Construction method and application of platinum drug sensitivity prediction model

Abstract

The invention relates to the technical field of drug sensitivity prediction, and discloses a platinum drug sensitivity prediction model construction method and application thereof, and the method comprises the steps: constructing a training set; according to the training set, a plurality of candidate models are obtained through training through a plurality of machine learning algorithms; and screening the plurality of candidate models according to a preset model index to obtain at least two platinum drug sensitivity prediction models. Based on the cancer species data of the target group, the applied platinum drug type data, the immune cell infiltration level data of the tumor sample tissue and the PTC data, a plurality of high-precision platinum drug sensitivity prediction models are obtained through training by utilizing a machine learning algorithm, and the method can be used for accurately predicting the platinum drug sensitivity in combination with a hard voting method. The method is high in model prediction accuracy, suitable for people, more targeted, simple and clear in prediction result, short in service cycle and low in cost.

Description

Technical Field

[0001] The present invention relates to the technical field of drug sensitivity prediction, and in particular to a method for constructing a platinum drug sensitivity prediction model and its application. Background Art

[0002] Currently, chemotherapy regimens for cancer are mostly selected based on the clinical stage and pathological type of the cancer. However, due to the heterogeneity of tumors and their insensitivity to chemotherapy drugs, the treatment is somewhat blind. Chemotherapy drugs themselves are cytotoxic. During chemotherapy, the cancer may not be controlled, and adverse reactions may occur, causing unnecessary pain to the patient, affecting the quality of life and even the prognosis of the tumor. For example, platinum chemotherapy drugs include cisplatin, carboplatin, lobaplatin, oxaliplatin, etc., which can treat gastrointestinal tumors, gynecological tumors, lung cancer and other diseases. Although this type of drug can treat tumors, due to the blind medication regimen, it will also produce side effects such as liver and kidney damage and bone marrow suppression.

[0003] With the rapid development of molecular biology, it has been found that gene expression in cancer is closely related to the occurrence, development, metastasis, and prognosis of tumors, and is also related to the chemotherapy efficacy of tumors. In addition, more and more machine learning models are used to predict drug sensitivity. How to use machine learning methods to accurately predict patient drug sensitivity and assist in guiding personalized treatment of tumor patients is of great clinical significance. At present, the data used by most drug sensitivity prediction models mainly come from public databases, and there is a lack of real clinical data for Chinese people, which makes the prediction results inconsistent with the actual clinical effects, and the prediction results of the model are IC50 values, which cannot intuitively show the effect of medication.

[0004] Therefore, there is an urgent need for a method to construct a platinum drug sensitivity prediction model to assist in intuitively and accurately predicting patient drug sensitivity, guide personalized treatment for cancer patients, and improve patient cure rates. Summary of the invention

[0005] The present invention provides a method for constructing a platinum drug sensitivity prediction model and its application, so as to solve the defects that the prediction results of the existing drug sensitivity prediction model are low in accuracy and the prediction results are IC50 values, which cannot intuitively show the drug effect.

[0006] The present invention provides a method for constructing a training set, comprising:

[0007] Obtaining whole transcriptome sequencing data, cancer type data, type data of administered platinum drugs, and platinum drug sensitivity data of tumor sample tissues of the target population, wherein the target population includes cancer patient populations, and platinum drug sensitivity data includes platinum drug response data and platinum drug resistance data;

[0008] Obtaining gene expression data of the tumor sample tissue of the target group based on the whole transcriptome sequencing data of the tumor sample tissue of the target group;

[0009] According to the gene expression data of the tumor sample tissue of the target group, the immune cell infiltration level data of the tumor sample tissue of the target group is obtained;

[0010] The training set is constructed using the target population's cancer type data, the type of platinum drugs administered, and the immune cell infiltration level data of tumor sample tissue as sample data, and the target population's platinum drug sensitivity data as label data.

[0011] In one embodiment, the data of the type of platinum drug administered include any one of the following or any combination thereof: cisplatin data, carboplatin data, lobaplatin data, oxaliplatin data.

[0012] In one embodiment, obtaining gene expression data of tumor sample tissues of the target population based on whole transcriptome sequencing data of tumor sample tissues of the target population includes:

[0013] Obtain reference genome data;

[0014] Perform data quality control on the whole transcriptome sequencing data of tumor sample tissues of the target population;

[0015] Compare the whole transcriptome sequencing data after data quality control with the reference genome data to obtain a bam file;

[0016] The gene expression levels were calculated based on the bam file to obtain the gene expression data of the tumor sample tissues of the target group.

