Single-cell transcriptome cell annotation method and system fused with large language model

Abstract

According to the single-cell transcriptome cell annotation method and system fused with the large language model, cell type annotation is carried out by constructing the proprietary cue word and utilizing the large language model, and the accuracy and universality of cell annotation are improved. The method has remarkable advantages on cell annotation of non-mode species, and an automatic and intelligent cell annotation process is realized.

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics analysis technology, and in particular relates to single-cell transcriptome data processing and cell type annotation technology, specifically a single-cell transcriptome cell annotation method and system integrating a large language model. Background Art

[0002] The development of single-cell transcriptome sequencing technology is changing with each passing day. Its detection throughput is constantly increasing and its cost is continuously decreasing. It has been widely used in basic biological research and clinical research. This technology can analyze gene expression profiles at the single-cell level, providing key data support for a deep understanding of cell heterogeneity, cell function, disease pathogenesis, etc. For example, in tumor research, single-cell transcriptome sequencing helps to reveal the gene expression characteristics of different cell types in the tumor microenvironment, providing a theoretical basis for the precise treatment of tumors.

[0003] However, cell type annotation is a basic step in single-cell transcriptome sequencing analysis, and existing cell annotation methods have many limitations. Manual annotation is a traditional cell annotation method, which mainly relies on experts to compare highly expressed genes in cell subpopulations with typical marker genes to determine cell types. The accuracy of this method is highly dependent on expert experience, and different experts may have different annotations on the same data. Moreover, the manual annotation process is extremely time-consuming and inefficient when facing large-scale single-cell transcriptome data. It is difficult to meet the needs of high-throughput data analysis, so it cannot become a standardized large-scale application method. The software annotation method determines the cell type by comparing the correlation between the new sequencing data and the reference data set. Although software annotation tools represented by singleR and cellassign can achieve efficient and automated annotation, they also have many problems. Their accuracy is severely restricted by the accuracy of the reference data set. Differences in reference data sets of different tissues or species may lead to inaccurate annotation results. For example, the reference set of liver cells is difficult to be directly applied to the annotation of small intestinal cells. In addition, if there is erroneous information in the reference data set, it will directly affect the annotation effect. With the continuous expansion of the application scope of single-cell sequencing, it takes a lot of manpower and time to build high-quality reference sets for various tissue samples.

[0004] Large language models (such as Chat GPT-4, Zhipu Qingyan, etc.) have made remarkable achievements in the field of natural language processing. Their training process integrates a large number of human scientific and technological paper libraries and has the ability to understand biological knowledge. Recent studies have shown that large language models are effective in the biomedical field, such as literature retrieval, knowledge question answering and other tasks. However, in the field of single-cell transcriptome cell annotation, its application is still in the exploratory stage. The knowledge related to cell marker genes contained in existing literature is mainly concentrated on common model species such as humans and mice, and the information of non-model species is seriously insufficient. Summary of the invention

[0005] The present invention aims to overcome the shortcomings of existing cell annotation methods, integrate large language models into the cell annotation analysis process, and improve the versatility of the model in different species through strategies such as homology annotation, and develop a more efficient, accurate and universal single-cell transcriptome cell annotation method and system.

[0006] The technical solution of the single-cell transcriptome cell annotation method integrating a large language model of the present invention is as follows.

[0007] Data preprocessing: Preprocess the single-cell transcriptome data obtained by new sequencing, including data cleaning, quality control and construction of gene expression matrix. Use advanced data quality control algorithms to remove low-quality data and outliers, and perform standardization processing such as normalization on the data to ensure the accuracy and comparability of the data, providing a high-quality data foundation for subsequent analysis.

[0008] Obtain and sort the up-regulated genes of each cell subpopulation: For each gene in a given subpopulation, use the Wilcoxon rank-sum test (a non-parametric test method that is mainly used to compare whether there is a significant difference in the distribution of two independent samples) to compare its expression value with other cell subpopulations to determine the genes that are significantly up-regulated in each cell subpopulation. The specific criteria are: the gene is expressed in at least 25% of the cells in the target subpopulation, the P value is ≤0.01, and the log2 fold change is ≥0.36. Sort the up-regulated genes of each subpopulation based on the results of the differential analysis, and prioritize sorting from small to large according to the P value. When the P values ​​are the same, further sorting from large to small using the difference fold will help screen out genes that have significant expression differences in cell subpopulations and may play a key role in cell type identification.

[0009] Homologous alignment of non-model species (if it is a non-model species): When the annotation target cell is a non-model species (such as non-human, mouse, Arabidopsis), homologous alignment is required to determine the gene symbol number (a gene identifier). If the target species is an animal, select human or mouse as the reference model species, and if it is a plant, select Arabidopsis as the reference model species. Use BLASTP software (a software developed by NCBI for similarity search in protein sequence databases) for homology alignment, and set the homology threshold to E-value (expected value) <1×10 -5, Identity (consistency) ≥ 0.8, sort the genes according to E-value, select the gene with the smallest E-value as the homologous gene of the source reference model species, and then define the gene symbol number (a gene identifier) ​​of the non-model species based on the symbol number of the reference model species gene. Through precise homology comparison algorithm, accurately screen out homologous genes and correctly define the gene symbol number of the non-model species, so that the gene information of the non-model species can be effectively mapped to the gene system of the reference model species. This is crucial for mapping the gene information of non-model species to the known gene system of model species, and helps to use the knowledge resources of model species to annotate cells of non-model species.

