Method and device for distinguishing somatic mutations and artificial mutations in DNA sequencing data generated by formalin-fixed paraffin embedding sample by using deep learning

Abstract

The present invention provides a method for providing information about artificial mutation or somatic mutation prediction and a device for providing information about artificial mutation or somatic mutation prediction wherein the method for providing information about artificial mutation or somatic mutation prediction is implemented by a processor, the method comprises the steps of: receiving data or medical data obtained from a biological sample of an individual; generating, based on the received data, data required to distinguish between artificial mutations and somatic mutations; and predicting an artificial mutation or a somatic mutation in a sample obtained from the individual based on the generated data by using a neural network model configured to predict the artificial mutation or the somatic mutation using the data required to differentiate the somatic mutation inputted thereto.

Description

Technical Field

[0001] The present invention relates to a method and device for distinguishing somatic mutations and artificial mutations in DNA sequencing data generated by formalin-fixed paraffin-embedded samples using deep learning. Background Art

[0002] Factors that influence cancer development include germline mutations (including inherited predisposition) and somatic mutations that occur during somatic cell division that are confined to specific organs or tissues.

[0003] Somatic mutations include various types of mutations, such as single nucleotide variations, structural variations, and aneuploidy. Somatic single nucleotide mutations that occur when specific mutation and repair mechanisms fail due to exposure to chemicals, ultraviolet light, smoking, etc., can predict specific cancer types. Various structural variations affect normal gene function in cancer due to chromosome deletions, insertions, inversions, tandem duplications, translocations, and complex rearrangements.

[0004] Each cancer type involves major genes, and if variants in these genes are identified, the stage of cancer progression can be determined. Furthermore, even cancers occurring in the same organ may have different combinations of genetic variants (heterogeneity), so the prognosis, treatment, recurrence, etc. of the cancer may differ depending on these other genetic differences.

[0005] Since the first analysis of the human genome in 2003, the development and application of genome sequencing technology has expanded in various ways. With the development of analytical technology that can identify many gene mutations in a single test using next-generation sequencing (NGS), cancer genome diagnostics has begun to play an important role in effective cancer treatment. With the successful development of targeted anticancer therapies for mutant proteins produced by somatic mutations of major genes such as EGFR and PIK3CA, the clinical demand for cancer genome diagnostics is increasing.

[0006] For genomic analysis, ideally, nucleic acids need to be extracted from cancer tissues immediately within 10 minutes after the vascular supply is interrupted during surgery, or if necessary, frozen in liquid nitrogen and stored at extremely low temperatures below -70°C for DNA sequencing. However, this method cannot be applied to surgical tissues of all patients in the hospital who are not research targets, and frozen tissues are not the best samples for histopathological examination. Therefore, all tissues of human origin need to be formalin-fixed paraffin-embedded (FFPE) and then microscopically sectioned and histopathologically examined. FFPE blocks can be stored at room temperature for at least 10 years or permanently. The need for actual cancer genome analysis is determined after the pathological examination results using FFPE tissues. Due to the size of cancer tissue that may be obtained by surgery or biopsy, there is usually no excess tissue that can be frozen after obtaining the tissue required for diagnostic FFPE blocks, and many hospitals do not have the infrastructure for cryopreservation, so most clinical genomic analyses are performed using FFPE samples.

[0007] FFPE samples have the advantages of convenient preservation and being most suitable for histopathological examination, but there are disadvantages in genomic analysis. That is, the chemical reaction occurring during the formalin fixation process can induce artificial mutations (artifacts) unrelated to cancer, and during DNA sequencing using NGS technology, mutations up to 10 times more than the original mutations can be found. By preparing a panel of hotspot gene mutations targeting only certain important cancer genes, the presence or absence of gene mutations can be diagnosed with considerable accuracy during genomic analysis. However, in the analysis of exome or genome levels, it is difficult to distinguish between actual mutations and artificial mutations. The presence of artificial mutations leads to inaccuracy in measuring tumor mutation burden or discovering new antigens, which becomes a major obstacle to achieving cutting-edge medicine (e.g., predicting immunotherapy response, developing new antigen vaccines, etc.). So far, a method for removing artificial mutations from FFPE NGS DNA sequencing data has been developed, but since most actual mutations are removed while removing artificial mutations, clinical applicability is poor, and therefore a technique for removing most artificial mutations while retaining actual mutations is highly desirable.

[0008] The background technology of the present invention is prepared to further help understand the present invention. The contents described in the background technology of the present invention should not be understood as prior art. Public Content Technical issues

[0009] The inventors of the present invention realize that the vast majority of general clinical samples are FFPE tissues, and NGS analysis is difficult for old FFPE tissues or samples that have mutated during the fixation process due to the above-mentioned problems.

[0010] That is, the inventors recognized that when using existing FFPE samples to distinguish somatic mutations, the accuracy would be reduced due to artificial mutations caused by various factors, and therefore a method that could accurately predict only somatic mutations was needed, and then noticed that the accuracy could be improved when using a deep learning neural network model like the present invention.

[0011] Specifically, the inventors analyzed paired NGS data generated from frozen tissue and FFPE tissue derived from the same tissue, defined mutations found only in FFPE as artificial mutations, and performed deep learning using learning data including cosine similarity, realizing that somatic mutations can be predicted with high accuracy compared to artificial mutations.

[0012] From this, the present inventors have confirmed that by applying the learning neural network model of the present invention, it is possible to distinguish between artificial mutations and somatic mutations with high accuracy.

[0013] Therefore, the inventors have developed an information providing system for distinguishing artificial mutations generated by FFPE from somatic mutations based on the deep learning neural network model of the present invention.

[0014] Therefore, the inventors were able to recognize that by introducing an information providing system that distinguishes between artificial mutations generated by FFPE and somatic mutations, the limitations of traditional diagnosis and information providing systems can be compensated, thereby proposing an effective treatment method.

[0015] Therefore, an object of the present invention is to provide a method for providing information on somatic mutation prediction by removing artificial mutations and preserving somatic mutation samples based on the deep learning of the present invention, and a device using the method.

