A lung cancer early screening detection device based on TCR sequencing and a construction method and application thereof

Abstract

The application discloses a kind of based on TCR sequencing lung cancer early screening model construction method, first based on the TCR sequencing data of lung cancer patient and healthy person sample is constructed TCR sequence database enrichment, then the TCR sequencing of each sample is carried out and obtains the TCR characteristics of each sample, including V gene and J gene The proportion characteristics of gene, the statistical characteristics of TCR immune group, convergence sequence characteristics, TCR sequence amino acid proportion characteristics, different length TCR sequence proportion characteristics, different frequency TCR sequence proportion characteristics and TCR enrichment sequence characteristics;The TCR characteristics of each sample are screened using feature selection method, and a machine learning model is constructed by combining other information of the sample to obtain a lung cancer early screening model.Based on the lung cancer early screening model, a lung cancer early screening detection device is constructed, which can realize lung cancer early screening based on TCR sequencing information of the sample.

Description

Technical Field

[0001] This invention relates to the field of gene detection technology, specifically to a lung cancer early screening detection device based on TCR sequencing, its construction method, and its application. Background Technology

[0002] Lung cancer is the most common malignant tumor in my country. Early detection and appropriate measures are crucial for its prevention and treatment. The 5-year survival rate for stage IA lung cancer can reach over 70%, while the 5-year survival rate for stage IVA lung cancer is only about 10%. However, due to the lack of effective tumor screening methods, the proportion of tumors detected and diagnosed at an early stage is relatively small in current clinical practice, with stage I and II lung cancer accounting for only about 30%. Early detection also helps reduce the costs associated with cancer diagnosis and treatment; the cost of diagnosis and treatment for tumors detected at the metastatic stage is generally 2 to 3 times that of those detected at an early stage.

[0003] Currently, low-dose spiral CT is commonly used in clinical practice to screen for lung cancer in specific populations. However, this method has a high false positive rate, which can easily lead to overdiagnosis and treatment, increasing patient risks and medical burden. It also has radiation issues.

[0004] T cell receptors (TCRs) are specific antibodies on the surface of T cells responsible for recognizing antigens presented by the major histocompatibility complex (MHC), playing a crucial role in recognizing tumor-associated antigens. TCRs are heterodimers that can be composed of αβ or γδ chains, with αβ chains accounting for over 90% of TCRs.

[0005] TCR sequences are generated by rearrangement and insertion of gene fragments from the V gene (variable), D gene (diversity), and J gene (join) regions, exhibiting rich diversity. The complementary-determining region (CDR) of the TCR is used to bind to the antigen-MHC (pMHC) complex; among them, the CDR3 region of the β chain of the TCR is mainly responsible for the specific recognition of epitopes.

[0006] Approximately 10 per individual 11 T cells, composed of 10 6 ~10 8A T-cell immune repertoire consists of a unique TCR sequence; the T-cell immune repertoire refers to the complete TCR sequence of an individual. Analysis of the T-cell immune repertoire can provide information about an individual's immune status, past infection history, and immune response, potentially playing a crucial role in predicting tumor occurrence and development. TCR sequencing refers to the analysis of an individual's T-cell immune repertoire to identify the clonotype and number of TCR sequences. Currently, most TCR sequencing methods primarily focus on sequencing the CDR3 region of the TCR β chain.

[0007] Some studies have used TCR sequencing for early cancer screening, but most of these studies use statistical indicators of TCR immune repertoires and features such as the kmer diversity of TCR sequences to build machine learning models for early cancer screening. Models based on statistical indicators of TCR immune repertoires use too simple features and fail to fully utilize the rich information provided by TCR sequencing, which may make it difficult to capture early tumor-related signals. On the other hand, models based on the kmer diversity of TCR sequences often have too many input features, which can easily lead to overfitting.

[0008] Therefore, there is an urgent need for a method that can utilize TCR sequencing information for early tumor screening, enabling accurate early screening of tumors through the rich information provided by TCR sequencing. Summary of the Invention

[0009] The purpose of this invention is to overcome the above-mentioned shortcomings of the prior art and provide a lung cancer early screening detection device based on TCR sequencing, its construction method and application.

[0010] The primary objective of this invention is to provide a method for constructing an early lung cancer screening model based on TCR sequencing.

[0011] The second objective of this invention is to provide a lung cancer early screening detection device based on TCR sequencing.

[0012] To achieve the above objectives, the present invention is implemented through the following solution:

[0013] A method for constructing a lung cancer early screening model based on TCR sequencing includes the following steps:

[0014] S1. Establishment of an enriched TCR sequence database:

[0015] S11. Obtain the TCR immune repertoire of the experimental group and the TCR immune repertoire of the control group, and calculate the occurrence frequency of each TCR sequence in the TCR immune repertoire of the experimental group and the TCR immune repertoire of the control group respectively; the experimental group contains lung cancer patient samples, and the control group contains healthy human samples;

[0016] S12. Remove TCR sequences that appear ≤5 times from the control group TCR immune database obtained in step S11 to obtain the control group TCR sequence dataset;

[0017] Remove low-abundance TCR sequences from the control group TCR sequence dataset to obtain a control group enriched TCR sequence database; remove TCR sequences that appeared in the control group TCR sequence dataset from the experimental group TCR immune database obtained in step S11, and remove low-abundance TCR sequences to obtain an experimental group enriched TCR sequence database.

[0018] S2. Construction of an early lung cancer screening model:

[0019] A lung cancer early screening model is obtained by cross-validating the machine learning model using features from the training set. The training set includes positive and negative samples. The positive samples are blood and / or tissue samples of malignant pulmonary nodules. The negative samples are blood and / or tissue samples of benign pulmonary nodules, or blood samples from healthy individuals. The features include TCR features, clinical features, and / or imaging features. The TCR features include: the proportion of V and J genes, statistical features of the TCR immune repertoire, convergent sequence features, TCR sequence amino acid proportion features, TCR sequence proportion features of different lengths, TCR sequence proportion features of different frequencies, and TCR enriched sequence features.

[0020] The specific method for obtaining the TCR feature is as follows:

[0021] The TCR sequencing data of each sample in the training set are quality controlled, and then the TCRs that are aligned with the CDR3 region of TCRβ are converted into clonotypes files to obtain the data to be analyzed from the training set samples.

[0022] The proportion of V and J genes: Based on the data to be analyzed from the samples in the training set, the corrected clone counts of each V and J gene in each TCR sequence of the sample are obtained as the proportion of V and J genes in the sample.

[0023] Converging sequence characteristics: Based on the data to be analyzed from the samples in the training set, TCR sequences with 1 to 100 reads or more, and which do not contain stop codons and / or have frameshift mutations, are retained as the functional TCR sequences of the samples, and the characteristics of convergent sequences are calculated; the convergent sequences are TCR sequences with the same amino acid sequence but different nucleotide sequences.

