A method of predicting pathogenic germline mutations

Abstract

The application provides a prediction method of pathogenic embryonic mutation, comprising the following steps: obtaining a variation site set of an embryonic mutation to be predicted, and performing population frequency screening on the variation site; filtering variation sites which are not annotated or are only annotated as clinically uncertain by using a clinical variation database; calculating a pathogenic posterior probability score of a candidate pathogenic variation by using a variation Bayesian inference (VBI) model with multiple types of function annotations; and cross- verifying the posterior probability score, external pathogenic prediction tool scores and variation types to determine a final pathogenicity discrimination result. Compared with an existing method based on single site frequency or single function score, the present application provides a posterior probability evaluation model with more biological interpretation in the field of rare variation judgment, effectively improves the prediction accuracy in a complex scene, significantly improves the risk classification accuracy based on multi-level evidence fusion discrimination, and effectively reduces the misdiagnosis and missed diagnosis rate in clinical stratification.

Description

Technical Field

[0001] This invention relates to the field of biomedicine, specifically to a method for predicting pathogenic germline mutations. Background Technology

[0002] In precision medicine, accurate diagnosis of pathogenic variants is crucial for effective clinical decision-making. In recent years, with the increasing maturity and cost reduction of high-throughput sequencing technology, the number of genetic variants identified in the human genome has surged. However, the diagnostic rate of hereditary diseases remains below ideal. To improve the accuracy and consistency of variant interpretation, the American Academy of Medical Genetics and Genomics and the American Society for Molecular Pathology published guidelines in 2015 to aid in the interpretation of variant information. These guidelines categorize genetic variants into five main groups: pathogenic, possibly pathogenic, of unknown significance, possibly benign, and benign, to assess the pathogenicity and potential risk of variants. Although this system provides a unified interpretation framework for clinical testing institutions worldwide, many variants are still classified as of unknown significance, lacking the functional testing data required for reliable classification. Current research has developed various bioinformatics and machine learning-based methods to predict the potential impact of genetic variants on protein structure and function. For example, PolyPhen2 and SIFT are widely used to assess protein damage that may be caused by missense variants, while CADD scores all categories of variants by integrating multidimensional annotation information. These tools have played a vital role in scientific research and preclinical analysis, providing valuable references for interpreting variants in situations where experimental validation is lacking.

[0003] Current computational methods for predicting the pathogenicity of germline variants still have significant limitations in overall performance: First, results are inconsistent between different algorithms, and there is a lack of a unified probability quantification standard. Most current prediction tools are based on their own independent algorithmic logic and training datasets, leading to frequent contradictions in predictions for the same variant (e.g., one predicts it as harmful, while another predicts it as benign). Second, reliance on a single feature results in a high false positive rate, and there is a lack of effective integration of multidimensional evidence. Traditional prediction methods often over-rely on the single feature of sequence conservation. However, not all variants at conserved sites cause disease. This single-dimensional reliance leads to many benign variants being incorrectly predicted as pathogenic (high false positive rate), especially in complex fields such as pharmacogenomics. Therefore, there is an urgent need to construct an evaluation framework that can systematically integrate multi-source biological annotations and provide a unified posterior probability through statistical inference to address the problems of inconsistent results and high false positive rates in existing tools. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a method for predicting pathogenic germline mutations. The aim is to construct an evaluation framework that can systematically integrate multi-source biological annotations and provide a unified posterior probability through statistical inference, thereby resolving the issues of inconsistent results and high false positive rates in existing tools. The technical solution provided by this invention is as follows:

[0005] A method for predicting pathogenic germline mutations, comprising the following steps:

[0006] Step 1: Obtain the set of variant sites for the germline mutation to be predicted, and perform population frequency screening for each variant site in the set.

[0007] Step 2: For the selected set of variant sites, query the existing pathogenicity annotation information in the public clinical variant database. When a variant site is not annotated in the clinical variant database or is only annotated as having unclear clinical significance, input the variant site into Step 3 for further predictive analysis.

[0008] Step 3: Obtain the functional annotation values ​​of each dimension of the variant site and perform standardized preprocessing. Calculate the ascending ranking of the functional annotation value of variant site j under the k-th annotation type among all candidate variants. The prior pathogenicity probability was calculated. Where M is the total number of variants involved in the calculation; the prior pathogenicity probability is embedded into a variational Bayesian inference framework with a peak-slab prior distribution, and the variational EM algorithm is used to construct a variational distribution to approximate the true posterior distribution. The distribution parameters are iteratively optimized by maximizing the lower bound of evidence, and finally the posterior probability score reflecting that variant site j belongs to the pathogenic variant is calculated. ;

[0009] Step 4: Cross-validate the posterior probability score with the external pathogenicity prediction tool score and variant type to determine the final pathogenicity discrimination result.

