A method for developing clinical diagnostic indicator algorithms for gene detection panels

Abstract

The present invention discloses a method for developing a clinical diagnostic indicator algorithm for a gene detection panel. The method belongs to the field of cell gene detection, and the specific steps are: providing a gene site information table to filter the sequencing data; simulating the sequencing data, and using the transmitted data as virtual gene detection panel data; using an existing indicator analysis algorithm to analyze the sequencing data; using an existing indicator analysis algorithm to analyze the panel data; integrating the analysis results and performing model training; evaluating the performance of various computational models and selecting the optimal solution. The present invention is based on the whole genome and whole exome sequencing data of a public database, and extracts these site data from the sequencing data through the site information of each gene in the gene detection panel to construct virtual gene detection panel data, and performs algorithm development on the virtual panel detection data, thereby improving the quality and efficiency of gene detection panel product development.

Description

Technical Field

[0001] The present invention belongs to the field of cell gene detection, and relates to a method for developing a clinical diagnostic indicator algorithm for a gene detection panel; through a new data analysis model, the development and optimization of clinical diagnostic indicators for gene (site) detection panel sequencing samples are realized. Specifically, the present invention is based on multi-omics sequencing data (including but not limited to whole genome sequencing, whole exon sequencing, whole genome methylation sequencing, whole transcriptome sequencing, etc.), by simulating the distribution pattern of gene sites under a specific detection panel, read segment enrichment bias and other characteristics, to help developers build a digital detection panel. On this basis, the detection value calculated in the detection panel is fitted and analyzed with the original detection value using an artificial intelligence algorithm, so that the detection panel has a detection performance consistent with multi-omics sequencing data. The invention can greatly reduce the development and testing costs of the detection panel. Background Art

[0002] In the prior art, the method for developing clinical diagnostic indicator algorithms for gene detection panels mainly adopts the method of collecting a large number of samples and performing gene panel detection on the samples to generate a large amount of data for algorithm development. However, this method requires a lot of money, time and manpower, and once the initial site design of the gene panel is wrong, it may bring great risks to product development; and because the current high-throughput sequencing of whole genome and whole exome is relatively expensive and covers more detection sites that are not related to the disease. Therefore, some gene detection panels are usually designed independently to detect the mutation status of sites of some important genes related to the disease, which can not only reduce the detection cost, but also focus on increasing the sequencing depth of these specific gene sites, thereby improving the sensitivity and accuracy of the detection results. However, based on the sequencing data generated by these gene detection panels, when performing some clinical diagnostic indicator analysis (such as TMB, MSI and other indicators), due to factors such as the bias of the selected gene combination, the results obtained by the existing indicator calculation method cannot fully reflect the true state of the sample; 1. Traditional method; At present, the following two methods are mainly used to construct and optimize the panel: (1) construct the panel from scratch by sampling in large quantities.The specific steps are as follows: a: Collect a large number of samples (such as 100 cases, 500 cases, etc.), and perform specific omics sequencing (such as whole exome sequencing, etc.) and detection panel sequencing on each sample; b: Use similar analysis algorithms to analyze the two sequencing methods to obtain a specific score for a certain indicator; c: Fit the indicator score obtained based on the detection panel to the indicator score obtained by specific omics sequencing, so as to obtain a standard score that can be used for clinical evaluation and diagnosis; the biggest disadvantage of this type of method is that the initial sample collection cycle is long, the cost is high, and it takes a lot of manpower; and once the initial site design of the gene panel is incorrect, it may bring great risks to product development; (2) Optimize the panel prediction algorithm based on public data; the specific steps are as follows: a: Collect relevant omics sequencing data based on the public database, and according to the Pa nel, and capture the corresponding area of ​​the collected sequencing data to simulate the sequencing data of the detection panel; steps b and c are consistent with steps b and c of the first method; although this type of method greatly reduces the cost of the preliminary preparation of the panel optimization; however, due to the bias of the panel's own technology, the area actually captured and the detection depth of different areas will be very different from the sequencing data in the existing public database; therefore, simply capturing the corresponding area for subsequent simulation analysis has limited effect, and may even obtain the opposite result from the actual detection and analysis process; therefore, the scope of use of this type of method is limited and it is difficult to promote it on a large scale; therefore, it is urgent to develop a new indicator analysis algorithm based on these gene detection panel data; the present invention is mainly aimed at the scenario of clinical diagnostic indicator application development of gene detection panels. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a method for developing a clinical diagnostic indicator algorithm for a gene detection panel.