[0016] The objects of the present disclosure are not limited to the above objects, and those skilled in the art will clearly understand other objects not mentioned above from the following description. Technical Solution

[0017] One aspect of the present invention provides a method for providing information on somatic mutation prediction based on deep learning according to an embodiment of the present invention.

[0018] The method for providing information is implemented by a processor and includes receiving data and / or medical data about mutations obtained from a biological sample derived from an individual; generating data required to distinguish between artificial mutations and somatic mutations based on the received data; and predicting artificial mutations or somatic mutations in a sample obtained from the individual by using a neural network model, the neural network model being configured to use the calculated data as input and the data required to distinguish between somatic mutations to predict artificial mutations or somatic mutations.

[0019] At this time, the data and / or medical data obtained from the biological sample of the individual source can be data about mutations, specifically data about somatic mutations. The medical data can be next generation sequencing (NGS) data, specifically, a whole genome sequence (WGS) for sequencing the whole genome, a whole exome sequencing (WES) file, a targeted sequencing file, etc. The biological sample derived from the individual can be data about mutations observed in the sample, specifically, data about somatic mutations observed in the sample, but is not limited thereto.

[0020] According to a feature of the present invention, calculating the data required to distinguish between artificial mutations and somatic mutations based on the received data may also include vectorizing the received data and calculating similarity based on the vectorized data. At this time, when calculating the similarity based on the vectorized data, the similarity may be cosine similarity, but is not limited thereto.

[0021] When predicting artificial mutations or somatic mutations in samples obtained from individuals, a neural network model is used. The neural network model is configured to use the calculated data as input and use the data required to distinguish somatic mutations to predict artificial mutations or somatic mutations. The neural network model can be a deep neural network (DNN), but is not limited to this.

[0022] According to a feature of the present invention, when calculating the data required to distinguish between artificial mutations and somatic mutations based on the received data, the calculated data may include at least any one selected from the group consisting of: the number of reads of the reference sequence observed / the sequence depth at the corresponding position, the number of reads of the alternative sequence observed / the sequence depth at the corresponding position, the ratio of the difference between the number of sequences in which read 1 (read1) is mapped to the forward chain in the reads of the reference sequence observed and the number of sequences in which read 2 (read2) is mapped to the reverse chain in the reads of the reference sequence observed, the ratio of the difference between the number of sequences in which read 1 is mapped to the forward chain in the reads of the alternative sequence observed and the number of sequences in which read 2 is mapped to the reverse chain in the reads of the alternative sequence observed, the median of the insertion length in the reads of the alternative sequence observed divided by the median of the insertion length in the reads of the alternative sequence observed. The value obtained by the median insertion length in the read segment, the base length of read segment 1 and read segment 2 in the reference fragment, the number of bases of read segment 1 and read segment 2 in the reference fragment, the base length of read segment 1 and read segment 2 in the alternative fragment, the number of bases of read segment 1 and read segment 2 in the alternative fragment, the reference sequence (REF_3_BASES) including one base before and after the mutation position, the alternative sequence (ALT_3_BASES) including one base before and after the mutation position, the ratio of the difference between the number of sequences mapped to the forward chain in the read segment of the reference sequence and the number of sequences mapped to the reverse chain in the read segment of the reference sequence, and the ratio of the difference between the number of sequences mapped to the forward chain in the read segment of the alternative sequence and the number of sequences mapped to the reverse chain in the read segment of the alternative sequence. In addition, the calculated data may also include: the cosine similarity between vectors defined by (the number of sequences in which read 1 is mapped to the forward chain and the number of sequences in which read 2 is mapped to the reverse chain in the reads in which the reference sequence is observed), or the cosine similarity between a vector defined by (the number of sequences to be mapped to the forward chain and the number of sequences to be mapped to the reverse chain in the reads in which the reference sequence is observed) and a vector defined by (the number of sequences to be mapped to the forward chain and the number of sequences to be mapped to the reverse chain in the reads in which the alternative sequence is observed). However, the present invention is not limited thereto.

[0023] In one embodiment of the present invention, the cosine similarity can be expressed by the following formula 1.

[0024] [Formula 1]

[0025] According to a feature of the present invention, in predicting artificial mutations or somatic mutations in a sample obtained from an individual by using a neural network model (the neural network model is configured to use calculated data as input and to predict artificial mutations or somatic mutations using data required to distinguish somatic mutations), the sample obtained from the individual may be an FFPE sample, the individual may be an individual suffering from cancer, and the cancer may be any one of liver cancer, colon cancer, breast cancer, lung cancer and fibrosarcoma, but is not limited thereto.

[0026] Another aspect of the present invention provides a system for providing information, the system comprising an apparatus for providing information on somatic mutation prediction according to an embodiment of the present invention, an apparatus for predicting mutation, and a server for providing medical data.

[0027] Details of other embodiments are included in the detailed description and accompanying drawings of the invention. Beneficial Effects

[0028] According to the present invention, by using a deep learning neural network model to provide information about artificial mutations or somatic mutation predictions, information about somatic mutation predictions can be provided, thereby achieving personalized medicine.

[0029] In particular, by calculating data related to somatic mutations and determining a predicted result of an artificial mutation or a somatic mutation based on the data, an artificial mutation or a somatic mutation can be predicted more accurately.

[0030] That is, by using information about artificial mutations or somatic mutation predictions based on deep learning neural network models, it is possible to more accurately set treatment directions for individuals and help select personalized treatment methods.

[0031] According to the present invention, the limitations of using traditional FFPE samples can be compensated by applying a deep learning neural network model trained using learning data including cosine similarity to predict artificial mutations or somatic mutations to predict artificial mutations and somatic mutations with high accuracy.

[0032] The effects of the present invention are not limited to the above, and other various effects are also expected herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A system for providing information on artificial mutation or somatic mutation prediction using the apparatus for predicting artificial mutation or somatic mutation according to an embodiment of the present invention is shown.

[0034] Figure 2 An example of a configuration of an apparatus for predicting an artificial mutation or a somatic mutation according to an embodiment of the present invention is shown.

[0035] Figure 3 is a block diagram of a process of a method for providing information on artificial mutation or somatic mutation prediction according to an embodiment of the present invention.

