A method for constructing a prediction model for the killing effect of NK cells on tumor cells

Abstract

The present invention discloses a method for constructing a prediction model of the killing effect of NK cells on tumor cells, comprising: obtaining the killing effect of different NK cells on multiple tumor cells in a preset tumor category; constructing a data set according to the receptor expression profile of NK cells, the ligand expression profile of tumor cells and the killing effect; dividing the data set into a training set, a validation set and a test set; constructing a training model based on logistic regression; using the training set, the validation set and the test set to train, validate and test the training model in sequence, and using the training model that passes the test as the prediction model of the killing effect of NK cells on tumor cells. The prediction model of the killing effect of NK cells on tumor cells constructed by the present invention can effectively predict the killing effect of NK cells on tumor cells, which is conducive to selecting the most suitable NK cells for treatment according to the killing effect.

Description

Technical Field

[0001] The present invention relates to the field of information processing technology, and in particular to a method for constructing a prediction model for the killing effect of NK cells on tumor cells. Background Art

[0002] Natural killer (NK) cells are an important type of immune cell with a broad spectrum of anti-tumor activity. In recent years, NK cell therapy has shown great potential in tumor treatment. Currently, cell transfusion therapy is a research focus in NK tumor treatment strategies. By transferring NK cells from peripheral blood, stem cells, etc. into the patient's body, the number and activity of NK cells in the patient's body can be increased. However, there are significant differences in the response of different patients to NK cell therapy, which limits the clinical application of NK cells. The main reason is that there is heterogeneity between tumor cells and NK cells between individuals, and there are differences in the expression profiles of cell surface molecules from different sources. These differentially expressed molecules will affect the NK's killing integration signals on tumor cells, ultimately leading to different clinical efficacy of NK immunotherapy. At present, there is a lack of effective predictive methods to evaluate the effect of NK cell therapy on tumors. Summary of the Invention

[0003] In order to solve the technical problems existing in the background technology, the present invention proposes a method for constructing a prediction model of the killing effect of NK cells on tumor cells.

[0004] The present invention proposes a method for constructing a prediction model for the killing effect of NK cells on tumor cells, comprising:

[0005] Obtain the killing effects of different NK cells on various tumor cells in preset tumor categories;

[0006] Construct a data set based on NK cells, tumor cells, and killing effects;

[0007] Divide the dataset into training, validation, and test sets;

[0008] Build a training model based on logistic regression;

[0009] The training model was trained, verified and tested in sequence using the training set, validation set and test set, and the training model that passed the test was used as the prediction model for the killing effect of NK cells on tumor cells.

[0010] Preferably, the preset tumor category includes any one of breast cancer, gastric cancer, colorectal cancer, hematological tumor, liver cancer, melanoma and ovarian cancer.

[0011] Preferably, a data set is constructed based on NK cells, tumor cells, and killing effects, specifically including:

[0012] Screen out receptors in NK cells and corresponding ligands in tumor cells;

[0013] Based on the receptors in NK cells and the corresponding ligands in tumor cells, the relationship between the receptor expression profile of NK cells, the ligand expression profile of tumor cells, and the killing effect is obtained: the analysis result shows that the ligand expression profile of tumor cells and the receptor expression profile of NK cells interact with each other to jointly determine the killing effect;

[0014] Based on the relationships between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated prediction factor is constructed; wherein the integrated prediction factor represents the actual expression level when the NK cell receptor expression profile interacts with the corresponding tumor cell ligand expression profile;

[0015] The integrated predictors and the corresponding killing effects were combined to form a data set.

[0016] Among them, the relationship between the screened NK cell receptor expression spectrum, the tumor cell ligand expression spectrum and the killing effect is that the tumor cell ligand expression spectrum and the NK cell receptor expression spectrum interact with each other and jointly determine the killing effect.

[0017] Preferably, an integrated predictive factor is constructed based on the relationship between the screened receptors in NK cells and the corresponding ligands in tumor cells, as well as their expression profiles and killing effects, specifically including:

[0018] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the various NK cell receptor proteins screened out during flow cytometry testing;

[0019] The ratio of the average fluorescence intensity values ​​of the experimental group and the control group in the flow cytometry test of the receptor proteins of various NK cells was normalized to obtain the normalized fluorescence intensity value ratio of the NK cell receptors;

[0020] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various screened tumor cells in the flow cytometry test;

[0021] The ratio of the average fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various tumor cells in the flow cytometry test is normalized to obtain the normalized fluorescence intensity value ratio of the ligands of the tumor cells;

[0022] According to the relationship between the NK cell receptor expression profile, the tumor cell ligand expression profile and the killing effect, an integrated prediction factor was constructed using the normalized ratio of the fluorescence intensity values ​​of the NK cell receptor and the corresponding normalized ratio of the fluorescence intensity values ​​of the tumor cell ligand.

[0023] Where CP1 = min(MFI Ratio ligand ,MFI Ratio receptor );

[0024]

[0025] Where CP1 represents the integrated prediction factor based on the ratio of the fluorescence intensity values ​​of the receptor and the ligand, min() represents the minimum value, and MFI Ratio ligand MFIRatio represents the ratio of the normalized fluorescence intensity values ​​of the tumor cell ligands. receptor represents the ratio of the normalized fluorescence intensity values ​​of the NK cell receptors, MFI Ratio (normalized) represents the ratio of the normalized fluorescence intensity values, k represents the fluorescence channel coefficient, and MFI Ratio represents the ratio of the mean fluorescence intensity values.