[0010] Construct proprietary prompt words: record information such as species, tissues, and subgroup up-regulated ranking marker gene names as variables, define input and output formats, and construct prompt words for submission to the large language model. For model species (human, mouse, Arabidopsis), select the 10 top-ranked genes for each subgroup to construct prompt words, and for non-model species, select the 20 top-ranked genes for each subgroup to construct prompt words. The constructed prompt words have an accurate format and complete variable information, which can clearly and accurately convey the gene expression characteristics of cell subgroups to the large language model, and guide the large language model to make accurate cell type judgments. For example, taking human small cell lung cancer as an example, the constructed prompt word format is: "You are a biologist, and you are very good at cell annotation." (Chinese translation: You are a biologist and are very good at cell annotation).

[0011] Identify cell types of Human small cell lung cancer using the following markers separately for each row. Some can be a mixture of multiple cell types. (Chinese translation: Please use the markers listed in each row below to identify the cell types of human small cell lung cancer. Some may be a mixture of multiple cell types).

[0012] #The following is the data to be analyzed. One row represents onetype of cells and the corresponding cell markers. Cell ID is beforesemicolon, and gene markers is after semicolon.

[0013] 0: BTBD17, LINC01896, CALML3, TRIM9, GRP, KLK12, NPW, KIF19, DLL3,ESPL1.

[0014] 1: H3C2, H1-5, CENPA, DEPDC1, DLGAP5, UBE2C, PBK, ASPM, NUF2, SGO1.

[0015] 2: CD8A, CD8B, GZMK, TRGC2, CCL5, IFNG, GZMA, GZMH, LAG3, APOBEC3G.

[0016] 3: IL7R, LTB, CD6, P2RY10, ITK, SPOCK2, KLRB1, CD3E, TRAC, CD2.

[0017] 4: S100A12, FCN1, IL1B, CD300E, FOLR3, EREG, PTX3, LILRA5, CLEC4E, S100A8.

[0018] 5: SERPINB4, SERPINB3, MMP10, AQP5, BPIFB1, SLC6A14, SCGB1A1, MSLN,KRT17, SCGB3A1.

[0019] 6: CCL18, LILRB5, APOE, APOC1, C1QB, C1QA, MARCO, C1QC, OTOA, TREM2.

[0020] ## Please output the cell types of each row. The format is as follows. Please output the result directly without further explanation. (Chinese translation: ## Please output the cell types of each row. The format is as follows. Please output the result directly without further explanation).

[0021] 0 Monocytes (Chinese translation: monocytes).

[0022] 1 Neutrophil (Chinese translation: neutrophil).

[0023] 2 B cells” (Chinese translation: B cells).

[0024] Submit the large language model and determine the cell type: Submit the constructed prompt words to the large language model (such as Chat GPT-4) through the API (running the Chat GPT-4 application programming interface), and repeat the question and answer five times. Summarize the responses of the large language model, and determine the cell type and marker gene corresponding to each subpopulation based on the concept of majority in statistics. For example, if three out of five questions and answers answer that a subpopulation is "B cells", then the subpopulation is ultimately determined to be of the "B cell" type. By asking multiple questions and counting the majority of answers, the reliability and stability of cell type judgment is improved.

[0025] Furthermore, the present invention also proposes a single-cell transcriptome cell annotation system that integrates a large language model, including a data preprocessing module, a gene sorting and screening module, a homology comparison module (for non-model species), a prompt word construction module, a large language model interaction module, and a cell type determination module. The modules work together to achieve an automated and intelligent cell annotation process.

[0026] A single-cell transcriptome cell annotation system integrating a large language model of the present invention includes the following.

[0027] Data preprocessing module: responsible for receiving single-cell transcriptome data and performing quality control on the data, including removing low-quality data and outliers, to ensure the accuracy and reliability of the data. At the same time, the data is standardized to make the data of different samples comparable, providing a high-quality data foundation for subsequent analysis steps.

[0028] Gene sorting and screening module: realizes the above-mentioned function of obtaining and sorting up-regulated genes in each cell subpopulation, screens out significantly up-regulated genes according to the set criteria, and performs sorting operations. This module uses advanced statistical analysis algorithms to ensure the accuracy of gene screening and sorting, and provides key gene information for subsequent homologous comparison and prompt word construction.

[0029] Homology alignment module (for non-model species): When processing non-model species data, this module performs homology alignment operations. Corresponding to the above non-model species homology alignment steps, BLASTP software (a software developed by NCBI for similarity search in protein sequence databases) and other tools are used to perform efficient homology alignment, determine the symbol number (a gene identifier) ​​of the non-model species gene, and achieve the mapping of non-model species genes with model species genes, expanding the scope of application of cell annotation methods.

[0030] Prompt word construction module: Based on the species, tissue and other information provided by the data preprocessing module and the genes screened by the gene sorting and screening module, the module constructs proprietary prompt words according to predetermined rules. This module can flexibly adapt to different species and data characteristics and generate accurate and effective prompt words so that the large language model can accurately understand the data characteristics and task requirements.

