A method for predicting cell killing ability based on amplified natural killer cell transcriptome sequencing data

Abstract

The application discloses a kind of based on amplification natural killer cell transcriptome sequencing data cell killing capacity prediction method, belong to the field of cell engineering, comprising: by obtaining the gene expression profile of multiple in vitro amplification NK cells and corresponding killing capacity measured value, utilize elastic network regression algorithm to construct prediction model.Gene expression matrix is used as input feature when training, and killing capacity value is used as target variable, and the regularization parameter is determined by cross validation.During the measurement of sample, only need to carry out transcriptome sequencing, gene expression data obtained are input into model, and killing capacity prediction value can be output, and then the classification screening of high killing activity sample is realized.The present application does not need to rely on traditional killing function experiment, and can realize rapid, non-destructive, standardized killing activity prediction in the scene of cell therapy product quality control and donor screening.

Description

Technical Field

[0001] This invention belongs to the field of cell engineering, and in particular relates to a method for predicting cell killing ability based on amplified natural killer cell transcriptome sequencing data. Background Technology

[0002] Natural killer (NK) cells are important effector cells of the body's innate immune system, capable of directly recognizing and killing tumor cells and virus-infected cells. In recent years, with the development of NK cell immunotherapy, obtaining high-quantity, highly active NK cells through in vitro expansion has become an important technical route for clinical applications and industrial production. Because NK cells from different donor sources still exhibit significant individual differences after expansion, their cytotoxic activity and therapeutic potential are not consistent. Therefore, accurately evaluating the cytotoxic capacity of expanded NK cells is a crucial step in cell product quality control and donor selection. Currently, the evaluation of the cytotoxic capacity of expanded NK cells mainly relies on in vitro cytotoxicity experiments, including lactate dehydrogenase release assays, flow cytometry detection, or luminescent labeling target cell killing assays. These methods typically require co-culturing expanded NK cells with standardized target cells at a specific effector-to-target ratio, and evaluating the cytotoxic activity of NK cells by detecting target cell mortality or survival rates. Meanwhile, high-throughput transcriptome sequencing technology has been widely used in the field of NK cell research. Current research mainly uses transcriptome data to analyze the activation status, differentiation characteristics, metabolic status and immune regulation mechanism of NK cells, and to screen biomarkers related to cell function.

[0003] However, existing methods for evaluating the cytotoxic capacity of expanded NK cells still have the following shortcomings: First, existing evaluation methods require the establishment of an additional co-culture system of NK cells and target cells. The detection process depends on various factors such as the state of target cells, the effector-target ratio, culture conditions, and detection platform. Differences in results can easily arise between different laboratories or different batches of experiments, limiting the standardization and reproducibility of the evaluation results. Second, functional detection usually requires a certain number of expanded NK cells and target cell resources. When comparing and screening a large number of expanded NK cells from donor sources, the workload increases significantly, and the detection cost is high, which is not conducive to large-scale application. Third, existing technologies can only obtain cytotoxic capacity results for samples that have already been tested, lacking technical means to predict and anticipate cytotoxic activity based on intrinsic molecular characteristics of cells. The main reason for the above problems is that a stable and reliable quantitative correlation model between the gene expression characteristics of amplified NK cells and their cytotoxic ability has not yet been established. Although transcriptome sequencing can comprehensively reflect the functional state of cells, transcriptome data has characteristics such as high dimensionality, high redundancy, and strong collinearity. Traditional statistical analysis methods are difficult to use to screen out key features closely related to cytotoxic ability from massive gene expression information and construct predictive models with good generalization ability. Therefore, there is an urgent need to develop an evaluation method that can accurately predict the cytotoxic ability of amplified NK cells using their intrinsic molecular characteristics. This method would enable standardized evaluation and efficient screening of the cytotoxic activity of amplified NK cells without relying on traditional cell cytotoxicity experiments, thus meeting the application needs of cell therapy product quality control and donor screening. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method for predicting cell killing ability based on amplified natural killer cell transcriptome sequencing data, comprising: Gene expression data of multiple in vitro expanded natural killer cell samples were obtained, as well as the measured killing ability of each sample. Using gene expression data of each sample as input features and measured killing ability values ​​corresponding to each sample as target variables, an elastic network regression algorithm is used to train a cell killing ability prediction model. Gene expression data of the natural killer cell sample to be tested are obtained and input into the prediction model, and the predicted killing ability value of the sample to be tested is output.

