WO2022203093A1 - Method for diagnosing or predicting cancer occurrence - Google Patents

Abstract

The present invention relates to a method for diagnosing or predicting cancer and, specifically, to a method for diagnosing or predicting cancer, comprising steps in which: first analysis data on a cell-free DNA (cfDNA) concentration, second analysis data on copy number variation (CNV), and third analysis data on tumor marker expression amount are acquired; and a prediction unit uses a machine learning model trained to calculate a cancer occurrence probability through computing of the analysis data, so as to analyze the analysis data, and thus diagnose or predict cancer occurrence

Description

Methods for diagnosing or predicting the occurrence of cancer

The present invention relates to a method for diagnosing or predicting cancer, and more particularly, first analysis data for cfDNA (cell-free DNA) concentration, second analysis data for copy number variation (CNV), and tumor obtaining third analysis data on the expression level of the marker; and diagnosing or predicting whether cancer occurs by analyzing the analysis data using a machine learning model trained to calculate a cancer occurrence probability by a prediction unit calculating a cancer occurrence probability through an operation on the analysis data; is about

Cancer is the second leading cause of death in the world and is a fatal disease for humans.

Currently, the standard method for diagnosing cancer is a tissue biopsy, which is a method of diagnosing cancer by collecting tumor tissue using an endoscope or an injection needle. However, since the tissue biopsy has to use an invasive method, it is burdensome for doctors and patients, it takes a lot of time, and even if it is a tumor tissue, the biological characteristics may appear differently depending on the location from which it is collected, so the accuracy of information may be a problem. have.

The liquid biopsy designed to overcome the problems of the tissue biopsy is to non-invasively diagnose cancer or various diseases using blood, urine, spinal fluid, and the like. In 1869, Australian physician Thomas Ashworth reported that circulating tumor cells (CTCs, cancer tumor cells) in the blood were associated with cancer metastasis. It is attracting attention as a new alternative that can replace invasive diagnosis with blood sampling.

Among liquid biopsies, the most active research is a blood biopsy, which uses blood to diagnose mutant genes in cancer. In Korea, a blood biopsy for EGFR gene mutation is available, and it is currently the most active in lung cancer.

However, in the case of liquid biopsy, there is a difference in the degree of detection of cells in the blood depending on the carcinoma, the expression rate of mutations is different, and certain biomarkers are not expressed at all for a specific cancer, so the accuracy of diagnosis or cancer biomarkers is not high. Uncertainty has emerged as a problem.

On the other hand, multi-omics analysis is performed at various molecular levels such as genome, transcriptome, proteome, metabolome, epigenome, and lipodome. It means a holistic and integrated analysis of a large number of data generated in

Such multiomics analysis has become possible with the development of information processing capabilities according to the development of high-speed molecular biological analysis technology and the development of the computer industry, and the information that can be provided due to the comprehensive analysis of various data is compared with a single analysis. Because there are many, it can make a great contribution to the diagnosis and treatment of diseases caused by complex causes such as cancer.

[Prior art literature]

[Patent Literature]

(Patent Document 0001) Korean Patent Publication No. 10-2012-0077568

[Non-patent literature]

(Non-Patent Document 0001) Heidi Schwarzenbach et al., Cell-free nucleic acids as biomarkers in cancer patients

The present inventors have recognized the problems of the prior art and have tried to develop a method for diagnosing or predicting cancer with high accuracy. As a result, it has been demonstrated that high sensitivity and specificity can be achieved in cancer diagnosis or prediction when using a multiomics analysis method that performs integrated analysis of specific data with a machine learning algorithm, and thus the present invention has been completed.

An object of the present invention is to provide a method for diagnosing or predicting cancer with high sensitivity and specificity.

For the above purpose, the present invention provides first analysis data for cfDNA (cell-free DNA) concentration, second analysis data for copy number variation (CNV), and third analysis data for tumor marker expression level obtaining a; and diagnosing or predicting whether or not cancer occurs by analyzing the analysis data using a machine learning model trained to calculate the cancer occurrence probability by the prediction unit calculating the cancer occurrence probability through operation on the analysis data. provides

The first analysis data may include: extracting blood from a subject; and measuring the amount of cfDNA detected per 1 ml of the extracted blood.

