Prediction method and prediction model for curative effect of HER2 positive breast cancer T-DM1 drug

Abstract

The invention relates to the technical field of bioinformatics, and provides a prediction method and a prediction model for the curative effect of an HER2 positive breast cancer T-DM1 drug. The prediction method comprises the following steps: obtaining gene expression data of an HER2 positive tumor sample through RNA sequencing; according to the gene expression data, scoring a sample by using a cell cycle related gene set to obtain an enrichment score; predicting the drug curative effect of the T-DM1 according to the enrichment score; if the enrichment score is relatively low, it is indicated that the T-DM1 drug sensitivity of the sample is good, and if the enrichment score is relatively high, it is indicated that the T-DM1 drug sensitivity is poor. According to the prediction method and the prediction model, the drug sensitivity of T-DM1 can be conveniently and quickly predicted, a new strategy is provided for prognosis prediction of HER2 positive breast cancer, and support is provided for related research of cell cycle and T-DM1 curative effect.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and in particular to a prediction method and a prediction model for the efficacy of T-DM1 for HER2-positive breast cancer. Background Art

[0002] Breast cancer is the most common malignant tumor in women worldwide. HER2-positive breast cancer accounts for about 15%-20% of all breast cancers. It is characterized by high malignancy and poor prognosis. Anti-HER2 trastuzumab and pertuzumab are recommended for the first-line treatment of HER2-positive advanced breast cancer, which significantly improves the survival rate of these patients, but some patients still develop drug resistance and disease progression after treatment. T-DM1 (trastuzumab-mertansine) is an antibody-drug conjugate (ADC) composed of trastuzumab and microtubule inhibitor mertansine. It is recommended by international treatment guidelines for the second-line treatment of HER-2-positive advanced breast cancer and is widely used in clinical practice. However, the efficacy of T-DM1 is not stable. T-DM1 resistance limits its application in the treatment of HER2-positive breast cancer patients. There is an urgent need for efficacy prediction methods to guide clinical practice. Traditional predictive indicators for breast cancer treatment include TNM staging and Ki67 protein expression, which are non-specific and have poor predictive ability for HER2-positive breast cancer with strong heterogeneity. Therefore, it is clinically crucial to establish a prognostic model that can effectively predict the efficacy of T-DM1 in patients with HER2-positive breast cancer.

[0003] ADC resistance is often associated with the antibody itself, drug transport, lysosomal function, payload, and tumor clone heterogeneity. Tumor heterogeneity refers to the genetic, phenotypic, and functional differences between different tumor cells within the same tumor. This difference may lead to inconsistent tumor responses to treatment, thereby affecting the effectiveness of the drug. Previous studies have shown that the mRNA expression of HER2 and HER3 or the PIK3CA mutation in tumor cells is not related to the effect of T-DM1. Currently, there is no single, accurate tumor molecular marker that can fully predict the efficacy of T-DM1. Summary of the invention

[0004] The present invention provides a method for predicting the efficacy of T-DM1 drug for HER2-positive breast cancer based on cell cycle-related genes, and also provides a gene set constructed based on tumor gene expression used in the prediction method and a method for obtaining the gene set, so as to solve the above problems.

[0005] In a first aspect of the present invention, a method for predicting the efficacy of T-DM1 drug for HER2-positive breast cancer is provided, wherein gene expression data of HER2-positive tumor samples are obtained by RNA sequencing; based on the gene expression data, the samples are scored using a cell cycle-related gene set to obtain an enrichment score; the efficacy of T-DM1 drug is predicted based on the enrichment score; if the enrichment score is relatively low, it indicates that the sample has a good sensitivity to T-DM1 drug, and if the enrichment score is relatively high, it indicates that the sample has a poor sensitivity to T-DM1 drug.