[0024] The characteristics of convergent sequences are: the percentage of convergent sequences, the percentage of convergent sequence clone counts, and the percentage of convergent sequence clone counts after correction.

[0025] TCR enrichment sequence characteristics: (1) For the samples to be analyzed in the training set and the control group in step S1, enrich TCR sequence databases with the same length, retain TCR sequence pairs with edit distance ≤ 0.3×b, where b is the length of the TCR sequence, and select TCR sequence pairs with alignment_score ≥ 4.5×b as TCR sequence combinations;

[0026] Based on TCR sequence combinations, the TCR enriched sequence feature enrich_score1 is calculated according to Formula II, and the TCR enriched sequence feature enrich_type1 is calculated according to Formula III.

[0027] Formula II:

[0028] Among them alignment_score ij The alignment score for TCR sequences i and j of the same length; n ki Let i be the number of TCR sequences i in sample k, and let sample k belong to The samples in Enriching the TCR sequence database for the control group; query_seq_score j =ln(n j ), n j represents the number of TCR sequences j in the TCR sequence combination, where TCR sequence j originates from the sample's data to be analyzed; query_seqs represents the number of TCR sequences in the sample's data to be analyzed.

[0029] Formula III:

[0030] Where `enriched_seqs` represents the number of TCR sequences derived from the sample in the TCR sequence combinations; `query_seqs` represents the number of TCR sequences in the sample's data to be analyzed.

[0031] (2) Following the method shown in (1), based on the TCR sequence database to be analyzed and the experimental group shown in step S1, calculate the TCR enrichment features enrich_score2 and enrich_type2.

[0032] Preferably, the patient mentioned in step S11 is a lung cancer patient.

[0033] Preferably, the experimental group shown in step S11 contains no less than 50 patient samples, and the control group shown contains no less than 50 healthy person samples.

[0034] Preferably, the acquisition of the experimental group TCR immune library and the control group TCR immune library in step S11 specifically involves: performing TCR sequencing, data analysis, and quality control on the experimental group containing patient samples and the control group containing healthy human samples, respectively, to obtain the quality-controlled TCR immune library of each sample in the experimental group and the quality-controlled TCR immune library of each sample in the control group, and merging them to obtain the experimental group TCR immune library and the control group TCR immune library.

[0035] More preferably, the TCR sequencing is performed by constructing a library and sequencing it according to the primers and multiplex PCR amplification method described in the prior art CN114774517A.

[0036] More preferably, the data analysis specifically involves: quality control, followed by alignment with the CDR3 region of TCRβ to obtain the aligned TCR sequence.

[0037] More preferably, the quality control specifically involves: filtering the fastq files from TCR sequencing using a data filtering tool, removing adapters and low-quality sequences, and removing TCR sequences that do not contain TCR sequencing primers.

[0038] More preferably, the data filtering tool is Fastp, FastQC, flash, and / or seqkit.

[0039] More preferably, Fastp is used to filter, remove adapters and remove low-quality sequences based on default parameters, and seqkit is used to remove TCR sequences that do not contain TCR sequencing primers.

[0040] More preferably, the CDR3 region of the TCRβ is the CDR3 region of human TCRβ in the IMGT database.

[0041] More preferably, the CDR3 region of human TCRβ in the IMGT database can be found at: https: / / www.imgt.org / download / V-QUEST / IMGT_V-QUEST_reference_directory / Homo_sapians / TR / TRBV.fasta.

[0042] More preferably, the quality control involves removing TCR sequences with a number of TCR sequences less than 1 to 100.

[0043] More preferably, TCR sequences with a TCR sequence number < 10 are removed.

[0044] More preferably, removing low-abundance TCR sequences from the control group TCR sequence dataset in step S12 specifically involves removing TCR sequences from the control group TCR sequence dataset that appear ≤ n*a times; where n is 0.001 to 0.1 and a is the number of control group samples.

[0045] More preferably, the removal of low-abundance TCR sequences in step S12 specifically involves removing TCR sequences whose occurrence frequency is ≤ n1*b; where n1 is 0.001 to 0.1 and b is the number of samples in the experimental group.

[0046] Preferably, in step S2, k-fold cross-validation and / or leave-one-out crossvalidation are used to perform cross-validation training on the machine learning model.

[0047] More preferably, k = 10 in the k-fold crosses validation.

[0048] Preferably, the machine learning model in step S2 is a random forest, a generalized linear model, a gradient boosting model, a linear support vector machine, and / or a radial basis function kernel support vector machine.

[0049] Preferably, during the training process described in step S2, the expected output of positive samples in the training set is ≥0.5, and the expected output of negative samples is <0.5.

[0050] Preferably, the clinical features in step S2 include whether there is a history of smoking, whether one has quit smoking, whether one has had lung cancer, whether one has COPD, and / or whether one has a family history of lung cancer; the imaging features include the size of the lung nodules, the number of lung nodules, whether they are solid nodules, whether they are partially solid nodules, whether they are ground-glass nodules, whether there are spiculations, and / or the location of the lung nodules.

[0051] Among them, numerical features in clinical and imaging features use their numerical values ​​as feature values, while binary features are converted to 0 / 1 feature values.

[0052] Preferably, before performing cross-training in step S2, the features of the samples in the training set are screened using Ensemble Feature Selection and / or Boruta feature selection methods.

[0053] More preferably, the ensemble feature selection method includes recursive feature elimination, random forest, and / or glmnet.

[0054] More preferably, the Ensemble Feature selection method includes glmnet.

[0055] More preferably, data balancing is performed after filtering the features of the samples in the training set.

[0056] More preferably, data balancing is performed using oversampling and / or undersampling methods.

[0057] More preferably, the SMOTE algorithm is used to balance the features of the samples in the training set, so that the features of the samples in the training set are balanced between positive samples and negative samples.

[0058] Preferably, the TCR sequencing data in step S2 is the TCR sequencing data obtained by constructing a library and sequencing it according to the primers and multiplex PCR amplification method described in the prior art CN114774517A.

[0059] Preferably, the CDR3 region of TCRβ in step S2 is the CDR3 region of human TCRβ in the IMGT database.

[0060] More preferably, the CDR3 region of human TCRβ in the IMGT database can be found at: https: / / www.imgt.org / download / V-QUEST / IMGT_V-QUEST_reference_directory / Homo_sapians / TR / TRBV.fasta.

[0061] Preferably, the corrected clonecounts in step S2 is clonecounts with the value of ln.

[0062] Preferably, the statistical characteristics of the TCR immune repertoire in step S2 include the Shannon index, d50 index, reverse Simpson index, and homogeneity index.

[0063] More preferably, the method for obtaining the TCR immune repertoire is as follows: the TCR sequence of the data to be analyzed from the samples in the training set is used as the TCR immune repertoire of the samples, and the statistical characteristics of the TCR immune repertoire are calculated.

