Novel plasma extracellular vesicle long-chain RNA analysis and cancer early diagnosis method

Abstract

The invention discloses a novel plasma extracellular vesicle long-chain RNA (Ribonucleic Acid) analysis and cancer early diagnosis method, relates to the technical field of cancer biomarkers and classified biology, and is used for carrying out early diagnosis and prediction on PDAC based on plasma EVs long-chain RNA (EvlRNA). According to the method, a two-layer classifier is established, the first layer identifies abnormal and normal samples, and the second layer identifies whether the samples belong to PDAC or CP; according to the invention, a new concept, namely EvlRNA-Index, is provided for the first time, a cancer diagnosis model is established based on the EvlRNA-Index, and a good diagnosis effect is achieved. According to the research, the interpretability of a whole machine learning framework is researched and discussed, close association between important characteristics and an actual biological process is explored, 49 PDAC cancer related genes are identified and obtained, and important help is provided for biomedical researchers to understand a model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cancer biomarkers and classification biotechnology, and particularly relates to a new method for plasma extracellular vesicle long-chain RNA analysis and early diagnosis of cancer. BACKGROUND

[0002] Pancreatic ductal adenocarcinoma (PDAC) is a tumor that occurs in pancreatic duct epithelium, and is the main type of pancreatic cancer, accounting for more than 90%. PDAC has insidious onset and high invasiveness, and most patients are in the advanced stage when diagnosed. Although early cancer can be effectively treated by surgery and radiotherapy, advanced cancer is often uncontrollable, so in the process of disease progression, early screening, early diagnosis, early treatment, and management of cancer as a "chronic disease" are the most effective way to improve the cure rate of tumor patients, reduce treatment pain, improve prognosis, and reduce economic burden.

[0003] In the prior art, the early diagnosis technology of PDAC has far failed to meet people's expectations. For example, although the CA19-9 level is helpful for the prediction and efficacy judgment of pancreatic cancer, its sensitivity and specificity are very low. In the case of biliary tract infection, cholangitis, biliary obstruction or jaundice, the CA19-9 level can also be elevated. For patients negative for Lewis antigen, the level of CA19-9 usually does not increase. Therefore, there is an urgent need to find new PDAC diagnostic biomarkers, especially liquid biopsy biomarkers suitable for early detection and early diagnosis of PDAC.

[0004] Plasma extracellular vesicles are one of the important sources of materials for liquid biopsy. Extracellular vesicles (EVs) include exosomes and microvesicles, and are a special membrane-like, nanoscale endocytic vesicle secreted by most cell types. EVs contain various molecular components (including RNA, protein, lipid and metabolite), which reflect the source type of cells. Although initially considered as cell debris, more and more people now believe that EVs are an important medium for intercellular communication and a biomarker of disease.

[0005] EVs are associated with most pathological conditions, including cancer, cardiovascular disease, neurological disease, and infectious disease. Different kinds of RNA are present and stabilized in EVs. MicroRNAs have been well characterized and studied in EVs. However, the small number of microRNAs in EVs and the lack of specificity in their production limit their widespread use. Increasing evidence suggests that long RNAs in plasma EVs, including messenger RNAs (mRNAs), circular RNAs (circRNAs), and long noncoding RNAs (lncRNAs), have certain functions and clinical significance. For example, androgen receptor splice variant 7 (AR-V7) can be detected in the blood EVs of patients with castration-resistant prostate cancer and can be used as a predictive biomarker of hormone therapy resistance. In patients with melanoma and non-small cell lung cancer, CD274 mRNA in plasma-derived EVs is associated with response to anti-PD-1 antibodies. Nabet et al. found that unshielded RN7SL1 can be transferred to breast cancer cells through EVs and activate the pattern recognition receptor RIG-I, promoting cancer invasion. These findings reveal that plasma EVs contain a large number of valuable and functional EvlRNAs. Therefore, it is feasible to identify tumor-specific genes in the plasma EvlRNA library and perform early diagnosis of cancer, which is a non-invasive strategy for early diagnosis and treatment evaluation of human cancer, and helps early diagnosis and early treatment of cancer.

