A mutation signature prediction method, device, terminal device and storage medium

Abstract

The present application is applicable to the field of biological gene technology, and provides a prediction method, apparatus, terminal device and storage medium for mutation signatures. The prediction method includes: obtaining a first gene expression vector of somatic cells of a sample to be predicted, the first gene expression vector is used to describe the correspondence between the sample to be predicted and multiple gene expression levels; inputting the first gene expression vector into multiple constructed prediction models for prediction, and obtaining the prediction results output by each prediction model, the prediction model is used to describe the association between multiple gene expression levels and a mutation signature; determining the mutation signature of the somatic cells of the sample to be predicted according to the prediction results output by multiple prediction models. The mutation signature prediction method of the present application is not limited to the number of mutations in somatic cells, and has universality and accuracy.

Description

Technical Field

[0001] The present application belongs to the field of biological gene technology, and in particular, relates to a method, apparatus, terminal device and storage medium for predicting mutation signatures. Background Art

[0002] In recent years, the application of mutation signatures has made great progress, but the existing technology can only detect mutation signatures through somatic mutations. Specifically, the existing technology first counts the frequency of each mutation in somatic cells, and then uses various methods such as non-negative matrices to obtain the corresponding mutation signatures. However, when the number of somatic mutations is small, due to the sparsity of the mutation vector, the existing technology cannot accurately detect the corresponding mutation signature.

[0003] Therefore, when the number of somatic cell mutations is small, the existing technology has problems with the lack of universality and accuracy in predicting mutation signatures. Summary of the invention

[0004] The embodiments of the present application provide a mutation signature prediction method, apparatus, terminal device and storage medium, which solve the problem that the prediction of mutation signatures in the prior art is insufficient in universality and accuracy when the number of mutations in somatic cells is small.

[0005] A first aspect of an embodiment of the present application provides a method for predicting a mutation signature, comprising:

[0006] Acquire a first gene expression vector of a somatic cell of a sample to be predicted, where the first gene expression vector is used to describe a corresponding relationship between the sample to be predicted and expression amounts of multiple genes;

[0007] Inputting the first gene expression vector into the constructed multiple prediction models respectively for prediction, and obtaining the prediction results output by each prediction model, wherein the prediction model is used to describe the correlation between the expression levels of the multiple genes and a mutation signature;

[0008] According to the prediction results output by the multiple prediction models, the mutation signature of the somatic cells of the sample to be predicted is determined.

[0009] In one embodiment, the step of constructing a plurality of the prediction models comprises:

[0010] Obtaining cancer gene information and constructing a first data set and a second data set, wherein the first data set is a set of multiple first association relationship vectors, and the second data set is a set of multiple second gene expression vectors, wherein the first association relationship vector is used to describe the association relationship between the training sample and the multiple mutation signatures, and the second gene expression vector is used to describe the corresponding relationship between the training sample and the expression amounts of multiple genes;

[0011] Obtaining a linear regression value through a linear regression model based on the first data set and the second data set;

[0012] Based on the first data set and the fourth data set as training sets, and the third data set as a validation set, each machine learning model is trained until a first preset condition is met to stop the training and obtain each prediction model after training, wherein the fourth data set is a set of multiple third gene expression vectors, and the third gene expression vectors are used to describe the correspondence between the training samples and a preset number of gene expression quantities, and the preset number is the number of the linear regression values ​​of the second preset condition, and the third data set is a set of association relationships between multiple gene expression quantities and a mutation signature.

[0013] In one embodiment, the steps of obtaining cancer gene information and constructing a first data set and a second data set include:

[0014] Acquire the cancer gene information, wherein the cancer gene information includes a mutation dataset of somatic cells of the training sample, a gene expression dataset of the training sample, and a standard frequency dataset, wherein the standard frequency dataset is a frequency dataset of a mutation type corresponding to a mutation signature;

[0015] Obtaining a sample frequency dataset based on the somatic cell mutation dataset of the training sample, wherein the sample frequency dataset is a frequency dataset of mutation types of the training sample;

[0016] Based on the standard frequency data set and the sample frequency data set, obtaining the first data set by using a mutation type matrix calculation formula;

[0017] The second data set is obtained based on the gene expression data set of the training samples.

[0018] In one embodiment,

[0019] The mutation type matrix calculation formula is:

[0020] C n×i =A n×m B m×i

[0021] Among them, C n×i is the sample frequency dataset, representing a frequency dataset of i mutation types of n training samples;

[0022] A n×m is the first data set, representing the association relationship between n training samples and m mutation signatures;

[0023] B m×iis the standard frequency dataset, which represents a frequency dataset of m mutation signatures corresponding to i mutation types, wherein n, m, and i are all positive integers, n≧1, m≧1, and i≧1.

[0024] In one embodiment,

[0025] The machine learning model includes at least one of a random forest model, an extreme gradient boosting tree model and a support vector machine model.

[0026] In one embodiment, the first preset condition is:

[0027] The area under the ROC curve was ≥ 80%, and the accuracy was ≥ 80%.

[0028] In one embodiment, the second preset condition is:

[0029] The linear regression value whose sequence number is less than or equal to 600 in the sequence in which the linear regression values ​​are sorted from small to large.

