Construction method and application of false positive judgment model of mutation site

Abstract

The invention provides a construction method, system and product of a false positive judgment model of a mutation site obtained by sequencing after enzyme digestion fragmentation and related application. Specifically, the false positive judgment model is constructed on the basis of a plurality of characteristic parameters associated with mutation sites obtained by enzyme digestion fragmentation sequencing; the characteristic parameters comprise scores obtained by performing reverse complementary sequence alignment on the front and back extension 30bp of the mutation site, the sequence proportion of 15 basic groups at the tail end of the mutation site, the Bayesian conditional probability of the extracted mutation site, and the KS test statistic of the extracted mutation site. Furthermore, the invention also provides a method for constructing the model by utilizing the crowd detection rate and the corresponding false positive. According to the method, the data characteristics of the false positive mutation sites are introduced by utilizing digestion fragmentation, so that the reliability of the digestion false positive mutation sites is objectively evaluated under the condition that high depth and large data volume are not needed, and the detection accuracy of the mutation sites can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of gene sequencing, specifically to a method for constructing a false positive judgment model for mutation sites obtained by sequencing after enzyme digestion and fragmentation, a judgment system, and related applications. Background Technology

[0002] DNA library preparation is a crucial first step in generating high-quality sequencing data. Errors and inefficiencies in the library preparation process can significantly impact downstream analysis and lead to detrimental effects, including an increased incidence of false positives or false negatives. DNA fragmentation is a fundamental step in the preparation of short-read sequencing libraries based on hybridization capture. Fragmentation allows the DNA sample length distribution to better match the read lengths of the sequencing platform. Currently, there are two main methods for DNA fragmentation: physical methods (mechanical fragmentation, such as sonication and nebulization) and enzymatic digestion methods (i.e., non-specific endonuclease digestion). Sonication is efficient and consistent, but can be both expensive and time-consuming. Sonication equipment can cost tens of thousands of dollars, and unless high-throughput plate-based instruments are used, one to eight samples must be sonicated at a time, each sample taking several minutes. This time requirement can make large sequencing projects impractical for small laboratories and even very labor-intensive for large operations. Therefore, many research groups are exploring alternative methods for DNA fragmentation as part of the library preparation workflow, including enzymatic fragmentation and transposase-mediated fragmentation and labeling (“labeling”).

[0003] Enzyme fragmentation uses a mixture of enzymes to create breaks or nicks in the input genetic material. Enzyme fragmentation can be easily applied to multiple samples at once in 96-well plates and does not require specialized equipment. While the fragment distribution is not as concentrated as sonication, enzyme fragmentation results in lower sample loss during the process. Enzyme fragmentation is increasingly popular in high-throughput sequencing operations due to its ease of use, scalability, and low barrier to entry. However, because enzymes can act non-randomly, enzyme fragmentation can introduce significant bias or sequencing errors if the enzyme is partially blocked by DNA or selectively cleaves certain parts of the DNA.

[0004] To address the issue of false-positive mutations during enzyme digestion and fragmentation, one common method is to use adapters containing UMI tags during adapter ligation. UMIs (Unique Molecular Identifiers) are typically designed as completely random nucleotide chains (e.g., NNNNNN), partially degenerate nucleotide chains (e.g., NNNRNYN), or fixed nucleotide chains (when the template molecule is limited). Figure 1As can be seen, the same DNA molecule after fragmentation and end repair generates multiple reads after sequencing, with some bases showing mutations (highlighted in red). These are false-positive mutations generated during library amplification and excessively long sequencing runs. Introducing UMI tags to eliminate false-positive mutations generated during amplification and sequencing can improve the detection rate of low-frequency mutations. However, using UMI-tagged adapter data to reduce the impact of enzyme digestion on false positives requires high-depth data support, thus requiring a large amount of sequencing data and significantly increasing costs.