[0036] Figure 4 The process of the method for providing information on artificial mutation or somatic mutation according to various embodiments of the present invention is exemplarily shown.

[0037] Figure 5a is a diagram showing the results of evaluating the predictive performance of the deep learning neural network model of the present invention.

[0038] Figure 5b is a graph showing the results of evaluating the prediction performance of Comparative Example 1 (FFpolish pre-training).

[0039] Figure 5c is a diagram showing the results of evaluating the prediction performance of Comparative Example 2 (FFpolish retraining).

[0040] Figure 5d is a diagram showing the results of evaluating the prediction performance of Comparative Example 3 (SOBDetector).

[0041] Figure 6a The figure shows the comparison of the accuracy of predicting artificial mutations or somatic mutations in liver cancer samples using the deep learning neural network model of the present invention.

[0042] Figure 6b The figure shows a comparison of the accuracy of predicting artificial mutations or somatic mutations in colon cancer samples using the deep learning neural network model of the present invention.

[0043] Figure 6c The figure shows a comparison of the accuracy of predicting artificial mutations or somatic mutations in breast cancer samples using the deep learning neural network model of the present invention.

[0044] Figure 6d A comparison of the accuracy of predicting artificial mutations or somatic mutations in lung cancer samples using the deep learning neural network model of the present invention is shown.

[0045] Figure 6e A comparison of the accuracy of predicting artificial mutations or somatic mutations in fibrosarcoma samples using the deep learning neural network model of the present invention is shown.

[0046] Figure 7a A comparison of the accuracy of predicting artificial mutations or somatic mutations of liver cancer samples using a comparative example (FFpolish pre-training) is shown.

[0047] Figure 7bA comparison of the accuracy of predicting artificial mutations or somatic mutations of colon cancer samples using a comparative example (FFpolish pre-training) is shown.

[0048] Figure 7c A comparison of the accuracy of predicting artificial mutations or somatic mutations of breast cancer samples using a comparative example (FFpolish pre-training) is shown.

[0049] Figure 7d A comparison of the accuracy of predicting artificial mutations or somatic mutations of lung cancer samples using a comparative example (FFpolish pre-training) is shown.

[0050] Figure 7e Comparison of the accuracy of predicting artificial or somatic mutations of fibrosarcoma samples using a comparative example (FFpolish pre-trained) is shown. DETAILED DESCRIPTION

[0051] Advantages and features of the present invention and methods for achieving the present invention will be more clearly understood from the following embodiments described in detail with reference to the accompanying drawings.

[0052] However, the present invention is not limited to the following embodiments, but will be implemented in various forms. The embodiments are only used to complete the disclosure of the present invention and are intended to enable a person of ordinary skill in the art to which the present invention belongs to have a complete understanding of the scope of the present invention. The present invention is limited only by the scope of the claims. In conjunction with the description of the drawings, the same reference numerals may be used for the same components.

[0053] In this specification, expressions such as “having”, “may have”, “including” or “may include” refer to the existence of corresponding features (for example, values, functions, operations or components (such as parts)), and do not exclude the existence of additional features.

[0054] In this specification, expressions such as "A or B", "at least one of A and / or B", or "one or more of A and / or B" may include all possible combinations of the items listed together. For example, "A or B", "at least one of A and B", or "at least one of A or B" may refer to all of the following: (1) including at least one A, (2) including at least one B, or (3) including at least one A and at least one B.

[0055] As used herein, expressions such as "first", "second", etc. may describe various components regardless of their order and / or importance, and are only used to distinguish one component from another component, but do not limit the corresponding components. For example, a first user device and a second user device may represent different user devices regardless of their order or importance. For example, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component without departing from the scope of this specification.

[0056] When a certain component (for example, a first component) is referred to as being "(operably or communicatively) coupled / connected to" or "connected to" another component (for example, a second component), it should be understood that the component may be directly connected to the other component, or may be connected to the other component through another component (for example, a third component). When a certain component (for example, a first component) is referred to as being "directly coupled to" or "directly connected to" another component (for example, a second component), it should be understood that there is no another component (for example, a third component) between the certain component and the other component.

[0057] The expression "configured to" used in this article can be changed and used according to the situation, such as "suitable for", "capable of", "designed to", "adapted to", "manufactured to", or "capable of". The term "configured to" does not necessarily mean "specially designed to" in hardware. In some cases, the expression "the device is configured to" may mean that the device is "capable" together with other devices or components. For example, the phrase "the processor is configured to perform A, B, and C" may mean a dedicated processor (such as an embedded processor) for performing the corresponding operations, or a general-purpose processor (such as a CPU, GPU, or application processor) that can perform the corresponding operations by executing one or more software programs stored in a memory device.

[0058] The terms used herein are only used to describe specific embodiments and are not intended to limit the scope of other embodiments. Unless the context clearly stipulates otherwise, singular expressions may include plural expressions. The terms used herein, including technical or scientific terms, may have the same meaning as the meanings generally understood by those of ordinary skill in the field described in this specification. The terms defined in a general dictionary in the terms used herein may be interpreted as meanings identical or similar to those in the context of the relevant art, and unless otherwise defined in this specification, should not be interpreted as ideal or overly formalized meanings. In some cases, even the terms defined in this specification may not be interpreted as excluding embodiments of this specification.

[0059] The features of the various embodiments of the present invention may be coupled or combined with each other in part or in whole, and may be interlocked and operated in various technical ways that are fully understood by those skilled in the art, and the embodiments may be implemented independently or in relation to each other.

[0060] To make the explanation of this specification clear, the terms used in this specification will be defined below.

[0061] The term "individual" as used herein may refer to any subject in which a mutation, particularly a somatic mutation, is to be predicted. For example, the individual may be an individual in which a somatic mutation has occurred. In this case, the individual disclosed in this specification may be all mammals except humans, but is not limited thereto.