[0030] A second aspect of an embodiment of the present application provides a mutation signature prediction device, comprising:

[0031] A vector acquisition module, used to acquire a first gene expression vector of a somatic cell of a sample to be predicted, wherein the first gene expression vector is used to describe a corresponding relationship between the sample to be predicted and a plurality of gene expression amounts;

[0032] A model prediction module, used for inputting the first gene expression vector into a plurality of constructed prediction models for prediction, and obtaining prediction results output by each prediction model, wherein the prediction model is used for describing the correlation between the expression levels of the plurality of genes and a mutation signature;

[0033] The result acquisition module is used to determine the mutation signature of the somatic cells of the sample to be predicted according to the prediction results output by the multiple prediction models.

[0034] A third aspect of an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a prediction method as described in any one of the contents of the first aspect is implemented.

[0035] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the prediction method as described in any one of the contents of the first aspect is implemented.

[0036] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0037] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0038] The present application provides a method for predicting mutation signatures, which predicts mutation signatures of somatic cells through gene expression vectors of somatic cells. It is not limited to the number of mutations in somatic cells and has universality and accuracy.

[0039] This application uses machine learning methods to construct the association between mutation signatures and gene expression, and verifies the universality of the association between somatic mutation signatures and all gene expression and dysregulated gene expression of non-coding RNA. It can help doctors or engineers quickly obtain the dysregulated expression of genes corresponding to specific mutation signatures, and build a prediction model based on RNA sequencing data to predict the type of mutation signature, and can also obtain carcinogens or biomarkers at the same time. On the one hand, it provides a molecular explanation for the screening and diagnosis of clinical cancer, and on the other hand, it studies the target functional non-coding RNA, which is conducive to the development of targeted cancer therapeutics. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 It is a flowchart of a method for predicting a mutation signature provided in one embodiment of the present application;

[0042] Figure 2 It is a schematic diagram of a process for constructing various prediction models provided in an embodiment of the present application;

[0043] Figure 3 It is a schematic diagram of various mutation processes leading to characteristic mutation signatures provided by an embodiment of the present application;

[0044] Figure 4 This is a schematic diagram of the contribution of each mutation signature to 9,096 patients provided in an embodiment of the present application;

[0045] Figure 5 This is a schematic diagram of the association between skin melanoma and lung adenocarcinoma and mutation signatures, respectively, provided in an embodiment of the present application;

[0046] Figure 6 This is a schematic diagram of the ACC and AUC heat map of the SVM-type model based on all genes provided in one embodiment of the present application;

[0047] Figure 7This is a heat map diagram of ACC and AUC of the RF model based on all genes provided in one embodiment of the present application;

[0048] Figure 8 This is a heat map diagram of ACC and AUC of an XGboost-type model based on all genes provided in an embodiment of the present application;

[0049] Fig. 9 It is a schematic diagram of the ACC and AUC distribution of the SVM, RF, and XGboost models based on all genes and a Venn diagram of the excellent models provided in an embodiment of the present application;

[0050] Fig.10 Schematic diagram of the ROC of the SBS4 (smoking signature) model in LUAD and the SBS4 (smoking signature) model in LUSC based on all genes provided in one embodiment of the present application;

[0051] Fig.11 Schematic diagram of ROC of the SBS2 model in UCS based on all genes provided in one embodiment of the present application;

[0052] Fig.12 It is a schematic diagram of the gene type composition of all excellent models provided in one embodiment of the present application;

[0053] Fig.13 It is a heat map diagram of ACC and AUC of the SVM-type model based on protein-coding genes provided in one embodiment of the present application;

[0054] Fig.14 It is a heat map diagram of ACC and AUC of the RF model based on protein-coding genes provided in one embodiment of the present application;

[0055] Fig.15 This is a heat map diagram of ACC and AUC of an XGboost-type model based on protein-coding genes provided in one embodiment of the present application;

[0056] Fig.16 It is a schematic diagram of the ACC and AUC distribution and excellent model Venn diagram of the SVM, RF, and XGboost models based on protein-coding genes provided in one embodiment of the present application;

[0057] Fig.17 Schematic diagram of the ROC of the SBS4 (smoking signature) model in LUAD based on protein-coding genes and the ROC of the SBS4 (smoking signature) model in LUSC provided in one embodiment of the present application;

[0058] Fig.18 Schematic diagram of ROC of the SBS2 model in UCS based on protein-coding genes provided in one embodiment of the present application;

[0059] Fig.19 It is a schematic diagram of the validation AUC and AUC for 587 excellent models based on the third data set provided in an embodiment of the present application;

[0060] Fig. 20 It is a schematic diagram of the verification confusion matrix of the RF model and the SVM model on the third data set provided in an embodiment of the present application;

[0061] Fig.21 is the accuracy, specificity and sensitivity of the SVM and RF models from the third data set provided by an embodiment of the present application, and a ROC diagram on the validation data set generated by the SVM model and the RF model on the third data set;

[0062] Fig. 22 It is a schematic diagram of the structure of a mutation signature prediction device provided in an embodiment of the present application; DETAILED DESCRIPTION

[0063] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0064] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0065] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0066] There are currently three main approaches to exploit mutational signatures in biomedical research and applications:

[0067] First, in cancer epidemiology, mutational signatures can quantify exposure to known carcinogens or identify previously unknown carcinogens.

[0068] Second, in cancer treatment, mutational signatures can be used as biomarkers for failure of different DNA repair pathways. This provides opportunities for existing targeted therapies, such as DNA repair enzyme (Poly ADP-ribose Polymerase, PARP) inhibitors, and preliminary results show that patients respond positively to these therapies.

[0069] Third, in basic DNA repair and mutagenesis studies, mutational signature analysis can be used to characterize the origin of individual somatic mutations. More specifically, it is now possible to probabilistically attribute mutational signatures to each somatic mutation in a cancer genome. This will provide a better understanding of mutagenic oncogenomic and epigenomic modifiers.