[0006] Optionally, the cell cycle-related gene set is a Cell cycle activity gene set, including 100 genes, namely: CDK1, UBE2C, BIRC5, PTTG1, TOP2A, MKI67, TPX2, KIF2C, CDC20, HMMR, CCNB1, PBK, CDKN3, MAD2L1, NUF2, NCAPG, CENPF, CCNA2, ZWINT, HMGB2, CDCA8, KIF23, TROAP, NDC80, DLGAP5, ANLN, PLK1, AURKB, KIF20A, BUB1, CDCA3, SGO1, CENPA, RRM2, KIFC1, NCAPH, DEPDC1, CEP55, CENPW, SPC25, CENPE, GTSE1, TK1, NEK2, CKAP2L, UBE2T, RACGAP1, LMNB1, PKMYT 1. ASPM, FOXM1, TTK, ARHGAP11A, MYBL2, TYMS, FAM83D, KIF11, HJURP, KIF4A, CKAP2, MELK, KNL1, ATAD2, ECT2, TRIP13, FBXO5, SGO2, STMN1, RAD51AP1, CKS1B, CENPU, MND1, SMC4, CENPN, KPN A2, CENPK, CENPM, TACC3, KIF20B, CKS2, ESCO2, SKA3, PARPBP, CDCA5, PIMREG, UBE2S, AURKA, ASF1B, TMPO, KIF14, OIP5, ORC6, KNSTRN, SKA1, FANCI, GAS2L3, NCAPD2, KIF15, FAM111B, MCM10.

[0007] Optionally, scoring the sample using the cell cycle-related gene set includes: calculating the enrichment score of the sample using the ssGSEA algorithm.

[0008] Optionally, the cell cycle-related gene set is obtained by the following method: S1: Collect HER2-positive breast cancer tumor samples, perform single-cell RNA sequencing, and obtain single-cell sequencing data, which is the original gene expression data; divide each tumor sample into sensitive group and drug-resistant group according to the disease progression; S2: performing data processing and quality control on the original gene expression data to obtain pre-processed data; S3: performing dimension reduction, clustering and grouping on the preprocessed data, including: Data was normalized based on the original UMI count of each cell, and 2000 highly variable genes were identified by the FindVariableFeatures function. PCA dimension reduction was performed based on the normalized highly variable gene matrix to remove batch effects. The shared nearest neighbor similarity of the cell-PC matrix after removing the batch effect was calculated based on the FindNeighbors function, and several cell subpopulations were identified using the KNN algorithm through the FindClusters function. Cell subpopulations were defined based on gene markers; several cell subpopulations included epithelial cell populations. S4: Set the clustering resolution, further cluster the epithelial cell group, and identify m epithelial cell subpopulations; among the m epithelial cell subpopulations, select the epithelial cell subpopulation in which the proportion of the sensitive group is lower than that of the resistant group and the difference in proportion between the two groups is the most statistically significant, and record it as epithelial cell subpopulation A; extract the differential genes between epithelial cell subpopulation A and other epithelial cell subpopulations through the COSG function, and define the top n genes as the cell cycle-related gene set.

[0009] Optionally, the data processing and quality control described in step S2 includes: checking the sequencing depth, evaluating the sequencing quality, removing low-quality cells with more than 8,000 or less than 200 sequenced genes, removing cells with a mitochondrial gene ratio of more than 10% and diploid cells, and then removing contaminating genes based on the DecontX algorithm.

[0010] Optionally, in step S4, the clustering resolution is set to 0.4.

[0011] Optionally, in step S4, m is 10 and n is 100.

[0012] The second aspect of the present invention provides a method for constructing a T-DM1 efficacy prediction model for HER2-positive breast cancer based on cell cycle-related genes, comprising: S1: Collect HER2-positive breast cancer tumor samples, perform single-cell RNA sequencing, and obtain single-cell sequencing data, which is the original gene expression data; divide each tumor sample into sensitive group and drug-resistant group according to the disease progression; S2: performing data processing and quality control on the original gene expression data to obtain pre-processed data; Including: checking sequencing depth, evaluating sequencing quality, removing low-quality cells with more than 8,000 or less than 200 sequenced genes, removing cells with a mitochondrial gene ratio of more than 10% and diploid cells, and then removing contaminating genes based on the DecontX algorithm; S3: performing dimension reduction, clustering and grouping on the preprocessed data, including: Data was normalized based on the original UMI count of each cell, and 2000 highly variable genes were identified by the FindVariableFeatures function. PCA dimension reduction was performed based on the normalized highly variable gene matrix to remove batch effects. The shared nearest neighbor similarity of the cell-PC matrix after removing the batch effect was calculated based on the FindNeighbors function, and 10 cell subpopulations were identified using the KNN algorithm through the FindClusters function. Cell subpopulations were defined based on gene markers; the 10 cell subpopulations included epithelial cells. S4: The epithelial cell group was further clustered with a clustering resolution of 0.4, and 10 epithelial cell subpopulations were identified. Among the 10 epithelial cell subpopulations, the epithelial cell subpopulation whose proportion in the sensitive group was lower than that in the resistant group and whose proportion difference between the two groups was the most statistically significant was selected, and was recorded as epithelial cell subpopulation A; the differential genes between epithelial cell subpopulation A and other epithelial cell subpopulations were extracted by the COSG function, and the top 100 genes were defined as the Cell cycle activity gene set to obtain the prediction model.

