CN111540407A - A method for integrating multiple neurodevelopmental diseases to screen candidate genes - Google Patents

Abstract

The invention is applicable to the technical field of neurological diseases, and provides a method for screening candidate genes by integrating different types of variation data of multiple neurodevelopmental disease genes, and a candidate for a certain disease or a certain type of disease evaluated based on the method of maximum likelihood estimation The number of genes, based on the Bayesian model method and combined with the expected gene mutation rate, case mutation load and other data to identify candidate genes for neurological diseases The problem of insufficient statistical effect of variation provides a new method for identifying new candidate genes and enhancing gene reliability. To verify the reliability of the method by using known genes, new candidate genes can be found according to the distribution of the variation spectrum of the disease cohort, and the method can be extended to other major genetic diseases.

Description

A method for integrating multiple neurodevelopmental diseases to screen candidate genes

technical field

The invention belongs to the field of neurological diseases, and in particular relates to a method for integrating multiple neurodevelopmental diseases to screen candidate genes.

Background technique

Due to its high morbidity and disability rate, neurodevelopmental diseases bring great pain and burden to individuals and society. There are many types of these diseases and high clinical heterogeneity, which brings great difficulties to the diagnosis and subsequent targeted treatment of the disease. The diagnosis of disease is currently based on the phenotype described by the patient or family member and the behavior observed by the physician. These clinical phenotype information is basically based on the results summarized by previous experts, and there is no unified biological index for judgment.

The high heritability of neurodevelopmental diseases provides an excellent opportunity to study the correlations between diseases and to understand the pathogenic mechanisms of diseases. Previous studies have found high genetic similarity in different neurodevelopmental diseases, and attempted to find possible candidate genes through genome-wide association methods. However, the number of candidate genes found by this method is very small, which can explain very few pathogenic causes. Due to the high probability of overlap in clinical phenotypes and genetic mechanisms between different types of neurological diseases, it is possible to integrate multiple neurological diseases to find possible disease candidate genes. Currently, multiple research institutions use whole exome or whole genome sequencing methods to detect de novo variants (ie, variants that are only present in the patient but not detected in the parental genome) and can find many candidate genes. This de novo variant is highly functionally disruptive and can significantly alter individual phenotypes.

Previous methods have been used to find candidate genes by integrating disparate disease variant data. However, these methods currently have many limitations, including the following defects:

(1) Only consider whether the variant site is on a known possible candidate gene. Although this method can carry out phenotypic diagnosis, it cannot discover new candidate genes.

(2) For a gene with multiple potentially harmful variants, it is not necessarily a candidate gene. Parameters such as the mutation rate of a single gene, the number of patients tested, and the number of expected candidate genes also need to be considered.

(3) Previous methods only considered the relationship between DNA-level variation and disease, and did not consider tissue-specific expression level data, protein levels, and other information. The credibility of the obtained proof requires further analysis.

(4) The integrated analysis is not really considered, and the previous method only analyzes the individual disease and compares multiple disease candidates to study the genetic correlation between different diseases.

SUMMARY OF THE INVENTION

The invention provides a method for integrating multiple neurodevelopmental diseases and screening candidate genes, aiming to integrate multiple new ideas of neurodevelopmental diseases to identify new candidate genes and enhance gene reliability.

The present invention is achieved in this way, a method for integrating multiple neurodevelopmental diseases and screening candidate genes, comprising the following steps:

S1. Evaluate candidate genes by maximum likelihood estimation method to obtain the number of possible candidate genes for each disease;

S2. At the same time, taking into account the parameters of disease mutation load, mutation rate, and the number of candidate genes for evaluation, according to the loss-of-function mutation and deleterious missense mutation carried on each gene and the number of patients detected, a single gene is obtained by comprehensive evaluation. The Bayes factor of the level and the probability value of the false positive of the candidate gene according to the Bayes factor;

S3. Further comprehensively evaluate the reliability of the candidate gene according to the data at the gene level.

