System and method for predicting genetic diseases and discovering genetic variants related to disease induction

Abstract

To provide a system for predicting genetic diseases, and discovering disease-induced genetic variants.SOLUTION: A system 1000 for predicting genetic diseases and discovering disease-induced genetic variants includes a multiple instance learning model unit that derives both predicting a genetic disease of a patient, and discovering a disease-induced genetic variant using a multiple instance learning model that learns, by input data, an instance serving as genetic variant information on the patient and a bag of the instance, and processes whether a disease of the patient is a genetic disease caused by a genetic variant by a bag label.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a system and method for predicting hereditary diseases and discovering genetic variants involved in disease induction, and in particular to a system and method for predicting hereditary diseases and discovering genetic variants involved in disease induction, which uses multiple instance learning to determine whether a patient's disease is a hereditary disease and to discover genetic variants involved in disease induction that cause the disease from among the patient's many genetic variants. [Background technology]

[0002] A genetic disease is a disease caused by a mutation in a gene or chromosome. A disease-causing genetic variant is a genetic mutation that causes a genetic disease.

[0003] Interpretation of human genetic variants is the process of searching for one or more genetic variants that are involved in the induction of disease.

[0004] Generally, to identify genetic variants related to disease induction, the genomic information of patients with a specific disease (cases) is compared with that of healthy individuals (controls), and genetic variants that are significantly more common in the cases than in the controls are identified.

[0005] Recently, research has been conducted using artificial intelligence to determine whether a patient's disease is a hereditary disease or to search for genetic variants that may be the cause of the patient's disease.

[0006] Research has been conducted using single-instance learning to predict the genetic variants associated with each patient's disease for each genetic variant. However, this type of single-instance learning requires labels (instance labels) for each genetic variant, which limits its practical use.

[0007] The present invention proposes a method for simultaneously determining whether a patient's disease is a hereditary disease and identifying the genetic variants involved in disease induction among many genetic variants, using a multiple instance learning (MIL) model. Summary of the Invention [Problem to be solved by the invention]

[0008] This specification provides a system and method for predicting hereditary diseases and discovering genetic variants involved in disease induction, which uses multiple instance learning to simultaneously determine whether a patient's disease is a hereditary disease and to identify genetic variants involved in disease induction among the patient's many genetic mutations. [Means for solving the problem]

[0009] To solve these problems, a system for predicting hereditary diseases and discovering genetic variants related to disease induction according to an embodiment of the present invention includes a multiple instance learning model unit that uses input data to learn instances, which are information about a patient's genetic variants, and bags of the instances, and processes bag labels to determine whether the patient's disease is a hereditary disease caused by a genetic variant, thereby predicting a patient's hereditary disease and discovering genetic variants related to disease induction.

[0010] The system may include an input data processing unit that uses an attention mechanism to generate an attention weight, which is the degree to which the instance contributed to predicting the patient's genetic disease, and processes the input data by reflecting the attention weight for the instance.

[0011] The input data processing unit may include a genetic variant information embedding unit that embeds each of the instances with a low-dimensional vector of the same dimension using each neural network, and then projects the instances onto a manifold using a matrix of the same weight value to obtain the same embedding vector; and a genetic variant information pooling unit that generates attention weights for the embedding vectors using an attention mechanism and performs a pooling process that treats the embedding vectors as one.

[0012] If the count is equal to or exceeds a predetermined standard, the patient's disease is determined to be a genetic disease caused by a genetic variant, and in this case, a disease and induction-related genetic variant discrimination unit may be included that uses the attention weight of the instance to identify genetic variants involved in disease induction that cause the patient's disease.

[0013] The multiple instance learning model may be a multi-input model that uses input data with various vector sizes.

[0014] An instance label for the instance may be generated using the attention weight of the instance, and the multiple instance learning model may be retrained using the generated instance label.

[0015] A method for predicting hereditary diseases and discovering genetic variants associated with disease induction according to an embodiment of the present invention includes an input data processing step in which an input data processing unit uses instances, which are information about a patient's genetic variants, and a bag of the instances as input data, generates attention weights for the instances using an attention mechanism, and processes the input data; a hereditary disease presence / absence prediction step in which a multiple instance learning model unit predicts whether the patient's disease is a hereditary disease using a multiple instance learning model; and a disease- and induction-related genetic variant detection step in which, if the patient's disease is determined to be a hereditary disease, the disease- and induction-related genetic variant discrimination unit uses the attention weights for the instances to discover genetic variants associated with disease induction that cause the patient's disease.

