Method and device for identifying tumor-induced nucleated red blood cells

Abstract

The application provides a method and device for identifying nucleated red blood cells induced by tumors, and relates to the technical field of bioinformatics.The method for identifying nucleated red blood cells induced by tumors provided by the application can efficiently identify the nucleated red blood cells induced by tumors in the peripheral blood of a patient in small quantities by using high-throughput transcriptome data combined with machine learning technology, has higher sensitivity and specificity, and lays a foundation for the role of nucleated red blood cells induced by tumors in biological processes.The device for identifying nucleated red blood cells induced by tumors provided by the application can efficiently identify nucleated red blood cells induced by tumors, has high sensitivity and specificity, can be used for auxiliary diagnosis and prognosis prediction of tumors, provides more accurate auxiliary information for personalized treatment of tumors, and has good clinical application potential.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics, and in particular to a method and apparatus for identifying tumor-induced nucleated red blood cells. Background Technology

[0002] Human red blood cells retain their nuclei in their immature stage, and are called nucleated red blood cells (NRBCs). In healthy adults, homeostatic erythropoiesis occurs in the bone marrow (BM), producing approximately 2-3 million red blood cells per second. These cells differentiate in the bone marrow, denucleate, and are released into the peripheral blood; therefore, nucleated red blood cells are almost undetectable in the peripheral blood of healthy individuals. However, in certain disease states, such as when bone marrow hematopoiesis is impaired, the production site of blood cells migrates to organs outside the bone marrow, especially organs involved in fetal hematopoiesis. This phenomenon is called extramedullary hematopoiesis (EMH). EMH can be observed in infectious diseases, hematologic malignancies, and some solid tumors. In a study of tumor-bearing mice, it was found that the cells produced by extramedullary hematopoiesis in the spleen were mainly nucleated red blood cells. Extramedullary hematopoiesis is considered a response to hematopoietic stress and a way to replenish the blood cells needed by the body. However, in tumor research, it has been found that extramedullary hematopoiesis in patients with advanced tumors produces nucleated red blood cells with immunosuppressive capabilities, which weaken the patient's anti-tumor immunity.

[0003] Tumor-induced nucleated erythrocytes are closely related to patient prognosis. In a study involving 28 HCC samples, researchers performed immunofluorescence staining on liver cancer tissue sections and observed CD45+CD71+ nucleated erythrocytes infiltrating the liver cancer tissue. The abundance of CD45+CD71+ erythroid cells in the tumor tissue was found to be more accurate than the CLIP (Cancer of the Liver Italian Program Score) in predicting disease-free survival and overall survival. The CLIP grading system is a grading standard developed based on the combined clinical and pathological characteristics of liver cancer patients, mainly including indicators such as the degree of liver function impairment, tumor size, lymph node metastasis, and serum alpha-fetoprotein (AFP) levels. Based on different combinations of these indicators, liver cancer patients are classified into different grades. Tumor-induced nucleated erythrocytes have a strong immunosuppressive capacity, and their quantity is correlated with the effectiveness of immunotherapy. Studies have shown that patients with tumor-induced nucleated erythrocytes (CD45+ CD235a+ CD71+ CD11b+ CD33+ HLA-DR+) in CD45+ PBMCs, after receiving ICI (Immune checkpoint inhibitors) treatment, achieved partial remission in 57.14% of patients in the low EPC (Erythroid Progenitor Cell) level group, but only 20% of patients in the high EPC level group achieved remission.

[0004] The peripheral blood of cancer patients contains very few nucleated erythrocytes, and their sorting markers are unclear. This makes it difficult to capture a comprehensive picture of nucleated erythrocytes in cancer patients using flow cytometry (FACS) and bulk RNA-seq technologies. Single-cell sequencing technology can capture the transcriptomic features of rare cell types and identify continuous and complex changes in cell states. Single-cell transcriptome sequencing requires few experimental materials and offers high resolution. Its application makes it possible to identify the characteristics of rare cells under specific pathological conditions, and it is particularly suitable for studying the differences and molecular characteristics of nucleated erythrocytes in different states.