[0070] Therefore, mutational signatures are powerful tools that are frequently used in cancer epidemiology, cancer treatment, and basic science research. Mutational signatures can provide insights into tumorigenesis and may illuminate early prevention strategies.

[0071] Since mutation signatures can only be detected through somatic mutations, there is currently no comprehensive prediction of mutation signatures based on the association between somatic mutation types and gene expression levels.

[0072] The embodiment of the present application provides a method for predicting mutation signatures of somatic cells, by obtaining a first gene expression vector of the sample to be predicted based on the somatic cells of the sample to be predicted, the first gene expression vector is used to describe the correspondence between the sample to be predicted and multiple gene expression levels, and the first gene expression vector is input into multiple prediction models that have been constructed for prediction, and the prediction results output by the multiple prediction models are obtained. According to the prediction results output by the multiple prediction models, the mutation signature of the somatic cells of the sample to be predicted is determined. The present application solves the problem that the prior art cannot predict the mutation signature of somatic cells through the gene expression data of the sample to be predicted, and realizes the prediction of the mutation signature of somatic cells through the gene expression level of the sample to be predicted.

[0073] The embodiment of the present application uses machine learning methods to construct the association between mutation signatures and gene expression, verifies the universality of the association between somatic mutations and the dysregulated expression of all genes and non-coding RNAs, and can assist doctors or engineers to quickly obtain the dysregulated expression of genes corresponding to specific mutation signatures, and builds an excellent prediction model based on RNA sequencing data to predict the type of mutation signature, and can also obtain carcinogens or biomarkers at the same time. On the one hand, it provides a molecular explanation for the screening and diagnosis of clinical cancer, and on the other hand, it studies the target functional non-coding RNA, which is conducive to the development of cancer targeted therapeutic drugs.

[0074] The technical solution of the present application is described below through specific embodiments.

[0075] like Figure 1 As shown, the first aspect of the embodiment of the present application provides a method for predicting a mutation signature of a somatic cell, comprising:

[0076] S100, obtaining a first gene expression vector of a somatic cell of a sample to be predicted.

[0077] In a specific implementation, gene detection is performed on the somatic cells of the sample to be predicted that have mutated into cancer, and the gene expression data of the sample to be predicted is obtained by RNA sequencing in the somatic cells, and the somatic mutation data of the sample to be predicted is obtained by DNA sequencing in the somatic cells. For example, the sample to be predicted is a cancer patient, wherein the first gene expression vector is used to describe the correspondence between the sample to be predicted and the expression amounts of multiple genes. The gene expression data of 56,716 genes are obtained by RNA sequencing of somatic cells, so the first gene expression vector is a set or one-dimensional matrix of the correspondence between the sample to be predicted and the expression amounts of 56,716 genes. If the number of samples to be predicted is 1, the specific content of the first gene expression vector refers to the set of correspondences between patient 1 and the expression amounts of 56,716 genes in Table 5.

[0078] Human deoxyribonucleic acid (DNA) is a macromolecular polymer composed of deoxynucleotides, which are composed of bases, deoxyribose and phosphate. There are four types of bases: adenine (A), thymine (T), cytosine (C) and guanine (G).

[0079] Mutation type is a characteristic set of mutations in the bases of deoxynucleotides in human gene DNA. Currently, there are 96 mutation types. Among them, there are 6 basic types of base mutations: C→A, C→G, C→T, T→A, T→C, and T→G. Each mutation point has 1 nucleotide base upstream and 1 nucleotide base downstream. Each nucleotide base has 4 possibilities: A, T, C, and G. Therefore, there are 4×6×4=96 mutation types.

[0080] Specifically, part of the somatic mutation data set of each cancer patient is shown in Table 1. For example, the mutation position of chromosome chr1 of patient 1 is at sequence number 89063200, the direction is +, the basic type of base mutation is C→T, and the base mutation type is missense mutation.

[0081] Patient number chromosome Mutation location direction Reference base Mutated base Mutation type Patient 1 chr1 89063200 + C T Missense mutation Patient 1 chr8 29140137 + T C Missense mutation Patient 1 chr8 144504453 + A G Missense mutation Patient 1 chr10 129840932 + A G Synonymous mutation Patient 1 chr11 117409839 + T G Missense mutation Patient 2 chr1 86469105 + A C Missense mutation Patient 2 chr2 55673000 + T C Synonymous mutation Patient 2 chr2 178591239 + T A Synonymous mutation

[0082] Table 1

[0083] S200, inputting the first gene expression vector into a plurality of constructed prediction models for prediction, and obtaining prediction results output by each prediction model, wherein the prediction model is used to describe the correlation between the expression levels of a plurality of genes and a mutation signature.

[0084] Since the causes of mutation characteristics of 33 types of cancers caused by human somatic mutations are different, and there are cases where the same type of cancer has multiple mutation characteristics, and humans have approximately 56,716 genes, a certain model cannot accurately predict so many complex associations, so it is necessary to combine multiple machine learning models for prediction and then select the prediction results. The constructed prediction models are encapsulated in a prediction model tool, and the gene expression vector of at least one sample to be predicted is input or uploaded to the preset entry, window or variable in the prediction model tool. Each prediction model outputs at least one prediction result according to the mutation type of the type of cancer suffered by the sample to be predicted. Multiple prediction results can avoid the deviation of the prediction results caused by errors in some models or other influencing factors, and facilitate the selection of the most appropriate prediction result from multiple prediction results.