[0006] In recent years, with the growing availability of clinical data supporting diagnosis and prognosis, advances in science and technology have made it possible to study cancer using high-throughput biomedical data. However, due to the complexity of cancer, clinical bioinformatics analysis and genetic interpretation are challenging, and these data also need to be truly explored and analyzed. Most machine learning algorithms, especially complex algorithms that rely on neural networks, have good classification effects, but it is difficult to mine the knowledge learned during model training, and it is a "black box" model. For biological problems, not only do we need to build high-performance diagnostic models to meet prediction needs, but we also need to find out the rules used by the model during training, visualize the corresponding important feature weights, and explore the close relationship between important features and actual biological processes to help biomedical researchers understand the model. Therefore, how to design a model with explainability and visualize the corresponding important feature weights is a hot and difficult point in the field of machine learning in recent years.

[0007] In view of this, we propose a new method for analyzing plasma extracellular vesicle long-chain RNA and early diagnosis of cancer. SUMMARY

[0008] The application provides a new method for analyzing plasma extracellular vesicle long-chain RNA and early diagnosis of cancer to solve the above problems.

[0009] A new method for analyzing plasma extracellular vesicle long-chain RNA and early diagnosis of cancer, comprising the following steps:

[0010] S1: extracting a plasma EvlRNA sample from the patient's blood;

[0011] S2: establishing a diagnostic model, wherein the diagnostic model comprises two layers of classifiers, the first layer of classifiers is used to identify normal and non-normal samples, i.e., Normal or Non-Normal, and the second layer of classifiers is used to identify the type of disease suffered by the sample;

[0012] S3: training and testing the diagnostic model by a training data set S, wherein the training data set S can be expressed as:

[0013] S=S Non-Normal ∪S Normal

[0014] wherein S Non-Normal is a Non-Normal data set composed of pancreatic ductal adenocarcinoma PDAC and chronic pancreatitis CP samples, S Normal is a Normal data set only with Normal samples, ∪ is a symbol representing the union in the set, and the Non-Normal data set can be further divided into two categories:

[0015]

[0016] wherein the subscripts 1 and 2 represent "PDAC" and "CP", and the data set S is obtained by next-generation sequencing of the plasma EvlRNA sample.

[0017] S4: inputting the plasma EvlRNA sample information of the patient into the diagnostic model to predict the type of disease suffered by the patient through the diagnostic model;

[0018] wherein the diagnostic model is established based on plasma EVs long-chain RNA, i.e., EvlRNA.

[0019] In the above method, the type of disease includes pancreatic ductal adenocarcinoma PDAC and chronic pancreatitis CP.

[0020] In the above method, the diagnostic model uses EvlRNA-Index as input features.

[0021] In the above method, the workflow of the diagnostic model comprises calculating the index of the EvlRNA-Index using a method comprising one or more of SVD PCA, Nipals PCA, Probabilistic PCA and FastHCS.

[0022] In the above method, the index calculation process of the EvlRNA-Index comprises the following steps:

[0023] S1: Calculate using the method for plasma samples;

[0024] S2: Each plasma sample is scored with 10 EvlRNA-Index;

[0025] S3: Each plasma sample generates a 1x10 matrix, which is equivalent to 10 EvlRNA-Index information.

[0026] In the above method, the first layer classifier can be selected from one or more of the support vector machine SVM algorithm, random forest RF algorithm, deep learning DL algorithm and XGBoost algorithm, and the algorithm combination is realized by stacking the method used by the EvlRNA-Index index, and the algorithm combination is selected by the training data set, independent test data set and 5-fold cross validation.

[0027] In the above method, the first layer classifier can be selected from one or more of the support vector machine SVM algorithm, random forest RF algorithm, deep learning DL algorithm and XGBoost algorithm, and the algorithm combination is realized by stacking the method used by the EvlRNA-Index index, and the algorithm combination is selected by the training data set, independent test data set and 5-fold cross validation.

[0028] In the above method, the second layer classifier can be selected from one or more of the support vector machine SVM algorithm, random forest RF algorithm, deep learning DL algorithm and XGBoost algorithm, and the algorithm combination is realized by stacking the method used by the EvlRNA-Index index, and the algorithm combination is selected by the training data set, independent test data set and 5-fold cross validation.

[0029] In the above method, the training set and test set in the training data set S account for 70% and 30% respectively, the independent test data set comprises a plurality of Normal samples and Non-Normal samples randomly selected, and the randomly selected samples are not used for model training, feature selection or parameter optimization process.

[0030] In the above method, the first layer classifier uses the algorithm combination of Probabilistic PCA method and support vector machine SVM algorithm.

[0031] In the above method, the second layer classifier selects an algorithm combination of Probabilistic PCA method and random forest RF algorithm.