The candidate genes are evaluated by the maximum likelihood estimation method to obtain the number of possible candidate genes for each disease, specifically:

Defining the number of all loss-of-function variants and deleterious missense variants in the disease (K), the number of genes for which two or more deletion variants and deleterious missense variants of the same gene were detected in different patients (R), and loss-of-function variants and deleterious missense variants The proportion of the sense variant contributing to the disease (E);

Do one million permutation tests, and randomly select a certain proportion of candidate genes (T) from all coding genes in each permutation test (value range is 1-2500) and select a certain proportion according to the binomial distribution (K, E). Variation in proportions (C);

Assign the number of loss-of-function variants and deleterious missense variants that contribute to disease (C) to candidate genes (T) and delete deleterious variants that do not contribute to disease (K-C) to other genes;

Each permutation test counts the number of genes extracted twice or more. If the number is the same as the actual observed number, the candidate gene (T) extracted this time is considered to be the number of possible causative genes for the disease;

After all permutation tests are completed, a prior probability distribution curve is obtained according to the number of all possible disease-causing genes, and the number of candidate genes corresponding to the point with the highest probability is the number of genes most likely to be related to the disease.

Preferably, the disease mutation load, mutation rate, and parameters of the number of predicted candidate genes are taken into account at the same time, and a comprehensive evaluation is carried out according to the loss-of-function mutation and deleterious missense mutation carried on each gene and the number of patients detected, etc. Obtain the Bayes factor at the single gene level and the probability value of the false positive of the candidate gene according to the Bayes factor, specifically:

Observing the number of variants that appear in N families and the expected number 2Nμ, μ represents the mutation rate of a specific gene, and the Bayesian method is used to compare the two models of H0 and H1, where H0 is a gene that is not a disease-causing gene, and H1 is a gene that is Pathogenic gene; when H0 is established, the expected number of mutations is 2Nμ; when H1 is established, the expected number of mutations is 2Nμγ, where γ represents the relative risk, and γ greater than 1 indicates that it may be a pathogenic gene;

The Bayes factor is defined as the probability of becoming H1 divided by the probability of H0; the probability of H1 being positive is determined according to Bayes' theorem, a Bayes factor greater than 1 indicates that H1 is likely to be positive, and a Bayes factor greater than 100 indicates a strong The proof H1 is positive;

Multiply the Bayes factors obtained by different mutation types, each mutation type will get a separate Bayes factor, and multiply all the Bayes factors to get the gene-level Bayes factor.

Using the Bayesian FDR method to control the false positive rate, each gene will get an FDR value, that is, q-value;

It is defined that the q-value is less than a preset value A, and 0≦A≦1, indicating that the gene with probability A may be wrong.

Preferably, it also includes: before performing step S1, performing genome function annotation on the detected variants to classify the variants into loss-of-function variants, deleterious missense variants, tolerable missense variants, synonymous variants and non-function variants. coding region variation.

Preferably, the credibility of the candidate gene is specifically: the level of specific expression of the gene in disease-related tissues, the degree of co-expression and protein interaction with known disease-related genes, and the significant enrichment in known disease-related genes. pathways associated with neurodevelopmental diseases.

Compared with the prior art, the beneficial effects of the present invention are:

1. The reliability of the method is verified by existing genes, and new candidate genes can be researched according to the harmful variation of unknown genes.

2. Combined with parameters such as mutation rate, number of detection samples, number of candidate genes, etc., the reliability of candidate genes is evaluated statistically. There is a more reliable basis for the discovered disease-causing genes.

3. In order to make up for the shortcoming of the insufficient sample size of a single neurodevelopmental disease, more candidate genes were identified at the same time. Based on the genetic similarity of different neurodevelopmental diseases, this method integrates de novo variant data of multiple diseases to enhance the genetic statistical power.

Description of drawings

1 is a schematic flowchart of a method for integrating multiple neurodevelopmental diseases and screening candidate genes according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a method for integrating multiple neurodevelopmental diseases and screening candidate genes according to the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

Referring to Fig. 1, the present invention provides a technical solution: a method for integrating multiple neurodevelopmental diseases and screening candidate genes, comprising the following steps:

S1. Genome functional annotation is performed on the detected variants to classify the variants into loss-of-function variants, deleterious missense variants, harmless missense variants, synonymous variants and non-coding region variants. Loss-of-function variants include frameshift variants, splice site variants, stop variants, and stop codon deletion variants. De novo variants, transmitted and non-transitive variants, and case-control variants are considered from the perspective of variant inheritance. Among them, genes carrying loss-of-function variants and deleterious missense variants are more likely candidates.

S2. Evaluate candidate genes by maximum likelihood estimation method to obtain the number of possible candidate genes for each disease. And implement it through the following specific steps:

Define the number of all possible loss-of-function variants and deleterious missense variants (K) of the disease, the number of genes with two or more deletion variants and deleterious missense variants detected in different patients for the same gene (R), and loss-of-function and deleterious missense variants. Proportion of sense variants contributing to disease (E).