[0005] Although many reagent kit manufacturers have developed kits to address the false positive mutation problem caused by enzyme digestion library construction, striving to reduce false positive mutations in fragmented enzyme library construction from the source, the results are not ideal. Summary of the Invention

[0006] To address the shortcomings of using UMI tag adapters which require high-depth support, this invention starts with the data characteristics of enzyme digestion fragmentation and develops a method, judgment system, product, and related applications for constructing a false positive judgment model for mutation sites obtained by sequencing after enzyme digestion fragmentation.

[0007] In a first aspect, the present invention provides a method for constructing a false positive judgment model for mutation sites obtained by sequencing after enzyme digestion and fragmentation. The method includes constructing the false positive judgment model based on several characteristic parameters associated with the mutation sites obtained by enzyme digestion and fragmentation sequencing. The aforementioned characteristic parameters include: the score obtained by reverse complementary sequence alignment with 30 bp extensions before and after the mutation site, the sequence proportion of the mutation site in the last 15 bases, the Bayesian conditional probability of the extracted mutation site, and the KS test statistic of the extracted mutation site.

[0008] Specifically, the above construction methods include:

[0009] The characteristic parameters associated with mutation sites obtained from enzyme digestion and fragmentation sequencing are used as relevant modeling factors. The modeling factors of the mutation sites and the validation results of the mutation sites are used to construct a training sample set.

[0010] Based on the training sample set constructed above, several types of models are trained, and the trained models are evaluated. The best judgment model is determined according to the model evaluation index.

[0011] In one specific implementation, a false positive judgment model is obtained by using several feature parameters associated with mutation sites obtained from enzyme digestion and fragmentation sequencing as inputs and whether the mutation site is a false positive as the output (i.e., the verification result of the mutation site) to construct a model through machine learning methods.

[0012] Secondly, this invention also provides a method for constructing a false positive judgment model for mutation sites obtained by sequencing after enzyme digestion and fragmentation, comprising:

[0013] Obtain a reference database of mutation sites obtained from mechanical fragmentation sequencing. The database includes the population detection rate of the mutation site and the false positives of the mutation site set based on a predetermined population detection rate threshold.

[0014] Feature parameters associated with mutation sites obtained from enzyme digestion and fragmentation sequencing were used as relevant modeling factors.

[0015] The modeling factors of the above mutation sites and their corresponding false positives in the above reference database are used to construct a training sample set. The feature parameters include: the score obtained by reverse complementary sequence alignment with 30bp extensions before and after the mutation site, the sequence proportion of the mutation site in the last 15 bases, the Bayesian conditional probability of the extracted mutation site, and the KS test statistic of the extracted mutation site.

[0016] Several types of models are trained based on the training sample set, and the trained models are evaluated. The best judgment model is determined based on the model evaluation index.

[0017] Thirdly, based on the false positive judgment model constructed using the methods described in the first and second aspects above, this invention provides a false positive judgment system for mutation sites obtained through sequencing after enzyme digestion and fragmentation. This system includes:

[0018] A device for acquiring characteristic parameters associated with mutation sites in a test object based on enzyme digestion and fragmentation sequencing;

[0019] A device for processing the obtained feature parameter data using a false positive detection model;

[0020] A device for outputting the judgment result.

[0021] Fourthly, based on the false positive detection model constructed using the methods described in the first and second aspects above, this invention provides a false positive detection product for mutation sites obtained through sequencing after enzyme digestion and fragmentation. The system includes:

[0022] Memory, used to store programs;

[0023] A processor is configured to implement the following detection method by executing a program stored in the aforementioned memory;

[0024] The method includes:

[0025] Obtain data on characteristic parameters associated with mutation sites in the test object based on enzyme digestion fragmentation sequencing;

[0026] The obtained feature parameter data is input into the false positive judgment model for processing, and the false positive of the mutation site is output.