[0062] The term "biological sample" as used herein refers to all biological samples derived from an individual and can be stored in the form of FFPE, and may include urine, tissue, cell lysate, whole blood, plasma, serum, saliva, eye fluid, cerebrospinal fluid, sweat, milk, ascites, synovial fluid, peritoneal fluid, etc., but is not limited thereto. Preferably, the biological sample may be a tissue sample derived from an individual to determine whether there is a somatic mutation, but is not limited thereto. Specifically, the biological sample of the present invention may be a formalin-fixed paraffin-embedded (FFPE) sample, and the FFPE sample may refer to a sample obtained by fixing the tissue in formalin according to a standard method used when collecting and preserving cancerous tissue from a patient, hardening the tissue with paraffin to prepare a paraffin block so that it can be stored at room temperature, and by cutting the paraffin block into thin slices for various histological examinations.

[0063] The term "somatic mutation (somatic variation)" used herein refers to one of the cancer features identified in cells accompanied by genetic abnormalities, and therefore, the occurrence and type of cancer can be distinguished by identifying somatic variation. Therefore, in the present invention, the data and / or medical data obtained from a biological sample derived from an individual may be data on somatic mutations, and the medical data may be next-generation sequencing (NGS) data, but is not limited thereto, and may refer to somatic mutations observed in a biological sample derived from an individual, but is not limited thereto.

[0064] The term "artificial mutation (artifact)" used herein may refer to an artificial defect or the like that causes a non-cancerous change due to a physical or chemical change, but is not limited thereto.

[0065] The term "next generation sequencing (NGS)" used herein refers to next generation sequencing, which is one of the high-speed sequencing methods of genome base sequences.

[0066] At this point, NGS can be applied to genome and whole genome analysis for a variety of purposes, including clinical research. At the same time, most NGS analyses can be performed on multiple target samples simultaneously.

[0067] The term "NGS data" as used herein may refer to a file containing sequencing data analyzed for a target sample.

[0068] At this time, the NGS data may include whole genome sequencing (WGS) files for whole genome sequencing, whole exome sequencing (WES) files, RNA sequencing files, and targeted sequencing files for sequencing specific regions, particularly whole exome sequencing (WES) files, but not limited thereto.

[0069] The following will refer to Figure 1 and Figure 2 A system for providing information on artificial mutation or somatic mutation prediction based on an apparatus for predicting artificial mutation or somatic mutation according to one embodiment of the present invention and an apparatus for predicting artificial mutation or somatic mutation are described.

[0070] Figure 1 A system for providing information on artificial mutation or somatic mutation prediction using an apparatus for predicting artificial mutation or somatic mutation according to an embodiment of the present invention is shown.

[0071] Figure 2 An example of a configuration of an apparatus for predicting an artificial mutation or a somatic mutation according to an embodiment of the present invention is shown.

[0072] First, refer to Figure 1 , the system 1000 for providing information may be a system configured to provide information related to the prediction of artificial mutations or somatic mutations by applying a deep learning neural network model based on data obtained from a biological sample and / or medical data. At this time, the system 1000 for providing information may be composed of a server / device for providing data 100 for providing data obtained from a biological sample and / or medical data, a device for predicting artificial mutations or somatic mutations based on the received data, and a device for providing information 300 for visually providing information about the prediction of artificial mutations or somatic mutations in a sample obtained from an individual based on the received prediction data.

[0073] Here, the device 300 for providing information is an electronic device that provides a user interface to display information related to artificial mutation or somatic mutation prediction, and may include at least one of a smartphone, a tablet personal computer (PC), a laptop computer, and / or a PC.

[0074] The apparatus 300 for providing information on somatic mutation prediction may receive information related to artificial mutation or somatic mutation prediction results from the apparatus 200 for predicting somatic mutation, and display the received results through a display unit described below.

[0075] The device 200 for predicting somatic mutations may include a general-purpose computer, a laptop computer, and / or a data server, which performs various operations to determine information related to artificial mutations or somatic mutation prediction based on various data (particularly data on somatic mutations and / or medical data) that can be obtained from the server / device 100 for providing data. At this time, the server 100 for providing data may be a device that accesses a web server that provides a web page or a mobile web server that provides a mobile website, but is not limited thereto.

[0076] More specifically, the apparatus 200 for predicting somatic mutations may receive data on somatic mutations and / or medical data from a server 100 providing medical data, and predict artificial mutations or somatic mutations based on the received data, and provide information related thereto.

[0077] At this time, the device 200 for predicting somatic mutations applies a deep learning neural network model to predict artificial mutations or somatic mutations based on data and / or medical data about somatic mutations obtained from biological samples received from the server 100 for providing data.

[0078] In addition, the apparatus 200 for predicting a somatic mutation may provide a prediction result of an artificial mutation or a somatic mutation to the apparatus 300 for providing information on somatic mutation prediction.

[0079] Thus, the information provided from the apparatus 200 for predicting somatic mutations can be provided on a web page through a web browser installed in the apparatus 300 for providing information on somatic mutation prediction, or provided in the form of an application or program. In various embodiments, these data can be provided in the form of a platform included in a client-server environment.

[0080] Hereinafter, the apparatus 200 for predicting somatic mutations according to an embodiment of the present invention is described as receiving data obtained from medical data and / or data obtained from a biological sample from a server 100 for providing data to perform operations, but the apparatus 300 for providing information about somatic mutation prediction can perform all operations itself.

[0081] Next, we will refer to Figure 2 The components of the apparatus 200 for predicting somatic mutations of the present invention are described in detail.

[0082] refer to Figure 2The device 200 for predicting somatic mutations may include a communication interface 210, a memory 220, an I / O interface 230, and a processor 240. The components may communicate with each other via one or more communication buses or signal lines.

[0083] The communication interface 210 may be connected to the device 300 for providing information on somatic mutation prediction and the server 100 for providing medical data through a wired / wireless communication network to exchange data. For example, the communication interface 210 may receive data and / or individual data on somatic mutations from the server 100 for providing medical data, and may transmit the determined information on artificial mutations or somatic mutation prediction results to the device 300 for providing information on somatic mutation predictions.