[0005] Optionally, it also includes the step of constructing a training dataset: Each in vitro expanded human natural killer cell sample was divided into a first part and a second part. The first part was subjected to transcriptome sequencing to obtain the gene expression data; the second part was subjected to a killing function experiment to determine the measured value of the killing ability.

[0006] Optionally, the acquisition of the gene expression data includes: Total RNA was extracted from amplified natural killer cells and subjected to transcriptome sequencing. The sequenced reads were aligned to a reference genome, the read counts of each gene were counted, the expression level per million transcripts was calculated, and the expression level per million transcripts was transformed by log2(TPM+1) to obtain the gene expression matrix.

[0007] Optionally, the acquisition of the gene expression data further includes: Genes with TPM > 1 in at least 20% of the samples were filtered out, and the filtered TPM expression levels were transformed by log2(TPM+1) to obtain the gene expression matrix.

[0008] Optionally, the cell killing capability prediction model trained using the elastic network regression algorithm includes: The gene expression matrix of each sample is used as the feature variable, and the measured killing ability of each sample is used as the target variable. The results are then input into the glmnet algorithm package. Set the hybrid parameters of the elastic network, use cross-validation to select the regularization parameter that minimizes the mean square error, and use the selected regularization parameter value to fit the final model. The regression coefficients of each gene are extracted to obtain the prediction model.

[0009] Optionally, the determination of the measured value of the killing ability includes: using the expanded natural killer cells as effector cells, co-culturing them with labeled target cells at an effector-to-target ratio, detecting the target cell mortality rate, and calculating the specific killing rate as the measured value of the killing ability.

[0010] Optionally, the predicted killing ability value is a continuous value used to classify the killing activity of amplified natural killer cell samples. This includes using the median of the measured killing ability values ​​of each sample in the independent test set as a threshold, classifying samples with predicted values ​​higher than the threshold as high killing activity samples, and classifying samples with predicted values ​​lower than or equal to the threshold as low killing activity samples.

[0011] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0012] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0013] Compared with the prior art, the present invention has the following advantages and technical effects: This invention utilizes the paired training of a resilient network model with the transcriptome gene expression profiles of in vitro expanded NK cells and measured killing capacity values. The principle lies in the fact that resilient network regularization can effectively handle high-dimensional, strongly collinear transcriptome data, automatically selecting key genomes with the highest predictive value for killing capacity and suppressing overfitting, thereby establishing a highly generalizable predictive model. This method allows for the acquisition of predicted killing capacity values ​​for the tested expanded NK cells solely through transcriptome sequencing, eliminating the need for separate, time-consuming killing function experiments. It fundamentally avoids batch-to-batch variations caused by reliance on target cell status, effector-target ratio, and culture conditions in traditional methods, significantly improving the standardization and reproducibility of evaluation results. Meanwhile, the prediction process does not consume resources of expanded NK cells and target cells, significantly reducing the detection cost and workload during large-scale donor screening. More importantly, this method enables early prediction of cytotoxic activity based on intrinsic molecular characteristics of cells, allowing donor activity assessment to be completed early in cell product preparation. Cross-validation with 73 training samples and validation with 27 independent test sets showed a correlation of 0.73 between the model's predicted values ​​and the actual cytotoxic capacity values, and an AUC of 0.91 for identifying high-cytotoxic samples. This indicates that the method can accurately and reliably achieve high-throughput prediction of the cytotoxic capacity of expanded NK cells, effectively solving the technical challenge of existing technologies that rely on functional experiments and cannot predict in advance. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram showing the ranking of NK kills after the training set samples were amplified according to an embodiment of the present invention. Figure 3 The diagram shows the correlation analysis of the prediction model in this embodiment of the invention, where (A) is the training set; (B) is the test set; and (C) is the subject operating curve. Detailed Implementation

