Marker, primer group and artificial neural network detection model for detecting non-small cell lung cancer

Abstract

The invention provides a marker, a primer group and an artificial neural network detection model for detecting non-small cell lung cancer, and belongs to the technical field of molecular biology. According to the method, an artificial neural network pre-training model is constructed aiming at miRNA-seq of cancer tissues in a large-sample-size TCGA database, information learned in primary tumor tissues is finely adjusted and migrated into serum by utilizing a transfer learning method, miRNA-seq spectrums of the primary tumor tissues are mapped through miRNA spectrums in the serum, and the miRNA-seq spectrums of the primary tumor tissues are obtained. The invention develops an economic, efficient and time-saving NSCLC liquid biopsy diagnosis method, namely an artificial neural network detection model for the non-small cell lung cancer, and has important significance on NSCLC diagnosis. The method has become a key research and development plan project for research and development of a new liquid biopsy technology for lung cancer diagnosis in Hunan province, and the project number is 2023SK2039.

Description

Technical Field

[0001] The present invention relates to the technical field of molecular biology, and in particular to a marker, a primer set and an artificial neural network detection model for detecting non-small cell lung cancer. Background Art

[0002] According to global cancer statistics, new cases of lung cancer accounted for 12.4% of the total number of cancers in 2022, ranking first. The mortality rate of lung cancer accounts for 18.7% of the total number of cancer deaths, and it is still the leading cause of cancer death. Among them, 85% of lung cancer is non-small cell lung cancer (NSCLC), which is divided into lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC) and large cell carcinoma (LCC). Because most early NSCLC symptoms are not obvious, it is easy to miss the best time for treatment, resulting in a low 5-year survival rate.

[0003] At present, the commonly used lung cancer diagnosis methods in clinical practice, such as imaging, sputum exfoliative cytology and histopathology, have problems such as high false negative rate, poor sensitivity and high trauma. The heterogeneity of the internal structure of the tumor also reduces the accuracy of the detection. Since the abundance of marker molecules changes before the morphological changes of cancer cells occur and enters the body fluids through the body fluid circulation, clinical doctors can obtain comprehensive information about the lesions by detecting marker molecules in the patient's body fluids in a non-invasive sampling manner, thereby achieving early diagnosis of cancer.

[0004] Transfer learning refers to using big data in the source domain to build a pre-trained model, combined with small data in the target domain of similar tasks. There is no need to train the model from scratch. Only appropriate parameter adjustments are made to the pre-trained model to finally obtain the target domain model.

[0005] MicroRNAs (miRNAs) are a class of non-coding single-stranded RNA molecules with a length of 18-25 nucleotides. They may be abnormally expressed in cancer cells and tissues as tumor suppressor or oncogenes, and enter the body fluid circulation. They have high stability and are considered to be potential biomarkers in liquid biopsies. Currently, most liquid biopsies focus on a single indicator. Individual differences require that disease diagnosis must integrate multiple parameters to improve the accuracy of detection.

[0006] Using the expression data of miRNA in serum, an effective NSCLC classification model can be established through deep learning. With the widespread application of high-throughput sequencing and microarray technology in cancer research, a large amount of tumor tissue miRNA-seq data has been accumulated and can be queried in public databases such as The Cancer Genome Atlas (TCGA). However, as an emerging biomarker for liquid biopsy, serum miRNA is still in the exploratory stage of clinical application, and there is little accumulation of relevant data. In addition, due to different detection technologies such as microarray chips, high-throughput sequencing or qRT-PCR, there is a lack of standardized data analysis methods for data merging between platforms, resulting in poor reproducibility and consistency of results.

[0007] Therefore, developing a cost-effective and time-saving liquid biopsy diagnostic method for NSCLC is crucial for the diagnosis of NSCLC. Summary of the invention

[0008] The purpose of the present invention is to provide a marker, a primer set and an artificial neural network detection model for detecting non-small cell lung cancer.