[0084] On the other hand, the communication interface 210 for sending and receiving such data includes a communication port 211 and a wireless circuit 212, wherein the wired communication port 211 may include one or more wired interfaces, such as Ethernet, a universal serial bus (USB), FireWire, etc. In addition, the wireless circuit 212 may send and receive data with an external device via an RF signal or an optical signal. In addition, the wireless communication may use at least one of a plurality of communication standards, protocols, and technologies, such as GSM, EDGE, CDMA, TDMA, Bluetooth, Wi-Fi, VoIP, Wi-MAX, or any other suitable communication protocol.

[0085] The memory 220 may store various data used in the apparatus 200 for predicting somatic mutations. For example, the memory 220 may store data on somatic mutations, or store a deep learning model configured to predict artificial mutations or somatic mutations based on the data.

[0086] In various embodiments, the memory 220 may include a volatile or non-volatile recording medium capable of storing various data, instructions, and information. For example, the memory 220 may include at least one type of storage medium in a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (e.g., SD or XD memory, etc.), RAM, SRAM, ROM, EEPROM, PROM, a network storage device, a cloud, and a blockchain database.

[0087] In various embodiments, the memory 220 may store at least one configuration of an operating system 221 , a communication module 222 , a user interface module 223 , and one or more application programs 224 .

[0088] The operating system 221 (e.g., an embedded operating system such as LINUX, UNIX, MAC OS, WINDOWS, VxWorks, etc.) may include various software components and drivers for controlling and managing general system operations (e.g., memory management, storage device control, power management, etc.), and may support communication between various hardware, firmware, and software components.

[0089] The communication module 223 may support communication with other devices through the communication interface 210. The communication module 220 may include various software components for processing data received by the wired communication port 211 or the wireless circuit 212 of the communication interface 210.

[0090] The user interface module 223 may receive a user's request or input from a keyboard, a touch screen, a microphone, etc. through the I / O interface 230 and provide a user interface on a display.

[0091] The application 224 may include a program or module configured to be executed by one or more processors 230. Here, the application for providing information related to somatic mutation prediction may be implemented on a server cluster.

[0092] The I / O interface 230 may connect at least one of the I / O devices (not shown) (e.g., a display, a keyboard, a touch screen, and a microphone) of the apparatus 200 for predicting somatic mutations to the user interface module 223. The I / O interface 230 may receive user input (e.g., voice input, keyboard input, touch input, etc.) through the user interface module 223, and process instructions according to the received input.

[0093] The processor 240 is connected to the communication interface 210 , the memory 220 , and the I / O interface 230 to control the overall operation of the apparatus 200 for predicting somatic mutations, and may execute various instructions for providing information through an application program or program stored in the memory 220 .

[0094] The processor 240 may correspond to a computing device, such as a central processing unit (CPU), a graphics processing unit (GPU), or an application processor (AP). In addition, the processor 240 may be implemented in the form of an integrated chip (IC), such as a system on a chip (SoC) integrated with various computing devices. Alternatively, the processor 240 may include a module for calculating an artificial neural network model, such as a neural processing unit (NPU).

[0095] In various embodiments, the processor 240 can be configured to apply a deep learning neural network model to generate data required to distinguish between artificial mutations and somatic mutations based on data obtained from a biological sample and / or medical data, and to predict and provide artificial mutations or somatic mutations.

[0096] In an embodiment, the device 200 for predicting somatic mutations may receive medical data from the device 100 for providing data and / or medical data obtained from a biological sample derived from an individual, and apply a deep learning neural network model using the received data to provide a prediction result of a somatic mutation using a model learned to predict artificial mutations or somatic mutations. At this time, the data and / or medical data obtained from a biological sample derived from an individual may be data about somatic mutations, and the medical data may be next-generation sequencing (NGS) data, but is not limited thereto, and may represent data obtained by observing somatic mutations in a biological sample derived from an individual, but is not limited thereto. Specifically, the NGS data may be at least one of a WGS file, a WES file, an RNA sequencing file, and a target sequencing file, in particular a WES file, but is not limited thereto.

[0097] In the following, reference will be made to Figure 3 and Figure 4 A method for providing information on artificial mutation or somatic mutation prediction according to an embodiment of the present invention is described in detail.

[0098] Figure 3 is a block diagram of the process of a method for providing information on artificial mutation or somatic mutation prediction according to one embodiment of the present invention.

[0099] Figure 4 The process of the method for providing information on artificial mutation or somatic mutation according to various embodiments of the present invention is exemplarily shown.

[0100] First, refer to Figure 3 , the process of the method for providing information about artificial mutation or somatic mutation prediction according to an embodiment of the present invention is as follows.

[0101] First, data on mutations occurring in a biological sample from an individual and / or medical data are received ( S310 ).

[0102] Next, data required to distinguish between artificial mutations and somatic mutations is generated based on the received data ( S320 ).

[0103] Artificial mutations or somatic mutations in a sample obtained from an individual are predicted by using a neural network model, the neural network model being configured to use the calculated data as input and to predict artificial mutations or somatic mutations using data required for distinguishing somatic mutations (S330).

[0104] According to the features of the present invention, in the step (S310) of receiving data and / or medical data about mutations occurring in a biological sample derived from an individual, a whole exome sequencing (WES) file, an RNA sequencing file, or a targeted sequencing file may be received, and specifically, a WES file may be received, but is not limited thereto.

[0105] At this time, the WES file, the RNA sequencing file, and the targeted sequencing file may have a format of a BAM file, but the format of the NGS data is not limited thereto.

[0106] Next, the step of generating data required to distinguish between artificial mutations and somatic mutations based on the received data (S320) may also include vectorizing the received data and calculating similarity based on the vectorized data.

