KR102078200B1 - Analysis platform for personalized medicine based personal genome map and Analysis method using thereof - Google Patents

Abstract

The present invention has been made to improve the requirements for realizing the commercialized "Personal Map-Based Customized Medicine Platform" based precision medicine. The present invention is directed to improving the precision and recall or reproducibility of personal genomes. It is designed to increase the method and to use the common GWAS (genome wide association study) marker to perform genotyping-based patient stratification, and to improve and improve the problem in linkage with Mach Supercombe developed by Korea Electronics and Telecommunications Research Institute. will be. In addition, research results for the improvement of commercialization of personal genetic map-based customized medical platform for use as customized medical and diagnostic device including reporting database analysis module that can be easily provided to multiple users in genomic big data and supercom environment. to be. When the technology is commercialized, most countries' genomic data-based customized medicine will remain in Academia, and in the case of only partially available reality, a customized medical platform can be provided that can be used in the first whole area.

Description

Analysis platform for personalized medicine based personal genome map and analysis method using

The present invention compares a large number of full-length (exome or target) genome DBs constructed by the Genome Project with personal full-length genome information input, and personal genome map-based customization using a supercomputing system that analyzes and provides genetic information from the personal genome. Invention for the medical analysis platform.

Current trends in the IT market are in the order of Google, Facebook, Amazon, Cloud Computing and Ubiquitous. At the same time, the biomedical, bioinformation, and genome areas are also bio google, systems. Bio, personalized medicine and precision medicine are changing in line with the new trend. In particular, the post-human genome project has been actively developed to realize personalized medicine with the rapid development of the next generation sequencing technology.

Current generation sequencing technology is known to take about one week for sequencing (detoxing) and analyzing one human (x30) full-length dielectric. It is reported that more than 100,000 next-generation sequencers are currently available worldwide, with significant funds being invested in major developers of the third-generation sequencer (Ion Torrent: 2.5 generations, the third generation of Pacific BioScience).

In addition, globally, this field is the fastest developing and developing field among all businesses. If this trend continues, one full-length genome sequencing and analysis is expected to be below $ 1,000 in the next two to three years. The most versatile and immediately available technologies based on these next-generation technologies are clinical genomics, pharmaco-genomics, and translational medicine, and these clinical genomes have recently been medical genomics, and the medical genome, along with the patient stratification technology, has created a new discipline and neologism called the precision medicine that the president of the United States referred to.

As such, genome variation related information is increasing year by year, and the present invention will continuously expand the area of analysis accuracy by expanding verification data.

On the other hand, Applicant is carrying out the development of continuous technology to improve the technical requirements in the field of genetic analysis mentioned.

As a result of these efforts, we developed methods for precision medicine, such as bio big data, clinical information, protein and genomic information, and analytical systems to speed up their analysis. In particular, we have developed a graphical processing unit (GPU) based analysis system for analysis speed (Patent registration: 10-0996443), and the features of the RVR (records virtual rack) analysis tool, which is a technique for improving data comparison speed, Based on the file, an information retrieval method (patent registration: 10-0880531, patent registration: 10-1035959, and patent registration: 10-1117603) was developed.

In addition, alleles were applied to the protein based on RVR and graphic process unit (patent registration: 10-1400717), to define the variant calling and to efficiently determine the degree of rare variation between the control and the individual genome. A gene depth based ADISCAN analysis tool was developed (Patent Registration: 10-1460520, 10-1542529, and 10-2014-0020738).

And genotyping method for efficient genome information management, mutation discovery for disease cause, and patient stratification (Patent registration: 10-2015-0187554, 10-2015-0187556, and 10-2015- 0187559) and a method for calculating human haplo typing from genome information (patent application: 10-2016-0096996).

In addition, middleware specialized in storage operation for big data such as integrated dielectric DB is able to analyze thousands of dielectric bulk data at the same time in parallel distributed environment developed by Korea Electronics and Telecommunications Research Institute (ETRI). Systems (Patent Registrations 10-1460520, 10-1010219, 10-0956637, 10-093623, 10-2013-0005685, 10-2012-0146892 and 10-2013-0004519) have been developed.

Applicant developed Mach System from Korea Electronics and Telecommunications Research Institute, developed an optimized environment using bio big data for clinical application, and developed Korea's first supercomputing system linked with an integrated genome analysis system for precision medicine. .

In particular, although Mach-Fs (storage systems for ultra-fast I / O for buck data such as dielectrics) are tailored to general cloud computing environments, we have clearly defined reproducibility, precision, and system limitations. Mach-FsDx has been developed for use in diagnostics.

(001) Republic of Korea Patent No. 10-0880531 (002) Republic of Korea Patent No. 10-0996443 (003) Republic of Korea Patent No. 10-1035959 (004) Korea Patent Registration No. 10-1117603 (005) Republic of Korea Patent No. 10-1400717 (006) Korean Patent Registration No. 10-1460520 (007) Korean Patent Registration No. 10-1542529 (008) Republic of Korea Patent Application No. 10-2015-0187554 (009) Republic of Korea Patent Application No. 10-2015-0187556 (010) Republic of Korea Patent Application No. 10-2015-0187559 (011) Republic of Korea Patent Application No. 10-2016-0096996 (012) Korean Patent Registration No. 10-0834574 (013) Republic of Korea Registered Patent No. 10-1010219 (014) Korean Patent Registration No. 10-0956637 (015) Republic of Korea Registered Patent No. 10-0936238 (016) Republic of Korea Patent Application No. 10-2013-0005685 (017) Republic of Korea Patent Application No. 10-2012-0146892 (018) Republic of Korea Patent Application No. 10-2013-0004519

The present invention has been made to improve the requirements for realizing such commercialized personal genomic map-based precision medicine, the present invention improves the integrated dielectric DB that can improve the speed and efficiency of mutation detection of personal genome In addition, the accuracy and reproducibility of the method for improving common class wide association study (GWAS) marker-based patient stratification in order to increase the efficiency of patient stratification, and the algorithm method for variation that is a component of the integrated genome DB It is designed to improve and improve the problem of integration with the integrated dielectric database and Mach Supercomm, such as the task of increasing recall or reproducibility and defining accurate variation.

In addition, the present invention is intended for use as an integrated dielectric DB and Mach supercomputing-based customized medical and diagnostic device that includes a reporting database analysis module that can provide the detected mutation information to multiple users in a big data and supercomputing environment. To provide a personalized genomic map-based custom medicine platform.

According to a feature of the present invention for achieving the above object, the present invention provides an analysis platform comprising an integrated dielectric DB, biomedical information, protein database and system-related technologies for analyzing them. Has features to improve.

In addition, the present invention in order to operate the integrated dielectric database according to the present invention Korean Patent Nos. 10-0834574, 10-1010219, 10-0956637, 10-0936238 and Republic of Korea Patent Application No. 10- In order to improve the user environment, function and operation efficiency of Mach-FsDx for one diagnostic device by interworking with Mach Supercom System disclosed in 2013-0005685, 10-2012-0146892, 10-2013-0004519 Has

In order to provide commercialized platform-based services for personal genome analysis, optimization of variant calling, patient stratification and integrated genome systems is needed. To this end, the present invention applies the technical features 1 to 9 for the construction of variant calling, patient stratification and integrated genome system.

First, the present invention applies three technical features with respect to variant calling.

The variable calling is a standard of analysis results and should be standardized without bias.

However, currently known models of genome mutations include GATK, STRELKA, EBcall, MuTect, and SomaticSniper, which are developed based on the Bayesian model, and based on the method of probabilistic prediction and graph graphs (Fisher's Exact Test and string graphs). JointSNVMix, VarScan and Fermi. In addition, there are GNUmap, GATK, SOAPsnp, SAMTools, and SNVer, which are developed based on prior knowledge and statistical models.

However, the above three different models produce biased results because the model itself contains biases.

In other words, tools developed under the Bayesian model (GATK, STRELKA, EBcall, MuTect, and SomaticSniper) have been optimized based on prior knowledge, so that the variations known as prior knowledge are good at predicting, but new variations In some cases, the prediction is not good.

In the case of probabilistic prediction and graph model-based algorithms (jointSNVMix, VarScan, and Fermi), general statistics calculate how far a particular variation is from the normal distribution. However, actual mutations do not have this normality and may be sensitive to certain genes, certain diseases, and certain environmental factors.