[0010] Preferably, step 1 specifically includes: obtaining the minimum allele frequency of each variant site from a public population frequency database; retaining the variant site when the minimum allele frequency of the variant site is less than a preset threshold; otherwise, removing the variant site.

[0011] Preferably, the peak-flat prior distribution is defined as:

[0012]

[0013] in, Let j be the effect coefficient of the j-th mutation site. The variance of the prior distribution plate portion. It is the Dirac function concentrated at zero.

[0014] Preferably, the distribution parameters are iteratively optimized by maximizing the lower bound of evidence. and The expression for the lower bound of evidence is:

[0015]

[0016] Where q is the variational distribution and p is the true posterior distribution. Let q be the expected value under the variational distribution q, and y be the phenotypic observation.

[0017] Preferably, during the iteration process, the variational distribution of the variant site j is modeled as follows using the factorization assumption of the variational distribution:

[0018]

[0019] in, and Let represent the posterior mean and variance when the variant site is a pathogenic variant, respectively. This is an indicator function.

[0020] Preferably, in step 4, a variant site is classified as a pathogenic variant when it simultaneously meets the following criteria:

[0021] Condition 1: The posterior probability score of the variant site is greater than the preset judgment threshold of 0.7;

[0022] Condition 2: The variant site is simultaneously identified as potentially pathogenic by at least two external pathogenicity prediction tools;

[0023] Condition 3: The molecular effect type of this mutation site belongs to missense mutation, nonsense mutation, splicing site variation, or frameshift mutation that affects protein function.

[0024] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0025] 1. It systematically integrates various types of functional annotation information, such as epigenetic modifications, sequence conservation, and protein function, changing the previous screening mode that relied solely on allele frequencies and providing rich biological prior support for sparse and rare variant signals.

[0026] 2. In pathogenicity prediction, variational Bayesian inference (VBI) is used instead of the traditional Markov chain Monte Carlo (MCMC) method. By maximizing the lower bound of evidence (ELBO), a fast and approximate inference of sparse signals in complex genetic backgrounds is achieved.

[0027] 3. By integrating multidimensional priors through a variational Bayesian framework, the sensitivity and specificity of pathogenic mutation identification are significantly improved.

[0028] 4. A prediction system was constructed that integrates clinical database filtering, VBI posterior probability calculation, and multi-tool consensus judgment, overcoming the technical bottleneck of high false positive rate of single prediction tools. Attached Figure Description

[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0030] Figure 1 This is a flowchart of the main process of the method of the present invention;

[0031] Figure 2 This is a comparison chart of the prediction performance of the present invention and the traditional method;

[0032] Figure 3 This is a comparison chart of the cumulative hits of individuals with extreme phenotypes using the present invention and traditional methods. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] To make the above-mentioned objectives, features and effects of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0035] Example 1: A method for predicting pathogenic germline mutations, comprising the following steps:

[0036] Step 1: Obtain the set of variant sites for the germline mutation to be predicted. Perform population frequency screening on each variant site in the variant site set. Specifically, this includes: obtaining the minimum allele frequency (MAF) of each variant site from a public population frequency database. If the MAF of a variant site is less than a preset threshold of 0.001, the variant site is retained; otherwise, the variant site is removed.

[0037] Step 2: For the set of variant sites selected in Step 1, query the existing pathogenicity annotation information in the ClinVar database, and process it as follows based on the annotation results: When a variant site is marked as pathogenic or potentially pathogenic in ClinVar, directly determine the variant site as a pathogenic variant and terminate the subsequent prediction process for that variant site; when a variant site is marked as benign or potentially benign in ClinVar, directly determine the variant site as a non-pathogenic variant and terminate the subsequent prediction process for that variant site; when a variant site is not marked or is only marked as clinically ambiguous in the clinical variant database, input the variant site into Step 3 for further prediction analysis.

[0038] Step 3: Using a variational Bayesian inference (VBI) model incorporating multi-type functional annotations, calculate the posterior probability score of the candidate pathogenic variants. .

