Diffuse large B-cell lymphoma genotype classification method, device and storage medium

Abstract

The embodiment of the present application discloses a genotype classification method for diffuse large B-cell lymphoma, a computer device, and a computer-readable storage medium. The method includes the following steps: testing a sample according to a preset specific gene set to obtain variation detection data; preprocessing the variation detection data to obtain variation information; generating an initial feature matrix based on the variation information, screening the initial feature matrix to obtain a feature matrix, and the feature matrix is ​​used to characterize the gene variation contained in the specific gene set in the corresponding sample; obtaining a first genotype label and an important feature set, constructing a data set based on the feature matrix, the first genotype label, and the important feature set, and training a classification model; obtaining a second genotype label output by the classification model, and determining a genotype classification report based on the second genotype label. Therefore, the present application can reduce the cost of testing, can effectively predict the patient's genotype, and has a high clinical application value.

Description

Technical Field

[0001] The present application belongs to the field of biological detection technology, and in particular relates to a diffuse large B-cell lymphoma genotype classification method, a computer device, and a computer-readable storage medium. Background Art

[0002] Diffuse Large B Cell Lymphoma (DLBCL) is the most common subtype of non-Hodgkin's lymphoma (NHL), accounting for approximately 30% to 50% of NHL. Worldwide, approximately 150,000 new cases of DLBCL are reported annually, with the incidence trending upward. Studies have shown that DLBCL exhibits a high degree of genetic and phenotypic heterogeneity, and these differences are associated with patient clinical characteristics, treatment response, and overall survival. Therefore, exploring the root causes of DLBCL heterogeneity through molecular biological mechanisms and achieving accurate classification and diagnosis based on this knowledge is of great clinical significance.

[0003] Existing technologies have used whole-exome sequencing, transcriptome sequencing, DNA copy number variation analysis, and targeted amplicon resequencing of 372 genes to identify various DLBCL genotypes with varying prognoses and treatment deficiencies. Furthermore, a probabilistic DLBCL genotype classification tool has been developed based on cluster analysis and the naive Bayesian algorithm. However, this tool requires numerous genetic tests to generate reliable predictions, resulting in high costs. Accurately and efficiently classifying genetic tests remains a pressing technical challenge for those skilled in the art.

[0004] The preceding description is intended to provide general background information and does not necessarily constitute prior art. Summary of the Invention

[0005] Based on this, in response to the above problems, a diffuse large B-cell lymphoma genotype classification method, computer device and computer-readable storage medium are proposed, which can effectively classify DLBCL patients according to genotype.

[0006] The present application solves the technical problem by adopting the following technical solutions:

[0007] The present application provides a method for genotyping diffuse large B-cell lymphoma, comprising the following steps: testing a sample according to a preset specific gene set to obtain variation detection data; preprocessing the variation detection data to obtain variation information; generating an initial feature matrix according to the variation information, screening the initial feature matrix to obtain a feature matrix, wherein the feature matrix is ​​used to characterize the gene variation contained in the specific gene set in the corresponding sample; obtaining a first genotype label and an important feature set, constructing a data set according to the feature matrix, the first genotype label and the important feature set, and training a classification model; obtaining a second genotype label output by the classification model, and determining a genotype classification report according to the second genotype label.

[0008] In an optional embodiment of the present application, the variation detection data includes sequencing data and fluorescence in situ hybridization detection data; the sample is tested according to a preset specific gene set to obtain the variation detection data, including: performing high-throughput sequencing on the sample according to the specific gene set to obtain sequencing data; designing rearrangement probes according to the specific gene set, and using fluorescence in situ hybridization technology to obtain fluorescence in situ hybridization detection data of the sample.

[0009] In an optional embodiment of the present application, the variation information includes: gene mutation information, copy number variation information and gene rearrangement information; the variation detection data is preprocessed to obtain the variation information, including: performing a first preprocessing on the sequencing data to obtain gene mutation information and copy number variation information respectively; performing a second preprocessing on the fluorescence in situ hybridization detection data to obtain gene rearrangement information that can be used for modeling.

[0010] In an optional embodiment of the present application, obtaining an important feature set includes: obtaining preset feature data, generating a data set based on the preset feature data, a first genotype label and a feature matrix, the data set including a training set; screening the initial features of the training set according to a gene variation type set to obtain a first feature, the gene variation type set including gene variation types of MUTATION, TRUNC, HOMDEL, AMP, GAIN, HETLOSS, CompUp and FUSION; screening the first feature according to a specific gene set to obtain a second feature; obtaining a first genotype set, constructing a feature fine screening model based on the first genotype set, and screening the second feature according to the feature fine screening model to obtain a third feature; obtaining auxiliary features, and generating an important feature set based on the third feature and the auxiliary features.