The second analysis data may include: a) inputting sequence information of test nucleotide sequence fragments of cfDNA (cell-free DNA) from blood isolated from a subject; b) arranging homology positions by comparing the sequence information of the test nucleotide sequence fragments with a human reference genome database; c) segmenting a target chromosome among the test nucleotide sequence fragments to a predetermined size and increasing the segment size to generate a normalized two-dimensional matrix in which rows and columns represent segment sizes and positions, respectively; d) forming a two-dimensional matrix of Z-score values by calculating a Z-score value of the generated two-dimensional matrix; e) selecting the lowest Z-score value from among the Z-score values lower than the Z-score value calculated from the target chromosome of the reference genome sequence fragment; and f) gene copy number variation (CNV) by checking the position of the segment in which the Z-score value gradually increases from the row and column to which the lowest Z-score value belongs to the row with the smallest segment size. It may be obtained by a method comprising a; confirming the location and size of the occurrence.

Step a) may include: a-1) separating plasma by centrifuging blood collected from a subject; a-2) extracting cfDNA (cell-free DNA) from the separated plasma; a-3) preparing a library using the extracted cfDNA; and a-4) pooling the library and then decoding the nucleotide sequence by next generation sequencing (NGS).

Step c) may be to repeatedly perform steps c)-n to generate an n-th row by segmenting the chromosome to a size of 0.5 Mb * n;

Z-score values for all segment sizes of the two-dimensional matrix in step d) may be calculated using mean and standard deviation (SD) values calculated by normalizing samples of a control group having normal chromosomes. .

In step e), the portion showing the lowest Z-score value is selected among the values having the Z-score value lower than the reference value, and in step f), the segment size is the highest from the row and column to which the lowest Z-score value belongs. It may be to check the copy number variation by checking the position of the segment where the Z-score value gradually increases up to small rows.

The Z-score value is calculated according to Equation 1 below using the mean and standard deviation (SD) of the position (xy) of the two-dimensional matrix calculated from the target chromosome of the reference genome sequence fragment. it may be

[Equation 1]

(where Z=Z-score, M=matrix, cor.gc=correctable gc percent, xy=location of matrix)

The third analysis data may include: obtaining a biological sample from the subject; and measuring the concentration of the tumor marker in the biological sample.

The tumor marker may be Cyfra 21-1, CA 15-3, AFP, CEA and CA 19-9.

The biological sample may be isolated from blood, plasma or serum.

In the diagnosing or predicting step, the machine learning model performs a plurality of calculations to which a weight between a plurality of layers of the machine learning model is applied to the analysis data to calculate an output value indicating the probability of cancer cancer in the test target patient step; and estimating, by the prediction unit, whether cancer occurs according to the output value.

Before the acquiring step, the learning unit preparing the learning data, which is the analysis data of the patient for which cancer is known; setting, by the learning unit, a label according to whether or not cancer occurs in the learning data; calculating, by the machine learning model before the learning is completed, an output value indicating the probability of cancer occurrence by performing a plurality of calculations to which a weight between a plurality of layers of the machine learning model is applied to the learning data; calculating, by the learning unit, a loss that is a difference between an output value and a label; and performing, by the learning unit, an optimization of correcting the weight w of the machine learning model (MLM) using a backpropagation algorithm so that the loss is minimized.

Calculating the loss is a loss function by the learning unit

Calculate the loss according to , where L means loss, i is an index corresponding to the training data, x _i is the training data that is an input to the machine learning model, and y _i is the expected value for the training data. , y _i is a label corresponding to the i-th training data, and f (x _i ) is an output value calculated by the machine learning model with respect to the i-th training data.

In the diagnosis or prediction step, the machine learning model may be formed through a multilayer perceptron (MLP).