[0005] Accurate cell type identification is a prerequisite for studying the roles of different cells in biological processes. Over the past few decades, researchers have used techniques such as flow cytometry to sort different cell types and combined this with transcriptomics and proteomics to obtain cell markers. These marker genes are used for cell type annotation in single-cell data. Cell population identification typically relies on unsupervised cell clustering; however, the choice of clustering method and resolution affects the number of cell clusters formed, and manually selecting cell type-specific marker genes is time-consuming, subjective, and prone to errors. Currently, single-cell sequencing studies are becoming increasingly large-scale, with sample sizes often reaching hundreds. The traditional cell type annotation methods mentioned above are clearly insufficient to meet the needs of research analysis; therefore, automated cell type annotation methods are receiving increasing attention.

[0006] In view of this, the present invention is hereby proposed. Summary of the Invention

[0007] The primary objective of this invention is to provide a method for identifying tumor-induced nucleated erythrocytes, thereby laying the foundation for research on the role of tumor-induced nucleated erythrocytes in biological processes and solving the aforementioned technical problems.

[0008] A second objective of the present invention is to provide an apparatus for identifying tumor-induced nucleated red blood cells.

[0009] A third objective of this invention is to provide the application of the above-described apparatus for identifying tumor-induced nucleated red blood cells in the preparation of products for the auxiliary diagnosis of tumors.

[0010] To achieve the above objectives, the following technical solution is adopted:

[0011] In a first aspect, the present invention provides a method for identifying tumor-induced nucleated erythrocytes, comprising the following steps:

[0012] a. Obtain mouse single-cell transcriptome and bulk data, wherein the mouse single-cell transcriptome and bulk data include gene expression data of nucleated erythrocytes induced by tumor and gene expression data of nucleated erythrocytes of healthy mice;

[0013] b. The mouse single-cell transcriptome data is pseudo-bulk processed and then merged with the bulk data; the genes in the merged data are homologously converted to retain homologous genes between humans and mice, and then the intersection with the genes in the human single-cell transcriptome data is taken. Candidate genes that have a significant role in identifying tumor-induced nucleated erythrocytes are screened from the intersection genes; the human single-cell transcriptome data includes single-cell transcriptome data from peripheral blood of cancer patients and single-cell transcriptome data from bone marrow or peripheral blood of non-cancer patients;

[0014] c. Using the candidate genes screened in step b as features, a classification model is trained using the mouse single-cell transcriptome data. Then, the trained classification model is used to classify the human single-cell transcriptome data to identify tumor-induced nucleated red blood cells.

[0015] As a further technical solution, the mouse single-cell transcriptome data were pseudo-bulk processed using the AverageExpression function of the Seurat package.

[0016] As a further technical solution, the homologous gene of mice is homologously converted using the homologgene package.

[0017] As a further technical solution, the Boruta algorithm is used to screen candidate genes from the intersection genes that have a significant role in identifying tumor-induced nucleated red blood cells.

[0018] As a further technical solution, the classification model includes a random forest model.

[0019] In a second aspect, the present invention provides an apparatus for identifying tumor-induced nucleated red blood cells, comprising an acquisition module and a classification module;

[0020] The acquisition module is used to acquire single-cell transcriptome data of the subject.

[0021] The classification module is used to input the human single-cell transcriptome data into a pre-trained classification model, and to classify the human single-cell transcriptome data through the classification model to identify tumor-induced nucleated red blood cells.

[0022] The classification model was trained using the following method:

[0023] a. Obtain mouse single-cell transcriptome and bulk data, wherein the mouse single-cell transcriptome and bulk data include gene expression data of nucleated erythrocytes induced by tumor and gene expression data of nucleated erythrocytes of healthy mice;

[0024] b. The mouse single-cell transcriptome data is pseudo-bulk processed and then merged with the bulk data; the genes in the merged data are homologously converted to retain homologous genes between humans and mice, and then the intersection with the genes in the human single-cell transcriptome data is taken. Candidate genes that have a significant role in identifying tumor-induced nucleated erythrocytes are screened from the intersection genes; the human single-cell transcriptome data includes single-cell transcriptome data from peripheral blood of cancer patients and single-cell transcriptome data from bone marrow or peripheral blood of non-cancer patients;

[0025] c. Using the candidate genes selected in step b as features, train a classification model using the mouse single-cell transcriptome data to obtain a trained classification model.

[0026] As a further technical solution, the mouse single-cell transcriptome data were pseudo-bulk processed using the AverageExpression function of the Seurat package.