[0009] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0010] The present invention provides a marker for detecting non-small cell lung cancer. The marker for detecting non-small cell lung cancer is miR-139-5p, miR-10a-5p, miR-148a-3p, and miR-30d-5p.

[0011] The present invention also provides a screening method for detecting markers of non-small cell lung cancer, comprising the following steps:

[0012] (1) The miRNA expression profiles of NSCLC cancer tissues and adjacent tissues were screened in TCGA, and 409 miRNAs with P < 0.05 and |FoldChange| greater than 1.2 were selected as candidate biomarkers;

[0013] (2) Genetic algorithm was used to determine the optimal miRNA dimension in multi-target detection: the differentially expressed miRNAs in non-small cell lung cancer in TCGA were randomly encoded into an individual composed of binary numbers, representing a set of marker combinations, and N individuals were randomly generated as the initial population. The genetic algorithm was used to find individuals with higher fitness f(c) to determine the optimal miRNA dimension, and the miRNA dimension was finally determined to be 4;

[0014] (3) Coupling t-SNE dimensionality reduction and K-Means clustering to determine the optimal miRNA marker combination: The top four miRNA groups ranked by fitness were selected as candidate markers. The high-dimensional data were converted to low-dimensional space through t-SNE to retain the local structure of the data. K-means was used to further cluster the data in the low-dimensional space and match them with the true labels of the samples in the database. The most accurate group of miRNA markers for detecting non-small cell lung cancer was selected: miR-139-5p, miR-10a-5p, miR-148a-3p, and miR-30d-5p.

[0015] The present invention also provides a primer set for detecting the non-small cell lung cancer marker according to claim 1, which is composed of the following primers:

[0016] (1) miR-139-5p:

[0017] The stem-loop primer sequence shown in SEQ ID NO.1 used for reverse transcription is:

[0018] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACA CTGGA;

[0019] The forward primer sequence shown in SEQ ID NO.2 used in fluorescent quantitative PCR is:

[0020] CGCGTCTACAGTGCACGTGTC;

[0021] The universal primer sequence shown in SEQ ID NO.3 used in fluorescent quantitative PCR is:

[0022] AGTGCAGGGTCCGAGGTATT;

[0023] (2) miR-10a-5p:

[0024] The stem-loop primer sequence shown in SEQ ID NO.4 used for reverse transcription is:

[0025] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACC ACAAA;

[0026] The forward primer sequence shown in SEQ ID NO.5 used in fluorescent quantitative PCR is:

[0027] CGCGTACCCTGTAGATCCGAA;

[0028] The universal primer sequence shown in SEQ ID NO.3 used in fluorescent quantitative PCR;

[0029] (3) miR-148a-3p:

[0030] The stem-loop primer sequence shown in SEQ ID NO.6 used for reverse transcription is:

[0031] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACA CAAAG;

[0032] The forward primer sequence shown in SEQ ID NO.7 used in fluorescent quantitative PCR is:

[0033] GCGCGTCAGTGCACTACAGAA

[0034] The universal primer sequence shown in SEQ ID NO.3 used in fluorescent quantitative PCR

[0035] (4) miR-30d-5p:

[0036] The stem-loop primer sequence shown in SEQ ID NO.8 used for reverse transcription is:

[0037] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCT TCCA;

[0038] The forward primer sequence shown in SEQ ID NO.9 used in fluorescent quantitative PCR is:

[0039] GCGTGTAAACATCCCCGAC;

[0040] The universal primer sequence shown in SEQ ID NO.3 was used in fluorescent quantitative PCR.

[0041] The present invention also provides a serum miRNA detection product, which includes a primer set for detecting non-small cell lung cancer markers.

