Data processing device for predicting immunotherapy target and computer readable storage medium

Abstract

The invention discloses a data processing device for predicting an immunotherapy target in the field of bioinformatics and a computer readable storage medium. The technical problem to be solved by the invention is how to predict a target spot of immunotherapy of a tumor patient. According to the computer device, expression quantities of genes under different cell subgroups in subjects responding to immunotherapy and not responding to immunotherapy serve as pre-input, an immunotherapy response result serves as an output training model, input feature importance value sequences are obtained based on the model, and first model input features are obtained; the importance values of the features in the model are explained through Shapley Value, second model input features are obtained based on the ranking, and the intersection of the first model input features and the second model input features is selected to obtain potential targets related to immunotherapy. The invention can be applied to the development and / or preparation of products related to disease immunotherapy or products for layering patients.

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics, and in particular relates to a data processing device and a computer-readable storage medium for predicting immunotherapy targets. Background Art

[0002] The emergence of large amounts of data has also brought new challenges to data analysis. With the reduction of sequencing costs, the scale and complexity of biological data have increased significantly. The impact of data screening, noise reduction and other pre-processing on experimental results has become increasingly greater. Reasonable data processing methods have gradually become an essential means to obtain scientific conclusions. With the increase in data scale, the field of machine learning has also developed rapidly. From statistical learning to large models based on neural networks, technological innovations in the field of machine learning have also provided new tools for biological data processing. Single-cell sequencing is a high-resolution genomics technology that aims to comprehensively analyze the gene expression of a single cell. Traditional gene expression analysis techniques usually measure the overall gene expression levels of a large number of cells, while single-cell sequencing technology can reveal the differences between cells and explore the diversity and function of individual cells. With the rapid development of single-cell sequencing, the quality and quantity of tumor immune microenvironment sequencing data sets have made great leaps. By studying the single-cell sequencing data of patients, we can have a more comprehensive understanding of the expression changes of different cell types during the disease process, providing new possibilities for personalized treatment.

[0003] In the field of cancer immunotherapy, an important challenge is how to screen for features that are helpful in predicting immunotherapy response, thereby screening for potential therapeutic targets. A large number of machine learning algorithms can use single-cell sequencing data to predict patients' immunotherapy responsiveness, but which data is more important for prediction remains a challenging question.

[0004] LIME (Local Interpretable Model-Agnostic Explanations) is a machine learning model explanation tool that explains the decisions of black-box models by generating locally interpretable models. Machine learning models are often difficult to explain due to their complex internal structures and parameters. LIME does not attempt to understand the entire model, but explains the contribution of features to the prediction results at the feature level, which reflects the importance of features to some extent. This method first perturbs the features of a specific sample to generate some samples; then these perturbed samples are used as inputs to the model to be explained to obtain a set of prediction results; then, using techniques such as Lasso regression or tree-based feature importance, a set of feature subsets with the highest correlation with the prediction results is selected; then, an interpretable model for predicting model output is trained on the perturbed samples of these features, such as linear regression and decision trees; finally, this interpretable model is used to explain the original model's prediction of this sample.

[0005] Shapley Value is a concept in cooperative game theory that also plays a role in explaining the predictions of machine learning models. It assigns a relative importance score to each feature by traversing all feature combinations and calculating the average marginal contribution of each feature to the model prediction. The method relies on the idea that features may interact in complex ways, and their contributions to the prediction results should be evaluated jointly rather than individually. Like the LIME method, Shapley Value calculates the contribution of features on a single sample to the model prediction.

[0006] SBM is a method for measuring the similarity between rankings. elements, calculate their intersection size / , traverse Find the average and finally get the similarity of the two sorts. Summary of the invention

[0007] The technical problem to be solved by the present invention is how to predict the targets of immunotherapy for subjects (tumor patients) and / or how to screen or assist in screening targets related to immunotherapy based on single-cell sequencing data and / or how to screen or assist in screening targets that tumor patients respond to immunotherapy.

[0008] In order to solve the above technical problem, the present invention first provides a data processing device, including a memory, a processor, and a computer program stored in the memory, wherein the processor can execute the computer program to implement the following steps: S1) Data reception: receiving the expression level M of genes in cell subpopulations from single-cell sequencing data of subjects who have responded to immunotherapy and the expression level N of genes in cell subpopulations from single-cell sequencing data of subjects who have not responded to immunotherapy; S2) obtaining the first model input feature and the second model input feature: taking the expression level of the gene in the cell subpopulation with a difference between the expression level M and the expression level N as the pre-input feature, taking the immunotherapy response result as the output, cyclically sampling the pre-input feature as the input, using the machine learning algorithm to train the immunotherapy response model, obtaining the importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance value P1 from high to low, and selecting the pre-input feature with the highest ranking to obtain the first model input feature; using the Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input feature to the immunotherapy response model, obtaining the importance value P2 of the pre-input feature, sorting the importance value P2 from high to low, and selecting the pre-input feature with the highest ranking to obtain the second model input feature; S3) Obtaining immunotherapy targets: taking the intersection of the first model input features and the second model input features to obtain immunotherapy related targets.

