Individualized gene therapy target prediction method and device and organoid verification method

Abstract

The invention discloses an individualized gene therapy target prediction method and a related device, and the method comprises the steps: screening a cell line sample set through gene expression data and gene dependence effect data, then selecting N genes from the co-occurrence frequency of a plurality of genes with the lowest gene effect value in each cell line sample as target genes, and carrying out the prediction of the individualized gene therapy target. The target gene is a gene having important influence on cell growth inhibition, then a data set is constructed by taking the screened target gene as a label, and a regression model is trained to obtain a prediction model, so that the prediction model has the capability of predicting a gene target spot effective for treatment; finally, tumor transcriptome sequencing data of a patient are input into the prediction model, an output result serves as an individualized target gene prediction result, individualized treatment is conducted on the patient through the predicted target gene, and compared with related technologies, the screening efficiency of individualized treatment gene targets is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of biomedical technology, in particular to an individualized gene therapy target prediction method and related device, and an organoid verification method. BACKGROUND

[0002] Malignant tumors seriously threaten human health. Drug treatment of malignant tumors is in a historical turning period from chemical drugs, targeted drugs, immunotherapy drugs to cell therapy and gene therapy. The advantage of tumor gene therapy is particularly prominent. How to systematically select treatment targets for individualized gene therapy is an urgent scientific problem to be solved.

[0003] With the rapid development of molecular biology and bioinformatics, target gene therapy has become one of the important fields of cancer treatment, and has made significant progress in some specific types of cancer. Some targeted drugs have been developed and applied in clinical practice. These drugs can intervene in specific cancer-related genes to inhibit the growth and spread of cancer cells. Target gene therapy can not only be used as an independent treatment method, but also can be combined with traditional chemotherapy, radiotherapy and other methods to improve the overall treatment effect.

[0004] However, due to individual differences, target gene therapy faces some challenges. SUMMARY

[0005] Therefore, the present application provides an individualized gene therapy target prediction method and related device, which trains a prediction model by constructing a data set, realizes the prediction of individualized gene therapy targets, and effectively improves the screening efficiency of individualized treatment gene targets.

[0006] To solve the above problems, the technical scheme provided by the present application is as follows:

[0007] In one aspect, the present application provides an individualized gene therapy target prediction method, which comprises:

[0008] Obtaining gene expression data and gene dependence effect data;

[0009] Screening a cell line sample set based on the gene expression data and the gene dependence effect data;

[0010] Based on the co-occurrence frequency of a plurality of genes with the lowest gene effect value ranking in each cell line sample of the cell line sample set, N genes in the plurality of genes are selected as target genes;

[0011] Training a regression model based on the cell line sample set and the target genes to obtain a prediction model;

[0012] obtain an individualized gene therapy target point of the patient based on tumor transcriptome sequencing data of the patient and the prediction model.

[0013] In a possible implementation, the method further includes:

[0014] ranking genes corresponding to each cell line sample in the set of cell line samples according to gene effect values;

[0015] determining co-occurrence frequencies of a plurality of genes with lowest gene effect values in the set of cell line samples;

[0016] ranking the plurality of genes according to the co-occurrence frequencies in descending order, and taking genes ranked in top N positions as target genes.

[0017] In a possible implementation, the N>1, and the method further includes:

[0018] constructing the regression model with N outputs by using an XGBoost algorithm.

[0019] In a possible implementation, the training of the regression model based on the set of cell line samples and the target genes to obtain a prediction model includes:

[0020] inputting the cell line samples into the regression model to obtain output results;

[0021] evaluating differences between the output results and the target genes by using mean square errors;

[0022] optimizing parameters of the regression model based on the differences until an iteration number of the regression model reaches a preset threshold or the differences reach a preset standard, to obtain a prediction model.

[0023] In a possible implementation, after the training of the regression model based on the set of cell line samples and the target genes to obtain a prediction model, the method further includes:

[0024] evaluating the prediction model by using a hit probability, the hit probability referring to a proportion of cell line samples in which a number of predicted genes with lowest predicted gene effect values in each cell line sample in a plurality of cell line samples contains at least K genes in top M genes with lowest corresponding gene effect values, the K being greater than or equal to 1.

[0025] In another aspect, an embodiment of the present application provides a device for predicting an individualized gene therapy target point, the device comprising:

[0026] an acquisition unit, configured to acquire gene expression data and gene-dependent effect data;

[0027] a screening unit, configured to screen a cell line sample set based on the gene expression data and the gene-dependent effect data;

[0028] a determination unit, configured to determine N target genes from a plurality of genes with the lowest gene effect value ranking in each cell line sample in the cell line sample set based on a co-occurrence frequency of the plurality of genes;

[0029] a training unit, configured to train a regression model based on the cell line sample set and the target genes, to obtain a prediction model;

[0030] a prediction unit, configured to obtain an individualized gene therapy target point of a patient based on tumor transcriptome sequencing data of the patient and the prediction model.

[0031] In a possible implementation, the determination unit is configured to:

[0032] sort genes corresponding to each cell line sample in the cell line sample set according to gene effect values;

[0033] determine a co-occurrence frequency of a plurality of genes with the lowest gene effect value in the cell line sample set;

[0034] sort the plurality of genes according to the co-occurrence frequency in descending order, and determine the top N genes as target genes.

[0035] In a possible implementation, the N>1, and the apparatus further includes a construction unit configured to:

[0036] construct the regression model with N outputs by using an XGBoost algorithm.

[0037] In a possible implementation, the training unit is configured to:

[0038] input the cell line sample into the regression model to obtain an output result;

[0039] evaluate a difference between the output result and the target genes by using a mean square error;

[0040] optimize parameters of the regression model based on the difference until an iteration number of the regression model reaches a preset threshold or the difference reaches a preset standard, to obtain a prediction model.