[0027] Fifthly, based on the false positive judgment model constructed by the construction methods of the first and second aspects described above, the present invention provides a computer-readable storage medium storing a program that can be executed by a processor to implement the following method:

[0028] Obtain data on characteristic parameters associated with mutation sites in the test object based on enzyme digestion fragmentation sequencing;

[0029] The obtained feature parameter data is input into the false positive judgment model for processing, and the false positive of the mutation site is output.

[0030] In a sixth aspect, the present invention provides a computer-readable storage medium storing a program on which a false positive judgment model constructed by the construction methods of the first and second aspects described above is stored.

[0031] In a seventh aspect, the present invention provides a method for constructing the false positive judgment model of the first and second aspects described above, as well as the application of related models, systems, products, and computer storage media in the analysis of mutation sites in tumor patients.

[0032] The beneficial effects of this invention are as follows: Given that current methods for reducing false positives in mutation sites obtained through enzyme digestion and fragmentation sequencing require high depth and large data output, or have a single threshold, this invention utilizes the data characteristics of false-positive mutation sites introduced by enzyme digestion and fragmentation. This allows for objective evaluation of the reliability of false-positive mutation sites without requiring high depth or large data volumes, significantly improving the accuracy of mutation site detection. Especially in mutation site screening for cancer patients, this invention can help more patients save costs, obtain reliable mutation site variation analysis results, and receive timely and appropriate medication guidance, achieving precision medicine through low-cost operation. Attached Figure Description

[0033] Figure 1 In the background technology, the use of UMI-tagged adapters resulted in multiple reads after sequencing, some of which contained mutations in individual bases.

[0034] Figure 2 Characteristic analysis of false-positive variations introduced by enzyme digestion fragmentation;

[0035] Figure 3 This is a characteristic distribution map of false positive sites in the enzyme digestion fragmentation sequencing results, as shown in one specific embodiment of the present invention.

[0036] Figure 4 This is a graph showing the population detection rate and false positives (true and false positives, false variants) in one specific embodiment of the present invention.

[0037] Figure 5 This is a graph showing the population detection rate and false positive (false positive, true variant) analysis in one specific embodiment of the present invention. Detailed Implementation

[0038] In existing technologies, pairwise comparative analysis of somatic variations in the same tumor DNA samples prepared using ultrasonic fragmentation and enzyme digestion methods has revealed the following:

[0039] Compared to libraries processed by sonication, libraries processed by restriction enzyme fragmentation showed a significantly higher number of somatic point mutations and insertions / deletions due to recurring false positives. Furthermore, although derived from the same DNA sample, the number of mutations detected by enzyme digestion was 2.3–9.9 times that of those detected by sonication, and most of the additional mutations (mutation sites) detected by enzyme digestion appeared in multiple samples.

[0040] Further analysis of the mutation sites obtained by sequencing these fragmented enzyme digests revealed that these specific mutation signals are usually located 10 to 15 bases at the 5' or 3' edge of the sequence, and the reads specifically detected by enzyme digestion contain a higher proportion of soft shearing (average 50.8% vs 5.0%).

[0041] Meanwhile, during variation analysis of large cohorts, a large number of single nucleotide, insertion, and deletion variations were also found. Careful examination revealed that these false positive signals were byproducts of the enzymatic cleavage process. Other enzyme-based fragmentation kits were also evaluated and found to have similar issues. Furthermore, this study demonstrated good sequencing quality, indicating that the false positive variations resulting from enzyme fragmentation are not directly proportional to sequencing quality.

[0042] Through in-depth research and analysis of the data, the following characteristics were found in the variants of false positives generated by enzyme digestion and fragmentation:

[0043] 1) The detected false-positive variants have a relatively low allele frequency;

[0044] 2) Sequences that support false positive signals all have more soft splicing, and these soft spliced ​​sequences are highly conserved, but they are not consistent with the adapter primer sequences; if reverse complementation is performed, they can usually be found in the nearby reference sequence (there is a palindrome structure nearby).

[0045] 3) These have high base quality values, indicating that they are genuine molecular derivatives, rather than products of sequencing errors.