[0009] The data processing device mentioned above may be a computer device.

[0010] In the above-mentioned data processing device, the machine learning algorithm may be a random forest. The importance value P2 of the pre-input feature may be obtained by calculating the average value of the marginal contribution rate of the pre-input feature to the immunotherapy response model using the Shapley Value method. In the above-mentioned data processing device, the immunotherapy-responsive subject and the immunotherapy-unresponsive subject may be a tumor patient. The subject may also be a person suffering from one or more diseases. The immunotherapy response result may be a response to or no response to immunotherapy.

[0011] In order to solve the above technical problems, the present invention also provides a device for predicting or assisting in predicting immunotherapy targets, and the device may include the following modules: A1) Data receiving module: used to receive the expression level M of genes in cell subpopulations in the single-cell sequencing data of subjects who have responded to immunotherapy and the expression level N of genes in cell subpopulations in the single-cell sequencing data of subjects who have not responded to immunotherapy; A2) Obtaining the first model input feature and the second model input feature module: taking the expression levels of the genes in the cell subpopulations with differences in the expression levels M and N as pre-input features, taking the immunotherapy response results as outputs, cyclically sampling the pre-input features as inputs, using a machine learning algorithm to train an immunotherapy response model, obtaining the importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance values ​​P1 from high to low, and selecting the pre-input features with the highest ranking to obtain the first model input feature; using the Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input features to the immunotherapy response model, obtaining the importance value P2 of the pre-input features, sorting the importance values ​​P2 from high to low, and selecting the pre-input features with the highest ranking to obtain the second model input feature; A3) Immunotherapy target acquisition module: used to take the intersection of the first model input feature and the second model input feature to obtain immunotherapy related targets.

[0012] In the above device, the machine learning algorithm may be a random forest. The importance value P2 of the pre-input feature may be obtained by calculating the average value of the marginal contribution rate of the pre-input feature to the immunotherapy response model using the Shapley Value method. The immunotherapy-responsive subject and the immunotherapy-unresponsive subject may be tumor patients. The immunotherapy-responsive subject and the immunotherapy-unresponsive subject may be tumor patients.

[0013] The subject may also be a human suffering from one or more diseases. The immunotherapy response outcome may be a response or non-response to immunotherapy.

[0014] In order to solve the above technical problems, the present invention also provides a method for predicting or assisting in predicting immunotherapy targets, which may include the following steps: B1) Data reception: receiving the gene expression levels M of cell subpopulations in the single-cell sequencing data of subjects who have responded to immunotherapy and the gene expression levels N of cell subpopulations in the single-cell sequencing data of subjects who have not responded to immunotherapy; B2) Obtaining the first model input feature and the second model input feature: taking the expression level of the gene in the cell subpopulation with a difference in the expression level M and the expression level N as the pre-input feature, taking the immunotherapy response result as the output, cyclically sampling the pre-input feature as the input, using the machine learning algorithm to train the immunotherapy response model, obtaining the importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance value P1 from high to low, and selecting the pre-input feature with the highest ranking to obtain the first model input feature; using the Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input feature to the immunotherapy response model, obtaining the importance value P2 of the pre-input feature, sorting the importance value P2 from high to low, and selecting the pre-input feature with the highest ranking to obtain the second model input feature; B3) Obtaining immunotherapy targets: taking the intersection of the first model input features and the second model input features to obtain immunotherapy-related targets.

[0015] In the above method, the machine learning algorithm may be a random forest. The importance value P2 of the pre-input feature may be obtained by calculating the average value of the marginal contribution rate of the pre-input feature to the immunotherapy response model using the Shapley Value method. The immunotherapy responding subject and the immunotherapy non-responding subject may be tumor patients.

[0016] The subject may also be a human suffering from one or more diseases. The immunotherapy response outcome may be a response or non-response to immunotherapy.

[0017] The above method may not include the step of obtaining a biological sample from an animal. The above method may not be based on a living human or animal body, but only on data. The above method may be an information processing method in which all steps are implemented by a data processing device such as a computer.

