WO2022170909A1

WO2022170909A1 - Drug sensitivity prediction method, electronic device and computer-readable storage medium

Info

Publication number: WO2022170909A1
Application number: PCT/CN2022/071509
Authority: WO
Inventors: 马少华; 方璐; 范家旗; 冯懿琳; 王旭康; 王子天; 戴琼海
Original assignee: 清华大学深圳国际研究生院
Priority date: 2021-02-09
Filing date: 2022-01-12
Publication date: 2022-08-18
Also published as: CN112951327A; CN112951327B

Abstract

Disclosed are a drug sensitivity prediction method, an electronic device and a computer-readable storage medium, which relate to the technical field of drug testing. The method comprises: acquiring gene sequencing data of training cancer cell tissue and drug feature data; pre-processing the gene sequencing data according to the drug feature data, so as to obtain gene sample data; performing verification processing according to the gene sample data and the drug feature data, so as to obtain a prediction model and a gene prediction list; and by means of the gene prediction list and the prediction model, performing drug sensitivity prediction on cancer cell tissue to be tested. Therefore, drug reaction prediction of a clinic patient can be quickly and precisely realized, thereby reducing prediction costs and time costs, and improving the drug effect prediction efficiency.

Description

Drug susceptibility prediction method, electronic device and computer-readable storage medium

technical field

The present application relates to the technical field of drug detection, and in particular, to a drug sensitivity prediction method, an electronic device and a computer-readable storage medium.

Background technique

In the era of precision medicine, the prediction of drug responsiveness in cancer patients based on their clinical characteristics and genomics is essential to assist clinicians in formulating effective and less toxic treatment regimens. Predictive models for drug response are often trained on different datasets. Currently, the most widely used drug prediction models are based on supervised learning techniques. The supervised learning methods used include regression models and classification models. The former can generate specific drug sensitivity values, such as IC50 (The half maximal inhibition concentration), and the latter can generate drug response levels, such as high-sensitivity drug responses and low-sensitivity drug responses.

At present, there are a number of studies and methods dedicated to discovering the relationship between the genome/transcriptome and the effect of cancer medication, so as to assist the cancer medication regimen and improve the efficacy of cancer medication. However, the current research and solutions are still far from practical application and cannot be efficiently applied to clinical scenarios. For example, there are certain deficiencies regarding the use of supervised learning to predict drug responsiveness from the genome or transcriptome: data analysis is limited to existing databases, lacking experimental and clinical validation; methods are based on RNA sequencing technology rather than small gene sets , it is impossible to use rapid gene expression measurement methods, and RNA sequencing takes several days to several weeks, which is not suitable for the situation of intraoperative or immediate postoperative medication that is often required in clinical practice; drug efficacy prediction only stops at data analysis, and no suggestion is made. The specific and fast application scheme is difficult to apply in practice, with high cost and long time.

SUMMARY OF THE INVENTION

The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a drug sensitivity prediction method, which can quickly and accurately predict drug responsiveness to clinical patients, reduce prediction costs and time costs, and improve drug efficacy prediction efficiency.

The present application also proposes an electronic device having the above drug sensitivity prediction method.

The present application also provides a computer-readable storage medium having the above drug sensitivity prediction method.

The drug sensitivity prediction method according to the embodiment of the first aspect of the present application includes: acquiring gene sequencing data and drug characteristic data of the cancer cell tissue to be trained; preprocessing the gene sequencing data according to the drug characteristic data to obtain a gene sample data; perform verification processing according to the gene sample data and the drug characteristic data to obtain a prediction model and a gene prediction list; perform drug sensitivity prediction on the cancer cell tissue to be tested by using the prediction model and the gene prediction list.

The drug sensitivity prediction method according to the embodiment of the present application has at least the following beneficial effects: by acquiring the gene sequencing data and drug characteristic data of the cancer cell tissue to be trained, and preprocessing the gene sequencing data according to the drug characteristic data to obtain gene sample data, Validation processing is carried out according to the gene sample data and drug characteristic data to obtain a prediction model and a gene prediction list. Through the gene prediction list and prediction model, the drug sensitivity of the cancer cell tissue to be tested can be predicted, which can quickly and accurately realize the drug responsiveness of clinical patients. Prediction, reduce prediction cost and time cost, and improve the efficiency of drug efficacy prediction.

According to some embodiments of the present application, the gene sequencing data includes first sequencing data, and the drug characteristic data includes drug sensitivity data; correspondingly, the acquiring gene sequencing data and drug characteristic data of the cancer cell tissue to be trained, The method includes: acquiring the first sequencing data of the cancer cell tissue to be trained and the corresponding drug sensitivity data based on a genome database.

According to some embodiments of the present application, the preprocessing of the gene sequencing data according to the drug characteristic data to obtain the gene sample data includes: standardizing the first sequencing data to obtain the first sample data ; Screen the first sample data according to the drug sensitivity correlation coefficient of the first sample data and the drug sensitivity data to obtain second sample data; The sample data is scored and determined to obtain scoring parameters of the second sample data; the second sample data is screened based on the scoring parameters to obtain the gene sample data.

According to some embodiments of the present application, performing the verification process according to the gene sample data and the drug characteristic data to obtain a prediction model and a gene prediction list includes: acquiring the difference between the gene sample data and the drug sensitivity data. Drug susceptibility correlation coefficient, obtain the scoring parameters of the gene sample data, the gene sample data includes multiple gene fragments; according to the drug susceptibility correlation coefficient and the scoring parameters, the multiple gene fragments are sorted in descending order; The plurality of gene fragments arranged in descending order are subjected to verification processing to obtain the model parameters of the prediction model and the number of gene lists; the gene prediction list is generated according to the number of gene lists, and the prediction model is determined according to the model parameters. .