On the other hand, tools based on knowledge and statistical models (GNUmap, GATK, SOAPsnp, SAMTools, and SNVer) use a variety of statistics based on prior knowledge based on Bayesian models.

Accordingly, the present invention applies ADISCAN to exclude bias in variant definitions.

The basic concept of ADISCAN is that all human genes have drainages (mother's genes and dad's genes), and when fragmented and detected again, the ratio of different alleles in base fragments is 50:50. Under the assumption of detection, if the ratio of alleles detected is 50:50, it is defined as a hetero variant (homozygote), if it is 100: 0, it is defined as homo (only reference homo is collected), and if it is 0: 100, it is defined as alternative homo. (only alternative homos are collected).

And this can be expressed as non-biased ratio (50/50 = 1, 0/100 or 100/0 = 0), and all these numbers can be represented by tangent and various other functions (but not tangent) Is suitable for expressing property properties).

As such, the basic concept of ADISCAN has been disclosed in the applicant's patent registration Nos. 10-1400717 and 10-1542529.

Thus, the present invention, by using the ADISCAN, in performing the variation definition, further defined the following technical features.

That is, in the present invention, the ADISCAN is used to standardize the highest and lowest points of the variation sensitivity scores according to the allele depth in units of 0 to 1, and perform the variable definition. The scope of the Twilight Zone is defined and constant parameters are used (Feature 1).

In the present invention, the ADISCAN uses the ADISCAN to designate a standard material (NA12878), perform machine learning based on the standard material as a correct answer, and determine the above-described constant parameters, and false positives (false). All mutations that are subject to a positive are compared to the standard (NA12878) to detect errors (Feature 2).

In addition, the present invention performs a haplotype-based correction in order to improve the accuracy of variant calling of the human genome in performing mutation definition using the ADISCAN (Feature 3).

Next, in the context of patient stratification, two technical features apply.

Patient stratification is available for realistic clinical practice and needs to be standardized in an international environment.

The present invention performs patient stratification through genetic variation using an integrated genome data-based haplotype sequence, and the basic principle of such a stratification is disclosed in the applicant's prior patent documents 8, 9 and 10. There is a bar.

This method of patient stratification clusters gene haplotype-based gene units and multigenes in genes and multigene units, which is highly applicable when there is sufficient population genome information, but due to race international data. If difficult to secure, there was a problem of low usability.

That is, the efficiency is improved when the patient stratification is calculated in the gene unit based on the common GWAS (genomewide association study) marker known as the international consortium (international consortium), so the patient stratification of the common GWAS marker-based gene unit method is required.

However, public GWAS markers have a problem due to sparse distribution, that is, many markers are located between genes and between genes or introns, thus preventing linkage with protein functionality.

Currently, the GWAS marker is a public marker created by the GWAS catalog (https://www.ebi.ac.uk/gwas/, European Biological Information Agency, EBI), and the DB is a disease-related study that calculates using thousands or hundreds of thousands. It is the result of an international consortium and an important asset in patient stratification research.

However, since 500,000 mutations were required to represent 3 billion bases, it was used in a way of having 1 base at 12,000 bp of 6,000 base pairs before and after around one base (or mutation).

The problem is that most of these are also non-functional variations.

In order to solve such a problem, the present invention is applied to a technique for performing patient stratification of gene units based on linkage disequilibrium (LD) (Feature 4).

In addition, the present invention (Feature 5) is applied to secure the association through the connection between the bio-data in order to compensate for the leanness of the common marker.

Finally, in the context of the integrated dielectric system configuration, the present invention applies four technical features.

The basic configuration of the integrated dielectric system applied to the present invention has been disclosed in Korean Patent Registration No. 10-1460520 and Korean Patent Application No. 10-2015-0187554.

However, the integrated dielectric system, as disclosed, has to calculate and integrate a large number of human genome data simultaneously based on many CPUs in a supercomputing environment using 100-1,000 pc-clusters (server-class nodes). In such an environment, in order to construct a pc-cluster at a reasonable price, a partitioning structure of analytical data for performing standardized electric field dielectric analysis is required.

In order to provide such a standardized analytical data partitioning structure, the present invention divides 3 billion electric field dielectrics into a defined number of chunks and allocates an optimized memory size per node (multiple CPU). 6).

In addition, the present invention provides a memory-based CPU, GPGPU-based GPU and I / O performance (1000 simultaneous calculations), memory-cache-server and to construct a supercomb for genome analysis that improves the efficiency of electric field dielectric analysis It provides a standardized configuration (parallel 7) for distributed storage.

In addition, the present invention provides a limit and reproducibility test results (feature 8) to apply such hardware as a clinical diagnostic device.

And finally, the present invention provides a standardized database equipped with precomputation and common tools for large-scale systems, such as personal genetic map-based custom medicine platforms, for general research developers to use (Feature 9).

Using the personal genomic map-based customized medical analysis platform according to the present invention as described above, it becomes possible to commercialize the customized medical platform, enable precise execution of genome variant calling, and the result of patient stratification is It can be improved from the academic level to the commercialized level, and the integrated dielectric DB system has the effect of being automated in a semi-automated environment.

1 is a schematic diagram illustrating a core technology configuration in a personal genomic map-based customized medical analysis platform and an analysis method using the same according to the present invention.
Figure 2 is a conceptual diagram showing the machine learning concept of the algorithm for dielectric mutation (ADISCAN) according to the present invention.
3 is a conceptual diagram illustrating an algorithm filtering and half-type based error correction concept of a standard material-based dielectric variation correction according to the present invention.
Figure 4 is an illustration showing the basic concept of haplotyping applied to the personal genomic map-based custom medical analysis platform and an analysis method using the same according to the present invention.
5 is a conceptual diagram summarizing the history and concepts of genome technology for patient stratification according to the present invention.
Figure 6 is an exemplary view showing an example of utilizing the common GWAS markers and Bio-map for the stratification of patients in the integrated dielectric DB according to the present invention.
7 is an exemplary view showing a GWAS marker-based patient stratification generation method according to the present invention.
8 is an exemplary view showing an example of generation of GWAS marker-based patient stratification according to the present invention.
9 is a conceptual diagram showing the divided chunks for the human genome 3 billion bases for operation in the PC-cluster applied to the present invention.
10 is a conceptual diagram showing the configuration of the calculation equipment (a1-a6), storage (b1-b3) and Mach-FsDx for analyzing the genome data according to the present invention.
11A to 11F illustrate results obtained by evaluating Mach-Fs to Mach-FsDx according to the present invention for stability, limit and reproducibility.
12A and 12B are exemplary diagrams verifying the maximum performance of the MAHA-FsDx configuration scale with respect to the limits.
13 is a conceptual diagram showing the configuration of a multi-researcher database based on the integrated dielectric DB platform according to the present invention.
14 is a conceptual diagram showing the configuration of allele depth, genotype and haplotype DB of the integrated dielectric DB according to the present invention.
15 is a conceptual diagram illustrating the detection principle of SNV, INDEL and CNV in each chunk unit matrix according to the present invention.
FIG. 16 is a conceptual diagram illustrating patient stratification through gene and multigene-based genotyping in the integrated genome DB according to the present invention. FIG.

Hereinafter, a personal genomic map based customized medical analysis platform according to a specific embodiment of the present invention will be described in detail with reference to the accompanying drawings.

Before describing each feature of the present invention, the configuration of the gene analysis service to which the present invention is applied will be briefly described. Genetic analysis services collect samples such as blood from individual gene collection institutions such as hospitals, and refer the samples to DNA sequencing companies.

The DNA sequencing company performs DNA (NGS, next generration sequencing) readout from the collected samples. Of course, in recent years, since the DNA sequencing can be generated by various methods according to the technical development, the DNA translation generating method can be performed by various methods according to the technical level of the DNA sequencing company. The DNA decoded as described above is analyzed genetic information contained in the personal genome through a system such as the present invention, the analyzed information is transmitted to a diagnostic institution or consumer, such as a hospital.

Of course, when the DNA bulk (bulk, dummy) data is provided from the DNA sequencing company, it is also possible to form a high-density indexing file from it to analyze the genome sequences that are big data.

