Esophageal squamous cell carcinoma chemoradiotherapy efficacy prediction method and system based on spatial transcriptome niche score

Abstract

The application discloses an esophageal squamous cell carcinoma chemoradiotherapy efficacy prediction method and system based on spatial transcriptome niche score. The system comprises a data acquisition module, a niche identification module, a score calculation module and an efficacy prediction module. By acquiring spatial transcriptome sequencing data of esophageal squamous cell carcinoma tissue samples before chemoradiotherapy, the immune cell enrichment niche and the perivascular niche in the sample are identified and their abundance is calculated, the abundance is input into the niche score model to obtain the niche score, and the efficacy prediction information is output after comparison with the preset threshold. The application changes the source of the efficacy prediction information from the molecular expression level to the spatial organization configuration level, and can realize accurate prediction without complex gene detection and calculation, thereby providing a new auxiliary tool for esophageal squamous cell carcinoma chemoradiotherapy efficacy prediction, and having high clinical transformation potential and application value.

Description

Technical Field

[0001] This invention belongs to the field of biomedical informatics technology, specifically relating to a method and system for predicting the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma based on spatial transcriptome niche scoring. Background Technology

[0002] Esophageal squamous cell carcinoma (ESCC) is a distinctive malignant tumor prevalent in China, characterized by its high malignancy, strong heterogeneity, and significant clinical challenges. Even with the recent emergence of neoadjuvant immunotherapy, a considerable proportion of patients still exhibit treatment resistance, resulting in less-than-expected efficacy. Current research indicates that pretreatment testing of programmed cell death protein ligand 1 (PD-L1) expression levels and tumor mutational burden (TMB) cannot accurately predict patient response.

[0003] Spatial transcriptomics is a high-throughput sequencing technology that has emerged in recent years. This technology can simultaneously acquire gene expression profiles and their spatial coordinates without disrupting the tissue's spatial structure, thereby enabling precise analysis of the spatial distribution and local interaction networks of different cell populations within the tumor microenvironment. Based on spatial transcriptomics data, unsupervised clustering or neighborhood analysis can be used to identify "niches" with specific cellular composition and functional characteristics. These niches are often closely related to key biological processes such as immune infiltration and matrix remodeling.

[0004] Chinese patent CN121294668A discloses a combination of biomarkers and its application in evaluating the efficacy of chemoimmunotherapy for esophageal squamous cell carcinoma. It identifies immune-enriched niches using spatial transcriptome data, then screens differentially expressed genes within these niches and constructs a gene expression weighted scoring model. This technical solution provides an effective evaluation tool at the gene expression level within a niche. However, in its technical approach, niche analysis primarily plays a role in locating and screening immune-enriched regions; efficacy prediction information ultimately comes from the joint detection and weighted calculation of the expression levels of multiple genes within the located region. In other words, existing technologies utilize spatial transcriptome data primarily at the molecular expression level within immune niches; whether the spatial abundance relationship between different functional niches has predictive value for efficacy is currently unreported. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method and system for predicting the efficacy of chemoimmunotherapy for esophageal squamous cell carcinoma based on spatial transcriptome niche scoring. By analyzing the tumor microenvironment niche of spatial transcriptome data, immune cell-enriched niches and perivascular niches are identified, and a novel scoring model is constructed based on their abundance to achieve accurate prediction of the efficacy of chemoimmunotherapy in patients with esophageal squamous cell carcinoma. Unlike existing technologies that screen differentially expressed genes within niches and construct weighted scoring models, this invention expands the information source for efficacy prediction from the molecular expression level of a single niche to the spatial abundance relationship between different functional niches, providing a novel technical approach for evaluating the efficacy of chemoimmunotherapy for esophageal squamous cell carcinoma.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The first aspect of this invention proposes a system for predicting the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma based on spatial transcriptome niche scoring, comprising:

[0008] The data acquisition module is used to acquire spatial transcriptome sequencing data of esophageal squamous cell carcinoma tissue samples before chemoimmunotherapy.

[0009] The niche identification module is used to compare the neighborhood cell composition vectors of each anchor cell in the sample with the pre-determined cluster centers of each niche category based on the spatial transcriptome sequencing data. Each anchor cell is assigned to the niche category corresponding to the nearest cluster center, thereby identifying the immune cell-enriched niche and the perivascular niche of the sample, and calculating their abundance in the sample. The immune cell-enriched niche and the perivascular niche are pre-determined based on the proportion of each cell type in the cluster centers: the niche category with the highest combined proportion of T / NK cells, B cells, and myeloid cells is identified as the immune cell-enriched niche, and the niche category with the highest combined proportion of endothelial cells and pericytes is identified as the perivascular niche.

[0010] The scoring calculation module is used to input the abundance of the immune cell enrichment niche and the perivascular niche into the niche scoring model to obtain the niche score of the sample. The calculation expression of the niche scoring model is as follows:

[0011]

[0012] in, This represents the niche score of the sample. This indicates the abundance of immune cell enrichment niches in the sample. Indicates the abundance of the perivascular niche in the sample;

[0013] The efficacy prediction module is used to compare the niche score with a preset threshold and output efficacy prediction information based on the comparison result; wherein the preset threshold is 0.61, when the niche score is greater than or equal to 0.61, the module outputs prediction information of potential benefits of chemotherapy and immunotherapy; when the niche score is less than 0.61, the module outputs prediction information of potential non-benefits of chemotherapy and immunotherapy.