[0026] Preferably, an integrated predictive factor is constructed based on the relationship between the screened receptors in NK cells and the corresponding ligands in tumor cells, as well as their expression profiles and killing effects, specifically including:

[0027] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the various NK cell receptor proteins screened out during flow cytometry testing;

[0028] The ratio of the average fluorescence intensity values ​​of the experimental group and the control group in the flow cytometry test of the receptor proteins of various NK cells was normalized to obtain the normalized fluorescence intensity value ratio of the NK cell receptors;

[0029] The transcriptome data of the ligands of various tumor cells screened were obtained respectively;

[0030] Normalizing the transcriptome data of the ligands of various tumor cells to obtain normalized transcriptome data of the ligands of the tumor cells;

[0031] Based on the relationship between the NK cell receptor expression profile, tumor cell ligand expression profile, and killing effect, an integrated prediction factor was constructed using the normalized ratio of the NK cell receptor fluorescence intensity value and the corresponding normalized tumor cell ligand transcriptome data.

[0032] Where CP2 = min(RNAseq ligand ,MFI Ratio receptor );

[0033] in,

[0034] Where CP2 represents the integrated prediction factor constructed based on the ratio of the fluorescence intensity value of the receptor and the transcriptome data of the ligand, min() represents the minimum value, and MFI Ratio receptor represents the ratio of the normalized fluorescence intensity values ​​of the NK cell receptors, MFI Ratio (normalized) represents the ratio of the normalized fluorescence intensity values, k represents the fluorescence channel coefficient, MFI Ratio represents the ratio of the mean fluorescence intensity values, RNAseq ligand Represents the normalized transcriptome data of tumor cell ligands.

[0035] Preferably, when the preset tumor category is ovarian cancer, the receptor ligand pair formed by the receptor expression spectrum in NK cells and the ligand expression spectrum in the corresponding tumor cells includes NKG2D-MIC / ULBP, NKp30-B7H6, NKp44-PCNA, TRAIL-DR4 / 5, KIR2DL1-3-HLA-C, KIR2DL4-HLA-G, CD200R-CD200, TIGIT-CD155, SIGLEC-9-CA125, DNAM1-CD112, and CD2-CD58.

[0036] The present invention also proposes a system for constructing a prediction model of the killing effect of NK cells on tumor cells, comprising:

[0037] An acquisition module is used to obtain the killing effects of different NK cells on multiple tumor cells in a preset tumor category;

[0038] A dataset construction module is used to construct a dataset based on NK cells, tumor cells, and killing effects;

[0039] Partition module, used to divide the data set into training set, validation set and test set;

[0040] Model building module, building a training model based on logistic regression;

[0041] The training module is used to train, verify and test the training model in sequence using the training set, validation set and test set, and use the training model that passes the test as the prediction model of the killing effect of NK cells on tumor cells.

[0042] Preferably, the preset tumor category includes any one of breast cancer, gastric cancer, colorectal cancer, hematological tumor, liver cancer, melanoma and ovarian cancer.

[0043] Preferably, the dataset construction module is used to construct a dataset based on NK cells, tumor cells, and killing effects, specifically including:

[0044] The dataset construction module is used to screen receptors in NK cells and corresponding ligands in tumor cells;

[0045] Based on the receptors in NK cells and the corresponding ligands in tumor cells, the relationship between the receptor expression profile of NK cells, the ligand expression profile of tumor cells, and the killing effect is obtained: the analysis result shows that the ligand expression profile of tumor cells and the receptor expression profile of NK cells interact with each other to jointly determine the killing effect;

[0046] Based on the relationships between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated prediction factor is constructed; wherein the integrated prediction factor represents the actual expression level when the NK cell receptor expression profile interacts with the corresponding tumor cell ligand expression profile;

[0047] The integrated predictors and the corresponding killing effects were combined to form a data set.

[0048] In the present invention, the proposed method for constructing a prediction model of the killing effect of NK cells on tumor cells can construct a data set based on NK cells, tumor cells and killing effect; the data set is divided into a training set, a validation set and a test set; a training model based on logistic regression is constructed; the training model is trained, verified and tested in turn using the training set, validation set and test set, and the training model that passes the test is used as a prediction model of the killing effect of NK cells on tumor cells. The prediction model of the killing effect of NK cells on tumor cells constructed by the present invention can effectively predict the killing effect of NK cells on tumor cells, which is conducive to selecting the most suitable NK cells for treatment according to the killing effect, reducing unnecessary treatment costs and side effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of a method for constructing a prediction model for the killing effect of NK cells on tumor cells in one embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of the correspondence between the receptor expression profile of NK cells and the ligand expression profile of tumor cells in one embodiment of the present invention.

[0051] Figure 3 (1) is a linear relationship diagram between the comprehensive signal score of ovarian cancer cells and the normalized NK cell killing effect in one embodiment of the present invention.

[0052] Figure 3 (2) is a linear relationship diagram between the NK cell comprehensive signal score and the killing effect of NK cells on CR-D84 ovarian cancer cells in one embodiment of the present invention.

[0053] Figure 4This is a scatter diagram of the NK cell integrated signal score, tumor cell integrated signal score and killing effect in one embodiment of the present invention.

[0054] Figure 5 This is a nomogram of a prediction model for the killing effect of NK cells on ovarian cancer tumor cells in one embodiment of the present invention.

[0055] Figure 6 (1) is a diagram showing the importance ranking of receptor-ligand pairs in the prediction model of the killing effect of NK cells on ovarian cancer tumor cells in one embodiment of the present invention.