That is, the present invention relates to an analysis and diagnosis system for analyzing the genetic information contained in the individual genome from the DNA decoded information, hereinafter, personal genomic map-based customized medical analysis platform using integrated genome big data and In the analysis method using the same, with reference to the technical features 1 to 9, the technical configuration of the present invention will be described in detail.

1 is a schematic diagram illustrating a core technology configuration in a personal genomic map based customized medical analysis platform and an analysis method using the same according to the present invention. As shown here, the technical features 1 to 9 are applied to the implementation of variant calling, patient stratification, and integrated genome systems in order to provide commercialized platform based services commercialized in personal genome analysis. .

FIG. 2 is a conceptual diagram illustrating the machine learning concept of the algorithm for dielectric mutation (ADISCAN) according to the present invention, which illustrates the concept of technical feature 1 of the present invention.

As shown in FIG. 2, feature 1 of the present invention uses ADISCAN to standardize the highest and lowest points of variation sensitivity scores according to allele depths in units of 0 to 1 in performing variation definition. The scope of the Twilight Zone is not defined and it is a constant parameterization.

That is, a score function for machine learning is applied, and the score function applies a variable calling of the genome using a tangent function to which three constant parameters (a, b, and c) are applied, but the constant parameter ( a) is the variation sensitivity according to allele depth, constant parameter (b) indicates the highest and lowest points of the score in 0-1 units, and constant parameter (c) is the area where variable calling is not possible ( Set the scope of the Twilight Zone.

That is, the score function for ADISCAN according to the present invention is

It is calculated by the score function S (i) = tan (D-0.5) x a-log b (max (b, min (depth))) + c.

The score function formula above means that the allele depth ratio (= D), the smaller of the first allele depth and the second allele depth, b (constant) ) Can be expressed as the logarithm value of the logarithm base b (constant), the weight a (constant) of tan (tangent function), and c (constant) for a scale of 0 to 1. will be.

In this way, knowing the constants a, b and c, it is possible to make a non-biased determination of the mutation as homo, hetero or alternative homo.

To this end, constants were calculated by performing machine learning in the form of constants a, b, and c on the known answers.The constant parameter (a) is the variation sensitivity according to the allele depth, the constant parameter (b) is the highest point of the score and The lowest point is expressed in units of 0-1, and the constant parameter (c) optimizes the range of the Twilight Zone where no variant calling is possible.

Accordingly, according to feature 1 of the present invention, there is an effect that the reproducibility and accuracy of the algorithm for versatility can be maintained.

On the other hand, feature 2 of the present invention is to determine the variable by performing a machine learning using the data provided from the standard parameter (NA12878) the constant parameter of the score function.

3 is a conceptual diagram illustrating an algorithm filtering and half-type based error correction concept of a standard material-based dielectric variation correction according to the present invention.

As shown in FIG. 3, feature 2 of the present invention additionally corrects errors in variant calling of the dielectric, compared to standard (NA12878).

Human genomes have a number of repeating sequences in the process of evolution, and the repeated bases are scattered and aligned in several different genes.

In this case, the base fragments aligned in the reference genome are called a binary alignment map (BAM), and if a variant calling is performed based on the BAM, all of them are subjected to false positives.

Therefore, by filtering and correcting genes having overlapping base sequences in this way, it is possible to more precisely define variant calling of the human genome.

At this time, the reference material (NA12878) refers to the genome information and specimen information that are used most often by researchers around the world for comparison, and edited by Justin (Justin et. Al., Integrating human) sequence data sets provides a resource of benchmark SNP and indel genotype calls, Nature Biotechnology 32, 246251, 2014).

Here, the standard material can be used as a standard by making a cell line that permanently survives a cell line sample of one person, and detects all mutations present in one person through various platforms. It is a sample to share.

The NA12878 generates approximately 3.5 million correct answers using a variety of next-generation sequencing platforms (more than four).

On the other hand, the step before applying the tool for detecting the variation is called a binary alignment map (BAM). Given this BAM file, we use the ADISCAN mutation detection tool to detect mutations. By comparing the variation with the correct answer, we can select the wrong answer.

In this case, the reason for the error is that the mutation detection tool itself is not wrong, but the BAM file is made wrong. Thus, 3.5 million correct answers of NA12878 can be used to correct errors in BAM files.

Specifically, error correction of the BAM file is performed by the following three steps.

First, using the correct answer variation of NA12878, the calculation is performed using the most standard genome analysis pipeline. There are the most standard genome analysis pipeline stages 1-8, and the variation analysis according to the present invention is also performed by the following eight stages.

1) Trimming, 2) Mapping, 3) Sort BAMs, 4) Merge BAMs, 5) Remove Duplication, 6) Realign InDEL, 7) Recalibrate Scores, 8) Variant Calling

Using this pipeline, a standard analysis tool (BWA-MEM) is used to generate a binary alignment map (BAM), which is the basic data for detecting new mutations.

Second, we use ADISCAN, a mutation detection tool, to detect mutations, and calculate cumulative sums of genetic mutations that are incorrectly calculated by mutations.

Genes with one or more cumulative sums are selected and the cause is analyzed for one or more genes. That is, the cause is checked whether there is a duplication of the reference itself in which the base reads are duplicated, or whether there is a duplication in a specific, exon or intron.

Third, haplotypes are performed using a 1000 genome project (5008 haplotypes) for haplotyping on exon, intron, utr or intergenic regions.

Meanwhile, FIG. 4 illustrates a basic concept of haplotyping applied to a personal genomic map based customized medical analysis platform and an analysis method using the same according to the present invention.

The basic concept of the haplotyping method is disclosed in Patent Application No. 10-2016-0096996, which is summarized again, as shown in FIG. 4, (A) a base sequence fragment (fastq) ( B) align to human standard dielectric (GRCH38), re-extract the aligned sequence (fastq), and (C) several different types of haplotype databases (IMGT / HLA: HLA genes specially engineered specifically for instrumentation) Sort on the database containing the haflow type). And (D) successively deciding the ranks by removing false positives, and (E) finally separating the two haplotypes of one dad / mom. And (E) when the two strands of the haplotype resulting from the step (D) comes out, it is carried out through the process of combining and making a variation (genotype) to report the variation.

In this case, the haplotype according to the present invention uses 5008 Haplotype databases of 1000 genome projects (haplotype) instead of the IMGT / HLA database.

On the other hand, the third feature of the present invention is a method of calculating the variation of the full-length dielectric and a method of correcting the error (error) caused by sequencing (decoding) sequence fragments (read) aligned in the overlapping region In other words, it is a new efficient way to apply the haplotyping method for correcting the result of feature 2 of the present invention.

Haplotyping applied to the present invention includes 26 race-based HaplotypeDBs and 1,000 other established Haplotype DBs generated from the 1,000 Genome Project. Therefore, it is possible to perform haplo typing and variant calling by using the haplo typing method. The basic concept of the haplotyping will be described again.

At this time, the base fragments (read) arranged in the reference genome are extracted (fastq), and the extracted base fragments are aligned to the above-mentioned 1,000 genome Haplotype DB through an embambler (bwa-mem, etc.). In addition, two alleles are selected using the haplotyping method, and the two alleles A and B shown in FIG. 3 are combined to perform variant calling based on the combined result.

Hereinafter, features 4 and 5 of the present invention related to patient stratification will be described.

5 is a conceptual diagram summarizing the history and concept of the genome technology for patient stratification according to the present invention.

As shown, patient stratification has been actively progressed in SNP chip-based disease-related technologies since 2002 when the human genome was completed, and in particular, the international consortium of GWAS-based marker discovery was active, followed by the eQTL-related international consortium. Arose

As a result, 40,000 GWAS_Catalog GWAS markers and 6 million markers of eQTL were published, and the US NCBI began to database databases for Clinvar (Clinical variant) for rare, cancer and general diseases. Currently, about 100,000 rare disease markers have been published.

In addition, around 2007, sequencing equipment based on Illumina and Proton (PGM-ION) NGS-based commercialization was introduced, and new mutant or cancer-mutant-based diagnostic clinical genomes appeared, and in each of the International Oncology Consortium and the 1,000 Genome Project, 2,100 pairs of cancer genomes of cancer patients and 2600 normal genomes have been released. And, for the first time in the world, Syntheca Bio performed a large matrix of two big data sets of 3 billion bases * 6,800 electric field dielectrics. This is called the integrated dielectric DB.