[0017] In one embodiment, the data quality control of the whole transcriptome sequencing data of the tumor sample tissue of the target population includes:

[0018] According to the preset data quality control conditions, the whole transcriptome sequencing data of the tumor sample tissue of the target population is subjected to data quality control, wherein the preset data quality control conditions include:

[0019] Filter out reads whose base N content exceeds the preset content threshold;

[0020] Filter out reads whose sequencing quality values ​​are lower than the preset quality value threshold.

[0021] In one embodiment, the preset quality value threshold is 15.

[0022] In one embodiment, obtaining the immune cell infiltration level data of the tumor sample tissue of the target population based on the gene expression data of the tumor sample tissue of the target population includes:

[0023] According to the gene expression data of the tumor sample tissue of the target group, multiple immune cell infiltration level assessment results are obtained through multiple cell infiltration level assessment methods, and then the immune cell infiltration level data of the tumor sample tissue of the target group is obtained, wherein the multiple cell infiltration level assessment methods include any one of the following or any combination thereof: CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER, XCELL.

[0024] In one embodiment, the present invention provides a method for constructing a training set further comprising:

[0025] The training set is cleaned according to the preset cleaning rules, wherein the preset cleaning rules include any one of the following or any combination thereof:

[0026] For each piece of immune cell infiltration level data in the immune cell infiltration level data of the tumor sample tissue of the target population, when the value of the immune cell infiltration level evaluation result of >60% is 0, the immune cell infiltration level data is deleted;

[0027] For the immune cell infiltration level data of tumor sample tissues of the target population obtained by multiple cell infiltration level assessment methods, when ≥80% of the values ​​of the immune cell infiltration level assessment results obtained by a certain cell infiltration level assessment method are all 0, the immune cell infiltration level assessment results obtained by this cell infiltration level assessment method are deleted;

[0028] When the cancer type data represent the same cancer type and the type data of administered platinum drugs represent the same platinum drug, for the platinum drug sensitivity data of the target population, if the platinum drug sensitivity data indicate that the ratio of platinum drug response to platinum drug resistance is outside the preset ratio range, some of the platinum drug sensitivity data are randomly deleted until the ratio of platinum drug response to platinum drug resistance is within the preset range.

[0029] The present invention provides a method for constructing a platinum drug sensitivity prediction model, comprising:

[0030] Obtaining a training set obtained by the training set construction method described in any one of the above items;

[0031] According to the training set, multiple candidate models are trained through multiple machine learning algorithms;

[0032] According to preset model indicators, multiple candidate models are screened to obtain at least two final candidate models as at least two platinum drug sensitivity prediction models.

[0033] In one embodiment, at least two final candidate models include four final candidate models, wherein the four final candidate models are candidate models trained by decision tree (DecisionTree), random forest (RandomForest), adaptive boosting (Adaboost), and boosted tree model (xgboost) machine learning algorithms according to the training set.

[0034] In one embodiment, the preset model indicators include any one of the following items or any combination thereof: area under the ROC curve (AUC), F1 score (F1-score), accuracy (Accuracy), Matthews Correlation Coefficient (MCC).

[0035] The present invention provides a platinum drug sensitivity prediction system, comprising:

[0036] The data receiving module is used to: receive cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient;

[0037] A prediction module, for: obtaining a prediction result of the platinum drug sensitivity of the subject according to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, by combining at least two platinum drug sensitivity prediction models obtained by any of the above-mentioned methods for constructing a platinum drug sensitivity prediction model with a hard voting method;

[0038] The data sending module is used to send the predicted result of the platinum drug sensitivity of the subject to be tested back to at least one client.

