A method and device for detecting multi-lesion lung cancer

Abstract

The present invention discloses a method and device for detecting multi-focal lung cancer, which obtains all variant sites by obtaining transcriptome data of multi-focal lung cancer samples, and then performs preliminary filtering and deep filtering; based on the variant sites of dual primary lung cancer samples and metastatic lung cancer samples, all variant sites are screened for 2 rounds using hypothesis test parameters, variance analysis functions and correlation functions to obtain core variant sites; a machine learning algorithm is constructed using the core variant sites, and the machine learning algorithm is trained using the transcriptome data to obtain an optimized machine learning algorithm for detecting multi-focal lung cancer, which is used to diagnose whether the multi-focal lung cancer sample to be detected is dual primary lung cancer or metastatic lung cancer. The present invention can continuously update the core variant sites, and continuously train the machine learning algorithm for detecting multi-focal lung cancer, which can more accurately determine whether the multi-focal lung cancer sample is dual primary lung cancer or metastatic lung cancer, and help clinical diagnosis and treatment.

Description

Technical Field

[0001] The present invention relates to the field of medical diagnosis, and in particular to a method and device for detecting multi-focal lung cancer. Background Art

[0002] Epidemiological surveys in China show that lung cancer ranks first in both incidence and mortality among various tumors. The lung is an organ prone to both primary cancer and metastasis. When making a clinical diagnosis of multifocal lung cancer, a key point is to first distinguish between metastasis and multiple primary lung cancer: one is that one or more tumors are primary lesions and the others are metastatic lesions; the other is that the primary lesions are in extrapulmonary organs and the lesions in the lungs are all metastatic tumors. Multiple primary lung cancer (MPLC) refers to two or more primary malignant tumors occurring simultaneously or successively in the lungs of the same patient. Multiple primary lung cancer is a relatively rare form of lung cancer, while double primary lung cancer (DPLC) is a typical representative of MPLC. The clinical treatment and prognosis of the above two situations are completely different: for multifocal lung cancer caused by metastasis, there is basically no chance of surgery and the long-term survival rate is low; while the cases of multiple primary lung cancer are mostly early cancers with a higher 5-year survival rate after surgery.

[0003] In the diagnosis method of multi-lesion lung cancer, the multi-lesion lung cancer diagnosis system (CN 111304330A) proposed by Zhiben Medical Technology (Shanghai) Co., Ltd. determines whether the lesions are derived from multiple primary or metastatic sources based on the carcinogenic driver genes and hotspot mutations of lung cancer. However, it can only make judgments based on the list of 578 oncogenes in the patent, and does not involve updating the oncogene list based on multi-lesion lung cancer samples.

[0004] In order to solve the above problems, the present invention focuses on determining whether multifocal lung cancer is dual primary lung cancer or metastatic lung cancer, and proposes a multifocal lung cancer detection method and device. Summary of the invention

[0005] The present invention provides a method for detecting multi-lesion lung cancer, characterized in that it comprises the following steps:

[0006] S1: Obtain transcriptome data of multi-lesion lung cancer samples, including transcriptome data of samples that have been identified as dual primary lung cancer and metastatic lung cancer and transcriptome data of samples to be identified;

[0007] S2: performing preliminary filtering and deep filtering on the transcriptome data to obtain all variant sites;

[0008] S3: Based on the variant sites of the dual primary lung cancer samples and metastatic lung cancer samples, the hypothesis test parameters and variance analysis function are used to perform the first round of screening on all the variant sites;

[0009] S4: performing a second round of screening on the variant sites after the first round of screening using the correlation function, the hypothesis test parameters and the variance analysis function to obtain core variant sites;

[0010] S5: constructing a machine learning algorithm using the core variant sites, and training the machine learning algorithm using the transcriptome data to obtain an optimized machine learning algorithm for detecting multifocal lung cancer;

[0011] S6: Inputting the mutation site of the multi-focal lung cancer sample to be detected, and using the multi-focal lung cancer detection machine learning algorithm to determine whether the multi-focal lung cancer sample to be detected is dual primary lung cancer or metastatic lung cancer.

[0012] Further, the preliminary filtering includes:

[0013] S201: deleting the adapter sequence in the transcriptome data, and filtering the multifocal lung cancer samples whose mutation sites have a frequency of less than 5% in the population;

[0014] S202: for the transcriptome data after completing step S201, filtering the variant sites in the non-coding region, retaining the variant sites in the coding region, and removing the synonymous variant sites;

[0015] S203: Use a visualization tool for gene sequencing variant sites to assist in filtering operations.