[0018] In order to solve the above technical problem, the present invention further provides a computer-readable storage medium storing a computer program, wherein the computer program can enable a computer to execute the following steps: C1) Data reception: receiving the gene expression levels M of cell subpopulations in the single-cell sequencing data of subjects who have responded to immunotherapy and the gene expression levels N of cell subpopulations in the single-cell sequencing data of subjects who have not responded to immunotherapy; C2) Obtaining the first model input feature and the second model input feature: taking the expression level of the gene in the cell subpopulation with a difference between the expression level M and the expression level N as the pre-input feature, taking the immunotherapy response result as the output, cyclically sampling the pre-input feature as the input, using the machine learning algorithm to train the immunotherapy response model, obtaining the importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance value P1 from high to low, and selecting the pre-input feature with the highest ranking to obtain the first model input feature; using the Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input feature to the immunotherapy response model, obtaining the importance value P2 of the pre-input feature, sorting the importance value P2 from high to low, and selecting the pre-input feature with the highest ranking to obtain the second model input feature; C3) Obtaining immunotherapy targets: taking the intersection of the first model input features and the second model input features to obtain immunotherapy-related targets.

[0019] In the above-mentioned computer-readable storage medium, the machine learning algorithm may be a random forest. The importance value P2 of the pre-input feature may be obtained by calculating the average value of the marginal contribution rate of the pre-input feature to the immunotherapy response model using the Shapley Value method. The immunotherapy-responsive subject and the immunotherapy-unresponsive subject may be tumor patients.

[0020] The subject may also be a human suffering from one or more diseases. The immunotherapy response outcome may be a response or non-response to immunotherapy.

[0021] Any of the following applications of the data processing device described above also falls within the protection scope of the present invention: E1) Application in the development and / or manufacture of products for disease patient stratification; E2) Use in the development and / or preparation of immunotherapy products.

[0022] Any of the following applications of the above-mentioned device and / or the above-mentioned computer-readable storage medium also falls within the protection scope of the present invention: F1) Application in the development and / or manufacture of products for disease patient stratification; F2) Use in the development and / or manufacture of immunotherapeutic products.

[0023] The cell subpopulation-gene expression combination mentioned above can be a cell subpopulation and the expression level of a specific gene in the cell subpopulation. The specific gene of the cell subpopulation is the same gene expressed in each cell in the cell subpopulation.

[0024] The product mentioned above may be a reagent and / or a kit or a medicine.

[0025] The above application or method is a non-disease diagnosis application or method. The above application or method does not have the direct purpose of obtaining disease diagnosis results or health conditions of living human or animal bodies.

[0026] The above-mentioned application or method is not for the purpose of treating a disease. The above-mentioned application or method is not intended to restore or obtain health or reduce pain in a living human or animal body.

[0027] The present invention uses the characteristics of machine learning algorithms and interpretability methods to propose a screening process for important features related to the response to immunotherapy in cancer patients, and obtains a ranking of the importance of gene expression levels in a group of cell subpopulations related to immunotherapy response. For the gene expression levels of some top-ranked cell subpopulations, they are screened as potential therapeutic targets. Finally, the present invention screened out the CD8+ Tem cell subpopulation. NKG7 , ISG15+ macrophage subsets CXCR4 , CD8+ MAIT cell subsets CD27 The expression levels of these three potential therapeutic targets are closely related to the patient's treatment responsiveness. The present invention analyzes the possible processes that affect the responsiveness of immunotherapy through relevant literature. In summary, the present invention develops a prediction model for the responsiveness of cancer patients to immunotherapy, establishes a screening method for genes related to tumor immunotherapy sensitivity, and provides a reference solution for other sensitive gene screening problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the process of the present invention.

[0029] Figure 2 This is the process of cyclically screening input features of the present invention.

[0030] Figure 3 Rank the importance of cell subset-genes (top 20).

[0031] Figure 4 This is the ROC curve of the model trained using the gene expression data in cell subpopulations.

[0032] Figure 5 It is the Shapley-Value of the cell subpopulation-gene. The positive direction of the horizontal axis represents response, red represents high-level expression, and blue represents low-level expression.

[0033] Figure 6SBM correlation diagram for ranking the importance of features for different inputs. Raw represents cell subpopulation-gene expression level data, Ratio represents cell subpopulation ratio data, '+' represents data concatenation, '*' represents multiplying the gene expression data under the cell subpopulation with the corresponding cell subpopulation ratio (i.e. Raw*Ratio), I represents the same calculation method after changing the random seed, and Random represents multiple comparisons with the random sequence to take the average value.