[0011] In an optional embodiment of the present application, the feature fine screening model includes at least two binary classification models; the second feature is screened according to the feature fine screening model to obtain the third feature, including: constructing corresponding binary classification models according to the genotypes recorded in the first genotype set; inputting the second feature into the binary classification model to obtain a feature importance score, and repeating a preset number of times; taking the union of the features of each binary classification model whose feature importance score is greater than the stable threshold each time, and determining it as the third feature.

[0012] In an optional embodiment of the present application, a data set is constructed and a classification model is trained based on a feature matrix, a first genotype label and an important feature set, including: establishing a first classification model based on the important feature set and the first genotype label, the first classification model being able to complete the classification of the genotypes recorded in the first genotype set; obtaining a preset feature set, and establishing a second classification model based on auxiliary features, the preset feature set and the first genotype label, the second classification model being able to complete the classification of the genotypes recorded in the second genotype set; training the first classification model and the second classification model respectively according to the training set, and iteratively adjusting parameters through a cross-validation set; integrating the trained first classification model and the second classification model to complete the construction of the classification model.

[0013] In an optional embodiment of the present application, the data set also includes a test set, and obtaining the second genotype label output by the classification model includes: inputting the feature matrix generated according to the test set into the second classification model for genotype classification; if the genotype classification of the sample can be determined according to the second classification model, generating the corresponding second genotype label according to the classification result of the second classification model; if the genotype classification of the sample cannot be determined according to the second classification model, inputting the feature matrix into the first classification model for further classification; and generating the corresponding second genotype label according to the classification result of the first classification model.

[0014] In an optional embodiment of the present application, the first genotype set includes ST2 and A53; the second genotype set includes MCD, BN2, EZB and N1.

[0015] The present application also provides a computer device, comprising a processor and a memory: the processor is configured to execute a computer program stored in the memory to implement the aforementioned method.

[0016] The present application also provides a computer-readable storage medium storing a computer program, which implements the aforementioned method when the computer program is executed by a processor.

[0017] The embodiments of the present application have the following beneficial effects:

[0018] This application can detect gene mutations in the genome of the sample in a targeted manner through a preset specific gene set, thereby significantly reducing the detection cost. The important feature set determined by layer-by-layer screening can avoid the complicated manual feature selection process. The classification model obtained by constructing a data set using the feature matrix, the first genotype label and the important feature set and training can dig out the driver genes of each typing, making the model have good interpretability, and can effectively predict the patient's genotype, improve the accuracy of DLBCL genotype classification, and have high clinical application value.

[0019] The above description is only an overview of the technical solution of this application. In order to more clearly understand the technical means of this application, which can be implemented in accordance with the contents of the description, and to make the above and other purposes, features and advantages of this application more obvious and easy to understand, the following preferred embodiments are specifically described in detail with reference to the accompanying drawings. It should be understood that the above general description and the detailed description below are only exemplary and explanatory and do not limit this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] in:

[0022] Figure 1 A schematic diagram of a flow chart of a method for genotyping diffuse large B-cell lymphoma provided in Example 1;

[0023] Figure 2 A schematic diagram of the process of obtaining an important feature set provided in Example 2;

[0024] Figure 3 A schematic diagram of the application process of the DLBCL gene classification method provided in Example 3;

[0025] Figure 4 This is a schematic block diagram of the structure of a computer device provided in Example 4. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0027] Example 1

[0028] With the emergence of gene expression profiling, initial progress has been made in the molecular diagnosis of DLBCL subtypes. Based on the differences in gene expression profiles, two major "cell of origin (COO)" types were defined, including germinal center B-cell-like (GCB) and activated B-cell-like (ABC), and the rest were classified as "unclassified". However, the COO type cannot fully explain the heterogeneous response and prognosis of DLBCL patients after R-CHOP treatment or targeted therapy. This may be because gene expression profiling provides a phenotypic description of cancer rather than a more direct genetic description of tumor pathogenesis. LymphGen, a classification tool developed by existing technology, is currently the most authoritative DLBCL genotype model in the world. However, it is stated in the disclaimer that the model tool cannot be used for commercial purposes, which hinders its guiding role in the precise treatment of clinical DLBCL. In addition, studies have shown that DLBCL patients differ in characteristics, incidence, and survival rates. These differences are specifically reflected in differences between human populations in geographical regions (hereinafter referred to as populations). For example, the standards used for non-Chinese populations are not applicable to the Chinese population. The LymphGen tool was developed based on DLBCL patients from non-Chinese populations, and the key variant genes in the genotypes were also determined from these cases using the Fisher test. Considering the population differences in DLBCL genotypes, due to population differences, the LymphGen developed by existing technology is based on datasets mostly collected from people who have lived in non-Chinese regions for a long time, and the results obtained are not fully applicable to the Chinese population. Based on the above, there is an urgent need to develop a new DLBCL genotype prediction method that relies on relatively low-cost genetic testing projects and can use machine learning algorithms to explore the variant genes and their variant types that play a key role in typing. With the help of the powerful feature integration and classification capabilities of machine learning algorithms, satisfactory typing results can be achieved to meet the strong clinical demand for genotype conclusions for Chinese DLBCL patients. Based on this, the diffuse large B cell lymphoma genotype classification method provided by this application is proposed, including steps S110 to S150. In order to clearly describe the diffuse large B cell lymphoma genotype classification method provided by this embodiment, please refer to Figure 1 .