According to some embodiments of the present application, the gene sequencing data includes second sequencing data, and the drug feature data includes drug effect classification data; correspondingly, the acquiring gene sequencing data and drug feature data of the cancer cell tissue to be trained, Including: acquiring second sequencing data and drug effect grading data of the cancer cell tissue to be trained based on the genome atlas database.

According to some embodiments of the present application, the preprocessing of the gene sequencing data according to the drug characteristic data to obtain gene sample data includes: standardizing the second sequencing data to obtain third sample data; The third sample data is tested according to the drug effect classification data to obtain the gene sample data.

According to some embodiments of the present application, the performing verification processing according to the gene sample data and the drug characteristic data to obtain a prediction model and a gene prediction list includes: acquiring gene scores of multiple gene segments of the gene sample data ; Arrange the plurality of gene fragments in descending order according to the gene score; perform cross-validation on the plurality of gene fragments after the descending order to obtain the model parameters of the prediction model and the number of gene lists; According to the gene list The number and the corresponding plurality of gene segments generate the gene prediction list, and the prediction model is determined according to the model parameters.

According to some embodiments of the present application, the predicting the drug sensitivity of the cancer cell tissue to be tested by using the prediction model and the gene prediction list includes: obtaining the corresponding information of the cancer cell tissue to be tested according to the gene prediction list obtain the gene expression of the gene fragment; input the gene expression into the prediction model to obtain the drug sensitivity result of the cancer cell tissue to be tested.

An electronic device according to an embodiment of the second aspect of the present application includes: at least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions, the instructions are stored by the at least one processor A processor executes such that the at least one processor implements the drug susceptibility prediction method according to the first aspect when the at least one processor executes the instructions.

The electronic device according to the present application has at least the following beneficial effects: by executing the drug sensitivity prediction method mentioned in the embodiment of the first aspect, the drug responsiveness prediction for clinical patients can be quickly and accurately realized, and the prediction cost and time cost can be reduced, Improve the efficiency of drug efficacy prediction.

A computer-readable storage medium according to an embodiment of the third aspect of the present application, where the computer-readable storage medium stores computer-executable instructions for causing a computer to execute the drug-sensing method according to the first aspect method of prediction.

The computer-readable storage medium according to the present application has at least the following beneficial effects: by executing the drug sensitivity prediction method mentioned in the embodiment of the first aspect, the drug responsiveness prediction for clinical patients can be quickly and accurately realized, and the prediction cost and Time cost and improve the efficiency of drug efficacy prediction.

Additional aspects and advantages of the present application will be set forth, in part, from the following description, and in part will become apparent from the following description, or may be learned by practice of the present application.

Description of drawings

Fig. 1 is a specific flow chart of the drug sensitivity prediction method in the embodiment of the application;

FIG. 2 is a specific flowchart of step S200 of the drug sensitivity prediction method in the embodiment of the present application;

FIG. 3 is another specific schematic flowchart of step S200 of the drug sensitivity prediction method in the embodiment of the present application;

FIG. 4 is a specific flowchart of step S300 of the drug sensitivity prediction method in the embodiment of the present application;

5 is another specific schematic flowchart of step S300 of the drug sensitivity prediction method in the embodiment of the present application;

FIG. 6 is a specific flowchart of step S400 of the drug sensitivity prediction method in the embodiment of the present application;

FIG. 7 is a diagram of a specific application example of the drug sensitivity prediction method in the embodiment of the present application.

Detailed ways

The following describes in detail the embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, but should not be construed as a limitation on the present application.

It should be noted that, in the flowcharts, a logical order is shown, but in some cases, the steps shown or described may be performed in the order in a different flowchart. If it refers to "several", it means more than one, if it refers to "plurality", it means two or more, if it refers to "the following", it should be understood as including the number. The use of any and all examples or exemplary language ("for example," "such as," etc.) provided herein is merely intended to better illustrate the embodiments of the present application and does not impose limitations on the scope of the present application unless otherwise required . Greater than, less than, exceeding, etc. are understood as not including this number, and above, below, within, etc. are understood as including this number. If it is described that the first and the second are only for the purpose of distinguishing technical features, it cannot be understood as indicating or implying relative importance, or indicating the number of the indicated technical features or the order of the indicated technical features. relation.

It should be noted that, unless otherwise specified, the singular forms of "a", "the" and "the" used in the embodiments are also intended to include plural forms unless the context clearly indicates otherwise. Also, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The terms used in the specification herein are for the purpose of describing specific embodiments only, and not for the purpose of limiting the application. As used herein, the term "and/or" includes any combination of one or more of the associated listed items.

Based on this, the embodiments of the present application provide a drug sensitivity prediction method, an electronic device and a computer-readable storage medium, which can quickly predict the drug efficacy of cancer based on a small number of genes, avoiding the need for drug efficacy prediction to RNA sequencing, etc. Dependence on time-consuming sequencing technology, and reduce the cost of drug efficacy prediction.

In a first aspect, the embodiments of the present application provide a drug sensitivity prediction method.

In some embodiments, referring to FIG. 1 , a schematic flowchart of the drug sensitivity prediction method in the embodiment of the present application is shown. It specifically includes steps:

S100, acquiring gene sequencing data and drug characteristic data of the cancer cell tissue to be trained;

S200, preprocessing the gene sequencing data according to the drug characteristic data to obtain gene sample data;

S300, performing verification processing according to the gene sample data and drug characteristic data to obtain a prediction model and a gene prediction list;

S400, predicting the drug sensitivity of the cancer cell tissue to be tested through a prediction model and a gene prediction list.