[0027] As a further technical solution, the homologous gene of mice is homologously converted using the homologgene package.

[0028] As a further technical solution, the Boruta algorithm is used to screen out candidate genes from the intersection genes that have a significant role in identifying tumor-induced nucleated red blood cells;

[0029] And / or, the classification model includes a random forest model.

[0030] Thirdly, this invention provides the application of the above-described device for identifying tumor-induced nucleated red blood cells in the preparation of products for auxiliary diagnosis of tumors.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The method for identifying tumor-induced nucleated erythrocytes provided by this invention utilizes high-throughput transcriptome data combined with machine learning technology. This method can efficiently identify the sparse number of tumor-induced nucleated erythrocytes in the peripheral blood of patients, exhibiting higher sensitivity and specificity. This lays the foundation for research on the role of tumor-induced nucleated erythrocytes in biological processes. The potential applications of this method include understanding the role and regulatory mechanisms of tumor-induced nucleated erythrocytes in tumor immune responses, their predictive ability for disease progression, and their role in drug target identification.

[0033] The device for identifying tumor-induced nucleated red blood cells provided by this invention can efficiently identify tumor-induced nucleated red blood cells with high sensitivity and specificity. It can be used for the auxiliary diagnosis and prognosis prediction of tumors, providing more accurate auxiliary information for personalized treatment of tumors, and has good potential for clinical application. Attached Figure Description

[0034] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0035] Figure 1This is a schematic diagram of characteristic gene selection;

[0036] Figure 2 A schematic diagram for training a random forest classifier;

[0037] Figure 3 The confusion matrix describes the accuracy results of the random forest classifier;

[0038] Figure 4 To illustrate the classification results of the random forest classifier using the Sankey diagram;

[0039] Figure 5 To illustrate the data integration and annotation results in the UMAP diagram;

[0040] Figure 6 To illustrate the NRBC clustering results using a UMAP plot;

[0041] Figure 7 The UMAP plot shows the results of the random forest classifier classifying human nucleated red blood cells;

[0042] Figure 8 The UMAP plot shows the clustering results of nucleated erythrocytes;

[0043] Figure 9 Use Sankey diagrams to illustrate the relationship between sample types, prediction results, and clusters. Detailed Implementation

[0044] The embodiments and examples of the present invention will be described in detail below. However, those skilled in the art will understand that the following embodiments and examples are for illustrative purposes only and should not be considered as limiting the scope of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Unless otherwise specified, conventional conditions or conditions recommended by the manufacturer shall apply. Reagents or instruments whose manufacturers are not specified are all commercially available conventional products.

[0045] In a first aspect, the present invention provides a method for identifying tumor-induced nucleated erythrocytes, comprising the following steps:

[0046] a. Obtain mouse single-cell transcriptome and bulk data, wherein the mouse single-cell transcriptome and bulk data include gene expression data of nucleated erythrocytes induced by tumor and gene expression data of nucleated erythrocytes of healthy mice;

[0047] b. The mouse single-cell transcriptome data is pseudo-bulk processed and then merged with the bulk data; the genes in the merged data are homologously converted to retain homologous genes between humans and mice, and then the intersection with the genes in the human single-cell transcriptome data is taken. Candidate genes that have a significant role in identifying tumor-induced nucleated erythrocytes are screened from the intersection genes; the human single-cell transcriptome data includes single-cell transcriptome data from peripheral blood of cancer patients and single-cell transcriptome data from bone marrow or peripheral blood of non-cancer patients;

[0048] c. Using the candidate genes screened in step b as features, a classification model is trained using the mouse single-cell transcriptome data. Then, the trained classification model is used to classify the human single-cell transcriptome data to identify tumor-induced nucleated red blood cells.

[0049] It should be noted that, in this invention, normal nucleated red blood cells include nucleated red blood cells from the spleen of mice with PHZ (Phenylhydrazine)-induced anemia and nucleated red blood cells from the bone marrow of healthy mice.

[0050] The method for identifying tumor-induced nucleated red blood cells provided by this invention utilizes high-throughput transcriptome data combined with machine learning technology to efficiently identify the sparse number of tumor-induced nucleated red blood cells in the peripheral blood of patients. It has higher sensitivity and specificity, laying the foundation for research on the role of tumor-induced nucleated red blood cells in biological processes.