[0041] In a possible implementation, the apparatus further includes an evaluation unit configured to evaluate the prediction model using a hit probability, the hit probability indicating a proportion of cell line samples in which a first M genes with the lowest predicted gene effect value in each cell line sample contain at least K genes in a first M genes with the lowest gene effect value, K being greater than or equal to 1.

[0042] In another aspect, the embodiments of the present application provide an organoid verification method for verifying prediction accuracy of the individualized gene therapy target obtained by the method described in the gene control group.

[0043] In another aspect, the embodiments of the present application provide a computer device including a processor and a memory:

[0044] The memory is configured to store a computer program.

[0045] The processor is configured to execute the method according to any one of the above embodiments according to the computer program.

[0046] In another aspect, the embodiments of the present application provide a computer readable storage medium configured to store a computer program, the computer program being configured to implement the method according to any one of the above embodiments when executed by a computer device.

[0047] In another aspect, the embodiments of the present application provide a computer program product including a computer program, the computer program product being configured to cause a computer device to execute the method according to any one of the above embodiments when the computer program product is executed by the computer device.

[0048] As can be seen from the above technical solution, the technical solution screens a cell line sample set by using gene expression data and gene-dependent effect data, then selects N genes from co-occurrence frequencies of a plurality of genes with the lowest gene effect value in each cell line sample as target genes, the target genes being genes that have an important influence on cell growth inhibition, then constructs a data set by using the screened target genes as labels, trains a regression model, obtains a prediction model, so that the prediction model has the ability to predict gene targets effective for treatment, finally inputs tumor transcriptome sequencing data of a patient into the prediction model, outputs a result as an individualized target gene prediction result, so as to perform individualized treatment on the patient by using the predicted target genes, and compared with related technologies, the screening efficiency of individualized treatment gene targets is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0050] Figure 1 A flowchart of an individualized gene therapy target prediction method provided by the embodiments of the present application is shown in the figure.

[0051] Figure 2 A general flowchart of an individualized gene therapy target prediction and organoid verification method provided by the embodiments of the present application is shown in the figure.

[0052] Figure 3 A schematic diagram of the distribution ratio of target genes in cell line samples provided by the embodiments of the present application is shown in the figure.

[0053] Figure 4 A schematic diagram of the data division ratio provided by the embodiments of the present application is shown in the figure.

[0054] Figure 5 A scatter plot of predicted value-true value provided by the embodiments of the present application is shown in the figure.

[0055] Figure 6 A prediction trajectory diagram provided by the embodiments of the present application is shown in the figure.

[0056] Figure 7 A hit probability diagram provided by the embodiments of the present application is shown in the figure.

[0057] Figure 8 A heat map of TOP3 individualized target gene prediction results provided by the embodiments of the present application is shown in the figure.

[0058] Figure 9 A volcano plot of tumor transcriptome sequencing data differential gene screening provided by the embodiments of the present application is shown in the figure.

[0059] Figure 10 A TOP3 gene column chart of tumor transcriptome sequencing data differential gene screening provided by the embodiments of the present application is shown in the figure.

[0060] Figure 11 An RT-qPCR and Western blot result statistical diagram of verifying and screening target siRNA in 293T cell lines provided by the embodiments of the present application is shown in the figure.

[0061] Figure 12 A representative bright field image of three breast cancer patient-derived organoids provided by the embodiments of the present application is shown in the figure.

[0062] Figure 13 A schematic diagram of organoid construction and target gene knockdown provided for the embodiments of the present application;

[0063] Figure 14 A schematic diagram of imaging and quantitative analysis for evaluating the growth of breast cancer organoids after knocking down target genes provided for the embodiments of the present application;

[0064] Figure 15 A schematic diagram of organoid cell viability provided for the embodiments of the present application;

[0065] Figure 16 A schematic diagram of an individualized gene therapy target point prediction device provided for the embodiments of the present application. DETAILED DESCRIPTION

[0066] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0067] As described in the background, obtaining a safe and effective individualized treatment target is the key to tumor gene therapy, and sgRNA library screening is one of the options for obtaining a gene therapy target point. CRISPR sgRNA library screening can obtain genes that are essential for the survival and proliferation of cancer cells, and thus the use of corresponding siRNA for blocking is expected to achieve inhibition and killing of tumors. However, sgRNA library screening targets are time-consuming and require a large amount of cells, which is difficult to meet with primary culture.

[0068] In view of the above problems, the present application provides an individualized gene therapy target point prediction method, which screens a cell line sample set through gene expression data and gene dependence effect data, then selects N genes as target genes from the co-occurrence frequency of a plurality of genes with the lowest gene effect value in each cell line sample, the target genes are genes that have an important impact on cell growth inhibition, then constructs a data set with the selected target genes as labels, trains a regression model, and obtains a prediction model, so that the prediction model has the ability to predict gene targets effective for treatment, finally inputs the tumor transcriptome sequencing data of the patient into the prediction model, and outputs the result as an individualized target gene prediction result, so as to perform personalized treatment on the patient through the predicted target genes. Compared with related technologies, the screening efficiency of individualized treatment gene targets is effectively improved.

[0069] The scheme provided by the embodiments of the present application relates to the biomedical technology field, and is specifically explained through the following embodiments.

[0070] Referring to Figure 1 As shown in the figure, a flowchart of an individualized gene therapy target prediction method provided by the embodiments of the present application.

[0071] S101: Obtain gene expression data and gene-dependent effect data.

[0072] The gene expression data refers to data for measuring the transcriptional activity level of a specific gene in a cell line, reflecting whether the gene is active and the degree of activity.