[0016] If the patient's disease is determined to be a genetic disease, the method may further include a re-learning step of generating instance labels using the attention weights of the instances and re-learning the multiple instance learning model using the generated instance labels. [Effects of the Invention]

[0017] The present invention uses a multiple instance learning model to determine whether or not a patient has a genetic disease, even without the patient's genetic variant labels (instance labels).

[0018] Furthermore, the present invention can mine genetic variants involved in the induction of disease in patients by using attention weights for the genetic variants of the patient generated using an attention mechanism.

[0019] Furthermore, the present invention can simultaneously determine whether a patient has a hereditary disease and the genetic variants involved in disease induction by using an attention mechanism and a multiple instance learning model, even without the patient's genetic variant labels (instance labels).

[0020] Furthermore, the present invention can improve the performance of a multiple instance learning model by using attention weights to generate genetic variant labels (instance labels) for the patient and retraining the multiple instance learning model using the generated instance labels. [Brief explanation of the drawings]

[0021] [Figure 1] 1 is a diagram illustrating the configuration of a system for predicting hereditary diseases and discovering genetic variants related to disease induction, input data, and output data according to one embodiment of the present invention. [Figure 2] 1 is a diagram illustrating the configuration of a system for predicting hereditary diseases and discovering genetic variants related to disease induction, input data, and output data according to one embodiment of the present invention. [Figure 3] 1 is a block diagram of a system for predicting hereditary diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention. [Figure 4] 1 is a diagram illustrating the process of performing a system for predicting hereditary diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention. [Figure 5] FIG. 10 is a diagram showing the results of a test to discover genetic variants involved in disease induction using a system for predicting hereditary diseases and discovering genetic variants involved in disease induction according to one embodiment of the present invention. [Figure 6] 1 is a flowchart illustrating a method for predicting genetic diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] It should be understood that the terms "comprise" or "have" used herein do not preclude the presence or possible addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0023] As used herein, the singular includes the plural unless the context specifically states otherwise. For example, although an instance is described in the singular, it also includes the plural.

[0024] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention, which are devised to solve the above-mentioned problems, will be described in detail below with reference to the accompanying drawings.

[0025] 1 and 2 are diagrams showing the configuration of a system for predicting hereditary diseases and discovering genetic variants related to disease induction, input data, and output data according to one embodiment of the present invention.

[0026] Referring to Figures 1 and 2, a system 1000 for predicting genetic diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention can learn from input data 100 and generate output data 200.

[0027] The system 1000 for predicting genetic diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention can use a multiple instance learning model and an attention mechanism.

[0028] The multiple instance learning model is a type of guided learning that deals with the classification of a bag when many data points (=instances) are organized into one bag.

[0029] The system 1000 for predicting hereditary diseases and discovering genetic variants related to disease induction according to the present invention uses information on each genetic variant possessed by a patient as an instance, learns a bag of instances using input data 100, and generates a bag label indicating whether the patient's disease is a hereditary disease caused by a genetic variant, and outputs the bag label as output data 200. In this case, the multiple instance learning model may be a multi-input model that uses input data of various vector sizes.

[0030] As an example, the input data 100 may instantiate genetic variants carried by a patient, and may be a bag of various types of mutations (e.g., SNVs, SVs, CNVs, etc.).

[0031] A single nucleotide variant (SNV) is a single base change in the DNA sequence, where one base is replaced with another. For example, if a C changes to a T, this is known as a C-to-T mutation or SNP (Single Nucleotide Polymorphism).

[0032] A structural variant (SV) is a major structural change within a gene. This structural change typically occurs when two DNA base sequences are moved, deleted, duplicated, or inverted. This change can have a significant impact on the DNA sequence.

[0033] Copy number variants (CNVs) are the occurrence of two or more copies of a specific DNA sequence. CNVs are known to be a genetic cause of human diseases.

[0034] In this case, the system 1000 for predicting hereditary diseases and discovering genetic variants associated with disease induction according to the present invention can determine whether a patient's disease is a hereditary disease caused by a genetic variant using a multiple instance learning model, and can discover genetic variants associated with disease induction using an attention mechanism. That is, the output data 200 can be a prediction result of whether a patient's disease is a hereditary disease caused by a genetic variant, and a genetic variant associated with disease induction among various genetic variants possessed by the patient.

