal tissue or adjacent non-cancerous tissue and the corresponding cancer tissue in the sample tissue data.

[0086] The significantly different CpG sites must meet the following two conditions:

[0087] (1) The significance of the significantly different CpG sites is P < 0.05;

[0088] (2) The fold change of the significantly different CpG sites

[0089] The ChAMP (The Chip Analysis Methylation Pipeline) software, mentioned above, is used to analyze methylation chips 450K and EPIC. This software is based on Limma software to identify differentially methylated CpG sites.

[0090] The above analysis results show the significant differences in CpG sites between normal / adjacent tissues and corresponding cancer tissues.

[0091] Step S220: Calculate the absolute median difference of the significantly different CpG sites, and determine the candidate CpG sites based on the absolute median difference.

[0092] In this embodiment, in order to ensure that the significantly different CpG sites have variation information in all cancers (19 types of cancer) and that the fluctuations are balanced, the absolute median difference of the significantly different CpG sites is calculated.

[0093] The Median Absolute Deviation (MAD) mentioned above is a method for assessing the dispersion of data by calculating the distance between each observation and the median.

[0094] Specifically, for a given data sample set, first find the median, then subtract the absolute value of the median from the original data to form a new data sample, and finally find the median of the new data sample, which is the absolute median difference. For example, in the data sample set X = {2 3 8 7 9 6 4}, the median is 6. Subtracting the median from the original data and taking the absolute value forms a new data sample {4 3 2 1 3 0 2}, and the median of the new data sample is 2. Therefore, the absolute median difference of the original data sample set is 2.

[0095] The median absolute deviation (MAD) is a measure of statistical deviation. Furthermore, MAD is a robust statistic that is better suited to outliers in a dataset than standard deviation. Standard deviation uses the squared distance from the mean, so larger deviations have greater weight, and outliers can significantly impact the results. With MAD, a small number of outliers will not affect the final result.

[0096] Further reference Figure 4Step S220, calculating the absolute median difference of the significantly differentiating CpG sites and determining the candidate CpG sites based on the absolute median difference, includes:

[0097] Step S221: Calculate the median of the significantly differentially expressed CpG sites in each cancer.

[0098] Step S222: Calculate the absolute median difference of the significantly different CpG sites in each cancer based on the median;

[0099] Based on step S211, the median is first calculated, that is, the median of significantly different CpG sites in each cancer is calculated, and then the absolute median difference (MAD) of significantly different CpG sites in all cancers is calculated.

[0100] Furthermore, after sorting according to absolute median difference, the candidate CpG sites are those with a predetermined number of significantly different CpG sites.

[0101] Further, in step S220, the absolute median difference of the significantly differentiating CpG sites in each cancer is calculated, and the formula for calculating the absolute median difference is:

[0102] MAD = 1.4826 × median(|X i -X m |);

[0103] Where MAD is the absolute median difference; i is the i-th sample in the sample tissue data; X m The median;

[0104] Wherein, the absolute median difference (MAD) is ≥ 0.45.

[0105] In this embodiment, significantly differentially expressed CpG sites are identified through ChAMP software analysis. Furthermore, by first calculating the median and then obtaining the absolute median difference, the absolute median difference of each significantly differentially expressed CpG site in each cancer is calculated. After sorting according to the absolute median difference, the significantly differentially expressed CpG sites with a predetermined number of rankings are selected as candidate CpG sites. This allows for the selection of features for further training, improving the accuracy and efficiency of model training.

[0106] Example 3:

[0107] Reference Figure 5 The third embodiment of the present invention provides a method for tracing the primary tumor of cancer. Based on the above embodiment 1, step S300 involves constructing a cancer tracing classifier based on the sample tissue data with the primary tumor label and the candidate CpG sites, including:

[0108] Step S310: Establish a cancer dataset; the cancer dataset includes a training set and a validation set; in the cancer dataset, the ratio of the training set to the validation set is 7:3;

[0109] Step S320: Train a random forest model on the training set to obtain the trained cancer source tracing classifier;

[0110] The above describes how a random forest model is trained using a training set to obtain a cancer source tracing classifier.