[0013] The third aspect of the present invention provides an RNA composition for use in a HER2-positive breast cancer T-DM1 drug efficacy prediction system or the manufacture of a prediction product, wherein the RNA composition comprises transcription products of 100 genes, wherein the 100 genes are genes in the above-mentioned Cell cycle activity gene set.

[0014] The prediction method and gene set (i.e., prediction model) of the T-DM1 drug efficacy of the present invention can be used to predict the clinical efficacy of the drug T-DM1 for HER2-positive breast cancer, including predicting the drug resistance of T-DM1. The scheme of the present invention extracts differentially expressed genes related to the cell cycle, forms a gene set, and establishes a prediction model. The gene set is verified by GEO database data, proving that it can effectively predict the drug resistance of T-DM1. The prediction method and prediction model of the present invention can conveniently and quickly predict the drug sensitivity of T-DM1, provide a new strategy for the prognosis prediction of HER2-positive breast cancer, and also provide support for the research related to the cell cycle and the efficacy of T-DM1. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1It is an introduction to the process of clinical sample collection and sequencing involved in the embodiments of the present invention, including the display of patient information, subgroup clustering diagrams of single-cell sequencing data, differential gene heat maps and ratio diagrams.

[0016] Figure 2 It is a display of the Cancer cell_CDK1 tumor subpopulation (epithelial cell subpopulation) in the embodiment of the present invention, including the display that the current single genes ERBB2 and ERBB3 lack predictive efficacy, subpopulation clustering diagrams, differential gene heat maps and proportion maps of single-cell sequencing tumor cells, expression of cell cycle-related genes in each tumor subpopulation, and GO and KEGG functional enrichment display of the Cancer cell_CDK1 subpopulation.

[0017] Figure 3 It is the application of the embodiment of the present invention in the GEO database, including the display of exogenous clinical cohorts and the display of Cell cycle activity scores of patients with different prognoses. DETAILED DESCRIPTION

[0018] Preclinical studies have found that high activation of the CDK4 / 6 pathway in tumor cells mediates resistance to HER2 targeted therapy. CDK4 / 6 are key regulators of the cell cycle. They promote the cell cycle from the G1 phase to the S phase by binding to cell cycle proteins, thereby promoting cell proliferation. Therefore, in HER2-positive breast cancer, overactivation of cell cycle-related pathways may promote tumor cell proliferation and be associated with resistance to HER2 targeted therapy. However, research on the cell cycle and the efficacy of T-DM1 remains in preclinical models and lacks support and further verification from clinical data. Therefore, there is an urgent need for a T-DM1 clinical efficacy prediction model based on tumor gene expression to solve the above technical problems.

[0019] The present invention provides a method for predicting the efficacy of T-DM1 drug for HER2-positive breast cancer. The method obtains gene expression data of a HER2-positive tumor sample to be tested by RNA sequencing; scores the sample to be evaluated using a cell cycle-related gene set based on the gene expression data to obtain an enrichment score; predicts the efficacy of T-DM1 drug based on the enrichment score; if the score is relatively low, it indicates that the sample has a good sensitivity to T-DM1 drug, and if the score is relatively high, it indicates that the sample has a poor sensitivity to T-DM1 drug.