[0035] JPEG0007763431000001.jpgIt can be expressed as a value between 7170 and 1, and if it is above a pre-established standard value, the patient's disease can be predicted as a hereditary disease caused by a genetic variant.

[0036] As will be described later, attention weights for each genetic variant generated using an attention mechanism can be used to predict disease-causing genetic variants from among the many genetic variants possessed by a patient.

[0037] Figure 3 is a configuration diagram of a system for predicting hereditary diseases and discovering genetic variants related to disease induction in one embodiment of the present invention, and Figure 4 is a diagram for explaining the process of performing a system for predicting hereditary diseases and discovering genetic variants related to disease induction in one embodiment of the present invention.

[0038] Referring to Figures 3 and 4, a system 1000 for predicting hereditary diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention includes an input data processing unit 1100, a multiple instance learning model unit 1200, and a genetic variant discrimination unit 1300 related to disease and induction.

[0039] The input data processing unit 1100 uses an attention mechanism to generate an attention weight, which is the degree to which the instance contributed to the prediction of the patient's genetic disease, and can process the input data by reflecting the attention weight for each instance.

[0040] The attention mechanism is a processing method that allows deep learning models to learn and predict by focusing on important parts of the input data. In existing deep learning, all parts of the input data were weighted equally, making it difficult to recognize even parts of the input data that contained important information. By using the attention mechanism, the model learns to find important parts of the input data and calculates the importance by multiplying the input by this importance, allowing the model to better recognize important information.

[0041] The input data processing unit 1100 includes a genetic variant information embedding unit 1110 and a genetic variant information pooling unit 1130 .

[0042] The genetic variant information embedding unit 1110 embeds each instance with a low-dimensional vector of the same dimension using each neural network, and then projects it onto a manifold using a weight matrix and an activation function to obtain the same embedding vector.

[0043] The genetic variant information pooling unit 1130 can generate attention weights for the embedding vectors using an attention mechanism and perform a pooling process that treats the embedding vectors as one.

[0044] For example, if two types of mutations, SNV and SV, are transmitted as input values, a set of mutations (Bag) can be defined by the expression shown in Equation 1, and the index can be defined by the expression shown in Equation 2. JPEG0007763431000002.jpg has 10,170 feature values.

[0045] [Formula 1] JPEG0007763431000003.jpg6170

[0046] [Formula 2] JPEG0007763431000004.jpg30170

[0047] The genetic variant information embedding unit 1110 can embed SNV and SV mutations into low-dimensional vectors of the same dimension using respective neural networks, as shown in the following equations 3 and 4, and then project them onto a single manifold using a matrix with the same weights.

[0048] [Formula 3] JPEG0007763431000005.jpg18170

[0049] [Formula 4] JPEG0007763431000006.jpg18170

[0050] JPEG0007763431000007.jpg17170

[0051] JPEG0007763431000008.jpg16170

[0052] The genetic variant information pooling unit 1130 calculates the attention weight for the embedding vector calculated for each mutation, and then temporarily stacks each vector. The attention weight calculated in this case is regarded as the importance of each mutation, and mutations with a large importance value can be interpreted as genetic variants related to disease induction.

[0053] The genetic variant information pooling unit 1130 can calculate the importance of the embedding vector calculated for each individual variant by using an attention mechanism to calculate an attention weight for each variant. As shown in Equation 5, the embedding vector is passed through a two-layer neural network, and then passed through a softmax function as shown in Equation 6 to calculate the attention weight.

[0054] [Formula 5] JPEG0007763431000009.jpg29170

[0055] [Formula 6] JPEG0007763431000010.jpg19170

[0056] where e is a natural constant.

[0057] As shown in Equation 7 below, the embedding vector obtained from Equation 4 is converted into a row vector, and by dot-product-ing each column vector with the obtained attention weight, the embedding vector aggregated for the mutation axis can be obtained.

[0058] [Formula 7] JPEG0007763431000011.jpg19170

[0059] JPEG0007763431000012.jpg9170

[0060] The multiple instance learning model unit 1200 can predict the likelihood that a patient actually has a genetic disease by passing the aggregated embedding vector through a single-layer neural network, as shown in Equation 8 below.

[0061] [Formula 8] JPEG0007763431000013.jpg26170

[0062] The disease and induction-related genetic variant discrimination unit 1300 determines that the patient's disease is a hereditary disease caused by a genetic variant if the result is above a pre-set standard, and in this case, can use the attention weight value of the instance to discover genetic variants related to disease induction that cause the patient's disease.