[0042] Furthermore, the product includes any one of a detection reagent, a detection kit and a detection chip

[0043] The present invention also provides a method for constructing an artificial neural network detection model for non-small cell lung cancer, comprising the following steps:

[0044] (1) The miRNA-seq expression matrix of non-small cell lung cancer and adjacent tissue samples was selected as the source domain dataset, and an artificial neural network pre-training model was constructed to classify tumor and adjacent tissue samples;

[0045] (2) The non-small cell lung cancer serum samples and healthy serum samples were amplified using the primer sequences shown in SEQ ID NO. 1 to 9, and the expression levels of the non-small cell lung cancer markers miR-139-5p, miR-10a-5p, miR-148a-3p, and miR-30d-5p were used as the target domains. Based on the transfer learning method, the third and fourth hidden layers in the artificial neural network pre-training model were frozen, and the connection weight parameters between the first two layers of neurons were fine-tuned to obtain a non-small cell lung cancer artificial neural network detection model guided by the miRNA expression level in serum.

[0046] Furthermore, in step (1), when constructing the artificial neural network pre-training model, the source domain data set is divided into a training set and a test set at 60% and 40% respectively.

[0047] Furthermore, in step (2), when retraining the frozen artificial neural network pre-training model, the target domain is divided into a training set and a test set at 60% and 40%.

[0048] The present invention also provides a non-small cell lung cancer artificial neural network detection model constructed by the above method.

[0049] The beneficial effects of the present invention compared with the prior art are:

[0050] (1) The present invention uses a transfer learning method to construct an artificial neural network pre-training model for the miRNA-seq of cancer tissues in the large sample size TCGA database. The information learned from primary tumor tissues is fine-tuned and transferred to serum using the transfer learning method. By mapping the miRNA spectrum in serum to the miRNA-seq spectrum in primary tumor tissue, an economical, efficient and time-saving NSCLC liquid biopsy diagnostic method - an artificial neural network detection model for non-small cell lung cancer is developed.

[0051] (2) The present invention couples supervised learning (genetic algorithm) with unsupervised learning (K-Means) and combines t-SNE dimensionality reduction cross-validation to determine the miRNA marker combination. Unsupervised learning can better mine the clustering characteristics and internal structure of the data itself without labels, and combined with supervised learning, it helps to screen out better biomarkers.

[0052] (3) The diagnostic marker combination and detection primer provided by the present invention have a diagnostic efficacy of 94.4%, a sensitivity of 88.9%, a specificity of 100.0%, and an accuracy of 100.0% in the training set; a diagnostic efficacy of 91.7%, a sensitivity of 100.0%, a specificity of 83.3%, and an accuracy of 85.7% in the test set; and a diagnostic efficacy of 91.5%, a sensitivity of 92.2%, a specificity of 87.5%, and an accuracy of 97.7% in 59 serum samples, which is of great significance for the clinical diagnosis of NSCLC. The present invention has become a key research and development project for the development of new liquid biopsy technologies for lung cancer diagnosis in Hunan Province, with a project number of 2023SK2039. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0054] Figure 1 A flow chart for constructing a NSCLC classification model for the present invention;

[0055] Figure 2 The volcano plot of the differential analysis between NSCLC and healthy tissue samples in TCGA of Example 1 of the present invention, where P<0.05 and |Fold Change|>1.2 indicate that differentially expressed miRNAs are used for subsequent marker screening;

[0056] Figure 3 The genetic algorithm screening results of Example 1 of the present invention, wherein a is a flowchart of the genetic algorithm to determine the miRNA dimension, and b is the fitness J d (c) Trend graph of changes with increasing dimension;

[0057] Figure 4 It is a dimensionality reduction cluster diagram of the top 4 miRNA combinations ranked by fitness in Example 1 of the present invention; wherein a to b are ranked 1 to 4 respectively;

[0058] Figure 5 : The RT-qPCR experimental results in Example 2 of the present invention, wherein a is the amplification curve of miR-139-5p, and b is the standard curve;

[0059] Figure 6 : The RT-qPCR experimental results in Example 2 of the present invention, wherein a is the amplification curve of miR-10a-5p, and b is the standard curve;

[0060] Figure 7: The RT-qPCR experimental results in Example 2 of the present invention, wherein a is the amplification curve of miR-148a-3p, and b is the standard curve;

[0061] Figure 8 : The RT-qPCR experimental results in Example 2 of the present invention, wherein a is the amplification curve of miR-30d-5p, and b is the standard curve;