[0034] Figure 7 SBM correlation table for ranking the importance of features obtained by different methods. RF represents random forest model, SVM represents support vector machine, DL represents deep learning model, L represents LIME, S represents Shapley Value, and Random represents taking the average value after multiple comparisons with random sequences.

[0035] Figure 8 Comparison of the prediction performance of different models when the gene expression in cell subpopulations is used as input. RF stands for random forest, SVM stands for support vector machine, and DL stands for deep learning model. ACC stands for accuracy, i.e. correctly predicted samples / total number of samples; AUC stands for the area under the ROC curve. ACC and AUC are the average performance of the model after training on randomly divided data sets (100 random divisions).

[0036] Fig. 9 CD8+Tem. NKG7 、ISG15+ Mph. CXCR4 and CD8+ MAIT. CD27 The difference in gene expression between responders and non-responders, the difference in expression of each cell subset in this dataset and the possible mechanism of action. A is CD8+Tem. NKG7 The difference in gene expression between responders and non-responders; B is ISG15+ Mph. CXCR4 The difference in gene expression between responders and non-responders; C is CD8+ MAIT. CD27 The difference in gene expression between responders and non-responders; D is CD8+Tem. NKG7 Gene expression differences among cell subpopulations in this dataset; E is ISG15+ Mph. CXCR4 Gene expression differences among cell subsets in this dataset; F is CD8+ MAIT. CD27 Gene expression differences among cell subsets in this dataset; G is CD8+Tem. NKG7 Possible mechanism of action of the gene; H is ISG15+ Mph. CXCR4 Possible mechanism of action of the gene; I is CD8+ MAIT. CD27 Possible mechanism of action of the gene. DETAILED DESCRIPTION

[0037] The present invention is further described in detail below in conjunction with specific embodiments, and the examples provided are only for illustrating the present invention, rather than for limiting the scope of the present invention. The examples provided below can be used as a guide for further improvements by those of ordinary skill in the art, and do not constitute a limitation of the present invention in any way.

[0038] The experimental methods in the following examples, unless otherwise specified, are all conventional methods, and are performed according to the techniques or conditions described in the literature in the field or according to the product instructions. The materials, reagents, etc. used in the following examples, unless otherwise specified, can all be obtained from commercial channels.

[0039] The following examples used Python 3 software to process the data, and the experimental results were expressed as mean ± standard deviation. The Mann-Whitney test was used, and P < 0.05 (*) indicated a significant difference, P < 0.01 (**) indicated a very significant difference, and P < 0.001 (***) indicated an extremely significant difference.

[0040] The data used in the embodiments of the present invention are 2.42 million single-cell RNA-Seq sequencing data of 550 patients with 13 cancer types under a pan-cancer immune microenvironment (public data collected from the GEO database, the PMIDs of the relevant literature from which the data came are: 27124452; 30250229; 30388455; 30388456; 31227543; 31359002; 31588021; 32405063; 32497499; 32949350; 33711272; 33723257; 33861994; 33958794; 34290408; 34653364; 34653365; 34836966; 35108529; 35121991 and 36719749).

[0041] Example 1. Construction of a prediction model for immunotherapy responsiveness.

[0042] The present invention uses random forest algorithm (Random Forest), linear support vector machine (SVM), and deep neural network (DL) to establish a prediction model for the patient's responsiveness to immunotherapy. Specifically, the model takes features as input and whether the patient responds as output. Among them, the features refer to the gene expression levels under each cell subpopulation of the patient.

[0043] After the training, the random forest algorithm can directly generate a set of feature importance values; for the linear support vector machine, the present invention uses the absolute value of its weight as the feature importance value; for the deep neural network, the present invention uses the Shapley Value method to calculate its feature importance ( Figure 7 However, in order to more intuitively show the impact of feature changes on responsiveness prediction and screen the features used in the final model, the present invention also uses Shapley Value or LIME method to display the prediction model based on the random forest algorithm ( Figure 7 The feature importance values ​​of RF+S or RF+L in the linear support vector machine and the prediction model based on the linear support vector machine use the Shapley Value or LIME method ( Figure 7 SVM+S or SVM+L) is used to display its feature importance value. By sorting the feature importance values, features with high feature importance values ​​are selected as the final input features used in model training.

[0044] Among them, the Shapley Value and LIME methods reflect the importance of features by measuring the impact of a certain set of feature changes in the sample on the model output. The feature importance values ​​directly obtained in this way are at the individual level and have positive and negative values. In order to obtain the importance value at the data set level, the present invention takes the absolute value of the importance of all individual units and sums them up to obtain the importance ranking of the features.