[0073] The gene-dependent effect data refers to data for measuring the sensitivity of cell survival / proliferation to the absence of a specific gene function, reflecting the necessity and dependence of the gene on the cell.

[0074] For example, the gene expression data (CCLE_RNAseq_rsem_genes_tpm_20180929) and the gene-dependent effect data (CRISPRGeneEffect) are downloaded from the DepMap public database.

[0075] S102: Screen a cell line sample set based on the gene expression data and the gene-dependent effect data.

[0076] The embodiments of the present application do not make specific limitations on the screening method of the cell line, for example, the gene expression data and the gene-dependent effect data can be integrated by a cell line identifier, multiple cell lines with gene expression profiles and CRISPR screening data are formed into a cell line sample set, or multiple cell line samples are further screened to select a specified number of cell lines from cell lines corresponding to gene effect values at different levels, and a cell line sample set is constructed.

[0077] S103: Select N genes from the multiple genes as target genes based on the co-occurrence frequency of the multiple genes with the lowest gene effect value ranking in each cell line sample of the cell line sample set.

[0078] The co-occurrence frequency refers to the number of common occurrences. The target gene refers to a gene that has an important influence on cell growth inhibition, and the target gene exists in each cell line and has a low gene effect value in each cell line. The smaller the gene effect value, the more important the role of the gene in cell inhibition.

[0079] Specifically, based on the gene-dependent effect data, the gene effect values of each cell line sample are sorted according to the comprehensive criteria of modeling error, actual benefit (strong inhibition effect), cell line coverage and hit rate, and then a plurality of genes with low gene effect values in the cell line sample are selected, such as 10 genes, and then the occurrence frequencies of the 10 genes in different cell line samples are counted.

[0080] wherein N is greater than or equal to 1, and the specific value can be determined according to actual conditions, for example, 47 genes commonly occurring in a plurality of cell line samples, and N is 47, the 47 genes are target genes, and a plurality of genes with higher co-occurrence frequencies are selected from the 47 commonly occurring genes, such as 1, 2 or 10 genes, as target genes, and N is 1, 2 or 10 at this time, which is not specifically limited in the embodiments of the present application.

[0081] In one possible implementation, S103 includes:

[0082] A1: Sort the genes corresponding to each cell line sample in the cell line sample set according to the gene effect value.

[0083] A2: Determine the co-occurrence frequencies of a plurality of genes with the lowest gene effect values in the cell line sample set.

[0084] A3: Sort the plurality of genes in descending order of co-occurrence frequency, and select the top N genes as target genes.

[0085] Specifically, in each cell line sample, the genes are sorted according to the gene effect value, a plurality of genes with the lowest effect value are first selected, such as 10 genes, and the 10 genes are the genes that have an important influence on the growth inhibition of the cell line, then the co-occurrence frequencies of the plurality of genes corresponding to each cell line sample (i.e. the number of cell line samples * the number of the plurality of genes with the lowest effect value) in the cell line sample are counted, then the genes are sorted in descending order of co-occurrence frequency, and the top N genes are selected as target genes, for example, if 47 genes appear in each cell line, the gene effect values of these genes in the corresponding cell line sample are taken as the gene effect values of the target genes in the corresponding cell line sample.

[0086] Thus, the target genes are screened in a plurality of cell lines according to the gene effect value, so that the target genes not only exist in a plurality of cell lines, but also have a strong inhibitory effect on the growth of certain cell lines, so that when the target genes are used as a treatment target for treatment, they can play an important role in inhibiting cancer cells in a plurality of cell lines.

[0087] S104: Training a regression model based on the cell line sample set and the target genes to obtain a prediction model.

[0088] wherein the regression model refers to a pre-trained model, and embodiments of the present application do not make specific limitations thereto, for example, it can be a linear regression model (such as OLS, Ridge), a nonlinear regression model (such as a random forest model), or a deep learning model based on time series (LSTM), etc.

[0089] Since the value of N can be one or more, the output of the regression model can also be one or more, and embodiments of the present application do not make specific limitations thereto.

[0090] In one possible implementation, N > 1, and the regression model is constructed by the following way:

[0091] An XGBoost algorithm is used to construct a regression model with N outputs.

[0092] Specifically, the XGBRegressor is wrapped with the MultiOutputRegressor to construct the regression model, so that the regression model supports the multi-output regression task, and the output is N, which corresponds to the number of target genes.

[0093] Therefore, by training the multi-output regression model, the multiple genes that have inhibitory effect on cell growth can be more accurately identified, and the accuracy of the individualized target gene prediction result is effectively improved.

[0094] wherein the prediction model further comprises a missing value processing module, which ensures that the gene expression data has no missing values, and fills the missing values in the label (filled with 0).

[0095] For example, the cell line sample set is taken as input, and the target gene is taken as label, and the regression model is trained, so that the predicted gene effect value of the target gene in the output result corresponding to the cell line sample is continuously close to the gene effect value of the target gene corresponding to the cell line sample.

[0096] Firstly, the dataset is divided, for example, the merged dataset is divided into features (X, gene expression data) and labels (Y, CRISPRGeneEffect values of N target genes); the cell line samples (such as 667 samples) in the cell line sample set are randomly divided into a training set (466 cell lines) and an independent test set (201 cell lines) according to a ratio of 7:3, to ensure the unbiasedness of the model evaluation, and a random seed is used to ensure the repeatability of the results.

[0097] Then the regression model is trained by the training set: the difference between the output result of the regression model and the target gene is evaluated by a loss function to iteratively train the model parameters of the regression model, and when the difference is reduced to a preset condition or the iteration number of the model reaches a preset iteration number, the training is stopped to obtain a prediction model. The model parameters also include a learning rate (learning_rate=0.05), an estimator number (n_estimators=50), a maximum depth (max_depth=10), a minimum child weight (min_child_weight=1), a random seed (seed=0), and a subsample ratio (subsample=1).