In step S100, in the embodiment of the present application, it is necessary to obtain the gene sequencing data of the cancer cell tissue to be trained and the corresponding drug characteristic data of different drugs, wherein the gene sequencing data refers to the RNA (ribonucleic acid, Ribonucleic Acid) of the cancer cell tissue to be trained. ) sequencing data; the drug characteristic data of the drug to be tested refers to the sensitivity data or drug effect data of different drugs applied to the cancer cell tissue to be trained, such as the IC50 (half of the drug sensitivity of the drug related to the cancer cell tissue to be trained) maximal inhibition concentration) data, IC50 is the concentration corresponding to 50% inhibitory concentration, and half inhibition is used to measure the sensitivity of the antibody; the lower the IC50 value, the higher the sensitivity of the antibody; for example, about cancer cells. The drug effect classification data, and the clinical effect classification data are used to represent the effect of clinical drug use of the cancer cell tissue, and have different effect grades.

In some embodiments, the cancer cell tissue to be trained in the embodiments of the present application may be any cancer cell tissue selected from the gene database; it may also be a cancer cell tissue sample of a clinical patient obtained from the gene database; the The cancer cell tissue to be trained is used to provide training data for the subsequent establishment of the prediction model.

Taking the selected cancer cell tissue to be trained as an example, the gene sequencing data of the cancer cell tissue and the drug characteristic data of the drug to be tested can be obtained based on the genome database, wherein the genome database is the anticancer drug sensitivity genomics database (Genomics). of Drug Sensitivity in Cancer, GDSC) and Cancer Cell Line Encyclopedia (CCLE). Specifically, by consulting the anticancer drug sensitivity genomics database and the cancer cell line encyclopedia, the required relevant information, that is, the gene sequencing data of the cancer cell tissue and the drug characteristic data of the drug to be tested, are obtained.

The Genomics of Drug Sensitivity in Cancer (GDSC) database was developed by the Sanger Institute in the United Kingdom to collect the sensitivity and response of tumor cells to the drug to be tested. Variations in the cancer genome can affect the efficacy of clinical treatments, and different targets respond to drugs differently. Such data are therefore important for the discovery of potential tumor therapeutic targets. Data from GDSC comes from 75,000 experiments describing the response of about 200 anticancer drugs in more than 1,000 tumor cells. The oncogenome mutation information in this database comes from the COSMIC database, including oncogene point mutations, gene amplification and loss, tissue types, and expression profiles. Users can search the database from three levels: compound, oncogene and cell line. The response of oncogene or cell line to different drugs will be listed in detail, and the results will be displayed in a graphical interface, including statistical analysis, volcano Figures and related literature. The search results, as well as the entire database, can be downloaded by the user for subsequent analysis.

The Encyclopedia of Cancer Cell Lines integrates genetic information such as DNA mutations, gene expression, and chromosomal copy number through large-scale deep sequencing of 947 human cancer cell lines covering more than 30 tissue sources.

Search directly through the anticancer drug sensitivity genomics database and cancer cell line encyclopedia to obtain the first sequencing data corresponding to the cancer cell tissue and the corresponding drug sensitivity data, where the first sequencing data refers to the RNA sequencing of the cancer cell tissue Data, RNA sequencing data is the data obtained by RNA-seq (transcriptome sequencing) technology, and the transcriptome refers to the collection of all transcriptome products in a cell under a certain physiological condition. The research object of transcriptome sequencing is the sum of all RNAs that can be transcribed by a specific cell in a functional state, mainly including mRNA and ncRNA. Drug sensitivity data refers to IC50 data about the cancer cell tissue-related drug.

Taking the cancer cell tissue samples of clinical patients as an example, the second sequencing data and drug effect classification data corresponding to the cancer cell tissue samples of clinical patients were obtained based on The Cancer Genome Atlas (TCGA), among which the tumor genome atlas The database contains clinical data, genomic variation, mRNA (messenger RNA) expression, miRNA (micro RNA) expression, methylation and other data of various human cancers (tumors including subtypes), which are very important to cancer researchers. Data Sources.

In step S200, the acquired gene sequencing data of the cancer cell tissue is preprocessed according to the acquired drug characteristic data to obtain preprocessed gene sample data.

In some embodiments, taking the selection of cancer cells to be predicted as an example, referring to FIG. 2 , step S200 specifically includes the steps:

S211, standardize the first sequencing data to obtain first sample data;

S212, screening the first sample data according to the drug sensitivity correlation coefficient of the first sample data and the drug sensitivity data to obtain second sample data;

S213, scoring and determining the second sample data according to the drug sensitivity data to obtain scoring parameters of the second sample data;

S214: Perform screening processing on the second sample data based on the scoring parameter to obtain gene sample data.

In step S211, standardize the acquired first sequencing data to obtain first sample data, wherein the normalization processing refers to standardizing the gene length and sequencing of the first sequencing data, that is, the RNA sequencing data of the cancer cell tissue to be trained. depth, the TPM corresponding to the first sequencing data can be obtained (Transcripts Per Kilobase of exon model per Million mapped reads, the number of transcripts per kilobase of transcription per million mapped reads), and then according to the TPM of the first sequencing data The first sequencing data is screened to obtain the first sample data, for example, the first sequencing data whose TPM is lower than 1 is screened to obtain the screened first sequencing data, that is, the first sample data. In practical applications, the first sequencing data are multiple gene fragments. By standardizing the first sequencing data, the first sequencing data can be screened based on the gene expression levels of multiple genes, and those with TPM lower than 1 can be screened out. Gene fragments, retain gene fragments with a TPM higher than 1.