[0016] The depth filtering includes:

[0017] S211: comparing the variant sites that have undergone the preliminary filtering to the reference genome, and filtering sites with single nucleotide variations, insertions and deletions, and structural variations with coverage below 100 layers;

[0018] S212: For the variant sites that have completed the S211 step, filtering the inherited germline sites and filtering the noise; the noise refers to the germline variant sites that have an occurrence rate of 20% or more in white blood cells in the multifocal lung cancer sample.

[0019] The variant sites of the double primary lung cancer samples and the metastatic lung cancer samples are specifically: using a mixed linear model in a fitting regression function, the variant sites belonging to the transcriptome data of the samples that have been distinguished as double primary lung cancer and metastatic lung cancer are extracted from all the variant sites.

[0020] The first round of screening is as follows:

[0021] S31: The hypothesis test parameter selects the P value in statistics, which is the probability of obtaining the same or more extreme result as the sample when the null hypothesis is assumed to be true; let the P value be 1.0e -5 As a screening criterion, the P value was greater than 1.0e -5 The variant sites with a P value less than 1.0e -5 The mutation site;

[0022] S32: performing variance analysis and screening on the retained variant sites using the variance analysis function.

[0023] The second round of screening is as follows:

[0024] S41: using the correlation function to calculate and test the correlation coefficient of the variant sites after the first round of screening;

[0025] S42: Let the P value be 1.0e -10 As a screening criterion, cluster analysis is performed on the variant sites after the correlation function test;

[0026] S43: performing variance analysis and screening on the variant sites after the cluster analysis by using the variance analysis function.

[0027] The machine learning algorithm is at least one of a random forest or a neural network.

[0028] Preferably, the present invention relates to a multi-lesion lung cancer detection device, characterized in that it comprises:

[0029] A data acquisition module is used to acquire transcriptome data of multi-lesion lung cancer samples, including transcriptome data of samples that have been identified as dual primary lung cancer and metastatic lung cancer and transcriptome data of samples to be identified;

[0030] A data filtering module performs preliminary filtering and deep filtering on the transcriptome data of the multi-lesion lung cancer sample to obtain all mutation sites;

[0031] The site screening module performs two rounds of screening on all the variant sites in the dual primary lung cancer samples and metastatic lung cancer samples to obtain the core variant sites;

[0032] A machine learning module, which constructs and trains a machine learning algorithm based on the transcriptome data and the core variant sites to obtain the ability to classify and judge whether the multi-lesion lung cancer sample is a dual primary or a metastasis;

[0033] The input-output module inputs the core variant sites of the sample to be tested and converts the results determined by the machine learning module into a paper report or an electronic report.

[0034] Further, the preliminary filtering includes:

[0035] The adapter sequences in the transcriptome data were deleted, and the multifocal lung cancer samples whose mutation sites had a frequency of less than 5% in the population were filtered out; the mutation sites in the non-coding regions were filtered out, the mutation sites in the coding regions were retained, and the synonymous mutation sites were removed; and a visualization tool for gene sequencing mutation sites was used to assist the filtering operation.

[0036] The depth filtering includes:

[0037] The variant sites that have passed the preliminary filtering are compared with the reference genome to filter out sites with single nucleotide variations, insertions and deletions, and structural variations with less than 100 layers of coverage; further, the genetic germline sites are filtered to filter out noise.

[0038] The first round of screening is as follows:

[0039] The hypothesis test parameter selects the P value in statistics, and the P value is set to 1.0e -5 As a screening criterion, the P value was greater than 1.0e -5 The variant sites with a P value less than 1.0e -5 The variant sites are then subjected to variance analysis and screening by variance analysis function.

[0040] The second round of screening is as follows:

[0041] The correlation function is used to calculate and test the correlation coefficient of the variant sites after the first round of screening; and the P value is set to 1.0e -10 As a screening criterion, cluster analysis is performed on the variant sites after the correlation function test; then variance analysis is performed and screened on the variant sites after the cluster analysis using the variance analysis function.

[0042] The machine learning algorithm is at least one of a random forest or a neural network.

