A method and system for tumor-specific gene mutation and methylation combined bioinformatics analysis

Abstract

The application provides a tumor-specific gene mutation and methylation combined bioinformatics analysis method and system, which is applied to the technical field of bioinformatics analysis and comprises the following steps: obtaining mutation site data and methylation data of a sample; performing first-time judgment according to an overall omega value of the sample; if the overall omega value is greater than 1, the sample is determined as a cancer type; if the overall omega value is not greater than 1, a Zscore value is subjected to second-time judgment; if the Zscore value meets a judgment condition one, the sample is determined as the cancer type; if the Zscore value does not meet the judgment condition one, the Zscore value is subjected to third-time judgment; if the Zscore value meets a judgment condition two, the sample is determined as a healthy type; if the Zscore value does not meet the judgment condition two, a random forest auxiliary model is used for fourth-time judgment; and if the random forest auxiliary model determines that the sample belongs to a healthy sample with a probability less than a threshold value, a neural network model is used for fifth-time judgment. The mutation site data and the methylation data are jointly analyzed, and multiple detections are sequentially performed, so that the accuracy of a detection result is ensured.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of biological information analysis, and particularly relates to a tumor-specific gene mutation and methylation combined bioinformatics analysis method and system. BACKGROUND

[0002] Cancer is a major public health problem, and early detection of cancer is crucial for improving survival rates and reducing the burden of disease. Liquid biopsy is a non-invasive diagnostic method that can detect circulating tumor cells or nucleic acids derived from cancer in blood or urine. Tissue biopsy is the gold standard for cancer diagnosis, but it is invasive, expensive, and only provides information about the sampled site. In contrast, liquid biopsy is a less invasive method that analyzes blood or other body fluids for signs of cancer cells or genetic material. The potential of cell-free DNA (cfDNA) liquid biopsy in early detection of cancer signals and prediction of disease has been extensively studied. Cancer genome analysis using next-generation sequencing (NGS) technology involves sequencing and analyzing DNA from cancer cells to identify gene mutations known as hotspots (such as KRAS and PIK3CA), which are associated with the occurrence and development of cancer. There are several different approaches to cancer genome analysis using NGS, such as whole-genome sequencing and genomic analysis.

[0003] Hotspot region mutation analysis using targeted sequencing panels is mainly used to detect specific mutations commonly found in cancer, but this approach may have limitations in early detection of cancer. To further overcome the limitations of traditional mutation and fragment-based liquid biopsy detection, researchers have been exploring the use of cancer-specific methylation signals. This approach involves analyzing DNA methylation patterns specific to cancer cells and using these characteristics to detect cancer-specific methylation signals in cfDNA.

[0004] In addition, machine learning is a common research focus in the field of artificial intelligence and pattern recognition, and its theories and methods have been widely applied to solve complex problems in engineering applications and scientific fields. With the introduction and application of machine learning in computational biology, diagnostic models were constructed based on cfDNA methylation level characteristics to detect and locate potential tumors, overcoming the problem of insufficient resolution in conventional plasma diagnosis.

[0005] Utilizing machine learning methods for modeling and analysis is a highly feasible approach for early cancer diagnosis based on cfDNA. In June 2024, a joint research team from Imperial College London and the University of Cambridge trained an artificial intelligence model—EMethylNET—which trained and evaluated four different model types: logistic regression, support vector machine (SVM), gradient boosting decision tree (XGBoost), and deep neural network (DNN). For the first three models, binary and multi-class classification models were created, respectively. By analyzing DNA methylation patterns, the models were able to identify 13 different types of cancer (including breast cancer, liver cancer, lung cancer, and prostate cancer) from non-cancerous tissues. In 2022, a team from the Second General Hospital of Guangdong Province, in collaboration with a team from Tsinghua University, established a semi-no-reference deconvolution (SRFD) algorithm based on machine learning methods. This algorithm can extract tumor information and locate the tumor origin tissue using cfDNA methylation spectra. The research team ultimately established a Bayesian diagnostic model (SRFD-Bayes), which achieved an average localization accuracy of 76.9% for normal controls and all early-stage tumors.

[0006] In summary, existing gene detection and analysis methods primarily focus on detecting mutations or methylation status alone. For methods that detect mutations alone, liquid biopsy-based detection of mutations and fragments has limitations and is not highly accurate in early cancer detection. For methods that detect methylation status alone, methylation patterns may differ between different cancer subtypes and individuals; the difference in methylation levels between certain cancer types and healthy individuals is not significant, making it difficult to accurately distinguish between cancer and healthy samples. Furthermore, methylation changes may be influenced by factors other than cancer, leading to potential false positives. Summary of the Invention

[0007] In view of the above-mentioned problems in the prior art, the purpose of this invention is to provide a bioinformatics analysis method for tumor-specific gene mutations and methylation, which combines mutation site data and methylation data for joint analysis and performs multiple tests in sequence to ensure the accuracy of the test results.

[0008] A bioinformatics analysis method combining tumor-specific gene mutations and methylation includes the following steps:

[0009] Acquire data on each mutation site and methylation level of the sample to be tested;

[0010] Based on the reference threshold of the mutation site detection model, the Ω value of each mutation site in the sample to be tested is calculated, and the first judgment is made based on the overall Ω value.