[0046] 4) On these reads that support false positive signals, the variant signal is usually located 10 to 15 bases away from the 5' or 3' edge of the fragment.

[0047] This suggests that the possible reason for the introduction of false-positive variants by enzyme fragmentation is that some inverted repeat sequences formed neck loops after enzyme digestion. These neck loops can generate duplicate false-positive sequences (such as...) during downstream end repair and PCR amplification. Figure 2 (As shown).

[0048] This invention utilizes the data characteristics of the fragmented sequences after enzyme digestion (i.e., several characteristic parameters associated with mutation sites obtained from enzyme digestion and fragmented sequencing), such as the length of the inverse complementary sequence and the proportion of mutations in the terminal 15 bases, to construct a model to solve the false positive problem of enzyme digestion-introduced mutation sites, or at least to identify false positive variants introduced by enzyme digestion from a large number of variants, thereby distinguishing mutation sites with a high probability of false positives in sequencing results based on enzyme digestion and fragmentation. In other words, this invention aims to filter out some false positive signals introduced by enzyme digestion and fragmentation at the bioinformatics level, in order to improve the accuracy of the final detected variants.

[0049] Specifically, the characteristic parameters associated with the mutation sites obtained by the above-mentioned enzyme digestion and fragmentation sequencing include, but are not limited to: the scores obtained by reverse complementary sequence alignment with 30 bp extensions before and after the mutation site, the sequence proportion of the mutation site in the last 15 bases, the Bayesian conditional probability of the extracted mutation site, and the KS test statistic of the extracted mutation site.

[0050] The sequence proportion of mutations in the last 15 bases refers to the ratio of the number of reads whose mutated bases are located within 15 bp of the end of the reads to the total number of reads supporting that mutated site.

[0051] It should be understood that those skilled in the art can clearly obtain the above-mentioned characteristic parameters from the relevant sequencing data results through software or other means based on the description of this invention, and the methods of data acquisition will not be elaborated in this invention.

[0052] In one specific embodiment of the present invention, preferably, several feature parameters associated with mutation sites obtained from enzyme digestion and fragmentation sequencing are used as input, and whether the mutation site is a false positive is used as output. A model is constructed using machine learning methods to obtain a false positive judgment model. The machine learning methods include, but are not limited to: random forest, Naive Bayes, logistic regression, support vector machine, AdaBoost, K-nearest neighbor algorithm, neural network, passive attack algorithm, stochastic gradient descent, or XGBoost, etc. In this specific embodiment of the present invention, a random forest model is used to construct the false positive judgment model.

[0053] It should be understood that determining whether the above-mentioned mutation sites are false positives and using this result as output to construct the model requires validation for each mutation site. Due to the cost of validation, this invention further provides an alternative method: introducing mechanically fragmented sequencing results to determine false positives for mutation sites.

[0054] Specifically, this invention provides another method for constructing a false positive judgment model for mutation sites obtained by sequencing after enzyme digestion fragmentation. This method adds a step of obtaining a reference database of mutation sites based on mechanically fragmented sequencing. This database includes the population detection rate of mutation sites and false positives of mutation sites set based on a predetermined population detection rate threshold. Therefore, unlike the above-mentioned scheme that verifies whether a mutation site is a false positive, this construction method uses the obtained mechanically fragmented mutation site database and the population detection rate as a label to distinguish between true and false positive variants in the model construction.

[0055] It is important to note that by statistically analyzing mutation sites in a large population through mechanical fragmentation sequencing, we can obtain the population detection rate covering known mutation sites, i.e., the detection rate of a particular mutation site in different samples. A threshold is then set for this population detection rate to determine false positives for that mutation site. For example:

[0056] 1) When the population detection rate obtained by mechanical fragmentation sequencing of a certain mutation site is greater than the highest threshold, the mutation site is determined to be a false positive.

[0057] 2) When the population detection rate obtained by mechanical fragmentation sequencing of a certain mutation site is less than the minimum threshold, the mutation site is determined to be a true positive.