The machine learning model may include an input layer (IL), a plurality of hidden layers (HL1 to HLk), and an output layer (OL).

Each of the plurality of layers may include at least one node.

The plurality of nodes of the plurality of layers may all have an operation, and the operation may be performed through an activation function (F).

The cancer may be one or more selected from the group consisting of lung cancer, head and neck cancer, breast cancer, ovarian cancer, liver cancer, testicular cancer, colorectal cancer, thyroid cancer, pancreatic cancer, cervical cancer, bladder cancer, digestive cancer and gallbladder cancer.

The method for diagnosing or predicting cancer of the present invention can reduce the possibility of false positives and false negatives, and has excellent sensitivity and specificity, so that it can be usefully used for diagnosis or prediction of cancer.

1 is a diagram for explaining the configuration of a prediction apparatus for predicting cancer according to an embodiment of the present invention.

2 is a diagram for explaining a schematic configuration of a machine learning model (MLM) executed in a predicting apparatus for predicting cancer according to an embodiment of the present invention.

3 is a view for explaining a detailed configuration of a machine learning model (MLM) formed through a multilayer perceptron (MLP) according to an embodiment of the present invention.

4 is a diagram for explaining the configuration of any one node of a machine learning model (MLM) according to an embodiment of the present invention.

5 is a diagram for explaining a learning method of a machine learning model (MLM) according to an embodiment of the present invention.

6 is a flowchart illustrating a method for predicting cancer using a machine learning model (MLM) according to an embodiment of the present invention.

7 is a view showing the difference in sensitivity of cancer diagnosis between the method according to an embodiment of the present invention, when using CNV single data, and when using Grail's liquid biopsy.

Prior to the detailed description of the present invention, the terms or words used in the present specification and claims described below should not be construed as being limited to their ordinary or dictionary meanings, and the inventors should develop their own inventions in the best way. It should be interpreted as meaning and concept consistent with the technical idea of the present invention based on the principle that it can be appropriately defined as a concept of a term for explanation. Accordingly, the embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all the technical ideas of the present invention, so various equivalents that can replace them at the time of the present application It should be understood that there may be water and variations.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this case, it should be noted that the same components in the accompanying drawings are denoted by the same reference numerals as much as possible. In addition, detailed descriptions of well-known functions and configurations that may obscure the gist of the present invention will be omitted. For the same reason, some components are exaggerated, omitted, or schematically illustrated in the accompanying drawings, and the size of each component does not fully reflect the actual size.

The present invention provides a method for diagnosing or predicting cancer comprising the following steps.

acquiring first analysis data for cell-free DNA (cfDNA) concentration, second analysis data for copy number variation (CNV), and third analysis data for tumor marker expression level; and

Diagnosing or predicting whether or not cancer occurs by analyzing the analysis data using a machine learning model trained to calculate a cancer occurrence probability by a prediction unit calculating a cancer occurrence probability through an operation on the analysis data.

In the method for diagnosing or predicting cancer of the present invention, the first analysis data is data obtained by measuring a cell-free DNA (cfDNA) concentration.

In the present invention, “cfDNA (cell-free DNA)” refers to a small DNA fragment floating outside the cell and in various body fluids including blood due to cell death. cfDNA is present. Here, cfDNA derived from cancer cells among cfDNA is referred to as circulating tumor DNA (ctDNA), and as the cancer tissue in a cancer patient grows, the amount of ctDNA increases, and the amount of cfDNA tends to increase.

In the present invention, the first analysis data is

extracting blood from the subject; and

measuring the amount of cfDNA detected per ml of the extracted blood;

It may be obtained by a method including, but is not limited to, a known method capable of measuring the concentration of cfDNA, or a commercially available kit, or may be obtained by requesting a measurable related institution.

In the present invention, the first analysis data result may indicate a high or low cfDNA concentration based on a cfDNA concentration of 7.4 ng/ml, but is not limited thereto.

In the method for diagnosing or predicting cancer of the present invention, the second analysis data is data on (or confirming) copy number variation (CNV).