[0011] If the overall Ω value is greater than 1, it is determined that the type of the sample to be tested is a cancer type; if the overall Ω value is not greater than 1, the Zscore value of the methylation data is judged for the second time based on the first judgment condition;

[0012] If the first judgment condition is met, it is determined that the sample to be tested is a cancer type; if not, the Zscore value of the methylation data is judged for the third time based on the second judgment condition;

[0013] If the second judgment condition is met, it is determined that the sample to be tested is a healthy type; if not, the fourth judgment is made according to the random forest auxiliary model constructed based on the methylation data;

[0014] If the probability of the sample to be tested belonging to a healthy sample determined by the random forest auxiliary model constructed based on the methylation data is less than a threshold value, the fifth judgment is made according to the neural network model constructed based on the methylation data to determine whether the sample to be tested is a cancer type;

[0015] After determining that the type of the sample to be tested is a cancer type, the cancer type is determined according to the random forest model for cancer type prediction to output the probability value of the sample to be tested belonging to different cancer types.

[0016] Preferably, the corresponding mutation site detection model is constructed by setting a training set and integrating the mutation frequency of the corresponding mutation site data in the training set, the training set includes cancer samples and healthy samples, and each mutation site in the mutation site detection model has a corresponding reference threshold value, and the reference threshold value ranges from 0 to 0.02.

[0017] Preferably, the determination process of the reference threshold value of each mutation site mutation frequency is as follows:

[0018] Each mutation site of each sample in the training set is processed in a loop, and the training set is divided into corresponding cancer sample mutation frequency data set and healthy sample mutation frequency data set according to the mutation site;

[0019] A reference mutation frequency is set, and the proportion of samples in the corresponding healthy sample mutation frequency data set whose mutation frequency is greater than the reference mutation frequency is calculated, and whether each mutation site is reliable is determined according to the sample proportion value;

[0020] On the basis of the reliability of the mutation site, whether there is a difference between the cancer sample mutation frequency data set and the healthy sample mutation frequency data set is determined by T-test;

[0021] In the case of no difference: the reference threshold value is determined according to the maximum value of the mutation frequency in the healthy sample mutation frequency data set and the 0.99 percentile;

[0022] In the case of difference or inability to determine the difference: if the maximum value of the mutation frequency in the healthy sample mutation frequency data set is not greater than the benchmark mutation frequency, the reference threshold of the mutation site is set as the maximum value of the mutation frequency in the healthy sample mutation frequency data set in the training set; if the maximum value of the mutation frequency in the healthy sample mutation frequency data set is greater than the benchmark mutation frequency, the reference threshold of the mutation site is set as the benchmark mutation frequency.

[0023] Preferably, in the case of no difference between the mutation frequency data set of the cancer sample and the mutation frequency data set of the healthy sample, the reference threshold of the mutation site is determined as follows:

[0024] If the maximum value of the mutation frequency in the healthy sample mutation frequency data set of the mutation site is greater than or equal to 0.40, the reference threshold of the mutation site is set as 0.01;

[0025] If the maximum value of the mutation frequency in the healthy sample mutation frequency data set of the mutation site is less than 0.01 and 1.30 times of the 0.99 quantile is also less than 0.01, the reference threshold of the mutation site is set as 0.01;

[0026] If the maximum value of the mutation frequency in the healthy sample mutation frequency data set of the mutation site is greater than 0.01 and less than 0.40, and 1.30 times of the 0.99 quantile is also greater than 0.01, at this time, if the maximum value is greater than 1.30 times of the 0.99 quantile, the reference threshold of the mutation site is set as the maximum value of the healthy sample mutation frequency in the training set, if the maximum value is less than or equal to 1.30 times of the 0.99 quantile, the reference threshold of the mutation site is set as 1.30 times of the 0.99 quantile of the healthy sample mutation frequency in the training set;

[0027] If the maximum value of the mutation frequency in the healthy sample mutation frequency data set of the mutation site is greater than 0.01 and less than 0.40, and 1.30 times of the 0.99 quantile is less than or equal to 0.01, the reference threshold of the mutation site is set as the maximum value of the healthy sample mutation frequency in the training set;

[0028] If the maximum value of the mutation frequency in the healthy sample mutation frequency data set of the mutation site is less than or equal to 0.01 and 1.30 times of the 0.99 quantile is greater than 0.01, the reference threshold of the mutation site is set as 1.30 times of the 0.99 quantile of the healthy sample mutation frequency in the training set.

[0029] Preferably, the value of Ω represents the closeness of the mutation frequency of the mutation site in the to-be-tested sample to the cancer sample, and the overall value of Ω of the to-be-tested sample is the sum of the values of Ω of each mutation site in the to-be-tested sample, and the process of calculating the value of Ω of each mutation site in the to-be-tested sample is as follows:

[0030] If the mutation frequency of the mutation site in the sample to be tested is greater than or equal to 0.40, then Ω = 1;

[0031] If the mutation frequency of the mutation site in the sample to be tested is less than or equal to the reference threshold value, if the reference threshold value is less than 0.01, then Ω = 99.999*AF; if the reference threshold value is greater than or equal to 0.01, then Ω = 50*AF.

[0032] If the mutation frequency of the mutation site in the sample to be tested is greater than the reference threshold value, if the reference threshold value is less than 0.01 and the mutation frequency of the sample to be tested is greater than 0.005, then Ω = 99.999*AF+1; if the reference threshold value is less than 0.01 and the mutation frequency of the sample to be tested is less than or equal to 0.005, then Ω = 99.999*AF; if the reference threshold value is greater than or equal to 0.01 and the mutation frequency of the sample to be tested is greater than 0.005, then Ω = 50*AF+1; if the reference threshold value is greater than or equal to 0.01 and the mutation frequency of the sample to be tested is less than or equal to 0.005, then Ω = 50*AF.

[0033] Wherein, the AF value is used to represent the frequency of each allele at a certain mutation site; the AF value is a number between 0 and 1, wherein 0 represents that the frequency of the allele in all alleles is 0, and 1 represents that the frequency of the allele in all alleles is 1.