[0062] Fig. 9 : The detection results of the source domain in Example 3 of the present invention, wherein a is a trend diagram of the accuracy and loss function of the ANN model in the source domain TCGA training set as the number of training steps increases; b is a ROC curve diagram of the training set and the test set; c and d are confusion matrix diagrams of the training set and the test set, respectively;

[0063] Fig.10 This is a flow chart of implementing transfer learning of tissue and serum miRNA expression profiles through an artificial neural network model in Example 3 of the present invention;

[0064] Fig.11 The feature 1 and feature 2 distributions of the expression levels of four miRNAs in TCGA and clinical serum samples after dimensionality reduction in Example 3 of the present invention, wherein a is the data distribution diagram before correction, and b is the data distribution diagram after correction;

[0065] Fig.12 The detection results of the target domain in Example 3 of the present invention, wherein a is a trend diagram of the accuracy and loss function of the ANN model in the target domain clinical serum sample training set after transfer learning as the number of training steps increases; b is a ROC curve diagram of the training set and the test set; c is a confusion matrix diagram of the training set, and d is a confusion matrix diagram of the test set;

[0066] Fig.13 is the Ct value of the target miRNA in the clinical serum sample of Example 3 of the present invention;

[0067] Fig.14 The distribution of feature 1 and feature 2 after dimensionality reduction of the expression content of the additional serum samples and 4 miRNAs in TCGA in Example 3 of the present invention;

[0068] Fig.15 This is the result of further verifying the model effectiveness through serum samples in Example 3 of the present invention, where a is the ROC curve and b is the confusion matrix diagram. DETAILED DESCRIPTION

[0069] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as limiting the present invention, but should be understood as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0070] It should be understood that the terms described in the present invention are only for describing special embodiments and are not intended to limit the present invention. In addition, for the numerical range in the present invention, it should be understood that each intermediate value between the upper and lower limits of the scope is also specifically disclosed. Each smaller range between the intermediate value in any stated value or stated range and any other stated value or intermediate value in the described range is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded in the scope.

[0071] Unless otherwise indicated, all technical and scientific terms used herein have the same meanings as those generally understood by those skilled in the art. Although the present invention describes only preferred methods and materials, any methods and materials similar or equivalent to those described herein may also be used in the implementation or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials associated with the documents. In the event of a conflict with any incorporated document, the content of this specification shall prevail.

[0072] It will be apparent to those skilled in the art that various modifications and variations may be made to the specific embodiments of the present invention description without departing from the scope or spirit of the present invention. Other embodiments derived from the present invention description will be apparent to the skilled artisan. The present invention description and examples are exemplary only.

[0073] The words “include,” “including,” “have,” “contain,” etc. used in this document are open-ended terms, meaning including but not limited to.

[0074] The present invention couples supervised learning (genetic algorithm) with unsupervised learning (K-Means) to screen miRNA marker combinations (miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p), constructs an artificial neural network model (ANN) based on the transfer learning method, transfers the information learned in the primary cancer tissue to the serum, and develops a cost-effective, small data-driven NSCLC liquid biopsy detection method. For the specific process, see Figure 1 .

[0075] Example 1

[0076] Example 1 of the present invention provides a method for screening markers for detecting non-small cell lung cancer, and the specific steps are as follows:

[0077] (1) Screening of candidate markers

[0078] The miRNA-seq profiles in TCGA were cleaned, duplicate data were removed, and missing values ​​were filled by KNN. Finally, 506 miRNA expression profiles of 992 NSCLC cancer tissues and 90 adjacent tissues were obtained. Figure 2 For the differential expression analysis results, 409 miRNAs with P < 0.05 and |Fold Change| greater than 1.2 were selected as candidate markers.