[0029] Step S110: testing the sample according to a preset specific gene set to obtain variation detection data.

[0030] Step S120: Preprocess the variation detection data to obtain variation information.

[0031] In one embodiment, the variation detection data includes sequencing data and fluorescence in situ hybridization detection data; the sample is tested according to a preset specific gene set to obtain the variation detection data, including: performing high-throughput sequencing on the sample according to the specific gene set to obtain sequencing data; designing rearrangement probes according to the specific gene set, and using fluorescence in situ hybridization technology to obtain fluorescence in situ hybridization detection data of the sample.

[0032] In one embodiment, the variation information includes: gene mutation information, copy number variation information and gene rearrangement information; the variation detection data is preprocessed to obtain the variation information, including: performing a first preprocessing on the sequencing data to obtain the gene mutation information and the copy number variation information respectively; performing a second preprocessing on the fluorescence in situ hybridization detection data to obtain the gene rearrangement information that can be used for modeling.

[0033] In one embodiment, as described above, DLBCL has a high degree of genetic and phenotypic heterogeneity. According to the authoritative guidelines and literature related to DLBCL, such as the NCCN guidelines, the WHO 2016 hematopoietic and lymphoid tissue tumor classification guidelines, and the authoritative journals and literature in related fields such as Blood, the New England Journal of Medicine, Cell, Cancer Cell, etc., a plurality of genes closely related to the occurrence, development and typing of DLBCL tumors are determined in the preferred embodiment of the present application (specifically, 125 genes), that is, the genes recorded in the specific gene set (Panel), for the prediction of DLBCL genotype results. Therefore, the present application detects the sample once by a preset Panel to obtain variation detection data and use it for subsequent analysis and modeling, reducing the additional link of detecting CNV (Copy number variation, CNV, copy number variation) by comparative genomic hybridization (CGH) technology. Under the premise of ensuring sufficient detection information, the detection cost of the patient is greatly reduced. Specific variation detection data include sequencing data and fluorescence in situ hybridization detection data, so it is necessary to perform corresponding detection on the two types of variation detection data to obtain. Specifically, sequencing data can be obtained using high-throughput sequencing technology, while fluorescence in situ hybridization (FISH) data can be obtained using next-generation sequencing (NGS). By setting a specific panel to limit the scope of testing, NGS can be used to detect mutations within a specific gene panel, rather than across the entire genome, thereby reducing testing costs.

[0034] In one embodiment, the variation detection data is the data obtained through gene variation detection. It is understandable that the variation detection data obtained through preliminary detection is relatively rough as a whole: the information is redundant, complex, and primitive, making it difficult to analyze and process. For this reason, it is necessary to process the variation detection data into a form that can support modeling, that is, to obtain variation information based on preprocessing. Taking gene mutation information as an example, the preprocessing process can refer to the following description. For gene mutation information, the detection process can be to detect single nucleotide variations (SNVs) and small insertions / deletions (InDels) occurring in the entire exon region of the gene in the package and within a first preset length nearby. In a preferred embodiment, the first preset length can be 10bp. For copy number variation information, SNP probes (Single Nucleotide Polymorphisms) that detect copy number changes can be used to detect copy number variations greater than a second preset length and copy number variations at the gene exon level of a first preset number of genes. In a preferred embodiment, the second preset length can be 5Mbp, that is, CNVs of large genomic fragments larger than 5Mbp are detected; the first preset number can be related to the number of genes recorded in the specific gene set, that is, in a preferred embodiment, the first preset number can be the same as the number of genes recorded in the specific gene set, specifically 125. The samples are sequenced in high-throughput mode using the double-end sequencing (2×150bp) mode of the Novasek sequencing platform. The sequencing data needs to undergo a first preprocessing to extract gene mutation information and copy number variation information. The first preprocessing can include but is not limited to: (1) quality control, removing low-quality reads; (2) data alignment, aligning the filtered reads with the reference genome; (3) sorting and deduplication, removing duplicate sequences generated by PCR during library construction; (4) base quality recalibration; (5) variant site detection; (6) variant result quality control and filtering; (7) variant result annotation to obtain gene mutation information that can be used for subsequent modeling; (8) using CNV Radar for detection to obtain copy number variation information that can be used for modeling. The second preprocessing performed on fluorescence in situ hybridization data involves intuitively determining whether a gene has rearranged variants based on the fluorescence signal and a reference threshold, thereby obtaining gene rearrangement information that can be used for modeling. It is also worth noting that whether testing to obtain variant detection data or analyzing to obtain variant information that can be used for modeling, the overall process is relatively existing and fixed. The focus is on performing a one-time test on the sample based on a set panel, that is, based on a specific gene set, to ultimately obtain the required variant information, thereby reducing testing costs.