As shown in FIG. 5, when comparing the distances between the markers of the electric field dielectric and the SNP chip data, each marker of the SNP chip data is present per 12,000 of the electric field dielectric. That is, 12,000 / 2 * 500,000 markers ˜ = 3 billion bases. Therefore, in order to link the functional information to the GWAS marker found in the SNP chip, the eQTL marker, the BAV (bioactive variant) and the Clinvar marker with the functional information are based on the fact that the two locus are inherited by each other by the linkage disequilibrium (LD). You need to try to connect. In addition, the penetrance information of mutations is a measure of linking diseases that are very important in clinical practice, and the role of rare but low penetrance information in the penetrance information is little known. And this information is judged to affect the rare mutation or the activity of specific protein in individual stratification.

6 is an exemplary diagram showing an example of utilizing a common GWAS marker and bio-map for the stratification of patients in the integrated dielectric DB according to the present invention, Figure 14 is an allele depth of the integrated dielectric DB according to the present invention, geno A conceptual diagram showing the configuration of the type and haplotype DB, Figure 16 is a conceptual diagram showing the stratification of patients through the genetic and multigene-based genotype calculation in the integrated genome DB according to the present invention.

As shown in FIG. 6, feature 4 of the personal genomic map-based customized medical analysis platform and an analysis method using the same according to the present invention performs patient stratification using a common marker based on association imbalance.

That is, as shown in FIGS. 14 and 16, the patient stratification according to the present invention may be performed by using a single nucleotide polymorphism (SNP) and a rare variant (rare variant) marker between a control and an individual genome in a gene unit and multiple genes. Patient stratification is performed by detecting and calculating genotypes based on these.

The basic concept of such a patient stratification is as disclosed in Korean Patent Application Nos. 10-2015-0187554, 10-2015-0187556, and 10-2015-0187559.

However, this method of patient stratification (genotyping by gene haplotype-based gene units and multigene clustering in genes and multigene units) is highly useful when there is sufficient population information. If international data is difficult to obtain, it will be less efficient.

In other words, when the gene-based calculation of common GWAS marker-based patient stratification, which is known as the international consortium, is improved, the patient stratification of the common GWAS marker-based gene unit method is required.

However, since most common GWAS markers have a sparse distribution over a wide range of mutations, the markers tend not to explain the function of a particular disease.

Therefore, the present invention uses the index R ^ 2 according to the linkage disequilibrium (LD) to connect the non-functional variation with the functional variation based on the reference value (0.7) based on the reference value (R ^ 2> 0.7). Perform marker-based genetic stratification of patients.

In this case, the associated unbalance index may be calculated by the following arithmetic expression.

In this case, r ^ 2, that is, the linkage disequilibrium (D ~ = LD) index calculation calculates the strength of the relationship between two single-base polymorphisms.

The LD index is calculated using the difference between the haplotype frequency calculated from the alleles observed in each monobasic polymorphism and the randomly generated haplotype frequency.

D 'is the most widely used LD index. Currently, | D' | > 0.8, we believe that there is a strong association between the two monobasic polymorphisms.

In addition, if r ^ 2 is greater than 0.7, it is determined that there is a strong correlation between two single-base polymorphisms.

Where Pa is the A allele at the first locus, Pa is the a allele at the first locus, PB is the B allele at the second locus, and Pb is the frequency of the haplotype with the b allele at the second locus.

Meanwhile, FIG. 7 illustrates a GWAS marker-based patient stratification generation method according to the present invention.

As shown here, in the present invention, the GWAS marker genotypes in the patient stratification have major and minor alleles, and among them, the LOF markers and eQTL markers in which the GWAS markers and LDs are formed should be the same allele.

And the GWAS marker does not have the functional information, and the marker connected to the LD always needs the functional information.

Therefore, the patient stratification generation rule can be defined, and the relationship between genetic variation and expression can show a simple relationship diagram of GWAS marker, eQTL, LOF, and expressed gene.

Figure 8 shows an example of the generation of GWAS marker-based patient stratification according to the present invention.

As shown in FIG. 8, the GWAS marker (brown) is selected, the case where the eQTL marker (red) is the minor allele while having the minor allele of the GWAS marker is selected, and then the GWAS marker (brown) Select to have the minor allele of and the LOF marker (green).

And disease stratification-1 to disease stratification-N, and classify non-disease stratification groups.

In addition, if a patient responding to the drug when the actual drug is added later to the specific disease stratification-i, the drug may be related to the specific patient stratification marker.

It is also possible to identify mechanisms in the molecular biology domain between patient stratification markers and functional correlation information. Here, the eQTL marker can know the mRNA coding gene to be expressed, and the LOF (loss of function) directly describes the function of the GWAS marker.

Meanwhile, as described above, common GWAS markers, eQTLs and gene variants link functional information based on LD such as R ^ 2> 0.7, but eQTL and gene variants cannot provide functional information. Can be.

This is because the GWAS markers, one per 12,000 bp, are still created with too sparse distributions, which still prevents them from linking with functionality.

In order to solve this problem, the present invention is a biomap (bmap, cmap and vmap, patent registration No. 10-0880531, No. 10-0996443, No. 10-1035959, No. 10-1117603 and No. 10-1400717) Biodata indexing (rvr), biomap homologs based on homologous constructs (bmap), maps to maps (cmap), and protein-drug-mutant complexes (vmap) A technique (feature 5) is applied to explain the functionality of a common GWAS marker that is secured and made with a sparse distribution.

That is, according to the present invention, the method of using biomap-based bioinformation is connected to the eQTL marker by linkage disequilibrium (LD) as described above in the integrated genome DB of the personalized medicine platform. If the gene expressed by the disease mechanism-related functionality is not explained, as shown in feature 5, the biomap (bmap), mapxmap (cmap) and complex (mutant-drug-protein, vmap) information through the information We will discuss these features and quickly network and discover them based on RVR technology.

Here, the concepts of rvr, bmap, cmap, and vmap are disclosed in Patent Registration Nos. 10-0880531, 10-0996443, 10-1035959, 10-1117603, and 10-1400717. To sum it up briefly:

First, records virtual rack (rvr) is a file creation method for single data retrieval and a single data file retrieval method, and uses a rat (record allocation table) file for single file retrieval. rvr is a modification of the inverted indexing technique used in computer science to apply to biodata.

The point is that the characteristics of a large amount of bio data is bulk data or big data, so when a file is recorded that records address information of a file externally (rat: record allocation table), it is easy to handle a large amount of bio data including batch processing. Become.

This method can be used to process numerous types of bio-related big data such as sequencing, protein sequences, multiple alignment files, text information, protein 3D information, mutation information, image information, audio information, etc. Can be.

Secondly, a bmap (bio map) is a method for searching biomedical integrated information based on a cluster and a backbone database, and has been disclosed in Korean Patent Registration No. 10-1035959.

The bmap collects all the information related to the gene from the public DB based on homology based on homology and indexing based on rvr. Says how to search.

The contents of public data can generate gene map results based on Gene expression, Pathway, Disease, Nucleotide, Protein, Regulation, Interaction, Metabolite, SNP & mRNA SNPChip data, Drug and Protein Chemical, and Interaction related databases.

In the same way, you can also map drugs, mutations, interactions, and FDA drug maps.

Thirdly, cmap (cell map) is a system for providing biomedical function related information through the generation of multiple maps that can be interconnected, and is disclosed in Korean Patent Registration No. 10-1117603.

The cmap relates to a method of linking a keyword or ID used in a biomap to a biomap and a method of linking a biomap and a biomap.

In addition, the characteristics of the cmap may be classified into the following three types.

1) The query may be a gene ID, a compound ID, a disease ID, an important keyword, or a combination thereof.

Homology (similarity) Query means all contents that give homology (similarity) to Query, and it is used as lower case query.

A database is a group of records, a record is a unit of an object, and an object is each article, bio primitive information (e.g., base sequence, amino acid sequence, atomic coordinates, compounds, etc.). Coordinates, chemical symbols, amino acids, etc.).

2) The result of searching database DB with query Q is expressed as Q x DB, and the query and homology query collection is expressed as (q1, q2, .. qN). Therefore, a set of data (q1.db1, ... qN.db1, ... q1.dbN ... qN.dbN ') retrieved with Q and homology q is defined as a map.