[0039] In one embodiment, the prediction module includes:

[0040] A prediction submodule, for obtaining at least two platinum drug sensitivity prediction results for the subject according to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, respectively by using at least two platinum drug sensitivity prediction models obtained by any of the above-mentioned methods for constructing a platinum drug sensitivity prediction model;

[0041] A determination submodule is used to obtain a final platinum drug sensitivity prediction result of the subject according to at least two platinum drug sensitivity prediction results of the subject in combination with a hard voting method, wherein the hard voting method is: when there is a unique platinum drug sensitivity prediction result with the most votes among the at least two platinum drug sensitivity prediction results of the subject, the unique platinum drug sensitivity prediction result with the most votes is used as the final platinum drug sensitivity prediction result of the subject; when there is no unique platinum drug sensitivity prediction result with the most votes among the at least two platinum drug sensitivity prediction results of the subject, when the platinum drug sensitivity prediction model is trained by the construction method of the platinum drug sensitivity prediction model described in any of the above items, the platinum drug sensitivity prediction result of the platinum drug sensitivity prediction model with the highest platinum drug sensitivity prediction accuracy corresponding to the platinum drug type is used as the final platinum drug sensitivity prediction result of the subject.

[0042] The present invention also provides an electronic device, comprising a processor and a memory storing a computer program, wherein when the processor executes the computer program, the method for constructing a training set and the method for constructing a platinum drug sensitivity prediction model described above are implemented.

[0043] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for constructing a training set and the method for constructing a platinum drug sensitivity prediction model described above are implemented.

[0044] The present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute any of the above-mentioned methods for constructing a training set and methods for constructing a platinum drug sensitivity prediction model.

[0045] The present invention provides a method for constructing a platinum drug sensitivity prediction model and its application. Based on the cancer data of the target group, the type of platinum drugs used, the immune cell infiltration level data of tumor sample tissue, and the platinum drug sensitivity data (PTC data), a machine learning algorithm is used to train multiple high-precision platinum drug sensitivity prediction models, which can be used to accurately predict the sensitivity of platinum drugs in combination with the hard voting method. The platinum drug sensitivity prediction model constructed by the present invention has high prediction accuracy, with an average accuracy of more than 80%; the applicable population is more targeted, Chinese cancer patients; the prediction results are simple and clear, divided into platinum drug response and platinum drug resistance; the use cycle is short, starting from the acquisition of the tumor sample tissue of the cancer patient, the prediction results can be obtained within 7 days (including sequencing, analysis, and prediction); the cost is low, and the main cost is the cost of second-generation sequencing, which is decreasing year by year. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A schematic diagram of a process for constructing a platinum drug sensitivity prediction model provided by the present invention.

[0048] Figure 2 A schematic diagram of the structure of a platinum drug sensitivity prediction system provided by the present invention.

[0049] Figure 3 This is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0051] Combine the following Figure 1-Figure 3 The method for constructing a platinum drug sensitivity prediction model provided by the present invention and its application are described. It should be noted that the execution subject of the method for constructing a platinum drug sensitivity prediction model provided by the present invention and its application can be any network-side device / terminal-side device that meets the technical requirements, such as a platinum drug sensitivity prediction device, etc.

[0052] Figure 1 Schematic diagram of the method for constructing a platinum drug sensitivity prediction model provided by the present invention. Figure 1 The present invention provides a method for constructing a platinum drug sensitivity prediction model, which may include:

[0053] Step S110, constructing a training set, wherein the cancer type data of the target population, the type data of the applied platinum drugs, and the immune cell infiltration level data of the tumor sample tissue are used as sample data, and the platinum drug sensitivity data of the target population are used as label data.

[0054] In this embodiment, step S110 may include:

[0055] S1101. Obtain whole transcriptome sequencing data, cancer type data (cancer types may include gastric cancer, colorectal cancer, ovarian cancer, breast cancer, non-small cell lung cancer, oral cancer), type data of platinum drugs used, and platinum drug sensitivity data of tumor sample tissues of the target population, wherein the target population includes a cancer patient population, the platinum drug sensitivity data includes platinum drug response data and platinum drug resistance data, the type data of platinum drugs used includes any one of the following or any combination thereof: cisplatin data, carboplatin data, lobaplatin data, oxaliplatin data, and the platinum drug sensitivity data is obtained by detecting tumor sample tissues of the target population based on a microtumor drug sensitivity detection method (PTC).