[0020] Specifically, the RNA sequencing can use bulk RNA sequencing (Bulk RNA-seq) or single-cell sequencing, etc. After the sequencing data of the sample is obtained, the sequencing data of the sample is scored according to the cell cycle-related gene set to obtain an enrichment score (ES). Among them, the cell cycle-related gene set can be obtained by performing single-cell RNA sequencing on a large number of collected tumor samples, performing data processing and cluster analysis, and extracting differential genes between different cell subpopulations. The cell cycle-related gene set can be regarded as a prediction model. When the cell cycle-related gene set or prediction model is actually applied, the method includes: obtaining a puncture sample of a patient whose T-DM1 drug efficacy is to be predicted, performing RNA sequencing on the sample to obtain the original gene expression data of the sample; according to the original gene expression data of the sample and the cell cycle-related gene set, using, for example, the ssGSEA algorithm, an enrichment score is obtained. The relative high or low enrichment score of the sample among many other patient samples or reference samples can reflect its drug sensitivity. For samples with relatively low scores, it means that the sample has good T-DM1 drug sensitivity and no significant T-DM1 drug resistance; if the score is relatively high, it indicates that the T-DM1 drug sensitivity is poor and has a certain drug resistance. Through the above scheme, the present invention discloses a method for predicting the efficacy of T-DM1 drug for HER2-positive breast cancer.

[0021] Furthermore, in one embodiment of the present invention, a specific cell cycle-related gene set is provided, named "Cell cycle activity" gene set, which includes 100 genes, namely: CDK1, UBE2C, BIRC5, PTTG1, TOP2A, MKI67, TPX2, KIF2C, CDC20, HMMR, CCNB1, PBK, CDKN3, MAD2L1, NUF2, NCAPG, CENPF, CCNA2, ZWINT, HMGB2, CDCA8, KIF23, TROAP, NDC80, DLGAP5, ANLN, PLK1, AURKB, KIF20A, BUB1, CDCA3, SGO1, CENPA, RRM2, KIFC1, NCAPH, DEPDC1, CEP55, CENPW, SPC25, CENPE, GTSE1, TK1, NEK2, CKAP2L, UBE2T, RACGAP1, LMNB1, P KMYT1, ASPM, FOXM1, TTK, ARHGAP11A, MYBL2, TYMS, FAM83D, KIF11, HJURP, KIF4A, CKAP2, MELK, KNL1, ATAD2, ECT2, TRIP13, FBXO5, SGO2, STMN1, RAD51AP1, CKS1B, CENPU, MND1, SMC4, CENPN, K PNA2, CENPK, CENPM, TACC3, KIF20B, CKS2, ESCO2, SKA3, PARPBP, CDCA5, PIMREG, UBE2S, AURKA, ASF1B, TMPO, KIF14, OIP5, ORC6, KNSTRN, SKA1, FANCI, GAS2L3, NCAPD2, KIF15, FAM111B, MCM10. Using this gene set, the enrichment score obtained by the ssGSEA algorithm is the Cell cycle activity score.

[0022] In addition, the transcription products of the above 100 genes form an RNA composition, which includes the transcription products of the above genes. The RNA composition can be used to predict the efficacy of HER2-positive breast cancer T-DM1 drugs, or to construct a HER2-positive breast cancer T-DM1 drug efficacy prediction system. Further, the RNA composition can also form a HER2-positive breast cancer T-DM1 drug efficacy prediction product.

[0023] Another embodiment of the present invention provides a method for obtaining a cell cycle-related gene set, comprising the following steps: S1: Collect HER2-positive breast cancer tumor samples, perform single-cell RNA sequencing, and obtain single-cell sequencing data, which is the original gene expression data; divide each tumor sample into sensitive group and resistant group according to the disease progression.

[0024] S2: performing data processing and quality control on the original gene expression data to obtain pre-processed data.

[0025] It includes: checking sequencing depth, evaluating sequencing quality, removing low-quality cells with more than 8,000 or less than 200 sequenced genes, removing cells with a mitochondrial gene ratio of more than 10% and diploid cells, and then removing contaminating genes based on the DecontX algorithm.