3) Assorted map / Assorted map can be expressed in the same way as used in map M1 and map M2. That is, a comparison between a single map and a collection of maps that collect genes associated with a disease-related gene pool (same gene functional collection, a collection of similar functional drugs, a disease resistant gene collection, or a genome, etc.) is expressed as M1 x M2.

Fourth, the vmap (virtual map) relates to a simulation method of the whole atom-based polymer composite, and has been disclosed in Korean Patent Registration No. 10-1400717.

The vmap can calculate the difference in free energy by simulating a complex of protein-drug-variation, calculate the resistance of the drug to the target due to the variation, and when repeatedly performing the protein and drug, specific atoms / Fixing amino acids and giving them a lot of stress has the effect of amplifying weighting concepts on energy, making energy-based detection easier.

In particular, in the case of a vaccine, the method prepares various types of MHC (immunoreceptor protein) structures as templates, and then replaces amino acid sequences obtained by pattern analysis of various gene sequences related to immune function with known peptide binding sites in the MHC. By fixing the peptides and then conducting a simulation, a comparative analysis of the energy changes of each peptide can be used to predict peptides that bind stably with MHC through GPGPU-based screening.

Hereinafter, the technical features of the present invention related to the integrated dielectric system according to the present invention will be described.

9 is a conceptual diagram showing the divided chunk of the human genome 3 billion bases for operation in the PC-cluster applied to the present invention, Figure 14 is an allele depth, genotype and bottom of the integrated dielectric DB according to the present invention Conceptual diagram showing the configuration of the flow type DB.

As shown in FIG. 9, feature 6 of the present invention relates to full-length dielectric analysis and integrated dielectric DB, the basic configuration of which is disclosed in Korean Patent Registration No. 10-1460520 and Korean Patent Application No. 10-2015-0187554 There is a bar.

In reality, however, in a PC-cluster (supercomputing environment using 100 to 1,000 connected servers), PC-cluster can be processed at a reasonable cost while processing and integrating a large number of human genome data simultaneously based on a large number of CPUs. In order to configure, it is desirable to define the processing capacity per node (multiple CPU).

Accordingly, the present invention preferably allocates approximately 64GB of memory according to human genome analysis.

Therefore, the present invention is to integrate the full-length dielectric (3 billion base pairs * N people) in a pc-cluster environment, the full-length dielectric to a constant size (about 50,000,000 base pairs * N people = 1 chunk) By applying a technology (part 6) to divide and create 600 chunks, it improves the efficiency of PC-cluster based analysis.

This is because the economics of this set-up analysis platform are significantly increased because about 10-20 GB of memory is required to make 10,000-20,000 integrated samples per chunk into a multi-aligned environment.

On the other hand, Figure 10 is a conceptual diagram showing the configuration of the calculation equipment (a1-a6), storage (b1-b3) and Mach-FsDx for analyzing the genome data according to the present invention.

As shown in FIG. 10, feature 7 of the present invention is an optimization of a supercomputer environment for operating dielectric bulk data, testing and deploying hardware and network card speeds in a pipeline according to data I / O rates, and simultaneously hundreds and thousands It is related to the configuration of middleware-class parallel distributed storage environment (eg cloud computing) that can be used when calculating two dogs.

In this case, to construct a supercomputer for genome analysis, a memory-based ultrafast computational flow (a1) (but relatively small I / O), a large amount of data after ultrafast computation (additional I / O performance + additional ultrafast memory-cache-server + If you need large storage (a2), flow for a large amount of data (additional I / O performance + additional ultra-fast memory-cache-server + large storage) after the ultra-fast calculation of the GPGPU performing ultra-fast calculations (a3), In the thousands of general computing servers, a flow (a4) for performing thousands of calculations simultaneously, a flow for a medium speed (a5), and a data backup (a6) are required.

In addition, from the storage point of view, an additional high-speed CPU and storage for providing I / O performance, such as a memory disk driver (b1), parallel distributed ultra-fast mass I / O storage (b2), and general storage (b3), must be provided.

On the other hand, feature 8 of the present invention means that to utilize such large equipment (mach-fsdx) in the clinical, reproducibility and consistency must be ensured, and accordingly, the definition of the system limits is clearly defined. Therefore, the work of the clinical environment for Mach-FsDx (diagnostic equipment and medical devices) of the current Mach-Fs is needed.

The following shows examples of evaluation contents.

11A to 11F show the results obtained by evaluating Mach-Fs to Mach-FsDx according to the present invention for stability, limit and reproducibility.

Here, the verification test environment,

1.MAHA-FsDx Master Node: MDS (CPU L5520 4Core x2ea / 64GB Memory / 128GB Disk)

2.MAHA-FsDx Data Node: DS1 ~ 5 (CPU E5630 2.53GHz x1EA / 24GB Memory / 2TB x5ea)

3.MAHA-FsDx Capacity: Physical 50TB / usable 25TB (MAHA Replica 2 option (stability))

Compute Node: CPU Xeon E5-2630 v2 2.60GHz / 64GB Memory / RAID 11TB

5. Sample Data: CHA-NS161103-0001_R1.fastq.gz CHA-NS161103-0001_R2.fastq.gz Genome Target Sequence (TS) Sample (420MB x2ea);

The test method is

1.Reproducibility test: Test the same on Local Disk / awork01 and / maha4

2. Stability test: Cause failure of one of five MAHA DS nodes.

3. Limit test: Test with 2Jobs per compute node.

 A. no caching: 30/40/60 jobs results

 B. caching: 60/200/240/360 jobs results

 C. no caching large capacity 1.4PB: 392/582/776/784 Jobs results (additional)

4. Used genome analysis pipeline and raw data

As shown here, it can be seen that Mach-FsDx to which the features 7 and 8 according to the present invention are applied is superior in reproducibility, stability, and limitation as compared to Mach-Fs.

12A and 12B are examples of verifying the maximum performance of the MAHA-FsDx configuration scale with respect to the limits.

At this time, 5 units were configured in a general server, each server was composed of MAHA-FsDx with 5 2TB HDDs, 5 additional memory caches (SDRAM chching) were configured in a general server, and 10 general high capacity servers were configured. It consists of MAHA-FsDx with 24 6TB HDDs.

From these experimental results, it can be seen that the limit of MAHA-FsDx is significantly improved.

FIG. 13 is a conceptual diagram illustrating a configuration of an integrated dielectric DB platform-based multi-investigator database according to the present invention, and FIG. 15 is a conceptual diagram illustrating a detection principle of SNV, INDEL, and CNV in each chunk unit matrix according to the present invention.

As shown in FIG. 13, the ninth aspect of the present invention is to integrate and analyze data and tools that have been pre-calculated in a form that can be accessed by a large number of research developers in a large system such as a personal genomic map-based customized medicine platform. It provides a standardized database (mySQL and mongoDB) based user environment on middleware.

The present invention calculates and extracts the result data of the preceding data required by multiple researchers using various tools and common tools provided by the integrated dielectric DB. In particular, the functional information of SNV, INDEL and CNV, especially LOF (loss of function), is calculated using polyphen and SNP effecter, and GWAS marker, LD calculation information, eQTL, MAF, phenotype, etc. Equipped with statistical analysis tools, gene-based patient stratification and toxicity marker calculations are performed.

FIG. 14 shows an allele, genotype, and haplotype DB in which an integrated dielectric DB is divided into nucleotide sequences of 5,000 genomes of 3 billion nucleotides in chunks (50 million) and formed in 5,000 matrices. It is an example.

FIG. 15 is a schematic diagram of extracting data (SNV, INDEL, and CNV) that have been precomputed in a form accessible to the research developer according to feature 9 of the present invention, in each chunk unit.

In addition, FIG. 16 shows patient stratification through gene and multigene-based genotyping in the integrated genome DB according to the present invention.

The rights of the present invention are not limited to the embodiments described above, but are defined by the claims, and those skilled in the art can make various modifications and adaptations within the scope of the claims. It is self-evident.

Hereinafter, a brief summary of some of the underlying technologies for the implementation of the present invention will be described.

