CN117095744A - Copy number variation detection method based on single-sample high-throughput transcriptome sequencing data - Google Patents

Abstract

The invention discloses a copy number variation detection method based on single-sample high-throughput transcriptome sequencing data. The detection method first compares the genome sequencing data to the human reference genome, and calculates the comparison to the sequencing fragments on each gene. The amount of data, the expression matrix is obtained; then the expression matrix is input into the detection model to obtain the copy number variation detection results; the detection model is based on the deep neural network model and is obtained by training with the preprocessed known database samples as the data set; The detection method can better understand the genome structure and variation of a single individual; using lower cost and time, it not only provides gene expression level information, but also can obtain gene copy number information, which can better understand the function and function of the genome. Regulatory mechanisms and the impact of copy number variation on gene expression are of great significance for the study of genetic diseases, rare diseases and complex diseases.

Description

A copy number variation detection based on single-sample high-throughput transcriptome sequencing data method

Technical field

The invention relates to the technical field of bioinformatics analysis, and in particular to a copy number variation detection method based on single sample high-throughput transcriptome sequencing data.

Background technique

Copy Number Variations (CNV) refers to changes in the copy number of a region on a chromosome, that is, an increase or decrease in the number of DNA sequence repeats in the region. CNVs are common in the human genome and have been associated with the onset and progression of several human diseases.

CNVs can involve larger genome segments or even entire genes or multiple genes, thereby affecting gene expression and function. They may have important effects on human disease susceptibility, risk, and clinical manifestations. Some CNVs are closely related to the occurrence of genetic diseases, such as certain hereditary cancers, neurodevelopmental disorders (such as autism and intellectual disability), and certain congenital heart diseases.

In addition, CNV can also have an impact on drug response and individual responsiveness to drug treatment. Some CNVs can cause changes in the quantity or function of drug-metabolizing enzymes or drug targets, thereby affecting the metabolism and effects of drugs in the body.

Therefore, detect disease-associated CNVs and further understand the contribution and mechanisms of these variants to disease. These studies help improve understanding of disease, develop personalized medicine, and improve methods for disease prevention and treatment.

There are three main strategies for CNV analysis, namely whole genome (WGS), whole exome (WES) and targeted sequencing, and detection involves a variety of algorithms and methods. Here are some common algorithms:

1. Based on depth analysis (Read Depth): This method is based on the distribution density of sequencing reads on the genome to infer copy number variation. By comparing the read depth of the sample and the reference genome, regions with copy number gains or losses can be identified. However, read depth analysis has limitations for detecting smaller CNVs and is easily affected by factors such as sequencing depth and regional GC content.

2. Breakpoint analysis (split reads): This method detects CNVs by analyzing the breakpoint positions of copy number variations. It can use paired-end sequencing data or long-read sequencing data to find breakpoint regions and infer the location and size of copy number variations. However, breakpoint analysis requires high-quality sequencing data and precise breakpoint positioning, and is more challenging for complex structural variations.

3. Segmentation-based Methods: These methods divide the genome into continuous fragments and compare the read depth or other characteristics of each fragment. By detecting copy number variation between fragments, the presence of CNV can be determined. However, segmented comparison methods may suffer from false positives or false negatives when identifying small CNVs and complex structural variants.

In contrast, there are few detection solutions to resolve CNVs in transcriptome data, possibly for the following reasons:

1) Due to the noise and technical errors in transcriptome sequencing data itself, such as sequencing errors, alignment errors, and expression estimation errors, etc. These errors may affect the results of CNV detection and require appropriate data correction and correction. Traditional CNV analysis methods may not be applicable;

2) The gene expression signal and CNV signal in the data may interfere with each other, which will cause the detection of copy number variation to be interfered with and affected by gene expression;

3) The coverage of the genome by RNA-seq is relatively sparse, which means that some regions may have higher coverage, while other regions may have insufficient coverage. This uneven coverage affects CNV detection accuracy and sensitivity, making it highly challenging or even impossible for comparative depth-based methods to detect precise CNV breakpoints and small CNV fragments.