[0107] At this time, refer to Table 1 and Figure 4, when calculating the data required to distinguish between artificial mutations and somatic mutations based on the received data, the calculated data may include at least one selected from the following groups: the number of reads of the reference sequence observed / the sequence depth at the corresponding position, the number of reads of the alternative sequence observed / the sequence depth at the corresponding position, the ratio of the difference between the number of sequences in which read 1 is mapped to the forward chain in the reads of the reference sequence observed and the number of sequences in which read 2 is mapped to the reverse chain in the reads of the reference sequence observed, the ratio of the difference between the number of sequences in which read 1 is mapped to the forward chain in the reads of the alternative sequence observed and the number of sequences in which read 2 is mapped to the reverse chain in the reads of the alternative sequence observed, the median of the insertion length in the reads of the alternative sequence observed divided by the median of the insertion length in the reads of the alternative sequence observed The values ​​obtained by the median, the base length of read 1 and read 2 in the reference fragment, the number of bases of read 1 and read 2 in the reference fragment, the base length of read 1 and read 2 in the alternative fragment, the number of bases of read 1 and read 2 in the alternative fragment, the reference sequence including one base before and after the mutation position (REF_3_BASES), the alternative sequence including one base before and after the mutation position (ALT_3_BASES), the ratio of the difference between the number of sequences mapped to the forward chain in the reads of the reference sequence and the number of sequences mapped to the reverse chain in the reads of the reference sequence, and the ratio of the difference between the number of sequences mapped to the forward chain in the reads of the alternative sequence and the number of sequences mapped to the reverse chain in the reads of the alternative sequence. In addition, the calculated data may also include the cosine similarity between vectors defined by (the number of sequences in which read 1 is mapped to the forward chain and the number of sequences in which read 2 is mapped to the reverse chain in the reads where the reference sequence is observed), or the cosine similarity between a vector defined by (the number of sequences to be mapped to the forward chain and the number of sequences to be mapped to the reverse chain in the reads where the reference sequence is observed) and a vector defined by (the number of sequences to be mapped to the forward chain and the number of sequences to be mapped to the reverse chain in the reads where the alternative sequence is observed), but is not limited to this.

[0108] [Table 1]

[0109] As used herein, the term "read" may refer to a DNA including an adaptor, and the term "fragment" may refer to a DNA not including an adaptor.

[0110] The term "read 1" as used herein means reading in the 5' to 3' direction when the paired-end DNA is read from both ends when entering the NGS device, and "read 2" means reading in the 3' to 5' direction.

[0111] The term "depth" as used herein is used interchangeably with the term "read depth" to refer to the thickness or depth of a read.

[0112] The term "fr_ref" used in this article represents the sequence number of read 1 mapped to the forward strand among the reads observed for the reference sequence.

[0113] The term "rf_ref" used herein indicates the sequence number to which read 2 is mapped to the reverse strand.

[0114] The term "fr_alt" as used herein indicates the sequence number where read 1 is mapped to the forward strand among the reads where the alternative sequence is observed.

[0115] The term "rf_alt" used herein indicates the sequence number to which read 2 is mapped to the reverse strand.

[0116] The term "fs_ref" as used herein represents the sequence number to be mapped to the forward strand among the reads where the reference sequence is observed.

[0117] The term "rs_ref" as used herein refers to the sequence number to be mapped to the reverse strand.

[0118] The term "fs_alt" as used herein represents the sequence number to be mapped to the forward strand in the reads where the alternative sequence is observed.

[0119] The term "rs_alt" as used herein refers to the sequence number to be mapped to the reverse strand.

[0120] At this time, the model into which the data about mutations and / or medical data occurring in a biological sample originating from an individual is input can be based on various algorithms, such as a convolutional neural network (CNN) model, a deep neural network (DNN), a deep convolutional neural network (DCNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), a deep belief network (DBN), a single shot detector (SSD), and a support vector machine (SVM), specifically, a CNN model and a DNN model, more specifically a DNN model, but not limited to this.

[0121] Next, in step (S330), an artificial mutation or somatic mutation of a sample obtained from an individual is predicted using a neural network model, the neural network model being configured to use the calculated data as input, and to predict artificial mutations or somatic mutations using the data required to distinguish somatic mutations, and to determine and provide a prediction result of an artificial mutation or somatic mutation of a sample obtained from an individual. For example, the model may be configured to output whether the result value is a true artificial mutation, a false artificial mutation, a true somatic mutation, or a false somatic mutation. For example, the corresponding output may be expressed as a percentage, and a threshold may be set to determine whether the percentage is greater than the threshold value is a true artificial mutation or a false artificial mutation, or a true somatic mutation or a false somatic mutation.

[0122] At this time, the sample obtained from the individual may be an FFPE sample, and the individual may be an individual suffering from any one of liver cancer, colon cancer, breast cancer, lung cancer, and fibrosarcoma, but is not limited thereto.

[0123] Meanwhile, the process of providing information on artificial mutation or somatic mutation prediction is not limited to the above process.

[0124] According to the above-mentioned method for providing information about artificial mutations or somatic mutation predictions, the present invention provides a system for providing information about artificial mutations or somatic mutation predictions based on an algorithm, thereby making it easier to set treatment directions for each individual and facilitating the selection of tailored treatment methods.

[0125] That is, the present invention can supplement the limitations of traditional diagnosis and information provision systems by introducing a system for providing information about the prediction of artificial mutations or somatic mutations based on a neural network model, thereby predicting artificial mutations or somatic mutations with high accuracy and making it easier to set treatment directions for each individual.

[0126] Example 1: Prediction of artifactual or somatic mutation rates in FFPE samples

[0127] To generate a trainable dataset, cancer tissue formalin-fixed paraffin-embedded (FFPE) samples were collected and mutations that were also present in frozen fresh (FF) samples were predicted from 112,976 mutant colonies.

[0128] The results confirmed that 2,113 mutations were found in both FF and FFPE samples, and 110,863 mutations were found only in FFPE samples, indicating that only about 1.8% of the mutations were somatic mutations rather than artifactual mutations.

[0129] Subsequently, using the obtained samples and a deep learning neural network model, features highly correlated with somatic prediction were derived, and then the performance of somatic mutation prediction was evaluated.

[0130] Example 2: Evaluation of the prediction performance of artificial or somatic mutations in deep learning neural network models

[0131] The following will describe the performance evaluation results of somatic mutation prediction in the deep learning neural network model according to an embodiment of the present invention with reference to FIG5 .

[0132] Figure 5a to Figure 5d is a diagram showing the results of evaluating the performance of a deep learning neural network model according to an embodiment of the present invention and a comparative example.