[0043] The present invention obtains transcriptome data of multi-focal lung cancer samples, and then performs preliminary filtering and deep filtering to obtain all variant sites; based on the variant sites of double primary lung cancer samples and metastatic lung cancer samples, all the variant sites are screened for 2 rounds using hypothesis test parameters, variance analysis functions and correlation functions to obtain core variant sites; a machine learning algorithm is then constructed using the core variant sites, and the machine learning algorithm is trained using the transcriptome data to obtain an optimized machine learning algorithm for detecting multi-focal lung cancer, which is used to diagnose whether the multi-focal lung cancer sample to be detected is double primary lung cancer or metastatic lung cancer.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The present invention can obtain transcriptome data of multi-focal lung cancer samples from different medical databases in batches, and continuously update the existing core gene mutation sites through multiple rounds of filtering and screening to obtain core mutation sites with stronger correlation with multi-focal lung cancer;

[0046] 2. After obtaining the transcriptome data of multi-lesion lung cancer samples, the present invention obtains fewer core gene mutation sites through multiple rounds of filtering and screening than the number of 578 oncogenes in the multi-lesion lung cancer diagnosis system (CN 111304330A), effectively reducing the workload of gene testing;

[0047] 3. The present invention combines a machine learning algorithm, and can continuously improve the accuracy of judging whether a multi-lesion lung cancer sample is a dual primary lung cancer or a metastatic lung cancer through training by continuously updating the transcriptome data of the multi-lesion lung cancer sample and the transcriptome data of the clinical sample. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0049] Figure 1 It is a schematic diagram of a process flow of a multi-lesion lung cancer detection method according to the present invention. DETAILED DESCRIPTION

[0050] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0051] like Figure 1 As shown, a multi-lesion lung cancer detection method proposed by the present invention is characterized in that it comprises the following steps:

[0052] S1: Obtain transcriptome data of multi-lesion lung cancer samples, including transcriptome data of samples that have been identified as dual primary lung cancer and metastatic lung cancer and transcriptome data of samples to be identified;

[0053] S2: performing preliminary filtering and deep filtering on the transcriptome data to obtain all variant sites;

[0054] The preliminary filtering includes:

[0055] S201: deleting the adapter sequence in the transcriptome data, and filtering the multifocal lung cancer samples whose mutation sites have a frequency of less than 5% in the population;

[0056] S202: for the transcriptome data after completing step S201, filtering the variant sites in the non-coding region, retaining the variant sites in the coding region, and removing the synonymous variant sites;

[0057] S203: Use a visualization tool for gene sequencing variant sites to assist in filtering operations.

[0058] The visualization tool for gene sequencing variation sites can be selected from patents related to visualization of gene sequencing variation sites, or visualization software for gene sequencing variation sites.

[0059] The depth filtering includes:

[0060] S211: comparing the variant sites that have undergone the preliminary filtering to the reference genome, and filtering sites with single nucleotide variations, insertions and deletions, and structural variations with coverage below 100 layers;

[0061] S212: For the variant sites that have completed the S211 step, filtering the inherited germline sites and filtering the noise; the noise refers to the germline variant sites that have an occurrence rate of 20% or more in white blood cells in the multifocal lung cancer sample.

[0062] S3: using a mixed linear model in a fitted regression function, extracting variant sites belonging to the transcriptome data of the samples that have been identified as double primary lung cancer and metastatic lung cancer from all the variant sites; then, based on the variant sites of the double primary lung cancer samples and metastatic lung cancer samples, performing a first round of screening on all the variant sites using hypothesis test parameters and variance analysis functions;

[0063] The first round of screening is as follows:

[0064] S31: The hypothesis test parameter selects the P value in statistics, which is the probability of obtaining the same or more extreme result as the sample when the null hypothesis is assumed to be true; let the P value be 1.0e -5 As a screening criterion, the P value was greater than 1.0e -5 The variant sites with a P value less than 1.0e -5 of the mutation sites.

[0065] S32: performing variance analysis and screening on the retained variant sites using the variance analysis function.

[0066] S4: performing a second round of screening on the variant sites after the first round of screening using the correlation function, the hypothesis test parameters and the variance analysis function to obtain core variant sites;

[0067] The second round of screening is as follows:

[0068] S41: using the correlation function to calculate and test the correlation coefficient of the variant sites after the first round of screening;

[0069] S42: Let the P value be 1.0e -10 As a screening criterion, cluster analysis is performed on the variant sites after the correlation function test;

[0070] S43: performing variance analysis and screening on the variant sites after the cluster analysis by using the variance analysis function.