[0064] More preferably, when calculating the statistical characteristics of the TCR immune repertoire, if the number of TCR sequences in the data to be analyzed in the sample is greater than m, m TCR sequences are randomly selected as the TCR immune repertoire of the sample; m = 1E6 to 1E7.

[0065] More preferably, m = 1E6.

[0066] Preferably, when calculating the convergent sequence characteristics in step S2, TCR sequences with a TCR sequence read count ≥ 10 are retained.

[0067] Wherein, the proportion of convergent sequence number (Convergence_type) is the proportion of the number of convergent TCR sequence types in the functional TCR sequence to the total number of TCR sequence types in the functional TCR sequence; the proportion of convergent sequence clone counts (Convergence_freq) is the proportion of the number of convergent sequences in the functional TCR sequence to the total number of TCR sequences in the functional TCR sequence; the proportion of convergent sequence clone counts after correction (Convergence_score) is the proportion of the corrected number of convergent sequences in the functional TCR sequence to the total corrected number of TCR sequences in the functional TCR sequence; the corrected number of convergent sequences in the functional TCR sequence is the ln value of the total number of convergent sequences in the functional TCR sequence; and the corrected number of TCR sequences in the functional TCR sequence is the ln value of the total number of TCR sequences in the functional TCR sequence.

[0068] Preferably, the method for obtaining the amino acid proportion characteristics of the TCR sequence in step S2 is as follows: based on the functional TCR sequence of the sample, remove 1 to 4 amino acids from both ends of each TCR sequence, and calculate the proportion of each amino acid in the TCR sequence after the amino acid deletion.

[0069] More preferably, three amino acids are removed from both ends of each TCR sequence.

[0070] Preferably, the method for obtaining the proportion features of TCR sequences of different lengths and the proportion features of TCR sequences of different frequencies in step S2 is as follows: based on the functional TCR sequences of the samples, they are grouped by length, and the proportion of TCR sequences in each length group is calculated to obtain the proportion of TCR sequences of each length; based on the functional TCR sequences of the samples, they are grouped by frequency, and the proportion of TCR sequences in each frequency group is calculated to obtain the proportion of TCR sequences of each frequency.

[0071] More preferably, when calculating the proportion of TCR sequences of each length, they are grouped according to lengths of ≤10aa, 11-13aa, 14-16aa, 17-19aa, and ≥20aa.

[0072] More preferably, when calculating the proportion of TCR sequences at each frequency, they are grouped according to their occurrence frequency as ≥1E-2, 1E-3~1E-4, 1E-4~1E-5, 1E-5~1E-6, 1E-6~1E-7 and ≤1E-7.

[0073] In step S2, when calculating the TCR enriched sequence features, the TCR sequence combination shown contains multiple TCR sequence pairs. Each TCR sequence pair contains one TCR sequence from the sample data to be analyzed and one TCR sequence from the control group enriched TCR sequence database.

[0074] Preferably, in step S2, the Needleman-Wunsch algorithm is used to calculate the alignment_score.

[0075] More preferably, in the process of calculating alignment_score, the amino acid substitution scoring matrix is ​​Blosum62, the open penalty is -10, and the extend penalty is -1.

[0076] This invention also seeks protection for the lung cancer early screening model constructed by the above-described construction method.

[0077] This invention also claims protection for a lung cancer early screening detection device based on TCR sequencing, including a sequencing module, an analysis module, a prediction module, and a result output module;

[0078] The sequencing module is used to perform TCR sequencing on the sample to be tested and obtain the fastq file in the sequencing output file;

[0079] The analysis module performs the steps shown in step S2 of the above construction method on the fastq file to be tested obtained by the sequencing module to obtain the TCR features of the sample to be tested and to obtain the input features of the sample after screening.

[0080] The prediction module is a lung cancer early screening model constructed by the above construction method. The input features of the sample to be tested obtained by the analysis module after screening are used as input variables and input into the lung cancer early screening model to obtain the lung cancer prediction score of the sample to be tested.

[0081] The result output module is used to output the lung cancer prediction score of the test sample obtained by the prediction module.

[0082] Preferably, in the analysis module, the input features after screening the samples to be tested include: the proportion of TRBV2-6 gene, the proportion of TRBV18 gene, the proportion of TRBV19 gene, the proportion of TRBV2 gene, the proportion of TRBV27 gene, the proportion of TRBV28 gene, the proportion of TRBV3-1 gene, the proportion of TRBV5-1 gene, the proportion of TRBV5-4 gene, the proportion of TRBV5-5 gene, and the proportion of TRBV7-8 gene. Features, TRBV9 gene proportion, glycine proportion, leucine proportion, asparagine proportion, tryptophan proportion, proportion of TCR sequences with frequencies of 1E-5 to 1E-6, TCR enriched sequence characteristics (enrich_score), Shannon index, reverse Simpson index, homogeneity index, proportion of clone counts after convergent sequence correction (Convergence_score), smoking cessation status, smoking history, whether it is a ground-glass nodule, whether it is a solid nodule, presence of spiculation, and size of the lung nodule.

[0083] Among them, the TRBJ2-6 gene (Gene ID: 28623) in the input features of the sample to be tested belongs to the J gene of the TCR sequence; the TRBV18 gene (Gene ID: 28569), TRBV19 gene (Gene ID: 28568), TRBV2 gene (Gene ID: 28620), TRBV27 gene (Gene ID: 28560), TRBV28 gene (Gene ID: 28559), TRBV3-1 gene (Gene ID: 28619), TRBV5-1 gene (Gene ID: 28614), TRBV5-4 gene (Gene ID: 28611), TRBV5-5 gene (Gene ID: 28610), TRBV7-8 gene (Gene ID: 28590) and TRBV9 gene (Gene ID: 28586) belong to the V gene of the TCR sequence.

[0084] Preferably, the lung cancer prediction score is [0, 1], the input result is 0, the probability that the test sample is lung cancer is 0, the output result is 1, and the probability that the test sample is lung cancer is 1.

[0085] More preferably, the cutoff value of the lung cancer prediction score is 0.5. If the output result is ≥0.5, the sample to be tested is determined to be a positive sample (i.e., a malignant lung nodule); if the output result is <0.5, the sample is determined to be a negative sample (i.e., a benign lung nodule or a healthy sample).

[0086] Compared with the prior art, the present invention has the following beneficial effects:

[0087] This invention provides a method for constructing a lung cancer early screening model based on TCR sequencing. First, an enriched TCR sequence database is constructed based on TCR sequencing data from samples from lung cancer patients and healthy individuals. Then, TCR sequencing is performed on each sample to obtain its TCR characteristics, including the proportion of V and J genes, statistical characteristics of the TCR immune repertoire, convergent sequence characteristics, TCR sequence amino acid proportion characteristics, proportion of TCR sequences of different lengths, proportion of TCR sequences of different frequencies, and TCR enriched sequence characteristics. After filtering the TCR characteristics of the samples using a feature selection method, a machine learning model is constructed by combining other information from the samples to obtain the lung cancer early screening model. Based on this model, a lung cancer early screening detection device is constructed, enabling early lung cancer screening based on the TCR sequencing information of the samples. Attached Figure Description

[0088] Figure 1 The flowchart for constructing the early lung cancer screening model shown in Example 3 is as follows;

[0089] Figure 2 The ROC-AUC curves of the five candidate lung cancer early screening models trained in Example 3 are shown.

