Method for discriminating transgenic events with unknown foreign fragment insertions based on targeted sequencing

Abstract

The application discloses a method for identifying unknown exogenous fragment insertion transgenic events based on targeted sequencing, and relates to the field of transgenic technology identification. The application can accurately identify the insertion position of a transgenic event, and can effectively distinguish transgenic sequences from non-transgenic sequences even if an exogenous gene is inserted into a repeated sequence region. Through classification and counting of the sequencing data after quality control, and in combination with a machine learning model, the proportion fluctuation range is optimized, the accuracy of determining the transgenic event is significantly improved, the introduction of a Bayesian optimization method can further improve the accuracy and reliability of the model in judging the nature of the transgenic event, and deviation and error in the traditional method are avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of transgenic technology identification, in particular to a method for identifying unknown exogenous fragment insertion transgenic events based on targeted sequencing. BACKGROUND

[0002] Transgenic technology has the characteristics of strong targeting, high breeding efficiency and short cycle, and has become one of the important means of plant molecular breeding. Determining whether a transgenic event is homozygous or heterozygous is of great significance for ensuring the genetic stability of transgenic plants, improving breeding efficiency, ensuring consistent trait expression, conducting accurate application research, and meeting regulatory and safety requirements.

[0003] During the growth and development of crops, they are often affected by pests, weeds and diseases. The current control measures mainly rely on the use of large amounts of chemical pesticides and herbicides, which not only seriously affects the ecological environment and biodiversity, increases production costs and labor intensity, but also increases the risk of human poisoning. The most economical and effective method is to exploit and utilize pest-resistant, herbicide-tolerant and disease-resistant genes for variety breeding. In the breeding process, homozygous transgenic plants, which contain exogenous genes on both homologous chromosomes, are more genetically stable and can be directly used for production or further breeding. In the commercialization and regulatory approval process of transgenic products, homozygous plants are more likely to be approved, so it is of great significance to distinguish between homozygous and heterozygous transgenic events.

[0004] Most of the existing methods for judging the insertion of transgenic events require the specific insertion position of the exogenous sequence on the reference genome as a starting point. Once the exogenous sequence is inserted into a repetitive fragment, the capture efficiency and accuracy will be greatly reduced. The present application can deal with this situation and does not need to rely on the insertion position information on the reference genome to judge most materials. In addition, the existing technology also has certain bottlenecks in analyzing sequencing data. For the sequence of the transformant of transgenic crops, the alignment algorithm mainly identifies the transgenic event by matching the exogenous sequence with the reference genome. However, when the exogenous gene is inserted into a complex repetitive sequence region, multiple alignment often occurs, further exacerbating the ambiguity of the insertion position of the exogenous gene. SUMMARY

[0005] The present application provides a method for identifying unknown exogenous fragment insertion transgenic events based on targeted sequencing, which aims to provide a solution to the problem of inaccurate identification of transgenic events with uncertain insertion positions.

[0006] The method for identifying unknown exogenous fragment insertion transgenic events based on targeted sequencing comprises the following steps:

[0007] S1. Confirm the junction position of the foreign sequence and the reference genome sequence in the transformant sequence of the target transgenic event by comparing the model, which is the breakpoint;

[0008] S2. Perform second-generation sequencing on the sample to be tested to obtain original clean reads, and compare them with the transformant sequence to generate a BAM file; quality control the alignment information in the BAM file to ensure that the mapping quality is ≥40 and the alignment only includes one CIGAR operation;

[0009] S3. According to the alignment results after quality control, the reads are classified into three types: Ref_read, Cross_reads and GMO_reads; count the reads of the three types, and correct the count results by taking the logarithm of 10 as the transformation method;

[0010] S4. Calculate the proportion between different types of reads, and optimize the fluctuation range of the proportion through machine learning to determine the properties of the material.

[0011] Further, in step S3, the specific types of reads classified into three types of Ref_read, Cross_reads and GMO_reads are as follows:

[0012] Reads that are completely aligned with the reference genome part;

[0013] Reads that are partially aligned with the reference genome and partially aligned with the foreign sequence part, i.e. reads that cross the breakpoint;

[0014] Reads that are completely aligned with the foreign sequence part.