[0051] In some alternative implementations, the single-cell transcriptome data of bone marrow or peripheral blood from individuals without tumors includes single-cell transcriptome data of bone marrow or peripheral blood from healthy individuals, and single-cell transcriptome data of bone marrow or peripheral blood from frail elderly individuals.

[0052] In some optional implementations, during step c, the candidate genes are simultaneously optimized to obtain a core feature gene set during the classification model training process.

[0053] In some optional implementations, the core feature gene set includes: Fcer1g, Tyrobp, Apoe, H2-Ab1, Plac8, Lgals3, Id2, Fos, Sat1, Cebpb, Ifitm2, Lsp1, Rhd, Actb, Vim, Tspan33, Gm2a, Cd9, Urod, Slc43a1, Alad, Ctsc, Ctsh, Uros, Ifitm1, Junb, Hemgn, Tyms, Rgs2, Klf6, Hmgb3, Napsa, Gngt2, Lig1, Marksl1, Tk1, Gmnn, Hdgf, Dusp1, Is g20, Tspo2, Cenpq, Prkar2b, Chac2, Sepp1, Mt1, Nuf2, Cdc6, Arhgdib, Ckap2l, Hmbs, Pld4, Ccne1, Ccne2, Bub1, Selplg, Grn, Comt, Cdc42ep3, Acp1, Incenp, Ermap, Cdkn3, Pbk, Cdc20, Prdx2, Blvrb, Rad51ap1, Mcm4, Ly6e, Abcf1, Rfc1, Mad2l1, Nr4a1, Fn3k, Plk1, Dtl, Ska1, Vars, Tmem14c, Aurkb, Ect2, Lmna, Rbbp7, Klf4, Mki67, Bst2, Cmtm7, Gpi1, Clec12a, Gpsm3, B2m, Ninj1, Gadd45a, Rbmxl1, Tpd52, Anp32b, Gadd45b, Man2b1, Prnp, Pycard, Ypel4, Rgs12, Ctsb, Gpsm2, Fyb, Minpp1, Mpeg1, Rnf149, Canx, Fkbp1a, Ier5, Smarca4, Tuba1b, Psmb9, Slfn14, Snrpd1, Vamp5, Evi2a, Emb, Zeb2, Nr ros, Cd81, Ppia, Svip, Clec4a2, Mgst1, Tuba1a, Zyx, Capg, Plk2, Cdca7, Ccnf, Cct7, Lsm4, Eef1b2, Cd302, Aldh2, Nmral1, Ppid, H2afz, Sub1, Gapdh , Tmed3, Nfkbid, Ifngr2, Fcgr1, AI607873, Arrb2, Dapk2, Ahnak, Clec4d, Ube2e1, Tap1, Snca, Hmgb2, Pgls, Tbcel, Anxa3, Pilrb2, Bcl11a and Mrpl33.

[0054] In some alternative implementations, the mouse single-cell transcriptome data are pseudo-bulk processed using the AverageExpression function of the Seurat package.

[0055] In some alternative implementations, the homologous conversion of mouse genes is performed using the homologgene package.

[0056] In some alternative implementations, the Boruta algorithm is used to screen candidate genes from the intersection genes that have a significant role in identifying tumor-induced nucleated red blood cells.

[0057] In some alternative implementations, the classification model includes a random forest model.

[0058] In some alternative implementations, in mouse single-cell transcriptome data, tumor-induced nucleated erythrocytes are identified from cells annotated as nucleated erythrocytes.

[0059] In a second aspect, the present invention provides an apparatus for identifying tumor-induced nucleated red blood cells, comprising an acquisition module and a classification module;

[0060] The acquisition module is used to acquire single-cell transcriptome data of the subject.

[0061] The classification module is used to input the human single-cell transcriptome data into a pre-trained classification model, and to classify the human single-cell transcriptome data through the classification model to identify tumor-induced nucleated red blood cells.