[0056] Figure 6 (2) is a ROC curve diagram of the results of the external test set of the prediction model for the killing effect of NK cells on ovarian cancer tumor cells in one embodiment of the present invention.

[0057] Figure 7 This is a nomogram of a prediction model for the killing effect of NK cells on breast cancer tumor cells in one embodiment of the present invention.

[0058] Figure 8 This is a nomogram of a prediction model for the killing effect of NK cells on gastric cancer tumor cells in one embodiment of the present invention.

[0059] Figure 9 (1) is a diagram showing the importance ranking of receptor-ligand pairs in a prediction model for the killing effect of NK cells on breast cancer tumor cells in one embodiment of the present invention.

[0060] Figure 9 (2) is a ROC curve diagram of the results of the external test set of the prediction model for the killing effect of NK cells on breast cancer tumor cells in one embodiment of the present invention.

[0061] Figure 9 (3) is a diagram showing the importance ranking of receptor-ligand pairs in the prediction model of the killing effect of NK cells on gastric cancer tumor cells in one embodiment of the present invention.

[0062] Figure 9 (4) is a ROC curve diagram of the results of the external test set of the prediction model for the killing effect of NK cells on gastric cancer tumor cells in one embodiment of the present invention.

[0063] Figure 10 This is a nomogram of a prediction model for the killing effect of NK cells on transcriptome data of various tumor cells in one embodiment of the present invention.

[0064] Figure 11This is an ROC line graph of the test set of the prediction model for the killing effect of NK cells on transcriptome data of various tumor cells in one embodiment proposed by the present invention. DETAILED DESCRIPTION

[0065] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0066] Reference Figure 1 The present invention proposes a method for constructing a prediction model of the killing effect of NK cells on tumor cells, comprising:

[0067] Obtain the killing effects of different NK cells on various tumor cells in preset tumor categories;

[0068] Construct a data set based on NK cells, tumor cells, and killing effects;

[0069] Divide the dataset into training, validation, and test sets;

[0070] Build a training model based on logistic regression;

[0071] The training model was trained, verified and tested in sequence using the training set, validation set and test set, and the training model that passed the test was used as the prediction model for the killing effect of NK cells on tumor cells.

[0072] The present invention can construct a data set based on NK cells, tumor cells and killing effects; divide the data set into a training set, a validation set and a test set; construct a training model based on logistic regression; use the training set, validation set and test set to train, validate and test the training model in sequence, and use the training model that passes the test as a prediction model for the killing effect of NK cells on tumor cells.

[0073] The NK cell killing effect prediction model constructed in the present invention can effectively predict the killing effect of NK cells on tumor cells, which is conducive to selecting the most appropriate NK cells for treatment based on the killing effect, thereby reducing unnecessary treatment costs and side effects.

[0074] In this embodiment, the preset tumor category includes any one of breast cancer, gastric cancer, colorectal cancer, hematological tumor, liver cancer, melanoma and ovarian cancer.

[0075] In this example, a data set was constructed based on NK cells, tumor cells, and the killing effect, specifically including:

[0076] Screen out receptors in NK cells and corresponding ligands in tumor cells;

[0077] Based on the receptors in NK cells and the corresponding ligands in tumor cells, the relationship between the NK cell receptor expression profile, the tumor cell ligand expression profile and the killing effect is obtained:

[0078] An integrated prediction factor is constructed based on the relationships between the screened NK cell receptors and the corresponding tumor cell ligands, as well as the NK cell receptor expression profile, the tumor cell ligand expression profile, and the killing effect; wherein the integrated prediction factor represents the actual expression level when the NK cell receptor expression profile interacts with the corresponding tumor cell ligand expression profile;

[0079] The integrated predictors and the corresponding killing effects were combined to form a data set.

[0080] This embodiment is configured in such a way that the independent variables affecting the killing effect can be determined based on the relationship between the receptor expression spectrum of NK cells, the ligand expression spectrum of tumor cells and the killing effect, and the independent variables affecting the killing effect and the dependent variable of the killing effect are combined to form a data set for training the model, which is conducive to improving the accuracy of the prediction.

[0081] According to existing literature or papers, the relationship between the receptor expression spectrum of screened NK cells, the ligand expression spectrum of tumor cells and the killing effect is that the ligand expression spectrum of tumor cells and the receptor expression spectrum of NK cells interact with each other and jointly determine the killing effect.

[0082] In one embodiment, an integrated predictive factor is constructed based on the relationship between the screened receptors in NK cells and the corresponding ligands in tumor cells, as well as their expression profiles and killing effects, specifically including:

[0083] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the various NK cell receptor proteins screened out during flow cytometry testing;

[0084] The ratio of the average fluorescence intensity values ​​of the experimental group and the control group in the flow cytometry test of the receptor proteins of various NK cells was normalized to obtain the normalized fluorescence intensity value ratio of the NK cell receptors;

[0085] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various screened tumor cells in the flow cytometry test;

[0086] Normalizing the ratio of the average fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various tumor cells in the flow cytometry test to obtain the normalized fluorescence intensity value ratio of the ligands of the tumor cells;

[0087] According to the relationship between the NK cell receptor expression profile, the tumor cell ligand expression profile and the killing effect, an integrated prediction factor was constructed using the normalized ratio of the fluorescence intensity values ​​of the NK cell receptor and the corresponding normalized ratio of the fluorescence intensity values ​​of the tumor cell ligand.