[0035] Step S130: generating an initial feature matrix based on the variation information, and screening the initial feature matrix to obtain a feature matrix, which is used to characterize the gene variation contained in the specific gene set of the corresponding sample.

[0036] Step S140: Obtain the first genotype label and the important feature set, construct a data set based on the feature matrix, the first genotype label and the important feature set, and train a classification model.

[0037] In one embodiment, the variation information is first encoded into a format that is easy for the model to process. For example, the variation information can be converted into an initial feature vector based on the feature construction method of LymphGen. The x-axis of the initial feature matrix corresponds to each sample, and the y-axis corresponds to the different variation forms of different genes on the genome. The value of the initial feature matrix indicates whether a gene in a certain sample has a certain form of variation. There are three values: 0, 1, and -1, which correspond to the absence of the variation, the presence of the variation, and the failure to detect the variation, respectively. The initial feature matrix is ​​then screened to obtain a feature matrix. The feature matrix is ​​used to input the classification model and is used to characterize the gene variation contained in the specific gene set of the corresponding sample. The specific screening situation needs to be screened through the corresponding feature set. The specific steps will be described in detail later and will not be expanded here.

[0038] In one embodiment, step S140: obtaining an important feature set includes: obtaining preset feature data, generating a data set based on the preset feature data, a first genotype label and a feature matrix, the data set including a training set; screening the initial features of the training set according to a gene variation type set to obtain a first feature, the gene variation type set including gene variation types of MUTATION, TRUNC, HOMDEL, AMP, GAIN, HETLOSS, CompUp and FUSION; screening the first feature according to a specific gene set to obtain a second feature; obtaining a first genotype set, constructing a feature fine screening model based on the first genotype set, and screening the second feature according to the feature fine screening model to obtain a third feature; obtaining auxiliary features, and generating an important feature set based on the third feature and the auxiliary features.

[0039] In one embodiment, the feature fine screening model includes at least two binary classification models; the second feature is screened according to the feature fine screening model to obtain the third feature, including: constructing corresponding binary classification models according to the genotypes recorded in the first genotype set; inputting the second feature into the binary classification model to obtain a feature importance score, and repeating a preset number of times; taking the union of the features whose feature importance scores of each binary classification model are greater than the stable threshold each time, and determining them as the third feature.

[0040] In one embodiment, the first genotype set includes ST2 and A53; the second genotype set includes MCD, BN2, EZB, and N1.

[0041] In one embodiment, the first genotype label is a reference label used to establish a model. The first genotype label mentioned in this embodiment can be predicted by LymphGen. It can be obtained through the following six input files: (1) sample annotation file, including sample name, whether CNV data is available, whether BCL2 and BCL6 ectopic information are available; (2) mutation information file, including sample name, gene ID, mutation type and chromosome location information; (3) mutation detection gene list file, including all gene IDs that have undergone targeted sequencing; (4) copy number variation information file, including sample name, gene ID and copy number variation type; (5) copy number variation detection gene list file, including all gene IDs in the copy number variation detection region; (6) chromosome arm variation information file, including sample name, chromosome arm name and copy number variation type. This part of information is used to count copy number variations based on chromosome arms and define their specific variation types. After the mutation detection data obtained in step S110 is pre-processed according to step S120, mutation information that can be used for analysis can be obtained to construct the above-mentioned six input files. LymphGen can then be used to assign genotyping labels to these cases, namely the first genotype label. Specifically, the first genotype label can be one of the genotype labels recorded in the first genotype set and the second genotype set. The first genotype set and the second genotype set are distinguished as first and second because the genotypes recorded in the two differ in whether there are significant population differences in important genetic characteristics: for the genes recorded in the first genotype set, there are significant differences in different populations; while for the genotypes recorded in the second genotype set, there are no significant differences in different populations. In the overlapping embodiments of the present application, the first genotype set may include ST2 and A53; the second genotype set may include MCD, BN2, EZB and N1. In addition, it is understandable that as the research goes deeper, other new genotypes will be defined in the future. Therefore, the first genotype label may include MCD, BN2, EZB, N1, ST2, A53 and Other, and Other is also a sample category that cannot be accurately typed. The output file of LymphGen contains the confidence level of each case belonging to the MCD, BN2, EZB, N1, ST2, A53 typing and the final typing prediction result. Since LymphGen's probability model is constructed for the six specific genotypes mentioned above, there is the possibility that a case is labeled as two or more of these six typings. For this multi-label situation, in this application, only the typing corresponding to the maximum confidence level of the case is taken as the label.