Technical Summary of Korean Patent Application No. 10-2015-0187554 (Patent Document 8), Korean Patent Application No. 10-2015-0187556 (Patent Document 9) and Korean Patent Application No. 10-2015-0187559 (Patent Document 10)

Patent document 8, 9 and 10 disease cause excavation system is the analysis data input unit 100, the search control unit 200, the result report providing unit 300, HaploScan DB (400), ADISCAN DB (500), IDA DB ( 600), the bioactive DB 700 and the reference DB (800).

The analysis data input unit 100 is a part for receiving personal genomic information and receives DNA sequencing data.

The search control unit 200 detects genotypes, genotypes, rare mutations, disease mutations, and physiologically active variants of each gene from the input DNA sequencing. For this purpose, the search control unit 200 is a HaploScan engine ( 210, an ADISCAN engine 220, an IDA search engine 230, and a bioactive mutation search engine 240 are configured.

The HaploScan engine 210 compares the analysis data (input DNA sequency) with the Haplo MAPs 414 and 424 stored in the HaploScan DB 400 to be described later.

The structure of the HaploScan DB 400 and the search method of the HaploScan engine 210 will be described in detail later.

In addition, the ADISCAN engine 220 prepares each base included in the input analysis data by using the ADISCAN DB 500 and the ADISCAN method, and calculates the rarity of the group control group.

In addition, the IDA search engine 230 detects a known disease-related disease variation, and detects a disease variation by comparing the analysis data with the IDA DB 600 stored the known disease variation.

In addition, the physiologically active mutation search engine 240 detects genetic metabolism-related genetic variation, and largely determines whether the genetic variation is related to amino acids involved in protein-drug, protein-DNA, and protein-protein binding.

At this time, the physiologically active search engine 240 compares the analysis data with the BAV DB (700) to determine whether the mutation of the base corresponding to the amino acid associated with the protein binding stored in the BAV DB (700) of the analysis data .

On the other hand, the search control unit 200 is a Manhattan so that the diagnostic person (or user) can easily determine the genotype and significance (rare) of each base determined by the HaploScan engine 210 and ADISCAN engine 220 easily Plot and radial variation significance charts are used to generate result reports.

The generated result report is provided to the user through the result report providing unit 300.

Hereinafter, the database structure of the excavation system that causes the disease according to Patent Documents 8, 9 and 10 will be described.

The disease cause excavation system according to Patent Documents 8, 9 and 10 is largely comprised of HaploScan DB (400), ADISCAN DB (500), IDA DB (600), BAV DB (700) and Reference DB (800).

As shown in FIG. 3, the HaploScan DB 400 is a database in which genotypes of control genes are arranged to calculate genotypes from personal genomic information to be analyzed. The HaploScan DB 400 is shown in FIG. 2. And a single gene information database 410 and a multiple gene information database 420.

The single gene information database 410 is a database storing genotypes for a single gene, and includes a single gene Haplo map 414 and a single gene Haplo preconciliation information 412.

On the other hand, as shown in Figure 6, the single gene Haplo map 414 is stored by dividing (distributed) the variation distribution by the occupancy rate for the same gene of the entire control, 26 world using each gene The haplotype calculation of races, the frequency of specific traits, and the frequency of each sub-racial, are summarized.

The single gene haplo preconciliation information 412 stores information about each variation. In this case, the single gene Haplo preconciliation information 412 may be data that directly stores the variation information, or may be configured as an identification factor indicating a location of information stored in the Reperence DB 800 to be described later. That is, the single gene Haplo preconciliation information 412 provides frequency and various disease-related annotation information in 39,000 genes of humans and in each gene in 5,000 people of the world.

In addition, the multi-gene information database 420 is a database for providing variation distribution and information for the multi gene, and includes a multi-gene Haplo map 424 and a multi-gene Haplo preconciliation information 422. do.

In this case, the multigene Haplo map 424 is a genetic pattern in which the phenotype is specified by the multigene, and stores the distribution of variation for the relevant bases of the entire control group by the occupancy ratio for each phenotype, phenotype (phenotype) The haplotype calculations and the frequency of specific traits and the frequency of each sub-racial for 26 races using the causal variation of the world were calculated.

The multi-gene haplo preconciliation information 422 stores information about each variation. In this case, the multi-gene Haplo preconciliation information 422 may also be data that directly stores mutation information, or may be configured as an identification factor indicating a location of information stored in the Reperence DB 800 to be described later.

That is, the multigene haplo preconciliation information 422 provides various disease-related annotation information and the frequency of phenotype-related gene sets in 39,000 genes of humans and 5,000 global races.

The X axis of the HaploScan DB 400 is 3 billion nucleotide sequences, and there are 39,000 genes in the nucleotide sequences. If N (varies) were found in a specific gene (i) in its schema, the mutations could be clustered using both haplotype and genotype in Y axis: 5,000, and the clustered form would be HaploMap.

In this case, each cluster means each genotype. In the contents of the first GP * 47 * 0, the genotype occupies 47% of the global population, 0 bit different from the global average, and second. Genotype GP * 25 * 1 means 25% of the world's population, which means that it is 1 bit different from the world's average.

Multigene-based HaploMaps are also classified and classified in the same manner.

The ADISCAN DB 500 is a DB storing genome information of a control group, and specifically, the genome may use genome information known by performing a global genome project.

On the other hand, the ADISCAN DB 500 stores the full-length genome information of the control group, it may be divided and stored according to the classification criteria forming a genotype group, such as race.

In this case, the racial classification may be a division of five major categories, or may be a division of 26 subclasses, for the purpose of discriminating / detecting whether or not the mutant gene is reflected by the genetic characteristics of each race.

In addition, the IDA DB 600 is a place where known diseases and genetic variations related thereto are stored. Genetic variation information related to each disease and various literature information supporting these variations are organized and stored for various diseases.

In addition, the BAV DB 700 stores genetic information for determining the amino acid form of the binding position of various proteins.

Specifically, in the binding between protein-drug, protein-DNA and protein-protein, amino acids that affect these bindings and genetic information that affects those amino acids are stored.

Accordingly, when a large number of mutations occur in the bases for the amino acids that govern the binding of a particular metabolite, the examinee of the analytical data increases the possibility that normal in vivo processing of the metabolite becomes difficult.

That is, the BAV DB 700 stores the predicted protein binding position, the promoter position and the protein activity of the binding state, including known disease mutations.

The BAV DB 700 is a database storing bioactivity-related gene information, and stores information on resistance and sensitivity to genes, drugs, metabolites, and foods. In this case, the BAV DB 700 may also be established by linking well-known data secured by the public trust, for example, about 6,000 drug information (interacting protein and binding region information, etc.) known to the drug bank, Information on more than 12,000 metabolite information (such as interaction protein and binding region information) known to the Metabolite Bank and about the location of drug metabolic mutations in over 200 genes in the drug metabolizing enzyme and transporter gene (DMET) have.

On the other hand, the reference DB 800 is a DB that stores information about the variation of the known genome, it can be built in connection with the public information database as well as document information.

For example, PheWAS-GWAS (Genome wide association study) data and eMERGE (Electronic Medical Records and Genomics) data may be applied to the reference DB.

Although not shown, the search control unit 200 further stores a clinical information DB in which the environmental predisposition information of the subjects to be examined is to be considered together with the genetic characteristics in order to derive the prediction result of the disease cause based on the clinical information. It may be configured to include.

In this case, the clinical information DB stores the individual environmental factor result data, the group average, and the reference information.

The individual environmental factor result data may be clinical information data such as an individual's comprehensive examination data, and the group mean and reference information may use a community cohort study result provided by the Center for Disease Control.

Hereinafter, a method of analyzing genetic information using individual full-length genomes according to Patent Documents 8, 9, and 10 will be described in detail.

First, the genetic information analysis method using the individual full-length genomes according to Patent Documents 8, 9, and 10 begins with receiving the analysis data (DNA Sequencing) to be analyzed, the analysis data input unit (S100).

In this case, the analysis data may be provided in the form of Dumy consisting of DNA fragments. In this case, as shown in FIG. 15, the present invention generates and stores DNA sequencing in the form of an RVR file through highly integrated indexing on the provided Dumy data. do.

Thereafter, the genetic information analysis method using the individual full-length genome according to the present invention performs four types of analysis according to the analysis target.