[0079] (2) Using genetic algorithms to determine the optimal miRNA dimension in multi-target detection

[0080] Due to the imbalance in the number of cancer and para-cancerous tissue samples, the Synthetic Minority Over-sampling Technique (SMOTE) algorithm is first used to increase the number of para-cancerous tissue samples to prevent overfitting of the model. Next, the candidate miRNA is encoded as an individual composed of binary numbers, 1 represents that the miRNA at the corresponding position is selected, and 0 represents that it is not selected. Each individual represents a group of biomarkers. Randomly generate N individuals as the initial population. Based on the inter-class / intra-class distance as the fitness function, find the individual with the highest fitness f(c). Elite individuals are selected from the N individuals according to probability, and these elite parent individuals cross-link with each other and combine with the mutations of the offspring individuals to form a new population. Iterate M times, and select the best individual as the final result to determine the optimal miRNA dimension. The process is as follows: Figure 3 As shown in a. Figure 3 Figure b shows the changing trend of the fitness of the optimal individual as the miRNA dimension increases. It can be seen that the individual fitness gradually converges after the dimension is 4. Considering the cost and complexity of the experiment, the miRNA dimension is finally determined to be 4.

[0081] (3) Determine the optimal miRNA combination by coupling t-SNE dimensionality reduction with unsupervised clustering

[0082] The top 4 miRNA combinations were selected to transform the high-dimensional data into low-dimensional space through t-SNE, and then clustered in low-dimensional space using K-means to match the true label values ​​of samples in the database to obtain the final accuracy. The results show that combination 1 can achieve 87.8% classification accuracy without labels. Figure 4 As shown in (a), miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p were selected as the final marker group.

[0083] Example 2

[0084] In Example 2 of the present invention, specific primers were designed for the markers miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p, and detection was performed. The specific steps are as follows:

[0085] (1) The public database information of miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p sequences is shown in Table 1:

[0086] Table 1 Public database information of the above miRNA sequences

[0087]

[0088] (2) Serum sample preparation: About 5 mL of venous whole blood was collected from 66 NSCLC patients and 23 healthy subjects from Hunan Cancer Hospital and Xiangya Hospital of Central South University from 2023 to 2024. The blood was placed in a blood collection tube (without anticoagulant) and left at room temperature for 1 hour for coagulation. The blood was then centrifuged at 3000g for 10 minutes at 4°C, and the upper layer of serum was transferred to an RNase-free EP tube and stored in a -80°C refrigerator for later use.

[0089] The selected samples of patients with non-small cell lung cancer were all first-time patients with no history of medication or surgical treatment, and were diagnosed with non-small cell lung cancer by pathology after hospitalization. Healthy people were those who passed the physical examination at the physical examination center.

[0090] (3) Extraction of total RNA from serum (using miRNAKit Nucleic Acid Extraction Kit, Quanshijin Company)

[0091] A. Take out serum samples from -80℃ refrigerator, thaw at room temperature, centrifuge at 3000g for 5 min at 4℃ to remove cold precipitate, transfer 200μL supernatant into a new EP tube, add 1mL LB10 and shake to mix, and let stand at room temperature for 5min.

[0092] B. Continue to add 0.2 mL of RNA Extraction Agent, shake vigorously for 30 seconds, incubate at room temperature for 3 minutes, and centrifuge at 10,000 g for 15 minutes at 4°C. Transfer the upper colorless aqueous phase into a new RNase-free EP tube.

[0093] C. Add 1 / 3 of the volume of the transfer solution, anhydrous EtOH, to the aqueous phase, invert and mix, transfer it to the RNASpin Column, and centrifuge at 12000g for 30 seconds at room temperature. Transfer the effluent to an RNase-free EP tube, add 1.25 times the volume of the effluent, and gently invert and mix. Transfer it to the miRNA Spin Column, centrifuge at 12000g for 30 seconds at room temperature, discard the effluent, and repeat this operation until the solution is added.

[0094] D. Add 500uL of anhydrous EtOH and WB10 mixed solution (v:v=4:1), centrifuge at 12000g for 30s at room temperature, discard the effluent, repeat this operation once, and centrifuge at 12000g for 2min at room temperature to remove residual EtOH.