[0014] Preferably, in the niche identification module, the neighborhood cell composition vector is obtained in the following manner:

[0015] All cells in the sample that have completed cell type annotation are sequentially set as anchor cells. The N nearest neighbor cells to the anchor cell are searched using Euclidean distance. The number of each cell type in the neighbor cells is counted, and the proportion of each cell type to the total number of neighbor cells is calculated to obtain the multidimensional neighborhood cell composition vector.

[0016] Preferably, N is 50.

[0017] Preferably, the cell types are B cells, T / NK cells, myeloid cells, fibroblasts, endothelial cells, pericytes, fibroblast reticulocytes, and epithelial cells.

[0018] Preferably, the predetermined cluster centers for each niche category are obtained in the following manner:

[0019] The neighborhood cell composition vectors of all anchor cells in the discovery queue samples are aggregated and clustered to obtain multiple niche categories, and the average neighborhood cell composition vector of each category is calculated as the cluster center.

[0020] Preferably, the total proportion of T / NK cells, B cells and myeloid cells in the immune cell enrichment niche exceeds 50%; and the total proportion of endothelial cells and pericytes in the perivascular niche exceeds 25%.

[0021] Preferably, the abundance is the percentage of anchor cells belonging to the corresponding niche category in the sample out of the total number of annotated cells in the sample.

[0022] The second aspect of this invention proposes a method for predicting the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma based on spatial transcriptome niche scoring, comprising the following steps:

[0023] Spatial transcriptome sequencing data of ex vivo esophageal squamous cell carcinoma tissue samples prior to chemoimmunotherapy were obtained.

[0024] Based on the spatial transcriptome sequencing data, the neighborhood cell composition vectors of each anchor cell in the sample are compared with the pre-determined cluster centers of each niche category. Each anchor cell is assigned to the niche category corresponding to the nearest cluster center. From this, the immune cell enrichment niche and perivascular niche of the sample are identified, and their abundance in the sample is calculated. The immune cell enrichment niche and perivascular niche are pre-determined based on the proportion of each cell type in the cluster centers: the niche category with the highest combined proportion of T / NK cells, B cells, and myeloid cells is determined as the immune cell enrichment niche, and the niche category with the highest combined proportion of endothelial cells and pericytes is determined as the perivascular niche.

[0025] The abundance of the immune cell enrichment niche and the perivascular niche are input into the niche scoring model to obtain the niche score of the sample. The calculation expression of the niche scoring model is as follows:

[0026]

[0027] in, This represents the niche score of the sample. This indicates the abundance of immune cell enrichment niches in the sample. Indicates the abundance of the perivascular niche in the sample;

[0028] The niche score is compared with a preset threshold, and efficacy prediction information is output based on the comparison result; wherein the preset threshold is 0.61, when the niche score is greater than or equal to 0.61, prediction information of potential benefit of chemotherapy and immunotherapy is output; when the niche score is less than 0.61, prediction information of potential non-benefit of chemotherapy and immunotherapy is output.

[0029] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0030] A fourth aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0031] The beneficial effects of this invention are as follows:

[0032] (1) This invention extends the information source for efficacy prediction from the molecular expression level within the niche to the spatial abundance relationship between different functional niches. It constructs an efficacy prediction scoring model through the abundance ratio of a specific niche, providing a new technical path for evaluating the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma.

[0033] (2) Compared with the existing technology based on gene expression weighted scoring, the present invention eliminates the need for multi-step molecular-level data processing such as differential expression analysis, gene screening, expression level standardization and weighted calculation after niche identification. This significantly reduces intermediate data processing steps, effectively reduces the risk of error accumulation, and helps improve the accuracy and robustness of prediction results.

[0034] (3) The scoring model of the present invention only requires the abundance of two types of niches (immune cell enrichment niche and perivascular niche) as input, without the need to detect the expression level of multiple genes. While maintaining cross-cohort prediction efficacy, it simplifies the model structure and technical process, improves detection efficiency, and provides a convenient and reliable new auxiliary tool for predicting the efficacy of esophageal squamous cell carcinoma chemotherapy and immunotherapy. It has good clinical translation potential and application value. Attached Figure Description

[0035] Figure 1 This is a spatial cell distribution map of representative samples in the discovery cohort of Embodiment 1 of the present invention;

[0036] Figure 2 This is a niche distribution map of representative samples in the discovery queue of Embodiment 1 of the present invention;

[0037] Figure 3 This is a heatmap showing the relative enrichment trends of each niche between the pCR group and the Non-pCR group in the discovery queue of Embodiment 1 of the present invention.

[0038] Figure 4 ROC curves for differentiating pCR and Non-pCR efficacy using multiple candidate indicators (CN4, CN5, CN2, (CN2+CN5)) in the discovery cohort of Embodiment 1 of the present invention;

[0039] Figure 5 ROC curves for differentiating pCR and Non-pCR efficacy using multiple candidate indicators (CN4 / CN2, CN4 / CN5, CN4 / (CN2+CN5), (CN4+CN0+CN1) / (CN3+CN2+CN5)) in the discovery cohort of Embodiment 1 of the present invention;

[0040] Figure 6 This is the ROC curve of the discovery cohort of Embodiment 1 of the present invention, which uses a Lasso regularized logistic regression model to distinguish the efficacy of pCR and Non-pCR.