[0034] Preferably, the first judgment condition is whether the proportion of Zscore values greater than 3 in the methylation data is greater than 7%, or whether the maximum value of the Zscore values is greater than 10, or whether the average value of the Zscore values is greater than 0.4.

[0035] The second judgment condition is whether the proportion of Zscore values greater than 1 in the methylation data is less than 3%, or whether the maximum value of the Zscore values is less than 1.80, or whether the average value of the Zscore values is less than -1.05.

[0036] Preferably, in the random forest auxiliary model constructed based on the methylation data, if the probability of the sample to be tested belonging to the healthy sample is not less than 65%, the type of the sample to be tested is determined as the healthy type; if the probability of the sample to be tested belonging to the healthy sample is less than 65%, the fifth judgment is made according to the neural network model constructed based on the methylation data.

[0037] Preferably, in the neural network model constructed based on the methylation data, if the probability value of the sample to be tested belonging to the healthy sample is greater than 50%, the type of the sample to be tested is determined as the healthy type, and if the probability value of the sample to be tested belonging to the healthy sample is not greater than 50%, the type of the sample to be tested is determined as the cancer type.

[0038] Preferably, the process of calculating the Zscore value of the methylation data of the sample to be tested is as follows:

[0039] A training set is set, and the training set includes cancer samples and healthy samples, and the cancer samples include samples of different types of cancer patients;

[0040] The average number μ and the standard deviation σ of each methylation site of the healthy samples in the training set are calculated;

[0041] The Zscore value of the sample to be tested compared with the training set is calculated, and the calculation formula is: Wherein, X represents the methylation level value of the methylation site corresponding to the sample to be tested.

[0042] A second object of the present application is to provide a tumor-specific gene mutation and methylation combined bioinformatics analysis system, comprising a data processing module, a mutation site detection module, a methylation detection module and a cancer type detection module, the methylation detection module comprises a Zscore value detection model, a random forest auxiliary model and a neural network model; the data processing module is used for upstream analysis of the sample to be tested to obtain mutation site data of the sample and methylation data of the sample; the mutation site detection module is used for calculating the Ω value of each mutation site in the sample to be tested based on the reference threshold of the mutation site detection model, and outputting whether the sample to be tested is a cancer type according to the overall Ω value; the methylation detection module is used for judging the methylation data of the sample to be tested by the Zscore value detection model, the random forest auxiliary model and the neural network model in turn, and outputting whether the sample to be tested is a cancer type; the cancer type detection module is used for judging which type of cancer the sample to be tested belongs to based on the methylation data when the sample to be tested is a cancer type output by the mutation site detection module or the methylation detection module.

[0043] The tumor-specific gene mutation and methylation combined bioinformatics analysis method and system has the following beneficial effects: the mutation site detection module is used for cancer probability detection based on mutation site data, and after detecting that the sample is a non-cancer type, a plurality of models constructed based on machine learning are used to detect the methylation data in turn, the mutation site data and the methylation data are combined, and the plurality of models are used for joint judgment, so that the accuracy of sample type prediction can be effectively improved. After judging that the sample type is a cancer type, the analysis model can be used to judge the cancer type, and the corresponding cancer type can be output.

[0044] When the sample is analyzed upstream, the mutation site data and the methylation state can be analyzed at the same time. Compared with the method of separately detecting mutations or separately detecting methylation, the analysis method is completely consistent in the processes of sequence filtering and reference genome alignment in the analysis process, and the analysis time and computing resources can be saved. BRIEF DESCRIPTION OF DRAWINGS

[0045] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and are meant to explain the application without limiting the application to the embodiments shown. In the drawings:

[0046] Figure 1 is a system flow chart of the present application;

[0047] Figure 2 is a flow chart of upstream analysis of mutation site data of the present application;

[0048] Figure 3 is a flow chart of upstream analysis of methylation data of the present application;

[0049] Figure 4 is a schematic diagram of the correct prediction proportion of the top three types in the male sample cancer type prediction model constructed based on the present application;

[0050] Figure 5 is a schematic diagram of the correct prediction proportion of the top three types in the female sample cancer type prediction model constructed based on the present application. DETAILED DESCRIPTION

[0051] Embodiment one

[0052] As shown in Figure 1 , a tumor-specific gene mutation and methylation combined bioinformatics analysis method specifically comprises the following steps:

[0053] Step 1, upstream analysis of the sample to be tested is performed to obtain mutation site data of the sample and methylation data of the sample.

[0054] The mutation site data includes mutation frequency, sequencing depth and annotation results corresponding to the mutation site. The methylation data includes methylation level, and the calculation formula is:

[0055]

[0056] In addition, the results of the upstream analysis of the sample to be tested also include QC data of the mutation site and QC data of the methylation. The QC data of the mutation site includes alignment rate, sequencing depth, coverage and sequencing uniformity, and the QC data of the methylation includes enzyme cutting efficiency and sequencing depth. The enzyme cutting efficiency is used to evaluate the cutting effect of the methylation-sensitive restriction endonuclease. Poor cutting effect will lead to inaccurate calculation of the methylation level, thereby affecting the reliability of subsequent analysis. Generally, it should be greater than 98%, which is calculated according to the sequencing depth of the exogenous DNA (Lambda DNA) added in the experiment, and the calculation formula is:

[0057]

[0058] Specifically, as shown in Figure 2 the process of obtaining mutation site data is as follows:

[0059] (1) Use the software trim_galore to filter the adapters and low-quality sequences in the data.

[0060] (2) Use the software BWA to align the filtered sequences to the human reference genome (GRCh37 / hg19), and output the results as a BAM file.