[0004] Technical solution: A method for developing a clinical diagnostic indicator algorithm for a gene detection panel according to the present invention includes two transfer processes of constructing a virtual gene detection panel and developing a clinical indicator analysis algorithm for the data of the virtual gene detection panel;

[0005] 1. The specific delivery process of constructing the virtual gene detection panel is as follows:

[0006] (1) Provide information on all test sites involved in the designed genetic testing panel.

[0007] (2) Filtering the whole genome or whole exome sequencing data;

[0008] (3) Based on a series of sequencing-related parameters, the sequencing data retained within the detection site is simulated.

[0009] (4) Arrange and save the data (simulation) transmitted in step (3) as virtual gene detection panel data;

[0010] 2. The specific delivery process for developing a clinical indicator analysis algorithm for virtual gene detection panel data is as follows:

[0011] (5) Analyzing the whole genome or whole exome sequencing data input in step (2) using an existing index analysis algorithm;

[0012] (6) Analyze the virtual gene detection panel data provided in step (4) using the existing indicator analysis algorithm;

[0013] (7) Integrate the results of steps (5) and (6); correspond the result of each sample in step (5) to the corresponding sample in step (6) and mark it as the expected result of the sample;

[0014] Use appropriate machine learning algorithms to train models based on the integrated results;

[0015] (8) Evaluate the performance of various computational models and select the optimal solution.

[0016] Furthermore, in step (1), the information provided includes, but is not limited to, the location information of the site on the genome and the sequence information of the site.

[0017] Furthermore, in step (2), the whole genome or whole exome sequencing data is filtered specifically by: extracting the sequencing data based on the detection site information provided in step (1), and only retaining the sequencing data covered in the detection site;

[0018] Furthermore, in step (3), the parameters include but are not limited to the sequencing platform, the length of the sequence, the sequencing depth and the GC content of the sequence;

[0019] The simulation process includes but is not limited to refitting the read distribution and enrichment in the data transmitted in step (3) (sequencing data within the detection site) according to the parameter settings, so that the generated data is consistent with the sequencing data of the gene detection panel obtained under real conditions in terms of read distribution and enrichment.

[0020] Furthermore, in step (6), the analysis results analyzed by the indicator analysis algorithm are divided into two groups: a training set and a test set; the samples analyzed in this step must be consistent with the samples analyzed in step (5);

[0021] The training set and test set are obtained by randomly grouping the existing data in a ratio of 7:3, wherein 70% of the sample data is used as the training set for training the model; and the remaining 30% of the data is used as the test set for the final evaluation of the prediction performance of the model.

[0022] appendix:

[0023] Genetic testing panel: refers to the testing of not just one locus or one gene, but multiple loci and multiple genes at the same time; these loci and genes need to be selected and combined according to a standard to form a testing set; this testing set composed of gene loci is called a genetic testing panel.

[0024] Whole genome sequencing: All DNA fragments in the cell nucleus are collectively called the genome, and high-throughput sequencing of the genome is called whole genome sequencing.

[0025] Whole exome sequencing: There is a part of DNA fragment in the cell that can guide the encoding of proteins, and this part of DNA fragment is called "exon"; all fragments on DNA with these functions are called exome; high-throughput sequencing of exome is whole exome sequencing.

[0026] Sequencing depth: refers to the ratio of the total number of bases obtained by sequencing to the size of the genome. It is one of the indicators for evaluating the sequencing volume.