[0111] Further reference Figure 6 Step S320, training a random forest model on the training set to obtain the trained cancer source tracing classifier, includes:

[0112] Step S321: In the random forest model, establish a decision tree for the target number;

[0113] The target quantity can be 1000 or other quantities.

[0114] In this embodiment, 1000 decision trees are built and trained in the random forest model.

[0115] Step S322: Classify the data in the training set using the target number of decision trees to obtain classification results; the classification results are the percentage scores of each sample in the training set for each cancer, and the percentage scores are used as the probability of each sample having each cancer.

[0116] As mentioned above, after establishing 1000 decision trees, the classification results of the 1000 decision trees are combined during training to give the percentage score of each type of cancer, that is, the probability of each type of cancer.

[0117] A stratified sampling method is used, randomly selecting k×n samples from each decision tree, where k is the minimum number of samples for a certain cancer, and n is the number of cancer types, which can be 19 in this embodiment; each decision tree randomly selects... Each feature (rounded down, where x is the total number of candidate CpG sites).

[0118] Step S323: Use each classification as training for the random forest model to obtain the trained cancer source tracing classifier.

[0119] Step S330: Input the data from the sample tissue data, which are listed as probes in the validation set, into the cancer tracing classifier to obtain the probability of each sample having each cancer.

[0120] Step S340: For the probability and the primary lesion label of each sample, calculate the overall precision, recall and AUC value, and use the overall precision, recall and AUC value as evaluation indicators to evaluate the cancer source tracing classifier;

[0121] Step S350: If the overall accuracy, recall and AUC value of the cancer source tracing classifier all reach the corresponding preset thresholds, then the cancer source tracing classifier is determined to have passed the evaluation, that is, the evaluated cancer source tracing classifier is obtained.

[0122] The validation set data, with rows and columns representing samples and probes respectively, is input into the classifier. The cancer source tracing classifier is validated using the validation set to obtain the probability of each sample having a certain type of cancer. Overall precision, recall, and AUC (Area Under Curve) are used as evaluation metrics for the classifier's source tracing performance.

[0123] The aforementioned preset threshold is the preset evaluation standard corresponding to the overall accuracy, the recall, and the AUC value. If the evaluation standard is met, the cancer source tracing classifier is deemed to have passed the evaluation; otherwise, it fails the evaluation.

[0124] Furthermore, for cancer source tracing classifiers, NGS (High-Throughput Sequencing) scalability evaluation can be performed. To assess the scalability of the classifier in NGS data applications, this invention tests the NGS application performance of the source tracing classifier in GEO primary cancer or cancer cell lines and self-built NGS capture data. Similarly, overall accuracy and recall can be used as evaluation metrics.

[0125] After obtaining the cancer source tracing classifier, in order to remove redundant cancer information features from the random forest classifier or screen important CpG sites, the random forest model was evaluated using methods such as 10-fold cross-validation, stepwise feature removal, and average precision decline index, resulting in a DNA methylation panel.

[0126] For details, please refer to Figure 7 and Figure 8 In step S300, after constructing a cancer origination classifier based on the sample tissue data with the primary lesion label and the candidate CpG sites, the method further includes:

[0127] Step S400: Evaluate the random forest model using 10x cross-validation and target ratio stepwise feature elimination to obtain the number of non-redundant features L corresponding to the random forest model;

[0128] In order to eliminate redundant cancer information features from the random forest classifier, the accuracy of the random forest model was evaluated by using 10x cross-validation and gradually removing features at a certain ratio, thus obtaining the number of non-redundant features L.

[0129] Step S500: For the target number of decision trees, randomly shuffle the distribution of preset features in the decision trees;

[0130] Step S600: Calculate the average of the precision before shuffling and the precision after shuffling;

[0131] Step S700: Arrange the averages from largest to smallest, and take the top L features as the DNA methylation set (panel) corresponding to the cancer source tracing classifier;

[0132] The above describes the process of randomly shuffling the distribution of feature v on 1000 decision trees and calculating the average difference in accuracy between v1 before shuffling and v2 after shuffling.