[0062] The classification model was trained using the following method:

[0063] a. Obtain mouse single-cell transcriptome and bulk data, wherein the mouse single-cell transcriptome and bulk data include gene expression data of nucleated erythrocytes induced by tumor and gene expression data of nucleated erythrocytes of healthy mice;

[0064] b. The mouse single-cell transcriptome data is pseudo-bulk processed and then merged with the bulk data; the genes in the merged data are homologously converted to retain homologous genes between humans and mice, and then the intersection with the genes in the human single-cell transcriptome data is taken. Candidate genes that have a significant role in identifying tumor-induced nucleated erythrocytes are screened from the intersection genes; the human single-cell transcriptome data includes single-cell transcriptome data from peripheral blood of cancer patients and single-cell transcriptome data from bone marrow or peripheral blood of non-cancer patients;

[0065] c. Using the candidate genes selected in step b as features, train a classification model using the mouse single-cell transcriptome data to obtain a trained classification model.

[0066] The device for identifying tumor-induced nucleated red blood cells provided by this invention, based on the above-mentioned method of this invention, can efficiently identify tumor-induced nucleated red blood cells with high sensitivity and specificity. It can be used for auxiliary diagnosis and prognostic prediction of tumors, providing more accurate auxiliary information for personalized treatment of tumors, and has good clinical application potential.

[0067] In some alternative implementations, the mouse single-cell transcriptome data are pseudo-bulk processed using the AverageExpression function of the Seurat package.

[0068] In some alternative implementations, the homologous conversion of mouse genes is performed using the homologgene package.

[0069] In some alternative implementations, the Boruta algorithm is used to screen candidate genes from the intersection genes that have a significant role in identifying tumor-induced nucleated red blood cells.

[0070] In some alternative implementations, the classification model includes a random forest model.

[0071] In some alternative implementations, in mouse single-cell transcriptome data, tumor-induced nucleated erythrocytes are identified from cells annotated as nucleated erythrocytes.

[0072] Thirdly, this invention provides the application of the above-described device for identifying tumor-induced nucleated red blood cells in the preparation of products for auxiliary diagnosis of tumors.

[0073] The device provided by this invention can be used for the auxiliary diagnosis of tumors by identifying nucleated red blood cells induced by tumors.

[0074] The present invention will be further illustrated below with specific embodiments. However, it should be understood that these embodiments are merely for the purpose of more detailed illustration and should not be construed as limiting the present invention in any way.

[0075] The data used in the following examples are as follows:

[0076] Mouse single-cell transcriptome datasets GSE201715 and E-MTAB-8630, transcriptome datasets GSE109429 and GSM1208164, and human single-cell transcriptome data were collected.

[0077] Dataset GSE201715: Tumor-induced nucleated erythroid precursor cells (TEPC) produced in the spleen of tumor-bearing mice and PHZ-induced anemia-induced erythroid precursor cells (AEPC) in the spleen of mice.

[0078] PHZ-induced anemia in mice: C57BL / 6 mice were injected intraperitoneally with 200 μL of 50 mg / kg phenylhydrazine hydrochloride solution in PBS; spleen samples were collected three days later.

[0079] Dataset E-MTAB-8630: Erythroid precursor cells from bone marrow of healthy mouse (HEPC) in C57BL / 6 healthy mice.

[0080] Dataset GSE109429: Nucleated red blood cells from tumor-bearing mice.

[0081] Dataset GSM1208164: Nucleated red blood cells from the bone marrow of healthy mice.

[0082] Human single-cell transcriptome data: Peripheral blood single-cell transcriptome data from patients with 11 types of cancer (GSE169246: triple-negative breast cancer, GSE111014: chronic lymphocytic leukemia, GSE127465: non-small cell lung cancer, GSE188632: multiple myeloma, GSE155698: pancreatic ductal adenocarcinoma, GSE164522: colorectal cancer with liver metastasis, GSE107747: hepatocellular carcinoma (HCC), GSE117988 and GSE118056: Merkel tumor (MCC), GSE162025: nasopharyngeal carcinoma, GSM5930045: testicular seminoma, GSE162708: pancreatic neuroendocrine tumor (pNET), peripheral blood single-cell datasets from 15 healthy individuals (including frail elderly individuals), and single-cell transcriptome data from bone marrow of healthy individuals.