[0031] Large language model interaction module: responsible for interacting with large language models (such as Chat GPT-4), submitting prompt words to the large language model and receiving its replies. Through the optimized API (running Chat GPT-4 application programming interface) calling method, it ensures efficient communication with the large language model, realizes multiple questions and answers and results aggregation, and provides a basis for accurate determination of cell types.

[0032] Cell type determination module: Based on the responses obtained by the large language model interaction module, the cell type is determined according to the concept of majority in statistics. This module counts and analyzes the output results of the large language model, integrates multiple question and answer results, and draws the most reliable conclusion on cell type determination, improving the accuracy and stability of cell annotation.

[0033] The beneficial effects of the single-cell transcriptome cell annotation method and system integrated with a large language model of the present invention are as follows.

[0034] 1. Improve annotation accuracy.

[0035] By optimizing the construction of prompt words and taking the mode of multiple questions and answers, the present invention can more accurately determine the cell type. Compared with traditional manual annotation, it reduces the subjective error of human judgment; compared with software annotation, it is not affected by the limitations of the reference data set and can more accurately identify cell types, especially when dealing with complex cell types or rare cell types.

[0036] 2. Enhance versatility.

[0037] Through the homologous comparison strategy, the method of the present invention can process single-cell transcriptome data of model species and non-model species, effectively solving the challenges faced by existing methods in cell annotation of non-model species, greatly expanding the scope of application of cell annotation methods, and making them have application value in biological research of different species.

[0038] 3. Realize automation and intelligence.

[0039] The automated and intelligent cell annotation process and the efficient processing capabilities of the large language model enable the present invention to quickly process large-scale single-cell transcriptome data. Compared with time-consuming manual annotation, it reduces the degree of human intervention, significantly shortens the analysis time, improves the annotation efficiency, and helps to process massive single-cell transcriptome data in a short time, meeting the needs of high-throughput data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of a single-cell transcriptome cell annotation method integrating a large language model according to the present invention is shown.

[0041] Figure 2 A flow chart of a single-cell transcriptome cell annotation system integrating a large language model of the present invention is shown. DETAILED DESCRIPTION

[0042] The present invention is described in detail below with reference to the embodiments.

[0043] Example 1: Single-cell annotation of mouse liver.

[0044] Data preparation and preprocessing: Collect mouse liver single-cell transcriptome data, including information such as gene expression values. Perform quality control on the data, remove low-quality cells and genes, and ensure the reliability of the data. Eliminate genes with extremely low expression or not expressed in a large number of cells, as well as cell data with possible technical errors. Perform differential analysis on the preprocessed data, calculate the expression difference of each gene in different cell subpopulations, and provide a data basis for subsequent gene screening and sorting.

[0045] Gene screening and sorting: According to the gene screening criteria in the above invention content, the Wilcoxon rank sum test is used to determine the genes that are significantly upregulated in each subgroup of mouse liver cells. During the screening process, the expression of each gene in different subgroups is strictly compared to ensure that the screened genes have significant subgroup-specific expression differences. The screened upregulated genes are sorted, preferably arranged from small to large according to the P value, and the genes with the same P value are sorted from large to small according to the difference multiple. For a certain subgroup, a group of ordered upregulated gene lists are obtained after sorting, among which the genes ranked at the top have higher expression specificity and larger expression difference multiples in the subgroup.