[0042] In one embodiment, the important feature set is necessary for building a classification model and needs to be acquired in advance. The feature matrix and the first genotype label are used to generate a data set, which is then divided into a training set and a test set according to a preset ratio, for example, it can be divided according to a ratio of 6:4. In addition, the DLBCL cases of non-Chinese populations used by LymphGen are also added to the training set as auxiliary data to enhance the typing ability and robustness of the model, thereby obtaining a training set that can be used for model training. Even after preprocessing, the variation information in the above text is actually relatively complex and redundant. Therefore, identifying and screening the most critical features for DLBCL genotyping from the complex gene variation information is an important prerequisite for model construction. To this end, a series of screening of the initial features of the training set is required to determine the important feature set required for the final model construction.

[0043] In one embodiment, step S140: constructing a data set and training a classification model based on the feature matrix, the first genotype label and the important feature set, including: establishing a first classification model based on the important feature set and the first genotype label, the first classification model being able to complete the classification of the genotypes recorded in the first genotype set; obtaining a preset feature set, establishing a second classification model based on the auxiliary features, the preset feature set and the first genotype label, the second classification model being able to complete the classification of the genotypes recorded in the second genotype set; training the first classification model and the second classification model respectively according to the training set, and iteratively adjusting the parameters through the cross-validation set; integrating the trained first classification model and the second classification model to complete the construction of the classification model.

[0044] In one embodiment, to prevent data leakage, the test set should not be accessed during feature selection, model training, and cross-validation. Therefore, only data from the training set is used for feature screening. LymphGen introduces a significant amount of redundant information when constructing the feature matrix, such as simply mapping mutations to multiple different variant types. Therefore, the initial features of the training set are first pre-screened based on a set of gene mutation types to obtain the first feature, thereby removing the redundant information added by the simple mapping. Specifically, in a preferred embodiment, the set of gene mutation types includes mutations, TRUNC, HOMDEL, AMP, GAIN, HETLOSS, CompUp, and FUSION. A further coarse screening is then performed, where the first feature is screened based on a specific gene set to obtain the second feature. The specific gene set can be genes closely related to the occurrence, development, and typing of DLBCL tumors, as defined by the targeted detection panel. Only features within the first feature that belong to genes defined within the panel are retained; these retained features are therefore the second features. Furthermore, as mentioned above, some important features of genotypes vary significantly among different populations, i.e., the genotypes recorded in the first genotype set; some do not vary significantly among different populations, i.e., the genotypes recorded in the second genotype set. Based on this, a feature fine-screening model can be constructed based on the first genotype set, and the second feature can be screened according to the feature fine-screening model to obtain a third feature. Finally, auxiliary features are obtained, and an important feature set is generated based on the third feature and the auxiliary features. Among them, the auxiliary features may include: the presence or absence of copy number variation detection results, the presence or absence of BCL2 rearrangement detection results, and the presence or absence of BCL6 rearrangement detection results.

[0045] In one embodiment, step S140: constructing a data set and training a classification model based on the feature matrix, the first genotype label and the important feature set, including: establishing a first classification model based on the important feature set and the first genotype label, the first classification model being able to complete the classification of the genotypes recorded in the first genotype set; obtaining a preset feature set, establishing a second classification model based on the auxiliary features, the preset feature set and the first genotype label, the second classification model being able to complete the classification of the genotypes recorded in the second genotype set; training the first classification model and the second classification model respectively according to the training set, and iteratively adjusting the parameters through the cross-validation set; integrating the trained first classification model and the second classification model to complete the construction of the classification model.

[0046] In one embodiment, extreme gradient boosting tree (eXtreme Gradient Boosting, XGBoost) is an improved form of gradient boosting decision tree (Gradient Boosting Decision Tree, GBDT), with higher training efficiency and more controllable parameters. For this reason, in the present application's preferred embodiment, the application can build a classification model based on XGBoost. For the genotype recorded in the first genotype set with population differences, and the genotype recorded in the second genotype set without showing population differences, different features are used to build respective classification models: the first classification model that can complete the classification of the genotype recorded in the first genotype set and the second classification model that can complete the classification of the genotype recorded in the second genotype set. The first classification model is constructed using the key feature set and the first genotype signature. The second model requires auxiliary features, a preset feature set, and the first genotype signature. Auxiliary features can include the presence or absence of copy number variation test results, BCL2 rearrangement test results, and BCL6 rearrangement test results. The preset feature set, as described above, specifically includes the 97 features selected by LymphGen. This means that the second classification model uses a total of 100 features to characterize each case sample to classify the second genotype set. After construction, the first and second classification models can be trained separately using the training set. The training process involves: after obtaining a feature matrix based on the key feature set, the classification model is trained using the first genotype signature and the feature matrix, and the model parameters are iteratively adjusted using the cross-validation set. After the first and second classification models are trained, they are merged to complete the construction of the classification model.