[0083] Example 1

[0084] Figure 1 A schematic diagram of characteristic gene selection, such as Figure 1As shown, nucleated erythrocytes were first extracted from the collected mouse single-cell transcriptome datasets GSE201715 and E-MTAB-8630 based on author annotations. Then, pseudo-bulk processing was performed using the AverageExpression function in the Seurat package, treating it as bulk data, and this was merged with the bulk datasets GSE109429 and GSM1208164 for analysis. Homologous gene conversion was performed on the genes in the data using the homologgene package, retaining only human and mouse homologous genes. The intersection of the converted homologous genes with the homologous genes in the human single-cell dataset was then calculated. Finally, the Boruta algorithm was used to calculate candidate genes that significantly contribute to nucleated erythrocyte classification, retaining 891 genes initially identified as tentative.

[0085] NEPC (normal nucleated red blood cells) contains nucleated red blood cells from the spleen of mice with PHZ (Phenylhydrazine)-induced anemia and nucleated red blood cells from the bone marrow of healthy mice. After preliminary experiments, it was found that nucleated red blood cells from the spleen of anemic mice and nucleated red blood cells from healthy mice, although they come from different sources, have similar expression profiles, so they are classified as normal nucleated red blood cells.

[0086] Figure 2 A schematic diagram for training a random forest classifier, as shown below. Figure 2 As shown, the collected mouse single-cell datasets GSE201715 and E-MTAB-8630 were annotated into two classes: TEPC and NEPC. Only the 891 identified genes were retained as features. A random forest model (ntree = 600, sampsize = 200) was built using the randomForest package. 9,764 mouse cells from the GSE201715 and E-MTAB-8630 datasets were input into the model. A portion of the dataset was randomly sampled; blue represents the training set, and gray represents the validation set. The model was trained 10 times using 10-fold cross-validation, ultimately optimizing to 162 core feature genes. The specific genes are as follows:

[0087] Fcer1g, Tyrobp, Apoe, H2-Ab1, Plac8, Lgals3, Id2, Fos, Sat1, Cebpb, Ifitm2, Lsp1, Rhd, Actb, Vim, Tspan33, Gm2a, Cd9, Urod, Slc43a1, Alad, Ctsc, Ctsh, Uros, Ifitm1, Junb, Hemgn, Tyms, Rgs2, Klf6, Hmgb3, Napsa, Gngt2, Lig1, Marcksl1, Tk1, Gmnn, Hdgf, Dusp1, Isg20, Tspo2, Cenpq, Prkar2b, Chac2, Sepp1, Mt1, Nuf2, Cdc6, Arhgdib, Ckap2l, Hmbs, Pld4, Ccne1, Ccne2, Bub1, Selplg, Grn, Comt, Cdc42ep3, Acp1, Incenp, Ermap, Cdkn3, Pbk, Cdc20, Prdx2, Blvrb, Rad51ap1, Mcm4, Ly6e, Abcf1, Rfc1, Mad2l1, Nr4a1, Fn3k, Plk1, Dtl, Ska1, Vars, Tmem14c, Aurkb, Ect2, Lmna, Rbbp7, Klf4, Mki67, Bst2, Cmtm7, Gpi1, Clec12a, Gpsm3, B2m, Ninj1, Gadd45a, Rbmxl1, Tpd52, Anp32b, Gadd45b, Man2b1, Prnp, Pycard, Ypel4, Rgs12, Ctsb, Gpsm2, Fyb, Minpp1, Mpeg1, Rnf149, Canx, Fkbp1a, Ier5, Smarca4, Tuba1b, Psmb9, Slfn14, Snrpd1, Vamp5, Evi2a, Emb, Zeb2, Nrros, Cd81, Ppia, Svip, Clec4a2, Mgst1, Tuba1a, Zyx, Capg, Plk2, Cdca7, Ccnf, Cct7, Lsm4, Eef1b2, Cd302, Aldh2, Nmral1, Ppid, H2afz, Sub1, Gapdh, Tmed3, Nfkbid, Ifngr2, Fcgr1, AI607873, Arrb2, Dapk2, Ahnak, Clec4d, Ube2e1, Tap1, Snca, Hmgb2, Pgls, Tbcel, Anxa3, Pilrb2, Bcl11a and Mrpl33.

[0088] The mouse nucleated erythrocyte datasets GSE201715 and E-MTAB-8630 were used for 10-fold cross-validation of the model. These datasets contain nucleated erythrocytes from three different sources: bone marrow erythroid precursor cells (HEPC) from healthy C57BL / 6 mice, spleen erythroid precursor cells (TEPC) from tumor-bearing mice, and spleen erythroid precursor cells (AEPC) from mice with PHZ-induced anemia. Results are as follows... Figure 3 and Figure 4 As shown.