In step S212, the first sample data is screened according to the drug sensitivity correlation coefficient between the first sample data and the drug sensitivity data of the corresponding drug to be tested to obtain second sample data. The drug susceptibility correlation coefficient is the Pearson correlation coefficient (Pearson correlation coefficient) between the TPM of each gene in the first sample data and the drug related to the first sample data, that is, the IC50 data of a drug to be tested, where Pearson The correlation coefficient is used to measure the degree of correlation between two variables, and its value is between -1 and 1, where the two variables are the TPM of the gene and the IC50 data of the sample drug. By calculating the Pearson correlation coefficient of the TPM of the first sample data and the drug sensitivity data of the corresponding drug, that is, the IC50 data, the first sample data whose absolute value of the Pearson correlation coefficient is lower than 0.1 is screened out, that is, the Pearson correlation. For gene segments whose absolute value of coefficient coefficient is lower than 0.1, the second sample data is obtained.

In steps S213 and S214, after screening the first sample data through the drug sensitivity correlation coefficient to obtain the second sample data, the second sample is scored and determined according to the drug sensitivity data of the relevant drug, and the score of the second sample data is obtained. parameter, the second sample data is screened by the scoring parameter, and the screened gene sample data is obtained. For example, scoring is determined based on Fisher's linear discriminant method. By calculating the mean and standard deviation of the gene expression levels of some cancer cells to which the drug to be tested is applied, the scoring parameters are calculated based on the calculated values and standard deviations. The calculated scoring parameters are used to screen genes corresponding to cancer cells to obtain the screened genes, that is, gene sample data. In practical applications, the gene sample data is a collection of multiple gene fragments. In practical applications, the drug sensitivity data of the drug, that is, the mean E1 and the standard deviation STD1 of the gene expression levels of 15% of the gene fragments in the cancer cell tissue line with the highest IC50 data, are calculated, and then the drug sensitivity data of the drug are calculated, namely The mean E2 and standard deviation STD2 of the gene expression levels of 15% of the gene fragments in the cancer cell tissue with the lowest IC50 data, according to the calculated mean E1, mean E2, standard deviation STD1 and standard deviation STD2 of the gene expression levels, through the formula (E1-E2)/(STD1+STD2) is calculated to obtain the scoring parameter, and the second sample data with the highest scoring parameter is retained as the gene sample data obtained by screening, wherein the gene sample data includes several gene fragments, the gene fragment The selection of the number of s can be set according to actual needs, so as to filter the second sample data according to the number.

In some embodiments, taking the cancer cell tissue of a clinical patient as an example, referring to FIG. 3 , step S200 specifically includes the steps:

S221, standardize the second sequencing data to obtain third sample data;

S222, test the third sample data according to the drug effect classification data to obtain gene sample data.

In step S221, normalize the acquired second sequencing data to obtain third sample data, wherein the normalization processing refers to normalizing the second sequencing data, that is, the RNA sequencing data of the cancer cell tissue, by normalizing the gene length, and then normalizing the sequencing depth, The TPM corresponding to the second sequencing data can be obtained, and then the second sequencing data is screened according to the TPM of the second sequencing data to obtain third sample data. The second sequencing data is the third sample data. In practical applications, the second sequencing data are multiple gene fragments. By standardizing the first sequencing data, the second sequencing data can be screened based on the expression levels of multiple gene fragments, and those with TPM lower than 1 can be screened out. Gene fragments, retain gene fragments with a TPM higher than 1.

In step S222, the third sample data is inspected according to the graded data of the drug effect of the drug to be tested on the cancer cell tissue, and the gene sample data after inspection processing is obtained. Specifically, the third sample data is tested based on the Mann-Whitney U test method, the third sample data is divided into valid data or invalid data according to the drug effect classification data, and the third sample data corresponding to the valid data is calculated. The gene expression amount of the gene fragment and the gene expression amount of the gene fragment in the third sample data corresponding to the invalid data are taken as the calculated data value, and the gene fragment whose data value is less than a certain value, for example, the gene fragment less than 0.1, is used as the calculated data value. Genetic sample data. The grading data of the drug effect of the drug to be tested against a certain cancer cell tissue obtained from The Cancer Genome Atlas includes various data, such as "complete remission", "partial remission", "stable disease", "disease progression", etc. Among them, "complete remission", "partial remission", "stable disease" can be defined as "effective", and "disease progression" can be defined as "ineffective", then the third sample data can be divided into effective sample data or ineffective according to the drug effect classification data sample.

In step S300, verification is performed according to the gene sample data obtained by preprocessing and the drug characteristic data of the drug to be tested, and a prediction model and a gene prediction list of the drug to be tested are obtained. The prediction model of the drug to be tested refers to the verification based on a preset mathematical model, and the optimal parameters of the prediction model are obtained; the gene prediction list refers to the prediction of drug sensitivity of the drug to be tested in cancer cells and plays a key role in prediction Gene fragments that act.

In some embodiments, taking the selection of cancer cells to be predicted as an example, referring to FIG. 4 , step S300 specifically includes the steps:

S311, obtaining the drug sensitivity correlation coefficient between the gene sample data and the drug sensitivity data, and obtaining the scoring parameter of the gene sample data;

S312, arranging the multiple gene fragments in descending order according to the drug susceptibility correlation coefficient and the scoring parameter;

S313, performing verification processing on the plurality of gene fragments arranged in descending order to obtain model parameters of the prediction model and the number of gene lists;

S314: Generate a gene prediction list according to the number of gene lists, and determine a prediction model according to model parameters.