[0045] 1. Model input feature screening.

[0046] Since the number of gene combinations under cell subpopulations in the sequencing data (cell subpopulation-gene, i.e., cell subpopulation and a specific gene of the cell subpopulation) is as high as 3.8 × 10 6 The number of samples is much smaller than this level. Directly using the original data for model training will greatly reduce the efficiency of the algorithm and may cause the model to overfit on a few features. Therefore, the present invention first uses the Mann-Whitney test to screen out 1.3 × 10 5 The genes in the cell subpopulations with significant differences in expression between patients who respond to immunotherapy and those who do not respond to immunotherapy are used to obtain differentially expressed cell subpopulation-gene combinations, and then the following method is used to cyclically screen the features used in the final model. In addition, since Shapley Value and LIME are based on feature perturbations, the high dimensionality greatly increases their computational complexity. In order to avoid these negative effects, the present invention needs to pre-screen the features to reduce the feature dimension used for training.

[0047] First, the present invention uses the Mann-Whitney U test to screen cell subpopulation-gene combinations with significant differences in expression levels between patients who respond to immunotherapy and those who do not respond to immunotherapy, and uses each cell subpopulation-gene combination and the expression level of the genes in the combination (the expression level of the genes under the cell subpopulation) as the model pre-input features, and obtains the initialized average importance value of the pre-input features through random initialization).

[0048] Afterwards, the present invention cyclically samples some pre-input features as input to train the random forest model, and at the same time, updates the average importance value of the corresponding features according to the obtained model. Finally, the features used for model training are sorted according to the importance value ( Figure 2 ).

[0049] Feature screening is divided into multiple stages, and the number of sampled features decreases with each stage, and each stage is cycled 100 times. In each cycle, the probability distribution of each feature is calculated as follows (1), and the features are sampled and trained according to the probability distribution.

[0050] ; In formula (1), For the The average importance value of all features The initial values ​​are all set to 1. After the model is trained, the exponential moving average method is used to update the average importance of each feature based on the following formula (2) to obtain the updated average importance value of each feature: ; In formula (2), For the The average importance value of the features, Whether to select The indicator variable of the characteristic, It is The importance value of the feature in this cycle, α is the attenuation factor.

[0051] Then, based on the updated average importance value of each feature, the updated probability distribution of each feature is calculated, and the features are sampled again to train the model according to the updated probability distribution.

[0052] After multiple rounds of training, the top 200 features (important features) ranked by importance value (from high to low) are selected as the input of the model.

[0053] After cyclic screening, the present invention obtained a set of importance rankings. In subsequent experiments, the present invention used the top 200 features to train the immune response prediction model.

[0054] 2. Use gene expression level data of cell subpopulations to build models.

[0055] By using the important features related to the patient's responsiveness to immunotherapy (gene expression levels in each cell subpopulation) obtained using the feature importance ranking results in step 1 as input and whether the patient responds to immunotherapy as output, a prediction model for the patient's responsiveness to immunotherapy was established using the random forest algorithm, linear support vector machine, and deep neural network respectively; the accuracy of the prediction models constructed based on different methods using the feature importance ranking results was compared.

[0056] The ROC (Receiver Operating Characteristic) curve is a tool used to evaluate the performance of classification models. It plots the True Positive Rate as the vertical axis and the False Positive Rate as the horizontal axis, reflecting the performance of the model at different thresholds. The model trained based on the random forest algorithm using cell subpopulation-gene expression level data has a fuller ROC curve, indicating that its performance is good ( Figure 4 ).

[0057] The predictive performance of the model is shown in Figure 8 Among them, the prediction model based on the Random Forest (RF) algorithm performed best, with an accuracy (ACC) of 0.7922 and an AUC value of 0.8816.

[0058] 3. Compare the performance and importance ranking of different methods.

[0059] 3.1 Comparison of feature importance ranking results obtained by different methods.