In the present invention, the second analysis data may be obtained by a non-invasive method including the following steps.

a) inputting sequence information of test nucleotide sequence fragments of cfDNA (cell-free DNA) from blood isolated from a subject;

b) arranging homology positions by comparing the sequence information of the test nucleotide sequence fragments with a human reference genome database;

c) segmenting a target chromosome among the test nucleotide sequence fragments to a predetermined size and increasing the segment size to generate a normalized two-dimensional matrix in which rows and columns represent segment sizes and positions, respectively;

d) forming a two-dimensional matrix of Z-score values by calculating a Z-score value of the generated two-dimensional matrix;

e) selecting the lowest Z-score value from among the Z-score values lower than the Z-score value calculated from the target chromosome of the reference genome sequence fragment; and

f) Gene copy number variation (CNV) is determined by checking the position of the segment where the Z-score value gradually increases from the row and column to which the lowest Z-score value belongs to the row with the smallest segment size. Steps to ascertain where and where it happened.

In the present invention, “Copy Number Variation (CNV)” is a genetic change corresponding to a structural variation of the human genome, and unlike general sequences that exist in the form of 2n, deletion (0n, 1n state), amplification (3n It refers to a DNA fragment that shows a difference in the number of repeating sequences compared to the human standard reference genome (reference genome).

In the present invention, "subject" includes any human or non-human animal. The term “non-human animal” can be a vertebrate, such as non-human primates, sheep, dogs, and rodents, such as mice, rats, and guinea pigs. The subject may preferably be a human, and specifically may be a human possessing or expected to have tumor cells. In the present invention, the term “subject” may be used interchangeably with “individual” or “patient”.

In the present invention, the step a) is,

a-1) centrifuging the blood collected from the subject to separate plasma;

a-2) extracting cfDNA (cell-free DNA) from the separated plasma;

a-3) preparing a library using the extracted cfDNA; and

a-4) pooling the library and then decoding the nucleotide sequence by next generation sequencing (NGS).

Step a-4) divides genomic DNA into countless pieces by massive parallel sequencing, reads each piece at the same time to obtain nucleotide sequence data, and uses various bioinformatics techniques to obtain the fragments. It includes all analysis methods that decipher the genome by combining the nucleotide sequence data of

The base sequence may be, for example, Roche 454 (ie, Roche 454 GS FLX), Applied Biosystems' SOLiD system (ie, SOLiDv4), Illumina's GAIIx, HiSeq 2500 and MiSeq sequences Analyzer, Life Technologies' Ion Torrent semiconductor sequencing platform, Pacific Biosciences' PacBio RS, and Sanger's 3730xl. It may be obtained by Life Technologies' IonTorrent platform or Illumina's MiSeq, and more preferably, Life Technologies' IonTorrent Personal Genome Machine (IonTorrent PGM). have.

In the present invention, the chromosome of step c) is an autosomal, and may be one or more chromosomes selected from autosomes consisting of chromosomes 1 to 22.

In the present invention, step c) may be to repeatedly perform steps c)-n to generate an n-th row by segmenting the chromosome to a size of 0.5 Mb * n; where n is 1 to 24.

Specifically, the first row is generated by segmenting the chromosome to a size of 0.5 Mb, and the second row is generated by segmenting the chromosome to a size of 1 Mb, which is 0.5 Mb * 2, and in this way, the segment size is increased by 0.5 Mb. It could be adding rows to create a two-dimensional matrix where rows and columns represent segment sizes and positions. For example, by increasing the length of one chromosome from 0.5 Mb to 12 Mb at 0.5 Mb intervals to generate a two-dimensional matrix including segments according to each length, one matrix contains segments composed of only one length. exists, and a total of 24 matrices composed of segments each having a length can be generated (the number of copies by checking the Z-score step by step in a two-dimensional matrix indicating the size and position of the generated segment) (also referred to herein as a stair-matrix to identify a mutation).