[0015] Further, the specific classification process of reads classified into three types of Ref_read, Cross_reads and GMO_reads in step S3 is as follows:

[0016] When the CIGAR string only includes matching operations and is completely aligned to the reference genome, the read is Ref_reads;

[0017] When the CIGAR string includes matching and insertion / deletion operations, and the two regions of the alignment come from the reference genome and the foreign sequence respectively, the read is Cross_reads;

[0018] When the CIGAR string only includes matching operations and is completely aligned to the foreign sequence, the read is GMO_reads.

[0019] Further, step S4 specifically includes the following sub-steps:

[0020] S401. According to the classification result of the reads, the number of each type of reads is calculated, the proportion between different types of reads is obtained, and the proportion between different types of reads is taken as a feature of the model input;

[0021] S402. Maximize the classification accuracy as the optimization objective;

[0022] S403. Define the hyperparameter space of Bayesian optimization;

[0023] S404. Construct the objective function, adjust the classification accuracy by selecting different proportion thresholds and fluctuation ranges;

[0024] S405. Use Gaussian process as a surrogate model to predict the behavior of the objective function, and guide the next evaluation point according to the current evaluation result; take the expected improvement as the acquisition function to select an optimal hyperparameter combination that can improve the performance of the model; update the surrogate model through iteration, and use the acquisition function to guide Bayesian optimization to find the optimal hyperparameters; after each iteration, the evaluation result returned by the objective function;

[0025] S406. Obtain the optimal parameter combination through Bayesian optimization, and use the optimized proportion relationship to judge the transgenic properties of the material.

[0026] Further, in the step S402, the optimization objective is optimized by an optimization function, the output of the optimization function is the accuracy of the classifier, and the optimized proportion relationship is used to distinguish the transgenic events.

[0027] Further, the calculation process of the accuracy is specifically calculated according to a confusion matrix composed of the number of true transgenic reads , the number of true non-transgenic reads , the number of non-transgenic reads incorrectly judged as transgenic , and the number of transgenic reads incorrectly judged as non-transgenic , that is:

[0028] ;

[0029] Wherein, the represents the calculation result of the accuracy, the represents the number of true transgenic reads, the represents the number of true non-transgenic reads, the represents the number of non-transgenic reads incorrectly judged as transgenic, and the represents the number of transgenic reads incorrectly judged as non-transgenic.

[0030] Further, in the step S404, the objective function is constructed according to the accuracy rate, that is:

[0031] ;

[0032]

[0033] wherein the objective function is constructed as: , wherein the objective function is constructed as: , and wherein the objective function is constructed as: and , respectively, wherein the objective function is constructed as: , respectively, wherein the objective function is constructed as: , respectively, wherein the objective function is constructed as:

[0034] Further, the calculation processes of the , , and are respectively shown as:

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] wherein the objective function is constructed as: , wherein the objective function is constructed as: , wherein the objective function is constructed as: , wherein the objective function is constructed as: , wherein the objective function is constructed as: , wherein the objective function is constructed as: represents a proportion threshold of reads crossing the breakpoint, i.e. the proportion of reads crossing the breakpoint among total reads, the represents the number of reads crossing the breakpoint, the represents a proportion threshold of reads crossing the breakpoint, i.e. the proportion of reads crossing the breakpoint among total reads.

[0040] Further, in the step S405, the expected improvement is taken as the acquisition function, and a calculation process of selecting an optimal combination of hyperparameters capable of improving the expected improvement of the model performance is represented as:

[0041] ;

[0042] wherein the represents the expected improvement of the model with the hyperparameter configuration , the represents the current optimal proportion of reads in the Bayesian optimization process, the represents the predicted classification accuracy of reads of the model with the hyperparameter configuration , the represents a cumulative distribution function of a standard normal distribution, used for calculating the probability of the objective function value being lower than the current optimal value, the represents the standard deviation of the prediction error of the model with the hyperparameter configuration , the represents a probability density function of a standard normal distribution, used for calculating the probability density of the gap between the current optimal value and the predicted mean value.

[0043] Further, in the step S405, the surrogate model is specifically represented as:

[0044] ;

[0045] wherein the represents the surrogate model, the represents a Gaussian process function, the represents the predicted classification accuracy of reads of the model with the hyperparameter configuration , the represents a kernel function.