[0058] 3) When the population detection rate obtained by mechanical fragmentation sequencing of a certain mutation site is between the highest and lowest thresholds, the mutation site is determined to be located in the gray area of ​​the data and should be discarded in the subsequent modeling process.

[0059] The aforementioned maximum and minimum thresholds can be adjusted by those skilled in the art during the modeling process based on the actual situation of the analyzed mutation sites. For example, they can be set based on large datasets of mechanical or enzyme-mediated mutations. In one specific embodiment of the present invention, in the analysis of several mutation sites based on tumor patient samples, the aforementioned population detection rate thresholds are as follows: mutations with a population frequency greater than 50% are considered false positives, and those less than 5% are considered true positives. Mutation sites falling between these two thresholds are discarded in subsequent modeling.

[0060] Apart from the aforementioned mutation site database based on mechanical fragmentation and the output of the corresponding mutation site in the model construction being the false positive judgment result of the mutation site in the database, the specific methods of other model construction, such as the machine learning methods used, are completely consistent with the construction method mentioned in the first aspect of this invention.

[0061] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0062] Example 1: Construction of a False Positive Detection Model

[0063] 1. The population detection rate of different mutation sites in 1108 tumor patient samples was statistically analyzed. Specifically, the DNA of these 1108 tumor patient samples was fragmented by ultrasound and sequenced. The raw data were preprocessed and subjected to basic bioinformatics analysis: alignment to a reference genome, removal of duplicate reads and local re-alignment of indels to obtain the BAM (Bioassay Anthropometric Model), mutation detection of the BAM, and generation of VCF (Virtual Data File) files. Based on the analysis, the population detection rate of each mutation site in the above 1108 samples was calculated.

[0064] 2. Ultrasonic physical fragmentation and enzyme digestion fragmentation experiments were performed on 138 tumor patient samples (the 138 samples must cover the mutation sites to be detected in the above 1108 samples). Raw sequencing data of tumor tissue and control tissue from the two groups of 138 samples were obtained by high-throughput sequencing (NGS).

[0065] 3. Preprocessing and basic bioinformatics analysis were performed on the raw sequencing data of 276 cases (two sets of experimental data from 138 samples): alignment to the reference genome, removal of duplicate reads and local re-alignment of Indels to obtain BAM, variant detection of BAM, and generation of vcf files.

[0066] 4. Compare the VCF files obtained from the two sets of experimental data. Define the variants detected only in the ultrasonic physical disruption experimental data as "mechanical only" and the variants detected only in the enzyme digestion fragmentation experimental data as "enzyme digestion only".

[0067] Characteristic analysis of false positive sites in the "enzyme digestion only" data revealed, for example Figure 3 As shown, Figure 3The horizontal axis represents the read length, and the vertical axis represents the distribution of mutation positions on the reads. The red portion represents the distribution of false positive mutations introduced by enzyme digestion, and the blue portion represents the distribution of normal refs. Clearly, mutations are concentrated at the ends of the reads. Further clarification of several characteristic parameters of the mutation sites introduced by enzyme digestion fragmentation includes: the scores obtained from reverse complementary sequence alignment after extending 30 bp before and after the mutation site; the proportion of the mutation site in the last 15 bases; the Bayesian conditional probability of the extracted mutation site; and the KS test statistic of the extracted mutation site.

[0068] 5. We tentatively decided to use the reverse complementarity sequence scoring to distinguish between true positive and false positive sites in the two experimental data sets. Combining this result with an examination of the population detection rate and the discriminative power of each mutation site in the 1108 samples, we found that using the reverse complementarity sequence scoring to distinguish between true positive and false positive sites in the data was very effective. Specifically, as follows... Figure 4 and Figure 5 As shown, Figure 4 and Figure 5 The horizontal axis represents the population detection rate range, and the vertical axis represents the number of variants (showing the number of variants within each population frequency range).