[0056] S1102. Obtain gene expression data of the tumor sample tissue of the target group based on the whole transcriptome sequencing data of the tumor sample tissue of the target group. Specifically, in this step, data quality control can be first performed on the whole transcriptome sequencing data of the tumor sample tissue of the target group to filter out reads whose base N content exceeds a preset content threshold and reads whose sequencing quality value is lower than a preset quality value threshold (e.g., 15), and then the whole transcriptome sequencing data after data quality control is compared with the reference genome data to obtain a bam file, and then the gene expression amount (TPM) is calculated based on the bam file to obtain the gene expression data of the tumor sample tissue of the target group.

[0057] S1103. Based on the gene expression data of the tumor sample tissue of the target population, multiple immune cell infiltration level assessment results are obtained by using multiple cell infiltration level assessment methods, and then the immune cell infiltration level data of the tumor sample tissue of the target population is obtained, wherein the multiple cell infiltration level assessment methods include: CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER, and XCELL.

[0058] In this embodiment, the gene expression data (TPM) of the tumor sample tissue of the target group is used to evaluate the immune cell infiltration level of the tumor sample tissue of the target group by seven cell infiltration level evaluation methods, and seven immune cell infiltration level evaluation results of the tumor sample tissue of the target group are obtained. The seven immune cell infiltration level evaluation results are then merged to obtain the final immune cell infiltration level evaluation result of the tumor sample tissue of the target group as the immune cell infiltration level data of the tumor sample tissue of the target group.

[0059] S1104. Perform data cleaning on the training set according to preset cleaning rules, wherein the preset cleaning rules include:

[0060] For each piece of immune cell infiltration level data in the immune cell infiltration level data of the tumor sample tissue of the target population, when >60% of the immune cell infiltration level assessment results have a value of 0, the piece of immune cell infiltration level data is deleted (because one piece of immune cell infiltration level data includes seven immune cell infiltration level assessment results, that is, when >60% of the immune cell infiltration level assessment results have a value of 0 among the seven immune cell infiltration level assessment results, the piece of immune cell infiltration level data is deleted);

[0061] For the immune cell infiltration level data of tumor sample tissues of the target population obtained by multiple cell infiltration level assessment methods, when ≥80% of the values ​​in the immune cell infiltration level assessment results obtained by a certain cell infiltration level assessment method are all 0, the immune cell infiltration level assessment results obtained by this cell infiltration level assessment method are deleted (that is, for all immune cell infiltration level assessment results, if ≥80% of the values ​​in the immune cell infiltration level assessment results obtained by a certain cell infiltration level assessment method are all 0, the immune cell infiltration level assessment results of the tumor sample tissues obtained by this cell infiltration level assessment method are deleted);

[0062] When the cancer type data represent the same cancer type and the type data of administered platinum drugs represent the same platinum drug, for the platinum drug sensitivity data of the target population, if the platinum drug sensitivity data indicate that the ratio of platinum drug response to platinum drug resistance is outside the preset ratio range, some of the platinum drug sensitivity data are randomly deleted until the ratio of platinum drug response to platinum drug resistance is within the preset range.

[0063] S1105. Construct a training set using the target population's cancer type data, the administered platinum drug type data, and the tumor sample tissue's immune cell infiltration level data as sample data, and the target population's platinum drug sensitivity data as label data.

[0064] Step S120: According to the training set, multiple candidate models are trained by multiple machine learning algorithms respectively. The multiple machine learning algorithms may include machine learning algorithms with classification tasks, such as decision tree, naive Bayes, logistic regression, K-nearest neighbor, support vector machine, random forest, adaptive boosting, gradient boosting tree algorithm, etc.

[0065] Step S130: Screen multiple candidate models according to preset model indicators to obtain at least two final candidate models as at least two platinum drug sensitivity prediction models, wherein the preset model indicators include any one of the following items or any combination thereof: area under the ROC curve (AUC), F1 score, accuracy, and Matthews Correlation Coefficient (MCC).

[0066] In this embodiment, four final candidate models are obtained, which are candidate models trained by decision tree (DecisionTree), random forest (RandomForest), adaptive boosting (Adaboost), and gradient boosting tree algorithm (xgboost) machine learning algorithms. In this embodiment, four platinum drug sensitivity prediction models are obtained.