Do one million permutation tests, and randomly select a certain proportion of candidate genes (T) from all coding genes in each permutation test (value range is 1-2500) and select a certain proportion according to the binomial distribution (K, E). Variation in proportions (C). Assume that no more than 2,500 genes are likely to cause each disease.

Loss-of-function and deleterious missense variants that contribute to disease (C) are assigned to candidate genes (T) and deleterious variants that do not contribute to disease (K-C) are assigned to other genes.

Each permutation test counts the number of genes extracted twice or more. If the number is the same as the actual observed number, the candidate gene (T) extracted this time is considered to be the number of possible causative genes for the disease.

After all permutation tests are completed, a prior probability distribution curve is obtained according to the number of all possible disease-causing genes, and the number of candidate genes corresponding to the point with the highest probability is the number of genes most likely to be related to the disease.

S3. At the same time, taking into account the disease mutation load, mutation rate, and the number of predicted candidate genes parameters, according to the loss-of-function mutation and deleterious missense mutation carried by each gene and the number of patients detected, a comprehensive evaluation is carried out to obtain a single gene. The level of the Bayes factor and the probability value of the false positive of the candidate gene according to the Bayes factor. And implement it through the following specific steps:

Observe the number of variants that appear in N families and the expected number 2Nμ, where μ represents the mutation rate of a specific gene. First, the Bayesian method is used to compare the two models, H0 and H1, where H0 is a gene that is not a disease-causing gene, and H1 is the base because of disease-causing genes. When H0 holds, the expected number of mutations is 2Nμ. When H1 is established, the expected number of variants is 2Nμγ, where γ represents the relative risk, and γ greater than 1 indicates a possible pathogenic gene.

The Bayes factor is defined as the probability of being H1 divided by the probability of H0. According to Bayes' theorem, the probability of H1 being positive is determined, a Bayes factor greater than 1 indicates that H1 is likely to be positive, and a Bayes factor greater than 100 indicates a strong proof that H1 is positive.

Multiply the Bayes factors obtained by different mutation types, each mutation type will get a separate Bayes factor, and multiply all the Bayes factors to get the gene-level Bayes factor.

Using the Bayesian FDR method to control the false positive rate, each gene will get an FDR value, that is, q-value.

It is defined that the q-value is less than a preset value A, and 0≦A≦1, indicating that the gene with probability A may be wrong.

S4. Further comprehensively evaluate the credibility of the candidate gene according to the data at the gene level. The credibility of the candidate gene is specifically: the specific expression level of the gene in the disease-related tissue, the existence of co-expression and protein between the gene and the known disease-related gene The degree of interaction was significantly enriched in pathways known to be associated with neurodevelopmental diseases.

Among them, it is generally believed that a disease-causing gene will have a higher variant load, that is, the same patient sample will detect a greater number of variants than randomly. This can be easily obtained by analyzing the Poisson test. We can compare the observed number of variants in N families with the expected number 2Nμ, where μ represents the mutation rate of a particular gene. But the problem with this approach is that the power of each variant type is not considered, for example, loss-of-function variants and missense variants may have different effects on disease. This method considers the contribution of different variant types to the disease phenotype when detecting the variation load of each gene. For example, loss-of-function variants that are high-risk variants will be given a relatively high weight.

For different classes of neurodevelopmental diseases, the method firstly identifies candidate genes by disease-by-disease analysis. At the same time, given the genetic similarity between different neurodevelopmental disorders, we adopted an approach that integrated all neurodevelopmental de novo variants. This allows the identification of novel causative genes that are shared across different diseases but cannot be identified in a single disease due to insufficient sample size.

The genes identified above are further enhanced by gene expression, protein interaction, enrichment analysis and other methods. Whether the gene is expressed in a specific disease-related tissue, and whether there is co-expression or protein interaction with known genes. Whether the candidate gene set is enriched in known disease-related gene sets or signaling pathways.

The method for integrating multiple neurodevelopmental diseases and screening candidate genes of the present invention has the advantages of integrating multiple new ideas of neurodevelopmental diseases, identifying new candidate genes, and enhancing gene reliability. The reliability of the method can be verified by existing genes, and new candidate genes can be searched according to the consistent variation of unknown genes. Combined with the mutation rate, the number of detection samples, the number of candidate genes and other parameters, the reliability of the candidate genes is statistically evaluated. There is a more reliable basis for the discovered disease-causing genes. In order to make up for the shortcoming of the insufficient sample size of a single neurodevelopmental disease, more candidate genes were identified at the same time. Based on the genetic similarity of different neurodevelopmental diseases, this method integrates de novo variant data of multiple diseases to enhance the genetic statistical power.