[0088] Where CP1 = min(MFI Ratio ligand ,MFI Ratio receptor );

[0089]

[0090] Where CP1 represents the integrated prediction factor based on the ratio of the fluorescence intensity values ​​of the receptor and the ligand, min() represents the minimum value, and MFI Ratio ligand MFIRatio represents the ratio of the normalized fluorescence intensity values ​​of the tumor cell ligands. receptor represents the ratio of the normalized fluorescence intensity values ​​of the NK cell receptors, MFI Ratio (normalized) represents the ratio of the normalized fluorescence intensity values, k represents the fluorescence channel coefficient, and MFI Ratio represents the ratio of the mean fluorescence intensity values.

[0091] In another embodiment, an integrated predictive factor is constructed based on the relationship between the screened receptors in NK cells and the corresponding ligands in tumor cells, as well as their expression profiles and killing effects, specifically including:

[0092] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the various NK cell receptor proteins screened out during flow cytometry testing;

[0093] The ratio of the average fluorescence intensity values ​​of the experimental group and the control group in the flow cytometry test of the receptor proteins of various NK cells was normalized to obtain the normalized fluorescence intensity value ratio of the NK cell receptors;

[0094] The transcriptome data of the ligands of various tumor cells screened were obtained respectively;

[0095] Normalizing the transcriptome data of the ligands of various tumor cells to obtain normalized transcriptome data of the ligands of the tumor cells;

[0096] Based on the relationship between the NK cell receptor expression profile, tumor cell ligand expression profile, and killing effect, an integrated prediction factor was constructed using the normalized ratio of the NK cell receptor fluorescence intensity value and the corresponding normalized tumor cell ligand transcriptome data.

[0097] Where CP2 = min(RNAseq ligand,MFI Ratio receptor );

[0098] in,

[0099] Where CP2 represents the integrated prediction factor constructed based on the ratio of the fluorescence intensity value of the receptor and the transcriptome data of the ligand, min() represents the minimum value, and MFI Ratio receptor represents the ratio of the normalized fluorescence intensity values ​​of the NK cell receptors, MFI Ratio (normalized) represents the ratio of the normalized fluorescence intensity values, k represents the fluorescence channel coefficient, MFI Ratio represents the ratio of the mean fluorescence intensity values, RNAseq ligand Represents the normalized transcriptome data of tumor cell ligands.

[0100] In this embodiment, the receptor ligand pairs formed by the receptor expression profile in NK cells and the corresponding ligand expression profile in tumor cells include NKG2D-MIC / ULBP, NKp30-B7H6, NKp44-PCNA, TRAIL-DR4 / 5, KIR2DL1-3-HLA-C, KIR2DL4-HLA-G, CD200R-CD200, TIGIT-CD155, SIGLEC-9-CA125, DNAM1-CD112, and CD2-CD58.

[0101] Of course, the method for constructing a prediction model for the killing effect of NK cells on tumor cells proposed in the present invention is also applicable to constructing a prediction model for the NK treatment effect of transcriptome data.

[0102] The present invention also proposes a system for constructing a prediction model of the killing effect of NK cells on tumor cells, comprising:

[0103] An acquisition module is used to obtain the killing effects of different NK cells on multiple tumor cells in a preset tumor category;

[0104] A dataset construction module is used to construct a dataset based on NK cells, tumor cells, and killing effects;

[0105] Partition module, used to divide the data set into training set, validation set and test set;

[0106] Model building module, building a training model based on logistic regression;

[0107] The training module is used to train, verify and test the training model in sequence using the training set, validation set and test set, and use the training model that passes the test as the prediction model of the killing effect of NK cells on tumor cells.

[0108] Among them, the preset tumor categories include any one of breast cancer, gastric cancer, colorectal cancer, blood cancer, liver cancer, melanoma and ovarian cancer.

[0109] The dataset construction module is used to construct a dataset based on NK cells, tumor cells, and killing effects, specifically including:

[0110] The dataset construction module is used to screen receptors in NK cells and corresponding ligands in tumor cells;

[0111] Based on the screened receptors in NK cells and the corresponding ligands in tumor cells, the relationship between the receptor expression profile of NK cells, the ligand expression profile of tumor cells, and the killing effect is obtained: the analysis result shows that the ligand expression profile of tumor cells and the receptor expression profile of NK cells interact with each other to jointly determine the killing effect;

[0112] Based on the relationships between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated prediction factor is constructed; wherein the integrated prediction factor represents the actual expression level when the NK cell receptor expression profile interacts with the corresponding tumor cell ligand expression profile;

[0113] The integrated predictors and the corresponding killing effects were combined to form a data set.

[0114] The specific method for constructing the data set in this embodiment is the same as the above method embodiment, and will not be described in detail here.

[0115] The method for constructing the data set construction module in this embodiment is the same as the above method embodiment.

[0116] The present invention will be described below with reference to specific embodiments.

[0117] Example 1

[0118] This example proposes a method for constructing a prediction model for the killing effect of NK cells on ovarian cancer tumor cells, comprising:

[0119] Obtained the killing effect of 52 types of NK cells on 14 types of ovarian cancer cells;

[0120] A data set was constructed based on the killing effects of 52 types of NK cells on 14 types of ovarian cancer cells;

[0121] Divide the dataset into training, validation, and test sets;

[0122] Build a training model based on logistic regression;

[0123] The training model was trained, verified and tested in sequence using the training set, validation set and test set, and the training model that passed the test was used as the prediction model for the killing effect of NK cells on tumor cells.