4) Current sequencing analysis of CNV usually relies on the construction of a reference baseline. A reference baseline refers to the sequencing and analysis of a large number of individuals to determine the genomic variation in the normal population. By comparing to a reference baseline, variations in an individual's genome can be determined and the presence and copy number of CNVs inferred. However, there are also some limitations in the construction of reference baselines, such as:

i) Sample number and diversity limitations: The quality and representativeness of a reference baseline depends on the number and variety of samples included. If the number of reference baseline samples is limited or does not have sufficient diversity, some population-specific CNVs may not be accurately captured.

ii) Detection of rare and individual-specific CNVs: Reference baselines usually focus on common CNV variants. However, there may be differences between different samples and data sets, and accurate baseline information may not be provided for rare or individual-specific CNVs. . These variants may have important effects on an individual's disease susceptibility and phenotypic characteristics.

iii) Changes in the experimental environment: Changes in the experimental environment may include changes in temperature, humidity, lighting and other factors, or replacement of experimental equipment and reagent batches. These changes may lead to inconsistencies in experimental conditions, resulting in large differences between the samples used to construct the baseline and the samples actually analyzed, which will have a greater impact on the analysis results.

Therefore, most of the current transcriptome sequencing data are used to detect the expression of genes and transcripts to estimate gene activity, or to identify single nucleotide polymorphisms (SNPs) and short indels. However, they include There is a wealth of information about genomic variation in samples that is underexploited. Among these variations, copy number variations (CNVs) are very important for cancer research because they are the main genetic drivers of cancer. However, identifying CNVs from RNA-seq data is very challenging because the dynamic and highly uneven coverage of the genome by RNA-seq signals makes it difficult to distinguish between deletion and amplification events and dynamic changes in gene expression levels.

Therefore, traditional CNV analysis methods that only rely on reference baselines and depth-based analysis may have certain limitations when used on transcriptome data. There is an urgent need for a flexible and accurate method for detecting CNVs.

Contents of the invention

The purpose of the present invention is to provide a copy number variation detection method based on single sample high-throughput transcriptome sequencing data, overcome the limitations of traditional CNV analysis methods, and achieve flexible and accurate detection of CNV.

In view of this, the solution of the present invention is as follows:

A copy number variation detection method based on single sample high-throughput transcriptome sequencing data, including the following steps:

Compare the genome sequencing data to the human reference genome, calculate the amount of data compared to the sequenced fragments on each gene, and obtain the expression matrix;

Input the expression matrix into the detection model to obtain the copy number variation detection results;

The detection model is based on a deep neural network model and is trained using pre-processed known database samples as the data set; the pre-processing includes standardizing the gene expression of the database samples and classifying copy number variation types.

Further, before the data comparison, the genome sequencing data is preprocessed to remove low-quality sequences and excise continuous low-quality bases.

Further, before training, the copy number type of the database sample is converted into a numerical value that can be used by the deep learning algorithm.

Further, the deep network model initializes the weights and biases of the neural network in a random initialization manner.

Further, during the training process of the detection model, the hidden layer is activated and the sample gene expression is mapped to max(0,x), that is, when x is greater than 0, x is output, otherwise 0 is output.

Further, during the training process of the detection model, the output layer is activated, the input vector is normalized, and the probability value of each element is obtained.

Further, the cross-entropy loss function is used during the training process of the detection model and is minimized.

The invention also provides a copy number variation detection system based on single sample high-throughput transcriptome sequencing data, including:

Computational comparison module: Compare the genome sequencing data to the human reference genome, calculate the amount of data compared to the sequenced fragments on each gene, and obtain the expression matrix;

Detection module: input the expression matrix into the detection model to obtain the copy number variation detection results;

The detection model is based on a deep neural network model and is trained using pre-processed known database samples as the data set; the pre-processing includes standardizing the gene expression of the database samples and classifying copy number variation types.