In the present invention, the Z-score values for all segment sizes of the two-dimensional matrix in step d) are calculated using the mean and standard deviation (SD) values calculated by normalizing the samples of the control group having normal chromosomes. characterized in that

The Z-score value is calculated according to Equation 1 below using the mean and standard deviation (SD) of the position (xy) of the two-dimensional matrix calculated from the target chromosome of the reference genome sequence fragment. do.

[Equation 1]

(where Z=Z-score, M=matrix, cor.gc=correctable gc percent, xy=location of matrix)

In the present invention, step e) selects a portion showing the lowest Z-score value among values lower than the reference value, and step f) is segmented from the row and column to which the lowest Z-score value belongs. Copy number variation is confirmed by checking the location of the segment where the Z-score value gradually increases up to the smallest row.

In the present invention, the Z-score value calculated from the target chromosome of the reference genome sequence fragment may also be defined as a “reference value” and refers to a Z-score value obtained from a sample determined to be normal.

In the present invention, the lowest Z-score value in step e) may be determined as the segment having the darkest shade has the lowest Z-score value by displaying the chromosome as a heat map.

In the method for diagnosing or predicting cancer of the present invention, the third analysis data is data on the expression level of a tumor marker.

The third analysis data in the present invention is,

obtaining a biological sample; and

measuring a concentration of a tumor marker in the biological sample;

It may be obtained by a method comprising a it could be

In the present invention, the term “tumor marker” refers to a proteinaceous material produced in cancer cells in response to cancer growth or in normal tissues around it in response to cancer tissue and is a biomarker detected in blood, urine, or tissue samples. . Although cancer is diagnosed through some established tumor markers in clinical practice, most tumor markers can be elevated even in non-tumor conditions, so specific tumor markers alone are not suitable for diagnosis of cancer.

In the present invention, the tumor markers are Cyfra 21-1, CA 15-3 (cancer antigen 15-3), AFP (alpha-fetoprotein), CEA (onco-embryonic antigen), CA 19-9 (cancer antigen 19-9) ), B2M (beta-2 microglobulin), CA-125 (cancer antigen 125), calcitonin, Chromogranin A (CgA), hCG (chorionic gonadotropin), monoclonal immunoglobulin, PSA (prostate specific antigen), thyroid gland It may be at least one selected from the group consisting of globulin, and preferably, the tumor marker may be at least one selected from the group consisting of Cyfra 21-1, CA 15-3, AFP, CEA and CA 19-9. According to an embodiment of the present invention, the tumor marker may be a tumor marker group consisting of Cyfra 21-1, CA 15-3, AFP, CEA and CA 19-9, but is not limited thereto.

In the present invention, the result of the third analysis data may indicate a low risk or a high risk level according to the following criteria, but is not limited thereto.

- Cyfra 21-1:≤3.30 ng/ml

- CA 15-3:≤26.40 U/ml

- AFP:≤7.0 ng/ml

- CEA: non-smoking ≤3.8, smoking ≤5.5 ng/ml

- CA 19-9:≤34.00 U/ml

In the present invention, the biological sample may be a body odor material isolated from a subject's body, and may preferably be blood, plasma or serum due to the nature of the present invention using a non-invasive method.

Next, a step of diagnosing or predicting whether or not cancer occurs by analyzing the analysis data using a machine learning model trained to calculate the cancer occurrence probability by the prediction unit operating on the analysis data will be described below.

1 is a diagram for explaining the configuration of a prediction apparatus for predicting cancer according to an embodiment of the present invention. Referring to FIG. 1 , the prediction device 10 may be a device for performing a function of processing data according to a predetermined clock while temporarily storing data, and the prediction device 10 is a central processing unit (CPU). unit), a graphic processing unit (GPU), a neural processing unit (NPU), or the like. The prediction apparatus 10 includes a preprocessor 100 , a learning unit 200 , and a prediction unit 300 .