[0046] Prompt word construction: Extract the top 10 genes based on the subgroup gene ranking results and construct prompt words. During the construction process, accurately record variables such as mouse species, liver tissue, and gene information of each subgroup, and generate prompt words according to the predetermined format. The constructed prompt words are: “You are a biologist, and you are very good at cell annotation. Identify cell types of mouse Liver cells using the following markers separately for each row. Some can be a mixture of multiple cell types. #The following is the data to be analyzed. One row represents onetype of cells and the corresponding cell markers. Cell ID is beforesemicolon, and gene markers is after semicolon. 0: Gm46565, Lyve1, Bmp2, Oit3, Id4, Kdr, Stab2, Aass, Gm30938,4930578G10Rik 1: Vsig4, Timd4, C6, Cd5l, Clec4f, Cd207, Folr2, C1qa, Mmp12, Spic 2: Pax5, Fcmr, Ms4a1, Cr2, Ebf1, Chst3, Fcer2a, Cacna1i, Gm31243, Blk 3: Chil3, F13a1, Arhgef37, Sirpd, Ace, F10, Tppp3, Gpr141, Gm56663, Nupr1 4: Gja5, Nrg1, Adgrg6, Ednrb, Mecom, Vegfc, Sdc1, Gask1b, Jag1, Fbln2 5: Nsg2, Lef1, Sidt1, Cd8b1, Tcf7, Fgf13, Gm44174, Il7r, Ms4a4b, Gm2682 6: Izumo1r, Cd4, Ctla4, Tnfsf8, Icos, Cd5, Cd40lg, Slamf1, A430093F15Rik, Ikzf2 7: Stfa2l1, S100a8, S100a9, Il36g, Retnlg, Cxcr2, Slc7a11, Trem1, Dhrs9, Il1r2 8: Il4, Il12rb2, Cd40lg, Zbtb16, Kcnq5, Il2ra, Gramd3, Podnl1, Klrc2, Acoxl 9: Gzmk, Pdcd1, Eomes, Ccl5, Cd8a, Gm44174, Trgc2, Cd8b1, Tigit, Cst7 10: Gstm3, Bhmt, A2ml1, Rgn, Fabp2, Otc, Sult1d1, Akr1c6, Agmat, Glyat 11: Rspo3, Wnt9b, Bmp4, Lhx6, Selp, Vwf, Plppr5, Tgfb2, Rnf165, Cdh13 12: Cd209a, Mreg, Klri1, Il4i1, Clec10a, Flt3, Nectin1, Tnip3,Spint1, Tspan33 13: Pkhd1, Fermt1, Sftpd, Gm57079, Hnf4g, Gm19935, Sytl5, Cd200l1, Ankrd1, Nipal2 14: Bmp5, Pdgfra, Ednra, Pde3a, Tbx20, Tcf21, Nr1h5, Gdf2, Srpx2, Lamb1 15: Nefl, Stmn3, Gap43, Stmn2, Tubb3, Sncg, Tubb2b, mt-Atp6, mt-Cytb, Map1b 16: Klra6, Klra1, Klra9, Klra7, Rasgrf1, Itgad, Ccdc136, Gzmb, Cd160,Cd7 17: Esco2, Mxd3, Pclaf, Pimreg, Ska1, Pbk, Cenpf, Ccna2, Sgo1, Shcbp1 18: Zmat4, Ncr1, Klrb1a, Cd200r2, Klrb1b, Gm43647, Xcl1, Dscam, Klrb1c, Gm36723 19: Iglv1, Jchain, Tnfrsf17, Igha, Derl3, 5830418P13Rik, Igkc, Ccr10,Bhlha15, Eef1akmt3 20: Gm38410, Xcr1, Gcsam, Gpr141b, Snx22, Htr7, Tlr11, 4931406G06Rik, Ffar4, Mycl 21: Siglech, Gm21762, Klk1, Grm8, Cox6a2, Gm30605, Havcr1, Cd209d, Paqr5, Duxbl1 22: Skap1, Nkg7, Itk, Ccl5, Rpl37a, Cxcr6, Il2rb, Thy1, Gm2682, Cd226 23: Cma1, Klra4, Klra8, Klri2, Gzma, Adamts14, Gm43647, Prf1, Ncr1,Itga2 24: 6030498E09Rik, Fam20a, Nova2, Gmpr, Rgs7bp, Pde10a, Pakap, Ptch2,Klhl4, Gm57038 25: Chst4, Bnc1, Muc16, Upk3b, Myl7, Msln, Upk1b, Gm36535, Lrrn4,Pkhd1l1 26: Negr1, Fcrl5, Gm26620, Cd300e, 4931431B13Rik, Klra17, Adamdec1,Itgad, Fam89a, Sox5 27: Ebf1, Iglc2, Cd79a, Ighd, Ms4a1, Bank1, Pax5, Iglc3, Blk, Aff3 28: Ms4a2, Mcpt8, Cyp11a1, Fcer1a, Cd200r3, Cpa3, Il6, Gm32511,Alox15, Csrp3 ## Please output the cell types of each row. The format is asfollows. Please output the result directly without further explanation. 0 Monocytes 1 Neutrophil 2 B cells” (Chinese translation: B cells).

[0047] Submit the large language model and determine the results: Submit the prompt words to the Chat GPT-4 model through the API, and repeat the questioning 5 times independently. During the questioning process, ensure that the prompt words submitted each time are accurate and independent of the previous questions to obtain diverse model responses. Summarize the 5 responses of the model, and determine the cell type corresponding to each subpopulation based on the concept of majority in statistics. If 3 out of 5 responses of a subpopulation are judged as "Monocytes", the subpopulation is finally annotated as "Monocytes". The annotation results are sorted and analyzed, and compared with the manual annotation results. In the actual case of mouse liver single cell annotation, the results show that among the 28 subpopulations, 20 subpopulations are completely consistent with the manual annotation results, and another 3 subpopulations are manually annotated as unknown cells (unknow), but the large language model successfully returns the annotation results. This fully demonstrates the accuracy and effectiveness of the method of the present invention in mouse liver single cell annotation.