In steps S311 and S312, the drug susceptibility correlation coefficients corresponding to the multiple gene fragments in the gene sample data and the drug sensitivity data of the drug are obtained, and the scoring parameters of the gene sample data are obtained, wherein the drug susceptibility correlation coefficient refers to the The drug susceptibility correlation coefficient obtained in step S212, and the scoring parameter refers to the scoring parameter obtained in step S213. By combining the drug susceptibility correlation coefficient and the scoring parameter, under a certain weight distribution, each gene in the gene sample data is calculated. The scores corresponding to the fragments are sorted in descending order according to the calculated scores of each gene fragment. In practical applications, the drug susceptibility correlation coefficient is set as S1, the corresponding weight is 0.3, the scoring parameter, namely the Fisher discrimination score, is S2, and the corresponding weight is 0.7, then the score of the gene fragment is calculated as S=0.3*S1 +0.7*S2. Arrange in descending order according to the calculated gene fragments to obtain the gene sample data after descending order.

In step S313 and step S314, verification processing is performed on a plurality of gene segments in the gene sample data sorted in descending order to obtain model parameters of the prediction model and the number of gene lists. The verification process refers to sequentially selecting the first n gene fragments in the arranged gene samples, where the value of n can be set according to actual needs, for example, the value range of n is set to be 10 to 30 genes. Based on the K nearest neighbor regression model, enumerate the model parameters k of the regression model, select the nearest k neighbors, predict the drug sensitivity data corresponding to the cancer cell tissue to be trained, and perform 5-fold cross-validation to obtain the prediction result. In practical applications, based on the acquired gene sample data, a prediction model for a certain drug of the cancer tissue cells to be trained is established according to the K-nearest neighbor algorithm. The prediction model has the optimal model parameters, that is, has the optimal k value. , and according to the number of gene fragments in the specific gene sample data, that is, the value of n, for example, the first n gene fragments can obtain the optimal model parameters of the prediction model, then the first n gene fragments in the gene sample data constitute gene prediction. list. In a possible application example, the acquisition of model parameters corresponds to the case where the area under the curve (Area under curve, AUC) of the receiver operating characteristic curve (ROC) obtained after cross-validation based on the prediction model corresponds to the largest The number of gene fragments n and the proximity parameter k are determined as the final model parameters.

In a possible application example, taking the cancer cell tissue to be trained as a colorectal cancer cell line as an example, the GDSC database contains the first sequencing data of each cell line, that is, RNA sequencing data and the IC50 of the cell line under the action of different drugs According to the data, the drugs are selected from four chemotherapeutic drugs, paclitaxel, 5-fluorouracil, cyclophosphamide, and cisplatin, and the above conditions are exemplified.

The RNA sequencing data of colorectal cancer cell lines and the IC50 data of paclitaxel were obtained from the GDSC database. After preprocessing the RNA sequencing data, the RNA sequencing data were scored and ranked; When the k value of the K nearest neighbor regression model is 1 to 30, the IC50 data of paclitaxel is predicted, and the prediction results are cross-validated, the AUC value is calculated, and the maximum AUC value obtained by different K values is recorded, and the AUC The k value corresponding to the maximum value; then select the top 11 genes after the gene score sorting, and predict the IC50 data of paclitaxel for the cases where the k value of the K nearest neighbor regression model is 1 to 30, and cross the prediction results. Verify, calculate the new AUC value, record the new maximum AUC value obtained by different K values, and the corresponding k value when the new AUC value is the largest, for the case where the number n of the gene selection is 10 to 30 Repeat the above operations, and finally obtain the maximum AUC value and the n and k values corresponding to the AUC value. For the n _max and k _max as the model parameters of the K-nearest neighbor regression model. And according to the obtained n _max , it can be determined that the gene prediction list for drug susceptibility prediction of paclitaxel includes the top n _max genes ranked by gene score, and the optimal model parameter km _max of the K-nearest neighbor regression model.

The above operations are repeated for the other three drugs, 5-fluorouracil, cyclophosphamide, and cisplatin, to obtain the gene prediction list of each drug and the corresponding optimal parameter km _max of the K-nearest neighbor regression model. It can be seen that the corresponding prediction model and the optimal parameter k _max corresponding to the prediction model can be constructed and obtained for the four compound drugs, and there are corresponding gene prediction lists for the four kinds of compound drugs; The gene prediction lists of the four compound drugs are aggregated into a large gene prediction list set. When the cancer cell tissue to be tested needs to be predicted, the corresponding multiple key gene fragments can be extracted directly according to the gene prediction list set. Key gene fragments are not only specific to a single compound, thus ensuring data sufficiency.

When it is necessary to predict the drug susceptibility of a new colorectal cancer cell line to these four drugs, the K-nearest neighbor regression model established above can be used to predict its drug susceptibility, and the IC50 value of the drug effect can be predicted. Therefore, the drug reaction conditions corresponding to each drug can be judged, and an appropriate medication plan can be efficiently formulated according to the respective drug reaction conditions.

In some embodiments, taking the cancer cell tissue of a clinical patient as an example, referring to FIG. 5 , step S300 specifically includes the steps:

S321, obtaining gene scores of multiple gene segments of the gene sample data;

S322, arranging the multiple gene fragments in descending order according to the gene score;

S323, performing cross-validation on the plurality of gene fragments arranged in descending order to obtain model parameters of the prediction model and the number of gene lists;

S324: Generate a gene prediction list according to the number of gene lists and the corresponding multiple gene segments, and determine a prediction model according to model parameters.

In steps S321 and S322, the gene scores of multiple gene fragments in the gene sample data are obtained, wherein the gene scores are the P values of the gene fragments calculated by the Mann-Whitney U test method mentioned in step S221 The opposite numbers of , and according to the size of the obtained gene scores, the multiple gene fragments are sorted in descending order.