[0015] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0017] Example 1 like Figure 1 As shown, this embodiment provides a method for predicting cell killing ability based on amplified natural killer cell transcriptome sequencing data, including: Gene expression data of multiple in vitro expanded natural killer cell samples were obtained, as well as the measured killing ability of each sample. Using gene expression data of each sample as input features and measured killing ability values ​​corresponding to each sample as target variables, an elastic network regression algorithm is used to train a cell killing ability prediction model. Gene expression data of the natural killer cell sample to be tested are obtained and input into the prediction model, and the predicted killing ability value of the sample to be tested is output.

[0018] The specific process includes: Peripheral blood was collected from 73 healthy donors, and peripheral blood mononuclear cells (NK cells) were separated by Ficoll density gradient centrifugation. Irradiated genetically engineered K562 cells were used as feeder cells. NK cells were mixed with peripheral blood mononuclear cells at a 1:1 ratio and cultured in complete medium containing 5%–10% fetal bovine serum and 500 IU / mL recombinant human IL-2. Cells were counted and replenished with complete medium every other day. Feeder cells were added on day 7 for secondary activation induction, and all suspension cells were harvested on day 14. All suspension cells were collected, and flow cytometry analysis showed that the purity of CD3⁻CD56⁺ NK cells was >80%.

[0019] Take 1×10⁻⁶ of each expanded NK cell. 5 Human erythroleukemia K562 cells, labeled with GFP fluorescence, were used as effector cells and seeded in 96-well plates at an effector-to-target ratio of 1:4. A negative control containing only target cells and a control treating with 1% Triton X-100 to maximize target cell release were also included. The cells were incubated at 37°C and 5%... After co-culturing for 6 hours, 7-AAD / PI staining was added, and the target cell death rate was determined by flow cytometry. Specific killing rate (%) = (experimental group mortality rate - spontaneous mortality rate) / (maximum mortality rate - spontaneous mortality rate) × 100%, and this killing rate is taken as the true killing ability value of the sample.

[0020] Take another 5 × 10⁵ NK cells from each sample for expansion. 5Total RNA was extracted from samples using the TRIzol method, and RNA integrity was assessed using an Agilent 2100 Bioanalyzer. Sequencing libraries were prepared using the Illumina TruSeq chain-specific mRNA library construction kit, and paired-end 150 bp sequencing was performed on a NovaSeq 6000 sequencing platform. The sequencing data were aligned to the human reference genome GRCh38 using STAR software. Read counts for each gene were calculated using feature counts, and TPM expression levels were then calculated. Genes with TPM > 1 were filtered out and retained in at least 20% of the samples. The TPM values ​​were then transformed using log2(TPM+1) to obtain the final whole transcriptome gene expression matrix.

[0021] like Figure 2 and Figure 3 As shown, the gene expression matrix of the above 73 samples was used as the feature variable, and the corresponding killing ability value of each sample was used as the target variable. These values ​​were input into the glmnet package in R software. The elastic network mixing parameter α was set to 0.5, the response variable family was set to "gaussian", and 10-fold cross-validation was used to select the regularization parameter λ that minimizes the mean squared error. The optimal λ value was used to fit the final model, and the regression coefficients of each gene were extracted to obtain a prediction model for killing ability based on gene expression profiles.

[0022] Peripheral blood was collected from 27 additional healthy donors. Following the amplification system, cytotoxicity assay, and transcriptome sequencing procedures described above, independent gene expression matrices and actual cytotoxicity values ​​were obtained. The expression matrices were input into the prediction model, which output the predicted cytotoxicity value for each sample. The Pearson correlation coefficient between the predicted and measured values ​​was calculated. Using the median of the measured cytotoxicity values ​​from all 27 samples as a threshold, the samples were divided into high-cytotoxicity and low-cytotoxicity groups. ROC curves were plotted using the predicted values ​​to evaluate the model's ability to identify samples with high cytotoxicity.