[0032] Compared with the prior art, the application has the advantages of:

[0033] The method solves the problems of model transparency and interpretability. Traditional black box models often face great challenges in explaining the prediction results, especially in the case of complex analysis of high-dimensional data. The EvlRNA-Index proposed in this study not only simplifies the feature space, but also provides an intuitive and easy-to-understand explanation framework, making the prediction process transparent to non-technical users.

[0034] At the same time, using EvlRNA-Index as input features improves the prediction performance and generalization ability of the model. By calculating EvlRNA-Index, key patterns and trends in the data can be captured, improving the accuracy and robustness of the model. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 . is the EvlRNA-Index index calculation schematic diagram of the plasma EvlRNA sample in the application.

[0036] Figure 2 . The flow chart of the diagnostic model construction in the application.

[0037] Figure 3 A is the EvlRNA-Index correlation analysis result graph calculated based on the SVD PCA algorithm.

[0038] Figure 3 B is the EvlRNA-Index correlation analysis result graph calculated based on the Nipals PCA algorithm.

[0039] Figure 3 C is the EvlRNA-Index correlation analysis result graph calculated based on the Probabilistic PCA algorithm.

[0040] Figure 3 D is the EvlRNA-Index correlation analysis result graph calculated based on the FastHCS algorithm.

[0041] Figure 3 E is the EvlRNA-Index visualization graph calculated by the SVD PCA algorithm.

[0042] Figure 3 F is the EvlRNA-Index visualization graph calculated by the Nipals PCA algorithm.

[0043] Figure 3 G is the EvlRNA-Index visualization plot calculated by Probabilistic PCA algorithm.

[0044] Figure 3 H is the EvlRNA-Index visualization plot calculated by FastHCS algorithm.

[0045] Figure 4 A is the plot of the top 5 genes with the highest weights in each Probabilistic PCA-based EvlRNA index calculation process.

[0046] Figure 4 B is the plot of gene type classification.

[0047] Figure 4 C is the plot of comparison between 49 PDAC cancer-related genes identified in this study and genes reported in the literature.

[0048] Figure 5 A is the plot of gene expression analysis of selected genes using RNA-seq data from TCGA and GTEx databases.

[0049] Figure 5 B is the plot of protein expression analysis of protein-coding genes using CPTAC data.

[0050] Figure 6 is the plot of GO analysis results.

[0051] Figure 7 is the plot of overall survival and disease-free survival analysis results DETAILED DESCRIPTION

[0052] The following examples are for illustrative purposes only and are not intended to limit the scope of the present application.

[0053] REFERENCE Figures 1 to 7 A new method for plasma extracellular vesicle long-chain RNA analysis and early cancer diagnosis, comprising the following steps:

[0054] S1: extracting a plasma EvlRNA sample from the patient's blood;

[0055] S2: establishing a diagnostic model, wherein the diagnostic model comprises two layers of classifiers, the first layer of classifiers is used to identify normal and non-normal samples, i.e. Normal or Non-Normal, and the second layer of classifiers is used to identify the type of disease suffered by the sample;

[0056] S3: training and testing the diagnostic model by training data set S, wherein the training data set S can be expressed as:

[0057] S = S Non-Normal ∪ S Normal

[0058] where S Non-Normal is a Non-Normal dataset consisting of pancreatic ductal adenocarcinoma (PDAC) and chronic pancreatitis (CP) samples, S Normal is a Normal dataset with only Normal samples, ∪ is a symbol representing the union in set, and the Non-Normal dataset can be further divided into two categories:

[0059]

[0060] where the subscripts 1, 2 represent “PDAC” and “CP”, and the dataset S is obtained from next-generation sequencing of plasma EvlRNA samples.

[0061] S4: input the information of the patient’s plasma EvlRNA sample into the diagnostic model to predict the type of disease the patient has through the diagnostic model.

[0062] wherein the diagnostic model is established based on plasma EVs long-chain RNA, i.e., EvlRNA.

[0063] wherein the diagnostic model includes the calculation of the EvlRNA-Index, and four algorithms are used to calculate each sample Figure 1 ) when calculating the EvlRNA-Index. Each sample is scored with 10 EvlRNA-Index. Finally, each sample will generate a 1x10 matrix, which is equivalent to 10 EvlRNA-Index information (EvlRNA-Index1, EvlRNA-Index2, EvlRNA-Index3, …, EvlRNA-Index8, vlRNA-Index9, EvlRNA-Index10). The four algorithms used are SVD PCA, Nipals PCA, Probabilistic PCA, and FastHCS.