[0090] Figure 3 The output distribution diagrams for the five candidate lung cancer early screening models trained in Example 3 are shown below; A is the output distribution diagram for the trained random forest, B is the output distribution diagram for the trained gradient boosting model, C is the output distribution diagram for the trained generalized linear model, D is the output distribution diagram for the trained svmLinear, and E is the output distribution diagram for the trained svmRadial.

[0091] Figure 4 The ROC-AUC curve of the lung cancer early screening detection device in Example 4 is shown.

[0092] Figure 5 The following are the output result distribution diagrams of the lung cancer early screening detection device in Example 4: A is the output result distribution diagram of lung cancer early screening detection device 1, B is the output result distribution diagram of lung cancer early screening detection device 2, C is the output result distribution diagram of lung cancer early screening detection device 3, and E is the output result distribution diagram of lung cancer early screening detection device 4. Detailed Implementation

[0093] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. These embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Unless otherwise specified, the experimental methods used in the following embodiments are conventional methods; the materials and reagents used, unless otherwise specified, are commercially available.

[0094] The samples used in the embodiments of this invention were all from Shenzhen Haplos Biotechnology Co., Ltd., and informed consent from patients was obtained before the samples were used for testing.

[0095] Example 1: A method for obtaining and analyzing enriched TCR sequences in lung cancer

[0096] I. Experimental Methods

[0097] 1. TCR sequencing

[0098] DNA was extracted from samples in experimental groups 1, 2, and the control group using a DNA extraction kit. For each sample in each group, a TCR sequencing library was constructed using multiplex PCR amplification with 40 primers specific to the V gene and 12 primers specific to the J gene (primers and multiplex PCR amplification method are as shown in the prior art CN114774517A). The constructed TCR libraries were sequenced using a Novaseq-2000 sequencer, and the TCR sequencing fastq file for each sample in each group was obtained.

[0099] The sample types and sample sizes for experimental group 1, experimental group 2, and the control group are shown in Table 1.

[0100] Table 1. Sample types and sample sizes for experimental group 1, experimental group 2, and control group.

[0101] Group Sample type Sample size control group Healthy Blood 2699 Experimental group 1 Blood from lung cancer patients 3360 Experimental group 2 Lung Cancer Tissue 988

[0102] 2. Data analysis and quality control

[0103] (1) For each sample in each group, the TCR sequencing fastq file was filtered, adapter removed and low-quality sequences were removed using Fastp software with default parameters. The TCR sequences that did not contain TCR sequencing primers (i.e., the specific primers shown in technical CN114774517A) were removed using seqkit software with preset parameters. Then, the mixcr software was used to align the human TCRβ gene sequence to the CDR3 region in the IMGT database (https: / / www.imgt.org / download / V-QUEST / IMGT_V-QUEST_reference_directory / Homo_sapiens / TR / TRBV.fasta). The alignment results were processed using vdjtools to obtain the TCR immune repertoire for each sample in each group.

[0104] The TCR immune repertoire contains the types (clonotypes) and the number (clonecounts) of TCR sequences in the CDR3 region.

[0105] (2) Remove the TCR sequences with clone counts < 10 from the TCR immune library of each sample in each group obtained in step (1) to obtain the quality-controlled TCR immune library of each sample in experimental group 1, experimental group 2 and control group.

[0106] (3) Merge the TCR immune repositories of each sample in experimental group 1 obtained in step (2) to obtain the TCR immune repositories of experimental group 1; merge the TCR immune repositories of each sample in experimental group 2 obtained in step (2) to obtain the TCR immune repositories of experimental group 2; merge the TCR immune repositories of each sample in control group obtained in step (2) to obtain the TCR immune repositories of control group.

[0107] Count the number of times each TCR sequence appears in each TCR immune database (db_seq_count) and calculate the score of each TCR sequence (db_seq_score);

[0108] Where db_seq_score i =∑ k∈S ln(n ki ), n ki Let S represent the number of TCR sequence i in sample k, S represent experimental group 1, experimental group 2 or control group, and k∈S represent sample k from experimental group 1, experimental group 2 or control group.

[0109] (4) Based on the control group TCR immune database obtained in step (3), remove TCR sequences with a count (db_seq_count) ≤ 5 from the control group TCR immune database to obtain the control group TCR sequence dataset; based on the experimental group 1 TCR immune database obtained in step (3), remove TCR sequences that appeared in the control group TCR sequence dataset and remove TCR sequences that appeared ≤ 17 times in the experimental group 1 TCR immune database to obtain the experimental group 1 enriched TCR sequence database; based on the experimental group 2 TCR immune database obtained in step (3), remove TCR sequences that appeared in the control group TCR sequence dataset and remove TCR sequences that appeared ≤ 15 times in the experimental group 2 TCR immune database to obtain the experimental group 2 enriched TCR sequence database.

[0110] TCR sequences with a frequency ≤150 in the control group TCR sequence dataset were removed to obtain the control group enriched TCR sequence database.

[0111] II. Experimental Results

[0112] The number of enriched sequences in the TCR sequence databases of the control group, experimental group 1, and experimental group 2 is shown in Table 2.

[0113] Table 2 Number of enriched sequences

[0114] Group Sample type Enriched TCR sequence number Control group enriched TCR sequence database Healthy Blood 5480 Experimental group 1 enriched TCR sequence database Blood from lung cancer patients 3840 Experimental group 2 enriched TCR sequence database Lung Cancer Tissue 3652

[0115] The TCR sequences in each enriched TCR database were sorted according to their counts. The top 30 enriched sequences in the control group enriched TCR sequence database are shown in Table 3, the top 30 enriched sequences in the experimental group 1 enriched TCR sequence database are shown in Table 4, and the top 30 enriched sequences in the experimental group 2 enriched TCR sequence database are shown in Table 5.

[0116] Table 3. Top 30 enriched sequences from the control group's enriched TCR sequence database.

[0117]

[0118]

[0119] Table 4 shows the top 30 enriched sequences from the TCR sequence database for Experimental Group 1.

[0120]

[0121]

[0122] Table 5. Top 30 enriched sequences from the TCR sequence database of Experimental Group 2

[0123]

[0124] Example 2: Method for extracting TCR features for early lung cancer screening prediction

[0125] Using peripheral blood from the lungs as a sample, TCR sequencing was performed on the sample according to step 1 of Example 1, resulting in a fastq file of the sample's TCR sequencing data. The fastq file of the sample's TCR sequencing data was filtered, adapter removed, and low-quality sequences were removed using Fastp software with default parameters. TCR sequences that did not contain TCR sequencing primers (i.e., the specific primers shown in prior art CN114774517A) were removed using seqkit software with preset parameters. Then, the mixcr software was used to align the human TCRβ gene sequence to the CDR3 region in the IMGT database (https: / / www.imgt.org / download / V-QUEST / IMGT_V-QUEST_reference_directory / Homo_sapiens / TR / TRBV.fasta), and the alignment results were processed using vdjtools to obtain the sample's clonotypes file. The sample's clonotypes file records the types (clonotypes) and the number (clonecounts) of TCR sequences in the sample's CDR3 region.