[0047] Step S150: Obtain the second genotype label output by the classification model, and determine a genotype classification report based on the second genotype label.

[0048] In one embodiment, the data set also includes a test set, and step S150: obtaining a second genotype label output by the classification model includes: inputting the feature matrix generated according to the test set into the second classification model for genotype classification; if the genotype classification of the sample can be determined according to the second classification model, generating a corresponding second genotype label according to the classification result of the second classification model; if the genotype classification of the sample cannot be determined according to the second classification model, inputting the feature matrix into the first classification model for further classification; generating a corresponding second genotype label according to the classification result of the first classification model.

[0049] In one embodiment, the classification model actually includes a first classification model and a second classification model. Therefore, during the classification process, the two models can be classified sequentially. During the testing process, the test set included in the previously generated dataset is used for testing. Specifically, to obtain the second genotype label output by the classification model, the feature matrix generated by filtering the test set with the aforementioned important feature set is input into the first and second classification models. In a preferred embodiment, the feature matrix generated based on the test set can first be input into the second classification model to determine whether the sample has one of the genotypes listed in the second genotype set. If the second classification model can complete the classification, the corresponding second genotype label is generated based on the classification results of the second classification model. If the classification cannot be completed, or the classification result is uncertain, such as "Other," the feature matrix is ​​input into the first classification model for further classification. The corresponding second genotype label is generated based on the classification results of the first classification model, specifically determining whether the sample has a genotype listed in the first genotype set. For example, if the first genotype set includes ST2 and A53, the classification result can include one of ST2, A53, and "Other." The genotype classification report is then determined based on the second genotype signature output by the first or second classification model. It is worth noting that, unlike the first genotype signature used to train the classification model, the second genotype signature is the output of the classification model. In other words, the genotype classification report can be determined based on the second genotype signature output by the model.

[0050] Therefore, the present application can perform gene variation detection on the genome of the sample in a targeted manner through a preset specific gene set, thereby significantly reducing the detection cost. The important feature set determined by layer-by-layer screening can avoid the complicated manual feature selection process. The classification model obtained by constructing a data set using a feature matrix, a first genotype label, and an important feature set and training can dig out the driver genes of each typing, making the model have good interpretability, and can effectively predict the patient's genotype, improve the DLBCL genotype classification accuracy, and have a high clinical application value.

[0051] Example 2

[0052] Important feature sets are an important part of building a classification model. They need to be acquired in advance. The acquisition process can be referred to Figure 2 , Figure 2 The schematic diagram of the important feature set acquisition process provided in the second embodiment includes steps S210 to S270.

[0053] Step S210: obtaining preset feature data, and generating a data set according to the preset feature data, the first genotype label and the feature matrix, wherein the data set includes a training set.

[0054] In one embodiment, to evaluate the performance of the classification model to be constructed on the Chinese population, a dataset can be generated from feature information and divided into a training set and a test set in a ratio of 6:4. In addition, DLBCL cases from non-Chinese populations used by LymphGen are also added to the training set as auxiliary data to enhance the classification ability and robustness of the classification model. The initial features of the training set are mainly used for screening. To avoid data leakage, the test set should not be exposed during feature selection, model training, and cross-validation. Therefore, the use of the test set will not be explained in the subsequent steps.

[0055] Step S220: Screening the initial features of the training set according to the gene variation type set to obtain a first feature.

[0056] Step S230: Screen the first feature according to the specific gene set to obtain the second feature.

[0057] In one embodiment, LymphGen introduces a significant amount of redundant information when constructing its feature matrix, for example, by simply mapping mutations to multiple variant types. Therefore, the initial features of the training set are pre-screened based on a set of genetic variant types to obtain the first feature, thereby removing the redundant information added by this simple mapping. Specifically, in a preferred embodiment, the set of genetic variant types includes mutations, trunks, homdels, amps, gains, hetlosss, compups, and fusions.

[0058] In one embodiment, the specific gene set can be genes closely related to the occurrence, development, and classification of DLBCL tumors set by the targeted detection panel. Only the features of the first feature that belong to the genes set in the panel will be retained, and the retained features are also the second features.

[0059] Step S240: construct corresponding binary classification models according to the genotypes recorded in the first genotype set.

[0060] Step S250: input the second features into the binary classification model for training to obtain feature importance scores, and repeat the training a preset number of times.

[0061] Step S260: Take the union of the features whose feature importance scores are greater than the stability threshold in each binary classification model and determine it as the third feature.