[0069] Depend on Figure 4 and Figure 5 It can be seen that variants with a population detection rate greater than 50% are considered false positives, those less than 5% are considered true variants, and variants with a detection rate greater than 5% but less than 50% are classified as gray area data points and discarded in subsequent modeling. Subsequently, the population detection rate and the aforementioned threshold range of the population detection rate are chosen as labels to distinguish between true and false positive variants.

[0070] 6. The 138 processed samples were randomly divided into a training set 1 (91 cases) and a test set (47 cases) for model construction, and modeling was performed.

[0071] In the modeling phase, the false positive results corresponding to the population detection rate in the database constructed from the "enzyme digestion_Only" variant data and the corresponding 1108 samples were selected for modeling. Training set 1 was divided into training set 2 and validation set in a 7:3 ratio. Based on the analyzed mutation information file, a random forest model was constructed using R language and the random forest package.

[0072] 7. After tuning the model parameters and verifying its evaluation performance, the accuracy, precision, and recall of the model on the training set 2 and the validation set were all 1. This means that all the data in the training set and the validation set were correctly classified into the two categories, namely false positives and non-false positives, after being validated by the random forest model.

[0073] 8. When the model was applied to an independent test set of 47 samples, the accuracy, precision and recall were 0.994, 0.996 and 0.995, respectively, indicating a high accuracy.

[0074] Example 2: False Positive Test

[0075] 1. After ultrasonic mechanical fragmentation and enzyme digestion fragmentation experiments were performed on 6 samples and their controls, high-throughput sequencing was performed to obtain raw data.

[0076] 2. Align the raw data to the reference genome, remove duplicate reads and local realignment of Indels, and obtain the BAM files and variant detection result files for two sets of experiments, including 6 samples and control samples.

[0077] 3. Input the mutation results of the enzyme-digested fragmented experimental samples into the random forest classification model obtained in Example 1 to determine the true positive and false positive status.

[0078] The mutation results of the enzyme digestion fragmentation experiment samples after being judged by the random forest classification model were compared with the mutation results of the ultrasonic mechanical fragmentation experiment to evaluate the model's filtering effect. The final results are shown in Table 1 below:

[0079] Table 1 compares the variation results of the enzyme digestion fragmentation experiment samples after random forest classification with the variation results of the ultrasonic mechanical fragmentation experiment.

[0080]

[0081] As shown in Table 1 above, mutation sites that only appear after enzyme digestion and fragmentation sequencing exist in different samples. Among them, sample number T11024697 has 14 such mutation sites that only exist after enzyme digestion. After filtering by the false positive judgment model in Example 1, these 14 sites were all judged as false positives and filtered out. Therefore, after filtering by the false positive judgment model, the mutation sites in the sequencing results after enzyme digestion and fragmentation are completely consistent with the results after mechanical fragmentation, with a filtering efficiency of 100%.

[0082] This invention, based on high-throughput sequencing data, utilizes the data characteristics of enzyme digestion and fragmentation experiments to determine the true positive and false positive status of sample test results in a low-cost manner. This method has no manual interpretation component, is objective and highly sensitive, and has a significant false positive filtering effect.

[0083] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.

Claims

1. A method for constructing a false positive judgment model for mutation sites obtained by sequencing after enzyme digestion and fragmentation, characterized in that, The false positive judgment model is constructed based on several characteristic parameters associated with mutation sites obtained by enzyme digestion and fragmentation sequencing. The characteristic parameters include: the score obtained by reverse complementary sequence alignment with 30 bp extensions before and after the mutation site, the sequence proportion of the mutation site in the last 15 bases, the Bayesian conditional probability of the extracted mutation site, and the KS test statistic of the extracted mutation site.

2. The construction method as described in claim 1, characterized in that, The construction method includes: The characteristic parameters associated with mutation sites obtained from enzyme digestion and fragmentation sequencing are used as relevant modeling factors. The modeling factors of the mutation sites and the validation results of the mutation sites are used to construct a training sample set. Several types of models are trained based on the training sample set, and the trained models are evaluated. The best judgment model is determined based on the model evaluation index.