[0089] Figure 3 As shown in the confusion matrix, the classifier of this invention can distinguish between TEPCs and normal erythroid progenitor cells (NEPCs). The accuracy of the classifier in identifying TEPCs in mice is 98.01%.

[0090] Figure 4 To illustrate the classification results of the random forest classifier using the Sankey diagram, such as... Figure 4 As shown, a small number of cell types were misclassified in both TEPC and AEPC, but all nucleated erythrocytes from the bone marrow of healthy mice were correctly classified into NEPC. It's important to note that the random forest classifier was initially designed to classify nucleated erythrocytes into three categories: TEPC, AEPC, and HEPC. Nucleated erythrocytes from extramedullary hematopoiesis sites and bone marrow differ due to their different developmental environments, and theoretically, AEPC should be distinguishable from HEPC in the bone marrow. However, during both single-cell data integration and attempts to build the random forest classifier, it was found that HEPC and AEPC were mixed together and could not be distinguished well. Therefore, the random forest classifier was subsequently designed as a binary classification system for TEPC and NEPC.

[0091] Example 2

[0092] After quality control of each sample from peripheral blood single-cell datasets collected from 11 types of cancer patients and 15 healthy individuals (including frail elderly individuals), Celltypist was used for automated annotation. Then, using the sample's GEO Accession as batch information, the Harmony algorithm was employed to integrate the samples and remove batch effects between different datasets. The results are as follows: Figure 5 As shown in the figure, the data integration and annotation results are presented. It can be seen from the figure that the Celltypist annotation results are basically accurate, and Harmony clusters cells of the same type well. This indicates that NRBCs in peripheral blood can be detected using single-cell transcriptome sequencing technology, which lays the foundation for subsequent identification of TEPCs.

[0093] Literature reports the presence of nucleated erythrocytes (NRBCs) in the blood of frail elderly individuals, while the bone marrow is the site of erythrocyte differentiation and development under homeostasis. To test the accuracy of the classifier, we extracted NRBCs from single-cell transcriptome data of peripheral blood from 11 types of cancer patients, single-cell peripheral blood data from frail elderly individuals (derived from single-cell peripheral blood datasets of 15 healthy individuals), and single-cell transcriptome data of bone marrow from healthy individuals. We filtered these NRBCs according to the expression of cell type markers, retaining only cells expressing erythroid markers. Finally, we integrated, clustered, and grouped the data based on gene expression levels in the cells. Ultimately, we obtained 1,531 NRBCs from bone marrow samples from healthy adults, 319 NRBCs from peripheral blood samples from frail elderly individuals, and 704 NRBCs from peripheral blood samples from cancer patients. After dimensionality reduction and clustering, the samples were divided into four subgroups, as shown below. Figure 6 As shown, this indicates that these nucleated red blood cells exhibit heterogeneity.

[0094] The classifier provided by this invention is used to classify nucleated red blood cells in the above data, such as... Figure 7 As shown, when the classifier was used to identify TEPC induced by human tumors, we found that the random forest classifier could accurately distinguish between TEPC and NEPC.

[0095] The classifier results ( Figure 7 ) and previous clustering results ( Figure 6 By combining these methods, clustering results of nucleated erythrocytes can be obtained, such as... Figure 8 As shown, human nucleated red blood cells are divided into four clusters: N_C1, N_C2, N_C3 and T_C4, where N represents normal and T represents tumor.