In step S323, cross-validation is performed on the plurality of gene fragments arranged in descending order to obtain the model parameters of the prediction model and the number of gene lists, which are used as model parameters of a drug efficacy prediction model for a drug targeted by the plurality of gene fragments. The prediction model refers to the K-nearest neighbor classification model for predicting the efficacy of a drug on cancer cells of a clinical patient. The model has the optimal model parameters for the cancer tissue, and the model parameters include the optimal neighbor parameters and Gene fragment parameters. Specifically, the first n gene fragments in descending order are selected in sequence, and the value of n can be selected according to actual needs, for example, the value of n is 10 to 30, and the parameter k of the K nearest neighbor classification model is enumerated, and the parameter k Indicates that k neighboring points are selected, and the value of k can be selected according to actual needs. For example, the value of k is 1 to 30. According to the K nearest neighbor regression model corresponding to the enumerated parameter k, predict whether the drug is effective or not. ” or “invalid” to do 5-fold cross-validation, and calculate the accuracy and F1-score of the new prediction results according to the new prediction results obtained after cross-validation. Determine the accuracy rate obtained by each parameter k and n and the model parameters k and n corresponding to the largest value in the F1 score as the optimal model parameters of the prediction model, and the first n gene segments form a gene prediction list. The F1 score is an indicator used in statistics to measure the accuracy of the binary classification model.

In a possible application example, the above conditions are exemplified by taking the cancer cell tissue sample of a clinical patient as a colorectal cancer cell line and the drug to be tested as 5-fluorouracil.

The RNA sequencing data of colorectal cancer cell line samples from clinical patients and the efficacy grading data of 5-fluorouracil were obtained from the TCGA database. After standardizing the RNA sequencing data, genes with TPM lower than 1 were discarded, and the calculation was effective. The data value of the gene expression level corresponding to each gene fragment between the sample data and the invalid sample data, the P value obtained by the Mann-Whitney U test method, and the gene fragments are sorted in ascending order according to the size of the P value, and the P value is filtered out. For gene fragments with a value greater than 0.1, gene fragments with a P value of less than 0.1 were retained, and valid data or invalid data in the RNA sequencing data were marked according to the drug effect classification data.

Taking the value of n from 10 to 30 as an example, first take the top 10 gene fragments in colorectal cancer cells sorted according to the gene score, and use the K nearest neighbor classification model to predict 5- The IC50 data value of fluorouracil, and 5-fold cross-validation was performed for the IC50 data value, and the prediction result of 5-fluorouracil on colorectal cancer cell samples was found to be valid or invalid, and the K nearest neighbor classification corresponding to each model parameter was calculated through the prediction result. The accuracy of the prediction results of the model, record the maximum accuracy, and the parameter k and n values corresponding to the maximum accuracy; then go to the top 11 genes ranked by gene score, and repeat the enumeration of the parameter k of the K-nearest neighbor classification model If it is equal to 1 to 30, re-record the new maximum accuracy value and the corresponding parameter k and n values, or the current maximum accuracy is greater than the maximum accuracy obtained when n is equal to 10, then record the new accuracy. rate data and the corresponding parameter k value and n value, if it is not greater than that, there is no need to record; and so on, repeat the above operation for the case where n is equal to 10 to 30, and finally obtain the maximum accuracy in all cases and the corresponding n _max and k _max values. For example, for the prediction results of colorectal cancer and 5-fluorouracil, the obtained value of _nmax is 15 and the value of _kmax is 5, then the first 15 gene segments are selected as the gene prediction list, that is, the key prediction gene list.

In this embodiment of the present application, by executing step S300, a gene set required for drug efficacy prediction, that is, a gene prediction list, can be obtained, so that when applied to actual drug efficacy prediction, a small number of genes can be obtained to realize the prediction of drug efficacy. It can improve the prediction speed and reduce the prediction cost; and avoid the dependence of drug efficacy prediction on long-term sequencing technologies such as RNA sequencing, which can quickly predict the cancer drug efficacy of patients, and is suitable for individualized medicine during or after surgery. .

In step S400, drug sensitivity prediction is performed by using the generated prediction model and gene prediction list, for example, drug sensitivity prediction of a certain drug to be tested with respect to a certain cancer cell tissue. Specifically, through the established prediction model and the key genes in the gene prediction list, rapid drug sensitivity prediction of cancer cells is performed.

In some embodiments, referring to FIG. 6 , step S400 further includes:

S410, obtaining gene fragments corresponding to the cancer cell tissue to be tested according to the gene prediction list;

S420, obtaining the gene expression level of the gene fragment;

S430, the gene expression level is input into the prediction model, and the drug sensitivity result of the cancer cell tissue to be tested is obtained.

In step S410 and step S420, according to the gene prediction list obtained by executing step S300, corresponding key gene fragments are extracted from the cancer cell tissue to be tested, and the extracted number of the gene fragments is selected according to the number in the gene prediction list; After the corresponding gene fragments are extracted from cancer cell tissue based on qPCR technology or gene chip technology, the gene expression of each gene fragment can be quickly measured.