[0124] Among them, a data set was constructed based on the killing effects of 52 types of NK cells on 14 types of ovarian cancer cells, including:

[0125] The receptors in NK cells and the corresponding ligands in tumor cells are screened out; wherein, the receptor ligand pairs formed by the receptor expression profile in NK cells and the ligand expression profile in the corresponding tumor cells include NKG2D-MIC / ULBP, NKp30-B7H6, NKp44-PCNA, TRAIL-DR4 / 5, KIR2DL1-3-HLA-C, KIR2DL4-HLA-G, CD200R-CD200, TIGIT-CD155, SIGLEC-9-CA125, DNAM1-CD112, CD2-CD58, such as Figure 2 As shown;

[0126] Based on the receptors in the NK cells and the corresponding ligands in the tumor cells, the relationship between the receptor expression spectrum of the NK cells, the ligand expression spectrum of the tumor cells, and the killing effect is obtained: wherein the relationship between the screened receptor expression spectrum of the NK cells, the ligand expression spectrum of the tumor cells, and the killing effect is that the ligand expression spectrum of the tumor cells and the receptor expression spectrum of the NK cells interact with each other and jointly determine the killing effect;

[0127] Based on the relationship between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated prediction factor is constructed;

[0128] The integrated predictors and the corresponding killing effects were combined to form a data set.

[0129] This example verifies the load relationship between the interaction between the ligand expression profile of tumor cells and the receptor expression profile of NK cells, which jointly determine the killing effect. The specific verification process includes:

[0130] Determine whether the receptor expression profile of NK cells and the ligand expression profile of tumor cells are linearly correlated with the killing effect:

[0131] If both are yes, then determine whether there is a linear correlation between the receptor expression profile of NK cells and the ligand expression profile of tumor cells and the killing effect;

[0132] If not, determine whether there is a multilinear relationship between the NK cell receptor expression profile and the tumor cell ligand expression profile; wherein a multilinear relationship indicates that there is a high correlation between the two independent variables, the NK cell receptor expression profile and the tumor cell ligand expression profile;

[0133] If not, it indicates that there is a nonlinear interaction between the receptor expression profile of NK cells and the ligand expression profile of tumor cells; wherein, the nonlinear interaction means that the ligand expression profile of tumor cells and the receptor expression profile of NK cells interact with each other and jointly determine the killing effect.

[0134] Among them, judging whether the receptor expression profile of NK cells is linearly correlated with the killing effect includes:

[0135] CR-D84 ovarian cancer cells were selected from 14 ovarian cancer cell lines;

[0136] The killing effects of various NK cells on CR-D84 ovarian cancer cells were extracted from the results;

[0137] Based on the differences in receptor expression profiles of various NK cells, PCA principal component analysis was used to assign NK cell comprehensive signaling scores to various NK cells;

[0138] Linear regression analysis was performed on the NK cell comprehensive signal score of each NK cell and the killing effect on CR-D84 ovarian cancer cells, with the receptor expression profile as the independent variable and the killing effect as the dependent variable. The first result was as follows Figure 3 As shown in (1); based on the first result, it was determined whether there was a linear correlation between the NK cell comprehensive signal score and the killing effect; wherein, the P value in the first result was 0.0229, the root mean square error (RMSD) was 0.210, and the Spearman correlation coefficient r was 0.3464;

[0139] Therefore, it was determined that there was a linear correlation between the NK cell comprehensive signal score and the killing effect, which indicated that the receptor expression profile of NK cells was related to the killing effect.

[0140] Among them, determining whether the ligand expression profile of tumor cells is linearly correlated with the killing effect specifically includes:

[0141] Select several NK cell lines from different NK cell lines;

[0142] The killing effects of several selected NK cell lines on 14 types of ovarian cancer cells were extracted from the killing effects of different NK cells on 14 types of ovarian cancer cells;

[0143] One HO8910 tumor cell was selected from 14 ovarian cancer cell lines as a benchmark, and the killing effects of several selected NK cell lines on the 14 ovarian cancer cell lines were normalized;

[0144] Based on the differences in the expression profiles of various ovarian cancer cells, PCA principal component analysis was used to assign tumor cell comprehensive signal scores to the 14 ovarian cancer cell types;

[0145] Linear regression analysis was performed on the tumor cell comprehensive signal scores and normalized killing effects of 14 ovarian cancer cells, and the second result was obtained as follows: Figure 3 (2) As shown; judging whether there is a linear correlation between the tumor cell comprehensive signal score and the killing effect according to the second result;

[0146] Among them, the P value in the second result is 0.0039, the RMSD is 0.16, and the r is 0.7321, which indicates that the comprehensive signal score of tumor cells is linearly correlated with the killing effect, indicating that the ligand expression profile of tumor cells is correlated with the killing effect.

[0147] Among them, determining whether there is a multi-linear relationship between the receptor expression profile of NK cells and the ligand expression profile of tumor cells specifically includes:

[0148] The third result was obtained by linear regression analysis based on the NK cell comprehensive signal score, tumor cell comprehensive signal score and the killing effect of 52 types of NK cells on 14 tumor cells in the preset tumor categories. Figure 4 As shown;

[0149] Based on the third result, determine whether there is a multi-linear relationship between the receptor expression profile of NK cells and the ligand expression profile of tumor cells.

[0150] Specifically, a killing effect prediction model was constructed using linear regression with the molecular expression profiles of NK cells and tumor cells as input independent variables: Y = 0.438-0.020×Tumor score+0.029×NK score, where Tumor score is the comprehensive signal score of tumor cells and NK score is the comprehensive signal score of NK cells.