[0067] Among them, the decision tree machine learning algorithm builds a tree structure through a series of feature-based branch nodes. Starting from the root node, the data (including cancer data of the target population, data on the type of platinum drugs used, and data on the level of immune cell infiltration in tumor sample tissues) are gradually divided into different child nodes according to different feature values ​​until the leaf node is reached. Each leaf node corresponds to a category (platinum drug response or platinum drug resistance). It is easy to understand and explain, and can handle multiple types of data (cancer data of the target population, data on the type of platinum drugs used, and data on the level of immune cell infiltration in tumor sample tissues), including numerical and categorical data, with relatively high computational efficiency.

[0068] The Random Forest machine learning algorithm is an integrated learning algorithm that performs prediction and classification based on a combination of multiple decision trees. It has high accuracy and stability, can effectively avoid overfitting, can process high-dimensional data, including cancer type data of the target population, type data of platinum drugs administered, and immune cell infiltration level data of tumor sample tissues, and is insensitive to feature selection; it can evaluate the importance of features.

[0069] The adaptive boosting (Adaboost) machine learning algorithm initializes the data weight distribution in the training set so that the data of each tumor sample tissue has the same weight. In each round of iteration, a weak learner is trained and the error of the weak learner on the training set is calculated. The weight is updated according to the error, and multiple weak learners are weighted and combined to form a strong learner. It can improve the performance of weak learners, is not prone to overfitting, and has a certain robustness to outliers and noise data.

[0070] The core principle of the boosted tree model (xgboost) machine learning algorithm is the gradient boosted tree algorithm, which gradually optimizes the objective function (consisting of loss function and regularization term) by continuously adding new trees to fit the residual of the previous round of models; in the process of building the tree, an accurate greedy algorithm or approximate algorithm is used to select the optimal split node to maximize the gain after splitting. It is efficient, flexible and accurate.

[0071] After the platinum drug sensitivity prediction model is trained, it can be applied to the platinum drug sensitivity prediction system to obtain the predicted results of the subject's platinum drug sensitivity to different platinum drugs based on the subject's cancer type data, the type of platinum drug to be administered (there can be multiple types of platinum drugs to be administered), and the immune cell infiltration level data of the tumor sample tissue, combined with the hard voting method, so as to assist in intuitively and accurately predicting the patient's drug sensitivity, guide personalized treatment of cancer patients, and improve the patient's cure rate.

[0072] The following two examples are used to describe the construction and application of a platinum drug sensitivity prediction model.

[0073] Example 1: Construction of a platinum drug sensitivity prediction model

[0074] Example 1 For tumor tissue samples of 1021 cancer patients (target group) participating in modeling, the operation steps include:

[0075] 1. Use Illumina's NovaSeq X Plus platform to perform LncRNA (whole transcriptome) sequencing, with a sequencing data volume of 30G for each sample, to obtain the whole transcriptome sequencing data of tumor sample tissues of the target population.

[0076] 2. Bioinformatics analysis of sequencing samples: Fastp (v 0.23.2) was used to perform data quality control on the whole transcriptome sequencing data of tumor sample tissues of the target population to generate cleandata (whole transcriptome sequencing data after data quality control). STAR software (v 2.7.10a) was used to compare the cleandata data with the reference genome data (GRCh38) to obtain a bam file.

[0077] 3. Use RSEM (v 1.3.3) to calculate the gene expression through the generated bam file to obtain the TPM value, that is, to obtain the gene expression data of the tumor sample tissue of the target group.

[0078] 4. Use the gene expression data of the tumor sample tissue of the target group and seven cell infiltration level assessment methods (CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER, XCELL) to assess the immune cell infiltration level of the tumor sample tissue of the target group, and obtain the immune cell infiltration level data of the tumor sample tissue of the target group.

[0079] 5. The PTC drug sensitivity reactions of cancer patients are classified into two categories: platinum drug response (indicated by 0) and platinum drug resistance (indicated by 1).

[0080] 6. Integrated data set, the data columns include: cancer patient number, type number of platinum drugs administered, cancer type number, immune cell infiltration level of tumor sample tissue evaluated by different methods, and platinum drug sensitivity classification.

[0081] 7. By dividing the dataset into 70% training set and 30% test set, 10-fold cross validation, and using AUC and F1_score indicators, four final candidate models were selected as four platinum drug sensitivity prediction models.

[0082] The optimal key parameters of the model set in Example 1 when training the four final candidate models are shown in Table 1.

[0083] Table 1 Optimal key parameter settings for the model

[0084]

[0085]

[0086] The training results of the final candidate model in Example 1 are shown in Table 2.