[0041] Figure 7 ROC curves for differentiating pCR and Non-pCR efficacy using multiple candidate indicators (CN4, CN5, CN2, (CN2+CN5)) in the validation cohort of Embodiment 2 of the present invention;

[0042] Figure 8ROC curves for differentiating pCR and Non-pCR efficacy among multiple candidate indicators (CN4 / CN2, CN4 / CN5, CN4 / (CN2+CN5), (CN4+CN0+CN1) / (CN3+CN2+CN5)) in the validation queue of Embodiment 2 of the present invention;

[0043] Figure 9 This is the ROC curve of the validation queue in Embodiment 2 of the present invention, which uses a Lasso regularized logistic regression model to distinguish the efficacy of pCR and Non-pCR.

[0044] Figure 10 This is a schematic diagram of the esophageal squamous cell carcinoma chemotherapy and immunotherapy efficacy prediction system based on spatial transcriptome niche scoring in Embodiment 4 of the present invention. Detailed Implementation

[0045] To make the objectives and technical solutions of this invention clearer and more complete, the invention will be further described in detail below with reference to embodiments. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. Modifications or equivalent substitutions made by those skilled in the art based on their understanding of the technical solutions of this invention, without departing from the spirit and scope of the technical solutions of this invention, are all within the scope of protection of this invention.

[0046] Unless otherwise specified, the reagents, methods and equipment used in this invention are all conventional reagents, methods and equipment in this technical field.

[0047] Terminology Explanation:

[0048] Chemoimmunotherapy (chemotherapy combined with immunotherapy): refers to a treatment method that combines chemotherapy drugs with immune checkpoint inhibitors.

[0049] A cell bin (single-cell compartment) is a basic data unit in spatial transcriptomics technology, generated by an image-assisted cell segmentation algorithm, representing the spatial location and gene expression profile of a single cell. In this invention, for ease of description, "cell" refers to the aforementioned cell bin. For example, the cell bin that serves as the center point for neighborhood analysis is called the anchor cell, and other cell bins within a certain spatial range around it are called neighboring cells (also known as neighborhood cells).

[0050] It should be noted that in this invention, "cell type" refers to cell classification in a biological sense (such as B cells, endothelial cells, etc.), which is a concept at a different level from "cell" (i.e., cell bin) mentioned above as a data unit.

[0051] Example 1

[0052] This embodiment provides a method for constructing a niche scoring model for predicting the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma. The specific steps are as follows:

[0053] 1. Sample collection and spatial transcriptome sequencing

[0054] This embodiment retrospectively collected pre-treatment biopsy samples from 38 patients clinically diagnosed with surgically resectable locally advanced esophageal squamous cell carcinoma at Zhejiang Cancer Hospital. All biopsy samples were fixed in formalin and embedded in paraffin.

[0055] All patients received neoadjuvant chemoimmunotherapy (chemoimmunotherapy) with cisplatin, paclitaxel, and anti-PD-1 monoclonal antibody prior to surgical resection. Radical resection was performed after chemoimmunotherapy. Pathological response to chemoimmunotherapy was assessed based on the Becker classification of postoperative pathological samples. Becker 1a (residual tumor cells <10%) was classified as a pathologic complete response (pCR) group; Becker 1b (residual tumor cells 10%-50%), Becker 2 (residual tumor cells >50%), and Becker 3 (no regression or progression) were combined into a non-pathologic complete response (Non-pCR) group. In this embodiment, both the pCR and Non-pCR groups were defined according to the aforementioned Becker classification criteria. The pCR group consisted of 18 patients, and the Non-pCR group consisted of 20 patients. The sample collection process was reviewed by the hospital ethics committee and complied with international bioethical guidelines; informed consent was obtained from all participating patients.

[0056] Subsequently, serial sections were prepared from 38 paraffin-embedded pre-treatment puncture samples, and spatial transcriptome sequencing was performed on each sample using the BGI stereo-seq platform. After quality control of the sequencing data, the 38 samples were randomly divided into a discovery cohort and a validation cohort based on efficacy stratification. The discovery cohort consisted of 19 samples, including 9 pathological complete remission (pCR) samples and 10 non-pathological complete remission (Non-pCR) samples; the validation cohort consisted of 19 samples, including 9 pathological complete remission (pCR) samples and 10 non-pathological complete remission (Non-pCR) samples.

[0057] 2. Cell segmentation and annotation

[0058] In a preferred embodiment, for each sample in the discovery queue, a cell bin data carrying spatial transcriptome gene expression matrix is ​​registered with hematoxylin and eosin (H&E) images of adjacent slices using a cell segmentation algorithm (such as CellBin2). Single-cell segmentation is performed based on the cell nucleus morphology of the H&E image to generate cell bin data carrying spatial coordinates (x, y) and gene expression profiles.

[0059] Subsequently, the SPACEL (Spatial Architecture Characterization by DeepLearning) algorithm was used to deconvolve the cell bin data of all 19 discovery cohort samples. SPACEL, by constructing a cell neighborhood map and introducing a graph neural network, can simultaneously utilize the gene expression features and spatial neighborhood relationships of individual cell bins to predict the biological cell type corresponding to each cell bin. For each cell bin, SPACEL outputs its predicted probability for each reference cell type, and the cell type with the highest probability value is determined as the annotation type for that cell bin. Ultimately, a total of 8 cell types were annotated for the 19 samples, including B cells, endothelial cells, epithelial cells, fibroblasts, myeloid cells, pericytes, T / NK cells, and fibroblastic reticular cells (FRC). Figure 1 The spatial cell distribution map of representative samples is shown.