The preprocessor 100 performs preprocessing on data input to the machine learning model (MLM), that is, analysis data or learning data. These data include cfDNA concentration or input amount, copy number variation (CNV), and tumor marker (Cyfra 21-1, CA 15-3, AFP, CEA and CA 19-9) expression levels, etc. It can be data. That is, the preprocessor 100 may remove and normalize outliers of data such as the amount of cfDNA, CNV, and tumor markers.

The learning unit 200 is for learning a machine learning model (MLM). The operation of the learning unit 200 will be described in more detail below.

The prediction unit 300 calculates a cancer occurrence probability (Positive Predict Probability: 0.0 to 1.0) using the machine learning model (MLM) generated by the learning unit 200 , and according to this probability according to the calculated probability, cancer occurrence can be predicted. The operation of the prediction unit 300 will be described in more detail below.

2 is a diagram for explaining a schematic configuration of a machine learning model (MLM) executed in a prediction apparatus for predicting cancer according to an embodiment of the present invention, and FIG. 3 is a multilayer perceptron ( It is a view for explaining the detailed configuration of a machine learning model (MLM) formed through MLP: Multilayer Perceptron, and FIG. 4 shows the configuration of any one node of the machine learning model (MLM) according to an embodiment of the present invention. It is a drawing for explanation. 2 to 4 , the machine learning model (MLM) may be a multilayer perceptron (MLP). A machine learning model (MLM) includes multiple layers (IL, HL, OL). The plurality of layers includes an input layer (IL), a plurality of hidden layers (HL1 to HLk), and an output layer (OL).

In addition, each of the plurality of layers IL, HL, and OL includes at least one node. For example, as illustrated, the input layer IL may include n input nodes i1 to in, and the output layer OL may include one output node o. Also, the number of hidden layers HL may be k (HL1, HL2, ..., HLk). Among the hidden layers HL, the first hidden layer HL1 includes a number of hidden nodes h11 to h1a, the second hidden layer HL2 includes b number of hidden nodes h21 to h2b, and the kth hidden layer HLk ) may include z hidden nodes (hk1 to hkz).

All of the plurality of nodes of the plurality of layers have operations. This operation is performed through the activation function (F). In particular, a plurality of nodes of different layers are connected by a channel (indicated by a dotted line) having a weight (w: weight). In other words, the calculation result of one node is weighted and becomes the input of the next layer node.

That is, as shown in Figure 4, any one node (N) of any one layer of the machine learning model (MLM) is a weight (weight: w1) to the input (x1, x2, ... xn) from the node of the previous layer , w2, . These results (OUT) are transferred to the input of the next layer. On the other hand, an unexplained parameter b indicates a threshold, which receives a value obtained by applying a weight (weight: w1, w2, …wn) to the inputs (x1, x2, …xn), and the sum of these values is the threshold If there is no abnormality, the operation result of the corresponding node is inactivated.

Accordingly, the machine learning model (MLM) calculates an output value (0.0 to 1.0) by performing a plurality of operations to which a weight between a plurality of layers (IL, HL, OL) is applied to the input data. That is, a plurality of calculations in which weights are applied from the input layer IL to the plurality of hidden layers HL1 , HL2 , ... HLk and the output layer OL are performed on data to calculate an output value. The value of the output node o of the output layer OL becomes the output value, which indicates the probability of cancer occurrence (0.0 to 1.0). For example, if the output value of each output node o is 0.45, it indicates that the probability of cancer occurrence is 45%.

Next, a learning method of the aforementioned machine learning model (MLM) will be described. 5 is a diagram for explaining a learning method of a machine learning model (MLM) according to an embodiment of the present invention.

Referring to FIG. 5 , the learning unit 200 prepares learning data in step S110. The learning data means data for which cancer is known. Specifically, the learning data include data on the cfDNA concentration (or amount) of patients with known cancer, data confirming copy number variation (CNV), and tumor markers (Cyfra 21-1, CA 15-3, AFP) , CEA and CA 19-9) means data on expression levels. That is, the learning data includes an experimental group, which is data of a patient with cancer, and a control group, which is data of a patient who does not have cancer.