[0026] S3: Performing dimension reduction, clustering and grouping on the preprocessed data.

[0027] It includes: data normalization based on the original UMI count of each cell, identifying 2000 highly variable genes by the FindVariableFeatures function; PCA dimensionality reduction and removal of batch effects based on the normalized highly variable gene matrix; calculating the shared nearest neighbor similarity of the cell-PC matrix after removing the batch effect based on the FindNeighbors function, and identifying several cell subpopulations using the KNN algorithm through the FindClusters function; defining cell subpopulations based on gene markers; several cell subpopulations include epithelial cell populations.

[0028] S4: Set the clustering resolution, for example, 0.4, and further cluster the epithelial cell group to identify m epithelial cell subpopulations; among the m epithelial cell subpopulations, select the epithelial cell subpopulation whose proportion in the sensitive group is lower than that in the drug-resistant group and whose proportion difference between the two groups is the most statistically significant, and record it as epithelial cell subpopulation A; extract the differential genes between epithelial cell subpopulation A and other epithelial cell subpopulations through the COSG function, and define the top n genes as the cell cycle-related gene set. In some schemes, the m is 10 and the n is 100.

[0029] Based on the above gene set acquisition method, the present invention also provides a method for constructing a T-DM1 efficacy prediction model for HER2-positive breast cancer based on cell cycle-related genes.

[0030] In a more specific scheme, the transcriptome data of neoadjuvant therapy of T-DM1 was also obtained from the GEO database to verify the prediction model obtained by the construction method. Including: obtaining the transcriptome data of neoadjuvant therapy of T-DM1 from the GEO database, preprocessing the collected gene expression data, including data cleaning, standardization and missing value processing. Then the expression score of each gene in a single sample was calculated by the ssGSEA function, and the enrichment score was obtained based on the gene set. The enrichment score was statistically analyzed to verify the difference in scores between the complete pathological remission group (pCR group) and the non-complete pathological remission group (non-pCR group) in pathological response. Example

[0031] Tumor samples from HER2-positive breast cancer patients in the applicant's prospective, multicenter, phase II clinical trial (NCT06125834) were collected for single-cell RNA sequencing, including tumor samples from 8 participants before treatment and samples from 2 participants whose disease progressed after receiving T-DM1 treatment. The single-cell sequencing data of this center have been uploaded to the Human Genetic Resources Omics Original Database of the National Center for Bioinformation, with the number HRA009023, and are available for access and download.

[0032] Among these 8 participants, 4 had a progression-free survival (PFS) of more than 6 months and were considered the sensitive group. The other 4 had rapid disease progression within 6 months and were considered the resistant group together with the 2 who had disease progression after treatment ( Figure 1 ).

[0033] The sequencing data were quality controlled: low-quality cells with more than 8000 or less than 200 sequenced genes were removed, and cells with a mitochondrial gene ratio of more than 10% and diploid cells were removed, resulting in a total of 80,239 single cells. Then, the contaminating genes (contamination < 0.01) were removed based on the DecontX algorithm. Data were normalized based on the original UMI counts of each cell, and 2000 highly variable genes were identified by the FindVariableFeatures function. The first 2000 highly variable genes were normalized using the ScaleData function, and the resulting matrix was subjected to PCA dimensionality reduction, and the batch effect was removed using the Harmony function. Then, the shared nearest neighbor similarity of the cell-PC matrix after removing the effect was calculated based on the FindNeighbors function, and 10 cell subpopulations (resolution 1.2) were identified using the KNN algorithm through the FindClusters function. Cell subsets were defined based on common genetic markers and divided into fibroblasts, endothelial cells, epithelial cells, mast cells, melanocytes, parietal cells, mononuclear phagocytes, B cells, as well as T cells and natural killer (NK) cells ( Figure 1 ).