[0071] S5: constructing a machine learning algorithm using the core variant sites, and training the machine learning algorithm using the transcriptome data to obtain an optimized machine learning algorithm for detecting multifocal lung cancer; further, the machine learning algorithm is at least one of a random forest or a neural network.

[0072] Random forest is a classification model based on decision trees. It uses the bootstrap resampling method to extract multiple samples from the original sample, models each bootstrap sample with a decision tree algorithm, and then combines the predictions of multiple decision trees to obtain the final prediction result through voting.

[0073] A neural network (NN) is a mathematical model or computational model that mimics the structure and function of a biological neural network. A neural network is an adaptive system that connects a large number of artificial neurons for computation and is used to model complex relationships between input and output. Preferably, the neural network described in the present invention is at least a convolutional neural network or a deep neural network.

[0074] Preferably, the transcriptome data of multifocal lung cancer samples are divided into a training set and a validation set, and the validation set is used to adjust the parameters of the machine learning algorithm.

[0075] S6: Inputting the mutation site of the multi-focal lung cancer sample to be detected, and using the multi-focal lung cancer detection machine learning algorithm to determine whether the multi-focal lung cancer sample to be detected is dual primary lung cancer or metastatic lung cancer.

[0076] Furthermore, in the actual application of the present invention, as the amount of transcriptome data increases, including the actual diagnosis results of the multi-lesion lung cancer samples to be detected, the parameters of the machine learning algorithm are continuously adjusted, thereby obtaining a continuously optimized machine learning algorithm.

[0077] In one embodiment of the present invention, the specific steps include:

[0078] Transcriptome data of 410 patients with multifocal lung cancer were downloaded from the National Center for Biotechnology Information (NCBI), with two transcriptome data for each patient, for a total of 820 transcriptome data; 218 transcriptome data of 109 patients with multifocal lung cancer could be distinguished as dual primary lung cancer or metastatic lung cancer.

[0079] Delete all the adapter sequences in the transcriptome data, and filter the multifocal lung cancer samples whose mutation sites have a frequency of less than 5% in the population; further, filter the mutation sites in the non-coding regions of the remaining 760 transcriptome data, retain the mutation sites in the coding regions, and remove synonymous mutation sites; wherein the number and types of gene mutation detection algorithms can be adjusted according to sample requirements, such as different parameters for the same algorithm.

[0080] Furthermore, the variant sites that have passed the preliminary filtering are aligned with the reference genome using the Bwa (Burrow-Wheeler Aligner) software to filter out sites with single nucleotide variants (SNVs), insertions and deletions (indels), and structural variants (SVs) with less than 100 layers of coverage;

[0081] For the variant sites filtered in the above steps, the Gatk (The Genome Analysis Toolkit) software was used to filter the genetic germline sites and filter out the noise; the noise refers to the germline variant sites with an occurrence rate of 20% or more in multiple white blood cells in the multifocal lung cancer samples; finally, clean system variant sites were obtained, totaling 5,800 variant sites.

[0082] The visualization tool for gene sequencing variation sites may be a visualization method for gene sequencing variation sites (publication number: CN107944228A), or a visualization tool software ggplot2.

[0083] Next, the first round of screening was performed on 5,800 variant sites:

[0084] The mixed linear model of the Im function in R language was used to extract the variant sites in the transcriptome data of samples that have been identified as dual primary lung cancer and metastatic lung cancer from 5,800 variant sites;

[0085] The variant sites of the transcriptome data of the samples that have been identified as dual primary lung cancer and metastatic lung cancer are taken as samples, and the hypothesis test parameter is the P value in statistics. The P value is the probability of obtaining the same or more extreme results as the sample when the null hypothesis is assumed to be true; the P value is set to 1.0e -5 As a screening criterion, the P value was greater than 1.0e-5 The variant sites with a P value less than 1.0e -5 of the mutation sites.

[0086] The retained variant sites were then subjected to variance analysis using the aov function of the R language, and 180 variant sites were screened.

[0087] Preferably, a second round of screening is performed on 180 variant sites:

[0088] The R language cor.test function was used to calculate and test the coefficients of the 180 selected variant sites, and the two were compared with each other; then the coefficients were calculated according to the P value less than 1.0e -10 The screening criteria were used to perform cluster analysis on 180 variant sites; variance analysis was then performed on the variant sites of the cluster analysis using the aov function of R language to screen out a group of variant sites with large variance differences, and finally 28 core variant sites were obtained.