[0098] Then the gene effect value prediction results of the N target genes of each cell line sample are obtained by the trained prediction model, including the final prediction results on the training set and the final prediction results on the test set.

[0099] In a possible implementation, S104 includes:

[0100] B1: inputting the cell line sample set into the regression model to obtain an output result;

[0101] B2: evaluating the difference between the output result and the target gene by mean squared error;

[0102] B3: optimizing the parameters of the regression model based on the difference until the iteration number of the regression model reaches a preset threshold or the difference reaches a preset standard to obtain a prediction model.

[0103] Specifically, each cell line sample in the cell line sample set is input into the regression model to obtain an output result, then the difference between the multiple genes output by each cell line sample and the target gene is evaluated by mean squared error (MSE), and then the iteration optimization of the parameters of the regression model is realized by reducing the MSE, until the iteration number of the regression model reaches a preset threshold or the MSE reaches a preset standard, the model training is stopped, and a prediction model is obtained.

[0104] Thus, the difference between the predicted value and the true value is measured by mean squared error to optimize the model, and the prediction parameters of multiple gene targets can be simultaneously optimized in the output regression model, thereby effectively improving the prediction accuracy of the prediction model for individualized gene therapy targets.

[0105] S105: obtaining the individualized gene therapy target of the patient based on the tumor transcriptome sequencing data of the patient and the prediction model.

[0106] wherein the patient can be one or more, for each patient, by collecting tumor tissue and paracancer normal tissue of the patient and performing transcriptome sequencing, obtaining the tumor transcriptome sequencing data of the patient, and then inputting the tumor transcriptome sequencing data into the trained prediction model to predict the individualized gene therapy target point of the patient (e.g., taking the top 3 genes with the highest gene effect values as the treatment target points).

[0107] Thus, by using gene expression data and gene-dependent effect data, a cell line sample set is screened, and then the co-occurrence frequency of a plurality of genes with the lowest gene effect value in each cell line sample is determined, N genes are selected from the N genes as target genes, the target genes are genes that have an important influence on cell growth inhibition, a data set is constructed using the selected target genes as labels, a regression model is trained to obtain a prediction model, so that the prediction model has the ability to predict gene targets effective for treatment, and finally the tumor transcriptome sequencing data of the patient is input into the prediction model, and the output result is used as an individualized target gene prediction result, so that the patient is treated by the predicted target gene, and the screening efficiency of the individualized treatment gene target point is effectively improved compared with related technologies.

[0108] In a possible implementation, after S104, the method further includes:

[0109] The hit probability is used to evaluate the prediction model.

[0110] The hit probability refers to the proportion of cell line samples in which the top M predicted genes with the lowest predicted gene effect value in each cell line sample include at least K genes in the top M genes with the lowest corresponding gene effect value, wherein M and K are both greater than or equal to 1, and M is greater than or equal to K.

[0111] For example, by using the trained prediction model to predict each cell line sample in the test set: the gene effect values of N (e.g., 47) genes are predicted, and the top M (e.g., Top10) genes with the lowest predicted gene effect value (i.e., the strongest predicted inhibition effect) are selected. Then check whether the Top10 prediction list contains the genes with the lowest real gene effect value (i.e., the strongest real inhibition effect) in the cell line sample. By counting the number of intersections between the predicted gene Top10 list and the real gene Top10 list for each cell line sample, the proportion of cell lines in which the predicted gene Top10 list contains at least K genes in the real gene Top10 list is calculated, wherein k = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10; the hit probability directly quantifies the reliability of the top 10 candidate target points recommended by the prediction model for a specific cell line, which contains the real effective target point.

[0112] Thus, the hit probability can be used to evaluate the accuracy of the prediction model in predicting target genes in individual cell lines, so that the prediction model supports the selection of target points for individualized cancer treatment.

[0113] The embodiments of the present application also provide an organoid verification method for verifying the prediction accuracy of individualized gene therapy target points by a gene control group.

[0114] Specifically, bioinformatics analysis is performed on tumor transcriptome sequencing data of a plurality of patients to obtain one or more genes with the largest gene expression difference, and then the gene, the top 3 individualized treatment target genes predicted by the machine learning model, and the top 3 individualized gene target points predicted based on the tumor transcriptome sequencing data of each patient are used as a control group, and the knockdown effect of the genes in the corresponding patient cancer tissue organoids is verified.

[0115] For example, the tumor transcriptome sequencing data of three breast cancer patients is screened by bioinformatics analysis to obtain a differentially expressed gene TOP1.

[0116] Then, cancer tissue organoids derived from the three breast cancer patients are constructed, and the top 3 individualized treatment target genes predicted by the machine learning model, the top1 target gene screened from the differentially expressed genes of the tumor transcriptome sequencing data, and the top 3 individualized treatment target genes of each patient predicted by the tumor transcriptome sequencing data input prediction model are transfected in the three organoids, respectively.

[0117] Then, the knockdown effect of the target genes in the organoids is verified by RT-qPCR and Western blot. The dead and live cell ratio of the organoids after transfection is evaluated by the Calcein-AM / PI detection kit to verify the inhibitory effect of the individualized treatment target genes predicted by the model on the organoids. The cell viability of the organoids after transfection is evaluated by the CellTiter-Glo luminescent cell viability detection kit to verify the inhibitory effect of the individualized treatment target genes predicted by the model on the organoids from multiple angles.

[0118] Further, according to the cell viability in the organoids, the inhibitory effects of the top1 target genes screened by the machine learning model and the tumor transcriptome sequencing data on the organoids are compared to evaluate and compare the effectiveness of the two methods.