[0095] E. Place the miRNA Spin Column in a 1.5mL RNase-free EP tube, add 25uL enzyme-free water to the center of the spin column, and let stand at room temperature for 1min. Centrifuge at 12000g for 1min at room temperature to elute the miRNA. Use an ELISA reader to measure the sample concentration and purity.

[0096] (4) The miRNA 1st Strand cDNA Synthesis Kit commercially available from Novozymes was used to reverse transcribe the miRNA. The total volume of the reverse transcription was 20 uL, and the system was shown in Table 4.

[0097] Table 4 Reverse transcription system

[0098]

[0099]

[0100] The reaction conditions were 16℃30min, 55℃30min, and then 85℃5min for enzyme inactivation. Stem-loop primer sequence:

[0101] miR-139-5p:

[0102] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACA CTGGA;

[0103] miR-10a-5p:

[0104] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACC ACAAA;

[0105] miR-148a-3p:

[0106] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACA CAAAG;

[0107] miR-30d-5p:

[0108] GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCT TCCA;

[0109] (5) Real-time fluorescence quantitative PCR

[0110] TB commercially available from Takara was used. Premix Ex Taq TM II amplification kit was used to detect the expression levels of miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p. The total volume of the reaction system was 20uL. The system is shown in Table 5.

[0111] Table 5 Real-time fluorescence quantitative PCR system

[0112]

[0113]

[0114] Thermo Fisher's QuantStudio 1PCR instrument was used for real-time fluorescence quantitative PCR. The reaction program was shown in Table 6. Other settings were the system default values:

[0115] Table 6 Real-time fluorescence quantitative PCR reaction program

[0116]

[0117] Amplification primer sequences:

[0118] miR-139-5p forward primer CGCGTCTACAGTGCACGTGTC

[0119] miR-10a-5p forward primer CGCGTACCCTGTAGATCCGAA

[0120] miR-148a-3p forward primer GCGCGTCAGTGCACTACAGAA

[0121] miR-30d-5p forward primer GCGTGTAAACATCCCCGAC

[0122] The reverse primer of qPCR was a universal primer with the specific sequence of AGTGCAGGGTCCGAGGTATT.

[0123] The above system and reaction procedure were used to detect the logarithm and Ct value of the standard amount of miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p at different concentrations, and the amplification curves and standard curves of different concentrations were obtained. Figure 5-8 .

[0124] Figure 5-8 The results showed that the Ct of miR-139-5p was 29.13 at 500aM, and the standard curve equation was y = -3.53062lgC + 39.7254 (R 2 =0.9895), and the amplification efficiency was 91.97%. The Ct of miR-10a-5p at 50aM was 34.299, and the standard curve equation was y=-3.59329lgC+40.68001(R 2 =0.99882), and the amplification efficiency was 89.80%. The Ct of miR-148a-3p was 28.9 at 500aM, and the standard curve equation was y=-3.5465lgC+38.36332(R 2 =0.99819), the amplification efficiency was 91.41%; the Ct of miR-30d-5p at 50aM was 30.1, and the standard curve equation was y=-3.35921lgC+35.8514(R 2 =0.99827), and the amplification efficiency was 98.47%.

[0125] As can be seen from the above, the specific primer set of the present invention for detecting the markers miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p has good sensitivity.

[0126] Example 3

[0127] Example 3 of the present invention builds a non-small cell lung cancer artificial neural network detection model based on transfer learning, and the specific steps are as follows:

[0128] (1) In order to achieve efficient liquid biopsy prediction, the present invention selects miRNA-seq expression matrices of 90 NSCLC and 90 paracancerous tissue samples from the primary tumor database (TCGA) as the source domain dataset (source domain). The source domain is used to divide 180 samples in TCGA into training set and test set at 60% and 40% to construct an artificial neural network pre-training model, which can accurately classify tumor and paracancerous tissue samples. Fig. 9 As shown in a, during the training process, the accuracy of the training set gradually increases with the increase in the number of steps and then reaches convergence, and the loss function gradually decreases and then reaches convergence. Fig. 9Figure b shows the ROC curves of the training set and test set, with AUC values ​​of 0.98 and 1.00 respectively. Fig. 9 Figures c and d are the confusion matrices of the training set and test set, respectively. The sensitivity, specificity, accuracy and precision of the training set are 96.3%, 96.3%, 96.3% and 96.3% respectively; the sensitivity, specificity, accuracy and precision of the test set are 97.2%, 86.1%, 91.7% and 87.5% respectively.