[0048] Mouse liver single cell annotation comparison table: Cell subset Manual annotation AI annotation result First output Second output Third output Fourth output Fifth output 0 Endothelial cell Endothelial cell Endothelial cell Endothelialcell Endothelialcell Endothelial cell Endothelialcell 1 Kupffercell Kupffercell Kupffercell Kupffercell Kupffercell Kupffercell Kupffercell 2 B cell B cell B cell B cell B cell B cell B cell 3 Macrophage Macrophage Macrophage Macrophage Macrophage Neutrophil Macrophage 4 Endothelial cell Endothelial cell Endothelial cell Endothelialcell Endothelialcell Endothelial cell Endothelialcell 5 T cell T cell T cell T cell T cell T cell T cell 6 T cell T cell T cell T cell T cell T cell T cell 7 Neutrophil Neutrophil Neutrophil Neutrophil Neutrophil Neutrophil Neutrophil 8 T cell T cell T cell T cell T cell T cell T cell 9 T cell T cell T cell T cell T cell T cell T cell 10 Hepatocytes Hepatocyte Hepatocyte Hepatocyte Hepatocyte Hepatocyte Hepatocyte 11 Endothelial cell Stellatecell Hepaticstellatecell Stellatecell Stellatecell Endothelial cell Stellatecell 12 Dendriticcell Dendriticcell Dendriticcell Dendriticcell Dendriticcell Dendritic cell Dendriticcell 13 Cholangiocytes Hepatocyte Hepatocyte Hepatocyte Hepatocyte Hepatocyte Hepatocyte 14 HSC Mesenchymal cell Mesenchymal stemcell Mesenchymalcell Mesenchymalcell Endothelial cell Mesenchymalcell 15 unkown Neuron Neuron Neuron Neuron Neuronalcell Neuron 16 NK cell NK cell Naturalkillercell NK cell NK cell NK cell NK cell 17 Diving_cell Cell cyclecell Proliferating cell Cell cyclecell Cell cyclecell Celldivisioncycle Cell cyclecell 18 NK cell NK cell Naturalkillercell NK cell NK cell NK cell NK cell 19 Plasma_B B cell B cell B cell B cell B cell B cell 20 Dendriticcell Dendriticcell Dendriticcell Dendriticcell Dendriticcell Dendritic cell Dendriticcell 21 Plasmacytoid_Dendritic_cell Plasmacytoiddendriticcell Dendriticcell Plasmacytoid dendriticcell Plasmacytoid dendriticcell Dendritic cell Plasmacytoid dendriticcell 22 T cell T cell T cell T cell T cell T cell T cell 23 NK cell NK cell Naturalkillercell NK cell NK cell NK cell NK cell 24 Endo Neuron Neuron Neuron Neuron Neuronalcell Neuron 25 unkown Epithelialcell Cholangiocyte Epithelialcell Epithelialcell Epithelial cell Epithelialcell 26 Macrophages B cell B cell B cell B cell NK cell B cell 27 unkown B cell B cell B cell B cell B cell B cell 28 Basophill Mast cell Mast cell Mast cell Mast cell Mastcell Mast cell

[0049] Example 2: Annotation of sheep back muscle cells.

[0050] Data preparation and preprocessing (similar steps to Example 1): Single-cell transcriptome data of sheep back muscle were collected and subjected to strict quality control and differential analysis. Since sheep is a non-model species, special attention should be paid to data standardization and outlier processing during data preprocessing to ensure the accuracy of subsequent homology comparison.

[0051] Homologous comparison and gene screening and sorting: Gene symbol numbers were obtained by homologous comparison with human genes. BLASTP software was used to select genes according to the set homology threshold (E-value < 1 × 10 -5 , Identity≥0.8), compare the genes of sheep back muscle cells with human genes, screen out homologous genes and determine their symbol numbers. Based on the genes after homologous alignment, according to the screening criteria and sorting method in the invention content, determine the up-regulated genes of each subgroup and sort them. For a certain subgroup of sheep back muscle cells, after homologous alignment and screening and sorting, a group of representative up-regulated genes and their corresponding sorting information are obtained.