[0064] Two-layer cancer diagnosis model

[0065] To enable the first layer classifier to distinguish samples as Normal or Non-Normal, four machine learning methods were employed: Support Vector Machine (SVM), Random Forest (RF), Deep Learning (DL) and XGBoost to build the first layer classifier, the implementation of SVM was performed using the e1071 package (available at http: / / cran.r-project.org / / ). The randomForest package was used to implement RF. For deep learning, the h2o package was utilized, and XGBoost was implemented using the XGBoost package. For the first layer prediction, we used SVM, RF, DL or XGBoost as base classifiers due to their performance.

[0066] The second layer classifier identifies whether a sample belongs to PDAC or CP, four machine learning methods were selected: SVM, RF, DL and XGBoost to implement our second layer classifier. Figure 2 A flow chart showing the construction of the diagnostic model is shown.

[0067] After the model was constructed, the performance of the classifiers was evaluated in terms of sensitivity, specificity, accuracy and the Matthews Correlation Coefficient (MCC), which is a measure used in machine learning to evaluate the performance of binary classification, widely used in bioinformatics. This measure takes into account four factors: true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN), it is generally considered a relatively balanced measure, even when there is a significant difference in the number of samples between the two classes. In essence, the MCC is a correlation coefficient that describes the relationship between the actual classification and the predicted classification, with a range of -1 to 1. A value of 1 indicates a perfect prediction of the subject, a value of 0 indicates that the prediction is no better than a random prediction. -1 indicates that the predicted classification is completely inconsistent with the actual classification. The following formula can be used to calculate the four indicators:

[0068]

[0069]

[0070]

[0071] To compare the overall performance of the various models, the area under the receiver operating characteristic curve (AUC) was calculated based on the ROC curve, ranging from 0 to 1. The ROC curve describes the relationship between the true positive rate and the false positive rate. Specifically, the AUC indicates the likelihood that a randomly selected true target has a higher level than a randomly selected decoy target. The higher the AUC value, the better the predictive performance of the model. Five-fold cross-validation was used to evaluate the performance of the model.

[0072] To evaluate the performance of our model, we built an independent dataset. This dataset consists of 35 randomly selected Normal samples out of 117 Normal samples and 115 randomly selected Non-Normal samples out of 384 Non-Normal samples. These samples were not used in the training, feature selection, or parameter optimization process of the model.

[0073] Data mining analysis, the biological significance of the genes was determined by functional enrichment analysis through DAVID. The p-value optimized by Benjamin Hochberg was taken as the critical value at <0.05. Survival outcome evaluation, including overall survival (OS) and disease-free survival (DFS), was performed by Kaplan-Meier survival curves, supplemented by log-rank test comparison. According to the gene expression level (high or low), the cohort was stratified, and the median expression value was used as the cutoff standard. Statistical calculations included hazard ratio (HR) quantification and the corresponding 95% confidence interval.

[0074] After calculating the EvlRNA-Index( Figure 1 ), in order to study the relationship between these indicators, correlation analysis was performed on them using the corrplot software package. Figure 3 A is the correlation analysis result of EvlRNA-Index calculated based on SVD PCA algorithm. Figure 3 B is the correlation analysis result of EvlRNA-Index calculated based on Nipals PCA algorithm. Figure 3 C is the correlation analysis result of EvlRNA-Index calculated based on Probabilistic PCA algorithm. Figure 3 D is the correlation analysis result of EvlRNA-Index calculated based on FastHCS algorithm. The results show that these indicators are not correlated, indicating that they are not redundant as input features of the model. In addition, we visualize the top three indicators of each sample, indicating that it is difficult to accurately distinguish PDAC / CP and normal samples( Figure 3 E-F).

[0075] The first layer of the classifier identifies whether the sample is Normal or Non-Normal. Using RF, SVM, DL, and XGB, respectively, each algorithm uses the indices calculated based on conventional_SVD_PCA, the indices calculated based on FastHCS, the indices calculated based on Nipals_PCA, and the indices calculated based on Probabilistic_PCA, respectively, to establish the first layer classification model. Table 1 shows the performance of the first layer of the model. Among all the models, the model PDACandCPvsHealth-SVDPCAIndex-RF (1-5) showed the best performance effect, with an accuracy of 96.23%, an MCC value of 0.8947, and an AUC value of 0.9901, but the prediction ability was poor in independent data testing and 5-fold cross-validation (Table 2, Table 3). In summary, among all the models, the model PDACandCPvsHealth-Probabilistic PCAIndex-SVM (1-18) performed best in terms of internal stability and external predictability, with an accuracy of 91.51%, an MCC value of 0.7760, and an AUC value of 0.9560 in the training data set; an accuracy of 93.33%, an MCC of 0.8137, and an AUC value of 0.9717 in the independent data test set; and an accuracy of 91.19%, an MCC of 0.7430, and an AUC value of 0.9389 in five-fold cross-validation.