[0063] The disease- and induction-related genetic variant discrimination unit 1300 can classify genetic variants based on a contribution value obtained by calculating the inner product of the attention weight calculated from Equation 6 and the incohered genetic variant information. After this, the previously determined top-ranked genetic variants can be considered as genetic variants related to disease induction.

[0064] FIG. 5 is a diagram showing the results of a test to discover genetic variants involved in disease induction using a system for predicting hereditary diseases and discovering genetic variants involved in disease induction according to one embodiment of the present invention.

[0065] Figure 5 shows a visualization of the contribution of each mutation in a patient with a total of 157 mutations, including 100 single nucleotide variations (SNVs) and 57 structural variations (SVs), predicted using the system for predicting hereditary diseases and discovering genetic variants related to disease induction according to the present invention.

[0066] The x-axis in Figure 5 represents the SNV mutation index from 0 to 99, and the SV mutation index from 100 to 157. The likelihood of a genetic disease was predicted with a model confidence of 0.6, and it can be interpreted that the 128th mutation had the highest contribution. Since the system for predicting genetic diseases and discovering genetic variants associated with disease induction according to the present invention focused on the 128th mutation as a genetic variant associated with disease induction and obtained a value of 0.6, this can be interpreted as a genetic variant associated with disease induction.

[0067] In this way, the system 1000 for predicting hereditary diseases and discovering genetic variants related to disease induction according to the present invention can determine whether or not a patient has a hereditary disease by using a multiple instance learning model, even if the patient's genetic variant label (instance label) is not available.

[0068] Furthermore, the attention weights for the patient's genetic variants generated using the attention mechanism can be used to discover genetic variants that are involved in the induction of disease in the patient.

[0069] In other words, the system 1000 for predicting hereditary diseases and discovering genetic variants related to disease induction according to the present invention can simultaneously determine whether a patient has a hereditary disease and the genetic variants related to disease induction by using an attention mechanism and a multiple instance learning model, even if the patient's genetic variant labels (instance labels) are not available.

[0070] In addition, the system 1000 for predicting hereditary diseases and discovering genetic variants related to disease induction according to the present invention can improve the performance of the multiple instance learning model by generating genetic variant labels (instance labels) for the patient using the determined attention weights and re-training the multiple instance learning model using the generated instance labels.

[0071] FIG. 6 is a flowchart illustrating a method for predicting hereditary diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention.

[0072] A method for predicting hereditary diseases and discovering genetic variants related to disease induction according to one embodiment of the present invention includes an input data processing step (S10), a step of predicting the presence or absence of a hereditary disease (S20), a step of discovering genetic variants related to disease induction (S30), and a re-learning step (S40).

[0073] In the input data processing step (S10), the input data processing unit takes an instance, which is information about a patient's genetic variants, and a bag of the instances as input data, and uses an attention mechanism to generate attention weights for the instances to process the input data.

[0074] In the step of predicting the presence or absence of a hereditary disease (S20), the multiple instance learning model unit can predict whether or not the patient's disease is a hereditary disease using the multiple instance learning model.

[0075] The step of discovering genetic variants related to disease induction (S30) can discover genetic variants related to disease induction that cause the patient's disease by using the attention weighting value for the instance if the patient's disease is determined to be a hereditary disease by the disease and induction-related genetic variant discrimination unit.

[0076] In this case, if the patient's disease is not determined to be a hereditary disease, the discovery of genetic variants involved in the induction of the disease may not be carried out.

[0077] In the retraining step (S40), if the patient's disease is determined to be a hereditary disease, the instance label is generated using the attention weight of the instance, and the multiple instance learning model can be retrained using the generated instance label. By retraining the multiple instance learning model according to the method for predicting hereditary diseases and discovering genetic variants related to disease induction according to an embodiment of the present invention, the performance of the multiple instance learning model can be improved.

[0078] Although exemplary device configurations are described in the specification and drawings, the functional operations and subject matter described herein may be embodied in other types of digital electronic circuitry, computer software, firmware, or hardware, including the structures disclosed herein and their structural equivalents, or a combination of any one or more of these. The subject matter described herein may also be embodied as one or more computer program products, i.e., one or more modules of computer program instructions encoded on a program storage medium of any type, for controlling the operation of an apparatus according to the methods described above or for execution thereby. The computer-readable medium may be a machine-readable storage device, a machine-readable storage substrate, a memory device, or a combination of any one or more of these.