3. The construction method as described in claim 1, characterized in that, The construction method includes: taking several feature parameters associated with mutation sites obtained by enzyme digestion and fragmentation sequencing as input, and taking whether the mutation site is a false positive as output, constructing a model through machine learning methods to obtain the false positive judgment model.

4. The construction method as described in claim 3, characterized in that, The machine learning methods mentioned are random forest, Naive Bayes, logistic regression, support vector machine, AdaBoost, K-nearest neighbor algorithm, neural network, passive attack algorithm, stochastic gradient descent or XGBoost; Preferably, the machine learning method is random forest.

5. A method for constructing a false positive judgment model for mutation sites obtained by sequencing after enzyme digestion and fragmentation, characterized in that, The construction method includes: A reference database of mutation sites obtained based on mechanical fragmentation sequencing is obtained. The database includes the population detection rate of the mutation sites and the false positives of the mutation sites set based on a predetermined population detection rate threshold. Feature parameters associated with mutation sites obtained from enzyme digestion and fragmentation sequencing were used as relevant modeling factors. The modeling factor of the mutation site and its corresponding false positive in the reference database are used to construct a training sample set; the feature parameters include: the score obtained by reverse complementary sequence alignment with 30bp extensions before and after the mutation site, the sequence proportion of the mutation site in the last 15 bases, the Bayesian conditional probability of the extracted mutation site, and the KS test statistic of the extracted mutation site. Several types of models are trained based on the training sample set, and the trained models are evaluated. The best judgment model is determined based on the model evaluation index.

6. The construction method as described in claim 5, characterized in that, The false positive of the mutation site is set based on a predetermined population detection rate threshold as follows: a mutation site with a population frequency greater than 50% is considered a false positive, and a mutation site with a population frequency less than 5% is considered a true positive.

7. The construction method as described in claim 5, characterized in that, Using several feature parameters associated with mutation sites obtained by enzyme digestion and fragmentation sequencing as input, and the false positives corresponding to the mutation sites in the reference database as output, a model is constructed using machine learning methods to obtain the false positive judgment model. Preferably, the machine learning method is random forest, Naive Bayes, logistic regression, support vector machine, AdaBoost, K-nearest neighbor algorithm, neural network, passive attack algorithm, stochastic gradient descent or XGBoost; Preferably, the machine learning method is random forest.

8. A false positive determination system for mutation sites obtained by sequencing after enzyme digestion and fragmentation, characterized in that, include: A device for acquiring characteristic parameters associated with mutation sites in a test object based on enzyme digestion and fragmentation sequencing; A device for processing the obtained feature parameter data using a false positive detection model; A device for outputting the judgment result; The false positive judgment model is obtained by the construction method as described in any one of claims 1-7.

9. A product for identifying false positives at mutation sites obtained through enzyme digestion and sequencing, characterized in that, include: Memory, used to store programs; A processor is configured to implement the following detection method by executing a program stored in the memory; The method includes: Obtain data on characteristic parameters associated with mutation sites in the test object based on enzyme digestion fragmentation sequencing; The obtained feature parameter data is input into the false positive judgment model for processing, and the false positive of the mutation site is output. The false positive judgment model is obtained by the construction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The medium stores a program that can be executed by a processor to implement the following method: Obtain data on characteristic parameters associated with mutation sites in the test object based on enzyme digestion fragmentation sequencing; The obtained feature parameter data is input into the false positive judgment model for processing, and the false positive of the mutation site is output. The false positive judgment model is obtained by the construction method as described in any one of claims 1-7.

11. A computer-readable storage medium, characterized in that, The medium stores the judgment model obtained by the construction method as described in any one of claims 1-7.

12. The application of the construction method as described in any one of claims 1-7 in the analysis of mutation sites in tumor patients.

Images