[0119] The individualized gene therapy target points obtained by the prediction model are verified by organoid verification, and it is found that the prediction model can obtain high-accuracy potential treatment target points through tumor transcriptome sequencing and the prediction model. This provides a new way for individualized treatment.

[0120] The embodiments of the present application implement the above technical solutions in actual scenarios. The detailed implementation steps of the individualized gene therapy target prediction method and the organoid verification method are exemplarily described below. Referring to Figure 2 FIG. 1 shows a general flowchart of an individualized gene therapy target prediction and organoid verification method provided by the embodiments of the present application. The following is a detailed description of the key steps.

[0121] (1) Data download and dimensionality reduction processing.

[0122] First, gene expression data (CCLE_RNAseq_rsem_genes_tpm_20180929) and gene dependence effect data (CRISPRGeneEffect) are downloaded from the DepMap public database.

[0123] Then, the two databases are integrated by cell line identifier to obtain cell line samples with both gene expression profiles and CRISPR screening data.

[0124] Among them, the gene expression data is in units of Transcripts Per Million (TPM) and is logarithmically transformed (log2(TPM + 1)) to approximate normal distribution and stabilize variance. The target variable (label, Y), i.e., the target gene, is the CRISPRGeneEffect value, representing the cell growth inhibition effect after gene knockout.

[0125] In order to focus on the most therapeutically potential targets and reduce modeling complexity, the gene effect values in the whole genome range are not predicted. Instead, based on the CRISPRGeneEffect data, according to the comprehensive criteria of modeling error, actual benefit (strong inhibition effect), cell line coverage rate and hit rate, i.e., the gene effect values of each sample are sorted, and the 10 genes with the lowest effect values in each sample are selected. By counting the frequency of these genes appearing in different samples, 47 genes commonly appearing in multiple samples are selected as the final target variables as labels. These genes have a high co-occurrence frequency in different samples, indicating that they may have an important impact on cell growth inhibition.

[0126] Table 1 lists all 47 target genes after dimensionality reduction. Referring to Figure 3As shown in the figure, the target target gene provided in the embodiment of the application provides a distribution ratio of a target target gene in a cell line sample. As can be seen from the figure, the target target gene is contained in at least one strong target point in most (89.79%) cell line samples. Finally, a complete data set containing 667 cell lines is obtained, each cell line is characterized by the expression value of 57820 genes, and the label is the gene effect value of the 47 key genes. These genes are considered to be genes that have an important impact on cell growth inhibition, and are used as target variables Y of a multi-output regression model.

[0127] Table 1

[0128] BUB3 CDC27 FAU HSPE1 KIF11 LRR1 PCNA PHF5A PLK1 POLD1 PRPF38A PSMA3 PSMA6 PSMB3 PUF60 RAN RPL11 RPL12 RPL13 RPL15 RPL17 RPL18A RPL23 RPL31 RPL5 RPL8 RPS11 RPS15 RPS18 RPS19 RPS20 RPS29 RPS3 RPS8 RRM1 RRM2 SARS1 SF1 SF3B5 SMU1 SNRPA1 SNRPD3 SNRPF SNU13 SRSF3 UBL5 VCP

[0129] (2) Data cleaning and arrangement.

[0130] a. Handle missing values: Ensure that the gene expression data (X) has no missing values, and fill the missing values in the target variable (Y) (fill with 0).

[0131] b. Data set division: Divide the merged data set into features (X, gene expression data) and targets (Y, CRISPRGeneEffect values of 47 genes).

[0132] c. Data set segmentation: see Figure 4 As shown in the figure, a data division ratio provided in the embodiment of the application is shown. The 667 cell line data set is randomly divided into a training set (466 cell lines) and an independent test set (201 cell lines) according to a 7:3 ratio, to ensure the unbiasedness of the model evaluation, and a random seed is used to ensure the repeatability of the results.

[0133] (3) Model construction and hyperparameter optimization.

[0134] The XGBoost (Extreme Gradient Boosting) multi-output regression algorithm is used to construct the prediction model. XGBoost is an advanced gradient boosting decision tree framework, which is selected because of its excellent performance in processing high-dimensional features, capturing complex nonlinear relationships, and good explanation ability for feature importance. The experimental process is as follows:

[0135] a. Model Construction: XGBRegressor is wrapped using MultiOutputRegressor to support multi-output regression tasks. Each output corresponds to the CRISPRGeneEffect value of a gene. Optimized parameter settings include: learning rate (learning_rate=0.05), number of estimators (n_estimators=50), maximum depth (max_depth=10), minimum child weight (min_child_weight=1), random seed (seed=0), and subsample ratio (subsample=1).

[0136] b. Model training: Train the multi-output XGBoost regression model using the training set data (train_X and train_y).

[0137] c. Model prediction: The trained model is used to predict the test set data (test_X) to obtain the CRISPRGeneEffect value prediction results for 47 genes in each sample.

[0138] (4) Model performance evaluation.

[0139] The model performance was comprehensively evaluated on an independent test set (201 cell lines). The evaluation scheme used included the following.

[0140] a. See also Figure 5 The image shows a scatter plot of predicted and actual values ​​provided in an embodiment of this application, used to demonstrate the goodness of fit between the model's predicted and actual values. R² (coefficient of determination) is the core indicator for measuring the model's goodness of fit. The training set R² = 0.799, indicating that the model can explain 79.9% of the target variable variance in the training set, demonstrating good performance; the test set R² = 0.786, indicating that the model can explain 78.6% of the variance on unseen test data.

[0141] b. See also Figure 6 As shown, this is a predicted trajectory diagram provided in an embodiment of this application. It clearly shows the three most commonly used regression error indicators: MSE (mean squared error) = 0.3284; RMSE (root mean squared error) = 0.5731; MAE (mean absolute error) = 0.4585, indicating that the XGBoost model has good and consistent prediction ability on both the training and test sets.