That is, the genetic information analysis using the individual full-length genomes according to Patent Documents 8, 9 and 10 is 1) genotype determination (S200), 2) rare mutation calculation (S300), 3) disease mutation calculation (S400) and bioactive mutation calculation Four analyzes of S500 will be performed. Hereinafter, each of them will be described in detail.

[Genetic discrimination]

The HaploScan engine 210 compares the DNA sequency with the Haplo Frequency 412 and the MAP 414 stored in the HaploScan DB 400 and detects a population belonging to the genotype and information about the single gene and phenotype.

Specifically, the HaploScan engine 210 compares the i-th gene of the single gene Haplo Frequency 412 with respect to the i-th gene of the single gene Haplo Frequency 412 with respect to the i-th gene of the DNA sequencying (S211). It is determined in which cluster among the single gene classifications classified in the MAP 414 (S213, S215).

Thereafter, the HaploScan engine 210 repeats i = 1 to the end (about i = 39,000) to determine genotypes for all genes in the analysis data (S217 and S219).

In addition, the HaploScan engine 210 compares the DNA sequencying with the multigene Haplo Frequency 422 (S221), and a combination of multiple genes of the genome to be analyzed for each phenotype is classified into the multigene Haplo MAP 424. It is determined in which cluster among the classification of the multiple gene combinations included (S223, S225).

In this case, the HaploScan engine 210 repeatedly determines all genotypes of the analysis data for all phenotypes stored in the multigene information database 420 (S227 and S229).

Through this HaploScaning process, it is possible to define genotypes according to single gene mutations and multiple gene mutations included in the target genome.

[Rare mutation calculation]

Rare mutations are base mutations caused by very unusual specific genetic mutations, and are generally related to rare diseases. By detecting the presence or difference of specific bases, the possibility of occurrence of rare diseases can be determined.

To this end, the present invention first, as shown in Figure 14, the ADISCAN engine 220 selects the control (S310).

In this case, the control group is a control group that will determine the rareness of the variation, and may limit a specific race or a specific country.

Thereafter, the ADISCAN engine 200 calculates the variance index using the base of the control DB and the ADISCAN method for the base of a specific locus, and performs the same process for the whole genome (n = 1 to n = about 3 billion). (S320, S330, S340).

This calculates the rareness of the bases for the entire base sequence (S350).

On the other hand, ADISCAN (allelic depth and imbalance scanning) for calculating the rare mutation is a technique for screening markers that give a difference between normal and abnormal genes, allelic depth multiplicant difference, allele squared difference, allelic absolute value difference, It is determined by geometric allele differences, statistical allele differences, or allele imbalance rates.

[Calculation of Disease]

In the disease mutation detection, the IDA search engine 230 compares the analysis data with the mutation information of the IDA DB 600 to determine the risk of the disease (S410).

In this manner, after reviewing the analysis data for all diseases included in the IDA DB (S420), significant mutation-related diseases are calculated (S430).

[Calculation of physiological activity variation]

The physiologically active mutation detection is a bioactive mutant search engine 240 searches for BAV DB (physiologically active mutant DB) (S510), and detects information on the amino acids involved in the binding of the protein (S520).

In this case, the protein binding includes protein-drug, protein-DNA and protein-protein binding, and the amino acid information includes base information related to the amino acid.

Thereafter, the physiologically active mutation search engine 240 detects amino acids and metabolite information related to the mutations generated in the analysis data by comparing the base and the analysis data included in the amino acid information (S530, S540).

The physiologically active mutation search engine 240 repeatedly performs mutation detection for all amino acids and integrates the detected information to calculate physiologically active mutation information (S550 and S560).

* Then, the search control unit 200 integrates the determined or calculated genotype, rare mutation, disease variation and physiological activity variation, and generates a result report to be provided to the user (S600).

In this case, the search control unit 200 may calculate and provide a clinical information-based disease cause based on the clinical information of the examinee.

Specifically, predicting the cause of a disease requires personal health records (PHR) that include the results of current environmental factors (general examination data and clinical information). In particular, environmental factors require average and baseline information of the population (in the present invention, the baseline information utilizes the results of the second community cohort study provided by the Center for Disease Control and Prevention). Here, the PHR-trait is associated with the result and genotype of these environmental factors.

The disease cause relationship (Π) detection equation uses a logistic regression analysis method, and the variable χ is a value determined according to the genotype, rare mutation, disease variation, and bioactive variation calculated as described above, and the variable β is The value is determined from the PHR.

In other words, the disease cause relationship is Gene, Disease or Drug genotype (group or cluster of genotypes) vs .. The correlation of PHR (BMI, AGE, SEX, etc.) can be calculated.

Therefore, the gene-based disease cause is calculated by calculating the association between the current clinical condition (normal, disease, or phenotype) and the gene, disease, and drug genotype calculated from the 39,000 gene.

On the other hand, the excavation system of the disease causes according to Patent Documents 8, 9 and 10 generates the reporting data from the calculated gene mutation information.

 The resulting report, although slightly different depending on the output, basically uses a Manhattan plot and a radial variation chart to visualize the mutation gene.

The Manhattan plot is a graph that visualizes accumulated values as points by classifying the standard genes of the Genome Project according to genotypes based on non-symmetric variations of all known SNPs for 39,000 genes. do.

If the gene of the genome to be analyzed is displayed, mutation specificity of the gene to be analyzed can be easily recognized compared to the control.

This Manhattan plot can easily identify the mutation locus, as well as the degree of variation.

Meanwhile, the significant variations indicated by the Manhattan plot may be displayed in a radial variation chart according to the degree of variation and genetic characteristics.

In this case, by displaying the degree of genetic variation of the analysis target genome and the control average together, the degree of variation of the genome to be analyzed can be displayed clearly and clearly, and additionally including genetic characteristic information to generate a result report. It may be.

The result report generated by the above-described method is provided through a result report providing unit.

Technical summary of Korean Patent Application No. 10-2016-0096996 (Patent Document 11)

The haplotyping system and method according to Patent Document 11 may be applied to haplotyping on a human full-length genome, or may be applied to SNPs in a specific region.

Herein, the specific region refers to a region of a gene (or a combination of genes) related to performing a specific function, and typically, a region of human leukocyte antigen gene (HLA gene), which is responsible for the regulation of human immune system, and drug metabolism. It may be a gene responsible for the function (DMET gene region), a gene region related to immune cell expression (KIR gene) region, and a gene region related to blood characteristics (ABO gene).

Therefore, Patent Document 11 can be applied not only to haplo typing on a human full-length dielectric but also to haplo typing on a specific region such as HLA typing, DMET typing, KIR typing, and ABO typing.

Here, DMET (Drug Metabolizing Enzymes and Transporters) refers to protein enzymes (transporters) and enzymes involved in the absorption (absorption) and disposition of drugs, drug action.

For example, the cytochrome p450 enzyme family (CYPs), uptake transporters, and efflux transporters belong to this group. There are several genes in a family, and their gene sequences are similar to each other but have polymorphism.

DMET gene sequence differences between individuals not only affect drug reactions, side effects, disease sensitivity, etc., but also can be a criterion for selection of appropriate drugs, which is a research area that has recently attracted attention in pharmacogenetics.

KIR (Killer-cell Immunoglobulin-like Receptors) is a protein expressed on the surface of certain immune cells such as natural killer (NK) cells or T cells.

KIR regulates the ability of NK cells and T cells to kill cells by interacting with major histocompatibililty (MHC) class I on the surface of other cells.

Thus, this function of KIR is associated with sensitivity and responsiveness to infections, autoimmune diseases, cancer and the like.

And the KIR is very polymorphic (polymorphic) gene sequence is largely different from person to person, the amount or type of genes that each person has different.

On the other hand, ABO (blood type) is a gene that plays a major role in distinguishing ABO blood type and transfusion system. It is located in chromosome 9q34 and three alleles (A, B, O types) by conventional serum technique. Can be distinguished.

Subgroups exist in alleles of A, B, and O, and even in rare cases, the same blood type may cause a problem in which transfusion is impossible between subgroups.

In the following, in order to ensure the specificity of the description, HLA typing will be described as a representative embodiment.

The major histocompatibility complex regions are one of the most complex regions of the human genome and are responsible for the regulation of the immune system. Among them, the human leukocyte antigens (HLAs) gene are present in about 3Mbp stretch of chromosome 6 and play a big role in adaptive immune response, which suppresses and eliminates pathogens. In charge of.