[0060] In the process of constructing the immunotherapy response prediction model, the present invention attempts to use a variety of methods to rank the importance of features when screening the model input features to select the top ranked features as model input, including directly outputting the feature importance of the random forest model (used in step 1, Figure 7 RF stands for Random Forest Model), using Shapley Value to explain the importance of input features in the Random Forest model ( Figure 7 RF+S stands for using Shapley Value to traverse all feature combinations and calculate the average marginal contribution of each feature to the model prediction, thereby assigning a relative importance score to each feature to obtain the feature importance value), and using LIME to explain the feature importance of the input in the random forest model ( Figure 7RF+L stands for using LIME to explain the contribution of features to the prediction results), using the weight of a linear support vector machine to represent feature importance (SVM), and using a support vector machine combined with Shapley Value or LIME to represent feature importance ( Figure 7 SVM+S or SVM+L represents), using deep learning model combined with LIME method to represent feature importance ( Figure 7 According to these methods, the present invention can obtain a series of importance rankings, which are based on the relevance of SBM such as Figure 7 shown. Figure 7 The results show that the sequences obtained based on different models are quite different, which may be related to the prediction accuracy of the model itself; among them, RF+S and RF (used in step 1) have the highest similarity value, indicating that the feature rankings obtained by feature importance screening based on RF+S and RF are more similar (that is, the input features obtained by both are more stable) and are more credible to some extent.

[0061] At the same time, the genes finally used are obtained by a process based on the random forest algorithm, which is also related to the large difference in importance ranking. Since step 2 shows that the prediction performance of the prediction model obtained based on the random forest algorithm is the highest (the highest classification accuracy), the present invention finally selects the ranking with higher SBM correlation as the screening basis, that is, the features with the highest feature importance ranking in both the methods based on RF+S (obtaining the second model input feature) and the methods based on RF screening model important features (obtaining the first model input feature) (that is, taking the intersection of the first model input feature and the second model input feature) are selected as the final model input feature ( Figure 3 ) The final immunotherapy response model was constructed based on the random forest algorithm.

[0062] Among them, the features that serve as the final model input features (i.e., the features that rank high in importance from high to low in both the RF+S-based and RF-based methods of screening model important features) were screened as immunotherapy response screening targets.

[0063] 3.2 Analysis of immunotherapy response targets in feature importance ranking results.

[0064] Finally, the present invention screened and obtained CD8+ TEM cell subsets based on step 3.1. NKG7, ISG15+ macrophage subsets CXCR4 and CD8+MAIT cell subsets CD27 These three potential therapeutic targets include CD8+ TEM cells NKG7 and CD8+MAIT cells CD27The expression level in immunotherapy responders was significantly higher than that in non-responders ( Fig. 9 (A and C) ISG15+ macrophages CXCR4 The expression level in non-responders was significantly higher than that in responders ( Fig. 9 Middle B). Fig. 9 D, E and F show the differences in the expression levels of the above three genes in different cell subpopulations in this dataset. If the mean gene expression level is directly used as input, the impact of this difference will be ignored, which shows the advantage of finer screening granularity of this method.

[0065] For CD8+ T cells, NKG7 The expression level of genes in responders was significantly higher than that in non-responders ( Fig. 9 A). There is literature (Wen, T. et al. NKG7 is a T-cell-intrinsic therapeutic target for improving antitumor cytotoxicity and cancer immunotherapy. Cancer immunology research 10, 162-181, doi:10.1158 / 2326-6066.Cir-21-0539 (2022)) pointed out that the expression of natural killer cell granulin-7 (NKG7) was reduced in non-responders to immunotherapy, while NKG7 Knockdown resulted in reduced CD8+ T cell-mediated cytotoxicity and, phenotypically, the lack of small granules in the immune prominence between T cells and tumor cells. NKG7 It may affect the transport and release of cytolytic granules in CD8+ T cells by promoting Ca2+ release, thereby affecting their cytotoxicity ( Fig. 9 Middle G).

[0066] and CD27 It plays a key role in T cell activation. The binding of CD27 and CD70 can enhance the survival, proliferation and differentiation of T cells into effector T cells and memory T cells (Starzer, AM&Berghoff, AS New emerging targets in cancer immunotherapy: CD27 (TNFRSF7). ESMO open 4,e000629, doi:10.1136 / esmoopen-2019-000629 (2020)) ( Fig. 9Middle I) (Borst, J.,Hendriks, J.&Xiao, Y. CD27 and CD70 in T cell and B cell activation. Current Opinion in Immunology 17, 275-281, doi: https: / / doi.org / 10.1016 / j.coi.2005.04.004 (2005); Denoeud, J. & Moser, M. Role of CD27 / CD70 pathway of activation in immunity and tolerance. Journal of leukocyte biology 89, 195-203, doi:10.1189 / jlb.0610351 (2011)), the responders CD27 High expression level ( Fig. 9 Middle C) may be related to this.