[0046] The application has the following beneficial effects:

[0047] The application can accurately identify the insertion position of the transgenic event, can effectively distinguish the transgenic sequence from the non-transgenic sequence even if the exogenous gene is inserted into the repeated sequence region; by classifying and counting the sequencing data after quality control, and combining the machine learning model to optimize the proportion fluctuation range, the accuracy of determining the transgenic event is significantly improved; the introduction of the Bayesian optimization method can further improve the accuracy and reliability of the model in judging the nature of the transgenic event, and avoid the deviation and error in the traditional method. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The method flowchart of the method for determining unknown exogenous fragment insertion transgenic event based on targeted sequencing provided by the embodiment of the application is provided.

[0049] Figure 2 The flowchart of threshold value verification by experiment provided by the embodiment of the application is provided. DETAILED DESCRIPTION

[0050] The technical solutions of the application will be further described in detail below with reference to the accompanying drawings, but the protection scope of the application is not limited to the following description.

[0051] In order to make the purpose, technical solutions and advantages of the application clearer and more understandable, the application will be further described in detail in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application, that is, the described embodiments are only a part of the embodiments of the application, but not all the embodiments. The components of the embodiments of the application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0052] Therefore, the detailed description of the embodiments of the application provided in the drawings below is not intended to limit the scope of the claimed application, but only represents selected embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application. It should be noted that the relationship terms such as "first" and "second" and the like are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations.

[0053] Also, the term "comprise", "comprising", or any other variant thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0054] The features and characteristics of the present application are further described in detail below in connection with the embodiments.

[0055] wherein, as Figure 1 A method for identifying unknown exogenous fragment insertion transgenic events based on targeted sequencing, comprising the following steps:

[0056] S1. Confirming the intersection position of the exogenous sequence and the reference genome sequence in the transformant sequence of the target transgenic event by alignment model, which is the breakpoint;

[0057] S2. Sequencing the sample to be tested by second-generation sequencing to obtain original clean reads, and aligning the clean reads with the transformant sequence to generate a BAM file; quality control is performed on the alignment information in the BAM file to ensure that the mapping quality is ≥40, and the alignment only includes one CIGAR operation;

[0058] S3. According to the alignment results after quality control, the reads are classified into three types: Ref_read, Cross_reads and GMO_reads; count the reads of the three types, and correct the count results by taking the logarithm of 10 as the transformation method;

[0059] S4. Calculate the proportion between different types of reads, and optimize the fluctuation range of the proportion through experimental verification to determine the properties of the material.

[0060] Specifically, for step S1, the specific implementation principle is: through alignment, the intersection position of the exogenous sequence and the reference genome sequence in the transformant sequence of the target transgenic event is determined, which is called the breakpoint. In this step, a variety of alignment tools (such as BWA, Bowtie, Minimap2, etc.) are used to align clean reads with reference genome and GMO.fa. The alignment results are fused by weighted voting mechanism or weighted average method to ensure high accuracy of the alignment results, especially when dealing with complex regions (such as repetitive sequences, structural variation regions), multiple tools are used to reduce the error influence of a single tool.

[0061] For step S2, the specific implementation principle is: the sample to be tested is subjected to second-generation sequencing to obtain original clean reads, and is compared with GMO.fa to generate a BAM file. The comparison information in the BAM file is subjected to quality control to ensure that the mapping quality (Mapping Quality) is ≥40 and the comparison only contains one CIGAR operation. That is, the comparison is either a simple match without insertion or deletion, or a single insertion or deletion.

[0062] Further, in step S3, the reads are classified into three types of Ref_read, Cross_reads and GMO_reads, and the specific types are as follows:

[0063] The reads that are completely compared with the reference genome part;

[0064] The reads that are compared with the reference genome part and are compared with the exogenous sequence part, that is, the reads crossing the breakpoint;

[0065] The reads that are completely compared with the exogenous sequence part.

[0066] Further, as preferred in step S3 of the above embodiment, the "soft-clipping" operation (S) is introduced to process the partially compared reads close to the breakpoint. For example, 10S50M10S indicates that the front and rear parts of the read do not match the reference genome completely, but the middle part matches the reference genome completely, which can be considered as part of the reads crossing the breakpoint.