[0151] Among them, the determination coefficient R in the third result 2 is 0.008, the F value is 0.644, and the P value is 0.527. Figure 4The three-dimensional scatter plot shows a clear nonlinear trend, indicating that the constructed regression model may be meaningless and that the model's predictions are subject to significant bias. However, the P values ​​for the slopes of the combined signaling scores for both tumor cells and NK cells were all greater than 0.05, indicating that there was no significant linear relationship between the combined signaling scores of both cells and the NK cell cytotoxicity prediction model. This contrasts with previous results and suggests that the effects of tumor cells and NK cells on cytotoxicity are not independent, but rather that nonlinear interactions or multicollinearity exist. Therefore, when assessing whether there was multicollinearity between the NK cell receptor expression profile and the tumor cell ligand expression profile, a variance inflation factor (VIF) of 1.015 and a tolerance of 0.985 were used, indicating that there was no multicollinearity between the NK cell receptor expression profile and the tumor cell ligand expression profile, indicating that there was a nonlinear interaction, meaning that tumor cells and NK cells interacted and jointly determined the cytotoxicity.

[0152] Among them, based on the relationship between the screened receptors in NK cells and the corresponding ligands in tumor cells, as well as their expression profiles and killing effects, an integrated predictive factor is constructed, specifically including:

[0153] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the various NK cell receptor proteins screened out during flow cytometry testing;

[0154] The ratio of the average fluorescence intensity values ​​of the experimental group and the control group in the flow cytometry test of the receptor proteins of various NK cells was normalized to obtain the normalized fluorescence intensity value ratio of the NK cell receptors;

[0155] Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various screened tumor cells in the flow cytometry test;

[0156] Normalizing the ratio of the average fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various tumor cells in the flow cytometry test to obtain the normalized fluorescence intensity value ratio of the ligands of the tumor cells;

[0157] According to the relationship between the NK cell receptor expression profile, the tumor cell ligand expression profile and the killing effect, an integrated prediction factor was constructed using the normalized ratio of the fluorescence intensity values ​​of the NK cell receptor and the corresponding normalized ratio of the fluorescence intensity values ​​of the tumor cell ligand.

[0158] Where CP1 = min(MFI Ratio ligand ,MFI Ratio receptor );

[0159]

[0160] Where CP1 represents the integrated prediction factor based on the ratio of the fluorescence intensity values ​​of the receptor and the ligand, min() represents the minimum value, and MFI Ratio ligand MFIRatio represents the ratio of the normalized fluorescence intensity values ​​of the tumor cell ligands. receptor represents the ratio of the normalized fluorescence intensity values ​​of the NK cell receptors, MFI Ratio (normalized) represents the ratio of the normalized fluorescence intensity values, k represents the fluorescence channel coefficient, and MFI Ratio represents the ratio of the mean fluorescence intensity values.

[0161] Among them, such as Figure 5 As shown in Figure 2, the prediction model for the killing effect of NK cells on ovarian cancer tumor cells is

[0162] Logit(P)=0.013+0.022CP1(NKp30-B7H6)-0.321CP1(NKG2D-MIC /

[0163] ULBP)+0.149CP1(TRAIL-DR4 / 5)-0.214CP1(NKp44-PCNA)+

[0164] 0.285CP1(KIR2DL1-3-HLA-C)-0.064CP1(KIR2DL4-HLA-G)+

[0165] 0.365CP1(CD200R-CD200)+0.005CP1(SIGLEC-9-CA125)+

[0166] 0.561CP1(CD2-CD58)+0.203CP1(CD155 / CD112-TIGIT / DNAM1);

[0167]

[0168] Where P(W=1|X) represents the probability that the target variable W=1 given the feature X.

[0169] Among them, CP1 (CD155 / CD112-TIGIT / DNAM1) integrates two pairs of receptor ligands, TIGIT-CD155 and DNAM1-CD112.

[0170] Specifically,

[0171]

[0172] After the prediction model of NK cell killing effect on ovarian cancer tumor cells was constructed, the weight coefficients of each receptor ligand pair were observed through machine learning, and the importance ranking of the receptor ligands in the prediction model of NK cell killing effect on ovarian cancer tumor cells was obtained as follows: Figure 6 (1) shown.

[0173] Example 2

[0174] A separate external test set was used to evaluate the predictive performance of the NK cell killing effect prediction model for ovarian cancer tumor cells constructed in Example 1. The validation accuracy of the NK cell killing effect prediction model for ovarian cancer tumor cells constructed in Example 1 was 79.49%, and the area under the ROC curve (AUC) was 0.8842. Figure 6 (2) shown.

[0175] Example 3

[0176] This embodiment proposes a method for constructing a prediction model of the killing effect of NK cells on breast cancer tumor cells. The nomogram of the constructed prediction model of the killing effect of NK cells on breast cancer tumor cells is as follows: Figure 7 The prediction model of the killing effect of NK cells on breast cancer tumor cells in this embodiment is

[0177] Logit(P)=-0.867+0.467CP1(NKp30-B7H6)+0.292CP1(NKG2D-

[0178] MIC / ULBP)+0.027CP1(TRAIL-DR4 / 5)-0.514CP1(NKp44-PCNA)-0.36CP1(KIR2DL1-3-HLA-C)+0.308CP1(KIR2DL4-HLA-G)+

[0179] 0.613CP1(CD200R-CD200)-0.633CP1(SIGLEC-9-CA125)-0.154CP1(CD2-CD58)-0.191CP1(CD155 / CD112-TIGIT / DNAM1).