[0067] For macrophages, CXCR4 The expression level of genes in responders was significantly lower than that in non-responders ( Fig. 9 B). There is literature (Jaramillo-Valverde, L. et al. CXCR4 knockdown via CRISPR / CAS9 in a tumor-associated macrophage model decreases human breast cancer cellmigration. Cureus 13, e20842, doi:10.7759 / cureus.20842 (2021)) pointed out that knocking down CXCR4 Genes that negatively affect the activation of oncogenic pathways MAPK and PI3K in breast cancer cells and reduce the migration ability of breast cancer cells ( Fig. 9 There are also reports that breast cancer cell lines CXCR4 Gene silencing inhibits tumor growth and increases sensitivity to cisplatin chemotherapy. CXCR4 The higher expression level may be related to this.

[0068] 4. The proportion of cell subsets is not the determining factor for the top ranking of genes in some subsets.

[0069] The present invention found that the gene importance rankings in some cell subpopulations are generally high, such as TE07-CD8-Temra-CXCR1, M13-Mph-ISG15, etc. Figure 5 The results showed that in the importance ranking of all cell subsets-gene expression levels (quantities), the gene importance rankings of some subsets were generally at the top: for example, in the CD8+ Temra cell subset, if the expression levels of PRSS23, GNLY, and ZEB2 genes in the CD8+ Temra cell subset were high, the model would tend to predict no response; if the expression levels of PRSS23, GNLY, and ZEB2 genes in the ISG15+ macrophage subset were high, the model would tend to predict no response. CXCR4 High gene expression tends to bias the model towards predicting a response.

[0070] In order to study whether the top ranking of genes in these cell subpopulations is caused by the cell subpopulation ratio, the present invention multiplies the expression amount of genes in the cell subpopulation by the cell subpopulation ratio data ( Figure 7 Indicated by "*"), splicing ( Figure 7 The “+” in the figure indicates the importance of the experiment. The SBM correlation is shown in Figure 6 .

[0071] The cell subpopulation-gene expression amount ( Figure 6 Raw*Ratio represents () which is equivalent to the sum of gene expression levels in a specific cell subset in an individual divided by the number of cells in the cell subset. It actually reflects the gene expression levels at the cell subset level, and to some extent introduces the influence of the cell subset proportion, because the number of gene expression counts in the cell subset with a higher proportion will also be higher. Therefore, the differences in gene expression levels of cell subsets with a larger difference in proportion between immunotherapy responders and non-responders are likely to increase, making the model prediction more focused on these characteristics.

[0072] Due to the calculation method of SBM, the SBM mean of a large number of random sequences and a certain ranking converge to 0.5. The SBM values ​​of the two groups of importance rankings and Raw data obtained using Raw*Ratio (multiplying the expression of genes in cell subsets by the proportion of corresponding cell subsets) and Raw*Ratio+Ratio (multiplying the expression of genes in cell subsets by the proportion of corresponding cell subsets and then splicing with the expression data of genes in cell subsets) are close to 0.5, indicating a large difference; and the SBM values ​​between the sequences obtained from the two groups of data, Raw and Raw+Ratio (splicing the expression of genes in cell subsets by the proportion of corresponding cell subsets), and the SBM values ​​between the sequences obtained from the two groups of data, Raw*Ratio and Raw*Ratio+Ratio, are close to the SBM values ​​of the sequences obtained after changing the random seed and the original sequence, indicating that their sequences are relatively close, reflecting to some extent that the high ranking of the importance of the cell subset-gene combination is not caused by the proportion of cell subsets.

[0073] The present invention has been described in detail above. For those skilled in the art, without departing from the purpose and scope of the present invention, and without the need to carry out unnecessary experimental conditions, the present invention can be implemented in a wide range under equivalent parameters, concentrations and conditions. Although the present invention provides specific embodiments, it should be understood that the present invention can be further improved. In a word, according to the principles of the present invention, the application is intended to include any changes, uses or improvements to the present invention, including departure from the disclosed scope in the application, and changes made with conventional techniques known in the art.

Claims

1. A data processing device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the following steps: S1) Data reception: receiving the expression level M of genes in cell subpopulations from single-cell sequencing data of subjects who have responded to immunotherapy and the expression level N of genes in cell subpopulations from single-cell sequencing data of subjects who have not responded to immunotherapy; S2) obtaining a first model input feature and a second model input feature: taking the expression amount of a gene in a cell subpopulation with a difference in the expression amount M and the expression amount N as a pre-input feature, taking the immunotherapy response result as an output, cyclically sampling the pre-input feature as an input, using a machine learning algorithm to train an immunotherapy response model, obtaining an importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance values ​​P1 from high to low, and selecting the pre-input feature with a higher ranking to obtain the first model input feature; Using Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input feature to the immunotherapy response model, to obtain the importance value P2 of the pre-input feature, to sort the importance values ​​P2 from high to low, and to select the pre-input feature with the highest sorting to obtain the second model input feature; S3) Obtaining immunotherapy targets: taking the intersection of the first model input features and the second model input features to obtain immunotherapy related targets.