[0067] Further, in step S3, the reads are classified into three types of Ref_read, Cross_reads and GMO_reads, and the specific classification process is as follows:

[0068] When the CIGAR string only includes matching operations and is completely compared with the reference genome, the read is Ref_read;

[0069] When the CIGAR string includes matching and insertion / deletion operations, and the two regions of the comparison come from the reference genome and the exogenous sequence respectively, the read is Cross_read;

[0070] When the CIGAR string only includes matching operations and is completely compared with the exogenous sequence, the read is GMO_read.

[0071] Further, the flow of threshold value verification by experiment preferred in the above embodiment is proposed, and the specific implementation principle is as follows: Figure 2 That is:

[0072] (1) Determine the insertion position of the exogenous sequence on the transformant sequence:

[0073] By comparison, the intersection position of the exogenous sequence and the reference genome sequence in the transformant sequence of the target transgenic event (hereinafter referred to as GMO.fa) is determined, which is referred to as the breakpoint.

[0074] (2) Obtain the BAM file by sequencing:

[0075] The sample to be tested is subjected to second-generation sequencing, and the clean reads without any filtering are aligned with GMO.fa to obtain a BAM file. The information in the BAM file is quality controlled, the mapping quality of the default alignment is ≥40, and the alignment only contains one CIGAR operation, that is, the alignment is either a simple match without insertion or deletion, or a single insertion or deletion.

[0076] (3) Classify and count the reads that meet the conditions and pass the quality control.

[0077] All reads are specifically classified into the following three types:

[0078] reads that completely align to the reference sequence part (Ref_reads);

[0079] reads that align to part of the reference genome and part of the exogenous sequence, i.e., reads that cross the breakpoint (Cross_reads);

[0080] reads that completely align to the exogenous sequence part (GMO_reads). Count the above three types of reads. Considering the problem of skew distribution, use the logarithmic transformation method with base 10 to correct the counting results to reduce the influence of extreme values on the results and the skewness of the data.

[0081] (4) Calculate the proportion of different types of reads and optimize the fluctuation range of the proportion through model verification to determine the properties of the material.

[0082] The specific experimental verification threshold settings are as follows:

[0083] The target transgenic event X has two transformant sequences: RB (5' is the reference genome sequence, 3' is the exogenous sequence, the full length is 774 bp) and LB (5' is the exogenous sequence, 3' is the reference genome sequence, the full length is 726 bp). The breakpoint position of X on RB is 262 and the breakpoint position of X on LB is 256 by using blast software alignment. 25 real samples (10 homozygous positives, prefixed with P as the sample name; 10 heterozygous positives, prefixed with H as the sample name; 5 negative control samples, prefixed with N as the sample name) are selected, and the judgment threshold is set to be the theoretical value ± 0.2 for the experiment;

[0084] In the initial experiment, 8 positive heterozygous samples were misjudged as negative samples, and the correct discrimination rate was 68%. Due to the deviation of the theoretical value, the threshold fluctuation range needs to be continuously adjusted. After continuous attempts, the final judgment threshold is as follows: (1) the ratio of Ref_reads to Cross_reads is 0.7-1.4 (including the boundary, closed interval) and is considered to be a transgenic homozygous negative; (2) the ratio of Ref_reads to Cross_reads is 1.4-2.1 (left open and right closed interval) and is considered to be a transgenic heterozygous positive; (3) the ratio of GMO_reads to Cross_reads is 0.7-1.4 (open interval) and is considered to be a transgenic homozygous positive. Finally, 25 real samples (10 homozygous positives, prefixed with P as the sample name; 10 heterozygous positives, prefixed with H as the sample name; 5 negative control samples, prefixed with N as the sample name) are selected for the experiment.

[0085] Further, as a preferred threshold setting method of the above embodiment, a method of replacing the experimental verification process by machine learning is proposed. Specifically, the experimental verification process in step S4 is improved to be verified by machine learning, including the following steps:

[0086] S401. According to the classification result of reads, the number of each type of reads is calculated, the ratio between different types of reads is obtained, and the ratio between different types of reads is taken as the feature of model input;

[0087] S402. The maximum classification accuracy is taken as the optimization target;

[0088] S403. Define the hyperparameter space of Bayesian optimization;

[0089] S404. Construct the objective function, adjust the classification accuracy by selecting different proportion thresholds and fluctuation ranges;

[0090] S405. The Gaussian process is used as a surrogate model to predict the behavior of the target function, and the next evaluation point is guided according to the current evaluation result; the expected improvement is used as a collection function to select an optimal combination of hyperparameters that can improve the performance of the model; the surrogate model is updated through iteration, and the optimal hyperparameters are found by the Bayesian optimization guided by the collection function; after each iteration, the evaluation result returned by the target function;

[0091] S406. The optimal parameter combination is obtained by Bayesian optimization, and the optimized proportional relationship is used to judge the transgenic properties of the material.