[0027] TMB: tumor mutation burden; defined as the total number of somatic gene coding errors, base substitutions, gene insertions or deletions detected per million bases; TMB is the latest marker for evaluating the therapeutic effect of PD-1 antibodies, and its effect has been confirmed in the treatment of various tumors.

[0028] MSI: microsatellite instability; a type of short tandem repeat DNA sequence in the genome, generally composed of 1-6 nucleotides, arranged in tandem repeats; due to the difference in the number of repetitions of its core repeating unit, microsatellites have population polymorphism. The occurrence of MSI is caused by functional defects in DNA mismatch repair in tumor tissues; the MSI phenomenon accompanied by DNA mismatch repair defects is an important tumor marker in clinical practice.

[0029] Read segment: The sequence fragment obtained by sequencing.

[0030] Omics: mainly includes genomics, proteomics, metabolomics, transcriptomics, lipidomics, immunoomics, glycomics, imaging omics, ultrasound omics, etc.

[0031] Beneficial effects: Compared with the prior art, the advantages of the present invention are as follows: the present invention is based on whole genome and whole exome sequencing data from a public database (or accumulated by itself), and extracts these site data from the sequencing data through the site information of each gene in the gene detection panel to construct virtual gene detection panel data, and performs algorithm development on the virtual panel detection data, thereby improving the quality and efficiency of gene detection panel product development and greatly reducing development costs and risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is an operation flow chart of the present invention;

[0033] Figure 2 It is a schematic diagram of TMB values ​​of two groups of data analyzed by linear fitting algorithm in the present invention;

[0034] Figure 3 It is a statistical diagram of the signal value of a single probe of the Beta mixed model in the present invention;

[0035] Figure 4 It is a schematic diagram of GCIMP values ​​of two sets of data analyzed by linear fitting algorithm in the present invention. DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0037] The present invention intends to realize the construction of a virtual gene detection panel and the development of a clinical indicator calculation method through the following scheme.

[0038] As shown in the figure, the present invention is divided into two transfer processes and 8 main steps.

[0039] The first transfer process is mainly used to build a virtual gene detection panel, and its specific steps are as follows:

[0040] Step 1: providing information of all detection sites involved in the designed gene detection panel. Step 1 of the present invention is to obtain information of all detection sites involved in the gene detection panel;

[0041] The information includes but is not limited to the location information of the site on the genome, the sequence information of the site, etc.; the information will be passed to the second step;

[0042] In addition, this information can be directly provided by the staff who designed the genetic testing panel, etc.; or the specific site information can be determined by performing sequence comparison analysis (such as using BWA or other comparison tools) on the trial sequencing sample data of the genetic testing panel;

[0043] Step 2: Filter the sequencing data of the whole genome or whole exome (or other omics); specifically, extract the sequencing data based on the detection site information provided in step 1, and only retain the sequencing data covered in the detection site; the retained data will be passed to step 3;

[0044] Step 2 of the present invention is to capture local site data based on whole genome, whole exon, whole transcriptome, whole genome methylation or other omics sequencing data (hereinafter referred to as reference data set);

[0045] The capture of site data in the present invention is mainly based on the site coordinate information obtained in step 1, and the read segment data of the corresponding site is captured in the reference data set;

[0046] The reference data set in the present invention can be obtained by downloading from a public database or other platform; it can also be the sequencing data accumulated by the staff themselves;

[0047] The method used for data capture in the present invention includes but is not limited to using tools such as BWA and samtools to extract data of specific sites from the reference data set;

[0048] Step 3: Simulate the data retained in step 3 based on a series of sequencing-related parameters, including but not limited to the sequencing platform, sequence length, sequencing depth, GC content on the sequence, etc. The simulation process includes but is not limited to refitting the read distribution and enrichment of the data transferred in step 3 according to the parameter settings, so that the generated data is consistent with the sequencing data of the gene detection panel obtained in real situations in terms of read distribution, enrichment, etc. The fitted data will be further passed to step 4;