[0087] Table 2 Model training results

[0088]

[0089] After determining the optimal key parameters of the model, Example 1 uses the test set to verify the generalization ability of each final candidate model. The model test accuracy is shown in Table 3.

[0090] Table 3 Model test accuracy

[0091]

[0092] Example 2: The platinum drug sensitivity prediction model constructed in Example 1 was used to predict the platinum drug sensitivity of 209 cancer patients to platinum drugs (cisplatin, carboplatin, lobaplatin, and oxaliplatin).

[0093] The specific steps include:

[0094] 1. Obtain LncRNA sequencing data (i.e., whole transcriptome sequencing data) of tumor sample tissues from 209 cancer patients. The sample sequencing volume is guaranteed to be at least 30G. All 209 cancer patients are on a single medication, including: 84 patients with clinical drug response data, and drug response types are classified according to tag classification; 125 patients have drug sensitivity values ​​measured by PTC, and drug response types are classified according to drug sensitivity data.

[0095] 2. Perform bioinformatics analysis on the whole transcriptome sequencing data in 1, including data quality control, comparison with reference genome data, and quantitative gene expression.

[0096] 3. Using the gene expression data of tumor sample tissues from 209 cancer patients, seven methods (CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER, and XCELL) were used to evaluate the immune cell infiltration level of tumor sample tissues.

[0097] 4. Generate model input data d1, where the column names include the cancer patient number, the type number of the platinum drug to be administered, the cancer type number, and the immune cell infiltration level of the tumor sample tissue evaluated by different methods.

[0098] 5. The model input data d1 is input into the four platinum drug sensitivity prediction models constructed in Example 1, and combined with the hard voting method (the platinum drug sensitivity prediction results of the four platinum drug sensitivity prediction models are combined, and the voting results of the majority of categories are the final results. If there is a tie, the result of the model with the highest corresponding drug prediction accuracy in the test data is selected as the final result), and the platinum drug sensitivity prediction results of 209 cancer patients are obtained.

[0099] 6. The predicted results of platinum drug sensitivity of 209 cancer patients were compared with their actual drug response types. The comparison results of clinical samples and PTC samples are shown in Table 4 and Table 5 respectively:

[0100] Table 4 Comparison of consistency of clinical samples

[0101]

[0102] Table 5 PTC sample consistency comparison

[0103]

[0104] The accuracy of the platinum drug sensitivity prediction model was tested on 209 cases with clinical data and PTC drug sensitivity data. After comparison, the prediction effect of platinum drugs (cisplatin, carboplatin, lobaplatin, and oxaliplatin) was better. Among them, in clinical samples, the accuracy of cisplatin was 82.8%, the accuracy of carboplatin was 90%, the accuracy of lobaplatin was 87.5%, and the accuracy of oxaliplatin was 80.95%; in patients undergoing PTC testing, the accuracy of cisplatin was 86.35%, the accuracy of carboplatin was 89.5%, the accuracy of lobaplatin was 85.75%, and the accuracy of oxaliplatin was 82.25%.

[0105] The present invention provides a method for constructing a platinum drug sensitivity prediction model and its application, which at least has the following advantages:

[0106] Beneficial effects:

[0107] (1) High accuracy;

[0108] It can be seen from the results in Tables 4 and 5 that the platinum drug sensitivity prediction model constructed in the present invention has a very high prediction accuracy (>80%), whether compared with clinical medication data or drug sensitivity data obtained by PTC.

[0109] (2) Short cycle, more economical and efficient;

[0110] Starting from obtaining the tumor sample tissue of the cancer patient, the prediction result can be obtained within 7 days (including sequencing, analysis, and prediction), while it takes 2 weeks to obtain drug sensitivity data through PTC; as the sequencing cost decreases year by year, the cost of drug sensitivity testing based on gene expression data for patients will also decrease year by year.

[0111] (3) The modeling data used is clinical data specifically for Chinese people, which is ethnically specific and has high clinical effect consistency.

[0112] (4) The prediction results are simple and clear, divided into two types: response and resistance, which can be used as an auxiliary guidance for medication.

[0113] The platinum drug sensitivity prediction system provided by the present invention is described below. The platinum drug sensitivity prediction system described below and the platinum drug sensitivity prediction method described above can be referenced to each other.