2. The data processing device according to claim 1, characterized in that: The machine learning algorithm is random forest.

3. The data processing device according to claim 1 or 2, characterized in that: The immunotherapy-responsive subjects and the immunotherapy-unresponsive subjects are tumor patients.

4. A device for predicting or assisting in predicting immunotherapy targets, characterized in that: The device comprises the following modules: A1) Data receiving module: used to receive the expression level M of genes in cell subpopulations in the single-cell sequencing data of subjects who have responded to immunotherapy and the expression level N of genes in cell subpopulations in the single-cell sequencing data of subjects who have not responded to immunotherapy; A2) Obtaining a first model input feature and a second model input feature module: taking the expression amount of the gene in the cell subpopulation with a difference in the expression amount M and the expression amount N as a pre-input feature, taking the immunotherapy response result as an output, cyclically sampling the pre-input feature as an input, using a machine learning algorithm to train an immunotherapy response model, obtaining an importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance values ​​P1 from high to low, and selecting the pre-input feature with a higher ranking to obtain the first model input feature; Using Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input feature to the immunotherapy response model, to obtain the importance value P2 of the pre-input feature, to sort the importance values ​​P2 from high to low, and to select the pre-input feature with the highest sorting to obtain the second model input feature; A3) Immunotherapy target acquisition module: used to take the intersection of the first model input feature and the second model input feature to obtain immunotherapy related targets.

5. The device according to claim 4, characterized in that: The machine learning algorithm is random forest.

6. A method for predicting or assisting in predicting immunotherapy targets, characterized in that: The method comprises the following steps: B1) Data reception: receiving the gene expression levels M of cell subpopulations in the single-cell sequencing data of subjects who have responded to immunotherapy and the gene expression levels N of cell subpopulations in the single-cell sequencing data of subjects who have not responded to immunotherapy; B2) Obtaining the first model input feature and the second model input feature: taking the expression amount of the gene in the cell subpopulation with different expression amounts M and N as the pre-input feature, taking the immunotherapy response result as the output, cyclically sampling the pre-input feature as the input, using the machine learning algorithm to train the immunotherapy response model, obtaining the importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance value P1 from high to low, and selecting the pre-input feature with the highest sorting to obtain the first model input feature; Using Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input feature to the immunotherapy response model, to obtain the importance value P2 of the pre-input feature, to sort the importance values ​​P2 from high to low, and to select the pre-input feature with the highest sorting to obtain the second model input feature; B3) Obtaining immunotherapy targets: taking the intersection of the first model input features and the second model input features to obtain immunotherapy-related targets.

7. The method according to claim 6, characterized in that: The machine learning algorithm is random forest.

8. A computer-readable storage medium storing a computer program, wherein the computer program causes a computer to execute the following steps: C1) Data reception: receiving the gene expression levels M of cell subpopulations in the single-cell sequencing data of subjects who have responded to immunotherapy and the gene expression levels N of cell subpopulations in the single-cell sequencing data of subjects who have not responded to immunotherapy; C2) Obtaining the first model input feature and the second model input feature: taking the expression amount of the gene in the cell subpopulation with different expression amounts M and N as the pre-input feature, taking the immunotherapy response result as the output, cyclically sampling the pre-input feature as the input, using the machine learning algorithm to train the immunotherapy response model, obtaining the importance value P1 of the pre-input feature according to the immunotherapy response model, sorting the importance value P1 from high to low, and selecting the pre-input feature with the highest sorting to obtain the first model input feature; Using Shapley Value or LIME method to calculate the average marginal contribution rate of the pre-input feature to the immunotherapy response model, to obtain the importance value P2 of the pre-input feature, to sort the importance values ​​P2 from high to low, and to select the pre-input feature with the highest sorting to obtain the second model input feature; C3) Obtaining immunotherapy targets: taking the intersection of the first model input features and the second model input features to obtain immunotherapy-related targets.

9. Any of the following applications of the data processing device according to any of claims 1 to 3: E1) Application in the development and / or manufacture of products for disease patient stratification; E2) Use in the development and / or preparation of immunotherapy products.

10. Any of the following applications of the apparatus according to claim 4 or 5 and / or the computer-readable storage medium according to claim 8: F1) Application in the development and / or manufacture of products for disease patient stratification; F2) Use in the development and / or manufacture of immunotherapeutic products.

Images