[0062] In one embodiment, the genotype of record in the first genotype set has significant differences in different populations, and can include multiple genotypes.Therefore in the screening process, can set up binary classification model according to the genotype of record in the first genotype set respectively.Can be, use different random seeds to repeat the preset number of feature importance scores to each binary classification, and the feature that each score exceeds the preset threshold is taken intersection to obtain stable and critical feature.Specifically, for example the genotype label of record in the first genotype set can include ST2, for this reason then can the genotype label of other all non-ST2 be classified into a class and set up corresponding ST2 / non-ST2 binary classification model according to the second feature.This binary classification model is built based on tree model, can carry out feature importance scoring to input second feature after training ends, and repeat five times.The feature that five scores are all exceeded 0.01 is retained, and in conjunction with the feature that other binary classification models retain, take intersection and obtain the 3rd feature.

[0063] Step S270: Acquire auxiliary features, and generate an important feature set based on the third feature and the auxiliary features.

[0064] In one embodiment, the auxiliary features may include: the presence or absence of copy number variation detection results, the presence or absence of BCL2 rearrangement detection results, and the presence or absence of BCL6 rearrangement detection results. The auxiliary features and the third features are integrated to complete the construction of the important feature set.

[0065] Therefore, based on the method of Example 3 of the present application, the key gene features can be acquired, ensuring that the constructed classification model is stable, efficient, and has good classification prediction performance.

[0066] Example 3

[0067] The diffuse large B-cell lymphoma genotype classification method proposed in this application is used to determine the patient's genotype type. However, there are also obvious differences between genotypes: some have significant differences in different populations, while others have no significant differences in different populations. According to different genotype sets, a first classification model and a second classification model are established respectively: the first classification model can complete the classification of the genotypes recorded in the first genotype set; the second classification model can complete the classification of the genotypes recorded in the second genotype set. Therefore, in the classification process, the two can be classified in order. For the specific classification process, please refer to Figure 3 , Figure 3 This is a schematic diagram of the application process of the DLBCL gene classification method provided in Example 3, including steps S310 to S360.

[0068] Step S310: Input the feature matrix generated according to the test set into the second classification model to perform genotype classification.

[0069] In one embodiment, it is worth noting that the feature matrix for the samples to be classified is generated using a test set. The generation of the test set and the method for generating the feature matrix based on the test set have been described in detail in the previous embodiment. For details, please refer to the previous text and will not be repeated here.

[0070] Step S320: Determine whether the genotype classification of the sample can be determined according to the second classification model.

[0071] If the genotype classification of the sample can be determined according to the second classification model, step S330 is executed: generating a corresponding second genotype label according to the classification result of the second classification model.

[0072] If the genotype classification of the sample cannot be determined according to the second classification model, step S340 is executed: the feature matrix is ​​input into the first classification model for further classification.

[0073] In one embodiment, the feature matrix generated based on the test set can first be input into a second classification model, which can complete the classification of the genotypes recorded in the second genotype set. If the classification result can determine the genotype of the sample, a corresponding second genotype label is generated. If the second classification model cannot determine the genotype classification of the sample, that is, if the classification result is Other, the feature matrix is ​​input into the first classification model for further classification.

[0074] Step S350: Generate a corresponding second genotype label according to the classification result of the first classification model.

[0075] In one embodiment, a first classification model classifies an input feature matrix to determine whether the sample belongs to a genotype within a first genotype set or whether classification cannot be completed. A corresponding second genotype label is generated based on the classification result of the first classification model. It is worth noting that both the first and second classification models can determine only one genotype.

[0076] Step S360: Obtain the second genotype label output by the classification model, and determine a genotype classification report based on the second genotype label.

[0077] In one embodiment, the classification model includes a first classification model and a second classification model. A feature vector of a sample input will only output a second genotype label, and the second genotype label is used to determine a genotype classification report for auxiliary treatment or diagnosis.

[0078] Therefore, with the cooperation of the two models, the diffuse large B-cell lymphoma genotype classification method provided in this application has achieved satisfactory predictive performance in the six categories of genotyping and Other categories, and is more suitable for the clinical precise typing diagnosis of the highly heterogeneous DLBCL Chinese population.

[0079] Example 4

[0080] Figure 4 FIG1 shows an internal structure diagram of a computer device in an embodiment. The computer device can be a terminal or a server. Figure 4 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When the computer program is executed by the processor, the processor can implement the diffuse large B-cell lymphoma genotype classification method. The internal memory may also store a computer program. When the computer program is executed by the processor, the processor can implement the age recognition method. It will be understood by those skilled in the art that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0081] In one embodiment, a computer device is proposed, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps: Step S110: Detect the sample according to a preset specific gene set to obtain variation detection data. Step S120: Preprocess the variation detection data to obtain variation information. Step S130: Generate an initial feature matrix based on the variation information, screen the initial feature matrix to obtain a feature matrix, and the feature matrix is ​​used to characterize the gene variation contained in the specific gene set in the corresponding sample. Step S140: Obtain a first genotype label and an important feature set, construct a data set based on the feature matrix, the first genotype label and the important feature set, and train a classification model. Step S150: Obtain a second genotype label output by the classification model, and determine a genotype classification report based on the second genotype label.