[0049] Step 3 of the present invention is to fit the distribution characteristics of the data captured in the reference data set based on the data distribution characteristics in the gene detection panel, which mainly includes the following two methods:

[0050] First, we directly construct mathematical statistical models (such as Poisson distribution models, etc.) based on the sequencing platform, sequence length, sequencing depth, GC content and other parameters provided by the staff, and fit the number of reads at each point in the captured data to make the distribution characteristics of the number of reads consistent with the data distribution characteristics generated by the real genetic testing panel;

[0051] 2. Calculate the sequence length, sequencing depth, GC content and other information in the sample sequencing data of the gene detection panel through tools such as but not limited to BWA, samtools, flagstat, etc., and build a mathematical statistical model (such as Poisson distribution model, etc.) based on these parameter information to fit the number of reads at each site in the captured data so that the distribution characteristics of the number of reads are consistent with the data distribution characteristics generated by the real gene detection panel;

[0052] Step 4: Arrange and save the data transferred in step 3 as virtual gene detection panel data;

[0053] Step 4 of the present invention is to save the fitting data results in a format such as Fastq or BAM, and name it as virtual gene detection panel data for subsequent analysis.

[0054] The second transfer process is mainly used to develop clinical indicator analysis algorithms for virtual gene detection panel data. The specific steps are as follows:

[0055] Step 5: Analyze the whole genome or whole exome (or other omics) sequencing data input in step 2 using existing indicator analysis algorithms (including but not limited to TMB, MSI, etc.); since the standard calculation methods of most clinical indicators are based on whole genome / whole exome sequencing data, the results obtained in this step will be used as the gold standard for algorithm training in step 7;

[0056] Step 5 of the present invention is to calculate the corresponding index score for each sample in the reference data set using the index analysis algorithm widely used in the industry, and the calculated index score will be used as the gold standard;

[0057] The indicator scores here include but are not limited to the calculation methods of indicators such as MSI, HRD, TMB, etc.;

[0058] Step 6: The virtual gene detection panel data provided in step 4 is analyzed using the existing indicator analysis algorithm; the analysis results will be divided into two groups: training set and test set; the samples analyzed in this step must be consistent with the samples analyzed in step 5; the training set and test set data are obtained by randomly grouping the existing data in a ratio of 7:3, of which 70% of the sample data is used as the training set for training the model; the remaining 30% of the data is used as the test set to finally evaluate the prediction performance of the model;

[0059] Step 6 of the present invention is to calculate the corresponding index score for each sample in the virtual gene detection panel data using the index analysis algorithm widely used in the industry;

[0060] The indicator scores here include but are not limited to the calculation methods of indicators such as MSI, HRD, TMB, etc.;

[0061] Step 7: Integrate the results of steps 5 and 6; correspond the results of each sample in step 5 to the corresponding sample in step 6, and mark them as the expected results of the sample; use appropriate machine learning algorithms (including but not limited to support vector machines, deep learning algorithms, etc.) to perform model training based on the integrated results;

[0062] Step 7 of the present invention is to construct a prediction model using the index scores calculated in steps 5 and 6. The specific steps are as follows:

[0063] 1. Match the result of each sample in step 5 to the corresponding sample in step 6 and mark it as the expected result of the sample. Then divide the above paired sample results into two groups: training set and test set in a ratio of 1:1 (or 7:3, etc.);

[0064] Second, in the training set data, use a variety of machine learning algorithms (such as linear fitting, etc.) to train the model so that the scores calculated by the model based on the virtual gene detection panel data are similar to the scores calculated by the corresponding samples in the reference data set. Then evaluate the model prediction performance through the test set data;

[0065] Step 8: Evaluate the performance of various computational models and select the best solution;

[0066] Step 8 of the present invention is to select the best solution as the specific indicator calculation method of the virtual gene detection panel by comparing the prediction performance of each model in step 7.