[0180] A separate external test set was used to evaluate the predictive performance of the prediction model for the killing effect of NK cells on breast cancer tumor cells in this embodiment. The prediction results are as follows: Figure 9 As shown in (2), the prediction accuracy is 80.95% and AUC = 0.7852.

[0181] After the prediction model of NK cell killing effect on breast cancer tumor cells was constructed, the weight coefficients of each receptor ligand pair were observed through machine learning, and the importance ranking of the receptor ligand pairs of the prediction model of NK cell killing effect on breast cancer tumor cells was obtained as follows: Figure 9 (1) shown.

[0182] Example 4

[0183] This embodiment proposes a method for constructing a prediction model of the killing effect of NK cells on gastric cancer cells. The nomogram of the constructed prediction model of the killing effect of NK cells on gastric cancer cells is as follows: Figure 8 The prediction model of NK cell killing effect on gastric cancer cells is shown as follows:

[0184] Logit(P)=0.373+0.677CP1(NKp30-B7H6)+0.844CP1(NKG2D-MIC /

[0185] ULBP)+0.374CP1(TRAIL-DR4 / 5)+0.240CP1(NKp44-PCNA)-0.646CP1(KIR2DL1-3-HLA-C )+0.946CP1(KIR2DL4-HLA-G)-0.692CP1(CD200R-CD200)-0.081CP1(SIGLEC-9-CA125)+

[0186] 0.121CP1(CD2-CD5)-0.375CP1(CD155 / CD112-TIGIT / DNAM).

[0187] A separate external test set was used to evaluate the predictive performance of the prediction model for the killing effect of NK cells on gastric cancer cells in this embodiment. The prediction results are as follows: Figure 9 (4) The prediction accuracy is 84.85% and AUC is 0.8759.

[0188] After the prediction model of NK cell killing effect on gastric cancer tumor cells was constructed, the weight coefficients of each receptor ligand pair were observed through machine learning, and the importance ranking of the receptor ligands in the prediction model of NK cell killing effect on gastric cancer tumor cells was obtained as follows: Figure 9 (3) shown.

[0189] Example 5

[0190] This embodiment proposes a method for constructing a prediction model of the killing effect of NK cells on the transcriptome data of various tumor cells. The nomogram of the constructed prediction model of the transcriptome data of NK cells on various tumor cells is as follows: Figure 10The prediction model of NK cell killing effect on gastric cancer cells is shown as follows:

[0191] Logit(P)=0.029+0.720CP2(NKp30-B7H6)+0.394CP2(NKG2D-MIC /

[0192] ULBP)+0.138CP2(TRAIL-DR4 / 5)-0.088CP2(NKp44-PCNA)-0.587CP2(KIR2DL1-3-HLA-C)-0.594CP2(KIR2DL4-HLA-G)+

[0193] 0.176CP2(CD200R-CD200)+0.279CP2(SIGLEC-9-CA125)+

[0194] 0.574CP2(CD2-CD5)-0.185CP2(CD155 / CD112-TIGIT / DNAM).

[0195] An external test set was used to evaluate the prediction performance of the NK cell killing effect prediction model in this embodiment. The prediction results are as follows: Figure 11 As shown, the prediction accuracy is 70.21% and AUC = 0.7554.

[0196] Comparative Example 1

[0197] The existing random forest model was used to predict the separate external test set used in Example 2, with a prediction accuracy of 72.09% and an area under the ROC curve (AUC) of 0.74. At the same time, the model had an overfitting problem.

[0198] It can be seen from Example 2 and Comparative Example 1 that the prediction model of the killing effect of NK cells on tumor cells constructed by the method for constructing the prediction model of the killing effect of NK cells on tumor cells proposed in the present invention has a high prediction accuracy.

[0199] From the results of Examples 2-5, it can be seen that the method for constructing a prediction model for the killing effect of NK cells on tumor cells proposed in the present invention can construct a corresponding prediction model with high prediction accuracy for each tumor.

[0200] from Figure 6 (1) and Figure 9 (1) Figure 9 As can be seen in (3), different receptor ligands have different effects on different types of tumors.

[0201] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for constructing a prediction model for the killing effect of NK cells on tumor cells, characterized in that: include: Obtain the killing effects of different NK cells on various tumor cells in preset tumor categories; Construct a data set based on NK cells, tumor cells, and killing effects; Divide the dataset into training, validation, and test sets; Build a training model based on logistic regression; The training model is trained, verified, and tested using the training set, validation set, and test set in sequence, and the training model that passes the test is used as the prediction model for the killing effect of NK cells on tumor cells; Among them, a data set is constructed based on NK cells, tumor cells and killing effects, including: Screen out receptors in NK cells and corresponding ligands in tumor cells; Based on the receptors in the screened NK cells and the ligands in the corresponding tumor cells, the relationship between the receptor expression profile of the screened NK cells, the ligand expression profile of the tumor cells and the killing effect is obtained: Based on the relationships between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated prediction factor is constructed; wherein the integrated prediction factor represents the actual expression level when the NK cell receptor expression profile interacts with the corresponding tumor cell ligand expression profile; The integrated predictors and the corresponding killing effects were combined to form a data set.

2. The method for constructing a prediction model of the killing effect of NK cells on tumor cells according to claim 1, characterized in that: The preset tumor categories include any one of breast cancer, gastric cancer, colorectal cancer, hematological cancer, liver cancer, melanoma and ovarian cancer.

3. The method for constructing a prediction model of the killing effect of NK cells on tumor cells according to claim 1, characterized in that: The relationship between the screened NK cell receptor expression profile, tumor cell ligand expression profile and killing effect is that the tumor cell ligand expression profile and the NK cell receptor expression profile interact with each other and jointly determine the killing effect.