From a clinical point of view, the risk of rejection can be reduced if the HLA gene between donor and recipient is similar during organ transplantation. Therefore accurate HLA typing is a very important issue.

However, accurate HLA typing is very difficult because of the highly polymorphic, linkage disequilibrium and sequence similarity between HLA genes.

For example, there are thousands of alleles reported in the IMGT / HLA database for exons 2-4 of HLA-A gene in class I. HLA-A, B and C genes Alleles of the liver are very similar.

Because of the low resolution (2-digits), even if the same antigenic peptide (antigen peptide) can cause an allogeneic response due to the difference in amino acid (amino acid level) high resolution of the amino acid level (4- HLA typing is required up to the digits.

Conventional methods of high resolution HLA typing include polymerase chain reaction by sequence specific oligonucleotide (SSO) and sequence-based typing (SBT) methods. Depending on the labor force of the workforce, low-throughput and high costs are a problem.

Targeted amplicon sequencing (TAS) approach shows relatively high throughput (high-throughput) compared to the PCR method, so that HLA typing with high accuracy is possible by generating long reads of several hundred bases.

However, due to efficiency and cost, most of the data generated recently is genome-wide sequence, whole genome sequence (WGS) or whole exome sequence (WES), and such data has short sequence reads (~ 101bp) rather than long reads. . Therefore, there is a need for HLA typing with the same accuracy and economy as the TAS approach using such short sequence reads.

That is, the HLA typing according to Patent Document 11 is to provide HLA typing that is secured with accuracy and efficiency using short reads.

HLA typing using short sequence reads is divided into two categories.

One is to assemble short reads to create long contigs to determine the overall HLA type, and the other is to align short sequence reads with a known allele sequence as a reference. After aligning, the actual alleles are determined by the sorted information.

The assembly-based approach makes it difficult to solve false positive alleles due to phasing issues and shorter time when using short reads.

Alignment-based methods, on the other hand, are difficult to determine the actual alleles because the known alleles are very similar due to the high polymorphicity of the HLA gene region.

Despite these problems, HLAreporter was introduced as a combination-based method, PHLAT as an alignment-based method, and recently published HLAreporter and PHLAT were more accurate than previous HLA typing results. Indicates.

HLA typing according to Patent Document 11 performs highly accurate HLA typing on genome-wide short sequencing data, hereinafter referred to as HLAscan.

In the prior art, such as PHLAT, allele candidates were selected by aligned reads and depth coverage, but the HLAscan according to the present invention is a lead aligned with alleles. Use a read distribution of.

In addition, Patent Document 11 applies an algorithm for removing false positive alleles selected as a phase issue (hereinafter referred to as a 'high-glass operation algorithm').

As a result of the test using the HLAscan according to Patent Document 11, eleven 1000 genome samples, 51 HapMap samples, and five samples of themselves showed very improved accuracy compared to the prior art.

Hereinafter, the specific configuration of the HLAscan according to Patent Document 11 will be described.

HLAscan according to Patent Document 11,

1) in selecting Candidate alleles, provide a score function considering the distribution of aligned reads;

2) A unique read algorithm for detecting false positive alleles generated by a phase issue is provided.

HLAscan according to Patent Document 11 is basically based on an alignment-based approach, as shown in FIG. 2, and is largely divided into two steps.

Tier1 generates a binary alignment map (BAM) by aligning raw sequence reads generated from the NGS equipment with a whole genome reference. It is a process of selecting sequence reads corresponding to the HLA gene region.

The second stage (Tier2) first aligns the sequence reads with each allele in the IMGT / HLA database, respectively.

Using HLA-A as an example, there are 3182 known alleles of the HLA-A gene in the IMGT / HLA database, and HLAscan uses sequence reads collected with reference to these alleles. To align.

The final alleles are then determined using the aligned information.

At this time, Patent Document 11, in determining the final alleles from the sorted information, provides a score function in consideration of the aligned read distribution in order to select candidate group alleles; In selecting the final allele, we provide a unique read algorithm to solve the phase issue.

Hereinafter, specific contents of the score function and the eigenread calculation algorithm in consideration of the aligned read distribution map provided by HLAscan according to Patent Document 11 will be described.

Score function considering the distribution of aligned reads

HLAscan is a score function considering the distribution of aligned to select true alleles among thousands (about 8,000) allele models stored in the IMGT / HLA database. reads) to remove false alleles.

Alleles that remain unremoved in this process are called candidate alleles.

In this case, the score function considering the distribution of aligned reads determines that the allele of the reference is a false allele when the distribution on the reference of the aligned read is not uniformly distributed. Say.

For example, assuming that read_i (1≤i≤n) aligned to positions s_i to e_i of a reference sequence (ref) is given, 'read_i is a position (s_i + e_i) of ref. Is aligned to / 2 '. Consecutive positions of the reference whose read is not aligned are referred to as noread_j (1 ≦ j ≦ m).

In this case, the score function is

( c is a constant)

Can be calculated by

In this case, the reference whose calculated score is larger than the reference value may be determined as a reference to the false allele, and the allele may be excluded from the candidate.

Unique Lead  Algorithm

The eigenread operation algorithm according to Patent Document 11 includes 1) an algorithm for removing false positive candidate alleles when there are a large number of candidate alleles, and 2) no more than 3 candidate alleles. Alleles, if present, includes an algorithm for detecting and eliminating candidate alleles selected as a phase issue.

In addition, the eigenread operation algorithm according to Patent Document 11 can determine whether the final allele is an allele of a homozygote or a heterozygous allele on the basis of the determination result as described above.

For example, assume that sequence reads were collected from the gene to be typed and there were no sequencing errors in sequencing.

In this case, two reads A and B which have different sequences among t candidate allele_i (1 ≦ i ≦ t) have different allele_p and allele_q (1 ≦ p, q ≦ t) When mapped with 100% match to position x to y of each other and not mapped to other regions, the actual gene of the subject is included as long as it contains the sequence of Read A. Heterozygous with one strand, including strand and sequence of ReadB.

Therefore, candidate alleles that do not map to 100% match with either Lead A or Lead B are eliminated because they are false positive alleles.

In addition, the eigenread operation algorithm according to Patent Document 11 has a sequence read that is aligned only to itself for each candidate allele when there are three or less candidate alleles. The number of sequence reads is counted to select the first candidate allele and the second candidate allele in that order. The same process is repeated for the two candidate alleles selected.

If both candidate alleles have unique aligned reads, two alleles are produced as final output. In this case, it means that the allele is an allele of a heterozygote.

And when only one candidate allele has unique aligned reads (when aligned reads in one allele includes all aligned reads in another allele), only alleles with unique aligned leads Output the final result. In this case, the allele represents the allele of the homozygotes.

The present invention is a core technology of a customized medical platform that has recently been accelerated and developed. According to the present invention, the present invention is partially standardized and reproduced through a personal genetic map-based customized medical platform, beyond the level of detecting the degree of risk of toxicity. It is possible to apply commercialized customized medical platform by applying automated technology, which can greatly contribute to the medical development of human being.

Claims (3)

To analyze the supercom based integrated dielectric map and genomic data,
Different parallel distributed storage at a level that considers heterogeneous computing systems and data I / O based on computer speed;
Precomputation results and statistical tools are provided for use by multiple users in the integrated genome DB;
By creating an input file for the external statistics tool, a mySQL database is built for multiple users:
The integrated dielectric DB,
A human genome comprising 3 billion bases of divided chunks of defined size to operate on a supercomb (pc-cluster) to analyze genetic variation through comparison of the genome sequences;
The chunk is,
The genome sequences extracted from a plurality of different subjects are divided into the same order and size so as to be contrasted with each other, and are formed in a two-dimensional matrix form arranged according to the order of the subjects:
The size of the nucleotide sequence included in the chunk is increased or decreased according to the number of subjects, personal genomic map based personalized medical analysis platform, characterized in that the data size is set to 10 to 20 Gbytes.
The method of claim 1,
The parallel distributed storage,
(a) setting a limit on the I / O;
(b) personal genomic map based customized medical analysis platform, characterized in that it is set by performing the steps of confirming the consistency, reproducibility, and limitations of the data analysis results calculated at multiple nodes (multiple cpu sets).
delete

Images