[0052] Prompt word construction: Extract the top 20 genes (because it is a non-model species) based on the results of subgroup gene sorting to construct prompt words. When constructing prompt words, accurately fill in the sheep species, back muscle tissue and gene information to ensure that the prompt words can accurately reflect the characteristics of sheep back muscle cells. The constructed prompt words are: “You are a biologist, and you are very good at cell annotation. Identify cell types of sheep Dorsal muscle of using the followingmarkers separately for each row. Some can be a mixture of multiple celltypes. #The following is the data to be analyzed. One row represents onetype of cells and the corresponding cell markers. Cell ID is beforesemicolon, and gene markers is after semicolon. 0: PACRG, CCDC85A, PRAG1, BTNL9, LNX1, LRRC36, ZNF366, APOL3, DACH1,CYYR1, LRRC3B, ADGRL4, SHANK3, RASGRF2, CA4, TSPAN13, ROBO4, ABLIM3, Stox2,VWF 1: ncbi_102185059, UCP3, ncbi_108636275, PEBP4, MYBPC2, PFKM, PFKFB4,SH3RF2, DDIT4L, NOS1, MYH2, FBP2, ncbi_106503003, TMEM233, ASB4, Igfn1, MYH1,ATP2A1, AMPD1, ACTN3 <h2 style=";text-align:left;direction:ltr">2: KCNJ8, OTOGL, SEMA5B, GPRIN3, EGFLAM, CNTN4, ADRA1D, ncbi_108633370, SCG3, ncbi_102187240, RERG, NDUFA4L2, CPM, TMEM176B, CDH6, PDGFRB,COX4I2, C8orf48, GUCY1B1, THBS4<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 3: PAX7, CHRDL2, ST8SIA1, RYR3, MEGF10, ncbi_106503297, NCAM1, C1QTNF3, APOE, Asic2, Hs6st2, PPP1R14B, Adamts20, CDH4, CADM2, GPC3, TAGLN3, ncbi_102186022, ATRNL1, C3orf70<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 4: PRG4, FMO2, NCAM2, CYP2F3, DPP4, ADAM22, CFD, BMPER, RIPOR3, CDH18, FBN2, ncbi_102190915, NOX4, ADAM19, HSD11B1, MYOC, KAZN, LUM, C3, GSTM1<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 5: Myh7b, MYOZ2,<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 6: SERPINF1, PCOLCE2, PI16, COL14A1, MFAP5, COL1A1, VIT, S100A4, NOVA1, COL1A2, TNXB, FN1, UST, DPT, PPL, ABI3BP, PDZRN4, MEDAG, Fbln2, LVRN<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr">7: ADRA1D, CPM, OTOGL, KCNJ8, ncbi_102187240, RGS5, APOLD1, TMEM176B, ncbi_108633370, NES, SEMA5B, STAC, RERG, FAM162B, GPRIN3, PDGFRB, NDUFA4L2, TRPC3, EGFLAM, ANO1<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 8: ITK, CD2, THEMIS, CTSW, KLRK1, BCL11B, GRAP2, IL7R, KLRC1, GZMH, TRGC1, CD3E, CD3D, Klre1, ncbi_106502126, CD3G, ncbi_108637283, STAT4, Tox2, HCST<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 9: AZGP1, ACVR1C, MAL2, COL22A1, ABCD2, ACSM1, KLB, PLIN1, APOL6,NNAT, PKP2, ADIPOQ, CHL1, SPEF2, ASGR1, CIDEC, ADIG, SLC22A4, NALCN, SLC1A3<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 10: MYOCD, MYH11, ACTG2, KCNAB1, NRIP2, RCAN2, TGFB3, CNN1, LSAMP,DGKG, ITGA8, CKB, KCTD16, Sorbs2, SLC22A3, ANKRD6, COL4A5, RYR2, DSTN, LBH<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 11: TNNC1, MYL2, CKM, TNNI1, TNNI2, MB, ENO3, MYOZ1, MYLPF, TNNC2, DES, ACTA1, SLN, MYH7, C1QTNF9, KLF2, ALDOA, CRYAB, PGAM2, ID1<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 12: VSIG4, C4BPA, BLNK, ncbi_102188010, RAB3C, CD163, C1QA, F13A1,C1qc, C1QB, BRN, CSF1R, BTK, CTLA4, P2RY12, MRC1, ncbi_106503915, CD74, HLA-DRA, CD86<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr">13: TNNC1, CCN1, CRYAB, MB, TMEM176B, Klhl29, ACTA1, MYH1, CPM, ncbi_102187240, ATP2A1, FAM162B, Ppp1r27, Ntn4, CKM, Tnnt3, TNNI2, ADCY5, COX6A2, SEMA5B<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 14: MMRN1, CCL21, RELN, SCN3A, PTN, C8orf34, SEMA3D, NR2F1, KCNIP4, TC2N, PIEZO2, TNFAIP8L3, DLGAP1, TLL1, Pard6g, LYPD6, RAB11FIP1, RALYL, PKHD1L1, Bst1<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 15: CDH1, ARMC4, Clstn2, Map3k7cl, Sema3a, NCKAP5, EDA2R, CDKN1A,NR5A2, NRK, ART3, FST, ABI3BP, PARM1, Bmpr1b, GPC6, ARPP21, UNC45B, MYBPH, KCNQ5<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 16: BMX, GJA5, ncbi_106502978, PRDM16, APOA1, DKK3, THSD7A, HMCN1, EFNA5, ADAMTS1, PLCG2, COL8A1, NEBL, ncbi_102181308, HEY1, CTNNAL1, PCSK5, ABCB1, LTBP1, TSPAN12<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 17: NELL1, PLVAP, SYT15, VCAM1, ADGRG6, Klhl14, IL1R1, HMCN1, KIF26B,CYP7B1, THSD7A, FABP4, NOS1AP, CTNNAL1, MMP16, CALCRL, Ehd4, Rgs6, AQP1, TSPAN12<h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> 18: FMO2, CDH18, PDGFRA, PCSK6, PRG4, COL6A3, SHCBP1L, NOX4, DCN, CCDC80, SPRY2, GSN, NUPR1, BMPER, CYP2F3, ncbi_102189207, Col6a2, ncbi_102190915, LUM, RGMA<h2 style=";text-align:left;direction:ltr"> 19: ncbi_106503297, CHRDL2, NCAM1, RYR3, Asic2, PAX7, APOE, GREB1,ST8SIA1, MEGF10, CADM2, CDH4, TAGLN3, CDH6, PPP1R14B, TMEM176B, FAM162B,ANO1, Adamts20, THBS4 20: TRPM3, ID4, LAMA3, Slit3, CNN1, LRRC17, Prrx1, MAMDC2, TRPC6,FST, SEMA5A, Sh3gl2, RIMS1, ADGRL3, PTGDS, NHS, IGFBP5, LSAMP, NCKAP5, Synpo2 21: ST8SIA1, C1QTNF3, RYR3, CHRDL2, NCAM1, CDH4, SLC16A10, PAX7,Adamts20, ncbi_108636275, EDA2R, Hs6st2, FBXO32, BTC, LRRTM3, KCNN3, CHRNB1,ASB4, SLIT2, PFKM 22: ncbi_108638533, ncbi_106503398, ncbi_106503383, ncbi_102168946,ncbi_102191280, ncbi_108637754, GADL1, ncbi_108638502, KLHL38, NR5A2, MYBPH,LRRC2, EDA2R, NEURL1, PLAAT1, SCN4A, ARPP21, ncbi_108635579, TMEM182, CPED1 23: SLC28A3, SERPINA1, CSF3R, VNN2, MS4A8, SDS, S100A8, S100A12,NCF1, SELL, HCK, C1orf162, MCTP2, CCL23, C5AR1, PLEK, RIPOR2, ncbi_102175442,SYK, EMB 24: CDH19, ncbi_102189639, IL1RAPL2, UGT8, Xkr4, ATP10B, Nrxn3,GRIK2, SORCS1, PLP1, PRIMA1, ERBB3, PTPRZ1, GRIK3, ADAM23, PPP2R2B, TMEM178B,NLGN1, SCN7A, CNTNAP2 25: KRT8, KRT18, KRT19, CLDN20, CLDN1, ITGB4, ROR2, SBSPON, KLF5,ncbi_106503347, Il1rapl1, FZD1, Robo2, PDLIM2, SFRP5, EPHB2, ncbi_102172959,SMOC2, UNC5C, DOCK5 ## Please output the cell types of each row. The format is asfollows. Please output the result directly without further explanation. 0 Monocytes 1 Neutrophil 2 B cells” (Chinese translation: B cells).