4. The method for constructing a prediction model of the killing effect of NK cells on tumor cells according to claim 1, characterized in that: Based on the relationship between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated predictive factor was constructed, specifically including: Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the various NK cell receptor proteins screened out during flow cytometry testing; The ratio of the average fluorescence intensity values ​​of the experimental group and the control group in the flow cytometry test of the receptor proteins of various NK cells was normalized to obtain the normalized fluorescence intensity value ratio of the NK cell receptors; Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various screened tumor cells in the flow cytometry test; Normalizing the ratio of the average fluorescence intensity values ​​of the experimental group and the control group of the ligand proteins of various tumor cells in the flow cytometry test to obtain the normalized fluorescence intensity value ratio of the ligands of the tumor cells; Based on the relationship between the NK cell receptor expression profile, tumor cell ligand expression profile, and killing effect, an integrated prediction factor was constructed using the normalized ratio of the fluorescence intensity values ​​of the NK cell receptor and the corresponding normalized ratio of the fluorescence intensity values ​​of the tumor cell ligand. Where CP1 = min(MFI Ratio ligand ,MFI Ratio receptor ); in, Where CP1 represents the integrated prediction factor based on the ratio of the fluorescence intensity values ​​of the receptor and the ligand, min() represents the minimum value, and MFI Ratio ligand MFIRatio represents the ratio of the normalized fluorescence intensity values ​​of the tumor cell ligands. receptor represents the ratio of the normalized fluorescence intensity values ​​of the NK cell receptors, MFI Ratio (normalized) represents the ratio of the normalized fluorescence intensity values, k represents the fluorescence channel coefficient, and MFI Ratio represents the ratio of the mean fluorescence intensity values.

5. The method for constructing a prediction model of the killing effect of NK cells on tumor cells according to claim 1, characterized in that: Based on the relationship between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated predictive factor was constructed, specifically including: Obtain the ratio of the mean fluorescence intensity values ​​of the experimental group and the control group of the various NK cell receptor proteins screened out during flow cytometry testing; The ratio of the average fluorescence intensity values ​​of the experimental group and the control group in the flow cytometry test of the receptor proteins of various NK cells was normalized to obtain the normalized fluorescence intensity value ratio of the NK cell receptors; The transcriptome data of the ligands of various tumor cells screened were obtained respectively; Normalizing the transcriptome data of the ligands of various tumor cells to obtain normalized transcriptome data of the ligands of the tumor cells; Based on the relationship between the NK cell receptor expression profile, tumor cell ligand expression profile, and killing effect, an integrated prediction factor was constructed using the normalized ratio of the NK cell receptor fluorescence intensity value and the corresponding normalized tumor cell ligand transcriptome data. Where CP2 = min(RNAseq ligand ,MFI Ratio receptor ); in, Where CP2 represents the integrated prediction factor constructed based on the ratio of the fluorescence intensity value of the receptor and the transcriptome data of the ligand, min() represents the minimum value, and MFI Ratio receptor represents the ratio of the normalized fluorescence intensity values ​​of the NK cell receptors, MFI Ratio (normalized) represents the ratio of the normalized fluorescence intensity values, k represents the fluorescence channel coefficient, MFIRatio represents the ratio of the mean fluorescence intensity values, RNAseq ligand Represents the normalized transcriptome data of tumor cell ligands.

6. The method for constructing a prediction model of the killing effect of NK cells on tumor cells according to claim 1, characterized in that: The receptor ligand pairs formed by the receptor expression profiles in NK cells and the corresponding ligand expression profiles in tumor cells include NKG2D-MIC / ULBP, NKp30-B7H6, NKp44-PCNA, TRAIL-DR4 / 5, KIR2DL1-3-HLA-C, KIR2DL4-HLA-G, CD200R-CD200, TIGIT-CD155, SIGLEC-9-CA125, DNAM1-CD112, and CD2-CD58.

7. A system for constructing a prediction model of the killing effect of NK cells on tumor cells, characterized in that: include: An acquisition module is used to obtain the killing effects of different NK cells on multiple tumor cells in a preset tumor category; A dataset construction module is used to construct a dataset based on NK cells, tumor cells, and killing effects; Partition module, used to divide the data set into training set, validation set and test set; Model building module, building a training model based on logistic regression; The training module is used to train, verify, and test the training model using the training set, validation set, and test set in sequence, and the training model that passes the test is used as the prediction model of the killing effect of NK cells on tumor cells; The dataset construction module is used to construct a dataset based on NK cells, tumor cells, and killing effects, specifically including: The dataset construction module is used to screen receptors in NK cells and corresponding ligands in tumor cells; Based on the screened receptors in NK cells and the corresponding ligands in tumor cells, the relationship between the receptor expression profile of NK cells, the ligand expression profile of tumor cells and the killing effect is obtained: Based on the relationships between the screened NK cell receptors and the corresponding tumor cell ligands, as well as their expression profiles and killing effects, an integrated prediction factor is constructed; wherein the integrated prediction factor represents the actual expression level when the NK cell receptor expression profile interacts with the corresponding tumor cell ligand expression profile; The integrated predictors and the corresponding killing effects were combined to form a data set.

8. The method for constructing a prediction model of the killing effect of NK cells on tumor cells according to claim 7, characterized in that: The preset tumor categories include any one of breast cancer, gastric cancer, colorectal cancer, hematological cancer, liver cancer, melanoma and ovarian cancer.

Images