[0061] (3) Use MarkDuplicates in the GATK tool to remove PCR duplicates in the sequences.

[0062] (4) Use the software Vardict to detect mutation sites in the sequences. Mutation sites refer to changes in a single base nucleotide in the sequence of DNA or RNA, including transitions, transversions, insertions, or deletions.

[0063] (5) Use the software ANNOVAR and the annotation information database to annotate the mutation sites.

[0064] (6) Obtain the mutation site data.

[0065] As shown in Figure 3 the process of obtaining methylation data is as follows:

[0066] (1) Use the software trim_galore to filter low-quality sequences and remove adapters in the data.

[0067] (2) Use the software BWA to align the filtered sequences to the human reference genome (GRCh37 / hg19), and output the results as a BAM file.

[0068] (3) Use pysam to process the BAM file, filter sequences that do not align to the reference genome, sequences that are secondary or multiple aligned, and sequences with a R1 and R2 (read1 and read2 sequences in double-end sequencing, i.e., two sequences in two complementary strands with opposite directions) interval >2000bp, and remove PCR duplicates, and then filter sequences in the adapter region.

[0069] (4) Calculate the sequencing depth of the target region and the internal reference region and calculate the methylation level.

[0070] (5) Calculate the enzyme cutting efficiency based on the sequencing depth of the exogenous DNA (Lambda DNA) added.

[0071] Second step, based on the reference threshold of the mutation site detection model, calculate the omega value of each mutation site in the test sample, and make the first judgment according to the overall omega value of the test sample, to judge whether the test sample belongs to the cancer type.

[0072] The first judgment process: if the overall omega value of the test sample is greater than 1, it is determined that the type of the test sample is the cancer type; if the overall omega value of the test sample is not greater than 1, the second judgment is made based on the Zscore value of the methylation data according to the first judgment condition.

[0073] Specifically, the corresponding mutation site detection model is constructed by setting the training set and integrating the mutation frequency of the corresponding mutation site data in the training set. The training set includes cancer samples and healthy samples. It should be noted that male samples and female samples are detected separately.

[0074] Among them, the mutation frequency of each mutation site in the mutation site detection model has a corresponding reference threshold, and the determination process of the reference threshold of the mutation frequency of each mutation site is as follows:

[0075] Each mutation site of each sample in the training set is processed in a loop, and the training set is divided into corresponding cancer sample mutation data set and healthy sample mutation data set according to the mutation site.

[0076] Set the reference mutation frequency as A and calculate the proportion of samples in the healthy sample mutation frequency data set whose mutation frequency is greater than A corresponding to each mutation site. According to the sample proportion value, it is judged whether each mutation site is reliable. Specifically, set the reference mutation frequency as 0.01. For each mutation site, calculate the proportion of samples in the healthy sample mutation frequency data set whose mutation frequency is greater than 0.01. If the proportion of samples in the healthy sample mutation frequency data set whose mutation frequency is greater than 0.01 for this mutation site is greater than 1%, it is considered that this mutation site is not reliable, and all data of this mutation site are deleted in the subsequent analysis.

[0077] If the number of cancer samples and healthy samples of the mutation site measured in the training set is greater than 1, the cancer sample mutation frequency data set and the healthy sample mutation frequency data set are subjected to T test. T test is a statistical method mainly used to evaluate whether there is a significant difference between the average values of two samples. If the P value of T test is greater than 0.05, it is considered that there is no difference between the cancer samples and the healthy samples in the mutation site; if the P value of T test is less than 0.05, it is considered that there is a difference between the cancer samples and the healthy samples in the mutation site.

[0078] In the case that there is no difference between the cancer sample mutation data set and the healthy sample mutation data set corresponding to the mutation site:

[0079] If the maximum value of the mutation frequency in the mutation frequency data set of the healthy sample of the mutation site is greater than or equal to 0.40, the reference threshold of the mutation site is set to 0.01.

[0080] If the maximum value of the mutation frequency in the mutation frequency data set of the healthy sample of the mutation site is less than 0.01 and 1.30 times of the 0.99 quantile (in a set of data, 99% of the data is less than or equal to this value) is also less than 0.01, the reference threshold of the mutation site is set to 0.01.

[0081] If the maximum value of the mutation frequency in the mutation frequency data set of the healthy sample of the mutation site is greater than 0.01 and less than 0.40, and 1.30 times of the 0.99 quantile is also greater than 0.01, then if the maximum value is greater than 1.30 times of the 0.99 quantile, the reference threshold of the mutation site is set to the maximum value of the mutation frequency in the training set of the healthy sample, and if the maximum value is less than or equal to 1.30 times of the 0.99 quantile, the reference threshold of the mutation site is set to 1.30 times of the 0.99 quantile of the mutation frequency of the healthy sample in the training set.

[0082] If the maximum value of the mutation frequency in the mutation frequency data set of the healthy sample of the mutation site is greater than 0.01 and less than 0.40, and 1.30 times of the 0.99 quantile is less than or equal to 0.01, the reference threshold of the mutation site is set to the maximum value of the mutation frequency in the training set of the healthy sample.

[0083] If the maximum value of the mutation frequency in the mutation frequency data set of the healthy sample of the mutation site is less than or equal to 0.01 and 1.30 times of the 0.99 quantile is greater than 0.01, the reference threshold of the mutation site is set to 1.30 times of the 0.99 quantile of the mutation frequency in the training set of the healthy sample.