[0039] FAVORDB from the STARRpipeline was selected as the basis for annotation data construction based on prior distributions, from which 13 functional annotations were obtained for each rare variant. These 13 functional annotations specifically include 10 annotation principal components (aPCs) and 3 comprehensive scores: epigenetic modification (aPC-Epigenetic), evolutionary conservation (aPC-Conservation), protein function prediction (aPC-Protein), local sequence diversity (aPC-LocalDiversity), distance to coding region (aPC-Dist2Coding), mutation density (aPC-MutationDensity), transcription factor binding site (aPC-TF), sequencing mappability (aPC-Mappability), distance to promoter and terminator (aPC-Dist2TSSTES), and microRNA target site (aPC-MicroRNA); comprehensive pathogenicity prediction score (CADD), non-coding region functional potential score (LINSIGHT), and cross-species functional impact prediction (FATHMM-XF).

[0040] When constructing the prior probability of a variant being a pathogenic variant using its functional annotation values, the original functional annotation values ​​of each dimension are first standardized to eliminate dimensional differences. Then, a non-parametric method is used to calculate the ascending ranking of the annotation value of variant site j under the k-th annotation class among all candidate variants. The prior pathogenicity probability was calculated. , where M is the total number of variants involved in the calculation. This probability is embedded within a variational Bayesian inference framework with a peak-and-slab prior distribution, defined as:

[0041]

[0042] in, Let j be the effect coefficient of the j-th mutation site. The variance of the prior distribution plate portion. It is the Dirac function concentrated at zero.

[0043] Construction of variational distribution using variational EM algorithm To approximate the true posterior distribution The distribution parameters are iteratively optimized by maximizing the lower bound of evidence, ELBO. and At this point, the basic expression for ELBO is:

[0044]

[0045] in, Let be the expected value under the variational distribution q, and y be the phenotypic observation. During the iteration process, using the factorization assumption of the variational distribution, the variational distribution of the variant site j is modeled as:

[0046]

[0047] in, and Let represent the posterior mean and variance when the variant site is a pathogenic variant, respectively. The indicator function is used. By maximizing the lower bound of evidence ELBO, the variational parameters are continuously updated until the model converges. Finally, the posterior probability score reflecting that the variant site belongs to the pathogenic variant is calculated. (Value range is 0~1).

[0048] Step 4: Cross-validate the posterior probability score with the score from the external pathogenicity prediction tool and the variant type to determine the final pathogenicity determination. A variant is classified as pathogenic when it simultaneously meets the following criteria:

[0049] Condition 1: The posterior probability score is greater than the preset judgment threshold of 0.7;

[0050] Condition 2: The site must be identified as potentially pathogenic by at least two of the external prediction tools consisting of CADD, SIFT, PolyPhen-2, and REVEL scores. The specific criteria for each tool to determine potential pathogenicity are: CADD > 20, SIFT < 0.05, PolyPhen-2 ≥ 0.85, and REVEL ≥ 0.05.

[0051] Condition 3: The molecular effect type of this mutation site belongs to missense mutation, nonsense mutation, splicing site variation, or frameshift mutation that affects protein function.

[0052] For variant sites that do not fully meet the above conditions, they are classified as variants of undetermined significance. The final output is a list of all variants classified as pathogenic and their associated biological characteristics.

[0053] In a computer-simulated trial involving 50,000 samples (based on real human whole-exome WES data), the predictive performance of this method was comprehensively evaluated by setting different heritability, causal variation ratios, and functional annotation influence strengths. The results are as follows: Figure 2 As shown.

[0054] The proposed method in this invention provides phenotypic prediction gain (incremental R) in the test set. 2 This invention demonstrates advantages over traditional C+T and Burden methods: The method achieves the highest predictive gain in most simulation scenarios, and its performance systematically improves with increasing heritability, causal ratio of rare variants, and annotation influence, exhibiting good stability and scalability. Especially when annotation information has a moderate to strong influence, the predictive performance of this method demonstrates its ability to integrate functional information and assist in screening rare variants with predictive value.

[0055] Identifying high-risk individuals is crucial for clinical intervention. By analyzing the predicted number of true extreme individuals among the top 500 individuals in the differential ranking, experimental data validated the technical advantages of this invention. The results are as follows: Figure 3 As shown, in a simulated scenario with high heritability (0.5) and strong influence from annotation information, the method of this invention successfully identified 262 individuals with extremely high risk among the top 500 individuals with predicted differences, demonstrating a significantly higher hit rate than the Burden and C+T methods. The cumulative identification curve further validates the reliability of the scheme: the curve of the method of this invention exhibits a steeper upward trend in the front-end ranking, indicating that the method can effectively enrich pathogenic signals.