[0067] Embodiment 1:

[0068] Constructing a lung cancer gene detection panel TMB prediction algorithm:

[0069] TMB is tumor mutation burden, which represents the density of non-synonymous mutations in protein coding regions. In some cancer types, patients with higher TMB may benefit from immunotherapy.

[0070] 1. Download 100 lung cancer exome sequencing data from the GDC website; at the same time, download the gene detection panel data designed by the commercial kit of MSK-IMPACT as the object of pre-simulation;

[0071] 2. Extract the corresponding read segments of the exome sequencing data according to the site information on the gene detection panel;

[0072] 3. Construct a Poisson distribution model based on the number of reads at each site of the gene detection panel sequencing data, and record various parameter information in the model;

[0073] 4. For the reads extracted from exome sequencing, according to the parameters of the Bosson model constructed in 3, reads are added and deleted again, so that the relative distribution of the number of reads at each site obtained based on exome sequencing is consistent with the relative distribution of the number of reads at each site on the gene detection panel;

[0074] 5. Using conventional TMB calculation methods, calculate the exon sequencing data and the read segment data obtained based on exon extraction, and obtain the TMB scores of the two sets of data respectively;

[0075] 6. The two sets of data were analyzed using a linear fitting model to construct a prediction model so that the TMB score calculated based on the read data extracted from the exon can be predicted by the model to be similar to the TMB score calculated directly based on the exon data, as shown in the following example. Figure 2 described.

[0076] Embodiment 2:

[0077] Constructing a brain tumor DNA methylation panel G-CIMP prediction algorithm:

[0078] G-CIMP is an epigenetic feature in gliomas, which refers to the methylation of a large number of CpG islands in the tumor; patients with this feature generally have a better prognosis;

[0079] 1. Download 100 cases of brain cancer Illumina 450K DNA methylation chip data from the GDC website; also download 10 cases of Illumina 27K DNA methylation chip data as pre-simulation objects;

[0080] 2. Extract the corresponding data from Illumina 450K DNA methylation data according to the site information on the Illumina 27K DNA methylation Panel;

[0081] 3. Construct a Beta mixture model based on the signal value of each site in the Illumina 27K DNA methylation data (see Figure 3 ), record various parameter information in the model;

[0082] 4. For the data extracted from Illumina 450K DNA methylation data, according to the Beta mixture model parameters constructed in 3, the data are re-added and subtracted so that the relative distribution of the signal value of each site is consistent with the relative distribution of the signal value of each site in the Illumina 27K DNA methylation data;

[0083] 5. Using the conventional G-CIMP calculation method, calculate the Illumina 450K DNA methylation data and the data extracted based on the Illumina 450K DNA methylation, and obtain the G-CIMP scores of the two sets of data respectively;

[0084] 6. The two sets of data were analyzed using a linear fitting model to construct a prediction model so that the score calculated based on the data extracted from Illumina 450K DNA methylation can be predicted by the model to be similar to the G-CIMP score calculated directly based on the Illumina 450K DNA methylation data (see Figure 4 ).