[0053] Submit the large language model and determine the results (similar steps to Example 1): Submit the prompt words to the Chat GPT-4 model for 5 independent questions, and summarize the response results. When processing the annotation of sheep back muscle cells, since sheep are non-model species, some gene expression patterns and biological characteristics that are different from model species may be encountered, but the method of the present invention can still effectively annotate cell types through homology alignment and the powerful learning ability of the large language model. Determine the cell type according to the concept of mode in statistics and compare it with the manual annotation results. In the sheep back muscle cell annotation case, the results showed that among the 25 subpopulations, 16 subpopulations were consistent with the manual annotation results, and another 4 subpopulations were manually annotated as unknown cells (unknow), and the large language model successfully gave the annotation results. This further proves the feasibility and advantages of the method of the present invention in the annotation of single-cell transcriptome cells of non-model species.

[0054] Sheep back muscle cell annotation comparison table: Cell subset Manual annotation AI annotation result First output Second output Third output Fourth output Fifth output 0 Endothelialcells Endothelial cells Endothelial cells Endothelial Cells Endothelial Cells Endothelialcells Endothelialcells 1 Myofiber Myofiber Myofiber MuscleCells SkeletalMuscleCells Musclefibers Myofibers 2 Smoothmusclecells Smoothmusclecells Smoothmusclecells Fibroblasts SmoothMuscleCells Smoothmusclecells Pericytes 3 Satellite cells Satellitecells Satellitecells SatelliteCells SatelliteCells Satellitecells Satellitecells 4 Fibroblasts Fibroblasts Fibroblasts Fibroblasts Fibro-AdipogenicProgenitors (FAPs) Fibroblasts Fibro-adipogenicprogenitors(FAPs) 5 Myofiber Myofiber Myofiber MuscleCells CardiacMuscleCells Musclefibers Cardiomyocytes 6 Tenocyte Fibroblasts Fibroblasts Fibroblasts Fibroblasts Fibroblasts Tenocytes 7 Smoothmusclecells Smoothmusclecells Smoothmusclecells Pericytes VascularSmoothMuscleCells Smoothmusclecells Pericytes 8 T cell T cells T cells T Cells T Cells T cells T cells 9 Unknow Adipocytes Adipocytes Adipocytes Adipocytes Adipocytes Adipocytes 10 SmoothMuscleCells SmoothMuscleCells Myofiber SmoothMuscleCells Myofibroblasts Smoothmusclecells Smooth musclecells 11 Endothelial Myofiber Myofiber MuscleCells Myofibers Musclefibers Skeletalmuscle cells 12 B cell Macrophages Macrophages Macrophages Macrophages Macrophages Macrophages 13 Unknow Myofiber Myofiber MuscleCells Myofibers Musclefibers Myofibers 14 Unknow SchwannCells Fibroblasts Fibroblasts SchwannCells Schwanncells Schwann cells 15 Myoblast Endothelial cells Endothelial cells EpithelialCells MuscleProgenitorCells Schwanncells Fibroblasts 16 Endothelialcells Endothelial cells Endothelial cells SmoothMuscleCells Pericytes Endothelialcells Vascularsmooth musclecells 17 Endothelialcells Endothelial cells Endothelial cells Endothelial Cells VascularEndothelial Cells Endothelialcells Vascularendothelialcells 18 Fibroblasts Fibroblasts Fibroblasts Fibroblasts Fibro-AdipogenicProgenitors (FAPs) Fibroblasts Fibroblasts 19 Unknow Satellitecells Satellitecells SatelliteCells SatelliteCells Satellitecells Satellitecells 20 Pericytes Myofibroblasts Myofibroblasts SmoothMuscleCells Myofibroblasts Fibroblasts Myoendothelial progenitors 21 Satellite cells Satellitecells Satellitecells SatelliteCells SatelliteCells Satellitecells Satellitecells 22 Myofiber Myofiber Myofiber SmoothMuscleCells MuscleProgenitorCells Schwanncells Neuromuscularjunctioncells 23 Neutrophils Neutrophils Neutrophils Neutrophils Neutrophils Neutrophils Neutrophils 24 Schwanncell Schwanncells Schwanncells GlialCells SchwannCells Schwanncells Oligodendrocytes 25 Fibroblast Epithelial cells Epithelialcells EpithelialCells EpithelialCells Epithelialcells Epithelialcells