[0089] A random forest model was constructed using the 28 core variant sites, and the 820 transcriptome data were divided into two parts: a training set and a validation set, with allocation ratios of 0.8 and 0.2, respectively. A grid search method was used to automatically adjust the three key parameters of the random forest model: the maximum number of calculated values, the bootstrap parameter, and the maximum number of features, and an optimized random forest model for detecting multi-lesion lung cancer was trained.

[0090] Input multiple variant sites of multifocal lung cancer samples to be tested, and the multifocal lung cancer detection random forest model will judge whether the multifocal lung cancer of the subject is dual primary lung cancer or metastatic lung cancer. After testing, the average accuracy of the model reached 92%.

[0091] In some embodiments, the information of the multi-focal lung cancer sample to be tested also includes one or more of the subject's photo, age, gender, height, weight, eating habits, medication history, psychological condition, time from symptom onset to medical treatment, family history of genetic diseases, smoking frequency and exercise frequency.

[0092] Preferably, the present invention relates to a multi-lesion lung cancer detection device, characterized in that it comprises:

[0093] A data acquisition module is used to acquire transcriptome data of multi-lesion lung cancer samples, including transcriptome data of samples that have been identified as dual primary lung cancer and metastatic lung cancer and transcriptome data of samples to be identified;

[0094] A data filtering module performs preliminary filtering and deep filtering on the transcriptome data of the multi-lesion lung cancer sample to obtain all mutation sites;

[0095] The site screening module performs two rounds of screening on all the variant sites in the dual primary lung cancer samples and metastatic lung cancer samples to obtain the core variant sites;

[0096] A machine learning module, which constructs and trains a machine learning algorithm based on the transcriptome data and the core variant sites to obtain the ability to classify and judge whether the multi-lesion lung cancer sample is a dual primary or a metastasis;

[0097] The input-output module inputs the core variant sites of the sample to be tested and converts the results determined by the machine learning module into a paper report or an electronic report.

[0098] The preliminary filtering includes:

[0099] The adapter sequences in the transcriptome data were deleted, and the multifocal lung cancer samples whose mutation sites had a frequency of less than 5% in the population were filtered out; the mutation sites in the non-coding regions were filtered out, the mutation sites in the coding regions were retained, and the synonymous mutation sites were removed; and a visualization tool for gene sequencing mutation sites was used to assist the filtering operation.

[0100] The depth filtering includes:

[0101] The variant sites that have passed the preliminary filtering are compared with the reference genome to filter out sites with single nucleotide variations, insertions and deletions, and structural variations with less than 100 layers of coverage; further, the genetic germline sites are filtered to filter out noise.

[0102] The first round of screening is as follows:

[0103] The hypothesis test parameter selects the P value in statistics, and the P value is set to 1.0e -5 As a screening criterion, the P value was greater than 1.0e -5 The variant sites with a P value less than 1.0e -5 The variant sites are then subjected to variance analysis and screening by variance analysis function.

[0104] The second round of screening is as follows:

[0105] The correlation function is used to calculate and test the correlation coefficient of the variant sites after the first round of screening; and the P value is set to 1.0e -10 As a screening criterion, cluster analysis is performed on the variant sites after the correlation function test; then variance analysis is performed and screened on the variant sites after the cluster analysis using the variance analysis function.

[0106] The machine learning algorithm is at least one of a random forest or a neural network.