The present invention also provides a computer device, including a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the steps of the above detection method.

The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above detection method are implemented.

Compared with the existing technology, the beneficial effects of the present invention include but are not limited to:

The detection method proposed by the present invention uses a single sample for CNV analysis to better understand the genome structure and variation of a single individual; using lower cost and time, it not only provides information on the expression level of the gene, but also can obtain the copy number of the gene. Information can better understand the function and regulatory mechanism of the genome, as well as the impact of copy number variation on gene expression; the detection method can identify potential disease-related CNVs, thereby aiding the diagnosis and prediction of diseases. This is of great significance for the study of genetic diseases, rare diseases and complex diseases.

Description of the drawings

Figure 1 is a flow chart of the copy number variation detection method for single sample high-throughput transcriptome sequencing data according to the present invention.

Detailed ways

In order to make the purpose, technical solutions and beneficial technical effects of the present invention clearer, the present invention will be further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described in this specification are only for explaining the present invention and are not intended to limit the present invention.

Using a single sample for CNV analysis can address several pain points and limitations, including:

Difficulty in sample collection: Traditional CNV analysis often requires a large number of samples to obtain meaningful results. However, in some cases, obtaining a sufficient number of samples may be difficult or expensive. Using a single sample for CNV analysis can solve the problem of sample collection, making the analysis more convenient and feasible.

Constructing a reference baseline is costly and time-consuming: Traditional CNV analysis usually requires long sample collection time and high analysis costs. Using a single sample for analysis saves the time and cost of collecting samples and results can be obtained faster.

In order to solve the cost problem caused by the need to construct a baseline for detecting copy number variations in transcriptome data, which requires sequencing a large number of samples, and the problem that traditional DNA CNV detection methods cannot be used for RNAseq data, in one embodiment, a method based on a single The method for detecting copy number variations in high-throughput genome capture sequencing data of samples is shown in Figure 1, which specifically includes the following steps:

1. Use fastp software to process the patient's genome sequencing data for data quality, remove low-quality sequences, excise continuous low-quality bases, and then align the high-quality sequences to the human reference genome hg19.

2. Then use a gene annotation file (such as a GTF or GFF file) to determine the location of each gene. Gene annotation files provide transcript and exon position information for each gene. According to the alignment results, the reads are associated with the genes, and the data amount of the reads aligned to each gene is calculated to obtain an expression matrix.

3. We then downloaded the sample data from the public database. Each sample has the expression level of the gene and the corresponding copy number variation type, which is used to build the model. The specific steps are:

1) Data preprocessing: Calculate the mean (μ) and standard deviation (σ) of each gene expression (x), and standardize each sample gene using the following formula:

z=(x-μ)/σ;

2) Divide the copy number type into normal (copy number equal to 2), missing (copy number less than 2), increase (copy number greater than 2), and use One-Hot Encoding to convert the copy number type into available Numerical representation used by deep learning algorithms.

4. Then divide the gene expression and copy number data obtained by the above processing into a training set and a test set.

1) Initialize network parameters: Initialize the weights and biases of the neural network using random initialization.

2) Set up an input layer (with 3 features), two hidden layers (with 16 neurons each) and an output layer (with 3 neurons)

3) Activating the hidden layer: Use the ReLU activation function to activate the hidden layer and map the input x to max(0,x), that is, when x is greater than 0, x is output, otherwise 0 is output.

4) Activate the output layer: Use the Softmax activation function to activate the output layer. The input vector is normalized, converting the value of each element into a probability value between 0 and 1, and the sum of all elements is 1.

5) Parameter optimization: The cross-entropy loss function used in the training process of the model uses the Adam optimizer to optimize and minimize the cross-entropy loss function to improve the accuracy of the model.

6) Verify the model: Use the test set as the input of the model to verify the accuracy of the model, as shown in Table 1.