[0085] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for developing a clinical diagnostic indicator algorithm for a gene detection panel, characterized in that: It includes two transfer processes: constructing a virtual gene detection panel and developing a clinical indicator analysis algorithm based on the data of the virtual gene detection panel; 1. The specific delivery process of constructing a virtual gene detection panel is as follows: (1) Provide information on all detection sites involved in the designed genetic detection panel, including the location information of the site on the genome and the sequence information of the site; (2) Filtering the sequencing data of the whole genome or the whole exome; extracting the sequencing data based on the detection site information provided in step (1), and retaining only the sequencing data covered in the detection site; (3) Based on a series of parameters related to sequencing, simulating the sequencing data retained within the detection site; the parameters based on a series of parameters related to sequencing include the sequencing platform used, the length of the sequence, the sequencing depth, and the GC content of the sequence; The simulation process includes refitting the read distribution and enrichment of the sequencing data in the detection site according to the parameter settings, so that the generated data is consistent with the sequencing data of the gene detection panel obtained in the actual situation in the read distribution and enrichment; (4) Organize and save the simulated data and use it as virtual gene detection panel data; 2. The specific delivery process for developing a clinical indicator analysis algorithm for virtual gene detection panel data is as follows: (5) Analyze the whole genome or whole exome sequencing data that are filtered using the existing indicator analysis algorithm; (6) The provided virtual gene detection panel data is analyzed using the existing indicator analysis algorithm; the analysis results analyzed by the indicator analysis algorithm are divided into two groups: a training set and a test set; The training set and test set are randomly grouped according to the ratio of 7:3 for the existing data, where 70% of the sample data is used as the training set for training the model; the remaining 30% of the data is used as the test set to finally evaluate the prediction performance of the model; (7) Integrate the analysis results of steps (5) and (6): correspond the result of each sample in step (5) to the corresponding sample in step (6) and mark it as the expected result of the sample; Machine learning algorithms such as support vector machines and deep learning algorithms are used to train models based on the integrated results; (8) Evaluate the performance of various computational models and select the optimal solution.

2. A method for developing a clinical diagnostic indicator algorithm for a gene detection panel according to claim 1, characterized in that: Constructing a lung cancer gene detection panel TMB prediction algorithm: TMB is tumor mutation burden, which represents the density of nonsynonymous mutation distribution in protein coding regions; (1) Download the exome sequencing data of 100 lung cancer cases from the GDC website; at the same time, download the gene detection panel data designed by the commercial kit of MSK-IMPACT as the object of pre-simulation; (2) Extract the corresponding read segments of the exome sequencing data according to the site information on the gene detection panel; (3) Construct a Poisson distribution model based on the number of reads at each site of the gene detection panel sequencing data, and record various parameter information in the model; (4) For the reads extracted from exome sequencing, the reads are added and deleted again according to the constructed Bosong model parameters, so that the relative distribution of the number of reads at each site obtained based on exome sequencing is consistent with the relative distribution of the number of reads at each site on the gene detection panel; (5) Using conventional TMB calculation methods, the exon sequencing data and the read segment data obtained based on exon extraction were calculated to obtain the TMB scores of the two sets of data; (6) The two sets of data were analyzed using a linear fitting model to construct a prediction model so that the TMB score calculated based on the read data obtained by exon extraction can be predicted by the model to be similar to the TMB score calculated directly based on the exon data.

3. A method for developing a clinical diagnostic indicator algorithm for a gene detection panel according to claim 1, characterized in that: Constructing a brain tumor DNA methylation panel G-CIMP prediction algorithm: G-CIMP is an epigenetic feature in gliomas, which refers to the methylation of CpG islands in the tumor; (1) Download 100 cases of brain cancer Illumina 450K DNA methylation chip data from the GDC website; at the same time, download 10 cases of Illumina 27K DNA methylation chip data as pre-simulation objects; (2) Extract the corresponding data from Illumina 450K DNA methylation data according to the site information on the Illumina 27K DNA methylation panel; (3) Construct a Beta mixture model based on the signal value of each site of Illumina 27K DNA methylation data and record various parameter information in the model; (4) The data extracted from the Illumina 450K DNA methylation data were re-added and subtracted according to the constructed Beta mixture model parameters so that the relative distribution of the signal value of each site was consistent with the relative distribution of the signal value of each site in the Illumina 27K DNA methylation data; (5) Using the conventional G-CIMP calculation method, the Illumina 450K DNA methylation data and the data extracted based on the Illumina 450K DNA methylation were calculated to obtain the G-CIMP scores of the two sets of data; (6) The two sets of data were analyzed using a linear fitting model to construct a prediction model so that the score calculated based on the data extracted from Illumina 450K DNA methylation is predicted by the model to be similar to the G-CIMP score calculated directly based on the Illumina 450K DNA methylation data.

Images