[0060] 3. Neighborhood Analysis and Niche Clustering

[0061] This step performs neighborhood analysis on all annotated cell bins in the 19 samples found in the cohort. In a preferred embodiment, each cell (cell bin) in each sample is sequentially designated as an anchor cell within its own spatial coordinate system. For each anchor cell, the N nearest cells (excluding the anchor cell itself) within the sample are searched using Euclidean distance to form a neighboring cell set, where N is preferably 50. For anchor cells located at the edge of tissue sections, if the number of neighboring cells is less than 50, the actual number of neighboring cells found is used, and this actual number is used as the denominator when calculating the composition ratio. The number of the above 8 cell types in the neighboring cell set of each anchor cell is counted, and the proportion of each cell type to the total number of neighboring cells is calculated to obtain an 8-dimensional neighborhood cell composition vector. ,in, The compositional proportions of B cells, T / NK cells, myeloid cells, fibroblasts, endothelial cells, pericytes, fibroblast reticulocytes, and epithelial cells, respectively, are shown below. ,and .

[0062] Subsequently, the MiniBatchKMeans algorithm was used to aggregate and cluster the 8-dimensional neighborhood cell composition vectors of all anchor cells in the 19 samples. Anchor cells with similar local cell composition patterns were grouped into one class. The optimal number of clusters, k=6, was set, and a fixed random seed was used to ensure consistency of clustering results under the same input conditions. It should be noted that in this embodiment, the input for clustering is the local cell composition vector of each anchor cell, rather than the gene expression vector. This is because the gene expression vector reflects the molecular state of a single cell, while the neighborhood cell composition vector reflects the cell type composition of the local microenvironment in which the cell exists. Using the neighborhood cell composition vector as the clustering basis allows cells in similar microenvironments to be grouped into one class, thereby identifying functional spatial units with specific cell composition characteristics, i.e., cell niches.

[0063] After clustering, it was found that all anchor cells in the queue were assigned to 6 categories, i.e. 6 niches, named cell neighbor 0 (CN0), cell neighbor 1 (CN1), cell neighbor 2 (CN2), cell neighbor 3 (CN3), cell neighbor 4 (CN4) and cell neighbor 5 (CN5) respectively. Figure 2 The niche distribution map of representative samples is shown. It should be noted that the category numbers CN0~CN5 mentioned above are arbitrary labels output by the clustering algorithm, and their order does not constitute a limitation on each category; the attributes of each category are determined based on its average neighborhood cell composition vector (see the steps below). In practical applications, for the sample to be tested, the neighborhood cell composition vector of each anchor cell in the sample can be compared with the cluster centers of each category in the reference clustering results (e.g., the clustering results of the discovery queue in this embodiment), and each anchor cell can be assigned to the category corresponding to the nearest cluster center. This eliminates the need to re-perform clustering on a single sample to achieve niche category assignment.

[0064] 4. Niche Abundance Calculation

[0065] For each sample in the discovery cohort, the number of anchor cells in each niche category is counted, and the percentage of each anchor cell relative to the total number of annotated cells in that sample is calculated. This percentage is used as the abundance of each niche category for that sample. That is:

[0066]

[0067] in, Indicates the first Abundance of similar ecological niches, Indicates the first The number of anchor cells included in a niche. This represents the total number of annotated cells in the sample. When there are no anchor cells belonging to a certain category in a sample, the abundance of that niche in that sample is 0.

[0068] This abundance value aggregates single-cell-level niche tags into sample-level microenvironmental composition features, which are then used for subsequent sample-level intergroup comparisons and candidate indicator evaluations.

[0069] 5. Relative enrichment analysis of ecological niches between the pCR group and the Non-pCR group

[0070] In the discovery cohort, after calculating the abundance of each niche for each sample, the enrichment bias of each niche was analyzed between the pCR group (9 cases) and the Non-pCR group (10 cases). Specifically, the proportion of cells belonging to a certain niche category in the anchor cells aggregated within a certain efficacy group was taken as observed, and the proportion of cells belonging to that niche category in the anchor cells aggregated from all samples in the discovery cohort was taken as expected. The ratio of the two was calculated as the relative enrichment degree (R_o / e, observed / expected ratio). Figure 3 The heatmap shows that in the pCR group, CN4, CN0, and CN1 exhibit a relative enrichment trend (R_o / e > 1); while in the Non-pCR group, CN3, CN2, and CN5 exhibit a relative enrichment trend (R_o / e > 1). The enrichment differences between CN4 and CN2 / CN5 in the two groups are particularly significant.

[0071] 6. Construction of the Niche Scoring Model

[0072] Based on the above niche enrichment analysis results, it was found that different efficacy groups in the cohort exhibited differentiated niche enrichment characteristics, especially CN4, CN2, and CN5, where the relative enrichment differences between the two groups were most prominent. Accordingly, this embodiment constructs various candidate indicators for systematic evaluation around the abundance of the above three niches and their combination relationships, including single-niche abundance indicators (CN4, CN2, CN5), multi-niche combination indicators (CN2+CN5), ratio models with different structures (CN4 / CN2, CN4 / CN5, CN4 / (CN2+CN5), and ratios based on the merging of enrichment directions of all six niches (CN4+CN0+CN1) / (CN3+CN2+CN5)), and a Lasso regularized logistic regression model (using the abundance of all six niches as input features). Then, using each candidate indicator as a continuous variable and the pCR / Non-pCR efficacy group as a categorical variable, receiver operating characteristic (ROC) curve analysis was used to evaluate the efficacy discrimination ability of each candidate indicator, and the area under the curve (AUC) was calculated. The results are as follows. Figures 4-6 As shown:

[0073] (1) Single index of niche abundance: The AUCs of CN4, CN5 and CN2 were 0.900, 0.767 and 0.622, respectively;

[0074] (2) The sum of abundance of CN2 and CN5 (CN2+CN5): AUC was 0.767, which was not improved compared with CN5 alone;

[0075] (3) The ratio (CN4+CN0+CN1) / (CN3+CN2+CN5) constructed by completely merging all ecological niches according to the inter-group enrichment direction: AUC is 0.867;

[0076] (4) Using the abundance of all 6 ecological niches as input features, a multivariate model was constructed using Lasso regularized logistic regression and 5-fold cross-validation: the AUC was 0.7889;

[0077] (5) The ratios of CN4 constructed with CN2 and CN5 respectively: The AUC of CN4 / CN2 and CN4 / CN5 are both 0.911;

[0078] (6) The ratio of CN4 to (CN2+CN5) constructed by CN4 / (CN2+CN5): AUC is 0.922.

[0079] Among the candidate indicators mentioned above, CN4 / CN2, CN4 / CN5, and CN4 / (CN2+CN5) exhibited high discriminative power in the discovery queue. Based on this, this embodiment initially constructs the following three candidate niche scoring models:

[0080] NS1 = CN4 / CN2

[0081] NS2 = CN4 / CN5

[0082] NS3 = CN4 / (CN2+CN5)

[0083] In this context, NS1, NS2, and NS3 represent the niche scores of the samples, while CN4, CN2, and CN5 represent the abundance of the corresponding niche categories in the samples. It should be noted that CN0 to CN5 are niche category symbols obtained from clustering in this embodiment; their numbers are arbitrary labels output by the clustering algorithm and do not constitute a limitation on each niche category.

[0084] 7. Cell type composition analysis of ecological niches

[0085] To clarify the cell type composition basis of the aforementioned ecological niches, a neighborhood cell composition analysis was further performed on anchor cells under each ecological niche category in all samples of the discovery cohort. Specifically, for a given ecological niche category, an 8-dimensional neighborhood cell composition vector was extracted for each anchor cell in that category. These vectors were then averaged element-wise, i.e., the average proportion of all anchor cells in each cell type dimension was calculated. The resulting 8 averages were combined into a new 8-dimensional vector, which serves as the average neighborhood cell composition vector for that ecological niche, and also as the cluster center for that ecological niche category.

[0086] The analysis results (Table 1) show that the average neighborhood cell composition vectors of the six ecological niches exhibit the following characteristics:

[0087] In the average neighborhood cell composition vector of CN4, the combined proportion of T / NK cells, B cells, and myeloid cells is greater than 50%, the highest among the six niches. Therefore, the niche category with the highest combined proportion of T / NK cells, B cells, and myeloid cells is defined as the immune cell-enriched niche. In the average neighborhood cell composition vector of CN2, fibroblasts account for greater than 40%, the highest among the six niches. Therefore, the niche category with the highest proportion of fibroblasts is defined as the fibroblast-enriched niche. In the average neighborhood cell composition vector of CN5, the combined proportion of endothelial cells and pericytes is greater than 25%, also the highest among the six niches. Therefore, the niche category with the highest combined proportion of endothelial cells and pericytes is defined as the perivascular niche. In addition, the average neighborhood cell composition of CN0 is dominated by epithelial cells, accounting for more than 70%; in CN1, epithelial cells account for the highest proportion (about 31%), but T / NK cells, B cells and fibroblasts also account for a considerable proportion, and no single dominant type has been formed; in CN3, epithelial cells account for more than 40%, but much less than CN0, showing a transitional composition characteristic of the tumor epithelial region extending to the surrounding microenvironment.

[0088] Table 1. Average neighborhood cell composition vectors (cluster centers) for 6 niches

[0089]

[0090] Note: The sum of CN0 and CN1 is 100.01%, and the sum of CN4 is 99.99%. This is a minor rounding error caused by floating-point rounding and is not a calculation error.

[0091] Example 2

[0092] This embodiment further evaluates the candidate models and metrics constructed in Embodiment 1 independently in the validation queue.

[0093] Specifically, for the spatial transcriptome sequencing data of 19 samples in the validation cohort, the same cell segmentation and annotation, neighborhood analysis, and neighborhood cell composition vector construction were performed sequentially as in Example 1.

[0094] Regarding the assignment of niche categories, in a preferred embodiment, the neighborhood cell composition vectors of each anchor cell in each sample of the validation cohort are compared with the Euclidean distances of the cluster centers (i.e., the average neighborhood cell composition vectors of each category) of the six niche categories obtained in Example 1, and each anchor cell is assigned to the niche category corresponding to the nearest cluster center. The attributes of each niche category (such as immune cell enrichment niche, perivascular niche, etc.) directly follow the determination results of Example 1.