[0133] Furthermore, in step S600, the formula for calculating the average of the degree of precision reduction before shuffling and the degree of precision reduction after shuffling is as follows:

[0134]

[0135] Wherein, MDA is the average; v1 is the accuracy before shuffling; v2 is the accuracy after shuffling; and the target quantity is 1000.

[0136] In this embodiment, the random forest model is evaluated using methods such as 10-fold cross-validation, stepwise feature removal, and average precision decline index to obtain a DNA methylation panel. The design of the DNA methylation panel can further reduce sequencing costs and sequencing time, and improve the accuracy of data analysis.

[0137] In addition, refer to Figure 9 This embodiment also provides a device for tracing the primary cancer lesion, comprising:

[0138] The labeling module 10 is used to acquire sample tissue data and assign a corresponding primary lesion label to each cancer sample in the sample tissue data through a methylation chip.

[0139] The determination module 20 is used to determine candidate CpG sites based on the sample tissue data with the primary lesion label;

[0140] The construction module 30 is used to construct a cancer source tracing classifier based on the sample tissue data with the primary lesion label and the candidate CpG sites, so as to trace the source of cancer samples through the cancer source tracing classifier.

[0141] Furthermore, this embodiment also provides a primary cancer lesion tracing system, including a memory and a processor. The memory stores a primary cancer lesion tracing program, and the processor runs the primary cancer lesion tracing program to enable the primary cancer lesion tracing system to perform the primary cancer lesion tracing method as described above.

[0142] In addition, this embodiment also provides a computer-readable storage medium storing a primary cancer lesion tracing program, which, when executed by a processor, implements the primary cancer lesion tracing method as described above.

[0143] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0144] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention. The above are only preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method of cancer primary site tracing, characterized by, The method comprises: obtaining sample tissue data, and assigning a corresponding primary tumor label to each cancer sample in the sample tissue data through methylation data; determining candidate CpG sites according to the sample tissue data with the primary tumor label, comprising: According to the sample tissue data, determining a significant difference CpG site, comprising: analyzing the sample tissue data by ChAMP software to obtain an analysis result; wherein the analysis result is the significant difference CpG site of the normal tissue or the paracancerous tissue in the sample tissue data and the corresponding cancer tissue; the significance of the significant difference CpG site in the analysis result is P<0.05; the difference fold of the significant difference CpG site in the analysis result is >1.9; calculating the absolute median difference of the significant difference CpG site, and determining the candidate CpG site according to the absolute median difference, comprising: calculating the median of the significant difference CpG site in each cancer; calculating the absolute median difference of the significant difference CpG site in each cancer according to the median; the calculation formula of the absolute median difference is: ； wherein, is the absolute median deviation; i is the i-th sample in the sample tissue data; is the median; and the absolute median deviation MAD ≥ 0.45; constructing a cancer tracing classifier based on the sample tissue data with the primary tumor label and the candidate CpG sites, so as to trace a cancer sample through the cancer tracing classifier, comprising: establishing a cancer data set; the cancer data set comprises a training set and a validation set; training a random forest model on the training set to obtain the trained cancer tracing classifier; wherein training the random forest model on the training set to obtain the trained cancer tracing classifier specifically comprises: in the random forest model, a target number of decision trees are established; the data in the training set are classified through the target number of decision trees to obtain a classification result; the classification result is a proportion score of each sample in the training set to each cancer, and the proportion score is taken as the probability of each sample suffering from each cancer; each classification is taken as the training of the random forest model, and the trained cancer tracing classifier is obtained.

2. The method of claim 1, wherein the cancer primary source is traced by, The primary tumor label comprises thymoma, germ cell tumor, serous cystadenocarcinoma, sarcoma, endometrial carcinoma, liver cancer, lung adenocarcinoma, prostate cancer, B-cell lymphoma, malignant melanoma, thyroid cancer, neural tumor, gastrointestinal tumor, breast cancer, kidney cancer, and cervical squamous carcinoma.