[0034] Traditionally, high expression of ERBB2 and ERBB3 indicates good therapeutic effect in HER2-positive breast cancer, but our data showed the opposite ( Figure 2 ), suggesting that a single ERBB2 or ERBB3 cannot accurately predict the efficacy of T-DM1. The 43,995 epithelial cells were further clustered and 10 epithelial cell subpopulations were identified (with a clustering resolution of 0.4) Figure 2 ). Compared with resistant cases, sensitive cases showed a significantly lower proportion of Cancer cells_CDK1 (P=0.019) ( Figure 2 ), Cancer cells_CDK1 highly expressed cell cycle activity-related genes such as UBE2C, CDK1, TOP2A, PLK1, NEK2, NDC80, KIF20A, DLGAP5, CDC20 and CCNA2 ( Figure 2 ), indicating a high level of cell cycle activity. In addition, GO (gene ontology enrichment analysis) and KEGG (kyoto encyclopedia of genes and genomes) analysis showed that Cancer cells_CDK1 was enriched in pathways such as cell cycle, chromosome separation, and mitosis and nuclear division ( Figure 2 ).

[0035] Based on the COSG algorithm, the differentially expressed genes between Cancer cells_CDK1 and other epithelial subgroups were extracted, and the top 100 genes were used to construct a gene set "Cell cycle activity" and establish a prediction model. The activity gene set consists of CDK1, UBE2C, BIRC5, PTTG1, TOP2A, MKI67, TPX2, KIF2C, CDC20, HMMR, CCNB1, PBK, CDKN3, MAD2L1, NUF2, NCAPG, CENPF, CCNA2, ZWINT, HMGB2, CDCA8, KIF23, TROAP, NDC80, DLGAP5, ANLN, PLK1, AURKB, KIF20A, BUB1, CDCA3, SGO1, CENPA, RRM2, KIFC1, NCAPH, DEPDC1, CEP55, CENPW, SPC25, CENPE, GTSE1, TK1, NEK2, CKAP2L, UBE2T, RACGAP1, LMNB1, PKMYT1, and ASPM , FOXM1, TTK, ARHGAP11A, MYBL2, TYMS, FAM83D, KIF11, HJURP, KIF4A, CKAP2, MELK, KNL1, ATAD2, ECT2, TRIP13, FBXO5, SGO2, STMN1, RAD51AP1, CKS1B, CENPU, MND1, SMC4, CENPN, KPNA2, C Composed of ENPK, CENPM, TACC3, KIF20B, CKS2, ESCO2, SKA3, PARPBP, CDCA5, PIMREG, UBE2S, AURKA, ASF1B, TMPO, KIF14, OIP5, ORC6, KNSTRN, SKA1, FANCI, GAS2L3, NCAPD2, KIF15, FAM111B, MCM10.

[0036] The transcriptome data of 129 patients with HER2-positive breast cancer who received T-DM1 plus pertuzumab in a neoadjuvant setting were obtained from the GEO database ( Figure 3 After data preprocessing, the "Cell cycle activity" gene set constructed based on single-cell data was used to score the 129 samples before treatment using the ssGSEA algorithm. The cell cycle score (Cell cycle activity score) of the pCR group was significantly lower than that of the Non-pCR group ( Figure 3 ), suggesting that this gene set has a good predictive effect on T-DM1 resistance.

[0037] HER2-positive breast cancer is a highly heterogeneous tumor with different subtypes of cancer cells inside. Based on the above scheme, it can be seen that the present invention discovered a tumor cell subpopulation Cancer cells_CDK1 enriched in cell cycle-related pathways through single-cell transcriptome sequencing analysis of tumor samples from prospective clinical cohort patients in this center. In the clinical study of this center, patients with metastatic HER2-positive breast cancer with a high proportion of CDK1-positive tumor cells responded poorly to T-DM1; in the embodiment scheme of the present invention, differentially expressed genes related to the cell cycle were extracted to form a gene set, and a prediction model was established. The predictive effect of this gene set was also verified, that is, the scores of the cell cycle gene set in different samples were calculated by combining the data of the public data neoadjuvant cohort, and it was found that compared with patients with a pathological outcome of non-complete pathological remission (non-pCR), the cell cycle gene set score of patients in the complete pathological remission (pCR) group was low.

[0038] All these data indicate that high cell cycle activity of tumor cells predicts poor response to T-DM1, CDK4 / 6 inhibitors may reverse drug resistance, and the gene set we constructed based on tumor gene expression can be used to predict the clinical efficacy of T-DM1, which provides a new strategy for prognosis prediction of HER2-positive breast cancer.