Claims

1. A single-cell transcriptome cell annotation method integrating a large language model, characterized in that: The following steps are involved: S1, data preprocessing: Preprocess the newly sequenced single-cell transcriptome data, including data cleaning, quality control, and construction of gene expression matrix; S2, obtain and sort the up-regulated genes of each cell subpopulation: For each gene in a given subpopulation, its expression value was compared with other cell subpopulations using the Wilcoxon rank sum test to identify genes that were significantly upregulated in each cell subpopulation; Based on the results of differential analysis, the up-regulated genes of each subgroup were ranked, and the order was prioritized from small to large P values. When the P values ​​were the same, they were further ranked from large to small using the difference fold; S3, non-model species homology comparison (if it is a non-model species): When the target cell is annotated as a non-model species, homology comparison is required to determine the gene symbol number. If the target species is an animal, human or mouse is selected as the reference model species. If the target species is a plant, Arabidopsis is selected as the reference model species. Use BLASTP software to perform homology comparison to determine the homology relationship between the genes of non-model species and reference model species, and define the gene symbol number of non-model species based on the symbol number of the reference model species gene; S4, construct a unique prompt word: Record the names of the up-regulated gene markers of species, tissues, and subgroups as variables, define the input and output formats, and construct prompt words for submission to the large language model; S5, submit the large language model and determine the cell type: The constructed prompt words are submitted to the large language model, and the cell type and marker genes corresponding to each subpopulation are determined by multiple questions and answers and counting the majority of answers.

2. According to the single-cell transcriptome cell annotation method integrating a large language model according to claim 1, it is characterized in that: In the S2 step, the specific criteria for significantly upregulated genes are: the gene is expressed in at least 25% of the cells in the target subpopulation, the P value is ≤ 0.01 and the log2-fold change is ≥ 0.

36.

3. According to the single-cell transcriptome cell annotation method integrating a large language model according to claim 1, it is characterized in that: In the step S3, the parameters for homology comparison using BLASTP software were set as the threshold of homology: E-value<1×10-5, Identity≥0.8, and the gene with the smallest E-value was selected as the homologous gene of the source reference model species according to the E-value sorting.

4. According to the single-cell transcriptome cell annotation method integrating a large language model according to claim 1, it is characterized in that: In the step S4, the 10 most highly ranked gene construction prompt words are selected for each subgroup of model species, and the 20 most highly ranked gene construction prompt words are selected for each subgroup of non-model species.

5. According to the single-cell transcriptome cell annotation method integrating a large language model according to claim 1, it is characterized in that: In the step S5, the constructed prompt words are submitted to the large language model through the API, the questions and answers are repeated multiple times, the replies of the large language model are summarized, and the cell type and marker gene corresponding to each subpopulation are determined based on the concept of mode in statistics.

6. A single-cell transcriptome cell annotation system integrating a large language model, characterized in that: include: S1, data preprocessing module: Used to receive single-cell transcriptome data and perform quality control and standardization on the data; S2, gene sorting and screening module: Used to realize the function of step S2 in claim 1, obtain and sort the up-regulated genes of each cell subpopulation, and screen and sort the significantly up-regulated genes according to the set criteria; S3, homology alignment module (for non-model species): When processing non-model species data, it is used to perform the function of step S3 in claim 1 to determine the symbol number of the non-model species gene; S4, prompt word building blocks: Used to construct a proprietary prompt word according to a predetermined rule based on the information provided in step S2 of claim 1 and the gene screened out in step S3 of claim 1; S5, large language model interaction module: Responsible for interacting with the large language model, submitting prompt words to the large language model and receiving its responses; S6, cell type determination module: Based on the answer obtained in step S5 of claim 1, the cell type is determined according to the concept of mode in statistics.

7. According to claim 5, a single-cell transcriptome cell annotation system integrating a large language model is characterized in that: In the S2 step, the Wilcoxon rank sum test was used to screen the genes, and the up-regulated genes were ranked according to the P value and the difference fold.

8. According to the single-cell transcriptome cell annotation system integrating a large language model according to claim 5, it is characterized in that: In the S3 step, BLASTP software and other tools are used for homology comparison, and the homology threshold is set as E-value <1×10-5, Identity ≥0.

8.

9. The single-cell transcriptome cell annotation system integrating a large language model according to claim 5, characterized in that: In the S5 step, an optimized API calling method is used to ensure efficient communication with the large language model, achieve stable progress of multiple questions and answers and accurate aggregation of results, and avoid problems such as data transmission errors and communication interruptions that affect the determination of cell types.

10. The single-cell transcriptome cell annotation system integrating a large language model according to claim 5, characterized in that: In the step S6, when the large language model replies are counted, the results of each question and answer are accurately recorded and analyzed, and rigorous judgments are made according to the concept of mode in statistics. For situations with the same or similar frequencies of occurrence, the ability to further analyze and judge is possessed to improve the accuracy and reliability of cell type determination.

Images