[0092] Further, in step S402, the optimization target is optimized by an optimization function, and the output of the optimization function is the accuracy of the classifier. The optimized proportional relationship is used to distinguish transgenic events.

[0093] Further, the calculation process of the accuracy is calculated according to a confusion matrix composed of the number of true transgenic reads , the number of true non-transgenic reads , the number of non-transgenic reads incorrectly judged as transgenic , and the number of transgenic reads incorrectly judged as non-transgenic , that is:

[0094]

[0095] Among them, the represents the calculation result of the accuracy, the represents the number of true transgenic reads, the represents the number of true non-transgenic reads, the represents the number of non-transgenic reads incorrectly judged as transgenic, and the represents the number of transgenic reads incorrectly judged as non-transgenic.

[0096] Further, in step S404, the target function is constructed according to the accuracy, that is:

[0097]

[0098]

[0099] Among them, the represents the target function when the hyperparameter configuration is , the and represent the target function when the hyperparameter configuration is ​​the true number of transgenic reads, the true number of non-transgenic reads, the number of non-transgenic reads that are wrongly judged as transgenic, and the number of transgenic reads that are wrongly judged as non-transgenic, wherein the respectively represent the proportion threshold of reads that are completely aligned with the reference genome, the proportion threshold of reads that cross the breakpoint, and the proportion threshold of reads that are completely aligned with the exogenous sequence.

[0100] Further, the , , and computation processes are respectively represented as:

[0101] ;

[0102] ;

[0103] ;

[0104] ;

[0105] wherein the represents the proportion threshold of reads that are completely aligned with the exogenous sequence, and is used to reflect the insertion amount of the exogenous gene sequence in the transgenic event, the represents the number of reads related to the exogenous sequence that meet the conditions, i.e., the number of reads that are completely aligned with the exogenous sequence, in a specific interval, the represents the total number of all reads, the represents the proportion of reads that are completely aligned with the reference genome in the total reads, the represents the number of reads that are completely aligned with the reference genome sequence, i.e., reads that are completely matched with the reference genome, the represents the proportion threshold of reads that cross the breakpoint, i.e., the proportion of reads that cross the breakpoint in the total reads, the represents the number of reads that cross the breakpoint, the represents the proportion threshold of reads that cross the breakpoint, i.e., the proportion of reads that cross the breakpoint in the total reads.

[0106] Further, in the step S405, the expected improvement is taken as the collection function, and the computation process of the expected improvement of an optimal hyperparameter combination that can improve the performance of the model is represented as:

[0107] ;

[0108] wherein the represents the expected improvement of the hyperparameter configuration as the represents the current optimal reads ratio in the Bayesian optimization process, the represents the model predicted reads classification accuracy of the hyperparameter configuration as the represents the cumulative distribution function of the standard normal distribution, used to calculate the probability that the objective function value is lower than the current optimal value, the represents the standard deviation of the prediction error of the hyperparameter configuration as the represents the probability density function of the standard normal distribution, used to calculate the probability density of the gap between the current optimal value and the predicted mean value.

[0109] Specifically, when is close to or less than , and is large, the objective function under the current configuration has large uncertainty; if is close to , and is small, it means that the current configuration may have been very close to the optimal solution, and the optimization will gradually tend to converge.

[0110] Specifically, the expected improvement (EI) is used to optimize the configuration of the reads ratio, so as to more accurately judge the nature of the transgenic event; through the calculation of the expected improvement (EI) in the Bayesian optimization, the configuration that is most likely to bring improvement is selected from multiple candidate hyperparameters (such as the reads ratio threshold), which can reduce the number of experiments and accelerate the process of finding the optimal configuration.