[0076] To further verify our method, we evaluated the performance of the model on an independent data set. An independent data set containing 115 positive and negative samples (85 PDAC, 30 CP, 35 Normal) was used to evaluate the prediction ability of the model. As shown in Table 2, these models all showed good performance, indicating the reliability of these models. In addition, 5-fold cross-validation was also performed (Table 3). Considering the performance on the training set and the independent data set to select the best classifier, in order to get the best prediction result, we only selected the model PDACandCPvsHealth-Probabilistic PCAIndex-SVM (1-18) as the first layer classifier of our method.

[0077] Table 1 Performance of the first layer of the model on the training data set.

[0078]

[0079]

[0080] Table 2 Performance of the first layer of the model on the independent test data set.

[0081]

[0082]

[0083] Table 3. First layer model five-fold cross-validation results.

[0084]

[0085] The second layer of classifiers was used to diagnose the distinction between PDAC and CP. Four machine learning algorithms (RF, SVM, DL, and XGB) were selected to implement the second layer classification. The statistics of the second layer of models are summarized in Table 4.

[0086] The model PDACvsCP-Probabilistic PCA Index-RF (2-17) had an accuracy of 93.83%, a sensitivity of 95.00%, a specificity of 90.48%, a MCC of 0.8422, and an AUC value of 0.9698. The model PDACvsCP-Probabilistic PCA Index-SVM (2-18) had an accuracy of 96.30%, a sensitivity of 100.00%, a specificity of 85.71%, a MCC of 0.9035, and an AUC value of 0.9833.

[0087] In the independent dataset test, the model PDACvsCP-Probabilistic PCA Index-RF (2-17) showed better performance compared to the model PDACvsCP-Probabilistic PCA Index-SVM (2-18) (Table 5), indicating that the combination of the index processed by Probabilistic PCA and RF is a good choice with high internal stability and strong external prediction ability. It can also be seen from the five-fold cross-validation results that among all the models, the model PDACvsCP-Probabilistic PCA Index-RF (2-17) had the best performance with an accuracy of 98.13% (Table 6).

[0088] Table 4. Performance of the second layer of models on the training dataset.

[0089]

[0090]

[0091] Table 5. Performance of the second layer of models on the independent test dataset.

[0092]

[0093] Table 6. Five-fold cross-validation results of the second layer of models.

[0094]

[0095] Among all models, PDACandCPvsHealth-Probabilistic PCA Index-SVM(1-18) showed the best diagnostic performance in the first layer, and PDACvsCP-Probabilistic PCA Index-RF(2-17) showed the best diagnostic performance in the second layer. To investigate the potential mechanism of these EvlRNA indexes as input features of diagnostic models and improve the interpretability of models, we extracted the top 5 genes in the weight ranking during the calculation of each Probabilistic PCA-based EvlRNA index, resulting in 10 x 5 genes. After removing duplicate genes and gene name conversion, we finally obtained 49 genes (A). The gene types included protein-coding (51.02%), pseudogene (26.53%), IncRNA (18.37%), and others (4.08%) (B). Through database and literature retrieval, we found that this gene set contained 16 known PDAC cancer-related genes, including ANXA4, PF4, MUC3A, MAL2, EIF4G2, NEAT1, MALAT1, NRGN, SCARNA10, GAPDH, TUBA1C, CALM1, DUOX2, FXYD3, LGALS4, and LENG8 (C). Therefore, we inferred that the remaining genes were also PDAC cancer-related genes. Figure 4 Figure 4 Figure 4 C). Thus, we inferred that the remaining genes were also PDAC cancer-related genes.

[0096] To verify the relationship between these 49 genes and PDAC, we downloaded the RNA-seq data of PDAC (including 178 PDAC tissues and 171 normal pancreas tissue samples) from the TCGA and GTEx databases for analysis. We found that the expression of most genes was significantly upregulated or downregulated in cancer tissue samples (A). Gene expression at the protein level was closer to the original manifestation of the disease. Therefore, we used CPTAC to further investigate the protein expression data of PDAC, and the results showed that the protein expression of most protein-coding genes was significantly upregulated or downregulated in PDAC tissues (B), which further verified the reliability of these results. Figure 5 Figure 5 B), which further verified the reliability of these results.