[0095] Subsequently, for each sample in the validation cohort, the abundance of each niche category, the niche scores of each candidate indicator and each candidate model were calculated. Using pCR / Non-pCR efficacy grouping as the categorical variable, ROC curve analysis was employed to assess their independent discrimination ability. The results are as follows: Figures 7-9 As shown:

[0096] (1) Single index of niche abundance: The AUCs of CN4, CN5 and CN2 were 0.633, 0.722 and 0.633, respectively;

[0097] (2) The sum of abundance of CN2 and CN5 (CN2+CN5): AUC is 0.700;

[0098] (3) The ratio of complete merging constructed according to the inter-group enrichment direction (CN4+CN0+CN1) / (CN3+CN2+CN5): AUC is 0.822;

[0099] (4) Using the abundance of all 6 ecological niches as input features, a multivariate model was constructed using Lasso regularized logistic regression and 5-fold cross-validation: the AUC was 0.544;

[0100] (5) Candidate model NS1 (CN4 / CN2): AUC is 0.822;

[0101] (6) Candidate model NS2 (CN4 / CN5): AUC is 0.911;

[0102] (7) Candidate model NS3 (CN4 / (CN2+CN5)): AUC is 0.889.

[0103] Combination Figures 4-9 The combined analysis of the results from the discovery and validation cohorts shows that candidate model NS2 maintained the same high discriminative power in the validation cohort as in the discovery cohort. However, NS1 and NS3, which include fibroblast-enriched niches as denominator components, both exhibited varying degrees of power decline in the validation cohort. The single indicator CN4 had an AUC of 0.900 in the discovery cohort, which decreased to 0.633 in the validation cohort, indicating insufficient stability of single niche abundance indicators across different cohorts. The ratio of complete merging based on inter-group enrichment direction (CN4+CN0+CN1) / (CN3+CN2+CN5) also had a significantly lower AUC than NS2's 0.911, suggesting that in building models based on spatial niche abundance, including more niche categories does not necessarily improve predictive performance. The remaining candidate indicators generally had low AUCs in both the discovery and validation cohorts, typically ranging from 0.6 to 0.8; the Lasso model even dropped to 0.544 in the validation cohort, approaching a random level.

[0104] The above results show that among all candidate indicators, NS2, which only requires input of two niche abundances, exhibits the best and most stable predictive performance in the independent validation cohort, with its AUC significantly outperforming other single-indicator, multi-niche combination, and Lasso multivariate models. Therefore, NS2=CN4 / CN5 is determined as the preferred niche scoring model in this embodiment of the invention. Represented by functional niches, the general form of this preferred niche scoring model can be expressed as:

[0105]

[0106] in, This indicates the abundance of immune cell enrichment niches in the sample. This indicates the abundance of the perivascular niche in the sample. In this embodiment, , The abundances correspond to CN4 and CN5, respectively.

[0107] It is worth noting that, from the perspective of the functional composition of the microenvironment, CN4 is an immune cell enrichment niche, and its abundance reflects the degree of spatial infiltration of immune cells; CN5 is a perivascular niche, and its abundance reflects the perivascular microenvironment characteristics dominated by vascular endothelial cells and pericytes; CN2 is a fibroblast enrichment niche, and its abundance reflects the matrix barrier characteristics dominated by fibroblasts in the local microenvironment. From a spatial biology perspective, CN2 and CN5 characterize the matrix features in the tumor microenvironment from two dimensions: physical barrier and vascular permeability barrier, respectively. The combination of the two with the immune cell enrichment niche to construct the abundance ratio (i.e., NS3=CN4 / (CN2+CN5)) should, in theory, more comprehensively reflect the spatial balance between immunity and matrix. However, validation data show that the AUC of NS3 (CN4 / (CN2+CN5)) is 0.889, while the AUC of NS2 (CN4 / CN5), which only contains CN5, is 0.911, showing a clear efficacy gap; the AUC of NS1 (CN4 / CN2) is only 0.822. The above comparative results indicate that, in this spatial niche system, the introduction of fibroblast enrichment niche abundance information did not improve the model's predictive efficacy or cross-cohort stability. This further suggests that, at the spatial niche level, the influence of different functional niches on the efficacy of chemoimmunotherapy is not a simple linear additive relationship, and their spatial interactions and their impact on efficacy prediction cannot be directly inferred based on conventional understanding.

[0108] Example 3

[0109] For the preferred scoring model NS2 determined in Example 2, this example first uses the niche score of each sample as a continuous variable and the pCR / Non-pCR efficacy group as a categorical variable in the discovery queue. The optimal cutoff value is determined using the principle of maximizing the Youden's Index (sensitivity + specificity - 1). The Youden's Index reaches its maximum when the threshold is set to 0.61. Therefore, 0.61 is used as the critical value for efficacy prediction in this model; efficacy prediction is better for niche scores above this threshold than for scores below it. Specifically, when the niche score is greater than or equal to 0.61, it is classified as pCR (positive), predicting a potential beneficiary of chemoimmunotherapy; when the niche score is less than 0.61, it is classified as Non-pCR (negative), predicting a potential non-beneficial recipient of chemoimmunotherapy. When the numerator and / or denominator of the scoring model is 0, the sample is considered undetermined and requires further analysis using other indicators.

[0110] Furthermore, based on this threshold, the predictive efficacy of 19 samples in the discovery cohort was evaluated, and the corresponding confusion matrix is ​​shown in Table 2.

[0111] Table 2 shows the queue confusion matrix.

[0112]

[0113] Furthermore, the threshold determined in the discovery queue was directly applied to the validation queue, that is, the efficacy prediction assessment of 19 samples in the validation queue was carried out with a threshold of 0.61, and the corresponding confusion matrix is ​​shown in Table 3.

[0114] Table 3. Validation Queue Confusion Matrix

[0115]

[0116] The results indicate that the preferred model NS2 maintained high and stable predictive efficacy in both cohorts, supporting its use as a predictive tool for the efficacy of chemoimmunotherapy for esophageal squamous cell carcinoma.