In step S430 and step S440, the gene expression of each gene fragment is input into the prediction model after the standardization process performed by the cancer cell tissue to be trained, and output the corresponding prediction model of the cancer cell tissue to be tested. The prediction result of whether a drug is effective or not, the prediction result represents the drug sensitivity result of the drug corresponding to the prediction model to the current cancer cell tissue. It should be noted that, in practical applications, by performing steps S100 to S300, a plurality of different prediction models can be established, and different prediction models are established for different drugs to be tested, and the selection of drugs to be tested is for cancer. The type of cell tissue is selected. For example, in the examples of this application, colorectal cancer cell line is selected as the type of cancer cell tissue, and paclitaxel, 5-fluorouracil, cyclophosphamide, and cisplatin are selected for the colorectal cancer cell line. For chemotherapeutic drugs, steps S100 to S300 are respectively performed for the four chemotherapeutic drugs to generate a prediction model and a gene prediction list, and each chemotherapeutic drug corresponds to its own prediction model and gene prediction list. When it is necessary to predict the effect of a chemotherapeutic drug on clinical patients or on colorectal cancer cell lines, the corresponding prediction model is selected, the gene samples of the extracted cancer cells are input, and the corresponding gene expression levels are obtained. Predict drug effects. It can predict whether a certain chemotherapeutic drug is effective in predicting cancer cells of clinical patients; it can also predict the drug effect of a chemotherapeutic drug on cancer cells of the same line, that is, the IC50 data value.

In a possible application example, taking the prediction of IC50 data of a drug on a cancer cell line as an example, in the examples of this application, the RNA sequencing data of the cancer cell line to be trained in the GDSC and CCLE databases and the data indicating drug sensitivity are used as an example. IC50 data, preprocessing the acquired RNA sequencing data, the preprocessing includes filtering based on gene expression, filtering based on the correlation between gene expression and IC50 data, and filtering through Fisher's linear judgment, and finally retain some gene fragments ; Use the K-nearest neighbor regression model for cross-validation, enumerate the parameters of the K-nearest neighbor regression model, select the model parameters with the highest cross-validation accuracy, determine the optimal parameters of the prediction model and construct the generated gene prediction list; When the drug sensitivity of tissue is predicted, the key gene fragments of cancer cells can be obtained according to the gene prediction list, the gene expression of key gene fragments can be obtained by qPCR technology or gene chip technology, and the gene expression can be used as the input parameter of the prediction model. , and the prediction result of the drug sensitivity of the cancer cell tissue is obtained, that is, the predicted IC50 data value.

In a possible application example, taking the prediction of whether a drug is effective on cancer cells of clinical patients as an example, the RNA sequencing data of cancer tissue samples of clinical patients and the drug effect classification data of corresponding clinical drugs are obtained through the TCGA database. ; Preprocess the acquired RNA sequencing data, which includes filtering based on gene expression and filtering through the Mann-Whitney U test method to obtain preprocessed gene fragments; use K-nearest neighbor regression model for crossover Verify, enumerate the parameters of the K-nearest neighbor regression model, select the model parameters with the highest cross-validation accuracy, determine the optimal parameters of the prediction model, and construct the generated gene prediction list; When making predictions, the key gene fragments of cancer cells of clinical patients can be obtained according to the gene prediction list, the gene expression of key gene fragments can be obtained by qPCR technology or gene chip technology, and the gene expression can be standardized as the input of the prediction model. parameters, to obtain the prediction result of the drug corresponding to the prediction model on the cancer cell tissue of clinical patients, that is, drug efficacy prediction, to predict whether the drug is effective or ineffective, and the clinical patient can be prescribed an appropriate method for drug use according to whether the drug is effective.

In a possible application example, drug efficacy prediction is performed on tumor cancer cell tissues of clinical patients. Referring to FIG. 7 , there are several candidate drugs, such as candidate drug 1, candidate drug 2 and candidate drug 3, for the tumor cancer cell tissue. , each candidate drug has a corresponding gene prediction list, which are gene prediction list 1, gene prediction list 2 and gene prediction list 3, among which the gene prediction list contains about ten corresponding key genes in practical applications. The gene prediction lists of the three candidate drugs are assembled into a set based on the prediction list set; and, drug candidate 1, drug candidate 2 and drug candidate 3 have their own trained prediction models, namely prediction model 1, prediction model 2 and prediction model 2 respectively. Model 3; on the other hand, the tumor cancer cell tissue of clinical patients is obtained, and the gene expression level of the corresponding gene fragment is obtained by qPCR technology or gene chip technology combined with the gene prediction list, and the gene expression level is standardized as each candidate drug. The input of the corresponding prediction model, so as to obtain the corresponding drug efficacy prediction of candidate drug 1, candidate drug 2 and candidate drug 3, that is, predict the tumor cancer cell tissue of candidate drug 1, candidate drug 2 and candidate drug 3 for the clinical patient. Whether it is effective or not, based on the prediction results, individualized drug delivery plans for clinical patients are formulated to achieve precision medicine.

In the embodiment of the present application, by using qPCR technology or gene chip to quickly measure the gene expression level of the gene fragment, the time-consuming of the entire drug effect prediction can be shortened, and it is convenient to timely recommend the fish drug regimen during or after the clinical operation. It effectively avoids the dependence of drug efficacy prediction on RNA sequencing and other time-consuming sequencing technologies; and by reducing the gene set required for drug efficacy prediction, it can quickly and accurately predict the drug responsiveness of clinical patients, reducing prediction costs and time costs, Improve forecasting efficiency.

In a second aspect, an embodiment of the present application further provides an electronic device, including: at least one processor, and a memory communicatively connected to the at least one processor;

Wherein, the processor is configured to execute the drug sensitivity prediction method mentioned in the embodiment of the first aspect by calling the computer program stored in the memory.

As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs, such as the drug sensitivity prediction method mentioned in the embodiments of the first aspect of the present application. The processor implements the drug sensitivity prediction method mentioned in the embodiment of the first aspect above by running the non-transitory software program and instructions stored in the memory.