[0097] ​​​GO analysis of these genes showed that these genes were enriched in items related to molecular function regulator activity (GO:0098772), plasma membrane (GO:0005886), vesicle (GO:0031982), nucleus (GO:0005634), base-pairing miRNA inhibitor activity (GO:0140869), lncRNA-mediated post-transcriptional gene silencing (GO:0000512), regulation of miRNA catabolic process (GO:2000625), gene expression (GO:0010467), etc. Figure 6 ) and played an important role in the development of cancer.

[0098] We performed survival analysis on these 49 genes and found that the high expression of some genes was related to the survival prognosis of PDAC. Specifically, CCDC13-AS1 (OS, DFS), LENG8 (OS) and NRGN (OS, DFS) were associated with good prognosis of PDAC patients. On the contrary, PTP4A2 (OS), OST4 (OS), MAL2 (OS, DFS), GAPDH (DFS), TUBA1C (OS, DFS) and DUOX2 (OS) were associated with poor prognosis of PDAC patients. The Kaplan-Meier plots of these genes are shown in Figure 7 The above results show that the biomarkers identified by us are of great significance for the diagnosis and prognosis of PDAC patients.

[0099] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis, characterized in that, Includes the following steps: S1: Extract plasma EvlRNA samples from the patient's blood; S2: Establish a diagnostic model, which includes two classifiers. The first classifier is used to identify normal and abnormal samples, i.e., Normal or Non-Normal. The second classifier is used to identify the type of disease the sample has. S3: Train and test the diagnostic model using the training dataset S, wherein the training dataset S can be described as follows: S=S Non-Normal ∪S Normal Where S Non-Normal A non-normal dataset consisting of pancreatic ductal adenocarcinoma (PDAC) and chronic pancreatitis (CP) samples, S Normal This is a Normal dataset containing only Normal samples. ∪ is the symbol for the union of sets. Non-Normal datasets can be further divided into two categories: The subscripts 1 and 2 represent "PDAC" and "CP", and the dataset S is taken from next-generation sequencing of plasma EvlRNA samples; S4: Input the patient's plasma EvlRNA sample information into the diagnostic model, and predict the type of disease the patient has through the diagnostic model; The diagnostic model described above is based on plasma EVs long chain RNA, namely EvlRNA.

2. The novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 1, characterized in that, The types of diseases mentioned include pancreatic ductal adenocarcinoma (PDAC) and chronic pancreatitis (CP).

3. The novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 1, characterized in that, The diagnostic model uses EvlRNA-Index as an input feature.

4. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 1, characterized in that, The workflow of the diagnostic model includes calculating the EvlRNA-Index using methods such as SVDPCA, Nipals PCA, Probabilistic PCA, and FastHCS.

5. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 4, characterized in that, The EvlRNA-Index calculation process includes the following steps: S1: The plasma sample is calculated using the method described above; S2: Each plasma sample was scored using 10 EvlRNA-Index; S3: Each plasma sample generates a 1×10 matrix, which is equivalent to 10 EvlRNA-Index information.

6. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 5, characterized in that, The first-layer classifier can be one or more of the following algorithms: Support Vector Machine (SVM), Random Forest (RF), Deep Learning (DL), and XGBoost, combined with the method used for the EvlRNA-Index index to achieve an algorithm combination. The algorithm combination is selected through the training dataset, the independent test dataset, and 5-fold cross-validation.

7. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 5, characterized in that, The second-layer classifier can be one or more of the following algorithms: Support Vector Machine (SVM), Random Forest (RF), Deep Learning (DL), and XGBoost, combined with the method used for the EvlRNA-Index index to achieve an algorithm combination. The algorithm combination is selected by combining the training dataset, the independent test dataset, and 5-fold cross-validation.

8. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 5, characterized in that, The training dataset S comprises a training set and a test set, accounting for 70% and 30% respectively. The independent test dataset consists of multiple normal samples and non-normal samples randomly selected from each other. These randomly selected samples are not used for model training, feature selection, or parameter optimization.

9. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 5, characterized in that, The first-layer classifier uses a combination of Probabilistic PCA and Support Vector Machine (SVM) algorithms.

10. A novel method for plasma extracellular vesicle long RNA analysis and early cancer diagnosis according to claim 5, characterized in that, The second-layer classifier uses a combination of Probabilistic PCA and Random Forest (RF) algorithms.

Images