[0114] Reference Figure 2 The present invention provides a platinum drug sensitivity prediction system, which may include:

[0115] The data receiving module is used to: receive cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient;

[0116] A prediction module, for: obtaining a prediction result of the platinum drug sensitivity of the subject according to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, by combining at least two platinum drug sensitivity prediction models obtained by any of the above-mentioned methods for constructing a platinum drug sensitivity prediction model with a hard voting method;

[0117] The data sending module is used to send the predicted result of the platinum drug sensitivity of the subject to be tested back to at least one client.

[0118] In one embodiment, the prediction module includes:

[0119] A prediction submodule, for obtaining at least two platinum drug sensitivity prediction results for the subject according to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, respectively by using at least two platinum drug sensitivity prediction models obtained by any of the above-mentioned methods for constructing a platinum drug sensitivity prediction model;

[0120] A determination submodule is used to obtain a final platinum drug sensitivity prediction result of the subject according to at least two platinum drug sensitivity prediction results of the subject in combination with a hard voting method, wherein the hard voting method is: when there is a unique platinum drug sensitivity prediction result with the most votes among the at least two platinum drug sensitivity prediction results of the subject, the unique platinum drug sensitivity prediction result with the most votes is used as the final platinum drug sensitivity prediction result of the subject; when there is no unique platinum drug sensitivity prediction result with the most votes among the at least two platinum drug sensitivity prediction results of the subject, when the platinum drug sensitivity prediction model is trained by the construction method of the platinum drug sensitivity prediction model described in any of the above items, the platinum drug sensitivity prediction result of the platinum drug sensitivity prediction model with the highest platinum drug sensitivity prediction accuracy corresponding to the platinum drug type is used as the final platinum drug sensitivity prediction result of the subject.

[0121] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to perform the following steps:

[0122] Receiving cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient;

[0123] According to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, at least two platinum drug sensitivity prediction models obtained by the construction method of the platinum drug sensitivity prediction model described in any one of the above are combined with the hard voting method to obtain the platinum drug sensitivity prediction result of the subject;

[0124] The predicted result of the platinum drug sensitivity of the subject is sent back to at least one client.

[0125] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0126] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and when the computer program is executed by a processor, the computer can perform the following steps:

[0127] Receiving cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient;

[0128] According to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, at least two platinum drug sensitivity prediction models obtained by the construction method of the platinum drug sensitivity prediction model described in any one of the above are combined with the hard voting method to obtain the platinum drug sensitivity prediction result of the subject;

[0129] The predicted result of the platinum drug sensitivity of the subject is sent back to at least one client.

[0130] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented to perform the following steps when executed by a processor:

[0131] Receiving cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient;

[0132] According to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, at least two platinum drug sensitivity prediction models obtained by the construction method of the platinum drug sensitivity prediction model described in any one of the above are combined with the hard voting method to obtain the platinum drug sensitivity prediction result of the subject;

[0133] The predicted result of the platinum drug sensitivity of the subject is sent back to at least one client.

[0134] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0135] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a training set, comprising: Obtaining whole transcriptome sequencing data, cancer type data, type data of administered platinum drugs, and platinum drug sensitivity data of tumor sample tissues of the target population, wherein the target population includes cancer patient populations, and platinum drug sensitivity data includes platinum drug response data and platinum drug resistance data; Obtaining gene expression data of the tumor sample tissue of the target group based on the whole transcriptome sequencing data of the tumor sample tissue of the target group; According to the gene expression data of the tumor sample tissue of the target group, the immune cell infiltration level data of the tumor sample tissue of the target group is obtained; The training set is constructed using the target population's cancer type data, the type of platinum drugs administered, and the immune cell infiltration level data of tumor sample tissue as sample data, and the target population's platinum drug sensitivity data as label data.

2. The method for constructing a training set according to claim 1, characterized in that: The method of obtaining gene expression data of tumor sample tissues of the target population based on the whole transcriptome sequencing data of tumor sample tissues of the target population includes: Obtain reference genome data; Perform data quality control on the whole transcriptome sequencing data of tumor sample tissues of the target population; Compare the whole transcriptome sequencing data after data quality control with the reference genome data to obtain a bam file; The gene expression levels were calculated based on the bam file to obtain the gene expression data of the tumor sample tissues of the target group.