[0129] (2) The expression levels of miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p in the serum of 15 NSCLC patients and 15 healthy samples collected in Example 2 were diagnosed using the method of Example 2 as the target domain. The expression levels of miRNA in the serum of 15 NSCLC patients and 15 healthy samples in the target domain were also divided into a training set and a test set at 60% and 40% respectively. Based on the transfer learning method, the third and fourth layers of the hidden layer in the artificial neural network pre-training model were frozen, and the connection weight parameters between the first two layers of neurons were fine-tuned to obtain an artificial neural network detection model for non-small cell lung cancer guided by the expression level of miRNA in serum, such as Fig.10 The process is shown.

[0130] In the process of transfer learning, in order to eliminate the difference in the distribution of miRNA expression data between the two different domains of tissue and serum, the expression levels of the four miRNAs were reduced to a specific feature space, and the cluster centers of the tissue and serum data were calculated to obtain the offset between the centroids of the two. By translating the tissue data to match the serum data, the distribution offset of miRNA markers in tissue and serum can be effectively corrected ( Fig.11 ), thereby transferring cancer lesion markers from tissue to blood, performing transduction learning, and achieving accurate cancer liquid biopsy. Fig.12 shown. Fig.12 The a in the middle shows that the accuracy of the training set gradually increases as the number of steps increases during the training process and then converges, and the loss function gradually decreases and then converges. Fig.12 Figure b shows the ROC curves of the training set and test set, with AUC values ​​of 1.00 and 0.97 respectively. Fig.12 Figures c and d are the confusion matrices of the training set and test set, respectively. The sensitivity, specificity, accuracy, and precision of the training set are 88.9%, 100.0%, 94.4%, and 100.0%; the sensitivity, specificity, accuracy, and precision of the test set are 100.0%, 83.3%, 91.7%, and 85.7%.

[0131] (3) In order to verify the effectiveness of the artificial neural network detection model for non-small cell lung cancer based on the expression of miRNA in serum obtained by transfer learning in step (2), the method of Example 2 was further used to detect the expression of miRNA in the serum of the remaining 51 NSCLC samples and 8 healthy samples. The Ct values ​​of the target miRNA in all serum samples were shown in Table 2. Fig.13 The transfer learning method in step (2) is used to correct the distribution deviation of miRNA between tissues and serum. The corrected data distribution is as follows Fig.14 shown. Fig.15 The ROC curve in a shows that the AUC value is 0.95. Fig.15 The confusion matrix in b shows that the sensitivity is 92.2%, the specificity is 87.5%, the accuracy is 91.5%, and the precision is 97.9%, which has a good classification effect and is of great significance for the clinical diagnosis of NSCLC.

[0132] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A marker for detecting non-small cell lung cancer, characterized in that: The markers for detecting non-small cell lung cancer are miR-139-5p, miR-10a-5p, miR-148a-3p, and miR-30d-5p.

2. A screening method for detecting markers for non-small cell lung cancer according to claim 1, characterized in that: The steps include: (1) Screen the miRNA expression profiles of NSCLC cancer tissues and adjacent tissues in TCGA, and select miRNAs with P < 0.05 and |FoldChange| greater than 1.2 as candidate markers; (2) Using genetic algorithm to determine the optimal miRNA dimension in multi-target detection: randomly encode the differentially expressed miRNAs in non-small cell lung cancer in TCGA into an individual composed of binary numbers, representing a set of marker combinations, randomly generate N individuals as the initial population, and use genetic algorithm to find individuals with higher fitness f(c) to determine the optimal miRNA dimension. The miRNA dimension was finally determined to be 4; (3) Coupling t-SNE dimensionality reduction and K-Means clustering to determine the optimal miRNA marker combination: The top four miRNA groups ranked by fitness were selected as candidate markers. The high-dimensional data were converted to low-dimensional space through t-SNE to retain the local structure of the data. K-means was used to further cluster the data in the low-dimensional space and match them with the true labels of the samples in the database. The most accurate group of miRNA markers for detecting non-small cell lung cancer was selected: miR-139-5p, miR-10a-5p, miR-148a-3p and miR-30d-5p.