[0085] In a specific implementation, the first gene expression vector of the sample to be predicted is input into a plurality of pre-built prediction models for computational prediction. In one embodiment, the prediction models include: Random Forests (RF), eXtreme Gradient Boosting (XGboost) and Support Vector Machine (SVM). In actual implementation, it is not limited to the above three machine learning models, and a suitable machine learning model is selected according to whether the prediction results meet the requirements of the evaluation index.

[0086] like Figure 2 As shown, in one embodiment, the steps of constructing multiple prediction models include:

[0087] S210, obtaining cancer gene information and constructing a first data set and a second data set.

[0088] Among them, cancer gene information of training samples is obtained, and a first data set and a second data set for training a machine learning model are constructed, the first data set is a set of multiple first association relationship vectors, the first association relationship vectors are used to describe the association relationship between the training samples and multiple mutation signatures, and the second data set is a set of multiple second gene expression vectors, the second gene expression vectors are used to describe the correspondence between the training samples and multiple gene expression amounts.

[0089] The processes that cause somatic mutations include enzymatic modification, mutagens, gene editing errors, and other lifestyle habits such as smoking. Mutagens include mutagens caused by internal factors in the human body and external factors such as drugs. External factors include physical factors such as ultraviolet radiation, chemical factors such as aromatic amines, and biological factors such as viruses. Gene editing errors include errors caused by DNA repair and DNA replication. These characteristic types of somatic mutations are called mutation signatures, such as mutation signatures of internal mutation processes, DNA repair, tobacco, ultraviolet radiation, polymerase mutations, chemotherapy, external carcinogens, and unknown carcinogens.

[0090] In one embodiment, obtaining cancer gene information and constructing a first data set and a second data set includes:

[0091] S211, obtaining cancer gene information, the cancer gene information includes a somatic mutation dataset of a training sample, a gene expression dataset of the training sample, and a standard frequency dataset, the standard frequency dataset being a frequency dataset of a mutation type corresponding to a mutation signature.

[0092] Specifically, the gene expression data of 10,274 cancer patients of 33 cancer types and the somatic mutation data of 10,179 patients (GRCh38 version) were downloaded from the Cancer Genome Atlas database (TCGA), and the two data sources were fused with each other, and the gene expression data set and somatic mutation data set of 9,096 cancer patients after fusion were used as the gene expression data set of the training sample and the somatic mutation data set of the training sample, respectively. The number of mutation signatures is calculated based on the somatic mutation data of the training sample and the type of mutation signature. In this embodiment, preferably, the number of mutation signatures is less than or equal to 49. In actual implementation, the number of mutation signatures is not limited to 49 mutation signatures, and according to the needs of the prediction model, it can also be 50, 75, 100 other numbers of mutation signatures.

[0093] In this embodiment, the calculation process, distribution, and related properties of the mutation signature of the training sample are analyzed. Specifically, various mutation processes may lead to characteristic mutation signatures (such as Figure 3 ), the contribution of each mutation signature to 9096 patients (as shown in Figure 4 ), parsing each cancer type by category of contributing mutation signatures (not shown), association of skin melanoma (SKCM) and lung adenocarcinoma (LUAD) with mutation signatures (as shown in Figure 5As shown in Figure 3, UV damage is the main mutation signature in SKCM and smoking-related signature is the main mutation signature in LUAD. The training samples may show associations to multiple mutation signatures, but only the mutation signature with the largest contribution is counted.

[0094] Since the frequencies of the 96 mutation types in different cancers are different, the frequencies of the 96 mutation types can be combined as a fixed mutation pattern to characterize a certain type of cancer. Cancer develops from somatic mutations. The characteristic type of somatic mutation is a mutation signature, so each mutation signature is a collection of frequency data of the 96 mutation types. The mutation type frequency is also called the trinucleotide motif frequency or mutation motif frequency. Among them, the frequency of a mutation type is the ratio of the number of times the mutation type appears in a training sample to the total number of times the 96 mutation types appear.

[0095] The frequency data set of 49 mutation signatures corresponding to mutation types was downloaded from the Catalogue Of Somatic Mutations In Cancer (COSMIC) database to obtain the standard frequency data set. The standard frequency data set is a frequency data set of 96 mutation types corresponding to 49 mutation signatures. The standard frequency data set is also a 49×96 probability matrix, denoted as B∈R m×96 , where m = 49. Part of the content of the standard frequency data set is shown in Table 2.

[0096]

[0097] Table 2

[0098] S212, obtaining a sample frequency dataset based on the somatic cell mutation dataset of the training sample, where the sample frequency dataset is a frequency dataset of mutation types of the training sample.

[0099] In one embodiment, a frequency dataset of mutation types of multiple training samples is obtained based on the mutation dataset of somatic cells of the training samples, thereby obtaining a sample frequency dataset. The sample frequency dataset is a frequency dataset of 96 mutation types of n training samples. The sample frequency dataset is also an n×96 matrix, denoted by C∈R n×96 , where n is a positive integer. In this embodiment, n is 9096, and part of the content of the sample frequency data set is shown in Table 3.

[0100]

[0101] Table 3

[0102] S213, based on the standard frequency data set and the sample frequency data set, obtain a first data set through a mutation type matrix calculation formula.

[0103] In one embodiment, a first data set is obtained by performing matrix operation transformation based on the standard frequency data set and the sample frequency data set through a mutation type matrix calculation formula. The first data set is a set of n first association relationship vectors. The first association relationship vector is used to describe the association relationship between the training sample and the m mutation signatures. The first data set is also an n×m matrix, denoted as A n×m , where n=9096 and m=49. Part of the data in the first data set is shown in Table 4.