[0079] As described above, the present invention is not limited to the above-described embodiments and accompanying drawings, and it is apparent to those skilled in the art to which the present invention pertains that various substitutions, modifications, and alterations are possible within the scope of the technical idea of ​​the present invention. [Explanation of symbols]

[0080] 100 input data 200 output data 1000 Genetic disease prediction and disease-causing genetic variant discovery system 1100 Input Data Processing Department 1110 Genetic variant information embedding section 1120 Genetic Variant Information Pooling Department 1200 Multiple Instance Learning Model 1300 Disease and Induction Related Genetic Variant Identification Division

Claims

1. Regarding data that compiles multiple instances of patient genetic variant information into one bag, Using an attention mechanism, an attention weight, which is the degree to which the instance contributed to the prediction of the patient's genetic disease, is generated for each of the plurality of instances, and the attention weight is reflected for each of the plurality of instances, thereby an input data processing unit that processes the data collected as the bag; The data processed by the input data processing unit and collected as the bag is used as input data, outputting a bag label indicating whether the patient's disease is a hereditary disease caused by the genetic variant; a multiple instance learning model unit that inputs the data processed by the input data processing unit into a trained multiple instance learning model and outputs the bag label from the multiple instance learning model, and that, after the output is determined to be a genetic disease, identifies a genetic variant that induces the disease in the patient using the attention weights for each of the multiple instances; A system for predicting genetic diseases and discovering genetic variants involved in disease induction, including:

2. The input data processing unit a genetic variant information embedding unit that embeds each of the plurality of instances into a low-dimensional vector of the same dimension using each neural network, and then projects the embedding onto a manifold using a weight matrix and an activation function to obtain the same embedding vector; a genetic variant information pooling unit that generates attention weights for the embedding vectors using an attention mechanism and performs a pooling process that treats the embedding vectors as one; The system for predicting genetic diseases and discovering genetic variants involved in disease induction according to claim 1, comprising:

3. the bag label indicates a value to indicate whether the patient's disease is a hereditary disease caused by the genetic variant; a disease and induction-related genetic variant discrimination unit that determines that the patient's disease is a genetic disease caused by a genetic variant if the value indicated by the output bag label is equal to or greater than a predetermined standard, and after determining that the patient's disease is a genetic disease caused by a genetic variant, uses the attention weights for each of the multiple instances to discover genetic variants involved in disease induction that cause the patient's disease; The system for predicting genetic diseases and discovering genetic variants involved in disease induction according to claim 2, further comprising:

4. The multiple instance learning model is a multi-input model that uses input data with various vector sizes. The system for predicting genetic diseases and discovering genetic variants involved in disease induction according to claim 1.

5. generating instance labels for each of the plurality of instances using the attention weights of the respective plurality of instances; and retraining the multiple instance learning model using the generated instance labels. The system for predicting hereditary diseases and discovering genetic variants involved in disease induction according to claim 3.

6. The information processing device collects data in which multiple instances of information on the patient's genetic variants are organized into a single bag, Using an attention mechanism, an attention weight, which is the degree to which the instance contributed to the prediction of the patient's genetic disease, is generated for each of the plurality of instances, and the attention weight is reflected for each of the plurality of instances, thereby an input data processing step of processing the data collected as the bag; The information processing device, The data collected as the bag processed in the input data processing step is used as input data, outputting a bag label indicating by a value whether the patient's disease is a hereditary disease caused by the genetic variant; a hereditary disease discrimination step of inputting the data compiled as bags processed in the input data processing step into a trained multiple instance learning model and outputting the bag label from the multiple instance learning model, in which the information processing device discriminates that the patient's disease is a hereditary disease caused by a genetic variant if the value indicated by the output bag label is equal to or greater than a pre-set standard; a step of discovering genetic variants involved in disease induction, in which the information processing device, after determining that the patient's disease is a genetic disease caused by a genetic variant in the genetic disease determination step, discovers genetic variants involved in disease induction that induce the patient's disease using the attention weights for each of the plurality of instances; A method for predicting genetic diseases and discovering genetic variants involved in disease induction, including:

7. a re-learning step in which, after the information processing device has determined that the patient's disease is a genetic disease caused by a genetic variant in the genetic disease determination step, it generates instance labels using attention weights of each of the plurality of instances and re-learns the multiple instance learning model using the generated instance labels. The method for predicting genetic diseases and discovering genetic variants involved in disease induction according to claim 6, further comprising:

Images