[0126] Using the clonotypes file of the sample as the data to be analyzed, the TCR features of the sample were extracted, including the proportion of each V gene (variable region gene) and J gene (joining region gene) in the TCR sequence, the statistical features of the TCR immune repertoire, convergent sequence features, amino acid proportion features of the TCR sequence, proportion features of TCR sequences of different lengths, proportion features of TCR sequences of different frequencies, and TCR enriched sequence features, as detailed below:

[0127] 1. Characteristics of the proportions of V gene and J gene:

[0128] Based on the data to be analyzed in the sample, the clone counts of each V gene and J gene in each TCR sequence of the sample are obtained and the ln value is taken to obtain the adjusted clone counts of each V gene and J gene in the sample, which are used as the proportion features of V gene and J gene in the sample.

[0129] 2. Statistical characteristics of the TCR immune repertoire:

[0130] Based on the data to be analyzed from the sample, 10 were randomly selected. 6The TCR sequences of the samples (all sequences are selected if the number is insufficient) are used as the TCR immune repertoire of the samples. Based on the TCR immune repertoire of the samples, the statistical characteristics of the TCR immune repertoire of the samples are calculated using vdjtools, including Shannon index (formula IV), d50 index (TCR diversity index), reverse Simpson index (formula V) and homogeneity index (formula VI).

[0131] Formula IV:

[0132] Formula V:

[0133] Formula VI:

[0134] in denoted as the set of TCR sequences in the sample's TCR immune repertoire; C represents the number of different types of TCR sequences in the sample's TCR immune repertoire; p i The ratio of the number of TCR sequences i in the sample's TCR immune repertoire to the total number of TCR sequences in the sample's TCR immune repertoire. n i Let i be the number of TCR sequences. This represents the number of TCR sequences in the TCR immune repertoire of the sample.

[0135] 3. Characteristics of convergent sequences:

[0136] Non-functional TCR sequences (TCR sequences containing stop codons or exhibiting frameshift mutations) are removed from the data to be analyzed in the sample, and TCR sequences with <10 reads are also removed to obtain the functional TCR sequences of the sample. The characteristics of convergent sequences are then calculated based on the functional TCR sequences.

[0137] Convergent sequences are TCR sequences that have the same amino acid sequence but different nucleotide sequences;

[0138] The characteristics of convergent sequences include: the percentage of convergent sequences (Convergence_type), the percentage of convergent sequence clone counts (Convergence_freq), and the percentage of convergent sequence clone counts after correction (Convergence_score).

[0139] Where Convergence_type is the proportion of the number of convergent TCR sequence types in the functional TCR sequence to the total number of TCR sequence types in the functional TCR sequence;

[0140] Convergence_freq is the proportion of convergent sequences in a functional TCR sequence to the total number of TCR sequences in a functional TCR sequence;

[0141] Convergence_score is the proportion of the corrected number of converging sequences in a functional TCR sequence to the corrected number of TCR sequences in a functional TCR sequence.

[0142] The corrected number of convergent sequences in a functional TCR sequence is the same as the number of convergent sequences in the functional TCR sequence, expressed as ln.

[0143] 4. Characteristics of amino acid proportions in the TCR sequence:

[0144] Delete three amino acids from both ends (5' and 3' ends) of each TCR sequence obtained in step 3 to obtain the TCR sequence of the sample after amino acid deletion, and calculate the proportion of each amino acid in the TCR sequence of the sample after amino acid deletion.

[0145] 5. Characteristics of the proportion of TCR sequences of different lengths and the proportion of TCR sequences of different frequencies:

[0146] For the functional TCR sequences obtained in step 3, they are grouped according to Table 6 based on the length of the TCR sequences and according to Table 7 based on the frequency of the TCR sequences. The proportion of the number of TCR sequences in each group to the total number of TCR sequences in the functional TCR sequences is calculated for different length groups and different frequency groups. This yields the proportion of TCR sequences of different lengths and the proportion of TCR sequences of different frequencies.

[0147] Table 6. TCR sequence length grouping

[0148] TCR sequence length ≤10 11～13 14～16 17～19 ≥20 Group A B C D E

[0149] Table 7 TCR sequence frequency grouping

[0150] Frequency of TCR sequence occurrence ≥1E-2 1E-3~1E-4 1E-4~1E-5 1E-5~1E-6 1E-6~1E-7 ≤1E-7 Group 1 2 3 4 5 6

[0151] 6. TCR enrichment sequence characteristics:

[0152] Each TCR sequence in the sample's data to be analyzed is compared with the TCR sequences in the control group enriched TCR sequence database obtained in Example 1 for sequence similarity. For TCR sequences in the sample's data to be analyzed and TCR sequences in the control group enriched TCR sequence database with a length of b (b>0), TCR sequence pairs are obtained. Each TCR sequence pair contains one TCR sequence from the sample's data to be analyzed and one TCR sequence from the control group enriched TCR sequence database, and the TCR sequences in the TCR sequence pair satisfy the edit distance ≤ 0.3*b. The edit distance is the minimum number of editing operations required to convert one TCR sequence into another.

[0153] For each TCR sequence pair, the alignment_score is calculated using the Needleman-Wunsch algorithm, and TCR sequence pairs with alignment_score ≥ 4.5 × b are used to obtain the TCR sequence combination;

[0154] In the process of calculating alignment_score, the amino acid substitution scoring matrix is ​​Blosum62, the open penalty is -10, and the extend penalty is -1.

[0155] Based on the TCR sequence combination, the TCR enriched sequence feature enrich_score1 of the sample is calculated according to Formula II, and the TCR enriched sequence feature enrich_type1 of the sample is calculated according to Formula III.

[0156] Formula II:

[0157] Among them alignment_score ij The alignment score for TCR sequences i and j of the same length; n ki Let be the number of TCR sequences i in sample k of the TCR sequence combination, and let sample k belong to . The samples in To enrich the TCR sequence database; query_seq_score j =ln(n j ), n j `query_seqs` represents the number of TCR sequences j derived from the sample's data to be analyzed within the TCR sequence set; `query_seqs` represents the number of TCR sequences in the sample's data to be analyzed.

[0158] Formula III:

[0159] Where enriched_seqs is the number of TCR sequences from the sample's data to be analyzed in the TCR sequence combination; query_seqs is the number of TCR sequences in the sample's data to be analyzed.