3. The method of claim 1, wherein the cancer primary source is traced by, The method further comprises: inputting the data of the samples in the sample tissue data into the cancer tracing classifier to obtain the probability of each sample suffering from each cancer; calculating the overall accuracy, recall rate and AUC value of each sample according to the probability and the primary tumor label of each sample, and taking the overall accuracy, recall rate and AUC value as evaluation indexes for evaluating the cancer tracing classifier; if the overall accuracy, recall rate and AUC value of the cancer tracing classifier all reach the corresponding preset threshold, it is determined that the cancer tracing classifier passes the evaluation, and the evaluated cancer tracing classifier is obtained.

4. The method for tracing the primary cancer lesion as described in claim 1, characterized in that, The proportion of the training set and the validation set in the cancer data set is 7:

3.

5. The method for tracing the primary cancer lesion as described in claim 1, characterized in that, After constructing the cancer tracing classifier based on the sample tissue data with the primary tumor label and the candidate CpG sites, the method further comprises: evaluating the random forest model by using 10-fold cross-validation and target proportion step-by-step feature elimination method to obtain the non-redundant feature number L corresponding to the random forest model; randomly disturbing the distribution of a preset feature in the target number of decision trees; calculating the average reduction of the accuracy before and after disturbance; arranging the average number from large to small, and taking the first L features as the DNA methylation set corresponding to the cancer tracing classifier.

6. The method of claim 5, wherein the cancer primary source tracing method is characterized by, The calculation formula of the average reduction of the accuracy before and after disturbance is: ； MDA = 0.0000 Accuracy before shuffling Accuracy after shuffling; Target number = 1000.

7. A cancer primary lesion provenance device, comprising: The method comprises: The marking module is configured to obtain sample tissue data, and assign a corresponding primary tumor label to each cancer sample in the sample tissue data based on methylation data; The determining module is configured to determine candidate CpG sites based on the sample tissue data with the primary tumor label; The method comprises the following steps: According to the sample tissue data, a significant difference CpG site is determined, including: by ChAMP software, analyzing the sample tissue data to obtain an analysis result; wherein the analysis result is the significant difference CpG site of the normal tissue or the paracancerous tissue in the sample tissue data and the corresponding cancer tissue; the significance of the significant difference CpG site in the analysis result is P<0.05; the difference fold of the significant difference CpG site in the analysis result is >1.9; the absolute median difference of the significant difference CpG site is calculated, and the candidate CpG site is determined according to the absolute median difference, including: calculating the median of the significant difference CpG site in each cancer; according to the median, the absolute median difference of the significant difference CpG site in each cancer is calculated; The constructing module is configured to construct a cancer provenance classifier based on the sample tissue data with the primary tumor label and the candidate CpG sites, so as to trace the provenance of a cancer sample through the cancer provenance classifier, including: establishing a cancer dataset; the cancer dataset comprises a training set and a validation set; training a random forest model on the training set to obtain the trained cancer provenance classifier; wherein training the random forest model on the training set to obtain the trained cancer provenance classifier specifically comprises: In the random forest model, a target number of decision trees are established; the data in the training set are classified through the target number of decision trees to obtain a classification result; the classification result is a proportion score of each sample in the training set for each cancer, and the proportion score is used as the probability of each sample suffering from each cancer; each classification is used as the training of the random forest model, and the trained cancer provenance classifier is obtained.

8. A cancer primary site provenance system, comprising: The cancer primary site tracing system comprises a memory and a processor, the memory stores a cancer primary site tracing program, and the processor runs the cancer primary site tracing program to enable the cancer primary site tracing system to perform the cancer primary site tracing method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a cancer primary site tracing program, and the cancer primary site tracing program is executed by a processor to implement the cancer primary site tracing method according to any one of claims 1-6.

Images