[0039] The present invention has been described in detail above in conjunction with specific implementation methods and exemplary examples, but these descriptions cannot be understood as limiting the present invention. Those skilled in the art will appreciate that, without departing from the spirit and scope of the present invention, various equivalent substitutions, modifications or improvements may be made to the technical solution of the present invention and its implementation methods, all of which fall within the scope of the present invention.

Claims

1. A method for predicting the efficacy of T-DM1 drug for HER2-positive breast cancer, characterized in that: Gene expression data of HER2-positive tumor samples are obtained by RNA sequencing; based on the gene expression data, the samples are scored using a cell cycle-related gene set to obtain an enrichment score; the drug efficacy of T-DM1 is predicted based on the enrichment score; if the enrichment score is relatively low, it indicates that the sample has a better drug sensitivity to T-DM1, and if the enrichment score is relatively high, it indicates that the drug sensitivity to T-DM1 is poor.

2. The prediction method according to claim 1, characterized in that: The cell cycle-related gene set is a Cellcycle activity gene set, which includes 100 genes, namely: CDK1, UBE2C, BIRC5, PTTG1, TOP2A, MKI67, TPX2, KIF2C, CDC20, HMMR, CCNB1, PBK, CDKN3, MAD2L1, NUF2, NCAPG, CENPF, CCNA2, ZWINT, HMGB2, CDCA8, KIF23, TROAP, NDC80, DLGAP5, ANLN, PLK1, AURKB, KIF20A, BUB1, CDCA3, SGO1, CENPA, RRM2, KIFC1, NCAPH, DEPDC1, CEP55, CENPW, SPC25, CENPE, GTSE1, TK1, NEK2, CKAP2L, UBE2T, RACGAP1, LMNB1, PKMYT 1. ASPM, FOXM1, TTK, ARHGAP11A, MYBL2, TYMS, FAM83D, KIF11, HJURP, KIF4A, CKAP2, MELK, KNL1, ATAD2, ECT2, TRIP13, FBXO5, SGO2, STMN1, RAD51AP1, CKS1B, CENPU, MND1, SMC4, CENPN, KPN A2, CENPK, CENPM, TACC3, KIF20B, CKS2, ESCO2, SKA3, PARPBP, CDCA5, PIMREG, UBE2S, AURKA, ASF1B, TMPO, KIF14, OIP5, ORC6, KNSTRN, SKA1, FANCI, GAS2L3, NCAPD2, KIF15, FAM111B, MCM10.

3. The prediction method according to claim 1, characterized in that: The scoring of samples using the cell cycle-related gene set includes: calculating the enrichment score of the samples using the ssGSEA algorithm.

4. The prediction method according to claim 1, characterized in that: The cell cycle-related gene set is obtained by the following method: S1: Collect HER2-positive breast cancer tumor samples, perform single-cell RNA sequencing, and obtain single-cell sequencing data, which is the original gene expression data; divide each tumor sample into sensitive group and drug-resistant group according to the disease progression; S2: performing data processing and quality control on the original gene expression data to obtain pre-processed data; S3: performing dimension reduction, clustering and grouping on the preprocessed data, including: Data was normalized based on the raw UMI counts of each cell, and 2000 highly variable genes were identified by the FindVariableFeatures function. PCA dimensionality reduction was performed based on the normalized highly variable gene matrix to remove batch effects. The shared nearest neighbor similarity of the cell-PC matrix after removing the batch effect is calculated based on the FindNeighbors function, and several cell subpopulations are identified using the KNN algorithm through the FindClusters function; cell subpopulations are defined based on gene markers; several cell subpopulations include epithelial cell populations; S4: Set the clustering resolution, further cluster the epithelial cell group, and identify m epithelial cell subpopulations; among the m epithelial cell subpopulations, select the epithelial cell subpopulation in which the proportion of the sensitive group is lower than that of the resistant group and the difference in proportion between the two groups is the most statistically significant, and record it as epithelial cell subpopulation A; extract the differential genes between epithelial cell subpopulation A and other epithelial cell subpopulations through the COSG function, and define the top n genes as the cell cycle-related gene set.