[0160] Each TCR sequence in the sample's data to be analyzed is compared with the TCR sequence in the enriched TCR sequence database of experimental group 1 obtained in Example 1 according to the method shown above, and the TCR enriched sequence feature enrich_score2 and TCR enriched sequence feature enrich_type2 of the sample are calculated.

[0161] Each TCR sequence in the sample's data to be analyzed is compared with the TCR sequence in the enriched TCR sequence database of experimental group 2 obtained in Example 1 according to the method shown above, and the TCR enriched sequence feature enrich_score3 and TCR enriched sequence feature enrich_type3 of the sample are calculated.

[0162] Example 3: A lung cancer early screening model based on TCR sequencing

[0163] I. Model Training

[0164] A schematic diagram illustrating the construction process of an early lung cancer screening model is shown below. Figure 1 As shown.

[0165] 1. Composition of training and test sets

[0166] The sample set containing 687 malignant pulmonary nodule samples (positive samples) and 202 benign pulmonary nodule samples (negative samples) was used as the test sample. The test sample was divided into training set and test set in a ratio of 85:15, as shown in Table 8. There was no significant difference in the proportion of benign and malignant pulmonary nodule samples between the training set and the test set during the division process.

[0167] Table 8 Samples of Training and Test Sets

[0168] Types of lung nodules training set test set total Malignant (positive) 583 104 687 Benign (negative) 172 30 202

[0169] Meanwhile, each sample in the training and testing sets also contains the clinical and imaging characteristics of each sample.

[0170] Clinical features include: whether there is a history of smoking, whether smoking has been quit, whether there has been a history of lung cancer, whether there is a family history of lung cancer, and whether there is COPD (chronic obstructive pulmonary disease);

[0171] Imaging features include: size of lung nodules, number of lung nodules, whether they are solid nodules, whether they are partially solid nodules, whether they are ground-glass nodules, presence of spiculations, and location of lung nodules.

[0172] 2. Feature selection for training the model

[0173] (1) Extraction of TCR features for each sample

[0174] The samples in the training set and the test set were processed according to the method shown in Example 2, and the TCR features of the samples in the training set and the test set were obtained respectively (the proportion of each V gene and J gene in the TCR sequence, the statistical features of the TCR immune repertoire, the convergent sequence features, the amino acid proportion features of the TCR sequence, the proportion features of TCR sequences of different lengths, the proportion features of TCR sequences of different frequencies, and the TCR enriched sequence features).

[0175] (2) Feature Filtering

[0176] The TCR features, clinical features, and imaging features of the samples in the training set are combined and used as the input features of the samples in the training set. Among them, the numerical features in the clinical features and imaging features of the samples are used as their corresponding feature values, and the binary type features (yes / no, present / absent) are converted into 1 / 0 inputs, where 1 represents yes / present and 0 represents no / absent.

[0177] Using the positive and negative results of the training set samples as the prediction target, feature selection is performed as follows:

[0178] The caret software package is used to perform ensemble feature selection on the input features of the training set samples to obtain the feature importance score of each input feature of the training set samples. Based on the feature importance score of each input feature of the training set samples, the 30 input features with the highest feature importance scores are selected as the ensemble features of the samples.

[0179] Ensemble Feature Selection includes three feature selection methods: recursive feature elimination, random forest, and glmnet. The feature importance score is obtained by summing the scores of each input feature of the sample obtained by each feature selection method after normalization.

[0180] The R boruta package is used to perform boruta feature filtering on the input features of the samples in the training set, resulting in boruta-determined features and boruta-undetermined features of the samples.

[0181] The ensemble features of the samples, the boruta-determined features of the samples, and the boruta-undetermined features of the samples are merged to obtain the types of input features after filtering, as shown in Table 9, and the input features after filtering of the samples in the training set are obtained.

[0182] Table 9. Types of input features after filtering

[0183]

[0184]

[0185] Based on the types of filtered input features shown in Table 9, the filtered input features for each sample in the test set are obtained.

[0186] 3. Training of lung cancer early screening models

[0187] The SMOTE algorithm is used to oversample the input features of the training set samples obtained in step 2 until the input features of the malignant lung nodule samples (positive samples) and the input features of the benign lung nodule samples (negative samples) in the training set are balanced, thus obtaining the input features of the training set samples after data balance.

[0188] After data balancing, the number of benign lung nodule samples in the training set was 408, reaching 70% of the number of malignant lung nodule samples.

[0189] Random Forest (RF), Generalized Linear Model (GLMNNET), Gradient Boosting Model (GBM), Linear Support Vector Machine (SVMLinear), and Radial Basis Function Kernel Support Vector Machine (SVMRadial) were used as candidate lung cancer early screening models. Ten-fold cross-validation was employed, with the input features of the filtered samples in the balanced training set as input variables, which were then fed into each of the five candidate models for training. The hyperparameters of the candidate models were optimized using the grid search hyperparameter optimization method until the output results of the candidate models showed no significant difference from the actual lung nodule types (positive and negative results) in the training set. This resulted in five well-trained candidate lung cancer early screening models: a trained Random Forest, a trained Generalized Linear Model, a trained Gradient Boosting Model, a trained Linear Support Vector Machine, and a trained Radial Basis Function Kernel Support Vector Machine.

[0190] During the training process, the expected output of malignant pulmonary nodule samples in the candidate lung cancer early screening model is ≥0.5, and the expected output of benign pulmonary nodule samples in the candidate lung cancer early screening model is <0.5.

[0191] The output of the candidate lung cancer early screening model is [0, 1]. The input is 0, the probability that the sample is lung cancer is 0, and the output is 1, the probability that the sample is lung cancer is 1.

[0192] If the cutoff value is 0.5 and the output result is ≥0.5, the sample is judged as a positive sample (i.e., a malignant lung nodule); if the output result is <0.5, the sample is judged as a negative sample (i.e., a benign lung nodule).

[0193] The filtered input features of each sample in the test set obtained in step 2 are used as input variables and input into the five trained candidate lung cancer early screening models respectively. The output results of the five trained candidate lung cancer early screening models are obtained respectively. Based on the actual lung nodule type (positive or negative result) of each sample in the test set, the indicators (sensitivity, specificity, post pred value, negative pred value, F1 value) of each trained lung cancer early screening model are calculated according to formulas IV to VIII.

[0194] Formula VI: Sensitivity = TP / (TP + FN);

[0195] Formula V: Specificity = TN / (TN + FP);

[0196] Formula VI: Pos Pred Value = TP / (TP + FP);

[0197] Formula VII: Neg Pred Value=TN / (TN+FN);

[0198] Formula VIII: F1 = 2TP / (2TP + FP + FN);

[0199] TP represents true positive, FP represents false positive, TN represents true negative, and FN represents false negative.

[0200] The AUC value was calculated based on the output results of the five trained candidate lung cancer early screening models, and the ROC-AUC curve and output result distribution were plotted.