[0104]

[0105] Table 4

[0106] The calculation formula of mutation type matrix is:

[0107] C n×i =A n×m B m×i

[0108] Among them, C n×i is the frequency dataset of samples, representing the frequency dataset of i mutation types of n training samples;

[0109] A n×m represents the first data set, which represents the association between n training samples and m mutation signatures;

[0110] B m×i is a standard frequency dataset, which means that m mutation signatures correspond to i mutation types. n, m, and i are all positive integers, n≧1, m≧1, and i≧1.

[0111] S214, obtaining a second data set based on the gene expression data set of the training samples.

[0112] In one embodiment, a set of 9096 second gene expression vectors is obtained based on the gene expression data set of 9096 training samples, and the second gene expression vectors are used to describe the corresponding relationship between the training samples and the expression levels of 56716 genes, thereby obtaining a second data set. The contents of some gene expression vectors in the second data set are shown in Table 5.

[0113]

[0114] Table 5

[0115] S220: Obtain a linear regression value through a linear regression model based on the first data set and the second data set.

[0116] In one embodiment, based on a set of 9096 first association relationship vectors of a first data set, the first association relationship vector is used to describe the association relationship between a training sample and 49 mutation signatures, and a first association relationship vector y of the association relationship between a training sample and the 49 mutation signatures is obtained; based on a set of 9096 second gene expression vectors of a second data set, the second gene expression vector is used to describe the correspondence between a training sample and the expression amounts of 56716 genes, and a second gene expression vector x of the correspondence between a training sample and the expression amounts of 56716 genes is obtained.

[0117] The second gene expression vector x of a training sample and the first association vector y of a training sample are used to obtain a linear regression value p through a univariate linear regression model. The linear regression value p is used to describe the association between the second gene expression vector x of a training sample and the first association vector y of a training sample. The smaller the p value, the stronger the correlation between the change of the mutation signature and the change of the gene expression amount. Since there are 56716 gene expression amounts in each second gene expression vector, a total of 56716 p values ​​are obtained for one mutation signature among the 49 mutation signatures of a cancer type.

[0118] The univariate linear regression model is:

[0119] y=θ 0 +θ 1 x

[0120] Among them, θ 0 represents the intercept of the straight line;

[0121] θ 1 It represents the slope of the line;

[0122] x represents a second gene expression vector of a training sample in the second data set;

[0123] y represents a first association relationship vector of a training sample in the first data set.

[0124] S230, training each machine learning model based on the first data set and the fourth data set as training sets, and the third data set as a validation set, until the first preset condition is met, and then stopping the training to obtain each trained prediction model.

[0125] In one embodiment, the fourth data set is a collection of multiple third gene expression vectors, and the third gene expression vector is used to describe the correspondence between the training sample and the preset number of gene expression amounts, and the preset number is the number of linear regression values ​​of the second preset condition. The second preset condition is the linear regression value whose sequence number is less than or equal to Z in the sequence of linear regression values ​​sorted from small to large, and Z is a positive integer. In this embodiment, preferably, Z takes a value of 600, that is, the preset number is 600. This embodiment does not make a specific restriction on the value of Z, and it is selected according to the prediction accuracy of the model, for example, Z is a positive integer such as 300, 400, 500, 800, 1000, etc.

[0126] The linear regression p values ​​are sorted in ascending order from small to large, and the first 600 linear regression p values ​​in the ascending order are selected, and then 600 second correlation vectors are obtained based on these 600 linear regression p values. The second correlation vector is used to describe the correlation between the second gene expression vector and the first correlation vector. Therefore, the corresponding 600 gene expression quantities are obtained from the 600 second correlation vectors. These 600 gene expression quantities are the gene expression quantities with the strongest correlation with a mutation signature among the 56716 gene expression quantities. Obtaining 600 gene expression quantities is equivalent to obtaining the corresponding gene expression.

[0127] In one embodiment, the third gene expression vector is used to describe the correspondence between the training sample and the 600 gene expression levels, and the fourth data set is a set of 9096 third gene expression vectors. The fourth data set can also be represented by a 9096×600 matrix.

[0128] The random forest machine learning model (RF model), the extreme gradient boosting tree machine learning model (XGboost model) and the support vector machine machine learning model (SVM model) are used to simulate the correlation between mutation signatures and gene expression. The training model is recorded as:

[0129] y=f(X)

[0130] Among them, f(·) represents different machine learning prediction models;

[0131] X represents the fourth data set, that is, X is a set of 9096 third gene expression vectors;

[0132] y represents a first association relationship vector of a training sample in the first data set.

[0133] The first data set and the fourth data set are used as training sets to train the random forest machine learning model (RF model), the extreme gradient boosting tree machine learning model (XGboost model) and the support vector machine machine learning model (SVM model) respectively until the first preset condition is met, and then the validation set is used to verify the trained machine learning models until the verification results also meet the first preset condition to obtain the trained prediction models.

[0134] The validation set is the third data set, and the third data set is a collection of correlation relationships between multiple gene expression levels and a mutation signature.

[0135] In one embodiment, an independent gene expression dataset (GSE29016) of lung cancer patients with known smoking status is downloaded from the Gene Expression Omnibus (GEO) to form a third dataset, which is a collection of associations between multiple gene expression levels and mutation signatures of smoking status, and is used to verify whether the association between smoking-related mutation signatures and the expression levels of each gene in the training sample meets the first preset condition.