3. The method for constructing a training set according to claim 2, characterized in that: The data quality control of the whole transcriptome sequencing data of the tumor sample tissue of the target population includes: According to the preset data quality control conditions, the whole transcriptome sequencing data of the tumor sample tissue of the target population is subjected to data quality control, wherein the preset data quality control conditions include: Filter out reads whose base N content exceeds the preset content threshold; Filter out reads whose sequencing quality values ​​are lower than the preset quality value threshold.

4. The method for constructing a training set according to claim 3, characterized in that: The step of obtaining the immune cell infiltration level data of the tumor sample tissue of the target group based on the gene expression data of the tumor sample tissue of the target group includes: According to the gene expression data of the tumor sample tissue of the target group, multiple immune cell infiltration level assessment results are obtained through multiple cell infiltration level assessment methods, and then the immune cell infiltration level data of the tumor sample tissue of the target group is obtained, wherein the multiple cell infiltration level assessment methods include any one of the following or any combination thereof: CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER, XCELL.

5. The method for constructing a training set according to claim 4, characterized in that: Also includes: The training set is cleaned according to the preset cleaning rules, wherein the preset cleaning rules include any one of the following or any combination thereof: For each piece of immune cell infiltration level data in the immune cell infiltration level data of the tumor sample tissue of the target population, when the value of the immune cell infiltration level evaluation result of >60% is 0, the immune cell infiltration level data is deleted; For the immune cell infiltration level data of tumor sample tissues of the target population obtained by multiple cell infiltration level assessment methods, when ≥80% of the values ​​of the immune cell infiltration level assessment results obtained by a certain cell infiltration level assessment method are all 0, the immune cell infiltration level assessment results obtained by this cell infiltration level assessment method are deleted; When the cancer type data represent the same cancer type and the type data of administered platinum drugs represent the same platinum drug, for the platinum drug sensitivity data of the target population, if the platinum drug sensitivity data indicate that the ratio of platinum drug response to platinum drug resistance is outside the preset ratio range, some of the platinum drug sensitivity data are randomly deleted until the ratio of platinum drug response to platinum drug resistance is within the preset range.

6. A method for constructing a platinum drug sensitivity prediction model, characterized in that: include: Obtain a training set obtained by the method for constructing a training set according to any one of claims 1 to 5; According to the training set, multiple candidate models are trained through multiple machine learning algorithms; According to preset model indicators, multiple candidate models are screened to obtain at least two final candidate models as at least two platinum drug sensitivity prediction models.

7. The method for constructing a platinum drug sensitivity prediction model according to claim 6, characterized in that: The at least two final candidate models include four final candidate models, wherein the four final candidate models are candidate models trained by decision tree, random forest, adaptive boosting, and boosted tree model machine learning algorithms according to the training set.

8. A platinum drug sensitivity prediction system, characterized in that: include: The data receiving module is used to: receive cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient; A prediction module, used to: obtain a prediction result of the platinum drug sensitivity of the subject according to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, by combining at least two platinum drug sensitivity prediction models obtained by the method for constructing a platinum drug sensitivity prediction model according to claim 6 or 7, with a hard voting method; The data sending module is used to send the predicted result of the platinum drug sensitivity of the subject to be tested back to at least one client.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor performs the following steps: Receiving cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient; According to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, at least two platinum drug sensitivity prediction models obtained by the method for constructing a platinum drug sensitivity prediction model according to claim 6 or 7 are combined with a hard voting method to obtain a platinum drug sensitivity prediction result of the subject; The predicted result of the platinum drug sensitivity of the subject is sent back to at least one client.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is executed by the processor to perform the following steps: Receiving cancer type data of a subject to be tested, type data of platinum drugs to be administered, and immune cell infiltration level data of tumor sample tissue sent by at least one client, wherein the subject to be tested is a cancer patient; According to the cancer type data of the subject, the type data of the platinum drug to be administered, and the immune cell infiltration level data of the tumor sample tissue, at least two platinum drug sensitivity prediction models obtained by the method for constructing a platinum drug sensitivity prediction model according to claim 6 or 7 are combined with a hard voting method to obtain a platinum drug sensitivity prediction result of the subject; The predicted result of the platinum drug sensitivity of the subject is sent back to at least one client.

Images