3. A primer set for detecting the non-small cell lung cancer marker according to claim 1, characterized in that: The following primers composition: (1) miR-139-5p: The stem-loop primer sequence shown in SEQ ID NO.1 used for reverse transcription is: GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACACTGGA; The forward primer sequence shown in SEQ ID NO.2 used in fluorescent quantitative PCR is: CGCGTCTACAGTGCACGTGTC; The universal primer sequence shown in SEQ ID NO.3 used in fluorescent quantitative PCR is: AGTGCAGGGTCCGAGGTATT; (2) miR-10a-5p: The stem-loop primer sequence shown in SEQ ID NO.4 used for reverse transcription is: GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCACAAA; The forward primer sequence shown in SEQ ID NO.5 used in fluorescent quantitative PCR is: CGCGTACCCTGTAGATCCGAA; The universal primer sequence shown in SEQ ID NO.3 used in fluorescent quantitative PCR; (3) miR-148a-3p: The stem-loop primer sequence shown in SEQ ID NO.6 used for reverse transcription is: GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACACAAAG; The forward primer sequence shown in SEQ ID NO.7 used in fluorescent quantitative PCR is: GCGCGTCAGTGCACTACAGAA The universal primer sequence shown in SEQ ID NO.3 used in fluorescent quantitative PCR (4) miR-30d-5p: The stem-loop primer sequence shown in SEQ ID NO.8 used for reverse transcription is: GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACCTTCCA; The forward primer sequence shown in SEQ ID NO.9 used in fluorescent quantitative PCR is: GCGTGTAAACATCCCCGAC; The universal primer sequence shown in SEQ ID NO.3 was used in fluorescent quantitative PCR.

4. A serum miRNA detection product, characterized in that: The serum miRNA detection product includes the primer set for detecting non-small cell lung cancer markers as described in claim 2.

5. The serum miRNA detection product according to claim 4, characterized in that: The product includes any one of a detection reagent, a detection kit and a detection chip.

6. A method for constructing an artificial neural network detection model for non-small cell lung cancer, characterized in that: The steps include: (1) The miRNA-seq expression matrix of non-small cell lung cancer and adjacent tissue samples was selected as the source domain dataset, and an artificial neural network pre-training model was constructed to classify tumor and adjacent tissue samples; (2) The non-small cell lung cancer serum samples and healthy serum samples were amplified using the primer sequences shown in SEQ ID NO.1~9, and the expression levels of the non-small cell lung cancer markers miR-139-5p, miR-10a-5p, miR-148a-3p, and miR-30d-5p were used as the target domains. Based on the transfer learning method, the third and fourth hidden layers in the artificial neural network pre-training model were frozen, and the connection weight parameters between the first two layers of neurons were fine-tuned to obtain a non-small cell lung cancer artificial neural network detection model guided by the miRNA expression level in serum.

7. The method for constructing an artificial neural network detection model for non-small cell lung cancer according to claim 6, characterized in that: In step (1), when constructing the artificial neural network pre-training model, the source domain data set is divided into a training set and a test set at 60% and 40% respectively.

8. The method for constructing an artificial neural network detection model for non-small cell lung cancer according to claim 6, characterized in that: In step (2), when retraining the frozen artificial neural network pre-training model, the target domain is divided into a training set and a test set at 60% and 40%.

9. An artificial neural network detection model for non-small cell lung cancer constructed according to the method of any one of claims 6 to 8.

Images