[0111] Further, in the step S405, the surrogate model is specifically represented as:

[0112] ;

[0113] wherein the represents the surrogate model, the represents the Gaussian process function, the represents the model predicted reads classification accuracy of the hyperparameter configuration as the represents the kernel function.

[0114] Specifically, the Gaussian Process (GP) is a random process on the input space, which can be regarded as an infinite-dimensional vector, where each dimension represents a function value of a corresponding input point; for example, the kernel function selects the Matern kernel, which has more flexibility in controlling smoothness compared to the commonly used Gaussian (RBF) kernel, and the specific process is as follows:

[0115] The basic form of the Matern kernel is represented as:

[0116] ;

[0117] wherein the represents two different input vectors, i.e., the feature vectors of reads of different categories, the represents the smoothness parameter of the Matern kernel, the length scale parameter, which controls the range of influence between data points, and the Table shows the Matern kernel.

[0118] When approaches infinity, the Matern kernel approaches the RBF kernel, representing a very smooth function;

[0119] When , the Matern kernel represents a discontinuous function, in which case the kernel function has the weakest control over the smoothness between points, and the effect is similar to that of a linear kernel function;

[0120] When is larger, it means that the similarity range between data points is larger, i.e., the distance between two points has less effect on the kernel function value;

[0121] When is smaller, it means that the similarity range between data points is smaller, i.e., the distance between two points has a greater effect on the kernel function value.

[0122] Further, through the Matern kernel, the covariance value is calculated for each pair of data points, which is used to reflect the similarity between reads of different categories.

[0123] Further, the Gaussian Process regression model learns through the covariance matrix provided by the Matern kernel and predicts the proportional fluctuation of reads of different categories based on the training data. On this basis, Bayesian optimization is used to adjust the hyperparameters in the model, optimize the proportional fluctuation range and improve the accuracy of material property judgment.

[0124] The foregoing is considered as illustrative only of the principles of the application. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the application to the exact construction and operation described. Accordingly, all such variations are intended to be included within the scope of the present application as defined in the claims below and their equivalents.

Claims

1. A method for discriminating transgenic events with unknown foreign fragment insertions based on targeted sequencing, characterized in that, Includes the following steps: S1. By comparing the model, the intersection position of the exogenous sequence and the reference genome sequence in the transformant sequence of the target transgenic event is identified, and the position is the breakpoint; S2. Perform next-generation sequencing on the sample to be tested to obtain the original clean reads, and align them with the transformed sequences to generate a BAM file; Perform quality control on the alignment information in the BAM file to ensure that the mapping quality of the alignment is ≥40 and that the alignment includes only one CIGAR operation; S3. Based on the comparison results after quality control, the reads are classified into three types: Ref_read, Cross_reads, and GMO_reads; the three types of reads are counted, and the counting results are corrected by a logarithmic transformation with base 10. S4. Calculate the ratio between different types of reads, and optimize the range of ratio fluctuations through machine learning to determine the properties of the material; The specific types of reads, categorized into Ref_reads, Cross_reads, and GMO_reads, are as follows: Reads that are fully aligned to a reference genome portion; Reads that are aligned with a reference genome portion and with a foreign sequence portion, i.e., reads that cross the breakpoint; Reads that are partially and completely aligned to the exogenous sequence; Step S4 specifically includes the following sub-steps: S401. Based on the classification results of reads, calculate the number of reads in each category, obtain the proportion of reads in different categories, and use the proportion of reads in different categories as the feature input to the model; S402. Maximize classification accuracy as the optimization objective; S403. Define the hyperparameter space for Bayesian optimization; S404. Construct an objective function and adjust the classification accuracy by selecting different scale thresholds and fluctuation ranges; S405. Use the Gaussian process as a surrogate model to predict the behavior of the objective function and guide the next evaluation point based on the current evaluation results; use the desired improvement as the acquisition function to select an optimal combination of hyperparameters that can improve the model performance; The surrogate model is updated iteratively, and the Bayesian optimization is guided by the acquisition function to find the optimal hyperparameters; after each iteration, the objective function returns the evaluation result. S406. Obtain the optimal parameter combination through Bayesian optimization, and use the optimized proportional relationship to determine the transgenic nature of the material.