[0133] First, in order to evaluate the deep learning neural network model according to the embodiment of the present invention, models using various somatic mutation-related data were constructed as comparative examples 1 to 3. More specifically, the experiment was conducted by comparing the FFpolish (pre-trained) model as comparative example 1, the FFpolish (re-trained) model as comparative example 2, and the SOBDetector model as comparative example 3 with the neural network model (DEEPOMICS(R)FFPE) of the present invention.

[0134] Precision, recall, F1 score, accuracy and specificity are calculated as follows.

[0135]

[0136]

[0137]

[0138]

[0139]

[0140] The results are as follows Figure 5a As shown, it is confirmed that the neural network model according to the embodiment of the present invention ( FFPE) removed 99.6% of artifactual mutations while retaining 77.8% of actual somatic mutations.

[0141] On the other hand, Figure 5b As shown, in Comparative Example 1 (FFpolish (pre-training)), it was confirmed that 92.6% of the artificial mutations were removed and 69.5% of the actual somatic mutation samples were retained, thus confirming that the ability to remove artificial mutations and preserve actual somatic mutation samples was lower than that of the neural network model according to an embodiment of the present invention.

[0142] like Figure 5c As shown, in the case of comparative example 2 (FFpolish (retraining)), the artificial mutation removal rate is 99.6%, showing a high removal ability, but it is confirmed that the actual somatic mutation preservation rate is 39.8%. In the model of comparative example 2, it is confirmed that when removing artificial mutations, samples with actual somatic mutations are also removed, thereby affecting the prediction results.

[0143] like Figure 5d As shown, in Comparative Example 3 (SOBDetector), the artificial mutation removal rate was 72.1% and the actual somatic mutation retention rate was 84.5%, confirming that the artificial mutation removal rate was significantly lower.

[0144] When the neural network model according to the embodiment of the present invention is used, the F1score is also 0.821, which is superior to other comparative examples. As described above, it is confirmed that while the artificial mutation removal ability is excellent, the samples with actual somatic mutations are not removed but retained.

[0145] Evaluation 2: Accuracy of prediction of artifactual or somatic mutations by cancer type-specific algorithms

[0146] The following will refer to Figures 6a to 6e and Figures 7a to 7e Results of evaluating the prediction performance of a neural network model according to an embodiment of the present invention for somatic mutations in various cancer types are described.

[0147] Figures 6a to 6e A comparison of the accuracy of predicting artificial mutations or somatic mutations for liver cancer, colon cancer, breast cancer, lung cancer, and fibrosarcoma using a neural network model according to an embodiment of the present invention is shown.

[0148] Figures 7a to 7e A comparison of the accuracy of artificial or somatic mutation prediction for liver cancer, colon cancer, breast cancer, lung cancer, and fibrosarcoma using the FFPolish model of Comparative Example 1 of Evaluation 1 is shown.

[0149] The results are as follows Figures 6a to 6e As shown, when the neural network model according to the implementation scheme of the present invention is adopted, the ability to remove artificial mutations is excellent in liver cancer, colon cancer, breast cancer, lung cancer, and fibrosarcoma samples, which are 89.9%, 88.2%, 99.8%, 96.5%, and 89.7%, respectively, and the somatic mutation preservation rates are also shown to be 88.5%, 89.3%, 63.9%, 76.4%, and 90.2%, respectively, which confirms that the ability to remove artificial mutations is excellent and the ability to preserve actual somatic mutation samples is also excellent.

[0150] On the other hand, in the case of the FFpolish model of the comparative example, Figures 7a to 7e As shown, it is confirmed that the ability to remove artificial mutations is lower than the model of the present invention in all samples, which are 85.3% for liver cancer, 63.4% for colon cancer, 92.8% for breast cancer, 91.2% for lung cancer, and 71.8% for fibrosarcoma, respectively. The somatic mutation samples retained are 69.1% for liver cancer, 78.9% for colon cancer, 60.6% for breast cancer, 74.3% for lung cancer, and 70.7% for fibrosarcoma. Therefore, it is confirmed that the probability of removing actual somatic mutation samples while removing artificial mutations is higher than that of the neural network model according to the embodiment of the present invention.

[0151] Although the embodiments of the present invention have been described in detail with reference to the accompanying drawings, the present invention is not limited thereto and can be implemented in a variety of different forms without departing from the technical concept of the present invention. Therefore, the embodiments of the present invention are only for illustrative purposes and are not intended to limit the technical concept of the present invention. The scope of the technical concept of the present invention is not limited thereto. Therefore, it should be understood that the above-mentioned embodiments are illustrative in all aspects and are not restrictive. The scope of protection of the present invention should be interpreted based on the following claims, and all technical concepts within their equivalent ranges should be interpreted as falling within the scope of the present invention.

Claims

1. A method for providing information about artificial mutation or somatic mutation prediction implemented by a processor, the method comprising the following steps: receiving data or medical data obtained from a biological sample derived from an individual; Calculating data required to distinguish the artificial mutation from the somatic mutation based on the received data; as well as The artificial mutation or the somatic mutation in the sample obtained from the individual is predicted by using a neural network model, wherein the neural network model is configured to use the calculated data as input and to predict the artificial mutation or the somatic mutation using the data required to distinguish the somatic mutation.

2. The method of claim 1, wherein the medical data is next generation sequencing (NGS) data regarding somatic mutations.

3. The method according to claim 2, wherein the NGS data is any one of a whole genome sequencing (WGS) file for sequencing the entire genome, a whole exome sequencing (WES) file, an RNA sequencing file, and a targeted sequencing file for sequencing a specific region.