[0096] Single-cell transcriptome data from peripheral blood of 11 types of cancer patients, 15 single-cell datasets from peripheral blood of healthy individuals (including frail elderly individuals), and single-cell transcriptome data from bone marrow of healthy individuals were collected and classified using the random forest classifier constructed above. The results were represented using a Sankey diagram. Figure 9 (Not all tumor types are shown in the figure because nucleated red blood cells were not identified in the unshown tumors.) As shown, TEPC mainly comes from hepatocellular carcinoma (HCC), multiple myeloma (MM), and Merkel tumor (MCC), while NRBC in peripheral blood of patients with pancreatic neuroendocrine tumors (pNET) are mostly NEPC. This indicates that cancer patients may also produce a large number of nucleated red blood cells in their peripheral blood due to physical weakness, etc. These nucleated red blood cells are normal and not tumor-induced. Nucleated red blood cells in the frail sample from elderly and frail individuals and in the normal-BM bone marrow sample from healthy individuals are almost entirely NEPC. TEPC is concentrated in the T_C4 subset, therefore we believe that the T_C4 subset represents tumor-induced NRBC.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying tumor-induced nucleated red blood cells for purposes other than disease diagnosis, characterized by, The method comprises the following steps: a. obtaining mouse single-cell transcriptome and bulk data, wherein the mouse single-cell transcriptome and bulk data both comprise gene expression data of nucleated red blood cells induced by tumors and gene expression data of nucleated red blood cells of healthy mice; b. performing pseudo-bulk processing on the mouse single-cell transcriptome data, and then merging the mouse single-cell transcriptome data with the bulk data; performing homologous conversion on genes in the merged data, retaining homologous genes of humans and mice, and then taking an intersection of the genes in the human single-cell transcriptome data, and screening candidate genes having a significant effect on identifying nucleated red blood cells induced by tumors from the intersection genes; the human single-cell transcriptome data comprises single-cell transcriptome data of peripheral blood of tumor patients and single-cell transcriptome data of bone marrow or peripheral blood of people without tumors; c. using the candidate genes screened in the step b as features, training a classification model by using the mouse single-cell transcriptome data, and then performing classification processing on the to-be-detected human single-cell transcriptome data by using the trained classification model, to identify nucleated red blood cells induced by tumors; pseudo-bulk processing is performed on the mouse single-cell transcriptome data by using the AverageExpression function of the Seurat package; homologous conversion is performed on the genes of mice by using the homologene package; Boruta algorithm is used to screen candidate genes having a significant effect on identifying nucleated red blood cells induced by tumors from the intersection genes; the classification model comprises a random forest model.

2. An apparatus for identifying nucleated red blood cells induced by a tumor, characterized by, The method comprises an obtaining module and a classification module; the obtaining module is configured to obtain to-be-detected human single-cell transcriptome data; the classification module is configured to input the human single-cell transcriptome data into a pre-trained classification model, perform classification processing on the human single-cell transcriptome data by using the classification model, and identify nucleated red blood cells induced by tumors; the classification model is trained by using the following method: a. obtaining mouse single-cell transcriptome and bulk data, wherein the mouse single-cell transcriptome and bulk data both comprise gene expression data of nucleated red blood cells induced by tumors and gene expression data of nucleated red blood cells of healthy mice; b. performing pseudo-bulk processing on the mouse single-cell transcriptome data, and then merging the mouse single-cell transcriptome data with the bulk data; performing homologous conversion on genes in the merged data, retaining homologous genes of humans and mice, and then taking an intersection of the genes in the human single-cell transcriptome data, and screening candidate genes having a significant effect on identifying nucleated red blood cells induced by tumors from the intersection genes; the human single-cell transcriptome data comprises single-cell transcriptome data of peripheral blood of tumor patients and single-cell transcriptome data of bone marrow or peripheral blood of people without tumors; c. using the candidate genes screened in the step b as features, training a classification model by using the mouse single-cell transcriptome data, and then performing classification processing on the to-be-detected human single-cell transcriptome data by using the trained classification model, to identify nucleated red blood cells induced by tumors; pseudo-bulk processing is performed on the mouse single-cell transcriptome data by using the AverageExpression function of the Seurat package; homologous conversion is performed on the genes of mice by using the homologene package; Boruta algorithm is used to screen candidate genes having a significant effect on identifying nucleated red blood cells induced by tumors from the intersection genes; the classification model comprises a random forest model. The mouse single-cell transcriptome data is subjected to pseudo-bulk processing by using the AverageExpression function of the Seurat package; The genes of the mouse are subjected to homologous conversion by using the homologene package; Boruta algorithm is used to screen candidate genes having a significant effect on the identification of tumor-induced nucleated red blood cells from the intersection genes; The classification model comprises a random forest model.

3. Use of the device for identifying tumor-induced nucleated red blood cells according to claim 2 in the preparation of a product for aiding the diagnosis of tumors.

Images