[0084] In the case that the corresponding mutation sites of the cancer sample mutation data set and the healthy sample mutation data set are different or the difference cannot be determined:

[0085] If the maximum value of the mutation frequency in the mutation frequency data set of the healthy sample of the mutation site is less than or equal to 0.01, the reference threshold of the mutation site is set to the maximum value of the mutation frequency in the training set of the healthy sample; if the maximum value of the mutation frequency in the mutation frequency data set of the healthy sample of the mutation site is greater than 0.01, the reference threshold of the mutation site is set to 0.01.

[0086] The values used for judgment in the determination process of the reference threshold of the mutation frequency of each mutation site in the mutation site detection model are determined by multiple adjustments and tests in the preliminary research.

[0087] Ω value represents the closeness of the mutation frequency of the mutation site in the sample to be tested to the cancer sample. The overall Ω value of the sample to be tested is the sum of the Ω values of each mutation site in the sample to be tested. The process of calculating the Ω value of each mutation site in the sample to be tested is as follows:

[0088] If the mutation frequency of the mutation site in the sample to be tested is greater than or equal to 0.40, then Ω = 1.

[0089] If the mutation frequency of the mutation site in the sample to be tested is less than or equal to the reference threshold, if the reference threshold is less than 0.01, then Ω = 99.999*AF; if the reference threshold is greater than or equal to 0.01, then Ω = 50*AF.

[0090] If the mutation frequency of the mutation site in the sample to be tested is greater than the reference threshold, if the reference threshold is less than 0.01 and the mutation frequency of the sample to be tested is greater than 0.005, then Ω = 99.999*AF+1; if the reference threshold is less than 0.01 and the mutation frequency of the sample to be tested is less than or equal to 0.005, then Ω = 99.999*AF; if the reference threshold is greater than or equal to 0.01 and the mutation frequency of the sample to be tested is greater than 0.005, then Ω = 50*AF+1; if the reference threshold is greater than or equal to 0.01 and the mutation frequency of the sample to be tested is less than or equal to 0.005, then Ω = 50*AF.

[0091] Wherein, AF value is used to represent the frequency of each allele at a certain mutation site. AF value is a number between 0 and 1, where 0 represents that the frequency of the allele in all alleles is 0, and 1 represents that the frequency of the allele in all alleles is 1.

[0092] Step 3, based on the first judgment condition, the Zscore value of the methylation data is judged for the second time.

[0093] If the Zscore value of the methylation data meets the first judgment condition, it is determined that the type of the sample to be tested is the cancer type; if the Zscore value of the methylation data does not meet the first judgment condition, the third judgment is made based on the second judgment condition.

[0094] The first judgment condition is whether the proportion of Zscore values greater than 3 is greater than 7%, or whether the maximum value of Zscore value is greater than 10, or whether the average value of Zscore value is greater than 0.4.

[0095] Step 4, based on the second judgment condition, the Zscore value of the methylation data is judged for the third time.

[0096] If the Zscore value of the methylation data meets the second judgment condition, it is determined that the type of the test sample is a healthy type; if the Zscore value of the methylation data does not meet the second judgment condition, a fourth judgment is made according to the random forest auxiliary model constructed based on the methylation data.

[0097] The second judgment condition: whether the proportion of the Zscore values greater than 1 in the methylation data is less than 3%, or whether the maximum value of the Zscore values is less than 1.80, or whether the average value of the Zscore values is less than -1.05.

[0098] The process of calculating the Zscore value of the methylation data of the test sample is as follows:

[0099] The training set is set. The training set includes cancer samples and healthy samples, and the cancer samples include samples of different types of cancer patients. It should be noted that male samples and female samples are detected separately.

[0100] The average μ and standard deviation σ of each methylation site of the healthy samples in the training set are calculated.

[0101] The Zscore value of the test sample compared with the training set is calculated, and the calculation formula is: Where X represents the methylation level value of the test sample corresponding to the methylation site.

[0102] In the process of judging the Zscore value of the methylation data:

[0103] When the proportion of the Zscore values greater than 3 in the methylation data is greater than 7%, or the maximum value of the Zscore values is greater than 10, or the average value of the Zscore values is greater than 0.4, the type of the test sample is a cancer type.

[0104] When the proportion of the Zscore values greater than 1 in the methylation data is less than 3%, or the maximum value of the Zscore values is less than 1.80, or the average value of the Zscore values is less than -1.05, the type of the test sample is a healthy type.

[0105] When the Zscore value of the methylation data does not belong to the above two cases, the fourth judgment is made based on the random forest auxiliary model constructed based on the methylation data.

[0106] In the random forest auxiliary model constructed based on the methylation data, if the probability of the test sample belonging to the healthy sample is not less than 65%, it is determined that the type of the test sample is a healthy type; if the probability of the test sample belonging to the healthy sample is less than 65%, a fifth judgment is made according to the neural network model constructed based on the methylation data.

[0107] Specifically, the process of constructing a random forest auxiliary model based on methylation data is as follows:

[0108] A training set is set. The training set includes cancer samples and healthy samples, and the cancer samples include samples of different types of cancer patients. It should be noted that male samples and female samples are detected separately.

[0109] A random forest auxiliary model is constructed, and the random forest auxiliary model is trained according to the training set. The random forest auxiliary model includes a plurality of decision trees, and each decision tree is extracted from the training set by replacement extraction.

[0110] The process of constructing a random forest auxiliary model based on methylation data is the same as that of constructing a random forest model in the prior art, that is, first, a new sub-training set D is formed by sampling several times from the training set with replacement sampling, then m features are randomly selected, and then a complete decision tree is learned using the new training set D and the m features. Repeat the above steps K times to get K decision trees to form a random forest.

[0111] In the neural network model constructed based on methylation data, if the probability value of the test sample belonging to the healthy sample is greater than 50%, the type of the test sample is determined to be healthy type, and if the probability value of the test sample belonging to the healthy sample is not greater than 50%, the type of the test sample is determined to be cancer type.