[0201] II. Experimental Results

[0202] The index results of each trained candidate lung cancer early screening model are shown in Table 10, and the AUC values ​​are shown in Table 11.

[0203] Table 10: Indicator Results of Each Trained Candidate Lung Cancer Early Screening Model

[0204] Model randomForest gbm glmnet svmLinear svmRadial Sensitivity 0.88 0.89 0.89 0.88 0.86 Specificity 0.60 0.60 0.70 0.57 0.63 Pos Pred Value 0.88 0.89 0.91 0.88 0.89 Neg Pred Value 0.60 0.62 0.66 0.57 0.56 F1 0.88 0.89 0.90 0.88 0.87

[0205] Table 11 AUC values ​​of each trained candidate lung cancer early screening model

[0206] Model randomForest gbm glmnet svmLinear svmRadial AUC 0.82 0.86 0.83 0.82 0.84

[0207] The ROC-AUC curves of the five trained candidate lung cancer early screening models are shown below. Figure 2 As shown in the figure; the output distribution of the five trained candidate lung cancer early screening models is shown in the figure. Figure 3 As shown, A is the output distribution of the trained random forest, B is the output distribution of the trained gradient boosting model, C is the output distribution of the trained generalized linear model, D is the output distribution of the trained svmLinear, and E is the output distribution of the trained svmRadial.

[0208] The results showed that among the five trained lung cancer early screening models, the trained generalized linear model (glmnet) had the highest F1 score (≥0.90), indicating that the trained generalized linear model can more accurately detect samples and more accurately classify samples as benign or malignant lung nodules. Furthermore, the ROC-AUC curves of the five trained lung cancer early screening models showed that the AUC value of the trained generalized linear model (glmnet) was 0.83 > 0.8, indicating that the model's prediction results have excellent reliability.

[0209] Figure 3 The results show that the p-value of the output distribution of the trained random forest is 7.33E-0.8, the p-value of the output distribution of the trained gradient boosting model is 2.66E-09, the p-value of the output distribution of the trained generalized linear model is 3.73E-08, the p-value of the output distribution of the trained svmLinear is 1.06E-07, and the p-value of the output distribution of the trained svmRadial is 1.27E-08.

[0210] Therefore, the trained generalized linear model (glmnet) is a lung cancer early screening model based on TCR sequencing.

[0211] Example 4: A lung cancer early screening device based on TCR sequencing

[0212] I. Composition of Lung Cancer Early Screening Detection Device

[0213] A lung cancer early screening detection device based on TCR sequencing includes a sequencing module, an analysis module, a prediction module, and a result output module;

[0214] The sequencing module is used to perform TCR sequencing on the sample to be tested and obtain the fastq file from the sequencing output file;

[0215] The analysis module is used to perform the steps shown in Example 2 on the fastq obtained by the sequencing module to obtain the TCR features of the sample to be tested and to obtain the input features of the sample after screening.

[0216] The types of input features after screening the samples to be tested are shown in Table 9.

[0217] The prediction module is the lung cancer early screening model based on TCR sequencing shown in Example 3. The input features of the sample to be tested obtained by the analysis module after screening are used as input variables and input into the lung cancer early screening model to obtain the prediction score of the sample to be tested.

[0218] The results output module is used to output the predicted scores of the test samples obtained by the prediction module;

[0219] The predicted score is [0, 1]. A predicted score of 0 indicates that the sample has lung cancer, and a predicted score of 1 indicates that the sample has lung cancer.

[0220] A cutoff value of 0.5 and a prediction score ≥ 0.5 indicate that the sample is positive (i.e., a malignant pulmonary nodule); an output result < 0.5 indicates that the sample is negative (i.e., a benign pulmonary nodule or a healthy person).

[0221] The lung cancer early screening model based on TCR sequencing shown in Example 3 in the prediction module is replaced with the trained random forest shown in Example 3 to obtain lung cancer early screening detection device 1.

[0222] Replace the lung cancer early screening model based on TCR sequencing shown in Example 3 in the prediction module with the gradient boosting model trained in Example 3 to obtain lung cancer early screening detection device 2.

[0223] Replace the TCR sequencing-based lung cancer early screening model shown in Example 3 in the prediction module with the trained svmLinear shown in Example 3 to obtain lung cancer early screening detection device 3.

[0224] The lung cancer early screening model based on TCR sequencing shown in Example 3 in the prediction module is replaced with the svmRadial trained in Example 3 to obtain the lung cancer early screening detection device 4.

[0225] II. Validation of Lung Cancer Early Screening Detection Device

[0226] Using a sample set containing 166 malignant pulmonary nodule samples (positive samples) and 52 benign pulmonary nodule samples (negative samples) as an independent validation set, each sample in the independent validation set was used as the test sample. The lung cancer early screening detection device shown in step one was used to detect the samples, and the predicted scores of each sample in the independent validation set were obtained. The indicators of the lung cancer early screening detection device (sensitivity, specificity, post pred value, negative pred value, F1 value) were calculated. Based on the predicted scores of each sample in the independent validation set, the AUC value was calculated, and the ROC-AUC curve and the output result distribution graph were plotted.

[0227] The lung cancer early screening devices 1 to 4 shown in step 1 were validated using independent validation sets according to the method described above, and the results were recorded.

[0228] III. Experimental Results

[0229] The index results of the lung cancer early screening detection device are shown in Table 12, and the AUC values ​​are shown in Table 13.

[0230] Table 12 Indicator Results of Lung Cancer Early Screening Detection Device

[0231]

[0232] Table 13 AUC values ​​of lung cancer early screening devices

[0233]

[0234] The ROC-AUC curve of the lung cancer early screening device is shown below. Figure 4 As shown in the figure; the distribution of output results from the lung cancer early screening device is shown in the figure. Figure 5 As shown, A is the output result distribution diagram of lung cancer early screening device 1, B is the output result distribution diagram of lung cancer early screening device 2, C is the output result distribution diagram of lung cancer early screening device 3, and E is the output result distribution diagram of lung cancer early screening device 4.

[0235] The results showed that the F1 value of the lung cancer early screening device was 0.87 and the AUC value was 0.84, indicating that the lung cancer early screening device can accurately detect lung cancer in the test samples.

[0236] Figure 5The results showed that the p-value of the output results of lung cancer early screening device 1 was 8.36E-11, the p-value of the output results of lung cancer early screening device 2 was 1.80E-12, the p-value of the output results of lung cancer early screening device 3 was 7.75E-13, and the p-value of the output results of lung cancer early screening device 4 was 6.86E-12.