[0117] It should be noted that, in practical applications, the following procedure can be used to predict the efficacy of test samples from subjects:

[0118] (1) Obtain spatial transcriptome sequencing data of esophageal squamous cell carcinoma tissue samples from subjects before chemotherapy and immunotherapy. Following the methods described in steps 1 to 3 of Example 1, complete cell segmentation and annotation, neighborhood analysis, and construct neighborhood cell composition vectors for each anchor cell in the sample.

[0119] (2) For each anchor cell in the sample to be tested, its neighborhood cell composition vector is compared with the Euclidean distance of the cluster centers of the six niche categories obtained in Example 1 (i.e., the average neighborhood cell composition vector of each category), and each anchor cell is assigned to the niche category corresponding to the nearest cluster center; the functional attributes of each niche category are determined according to the determination result of step 7 in Example 1, that is, among all niche categories, the category with the highest total proportion of T / NK cells, B cells and myeloid cells is the immune cell enrichment niche, and the category with the highest total proportion of endothelial cells and pericytes is the perivascular niche;

[0120] (3) Count the number of anchor cells belonging to the immune cell enrichment niche and the perivascular niche in the sample to be tested, and calculate their abundance according to the method described in step 4 of Example 1. and And substitute it into the niche scoring model Calculate the niche score for this sample; when and / or When the value is 0, the sample is considered undetermined and requires further analysis in conjunction with other indicators;

[0121] (4) Compare the niche score with a preset threshold (preferably 0.61). Subjects with scores greater than or equal to the threshold are predicted as potential beneficiaries of chemotherapy and immunotherapy, while subjects with scores lower than the threshold are predicted as potential non-beneficiaries. Output efficacy prediction information.

[0122] Example 4

[0123] See Figure 10 This embodiment further provides a system for predicting the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma based on spatial transcriptome niche scoring. This system integrates the methods described in the above embodiments to predict the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma patients. Specifically, it includes the following modules:

[0124] The data acquisition module is used to acquire spatial transcriptome sequencing data of esophageal squamous cell carcinoma tissue samples before chemoimmunotherapy.

[0125] The niche identification module is used to compare the neighborhood cell composition vector of each anchor cell in the sample with the pre-determined cluster centers of each niche category based on the spatial transcriptome sequencing data, classify each anchor cell into the niche category corresponding to the nearest cluster center, identify the immune cell enrichment niche and perivascular niche of the sample, and calculate the abundance of the two in the sample respectively.

[0126] The scoring calculation module is used to input the abundance of the immune cell enrichment niche and the perivascular niche into the niche scoring model to obtain the niche score of the sample. The calculation expression of the niche scoring model is as follows:

[0127]

[0128] in, This represents the niche score of the sample. This indicates the abundance of immune cell enrichment niches in the sample. Indicates the abundance of the perivascular niche in the sample;

[0129] The efficacy prediction module is used to compare the niche score with a preset threshold and output efficacy prediction information based on the comparison result.

[0130] For detailed descriptions of the functions of each module, please refer to the relevant descriptions in the aforementioned embodiments, which will not be repeated here.

[0131] It should be noted that in the data acquisition module, the esophageal squamous cell carcinoma tissue sample of the subject refers to the puncture tumor tissue of the esophageal squamous cell carcinoma patient before receiving chemotherapy and immunotherapy. After being fixed in formalin and embedded in paraffin, spatial transcriptome data is acquired using the stereo-seq technology platform. Then, based on the method of the aforementioned embodiment, the abundance value information of the corresponding ecological niche is identified and calculated.

[0132] In the efficacy prediction module, this embodiment preferably sets the preset efficacy threshold to 0.61. Subjects with scores higher than or equal to this threshold are predicted to have a good response to chemotherapy and immunotherapy and are potential beneficiaries of chemotherapy and immunotherapy; subjects with scores lower than this threshold are predicted to have a poor response to chemotherapy and immunotherapy and are potential non-beneficiaries.

[0133] Example 5

[0134] This embodiment further provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the following method:

[0135] Spatial transcriptome sequencing data of ex vivo esophageal squamous cell carcinoma tissue samples prior to chemoimmunotherapy were obtained.

[0136] Based on the spatial transcriptome sequencing data, the neighborhood cell composition vectors of each anchor cell in the sample are compared with the pre-determined cluster centers of each niche category. Each anchor cell is assigned to the niche category corresponding to the nearest cluster center. The immune cell enrichment niche and perivascular niche of the sample are identified, and their abundance in the sample is calculated.

[0137] The abundance of the immune cell enrichment niche and the perivascular niche are input into the niche scoring model to obtain the niche score of the sample. The calculation expression of the niche scoring model is as follows:

[0138]

[0139] in, This represents the niche score of the sample. This indicates the abundance of immune cell enrichment niches in the sample. Indicates the abundance of the perivascular niche in the sample;

[0140] The niche score is compared with a preset threshold, and therapeutic effect prediction information is output based on the comparison result.

[0141] Example 6

[0142] This embodiment further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the following method:

[0143] Spatial transcriptome sequencing data of ex vivo esophageal squamous cell carcinoma tissue samples prior to chemoimmunotherapy were obtained.

[0144] Based on the spatial transcriptome sequencing data, the neighborhood cell composition vectors of each anchor cell in the sample are compared with the pre-determined cluster centers of each niche category. Each anchor cell is assigned to the niche category corresponding to the nearest cluster center. The immune cell enrichment niche and perivascular niche of the sample are identified, and their abundance in the sample is calculated.