[0142] C. To evaluate the model's practicality in selecting the most effective targets, hit probability was used as the core metric. For each cell line in the test set: the model predicted the gene effect values ​​of 47 genes and selected the top 10 genes with the lowest predicted values ​​(i.e., the strongest predicted inhibitory effect). It was checked whether this Top 10 prediction list contained the gene with the lowest actual gene effect value (i.e., the strongest actual inhibitory effect) in the cell line. The number of intersections between the predicted Top 10 gene label list and the actual Top 10 gene label list for each sample was counted, and the proportion of cell lines containing at least k actual potent targets in the predicted Top 10 list was calculated, where k = 1, 2, 3, 4, 5. This proportion is P(Hit ≥ k), which directly quantifies the reliability of the model's recommendation of the top 10 candidate targets for a specific cell line containing truly effective targets.

[0143] This model aims to predict key gene targets in individual cell lines to support target selection for personalized cancer therapy. The model performs excellently on the test set. See also Figure 7 The diagram shown illustrates a hit probability provided in an embodiment of this application. When at least 1, 2, 3, 4, and 5 depleted genes are hit, the model's accuracy is 100%, 99.50%, 95.52%, 82.08%, and 52.23%, respectively. This indicates that the prediction model can provide at least one highly reliable list of depleted genes for each cell line.

[0144] (5) Validation of tumor transcriptome sequencing data.

[0145] We collected breast tumor tissue and adjacent normal tissue from three breast cancer patients and performed transcriptome sequencing. The sequencing data were used as new external input features and then fed into the prediction model for prediction.

[0146] See Figure 8 The image shown is a heatmap of the top 3 personalized target gene prediction results provided in an embodiment of this application. The machine learning model predicted RAN as the top target gene for all three patients.

[0147] To evaluate the predictive ability of the prediction model for target genes and its comparison with traditional target gene screening methods, differential expression analysis was performed using DESeq2 on transcriptome sequencing data. Genes with |log2(fold change)|>0.5 and adjusted P<0.05 were selected as differentially expressed genes. See [link to relevant documentation] Figure 9 The image shown is a volcano plot for differential gene screening based on tumor transcriptome sequencing data, provided in an embodiment of this application. (See also...) Figure 10The image shown is a bar chart of the top 3 differentially expressed genes in tumor transcriptome sequencing data provided in this application embodiment, in which PIP was ultimately identified as the control target gene.

[0148] (6) Validate the effect of knocking down target genes in breast cancer organoids.

[0149] Based on the above screening of therapeutic target genes, knockdown of five therapeutic target genes in three groups will be performed on organoids derived from three breast cancer patients to verify the growth status of the organoids. These three groups of target genes include: the top three personalized therapeutic target genes predicted by the machine learning model; the top one therapeutic target gene selected through differential gene analysis of tumor transcriptome sequencing data from the three breast cancer patients; and the top three personalized therapeutic target genes output by the prediction model after using tumor transcriptome sequencing data from the three breast cancer patients as external data. Among these, target genes RAN and HSP10 overlap with the top three personalized therapeutic target genes predicted by the machine learning model. To conserve organoid resources, only gene SF3B5 will be studied in the third group of target genes.

[0150] First, siRNA screening was performed in the 293T cell line, and the knockdown effects of siRNA-mediated knockdown of five target genes were verified by RT-qPCR and Western blot analysis. See [link to relevant documentation]. Figure 11 The figure shows the statistical results of RT-qPCR and Western blot for validating and screening target siRNAs in the 293T cell line, as provided in an embodiment of this application. Compared with the control group transfected with the negative control siRNA sequence (si-NC), siRNAs with significantly reduced mRNA and protein expression were selected as siRNAs for organoid transfection. The target gene siRNA sequences are shown in Table 2.

[0151] Table 2

[0152] siRNA Sequence RAN-Homo-430 GUGGCAACAAAGUGGAUAUTT AUAUCCACUUUGUUGCCACTT RAN-Homo-230 CACCAACAGAGGACCUAUUTT AAUAGGUCCUCUGUUGGUGTT RAN-Homo-325 GUGCCAUCAUAAUGUUUGATT UCAAACAUUAUGAUGGCACTT RAN-Homo-580 GAGACCCUAACUUGGAAUUTT AAUUCCAAGUUAGGGUCUCTT HSPE1 (HSP10)-Homo-247 GCGUGAAAGUUGGAGAUAATT UUAUCUCCAACUUUCACGCTT HSPE1 (HSP10)-Homo-286 GAGGCACCAAAGUAGUUCUTT AGAACUACUUUGGUGCCUCTT HSPE1 (HSP10)-Homo-345 CUUGGAAAGUACGUAGACUTT AGUCUACGUACUUUCCAAGTT SNRPD3-Homo-439 GGGACGUGGAAACAUCUUUTT AAAGAUGUUUCCACGUCCCTT SNRPD3-Homo-106 GCCGAUUAAAGUACUGCAUTT AUGCAGUACUUUAAUCGGCTT SNRPD3-Homo-265 GGAGCAGGUAUACAUCCGUTT ACGGAUGUAUACCUGCUCCTT SNRPD3-Homo-322 GAAGAACGCACCCAUGUUATT UAACAUGGGUGCGUUCUUCTT PIP-Homo-372 CAGCCGUCGUUGAUGUUAUTT AUAACAUCAACGACGGCUGTT PIP-Homo-304 CUGCCUAUGUGACGACAAUTT AUUGUCGUCACAUAGGCAGTT PIP-Homo-278 GGUGCAUUUAACUAUAAGUTT ACUUAUAGUUAAAUGCACCTT SF3B5-Homo-188 GACUCGUACUGCUCCUACATT UGUAGGAGCAGUACGAGUCTT SF3B5-Homo-210 GCCACUUCGACCUUCUCAATT UUGAGAAGGUCGAAGUGGCTT

[0153] See Figure 12 The image shown is a bright-field diagram representing organoids from three breast cancer patients, provided in an embodiment of this application. See also... Figure 13 The diagram shown is a flowchart of organoid construction and target gene knockdown provided in an embodiment of this application.