[0056] For a binary phenotype simulation scenario with a 5% prevalence, the accuracy of classification prediction was evaluated by calculating the Net Reclassification Index (NRI) and the Integrated Discriminant Improvement Index (IDI). The results show that the method of this invention exhibits a significant gain in risk class adjustment capability, with the median NRI reaching 17.06% (P < 0.001) in a strong genetic background. This is not only superior to the Burden method's 7.04%, but also significantly higher than the traditional C+T method's 1.20% (see Table 1 for details).

[0057] Table 1. Median NRI results for each method when the prevalence of the disease in the population is 5%.

[0058]

[0059] Example 2: The computer-readable storage medium of this example stores a computer program that, when executed by a processor, implements the steps in the method for predicting pathogenic germline mutations in Example 1.

[0060] The computer-readable storage medium in this embodiment can be an internal storage unit of the terminal, such as the terminal's hard disk or memory; the computer-readable storage medium in this embodiment can also be an external storage device of the terminal, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc. equipped on the terminal; furthermore, the computer-readable storage medium can include both the terminal's internal storage unit and external storage devices.

[0061] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0062] Example 3: The computer device of this example includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for predicting pathogenic germline mutations in Example 1.

[0063] In this embodiment, the processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The memory can include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0064] Those skilled in the art will clearly understand that each implementation can be achieved using software plus the necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting pathogenic germline mutations, characterized in that, Includes the following steps: Step 1: Obtain the set of variant sites for the germline mutation to be predicted, and perform population frequency screening for each variant site in the set. Step 2: For the selected set of variant sites, query the existing pathogenicity annotation information in the public clinical variant database. When a variant site is not annotated in the clinical variant database or is only annotated as having unclear clinical significance, input the variant site into Step 3 for further predictive analysis. Step 3: Obtain the functional annotation values ​​of each dimension of the variant site and perform standardized preprocessing. Calculate the ascending ranking of the functional annotation value of variant site j under the k-th annotation type among all candidate variants. The prior pathogenicity probability was calculated. Where M is the total number of variants involved in the calculation; the prior pathogenicity probability is embedded into a variational Bayesian inference framework with a peak-slab prior distribution, and the variational EM algorithm is used to construct a variational distribution to approximate the true posterior distribution. The distribution parameters are iteratively optimized by maximizing the lower bound of evidence, and finally the posterior probability score reflecting that variant site j belongs to the pathogenic variant is calculated. ; Step 4: Cross-validate the posterior probability score with the external pathogenicity prediction tool score and variant type to determine the final pathogenicity discrimination result.

2. The method for predicting pathogenic germline mutations according to claim 1, characterized in that, Step 1 specifically includes: obtaining the minimum allele frequency of each variant site from the public population frequency database; retaining the variant site when the minimum allele frequency of the variant site is less than a preset threshold; otherwise, removing the variant site.

3. The method for predicting pathogenic germline mutations according to claim 1, characterized in that, The peak-flat prior distribution is defined as follows: ； in, Let j be the effect coefficient of the j-th mutation site. The variance of the prior distribution plate portion. It is the Dirac function concentrated at zero.

4. The method for predicting pathogenic germline mutations according to claim 3, characterized in that, Iterative optimization of distribution parameters is performed by maximizing the lower bound of evidence. and The expression for the lower bound of evidence is: ； Where q is the variational distribution and p is the true posterior distribution. Let q be the expected value under the variational distribution q, and y be the phenotypic observation.

5. The method for predicting pathogenic germline mutations according to claim 4, characterized in that, During the iteration process, the variational distribution of the variant site j is modeled using the factorization assumption of the variational distribution as follows: ； in, and Let represent the posterior mean and variance when the variant site is a pathogenic variant, respectively. This is an indicator function.

6. The method for predicting pathogenic germline mutations according to claim 1, characterized in that, In step 4, a variant site is classified as a pathogenic variant if it simultaneously meets the following criteria: Condition 1: The posterior probability score of the variant site is greater than the preset judgment threshold of 0.7; Condition 2: The variant site is simultaneously identified as potentially pathogenic by at least two external pathogenicity prediction tools; Condition 3: The molecular effect type of this mutation site belongs to missense mutation, nonsense mutation, splicing site variation, or frameshift mutation that affects protein function.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the method for predicting pathogenic germline mutations as described in any one of claims 1-6.

8. A computer device comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method for predicting pathogenic germline mutations as described in any one of claims 1-6.

Images