[0136] In this embodiment, preferably, the first preset condition is: area under curve (ORArea under the Curve of ROC, AUC) ≧ 80%, and accuracy (Accuracy, ACC) ≧ 80%, and the model that meets the first preset condition is identified as an excellent model.

[0137] Furthermore, when training and validating each machine learning model, a five-fold cross validation is used for the training and validation of all machine learning models. In the process of training and validating the model, mutation signatures greater than the median coefficient are recorded as positive samples, and those less than or equal to the median coefficient are recorded as negative samples.

[0138] According to the above content, based on 33 cancer types, 49 mutation signatures, and 3 machine learning models, there are 33×49×3=4851 initial machine learning models.

[0139] For the prediction model of the i-th mutation signature of a cancer type, the gene expression vector of the input training sample in this embodiment is predicted using three different prediction models, which are denoted as:

[0140] S i =f i svm ,f i rf ,f i xgb ]

[0141] Optionally, all training samples of the third data set can be randomly divided into two sets, 80% of the data forming a training set and the remaining 20% ​​of the data forming a validation set, the training set is used to train and select each machine learning model, and the validation set is used to evaluate the performance of each machine learning model.

[0142] After screening, 587 excellent prediction models were finally obtained, and the collection of excellent prediction models S e Denoted as:

[0143] S e ={s∈S|AUC(s)≥0.8&Acc(s)≥0.8}

[0144] Optionally, in this embodiment, two sets of 56716 gene expression levels and protein-coding genes therein are used for modeling training respectively.

[0145] For the total gene expression set of 56,716, 374 of the 996 classification problems were successfully solved by at least one of the three machine learning models. Figure 6 , Figure 7 , Figure 8 As shown, the SVM model performed best among 372 excellent models, while the RF model ranked second among 74 excellent models, and the XGBoost model had 29 excellent models. Each cell in the heat map represents a model defined by cancer type and mutation signature. If both ACC and AUC are higher than 0.8, the excellent model will be represented by "√" according to the average of AUC and ACC. Otherwise, the model will be considered not excellent and represented by a uniform solid color. Fig. 9 Schematic diagram of the ACC and AUC distribution of the SVM, RF, and XGboost models based on all genes provided in this embodiment and the Venn diagram of the excellent model.

[0146] Using SVM model analysis, it was found that the number of excellent models was highly unbalanced between different cancer types. For some cancer types, such as ovarian cancer (OVarian cancer, OV), kidney clear cell carcinoma (KIRC) and kidney papillary cell carcinoma (KIRP), the excellent models include prediction models that can predict more than 40 mutation signatures. On the contrary, for example, uterine carcinosarcoma (UCS) and uveal melanoma (UVM), the excellent models can only predict less than 5 mutation signatures.

[0147] The differences between cancer types reflect the heterogeneity within cancer, where cancer types differ in mutation and gene expression profiles. For the entire set of 56,716 genes, 26 pairs of cancer-mutation signature associations were excellent predictors in all three categories of prediction models.

[0148] like Fig.10 , Fig.11 As shown, this example lists the ROC curves corresponding to the three association prediction models that obtained the best performance in the five-fold cross validation, including: Single Base Substitutions 4 (SBS4, corresponding to the smoking signal) mutation signature in lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUng Squamous cellCarcinoma, LUSC), SBS2 (corresponding to AID / APOBEC protein) mutation signature in uterine carcinosarcoma (Uterine Carcinosarcoma, UCS). When using the SVM model for prediction, these classification schemes inputting the entire gene library achieved AUC>0.94.

[0149] like Fig.12 As shown, this embodiment also examines the gene type composition of the features used in the SVM prediction model, and determines that a large amount of non-coding RNA (Ribonucleic Acid) is used in the excellent model. In many cases, the proportion of non-coding RNA is higher than that of protein-coding RNA. Long non-coding RNA (Long non-coding RNA, LncRNA) and pseudogenes are the two most important types of non-coding RNA used by the SVM model. This result suggests that somatic mutations substantially affect non-coding RNA gene expression.

[0150] For the protein-coding gene set, when the gene set only includes protein-coding genes, only 84 prediction models in all 996 model trainings reached the standard of excellent models. Fig.13 , Fig.14 , Fig.15 As shown, the SVM model has 84 excellent prediction models, the RF model has 20 excellent prediction models, and the XGBoost model has 8 excellent prediction models. Among them, the SVM model obtained the highest ACC and AUC. Each cell in the heat map represents a model defined by cancer type and mutation signature. If both ACC and AUC are higher than 0.8, the excellent model will be represented by "√" according to the average value of AUC and ACC. Otherwise, the model will be considered not excellent and represented by a uniform solid color. Fig.16 A schematic diagram of the ACC and AUC distribution of the SVM, RF, and XGboost models based on protein-coding genes provided in this embodiment and a Venn diagram of the excellent models;

[0151] For the protein-coding gene set, there are 7 pairs of cancer-mutation signature associations that are excellent predictive models in all three types of prediction models. Fig.17 , Fig.18 As shown, the ROC curves of three high-performance models are listed, including the prediction model for gene expression associated with SBS4 (corresponding to smoking signature) mutation signature in lung adenocarcinoma (LUAD), the prediction model for gene expression associated with SBS4 (corresponding to smoking signature) mutation signature in lung squamous cell carcinoma (LUSC), and the prediction model for gene expression associated with SBS2 (AID / APOBEC) mutation signature in uterine carcinosarcoma. The AUC values ​​obtained by the three best prediction models were compared in the set of 56,716 all genes and the set of protein-coding genes. Compared with the set of 56,716 all genes, the AUC values ​​of the protein-coding gene set were consistently lower, so non-coding genes are essential features for enhancing model performance.