[0023] For any donor NK cell sample obtained through in vitro amplification, only RNA needs to be extracted for transcriptome sequencing to obtain the gene expression matrix. This expression matrix is ​​then input into a pre-trained elastic network prediction model to directly output the predicted killing ability of the sample. The entire process does not require consuming cells for killing function experiments. In cell therapy product quality control and high-throughput donor screening scenarios, it can rapidly and non-destructively achieve graded prediction of the killing activity of amplified NK cells, prioritizing samples predicted to have high killing activity for subsequent immunotherapy.

[0024] It is important to note that this method detects natural killer cells that have been expanded and cultured in vitro, not blood or tissue samples directly collected from the human body for pathological analysis. The entire procedure is performed in vitro (in culture dishes or multi-well plates) and does not involve any interventional treatment on the human or animal body.

[0025] The "predicted lethality value" output by the model is an evaluation of the quality / potency of the cell product (i.e., the activity level of the batch of cells), and belongs to the quality control parameters of cell products. This result itself does not directly determine whether the subject has a certain disease, nor can it directly guide the clinical prescribing of specific treatment plans.

[0026] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0027] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0028] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting cell killing ability based on amplified natural killer cell transcriptome sequencing data, characterized in that, include: Gene expression data of multiple in vitro expanded natural killer cell samples were obtained, as well as the measured killing ability of each sample. Using gene expression data of each sample as input features and measured killing ability values ​​corresponding to each sample as target variables, an elastic network regression algorithm is used to train a cell killing ability prediction model. Gene expression data of the natural killer cell sample to be tested are obtained and input into the prediction model, and the predicted killing ability value of the sample to be tested is output.

2. The method according to claim 1, characterized in that, It also includes the step of building the training dataset: Each in vitro expanded human natural killer cell sample was divided into a first part and a second part. The first part was subjected to transcriptome sequencing to obtain the gene expression data; the second part was subjected to a killing function experiment to determine the measured value of the killing ability.

3. The method according to claim 1, characterized in that, The acquisition of the gene expression data includes: Total RNA was extracted from amplified natural killer cells and subjected to transcriptome sequencing. The sequenced reads were aligned to a reference genome, the read counts of each gene were counted, the expression level per million transcripts was calculated, and the expression level per million transcripts was transformed by log2(TPM+1) to obtain the gene expression matrix.

4. The method according to claim 1, characterized in that, The acquisition of the gene expression data also includes: Genes with TPM > 1 in at least 20% of the samples were filtered out, and the filtered TPM expression levels were transformed by log2(TPM+1) to obtain the gene expression matrix.

5. The method according to claim 1, characterized in that, The cell killing capability prediction model trained using the elastic network regression algorithm includes: The gene expression matrix of each sample is used as the feature variable, and the measured killing ability of each sample is used as the target variable. The results are then input into the glmnet algorithm package. Set the hybrid parameters of the elastic network, use cross-validation to select the regularization parameter that minimizes the mean square error, and use the selected regularization parameter value to fit the final model. The regression coefficients of each gene are extracted to obtain the prediction model.

6. The method according to claim 1, characterized in that, The determination of the measured value of the killing ability includes: using the expanded natural killer cells as effector cells, co-culturing them with labeled target cells at an effector-to-target ratio, detecting the target cell mortality rate, and calculating the specific killing rate as the measured value of the killing ability.

7. The method according to claim 1, characterized in that, The predicted killing ability value is a continuous value used to classify the killing activity of amplified natural killer cell samples. It includes using the median of the measured killing ability values ​​of each sample in the independent test set as a threshold, classifying samples with predicted values ​​higher than the threshold as high killing activity samples, and classifying samples with predicted values ​​lower than or equal to the threshold as low killing activity samples.

8. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-7.

Images