2. The method of discerning unknown exogenous segment insertion into a transgenic event based on targeted sequencing of claim 1, wherein, In step S3, the specific classification process of categorizing reads into three types—Ref_read, Cross_reads, and GMO_reads—is as follows: When the CIGAR string contains only matching operations and is fully aligned to the reference genome, the read is a Ref_read; When a CIGAR string includes matching and insertion / deletion operations, and the two aligned regions are from the reference genome and the foreign sequence, respectively, then the read is a Cross_read; A read is a GMO_read when the CIGAR string contains only the matching operation and is fully aligned to the foreign sequence.

3. The method of discerning unknown exogenous segment insertion into a transgenic event based on targeted sequencing of claim 1, wherein, In step S402, the optimization objective is specifically optimized through an optimization function. The output of the optimization function is the accuracy of the classifier, and the optimized ratio is used to distinguish genetically modified events.

4. The method of discriminating unknown exogenous segment insertion into a transgenic event based on targeted sequencing of claim 3, wherein, The calculation process of the accuracy rate is calculated according to a confusion matrix composed of the number of true transgenic reads , the number of true non-transgenic reads , the number of non-transgenic reads incorrectly judged as transgenic , and the number of transgenic reads incorrectly judged as non-transgenic , that is: ； wherein the a calculation result indicating an accuracy rate, the a number of reads indicating a true transgenic, the a number of reads indicating a true non-transgenic, the a number of reads indicating a non-transgenic that is erroneously judged as a transgenic, the a number of reads indicating a transgenic that is erroneously judged as a non-transgenic.

5. The method of discerning unknown exogenous segment insertion into a transgenic event based on targeted sequencing of claim 1, wherein, In step S404, the objective function is constructed based on the accuracy, that is: ； wherein the represents an objective function with the hyperparameter configuration and represent the true number of transgenic reads, the true number of non-transgenic reads, the number of non-transgenic reads that are wrongly judged as transgenic, and the number of transgenic reads that are wrongly judged as non-transgenic, respectively, with the hyperparameter configuration represent the proportion threshold of reads that perfectly align to the reference genome, the proportion threshold of reads that cross the breakpoint, and the proportion threshold of reads that perfectly align to the exogenous sequence, respectively.​​ 6. The method of discerning unknown exogenous segment insertion into a transgenic event based on targeted sequencing of claim 5, wherein, The , , and computational flows are represented as: ； ； ； ； wherein the represents a proportion threshold of reads perfectly aligned to the exogenous sequence, for reflecting the insertion amount of the exogenous gene sequence in the transgenic event, the represents the number of reads related to the exogenous sequence that meet the conditions, i.e. the number of reads perfectly aligned to the exogenous sequence, in a specific interval, the represents the total number of all reads, the represents the proportion of reads perfectly aligned to the reference genome in the total reads, the represents the number of reads perfectly aligned to the reference genome sequence, i.e. reads perfectly matching the reference genome, the represents a proportion threshold of reads crossing the breakpoint, i.e. the proportion of reads crossing the breakpoint in the total reads, the represents the number of reads crossing the breakpoint, the represents a proportion threshold of reads crossing the breakpoint, i.e. the proportion of reads crossing the breakpoint in the total reads.

7. The method of discerning unknown exogenous segment insertion into a transgenic event based on targeted sequencing of claim 1, wherein, In step S405, the calculation process for selecting the optimal hyperparameter combination that can improve model performance, using the desired improvement as the acquisition function, is expressed as follows: ； wherein the represents the expected improvement of the hyperparameter configuration represents the current optimal proportion of reads in the Bayesian optimization process, the represents the model predicted classification accuracy of the hyperparameter configuration represents the cumulative distribution function of the standard normal distribution, used to calculate the probability that the objective function value is lower than the current optimal value, the represents the standard deviation of the prediction error of the hyperparameter configuration represents the probability density function of the standard normal distribution, used to calculate the probability density of the gap between the current optimal value and the predicted mean.​​​ 8. The method of discerning unknown exogenous segment insertion into a transgenic event based on targeted sequencing of claim 1, wherein, In step S405, the proxy model is specifically represented as follows: ； wherein the denotes a proxy model, the denotes a Gaussian process function, the denotes a hyperparameter configuration as the model predicted classification accuracy of reads, the denotes a kernel function.

Images