[0145] The abundance of the immune cell enrichment niche and the perivascular niche are input into the niche scoring model to obtain the niche score of the sample. The calculation expression of the niche scoring model is as follows:

[0146]

[0147] in, This represents the niche score of the sample. This indicates the abundance of immune cell enrichment niches in the sample. Indicates the abundance of the perivascular niche in the sample;

[0148] The niche score is compared with a preset threshold, and therapeutic effect prediction information is output based on the comparison result.

Claims

1. A predictive system for the efficacy of chemoimmunotherapy for esophageal squamous cell carcinoma based on spatial transcriptome niche scoring, characterized in that, include: The data acquisition module is used to acquire spatial transcriptome sequencing data of esophageal squamous cell carcinoma tissue samples before chemoimmunotherapy. The niche identification module is used to compare the neighborhood cell composition vectors of each anchor cell in the sample with the pre-determined cluster centers of each niche category based on the spatial transcriptome sequencing data. Each anchor cell is assigned to the niche category corresponding to the nearest cluster center, thereby identifying the immune cell-enriched niche and the perivascular niche of the sample, and calculating their abundance in the sample. The immune cell-enriched niche and the perivascular niche are pre-determined based on the proportion of each cell type in the cluster centers: the niche category with the highest combined proportion of T / NK cells, B cells, and myeloid cells is identified as the immune cell-enriched niche, and the niche category with the highest combined proportion of endothelial cells and pericytes is identified as the perivascular niche. The scoring calculation module is used to input the abundance of the immune cell enrichment niche and the perivascular niche into the niche scoring model to obtain the niche score of the sample. The calculation expression of the niche scoring model is as follows: ； in, This represents the niche score of the sample. This indicates the abundance of immune cell enrichment niches in the sample. Indicates the abundance of the perivascular niche in the sample; The efficacy prediction module is used to compare the niche score with a preset threshold and output efficacy prediction information based on the comparison result; wherein the preset threshold is 0.61, when the niche score is greater than or equal to 0.61, the module outputs prediction information of potential benefits of chemotherapy and immunotherapy; when the niche score is less than 0.61, the module outputs prediction information of potential non-benefits of chemotherapy and immunotherapy.

2. The system according to claim 1, characterized in that, In the niche identification module, the neighborhood cell composition vector is obtained in the following way: All cells in the sample that have completed cell type annotation are sequentially set as anchor cells. The N nearest neighbor cells to the anchor cell are searched using Euclidean distance. The number of each cell type in the neighbor cells is counted, and the proportion of each cell type to the total number of neighbor cells is calculated to obtain the multidimensional neighborhood cell composition vector.

3. The system according to claim 2, characterized in that, The value of N is 50.

4. The system according to claim 2, characterized in that, The cell types mentioned are B cells, T / NK cells, myeloid cells, fibroblasts, endothelial cells, pericytes, fibroblast reticulocytes, and epithelial cells.

5. The system according to claim 1, characterized in that, The pre-determined cluster centers for each niche category are obtained in the following manner: The neighborhood cell composition vectors of all anchor cells in the discovery queue samples are aggregated and clustered to obtain multiple niche categories, and the average neighborhood cell composition vector of each category is calculated as the cluster center.

6. The system according to claim 1, characterized in that, In the immune cell enrichment niche, the combined proportion of T / NK cells, B cells, and myeloid cells exceeds 50%; in the perivascular niche, the combined proportion of endothelial cells and pericytes exceeds 25%.

7. The system according to claim 2, characterized in that, The abundance is the percentage of anchor cells belonging to the corresponding niche category in the sample out of the total number of annotated cells in the sample.

8. A method for predicting the efficacy of chemotherapy and immunotherapy for esophageal squamous cell carcinoma based on spatial transcriptome niche scoring, characterized in that, Includes the following steps: Spatial transcriptome sequencing data of ex vivo esophageal squamous cell carcinoma tissue samples prior to chemoimmunotherapy were obtained. Based on the spatial transcriptome sequencing data, the neighborhood cell composition vectors of each anchor cell in the sample are compared with the pre-determined cluster centers of each niche category. Each anchor cell is assigned to the niche category corresponding to the nearest cluster center. From this, the immune cell enrichment niche and perivascular niche of the sample are identified, and their abundance in the sample is calculated. The immune cell enrichment niche and perivascular niche are pre-determined based on the proportion of each cell type in the cluster centers: the niche category with the highest combined proportion of T / NK cells, B cells, and myeloid cells is determined as the immune cell enrichment niche, and the niche category with the highest combined proportion of endothelial cells and pericytes is determined as the perivascular niche. The abundance of the immune cell enrichment niche and the perivascular niche are input into the niche scoring model to obtain the niche score of the sample. The calculation expression of the niche scoring model is as follows: ； in, This represents the niche score of the sample. This indicates the abundance of immune cell enrichment niches in the sample. Indicates the abundance of the perivascular niche in the sample; The niche score is compared with a preset threshold, and efficacy prediction information is output based on the comparison result; wherein the preset threshold is 0.61, when the niche score is greater than or equal to 0.61, prediction information of potential benefit of chemotherapy and immunotherapy is output; when the niche score is less than 0.61, prediction information of potential non-benefit of chemotherapy and immunotherapy is output.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in claim 8.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in claim 8.

Images