[0152] A third data set of lung cancer independent expression data sets with known mutation signatures of smoking status was used to validate the RF prediction model and SVM prediction model of gene expression in excellent lung squamous cell carcinoma (LUSC) associated with SBS4 mutation signature. Since the SBS4 mutation signature is significantly associated with the amount of smoking, this example specifies smoking status as a surrogate for the SBS4 mutation signature.

[0153] like Fig. 20 , Fig.21 As shown, the training samples in the third data set are divided into 56 smokers and 10 never smokers, that is, the third data set is divided into SBS4 positive and SBS4 negative, positive corresponds to smoking, and negative corresponds to non-smoking. The overall confusion matrix of the verification results of the RF prediction and SVM prediction models are compared respectively. The verification accuracy ACC obtained by the RF prediction model is 84.8%, the sensitivity is 80.0% and the specificity is 85.7%. The verification accuracy ACC obtained by the SVM prediction model is 80.3%, the sensitivity is 80.0% and the specificity is 80.4%, and the ROC curves of the RF prediction and SVM prediction models are compared.

[0154] like Fig.19 As shown, the third dataset was used to verify the performance of 587 prediction models of 31 cancer types using the method of 80% training set and 20% validation set, and the ACC and AUC of all prediction models were greater than 80%.

[0155] Among the three machine learning models selected in this embodiment, the SVM prediction model obtained the highest AUC value, and the performance of the RF prediction model and the XGboost prediction model was moderate. On average, the gene expression of each cancer type can predict about 65% of the known mutation signatures. The prediction results show that the regulatory effect between somatic mutations and gene expression changes is not universal, that is, not every somatic mutation in the sample to be predicted can affect gene expression.

[0156] S300, determining a mutation signature of somatic cells of a sample to be predicted according to prediction results output by a plurality of prediction models.

[0157] In one embodiment, the optimal prediction model among the RF prediction model, the XGboost prediction model and the SVM prediction model outputs a prediction result. According to the prediction result, the mutation signature of the somatic cells of the sample to be predicted is determined so as to discover the distribution of the mutation signature of the somatic cells of the sample to be predicted, which is beneficial to finding the source of somatic mutations and is beneficial to the development of targeted cancer therapeutics.

[0158] The present embodiment provides a method for predicting a mutation signature of somatic cells, which obtains a first gene expression vector of a sample to be predicted based on the somatic cells of the sample to be predicted, and inputs the first gene expression vector into multiple constructed prediction models for prediction, thereby obtaining prediction results output by multiple prediction models. According to the prediction results output by the multiple prediction models, the mutation signature of the somatic cells of the sample to be predicted is determined, which is not limited to the number of mutations in the somatic cells and has universality and accuracy.

[0159] This example also uses machine learning methods to construct the association between mutation signatures and gene expression, and verifies the universality of the association between somatic mutation signatures and all gene expression and dysregulated gene expression of non-coding RNA, which can help doctors or engineers quickly obtain the dysregulated expression of genes corresponding to specific mutation signatures, and build an excellent prediction model based on RNA sequencing data to predict the type of mutation signature, and can also obtain carcinogens or biomarkers at the same time. On the one hand, it provides a molecular explanation for the screening and diagnosis of clinical cancer, and on the other hand, it studies the target functional non-coding RNA, which is conducive to the development of cancer targeted therapeutic drugs.

[0160] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0161] Corresponding to the prediction method described in the above embodiment, Fig. 22 A structural block diagram of a somatic cell mutation signature prediction device provided in an embodiment of the present application is shown. For ease of explanation, only the part related to the embodiment of the present application is shown.

[0162] A second aspect of an embodiment of the present application provides a mutation signature prediction device, comprising:

[0163] A vector acquisition module 10 is used to acquire a first gene expression vector of a somatic cell of a sample to be predicted, where the first gene expression vector is used to describe a corresponding relationship between the sample to be predicted and a plurality of gene expression levels;

[0164] A model prediction module 20, used to input the first gene expression vector into a plurality of constructed prediction models for prediction, and obtain prediction results output by each prediction model, wherein the prediction model is used to describe the correlation between the expression levels of a plurality of genes and a mutation signature;

[0165] The result acquisition module 30 is used to determine the mutation signature of the somatic cells of the sample to be predicted according to the prediction results output by the multiple prediction models.

[0166] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0167] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0168] A third aspect of an embodiment of the present application provides a terminal device, including:

[0169] A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the prediction method as described in any one of the first aspects above is implemented.

[0170] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the prediction method as described in any one of the contents of the first aspect above is implemented.

[0171] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0172] The mutation signature prediction method provided in the embodiment of the present application can be applied to terminal devices such as mobile phones, tablet computers, wearable devices, vehicle-mounted devices, augmented reality (AR) / virtual reality (VR) devices, laptop computers, ultra-mobile personal computers (UMPC), netbooks, personal digital assistants (PDA), etc. The embodiment of the present application does not impose any restrictions on the specific type of the terminal device.

[0173] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc.

[0174] The computer-readable medium may include at least: any entity or device capable of carrying computer program codes to a camera / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electric carrier signal, a telecommunication signal, and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disk. In some jurisdictions, according to legislation and patent practice, a computer-readable medium cannot be an electric carrier signal or a telecommunication signal.