[0082] In one embodiment, the present application further proposes a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the aforementioned method.

[0083] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0084] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0085] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for genotyping diffuse large B-cell lymphoma, characterized in that: The steps include: Testing the sample according to a preset specific gene set to obtain mutation detection data; the specific gene set is a gene set closely related to the occurrence, development and typing of DLBCL tumors; Preprocessing the variation detection data to obtain variation information; generating an initial feature matrix based on the variation information, and screening the initial feature matrix to obtain a feature matrix, wherein the feature matrix is ​​used to characterize the presence of gene variation included in the specific gene set in the corresponding sample; Obtaining a first genotype label and an important feature set, constructing a data set based on the feature matrix, the first genotype label and the important feature set, and training a classification model; A second genotype label output by the classification model is obtained, and a genotype classification report is determined according to the second genotype label.

2. The method for genotyping diffuse large B-cell lymphoma according to claim 1, wherein: The mutation detection data includes sequencing data and fluorescence in situ hybridization detection data; The method of testing the sample according to a preset specific gene set to obtain variation detection data includes: Performing high-throughput sequencing on the sample according to the specific gene set to obtain the sequencing data; Rearrangement probes are designed according to the specific gene set, and the fluorescence in situ hybridization detection data of the sample is obtained using fluorescence in situ hybridization technology.

3. The method for genotyping diffuse large B-cell lymphoma according to claim 2, wherein: The variation information includes: gene mutation information, copy number variation information and gene rearrangement information; The preprocessing of the variation detection data to obtain variation information includes: performing a first preprocessing on the sequencing data to obtain the gene mutation information and copy number variation information; A second preprocessing is performed on the fluorescence in situ hybridization detection data to obtain the gene rearrangement information that can be used for modeling.

4. The method for genotyping diffuse large B-cell lymphoma according to claim 1, wherein: The obtaining of the important feature set includes: Acquire preset feature data, and generate the data set according to the preset feature data, the first genotype label, and the feature matrix, wherein the data set includes a training set; Screening the initial features of the training set according to a set of gene variation types to obtain a first feature, wherein the set of gene variation types includes gene variation types of MUTATION, TRUNC, HOMDEL, AMP, GAIN, HETLOSS, CompUp, and FUSION; Screening the first feature according to the specific gene set to obtain a second feature; Obtaining a first genotype set, constructing a feature fine screening model based on the first genotype set, and screening the second feature based on the feature fine screening model to obtain a third feature; Auxiliary features are obtained, and the important feature set is generated according to the third feature and the auxiliary features.

5. The method for genotyping diffuse large B-cell lymphoma according to claim 4, wherein: The feature fine screening model includes at least two binary classification models; The step of screening the second feature according to the feature fine screening model to obtain the third feature includes: According to the genotypes recorded in the first genotype set, corresponding binary classification models are constructed respectively; Inputting the second features into the binary classification model respectively to obtain feature importance scores, and repeating the process for a preset number of times; The union of the features whose feature importance scores of each binary classification model are greater than the stability threshold is taken as the third feature.

6. The method for genotyping diffuse large B-cell lymphoma according to claim 4, wherein: The step of constructing a data set and training a classification model based on the feature matrix, the first genotype label, and the important feature set includes: Establishing a first classification model based on the important feature set and the first genotype label, wherein the first classification model is capable of completing the classification of the genotypes recorded in the first genotype set; Obtaining a preset feature set, and establishing a second classification model based on the auxiliary feature, the preset feature set, and the first genotype label, wherein the second classification model is capable of completing the classification of the genotypes recorded in the second genotype set; Training the first classification model and the second classification model respectively according to the training set, and iteratively adjusting parameters through a cross-validation set; The trained first classification model and the second classification model are integrated to complete the construction of the classification model.

7. The method for genotyping diffuse large B-cell lymphoma according to claim 6, wherein: The data set also includes a test set. The obtaining of the second genotype label output by the classification model includes: Inputting the feature matrix generated according to the test set into the second classification model to perform genotype classification; If the genotype classification of the sample can be determined according to the second classification model, generating a corresponding second genotype label according to the classification result of the second classification model; If the genotype classification of the sample cannot be determined according to the second classification model, the feature matrix is ​​input into the first classification model for further classification; and a corresponding second genotype label is generated according to the classification result of the first classification model.

8. The method for genotyping diffuse large B-cell lymphoma according to claim 6, wherein: The first genotype set includes ST2 and A53; the second genotype set includes MCD, BN2, EZB and N1.

9. A computer device, characterized in that: including processor and memory; The processor is configured to execute the computer program stored in the memory to implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images