[0154] Then, the well-growing organoids were passaged and transfected with the target gene siRNA. RT-qPCR and Western blot were performed at 48 and 72 hours post-transfection, respectively.

[0155] See Figure 14Figure 1 shows a schematic diagram of the imaging and quantitative analysis of the growth of breast cancer organoids after knocking down target genes, wherein A shows the knocking down of the top three target genes screened by machine learning, the target genes screened by conventional differential gene analysis, and the top three target genes screened by external data verification in three patient-derived breast cancer organoids, respectively. Calcein-AM / PI can effectively distinguish between live and dead cells in patient-derived breast cancer organoids. B-D show the quantitative results of the percentage of Calcein-AM / PI staining areas in three patient-derived breast cancer organoids after knocking down five target genes for 72 hours, respectively. E and F show the RT-qPCR and Western blot results of the target genes before and after knocking down. NC, untreated negative control group. All data are expressed as the mean of three independent experiments standard deviation.

[0156] As shown in E and F, the expression of the target genes at the RNA level and the protein level was significantly reduced compared with the negative control siRNA sequence (si-NC) group, indicating that the knocking down of the target genes in patient-derived organoids was successfully achieved.

[0157] Calcein-AM / PI staining was performed on the organoids transfected for 72 hours, followed by observation under a fluorescence microscope. This technique can distinguish between live cells (green fluorescence, corresponding to the lower half of the shadow) and dead cells (red fluorescence, corresponding to the upper half of the shadow), clearly showing the growth of the organoids. As shown in A, compared with the negative control group, the proportion of dead cells in the organoids transfected with the target genes was higher, and the organoids had different degrees of disintegration; while the control organoids remained structurally intact and contained a significant proportion of live cells.

[0158] As shown in B, C, and D, the quantitative statistical graphs of the percentage of Calcein-AM / PI staining areas in three patient-derived organoids after knocking down five target genes, wherein the P1 and P2 patient-derived organoids of the individualized treatment target gene RAN predicted by the machine learning model had the highest cell death rate; while for the P3 patient-derived organoids, the individualized treatment target gene SF3B5 output by the machine learning model after knocking down the external data had the highest cell death rate. The above results show that the individualized treatment target genes predicted by the machine learning model for transcriptome data can be verified in organoids, and the potential therapeutic effect of the individualized treatment target genes predicted by the machine learning model for transcriptome data is stronger than that of the treatment target genes screened by differential gene analysis of transcriptome sequencing data. The target gene RT-qPCR primer sequences are shown in Table 3.

[0159] Table 3

[0160] Gene sequence RAN Fwd GAAAAGCCCTTCCTCTGGCT RAN Rev TGCTCATACTGTGCTGCCAA HSP10 Fwd TCTGAGGCGGAGGGAGTAAT HSP10 Rev CAGCGACTACTGTTGCTTGC SNRPD3 Fwd GGATGTCCTGGGAGAATGCC SNRPD3 Rev TCTGGGGCTTAACCTGCAAC PIP Fwd TTGCTCCAGCTCCTGTTCAG PIP Rev TTCCGAGTGTTGTCCTGAGC SF3B5 Fwd GCGACTCGTACTGCTCCTAC SF3B5 Rev CTCGCGCTTTGCTCTCATTC

[0161] In order to multi-angle evaluate the growth state of the organoids after knocking down the target genes, the CellTiter-Glo® Luminescent Cell Viability Assay Kit was used to detect the cell viability of the organoids after knocking down the target genes. See Figure 15 As shown in FIG. 6, which is a schematic diagram of organoid cell viability provided by the embodiments of the present application, A-C are the cell viability of the organoids after knocking down five target genes in three patient-derived organoids (the ratio of relative fluorescence corresponds to the cell viability of the organoids), and the results show that the cell viability is the lowest after knocking down the target gene RAN in the three patient-derived organoids, further verifying the effectiveness of the machine learning model in predicting the output of individualized treatment target genes. D is the cell viability of the organoids after knocking down the top1 target gene screened by the machine learning and the top1 target gene screened by the conventional differential factor, and the cell viability of the organoids after knocking down the top1 target gene screened by the machine learning and the top1 target gene screened by the conventional differential factor is compared, and the results show that the cell viability of the organoids after knocking down the target gene RAN is lower.

[0162] The above findings collectively indicate that the machine learning model is effective in predicting individualized treatment target genes, and RAN can be a potential treatment target, worthy of further in vivo research.

[0163] Therefore, the technical scheme of the present application uses tumor transcriptome data and sgRNA library screening data for machine learning to train a prediction model, so that some potential treatment targets can be obtained by inputting the tumor transcriptome sequencing data of a patient, saving resources and simplifying operations, and providing a new approach for individualized treatment. Moreover, the organoids can stably retain the key characteristics of the primary tumors from which they are derived, and thus the potential treatment targets obtained can be screened and verified one by one in patient breast cancer organoids, which helps to explore the biological mechanism and evaluate the inhibitory effect of the candidate target genes on tumors.

[0164] On the basis of the above embodiments, the embodiments of the present application provide an individualized gene therapy target point prediction device, as shown in FIG. 7, which is a schematic diagram of an individualized gene therapy target point prediction device provided by the embodiments of the present application. The device 200 comprises: Figure 16

[0165] The acquisition unit 201 is configured to acquire gene expression data and gene-dependent effect data.