[0175] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0176] Those of ordinary skill in the art will appreciate that the single modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0177] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for predicting a mutation signature, characterized in that: include: Acquire a first gene expression vector of a somatic cell of a sample to be predicted, where the first gene expression vector is used to describe a corresponding relationship between the sample to be predicted and expression amounts of multiple genes; Inputting the first gene expression vector into the constructed multiple prediction models respectively for prediction, and obtaining the prediction results output by each prediction model, wherein the prediction model is used to describe the correlation between the expression levels of the multiple genes and a mutation signature; The step of constructing multiple prediction models includes: obtaining cancer gene information and constructing a first data set and a second data set, wherein the first data set is a set of multiple first association relationship vectors, and the second data set is a set of multiple second gene expression vectors, wherein the first association relationship vector is used to describe the association relationship between the training sample and multiple mutation signatures, and the second gene expression vector is used to describe the correspondence between the training sample and multiple gene expression amounts; obtaining a linear regression value through a linear regression model based on the first data set and the second data set; training each machine learning model based on the first data set and the fourth data set as the training set, and the third data set as the validation set, until the first preset condition is met, the training is stopped to obtain each prediction model after training, wherein the fourth data set is a set of multiple third gene expression vectors, and the third gene expression vector is used to describe the correspondence between the training sample and a preset number of gene expression amounts, and the preset number is the number of the linear regression values ​​of the second preset condition, and the third data set is a set of association relationships between multiple gene expression amounts and a mutation signature; the machine learning model includes at least one of a random forest model, an extreme gradient boosting tree model, and a support vector machine model; The steps of obtaining cancer gene information and constructing the first data set and the second data set include: obtaining the cancer gene information, the cancer gene information including the somatic mutation data set of the training sample, the gene expression data set of the training sample and the standard frequency data set, the standard frequency data set being the frequency data set of the mutation type corresponding to the mutation signature; obtaining a sample frequency data set based on the somatic mutation data set of the training sample, the sample frequency data set being the frequency data set of the mutation type of the training sample; obtaining the first data set based on the standard frequency data set and the sample frequency data set by a mutation type matrix calculation formula, the mutation type matrix calculation formula being: ,in, is the sample frequency dataset, representing a frequency dataset of i mutation types of n training samples; is the first data set, representing the association relationship between n training samples and m mutation signatures; is the standard frequency data set, indicating m A frequency dataset of i mutation types corresponding to mutation signatures, wherein n, m, and i are all positive integers, n≧1, m≧1, and i≧1; and obtaining the second dataset based on the gene expression dataset of the training samples; According to the prediction results output by the multiple prediction models, the mutation signature of the somatic cells of the sample to be predicted is determined.

2. The prediction method according to claim 1, characterized in that: The first preset condition is: the area under the ROC curve ≧80%, and the accuracy ≧80%.

3. The prediction method according to claim 1, characterized in that: The second preset condition is: the linear regression value whose sequence number is less than or equal to Z in the sequence of the linear regression values ​​arranged from small to large, where Z is a positive integer.

4. A prediction device for mutation signatures, characterized in that: include: A vector acquisition module, used to acquire a first gene expression vector of a somatic cell of a sample to be predicted, wherein the first gene expression vector is used to describe a corresponding relationship between the sample to be predicted and a plurality of gene expression amounts; A model prediction module is used to input the first gene expression vector into the constructed multiple prediction models for prediction, and obtain the prediction results output by each prediction model, wherein the prediction model is used to describe the association between the multiple gene expression levels and a mutation signature; the steps of constructing multiple prediction models include: obtaining cancer gene information and constructing a first data set and a second data set, wherein the first data set is a collection of multiple first association relationship vectors, and the second data set is a collection of multiple second gene expression vectors, wherein the first association relationship vector is used to describe the association between the training sample and the multiple mutation signatures, and the second gene expression vector is used to describe the corresponding relationship between the training sample and the multiple gene expression levels; based on the first data set and the The second data set obtains a linear regression value through a linear regression model; based on the first data set and the fourth data set as training sets, and the third data set as a validation set, each machine learning model is trained until the first preset condition is met to stop the training and obtain each prediction model after training, wherein the fourth data set is a set of multiple third gene expression vectors, and the third gene expression vectors are used to describe the corresponding relationship between the training samples and a preset number of gene expression quantities, the preset number is the number of the linear regression values ​​of the second preset condition, and the third data set is a set of the association relationship between multiple gene expression quantities and a mutation signature; the machine learning model includes at least one of a random forest model, an extreme gradient boosting tree model, and a support vector machine model; The steps of obtaining cancer gene information and constructing the first data set and the second data set include: obtaining the cancer gene information, the cancer gene information including the somatic mutation data set of the training sample, the gene expression data set of the training sample and the standard frequency data set, the standard frequency data set being the frequency data set of the mutation type corresponding to the mutation signature; obtaining a sample frequency data set based on the somatic mutation data set of the training sample, the sample frequency data set being the frequency data set of the mutation type of the training sample; obtaining the first data set based on the standard frequency data set and the sample frequency data set by a mutation type matrix calculation formula, the mutation type matrix calculation formula being: ,in, is the sample frequency dataset, representing a frequency dataset of i mutation types of n training samples; is the first data set, representing the association relationship between n training samples and m mutation signatures; is the standard frequency data set, indicating m A frequency dataset of i mutation types corresponding to mutation signatures, wherein n, m, and i are all positive integers, n≧1, m≧1, and i≧1; and obtaining the second dataset based on the gene expression dataset of the training samples; The result acquisition module is used to determine the mutation signature of the somatic cells of the sample to be predicted according to the prediction results output by the multiple prediction models.

5. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the prediction method according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the prediction method according to any one of claims 1 to 3 is implemented.

Images