5. The prediction method according to claim 4, characterized in that: The data processing and quality control described in step S2 include: checking sequencing depth, evaluating sequencing quality, removing low-quality cells with more than 8,000 or less than 200 sequenced genes, removing cells with a mitochondrial gene ratio of more than 10% and diploid cells, and then removing contaminating genes based on the DecontX algorithm.

6. The prediction method according to claim 4, characterized in that: In step S4, the clustering resolution is set to 0.

4.

7. The prediction method according to claim 4, characterized in that: In step S4, m is 10 and n is 100.

8. A method for constructing a T-DM1 efficacy prediction model for HER2-positive breast cancer based on cell cycle-related genes, characterized in that: include: S1: Collect HER2-positive breast cancer tumor samples, perform single-cell RNA sequencing, and obtain single-cell sequencing data, which is the original gene expression data; divide each tumor sample into sensitive group and drug-resistant group according to the disease progression; S2: performing data processing and quality control on the original gene expression data to obtain pre-processed data; Including: checking sequencing depth, evaluating sequencing quality, removing low-quality cells with more than 8,000 or less than 200 sequenced genes, removing cells with a mitochondrial gene ratio of more than 10% and diploid cells, and then removing contaminating genes based on the DecontX algorithm; S3: performing dimension reduction, clustering and grouping on the preprocessed data, including: Data was normalized based on the original UMI count of each cell, and 2000 highly variable genes were identified by the FindVariableFeatures function. PCA dimension reduction was performed based on the normalized highly variable gene matrix to remove batch effects. The shared nearest neighbor similarity of the cell-PC matrix after removing the batch effect was calculated based on the FindNeighbors function, and 10 cell subpopulations were identified using the KNN algorithm through the FindClusters function. Cell subpopulations were defined based on gene markers; the 10 cell subpopulations included epithelial cells. S4: The epithelial cell group was further clustered with a clustering resolution of 0.4, and 10 epithelial cell subpopulations were identified. Among the 10 epithelial cell subpopulations, the epithelial cell subpopulation whose proportion in the sensitive group was lower than that in the resistant group and whose proportion difference between the two groups was the most statistically significant was selected, and was recorded as epithelial cell subpopulation A; the differential genes between epithelial cell subpopulation A and other epithelial cell subpopulations were extracted by the COSG function, and the top 100 genes were defined as the Cell cycle activity gene set to obtain the prediction model.

9. Use of an RNA composition in the manufacture of a HER2-positive breast cancer T-DM1 drug efficacy prediction system or prediction product, characterized in that: The RNA composition includes the transcription products of 100 genes, and the 100 genes are: CDK1, UBE2C, BIRC5, PTTG1, TOP2A, MKI67, TPX2, KIF2C, CDC20, HMMR, CCNB1, PBK, CDKN3, MAD2L1, NUF2, NCAPG, CENPF, CCNA2, ZWINT, HMGB2, CDCA8, KIF23, TROAP, NDC80, DLGAP5, ANLN, PLK1, AURKB, KIF20A, BUB1, CDCA3, SGO1, CENPA, RRM2, KIFC1, NCAPH, DEPDC1, CEP55, CENPW, SPC25, CENPE, GTSE1, TK1, NEK2, CKAP2L, UBE2T, RACGAP1, LMNB1, PKMYT1, ASPM, FOXM1, TTK, ARHGAP11A, MYBL2, TYMS, FAM83D, KIF11, HJURP, KIF4A, CKAP2, MELK, KNL1, ATAD2, ECT2, TRIP13, FBXO5, SGO2, STMN1, RAD51AP1, CKS1B, CENPU, MND1, SMC4, CENPN, K PNA2, CENPK, CENPM, TACC3, KIF20B, CKS2, ESCO2, SKA3, PARPBP, CDCA5, PIMREG, UBE2S, AURKA, ASF1B, TMPO, KIF14, OIP5, ORC6, KNSTRN, SKA1, FANCI, GAS2L3, NCAPD2, KIF15, FAM111B, MCM10.

Images