[0112] A neural network model is a machine learning model composed of multiple neuron layers, each neuron layer receives the output of the previous layer as input and calculates the output through a series of nonlinear transformations and weight adjustments. Then it is trained through the backpropagation algorithm, by calculating the error between the predicted output and the true output, and using gradient descent method to update the weights and bias values in the network until the network reaches a predetermined performance level.

[0113] The neural network model in this embodiment is trained by a training set of methylation level data including cancer samples and healthy samples, the input layer of which is the methylation level corresponding to 431 methylation sites; the output layer is 0 (representing healthy type samples) and 1 (representing cancer type samples); the weight value is updated using the backpropagation algorithm; the activation function used in the hidden layer of the neural network model is the rectified linear activation function; and the activation function used in the output layer is the normalized exponential function.

[0114] After determining that the type of the test sample is cancer type, the cancer type is determined according to the random forest model for cancer type prediction to output the probability value of the test sample belonging to different cancer types.

[0115] The output of the random forest model for cancer type prediction is divided into healthy type and specific cancer types, including bladder cancer, bile duct cancer, colorectal cancer, esophageal cancer, kidney cancer, liver cancer, lung cancer, pancreatic cancer, gastric cancer, and thyroid cancer.

[0116] In this embodiment, the parameters of the neural network model constructed by methylation data, the random forest auxiliary model based on methylation data, and the random forest model for cancer type prediction are optimized by ten-fold cross-validation method. Specifically, the ten-fold cross-validation method divides the data set into 10 parts, and in turn uses 9 parts as training data and 1 part as test data for testing. Each test will give the corresponding accuracy (or error rate), and the average of the accuracy (or error rate) of the 10 results is used as an estimate of the accuracy of the algorithm.

[0117] Embodiment Two

[0118] This embodiment proposes a tumor-specific gene mutation and methylation combined bioinformatics analysis system, which includes a data processing module, a mutation site detection module, a methylation detection module, and a cancer type detection module. The methylation detection module includes a Zscore value detection model, a random forest auxiliary model, and a neural network model.

[0119] The data processing module is used for upstream analysis of the test sample to obtain mutation site data and methylation data of the sample.

[0120] The mutation site detection module is used to cyclically process each mutation site in the test sample, compare the mutation frequency in each mutation site data with the corresponding reference threshold, and output whether the test sample is of a cancer type.

[0121] The methylation detection module is used to sequentially pass through the Zscore value detection model, the random forest auxiliary model, and the neural network model to judge the methylation data of the test sample, and output whether the test sample is of a cancer type.

[0122] The cancer type detection module is used to determine which cancer type the test sample belongs to based on the methylation data when the mutation site detection module or the methylation detection module outputs that the test sample is of a cancer type.

[0123] After the data processing module obtains the mutation site data and methylation data of each sample, the first cancer type judgment is performed by the mutation site detection module. If the cancer type is not detected by the mutation site detection module, the second cancer type judgment is performed by the Zscore value detection model according to the first judgment condition. If the cancer type is not detected, the third cancer type judgment is performed by the Zscore value detection model according to the second judgment condition. If the cancer type is not detected, the fourth cancer type judgment is performed by the random forest auxiliary model. If the cancer type is not detected, the fifth cancer type judgment is performed by the neural network model. After the cancer type is detected, the cancer type detection module outputs which cancer type the sample belongs to.

[0124] Example Three

[0125] In this embodiment, a male analysis model and a female analysis model of joint bioinformatics of methylation and mutation sites related to human tumors are constructed.

[0126] The training set of the male analysis model includes 1 bladder cancer (BLCA), 16 colorectal cancer (COADREAD), 36 esophageal cancer (ESCA), 19 liver cancer (LIHC), 69 lung cancer (LUNG), 2 pancreatic cancer (PAAD), 18 gastric cancer (STAD), 3 thyroid cancer (THCA), and 87 healthy samples. The test set includes 12 colorectal cancer (COADREAD), 15 esophageal cancer (ESCA), 8 liver cancer (LIHC), 39 lung cancer (LUNG), 14 gastric cancer (STAD), 1 thyroid cancer (THCA), and 35 healthy samples.

[0127] Based on the analysis method of Example One, the result statistical table of judging whether the sample belongs to a cancer sample or a normal sample in the male analysis model is obtained.

[0128] Table 1: Result statistical table of judging whether the sample belongs to a cancer sample or a normal sample in the male analysis model

[0129]

[0130] According to the data in Table 1, in the male analysis model, the specificity of the training set judging whether the sample belongs to a cancer sample or a normal sample is 100.00%, and the sensitivity is 99.39%. The specificity of the test set judging whether the sample belongs to a cancer sample or a normal sample is 91.43%, and the sensitivity is 89.89%.

[0131] The training set for the female analysis model included 1 bladder cancer (BLCA), 3 breast cancer (BRCA), 41 cervical cancer (CESC), 15 colorectal cancer (COADREAD), 8 esophageal cancer (ESCA), 1 chromophobe renal cell carcinoma (KICH), 8 liver cancer (LIHC), 90 lung cancer (LUNG), 4 ovarian cancer (OV), 11 gastric cancer (STAD), 4 thyroid cancer (THCA), 27 endometrial cancer (UCEC), and 77 healthy samples. The test set included 2 breast cancer (BRCA), 17 cervical cancer (CESC), 1 bile duct cancer (CHOL), 13 colorectal cancer (COADREAD), 3 esophageal cancer (ESCA), 4 liver cancer (LIHC), 40 lung cancer (LUNG), 6 ovarian cancer (OV), 12 gastric cancer (STAD), 1 thyroid cancer (THCA), 16 endometrial cancer (UCEC), and 38 healthy samples.