[0166] The screening unit 202 is configured to screen a cell line sample set based on the gene expression data and the gene-dependent effect data.

[0167] The determination unit 203 is configured to determine N target genes from a plurality of genes with the lowest gene effect value ranking based on the co-occurrence frequency of the plurality of genes in each cell line sample of the cell line sample set. ​

[0168] a training unit, configured to train a regression model based on the set of cell line samples and the target target gene, to obtain a prediction model;

[0169] a prediction unit, configured to obtain the individualized gene therapy target point of the patient based on tumor transcriptome sequencing data of the patient and the prediction model.

[0170] In a possible implementation, the determination unit is configured to:

[0171] sort the genes corresponding to each cell line sample in the set of cell line samples according to the gene effect value;

[0172] determine the co-occurrence frequency of a plurality of genes with the lowest gene effect value in the set of cell line samples;

[0173] sort the plurality of genes in descending order of the co-occurrence frequency, and select the top N genes as the target target gene.

[0174] In a possible implementation, N>1, and the device further includes a construction unit configured to:

[0175] construct the regression model with N outputs using the XGBoost algorithm.

[0176] In a possible implementation, the training unit is configured to:

[0177] input the cell line sample into the regression model to obtain an output result;

[0178] evaluate the difference between the output result and the target target gene using mean square error;

[0179] optimize the parameters of the regression model based on the difference until the number of iterations of the regression model reaches a preset threshold or the difference reaches a preset standard, to obtain a prediction model.

[0180] In a possible implementation, the device further includes an evaluation unit configured to evaluate the prediction model using a hit probability, the hit probability indicating the proportion of cell line samples in which the top M predicted genes with the lowest predicted gene effect value of each cell line sample include at least K genes in the top M genes with the lowest corresponding gene effect value, the K being greater than or equal to 1.

[0181] On the basis of the above embodiments, the embodiments of the present application provide a computer device, which comprises a processor and a memory:

[0182] The memory is configured to store a computer program;

[0183] The processor is configured to execute the individualized gene therapy target prediction method according to the computer program.

[0184] On the basis of the above-mentioned embodiments, the embodiments of the present application provide a computer readable storage medium for storing a computer program, the computer program realizing the execution of the above-mentioned individualized gene therapy target prediction method when executed by a computer device.

[0185] On the basis of the above-mentioned embodiments, the embodiments of the present application provide a computer program product including a computer program, which, when running on a computer device, causes the computer device to execute the above-mentioned individualized gene therapy target prediction method.

[0186] It should be noted that the embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the system or device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant part can be referred to the method part.

[0187] The above description of disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting personalized gene therapy targets, characterized in that, The method includes: Acquire gene expression data and gene-dependent effect data; Based on the gene expression data and the gene-dependent effect data, a set of cell line samples was selected; Based on the co-occurrence frequency of multiple genes with the lowest gene effect value in each cell line sample of the cell line sample set, N genes among the multiple genes are selected as target genes. The regression model is trained based on the cell line sample set and the target gene to obtain the prediction model; Based on the patient's tumor transcriptome sequencing data and the prediction model, the patient's individualized gene therapy target is obtained.

2. The method according to claim 1, characterized in that, The method of selecting N genes from the multiple genes with the lowest effect values ​​in each cell line sample of the cell line sample set as target genes includes: The genes corresponding to each cell line sample in the cell line sample set are sorted according to their gene effect values; Determine the co-occurrence frequencies of the genes with the lowest gene effect values ​​in the cell line sample set; The genes are sorted in descending order of their co-occurrence frequency, and the top N genes are selected as target genes.

3. The method according to claim 1, characterized in that, Where N>1, the method further includes: The regression model with N outputs is constructed using the XGBoost algorithm.

4. The method according to claim 1, characterized in that, The process of training the regression model based on the cell line sample set and the target gene to obtain the prediction model includes: The cell line sample is input into the regression model to obtain the output result; The mean squared error is used to evaluate the difference between the output result and the target gene; The parameters of the regression model are optimized based on the differences until the number of iterations of the regression model reaches a preset threshold or the differences reach a preset standard, thus obtaining a prediction model.

5. The method according to claim 1, characterized in that, After training the regression model based on the cell line sample set and the target gene to obtain the prediction model, the method further includes: The prediction model is evaluated using the hit probability, which refers to the proportion of cell line samples containing at least K genes among the M predicted genes with the lowest predicted gene effect values ​​in each cell line sample, where K is greater than or equal to 1.

6. An organoid validation method for validating the predictive accuracy of personalized gene therapy targets obtained by the method of claim 1 using a gene control group.

7. A personalized gene therapy target prediction device, characterized in that, The device includes: Acquisition unit, used to acquire gene expression data and gene-dependent effect data; A screening unit is used to screen a set of cell line samples based on the gene expression data and the gene-dependent effect data; The determination unit is used to identify N genes among the multiple genes with the lowest gene effect values ​​in each cell line sample of the cell line sample set as target genes. The training unit is used to train the regression model based on the cell line sample set and the target gene to obtain a prediction model; The prediction unit is used to obtain the patient's personalized gene therapy target based on the patient's tumor transcriptome sequencing data and the prediction model.

8. A computer device, characterized in that, The computer device includes a processor and memory: The memory is used to store computer programs; The processor is configured to perform the method according to any one of claims 1-5 according to the computer program.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program that, when executed by a computer device, performs the method described in any one of claims 1-5.

10. A computer program product comprising a computer program, characterized in that, When it is run on a computer device, it causes the computer device to perform the method described in any one of claims 1-5.

Images