The above descriptions are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (5)

1. A method for screening candidate genes by integrating a plurality of neural developmental diseases, comprising: the method comprises the following steps:

s1, evaluating candidate genes by a maximum likelihood estimation method to obtain the possible number of the candidate genes of each disease;

s2, taking the disease variation load, variation rate and the number parameters of the candidate genes into consideration, and carrying out comprehensive evaluation according to the functional deletion variation and the harmful missense variation carried on each gene, the number of detected patients and the like to obtain a Bayesian factor of a single gene level and a probability value for judging the false positive of the candidate genes according to the Bayesian factor;

and S3, further comprehensively evaluating the credibility of the candidate genes according to the data of the gene level.

2. The method of claim 1, wherein the screening for candidate genes for integration into multiple neurodevelopmental diseases comprises: the candidate genes are evaluated by a maximum likelihood estimation method to obtain the possible number of the candidate genes of each disease, and the method specifically comprises the following steps:

defining the number (K) of all loss-of-function and deleterious missense variations in the disease, the number (R) of genes for which the same gene has detected two or more loss-and deleterious missense variations in different patients, and the proportion (E) of loss-of-function and deleterious missense variations contributing to the disease;

performing one million times of replacement tests, randomly extracting a certain proportion of candidate genes (T) (the value range is 1-2500) from all coding genes in each replacement test, and extracting a certain proportion of variation numbers (C) according to binomial distribution (K, E);

assigning loss-of-function variations and deleterious missense variations (C) that contribute to disease to candidate genes (T) and deleterious variations (K-C) that do not contribute to disease to other genes;

counting the number of the extracted genes twice or more in each replacement test, and if the number is the same as the actually observed number, considering the candidate genes (T) extracted this time as the number of possible pathogenic genes of the disease;

after all replacement tests are finished, a prior probability distribution curve is obtained according to all possible pathogenic gene numbers, and the candidate gene number corresponding to the point with the highest probability is the gene number most possibly related to diseases.

3. The method of claim 1, wherein the screening for candidate genes for integration into multiple neurodevelopmental diseases comprises: meanwhile, parameters such as disease variation load, gene expected variation rate, predicted candidate gene number and the like are taken into consideration, comprehensive evaluation is carried out according to functional deletion variation and harmful missense variation carried on each gene, the number of detected patients and the like to obtain Bayes factors of a single gene level and probability values of candidate gene false positives are judged according to the Bayes factors, and the probability values are specifically as follows:

observing the variation number and the expected number 2N mu of variation occurring in N families, wherein mu represents the variation rate of a specific gene, and comparing H0 and H1 by using a Bayesian method, wherein H0 is that the gene is not a pathogenic gene, and H1 is that the gene is a pathogenic gene; when H0 is true, the number of desired variations is 2N μ; when H1 is true, the expected number of mutations is 2Nμ γ, where γ represents the relative risk and γ is greater than 1 indicates a possible causative gene;

defining a bayesian factor as the probability of becoming H1 divided by the probability of H0; determining the probability that H1 is positive according to Bayes theorem, wherein a Bayes factor larger than 1 indicates that H1 is possibly positive, and a Bayes factor larger than 100 indicates that H1 is proved to be positive;

and multiplying the Bayes factors obtained by different variation types, wherein each variation type can independently obtain one Bayes factor, and multiplying all Bayes factors to obtain a gene level Bayes factor.

Controlling false positive rate by using a Bayesian FDR method, wherein each gene can obtain an FDR value, namely q-value;

defining q-value smaller than a preset value A, and 0 ≦ A ≦ 1, which indicates that the gene obtained with probability of A may be wrong.

4. The method of claim 1, wherein the screening for candidate genes for integration into multiple neurodevelopmental diseases comprises: further comprising: before performing step S1, the detected variants are genomically functionally annotated to classify the variants into loss-of-function variants, deleterious missense variants, tolerable missense variants, synonymous variants, and non-coding region variants.

5. The method of claim 1, wherein the screening for candidate genes for integration into multiple neurodevelopmental diseases comprises: the reliability of the candidate gene is specifically as follows: the specific expression level of the gene in the disease-related tissues, the co-expression degree and the protein interaction degree of the gene and the known disease-related genes are obviously enriched on the known pathways related to the neurodevelopmental diseases.

Images