[0107] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. A method for detecting multifocal lung cancer, characterized in that: The following steps are involved: S1: Obtain transcriptome data of multi-lesion lung cancer samples, including transcriptome data of samples that have been identified as dual primary lung cancer and metastatic lung cancer and transcriptome data of samples to be identified; S2: performing preliminary filtering and deep filtering on the transcriptome data to obtain all variant sites; S3: Based on the variant sites of the dual primary lung cancer samples and metastatic lung cancer samples, the hypothesis test parameters and variance analysis function are used to perform the first round of screening on all the variant sites; S4: performing a second round of screening on the variant sites after the first round of screening using the correlation function, the hypothesis test parameters and the variance analysis function to obtain core variant sites; S5: constructing a machine learning algorithm using the core variant sites, and training the machine learning algorithm using the transcriptome data to obtain an optimized machine learning algorithm for detecting multifocal lung cancer; S6: inputting the mutation site of the multi-focal lung cancer sample to be detected, and using the multi-focal lung cancer detection machine learning algorithm to determine whether the multi-focal lung cancer sample to be detected is dual primary lung cancer or metastatic lung cancer; The preliminary filtering includes: S201: deleting the adapter sequence in the transcriptome data, and filtering the multifocal lung cancer samples whose mutation sites have a frequency of less than 5% in the population; S202: for the transcriptome data after completing step S201, filtering the variant sites in the non-coding region, retaining the variant sites in the coding region, and removing the synonymous variant sites; S203: Use visualization tools of gene sequencing variant sites to assist filtering operations; The depth filtering includes: S211: comparing the variant sites that have undergone the preliminary filtering to the reference genome, and filtering sites with single nucleotide variations, insertions and deletions, and structural variations with coverage below 100 layers; S212: For the variant sites that have completed the step S211, filtering the inherited germline sites and filtering the noise; The first round of screening is as follows: S31: The hypothesis test parameter selects the P value in statistics, and the P value is set to 1.0e -5 As a screening criterion, the P value was greater than 1.0e -5 The variant sites with a P value less than 1.0e -5 The mutation site; S32: performing variance analysis and screening on the retained variant sites by using the variance analysis function; The second round of screening is as follows: S41: using the correlation function to calculate and test the correlation coefficient of the variant sites after the first round of screening; S42: Let the P value be 1.0e -10 As a screening criterion, cluster analysis is performed on the variant sites after the correlation function test; S43: performing variance analysis and screening on the variant sites after the cluster analysis by using the variance analysis function.

2. A method for detecting multifocal lung cancer according to claim 1, characterized in that: The noise refers to germline variation sites with an occurrence rate of 20% or more in white blood cells in the multifocal lung cancer sample.

3. A method for detecting multifocal lung cancer according to claim 1, characterized in that: The mutation sites of the dual primary lung cancer samples and metastatic lung cancer samples are specifically: Using the mixed linear model in the fitting regression function, the variant sites belonging to the transcriptome data of the samples that have been distinguished as dual primary lung cancer and metastatic lung cancer are extracted from all the variant sites.

4. A method for detecting multifocal lung cancer according to claim 1, characterized in that: The machine learning algorithm is at least one of a random forest or a neural network.

5. A multi-lesion lung cancer detection device, characterized in that: include: A data acquisition module is used to acquire transcriptome data of multi-lesion lung cancer samples, including transcriptome data of samples that have been identified as dual primary lung cancer and metastatic lung cancer and transcriptome data of samples to be identified; A data filtering module performs preliminary filtering and deep filtering on the transcriptome data of the multi-lesion lung cancer sample to obtain all mutation sites; The site screening module performs two rounds of screening on all the variant sites in the dual primary lung cancer samples and metastatic lung cancer samples to obtain the core variant sites; A machine learning module, which constructs and trains a machine learning algorithm based on the transcriptome data and the core variant sites to obtain the ability to classify and judge whether the multi-lesion lung cancer sample is a dual primary or a metastasis; An input-output module, which inputs the core variant sites of the sample to be tested and converts the results determined by the machine learning module into a paper report or an electronic report; Also includes: The preliminary filtering includes: Deleting the adapter sequence in the transcriptome data, filtering the multifocal lung cancer samples whose mutation sites have a frequency of less than 5% in the population; then filtering the mutation sites in the non-coding regions, retaining the mutation sites in the coding regions, and removing the synonymous mutation sites; and using a visualization tool for gene sequencing mutation sites to assist in the filtering operation; The depth filtering includes: The variant sites that have undergone the preliminary filtering are compared with the reference genome to filter out sites with single nucleotide variations, insertions and deletions, and structural variations with coverage below 100 layers; the inherited germline sites are then filtered to filter out noise; The first round of screening is as follows: the hypothesis test parameter selects the P value in statistics, and the P value is set to 1.0e -5 As a screening criterion, the P value was greater than 1.0e -5 The variant sites with a P value less than 1.0e -5 The variant sites are then subjected to variance analysis and screening by using a variance analysis function on the retained variant sites; The second round of screening is as follows: using the correlation function to calculate and test the correlation coefficient of the variant sites after the first round of screening; and setting the P value to 1.0e -10 As a screening criterion, cluster analysis is performed on the variant sites after the correlation function test; then variance analysis is performed and screened on the variant sites after the cluster analysis using the variance analysis function; the machine learning algorithm is at least one of a random forest or a neural network.

Images