[0132] Based on the analysis method described in Example 1, Table 2 shows the statistical results of whether the female samples in the above-mentioned female analysis model belong to cancer samples or normal samples.

[0133] Table 2. Statistical table of results for classifying samples as either cancerous or normal in the female analysis model.

[0134]

[0135] As shown in Table 2, in the female analysis model, the specificity of the results in judging whether a sample belongs to cancer or a normal sample in the training set is 100.00%, and the sensitivity is 94.84%; the specificity of the results in judging whether a sample belongs to cancer or a normal sample in the test set is 92.11%, and the sensitivity is 89.57%.

[0136] Specificity refers to the proportion of negative test results among non-cancer individuals detected by the test; specificity = true negative population / non-cancer population. Sensitivity refers to the proportion of positive test results among cancer patients, reflecting the ability to identify positive cases; sensitivity = true positive population / all cancer patients. Ideally, both specificity and sensitivity should be 100%.

[0137] As shown in Tables 1 and 2, the combined analysis of methylation and mutation sites resulted in better specificity and sensitivity of the test set than the results of single-analysis.

[0138] like Figure 4 As shown, in the cancer type prediction analysis model for men, the probability of liver cancer (100%), lung cancer (100%), esophageal cancer (93.33%), and colorectal cancer (86.67%) being correctly predicted among the top three cancer types in terms of prediction probability is greater than 80%.

[0139] likeFigure 5 As shown in the female cancer type prediction analysis model, the probabilities of cervical cancer (100%), cholangiocarcinoma (100%), lung cancer (100%), and endometrial cancer (87.50%) in the test set being correctly predicted in the top three predicted cancer types are all greater than 80%.

[0140] Example Four

[0141] This example sample 1 is from a 63-year-old male patient diagnosed with stage 1 lung cancer. The probability of belonging to a cancer sample is 98.88% by methylation level; in cancer type prediction, the top three cancer types by probability are lung cancer (43.20%), esophageal cancer (17.00%), and gastric cancer (15.00%). The mutation site with the highest Omega value is PTEN:NM_000314.8:exon5:c.318del:p.D107Ifs*6, with a mutation frequency of 0.0052, which can produce multiple forms of C-terminally truncated PTEN protein. Truncation mutations close to the N-terminus result in loss of PTEN phosphatase function, which cannot negatively regulate the activity of the PI3K / AKT pathway (PMID: 11237521). Expression of PTEN truncation mutations in mouse embryonic fibroblasts indicates that these mutations are oncogenic, and PTEN cannot bind to centromeric chromatin compared to the full-length protein, thereby increasing the vulnerability of the genome (PMID: 17218262).

[0142] Table 3 Actual sample mutation site detection results

[0143]

[0144] This example sample 2 is from a 50-year-old female patient diagnosed with stage 1 lung cancer. The probability of belonging to a cancer sample is 95.74% by methylation level; in cancer type prediction, the top three cancer types by probability are lung cancer (31.20%), colorectal cancer (24.40%), and esophageal cancer (6.60%). The mutation site with the highest Omega value is CDKN2A:NM_000077.5:c.157A>G:p.Met53Val, with a mutation frequency of 0.0092, which is located in the ankyrin repeat of the p16 / INK4A protein. This mutation appears as a germline variation in melanoma families (PMID: 15146471). Expression of this mutation in an RB-negative osteosarcoma cell line and a melanoma cell line showed that, compared to the wild type, this mutation can be inactivated, exhibiting reduced binding to CDK4 and increased cell proliferation (PMID: 20340136, 19260062).

[0145] Table 4 Actual sample mutation site detection results

[0146]

[0147] The above merely provides preferred embodiments of the application, and is not used to limit the application. Although the application has been described in detail with reference to the foregoing embodiments, the technical solutions recorded in the foregoing embodiments can still be modified or some technical features can be replaced by equivalent features by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method of tumor-specific gene mutation and methylation combined bioinformatics analysis, characterized in that, The method comprises: Obtaining each mutation site data and methylation data of the sample to be tested; Based on the reference threshold of the mutation site detection model, the omega value of each mutation site in the sample to be tested is calculated, and the first judgment is made according to the overall omega value; if the overall omega value is greater than 1, it is determined that the sample to be tested is cancerous; if the overall omega value is not greater than 1, the second judgment is made based on condition one on the methylation data Zscore value; The omega value of each mutation site represents the closeness of the mutation frequency of the mutation site to the cancer sample; the overall omega value of the sample to be tested is the sum of the omega values of each mutation site in the sample to be tested; the process of calculating the omega value of each mutation site is as follows: If the mutation frequency of the mutation site is greater than or equal to 0.40, then omega = 1; If the mutation frequency of the mutation site is less than the reference threshold: if the reference threshold is less than 0.01, then omega = 99.999*AF; if the reference threshold is greater than or equal to 0.01, then omega = 50*AF; If the mutation frequency of the mutation site is greater than the reference threshold: if the reference threshold is less than 0.01 and the mutation frequency is greater than 0.005, then omega = 99.999*AF+1; if the reference threshold is less than 0.01 and the mutation frequency is less than or equal to 0.005, then omega = 99.999*AF; if the reference threshold is greater than or equal to 0.01 and the mutation frequency is greater than 0.005, then omega = 50*AF+1; if the reference threshold is greater than or equal to 0.01 and the mutation frequency is less than or equal to 0.005, then omega = 50*AF; The AF value is used to represent the frequency of each allele at a certain mutation site, and the AF value is between 0 and 1, 0 represents that the frequency of the allele in all alleles is 0, and 1 represents that the frequency of the allele in all alleles is 1; The condition one is that the proportion of methylation data Zscore value greater than 3 is greater than 7%, or the maximum value is greater than 10, or the average value is greater than 0.4; if condition one is met, it is determined that the sample to be tested is cancerous; if it is not met, the third judgment is made based on condition two on the methylation data Zscore value; The condition two is that the proportion of methylation data Zscore value greater than 1 is less than 3%, or the maximum value is less than 1.80, or the average value is less than -1.05; if condition two is met, it is determined that the sample to be tested is healthy; if it is not met, the fourth judgment is made according to the random forest auxiliary model constructed based on the methylation data; If the probability of the random forest auxiliary model judging that the sample to be tested belongs to the healthy sample is less than the threshold value, the fifth judgment is made according to the neural network model constructed based on the methylation data to determine whether the sample to be tested is cancerous; After determining that the sample to be tested is cancerous, the cancer type is determined according to the random forest model for cancer type prediction, and the probability value of the sample to be tested belonging to different cancer types is output. 2.The method of claim 1, wherein the method is characterized by, A corresponding mutation site detection model is constructed by setting a training set and integrating the mutation frequency of the corresponding mutation site data in the training set, the training set includes cancerous samples and healthy samples, the mutation frequency of each mutation site in the mutation site detection model has a corresponding reference threshold, and the reference threshold ranges from 0 to 0.