4. The method according to claim 1, wherein when the data required to distinguish the artificial mutation from the somatic mutation is calculated based on the received data, the calculated data includes at least one selected from the group consisting of: the number of reads of the reference sequence observed / the sequence depth at the corresponding position, the number of reads of the alternative sequence observed / the sequence depth at the corresponding position, the ratio of the difference between the number of sequences in which read 1 is mapped to the forward chain in the reads of the reference sequence observed and the number of sequences in which read 2 is mapped to the reverse chain in the reads of the reference sequence observed, the ratio of the difference between the number of sequences in which read 1 is mapped to the forward chain in the reads of the alternative sequence observed and the number of sequences in which read 2 is mapped to the reverse chain in the reads of the alternative sequence observed, the median of the insertion length in the reads of the alternative sequence observed divided by the median of the insertion length in the reads of the alternative sequence observed the value obtained by inputting the median of the length, the base length of read 1 and read 2 in the reference fragment, the number of bases of read 1 and read 2 in the reference fragment, the base length of read 1 and read 2 in the alternative fragment, the number of bases of read 1 and read 2 in the alternative fragment, the reference sequence including one base before and after the mutation position (REF_3_BASES), the alternative sequence including one base before and after the mutation position (ALT_3_BASES), the ratio of the difference between the number of sequences mapped to the forward chain in the reads of the reference sequence and the number of sequences mapped to the reverse chain in the reads of the reference sequence, and the ratio of the difference between the number of sequences mapped to the forward chain in the reads of the alternative sequence and the number of sequences mapped to the reverse chain in the reads of the alternative sequence.

5. The method according to claim 1, wherein: Calculating the data required to distinguish the artificial mutation from the somatic mutation based on the received data includes vectorizing the received data.

6. The method according to claim 5, wherein: Calculating the data required to distinguish the artificial mutation from the somatic mutation based on the received data also includes calculating similarity based on the vectorized data.

7. The method according to claim 4, wherein when the data required to distinguish the artificial mutation from the somatic mutation is calculated based on the received data, the calculated data further comprises: The cosine similarity between vectors defined by (the number of sequences at which read 1 is mapped to the forward strand and the number of sequences at which read 2 is mapped to the reverse strand in the reads where the reference sequence is observed), or the cosine similarity between a vector defined by (the number of sequences to be mapped to the forward strand and the number of sequences to be mapped to the reverse strand in the reads where the reference sequence is observed) and a vector defined by (the number of sequences to be mapped to the forward strand and the number of sequences to be mapped to the reverse strand in the reads where the alternative sequence is observed).

8. The method according to claim 1, wherein in determining the prediction of the artificial mutation or somatic mutation in the sample obtained from the individual based on the calculated data, The sample obtained from the individual is a FFPE sample.

9. The method according to claim 1, wherein in determining the prediction of the artificial mutation or somatic mutation in the sample obtained from the individual based on the calculated data, The individual is an individual suffering from cancer.

10. The method of claim 9, wherein the cancer is any one of liver cancer, colon cancer, breast cancer, lung cancer, and fibrosarcoma.

11. An apparatus for providing information on prediction of artificial mutations or somatic mutations, the apparatus comprising: a communication unit configured to receive data on somatic mutation or medical data obtained from a biological sample from a server for providing data; as well as a processor electrically connected to the communication unit, Wherein, the processor is configured to: generating data required to distinguish between artificial mutations and somatic mutations based on the received data; and The artificial mutation or the somatic mutation in the sample obtained from the individual is predicted by using a neural network model, wherein the neural network model is configured to use the calculated data as input and to predict the artificial mutation or the somatic mutation using the data required to distinguish the somatic mutation.

12. The apparatus of claim 11, wherein the medical data is next generation sequencing (NGS) data regarding the somatic mutation.

13. The apparatus according to claim 12, wherein the NGS data is any one of a whole genome sequencing (WGS) file for sequencing the entire genome, a whole exome sequencing (WES) file, an RNA sequencing file, and a targeted sequencing file for sequencing a specific region.

14. The apparatus according to claim 11, wherein the data required to distinguish the artificial mutation from the somatic mutation based on the received data includes at least one selected from the group consisting of: the number of reads of the reference sequence observed / sequence depth at the corresponding position, the number of reads of the alternative sequence observed / sequence depth at the corresponding position, the ratio of the difference between the number of sequences in which read 1 is mapped to the forward chain in the reads of the reference sequence observed and the number of sequences in which read 2 is mapped to the reverse chain in the reads of the reference sequence observed, the ratio of the difference between the number of sequences in which read 1 is mapped to the forward chain in the reads of the alternative sequence observed and the number of sequences in which read 2 is mapped to the reverse chain in the reads of the alternative sequence observed, the median of the insertion length in the reads of the alternative sequence observed divided by the median of the insertion length in the reads of the alternative sequence observed The values ​​obtained, the base length of read 1 and read 2 in the reference fragment, the number of bases of read 1 and read 2 in the reference fragment, the base length of read 1 and read 2 in the alternative fragment, the number of bases of read 1 and read 2 in the alternative fragment, the reference sequence (REF_3_BASES) including one base before and after the mutation position, the alternative sequence (ALT_3_BASES) including one base before and after the mutation position, the ratio of the difference between the number of sequences mapped to the forward chain in the reads of the reference sequence and the number of sequences mapped to the reverse chain in the reads of the reference sequence, and the ratio of the difference between the number of sequences mapped to the forward chain in the reads of the alternative sequence and the number of sequences mapped to the reverse chain in the reads of the alternative sequence.

15. The device of claim 11, wherein the processor is further configured to vectorize the received data. 16 . The apparatus according to claim 15 , wherein the processor is further configured to calculate similarity based on the vectorized data.

17. The apparatus according to claim 14, wherein: The data required to distinguish the artificial mutation from the somatic mutation based on the received data also includes: the cosine similarity between vectors defined by (the number of sequences in which read 1 is mapped to the forward chain and the number of sequences in which read 2 is mapped to the reverse chain in the reads in which the reference sequence is observed), or the cosine similarity between a vector defined by (the number of sequences to be mapped to the forward chain and the number of sequences to be mapped to the reverse chain in the reads in which the reference sequence is observed) and a vector defined by (the number of sequences to be mapped to the forward chain and the number of sequences to be mapped to the reverse chain in the reads in which the alternative sequence is observed).

18. The apparatus of claim 11, wherein: The sample obtained from the individual is a FFPE sample.

19. The apparatus of claim 11, wherein the individual is an individual suffering from cancer.

20. The apparatus according to claim 19, wherein the cancer is any one of liver cancer, colon cancer, breast cancer, lung cancer and fibrosarcoma.

Images