The memory may include a storage program area and a storage data area, wherein the storage program area may store an operating system and an application program required by at least one function; the storage data area may store and execute the drug sensitivity prediction method mentioned in the first aspect embodiment above. . Additionally, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory may optionally include memory located remotely from the processor, and these remote memories may be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The non-transitory software programs and instructions required to implement the drug sensitivity prediction method mentioned in the embodiment of the first aspect above are stored in the memory, and when executed by one or more processors, execute the method proposed in the embodiment of the first aspect above. A method for predicting drug susceptibility.

In a third aspect, the embodiments of the present application further provide a computer-readable storage medium storing computer-executable instructions, where the computer-executable instructions are used to: execute the drug sensitivity prediction method mentioned in the embodiments of the first aspect;

In some embodiments, the computer-readable storage medium stores computer-executable instructions that are executed by one or more control processors, eg, by a processor in the electronic device of the second aspect embodiment The execution can cause the above one or more processors to execute the drug sensitivity prediction method mentioned in the embodiment of the first aspect.

The device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

Those of ordinary skill in the art can understand that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and appropriate combinations thereof. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit . Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). As known to those of ordinary skill in the art, the term computer storage media includes both volatile and nonvolatile implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules or other data flexible, removable and non-removable media. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cartridges, magnetic tape, magnetic disk storage or other magnetic storage devices, or may Any other medium used to store desired information and which can be accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and can include any information delivery media, as is well known to those of ordinary skill in the art .

In the description of this specification, reference to the terms "some embodiments", "examples", "specific examples", or "some examples", etc., means that specific features or characteristics described in connection with the embodiments or examples are included in the present application at least one embodiment or example of . In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example.

Claims

A drug sensitivity prediction method, characterized in that it includes:

Obtain the gene sequencing data and drug characteristic data of the cancer cell tissue to be trained;

Preprocessing the gene sequencing data according to the drug characteristic data to obtain gene sample data;

Perform verification processing according to the gene sample data and the drug characteristic data to obtain a prediction model and a gene prediction list;

The drug sensitivity prediction of the cancer cell tissue to be tested is performed through the prediction model and the gene prediction list.
The drug sensitivity prediction method according to claim 1, wherein the gene sequencing data includes first sequencing data, and the drug characteristic data includes drug sensitivity data;

Correspondingly, the obtaining of the gene sequencing data and drug characteristic data of the cancer cell tissue to be trained includes:

The first sequencing data of the cancer cell tissue to be trained and the corresponding drug sensitivity data are acquired based on a genome database.
The drug sensitivity prediction method according to claim 2, wherein the gene sequencing data is preprocessed according to the drug characteristic data to obtain gene sample data, comprising:

standardizing the first sequencing data to obtain first sample data;

Screening the first sample data according to the drug sensitivity correlation coefficient of the first sample data and the drug sensitivity data to obtain second sample data;

Scoring and judging the second sample data according to the drug sensitivity data to obtain a scoring parameter of the second sample data;

The second sample data is screened based on the scoring parameter to obtain the gene sample data.
The method for predicting drug sensitivity according to claim 3, wherein the verification process is performed according to the gene sample data and the drug characteristic data to obtain a prediction model and a gene prediction list, comprising:

obtaining the drug sensitivity correlation coefficient between the genetic sample data and the drug sensitivity data, and obtaining the scoring parameter of the genetic sample data, where the genetic sample data includes a plurality of gene fragments;

Arrange the multiple gene fragments in descending order according to the drug susceptibility correlation coefficient and the scoring parameter;

Perform verification processing on the plurality of gene fragments arranged in descending order to obtain the model parameters of the prediction model and the number of gene lists;

The gene prediction list is generated according to the number of gene lists, and the prediction model is determined according to the model parameters.
The drug sensitivity prediction method according to claim 1, wherein the gene sequencing data includes second sequencing data, and the drug characteristic data includes drug effect classification data;

Correspondingly, the obtaining of the gene sequencing data and drug characteristic data of the cancer cell tissue to be trained includes:

The second sequencing data and drug effect grading data of the cancer cell tissue to be trained are obtained based on the genome atlas database.
The drug sensitivity prediction method according to claim 5, wherein the gene sequencing data is preprocessed according to the drug characteristic data to obtain gene sample data, comprising:

standardizing the second sequencing data to obtain third sample data;

The third sample data is tested according to the drug effect classification data to obtain the gene sample data.
The drug sensitivity prediction method according to claim 6, wherein the verification process is performed according to the gene sample data and the drug characteristic data to obtain a prediction model and a gene prediction list, comprising:

obtaining gene scores of multiple gene fragments of the gene sample data;

Ranking the plurality of gene fragments in descending order according to the gene score;

Cross-validation is performed on the plurality of gene fragments arranged in descending order to obtain the model parameters of the prediction model and the number of gene lists;

The gene prediction list is generated according to the number of gene lists and the corresponding plurality of gene segments, and the prediction model is determined according to the model parameters.
The drug sensitivity prediction method according to claim 4 or 7, wherein the predicting the drug sensitivity of the cancer cell tissue to be tested by using the prediction model and the gene prediction list includes:

Acquiring gene fragments corresponding to the cancer cell tissue to be tested according to the gene prediction list;

obtaining the gene expression level of the gene fragment;

The gene expression level is input into the prediction model, and the drug sensitivity result of the cancer cell tissue to be tested is obtained.
Electronic equipment, characterized in that it includes:

at least one processor, and,

a memory communicatively coupled to the at least one processor; wherein,

The memory stores instructions, which are executed by the at least one processor, so that when the at least one processor executes the instructions, the drug sensitivity prediction method of any one of claims 1 to 8 is implemented.
A computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to cause a computer to perform the drug sensitivity prediction according to any one of claims 1 to 8 method.