02. 3.The method of claim 2, wherein the method is characterized by, The determination process of the reference threshold of the mutation frequency of each mutation site is as follows: Circulating each mutation site of each sample in the training set, and dividing the training set into corresponding cancer sample mutation frequency data set and health sample mutation frequency data set according to the mutation site; Setting a reference mutation frequency and calculating the proportion of samples in the health sample mutation frequency data set corresponding to each mutation site whose mutation frequency is greater than the reference mutation frequency, and determining whether each mutation site is credible according to the sample proportion value; On the basis of the credibility of the mutation site, whether there is a difference between the cancer sample mutation frequency data set and the health sample mutation frequency data set is determined by T test; In the case of no difference: the reference threshold is determined according to the maximum value of the mutation frequency in the health sample mutation frequency data set and the 0.99 quantile; In the case of difference or inability to judge the difference: if the maximum value of the mutation frequency in the health sample mutation frequency data set is not greater than the reference mutation frequency, the reference threshold of the mutation site is set to the maximum value of the health sample mutation frequency in the training set; if the maximum value of the mutation frequency in the health sample mutation frequency data set is greater than the reference mutation frequency, the reference threshold of the mutation site is set to the reference mutation frequency. 4.The method of claim 3, wherein the method is characterized by, In the case of no difference between the cancer sample mutation frequency data set and the health sample mutation frequency data set corresponding to the mutation site, the reference threshold of the mutation site is determined as follows: If the maximum value of the mutation frequency in the health sample mutation frequency data set of the mutation site is greater than or equal to 0.40, the reference threshold of the mutation site is set to 0.01; If the maximum value of the mutation frequency in the health sample mutation frequency data set of the mutation site is less than 0.01 and 1.30 times of the 0.99 quantile is also less than 0.01, the reference threshold of the mutation site is set to 0.01; If the maximum value of the mutation frequency in the health sample mutation frequency data set of the mutation site is greater than 0.01 and less than 0.40, and 1.30 times of the 0.99 quantile is also greater than 0.01, at this time, if the maximum value is greater than 1.30 times of the 0.99 quantile, the reference threshold of the mutation site is set to the maximum value of the health sample mutation frequency in the training set, if the maximum value is less than or equal to 1.30 times of the 0.99 quantile, the reference threshold of the mutation site is set to 1.30 times of the 0.99 quantile of the health sample mutation frequency in the training set; If the maximum value of the mutation frequency in the health sample mutation frequency data set of the mutation site is greater than 0.01 and less than 0.40, and 1.30 times of the 0.99 quantile is less than or equal to 0.01, the reference threshold of the mutation site is set to the maximum value of the health sample mutation frequency in the training set; If the maximum value of the mutation frequency in the health sample mutation frequency data set of the mutation site is less than or equal to 0.01 and 1.30 times of the 0.99 quantile is greater than 0.01, the reference threshold of the mutation site is set to 1.30 times of the 0.99 quantile of the health sample mutation frequency in the training set. 5.The method of claim 1, wherein the method is characterized by, In the random forest auxiliary model constructed based on the methylation data, if the probability of the to-be-tested sample belonging to the healthy sample is not less than 65%, the type of the to-be-tested sample is determined as the healthy type; if the probability of the to-be-tested sample belonging to the healthy sample is less than 65%, the fifth determination is performed according to the neural network model constructed based on the methylation data. 6.The method of claim 1, wherein the method is characterized by, In the neural network model constructed based on the methylation data, if the probability value of the to-be-tested sample belonging to the healthy sample is greater than 50%, the type of the to-be-tested sample is determined as the healthy type; if the probability value of the to-be-tested sample belonging to the healthy sample is not greater than 50%, the type of the to-be-tested sample is determined as the cancer type. 7.The method of claim 1, wherein the method is characterized by, The process of calculating the methylation data Zscore value of the to-be-tested sample is as follows: A training set is set, and the training set includes cancer samples and healthy samples, and the cancer samples include samples of different types of cancer populations; The average μ and the standard deviation σ of each methylation site of the healthy samples in the training set are calculated. The Zscore value of the sample to be tested compared with the training set is calculated, and the calculation formula is: Wherein, X represents the methylation level value of the to-be-tested sample corresponding to the methylation site.

Images