[0237] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description and ideas, and it is neither necessary nor possible to exhaustively describe all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a lung cancer early screening model based on TCR sequencing, characterized in that, The method comprises the following steps: S1. Establishment of an enriched TCR sequence database: S11. Obtain the TCR immune repertoire of the experimental group and the TCR immune repertoire of the control group, and calculate the number of occurrences of each TCR sequence in the TCR immune repertoire of the experimental group and the TCR immune repertoire of the control group, respectively; the experimental group contains patient samples, and the control group contains healthy person samples; S12. Remove the TCR sequences with an occurrence number < 5 in the TCR immune repertoire of the control group obtained in step S11 to obtain a TCR sequence dataset of the control group; Remove the low-abundance TCR sequences in the TCR sequence dataset of the control group to obtain an enriched TCR sequence database of the control group; remove the TCR sequences contained in the TCR sequence dataset of the control group in the TCR immune repertoire of the experimental group obtained in step S11, and remove the low-abundance TCR sequences to obtain an enriched TCR sequence database of the experimental group; S2. Construction of a lung cancer early screening model: Cross-validation training is performed on the machine learning model using the characteristics of the samples in the training set to obtain a lung cancer early screening model; the training set comprises positive samples and negative samples, the positive samples are blood and / or tissue samples of malignant lung nodule samples, and the negative samples are blood and / or tissue samples of benign lung nodule samples or healthy person blood samples; The characteristics include TCR characteristics, clinical characteristics, and imaging characteristics; The TCR characteristics include V gene and J gene proportion characteristics, TCR immune repertoire statistical characteristics, convergent sequence characteristics, TCR sequence amino acid proportion characteristics, TCR sequence proportion characteristics of different frequencies, and TCR enriched sequence characteristics; The method for obtaining the TCR characteristics is as follows: The TCR sequencing data of each sample in the training set is subjected to quality control, and then the TCRs aligned with the CDR3 region of TCR beta are converted into clonotypes files to obtain the to-be-analyzed data of the training set samples; V gene and J gene proportion characteristics: based on the to-be-analyzed data of the samples in the training set, the corrected clonecounts of each V gene and J gene in each TCR sequence of the sample are obtained as the V gene and J gene proportion characteristics of the sample; Convergent sequence characteristics: based on the to-be-analyzed data of the samples in the training set, the TCR sequences with TCR sequence reads greater than or equal to 1-100 and without stop codons and / or frame shift mutations are retained as the functional TCR sequences of the sample, and the characteristics of the convergent sequences are calculated; the convergent sequences are TCR sequences with the same amino acid sequence but different nucleotide sequences; TCR enriched sequence characteristics: (1) for the to-be-analyzed data of the samples in the training set and the TCR sequences with the same length in the control group enriched TCR sequence database in step S1, retain TCR sequence pairs with an edit distance ≤ 0.3×b, wherein b is the length of the TCR sequence, and select TCR sequence pairs with an alignment_score ≥ 4.5×b as TCR sequence combinations; The TCR enrichment sequence feature enrich_score1 is calculated according to Formula II based on the TCR sequence combination, and the TCR enrichment sequence feature enrich_type1 is calculated according to Formula III. Formula II: ; wherein alignment_score ij is the alignment_score of TCR sequence i and TCR sequence j that are of the same length; , n ki is the number of TCR sequence i in sample k, sample k belongs to samples in the sample set, is the database of enriched TCR sequences of the control group; , n j is the number of TCR sequence j in the TCR sequence combination, TCR sequence j is derived from the to-be-analyzed data of the sample; query_seqs is the number of TCR sequences in the to-be-analyzed data of the sample; Formula III: ; Wherein enriched_seqs is the number of TCR sequences in the TCR sequence combination derived from the sample; query_seqs is the number of TCR sequences in the sample to be analyzed data; (2) According to the method shown in (1), the TCR sequence database is enriched according to the sample to be analyzed and the experimental group shown in step S1, and the TCR enrichment features enrich_score2 and enrich_type2 are calculated. The proportion of the V gene and the J gene is the proportion of the TRBJ2-6 gene, the proportion of the TRBV18 gene, the proportion of the TRBV19 gene, the proportion of the TRBV2 gene, the proportion of the TRBV27 gene, the proportion of the TRBV28 gene, the proportion of the TRBV3-1 gene, the proportion of the TRBV5-1 gene, the proportion of the TRBV5-4 gene, the proportion of the TRBV5-5 gene, the proportion of the TRBV7-8 gene, and the proportion of the TRBV9 gene. The statistical features of the TCR immune repertoire are Shannon index, reverse Simpson index, and uniformity index. The convergence sequence feature is the proportion of the corrected clonecounts of the convergence sequence Convergence_score; the Convergence_score is the proportion of the corrected number of the convergence sequence in the functional TCR sequence to the corrected number of the TCR sequence in the functional TCR sequence. The TCR sequence amino acid proportion feature is the proportion of glycine, the proportion of leucine, the proportion of asparagine, and the proportion of tryptophan. The proportion of TCR sequences with different frequencies is the proportion of TCR sequences with a frequency of 1E-5 to 1E-6. The TCR enrichment sequence feature is the TCR enrichment sequence feature enrich_score. The clinical features are whether to quit smoking and whether to have a history of smoking. The imaging features are whether it is a ground glass nodule, whether it is a solid nodule, whether there is a burr, whether it is a partial solid nodule, and the size of the pulmonary nodule.

2. The construction method of claim 1, wherein, In step S2, the machine learning model is cross-validated and trained using k-fold cross validation and / or leave-one-out cross validation.

3. The construction method of claim 1, wherein, Before cross-training in step S2, the features of the samples in the training set are screened using Ensemble Feature selection and / or boruta feature screening method.

4. The construction method according to claim 3, characterized in that, The Ensemble Feature selection feature screening method includes recursive feature elimination, random forest, and / or glmnet.

5. The construction method of claim 3, wherein, After screening the features of the samples in the training set, data balancing is performed.

6. The construction method of claim 1, wherein, The method for obtaining the TCR sequence amino acid proportion feature in step S2 is: based on the functional TCR sequence of the sample, removing 1-4 amino acids at both ends of each TCR sequence, and calculating the proportion of each amino acid after deleting the amino acid of the TCR sequence.

7. The construction method of claim 1, wherein, The method for obtaining the different frequency TCR sequence proportion feature in step S2 is: based on the functional TCR sequence of the sample, grouping the functional TCR sequence by frequency, calculating the number proportion of the TCR sequence of each frequency group, and obtaining the proportion of the TCR sequence of each frequency.

8. A lung cancer early screening detection device based on TCR sequencing, characterized in that, The sequencing module, the analysis module, the prediction module and the result output module are included. The sequencing module is used for TCR sequencing of the sample to be tested and obtaining the fastq file in the sequencing file. The analysis module executes the steps shown in step S2 in the construction method of claim 1 on the fastq file obtained by the sequencing module, obtains the TCR feature of the sample to be tested, and obtains the screened input feature of the sample to be tested. The prediction module is the lung cancer early screening model constructed by the construction method of claim 1, and the screened input feature of the sample to be tested obtained by the analysis module is input into the lung cancer early screening model as an input variable to obtain the lung cancer prediction score of the sample to be tested. The result output module is used for outputting the lung cancer prediction score of the sample to be tested obtained by the prediction module.

Images