Table 1:

Gene Prediction accuracy ENSG00000000457.14 0.9122 ENSG00000000460.17 0.9298 ENSG00000000938.13 0.9298 ENSG00000000971.16 0.9122 ENSG00000001460.18 0.9298

5. The last step is to input the expression moment of the sample into the above model to obtain the copy number prediction result of each gene.

In the above embodiments, the detection method can be used to detect the presence or absence of copy number variation in a region on a chromosome in a single sample, but cannot directly diagnose one or more diseases. Only as a single sample CNV result, in order to better understand the function and regulatory mechanism of the genome, as well as the impact of copy number variation on gene expression, it is of great significance for the study of genetic diseases, rare diseases and complex diseases.

The following is an example of CNV detection using certain genome sequencing data as an example:

(1) Sequencing data preprocessing

Obtain fastq data, the statistics are shown in Table 2.

Table 2:

Samples Totalreads Totalbases(bp) Q20(%) Q30(%) read1 54593963 8189094450 97.52 93.27 read2 54593963 8189094450 97.52 93.27

(2)Fastq data processing

After quality control, high-quality sequences were obtained, and the data statistics are shown in Table 3.

table 3:

(3) Sequence comparison with reference genome

The alignment of sequence data with the human reference genome hg19 is shown in Table 4.

Table 4:

(4) Calculate the number of reads for each gene, as shown in Table 5.

table 5:

(5) Enter the detection model to obtain the copy number variation detection results, as shown in Table 6.

Table 6:

chromosome Variation type ENSG00000007908.16 Gain ENSG00000007923.16 Gain ENSG00000007933.13 Gain ENSG00000007968.7 Normal ENSG00000008118.10 Normal ENSG00000008128.23 Normal ENSG00000008130.15 Normal ENSG00000009307.16 Normal

The present invention is not limited to what is described in the specification and embodiments, and therefore other advantages and modifications can be easily realized by those skilled in the art without departing from the spirit and scope of the general concept as defined by the claims and equivalent scopes. The invention is not limited to the specific details, representative arrangements and examples described herein.

Claims (10)

1. A copy number variation detection method based on single sample high throughput transcriptome sequencing data, comprising the steps of:

comparing the genome sequencing data to a human reference genome, and calculating the data quantity of the sequencing fragments on each gene to obtain an expression matrix;

inputting the expression matrix into a detection model to obtain a copy number variation detection result;

the detection model is obtained by training a data set by using a preprocessed known database sample based on a deep neural network model; the pretreatment comprises the standardization of the gene expression quantity of the database sample and the division of copy number variation types.

2. The method of claim 1, wherein the data alignment is preceded by pretreatment of genomic sequencing data to remove low quality sequences and excision of consecutive low quality bases.

3. The method of claim 1, wherein the database sample converts the copy number type to a value that can be used by a deep learning algorithm prior to training.

4. The method of claim 1, wherein the deep network model initializes weights and biases of the neural network in a random initialization manner.

5. The method according to claim 1, wherein the hidden layer is activated during the training of the detection model, and the sample gene expression level is mapped to max (0, x), i.e., x is output when x is greater than 0, otherwise 0 is output.

6. The method according to claim 1, wherein the output layer is activated during the training process of the detection model, the input vector is normalized, and the probability value of each element is taken.

7. The method according to claim 1, wherein the cross entropy loss function is used in the training process of the detection model and is subjected to a minimization process.

8. A copy number variation detection system based on single sample high throughput transcriptome sequencing data, comprising:

calculating a comparison module: comparing the genome sequencing data to a human reference genome, and calculating the data quantity of the sequencing fragments on each gene to obtain an expression matrix;

and a detection module: inputting the expression matrix into a detection model to obtain a copy number variation detection result;

the detection model is obtained by training a data set by using a preprocessed known database sample based on a deep neural network model; the pretreatment comprises the standardization of the gene expression quantity of the database sample and the division of copy number variation types.

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1-7 when the computer program is executed.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any of claims 1-7.