Next, the learning unit 200 sets a label on the training data in step S120. According to an embodiment, the learning unit 200 may set the label to 1 for the experimental group among the training data and set the label to 0 for the control group in the one-hot encoding method.

Subsequently, when the learning unit 200 inputs the learning data to the machine learning model (MLM), the machine learning model (MLM) performs a plurality of operations in which weights between a plurality of layers are applied to the learning data input in step S130. Thus, an output value representing the probability of cancer is calculated.

Then, in step S140 , the learning unit 200 calculates a loss through a loss function as in Equation 2 below.

[Equation 2]

In Equation 2, L means loss (L2 Loss). i is an index corresponding to the training data. f (x _i ) is the output value calculated by the machine learning model (MLM) according to the input (x _i ), and y _i is a label indicating the expected value. That is, y _i is a label corresponding to the i-th training data (x _i ). In addition, f (x _i ) is an output value calculated by the machine learning model (MLM) for the i-th training data (x _i ).

Next, the learning unit 200 performs optimization of correcting the weight w of the machine learning model (MLM) so that the loss that is the difference between the output value and the label of the machine learning model (MLM) is minimized in step S150. For this optimization, a back-propagation algorithm can be used.

Steps S110 to S150 described above are repeatedly performed using a plurality of different learning data. This iteration may be performed until the accuracy is calculated through the evaluation index, and a desired accuracy is reached.

Next, as described above, after learning of the machine learning model (MLM) is completed, cancer may be predicted using the machine learning model (MLM). These methods will be described. 6 is a flowchart illustrating a method for predicting cancer using a machine learning model (MLM) according to an embodiment of the present invention.

Referring to FIG. 6 , the preprocessor 100 analyzes the analysis data of the test target patient whose cancer is unknown in step S210 , that is, analysis data on the cfDNA concentration or amount, and the copy number variation (CNV) analysis. Data, and analysis data on the expression level of tumor markers (Cyfra 21-1, CA 15-3, AFP, CEA and CA 19-9) are input, and pretreatment is performed thereon in step S220, and then pretreatment in step S230 The analyzed data is input into a machine learning model (MLM).

Then, the machine learning model (MLM) calculates an output value indicating the probability of whether the patient to be examined has cancer by performing a plurality of calculations to which a weight between a plurality of layers is applied on the analysis data input in step S240. For example, if the output of the output node o of the output layer OL is 0.84, it means that the cancer occurrence probability is 84%.

Accordingly, the prediction unit 300 may estimate the cancer according to the output value in step S250 .

In the method for diagnosing or predicting cancer of the present invention, the cancer to be diagnosed is lung cancer, head and neck cancer, breast cancer, ovarian cancer, liver cancer, testicular cancer, colorectal cancer, thyroid cancer, pancreatic cancer, cervical cancer, bladder cancer, digestive cancer and gallbladder cancer. It may be one or more selected from.

The cancer diagnosis or prediction method of the present invention has improved sensitivity and accuracy even in a trace amount of cfDNA fraction, and according to an experimental example of the present invention, Grail liquid biopsy and As a result of confirming the occurrence of cancer with respect to the artificial sequence using the method of the present invention, it was confirmed that the method of the present invention can diagnose or predict the occurrence of cancer with a higher level of sensitivity and accuracy than Grail's liquid biopsy.

experimental example. Confirmation of sensitivity of the method for diagnosing or predicting cancer of the present invention

In order to confirm the sensitivity of the method of the present invention, 6 ctDNA fractions such as 2%, 3%, 4%, 6%, 8%, and 10% were applied based on 30 cfDNA samples of the normal group for a total of 180 samples. samples were made. The total chromosomes 13 and 18 and shorter 30M, 20M, and 10M samples were also made, respectively.

For the artificial sample, only Grail's liquid biopsy, CNV single assay (using a stair-matrix), and the cancer diagnosis or prediction method of the present invention were used (here, the second analysis data was obtained using a stair-matix) , the first analysis data was obtained by measuring and obtained by Leewon Diagnostics Co., Ltd. in Korea, and the third analysis data was obtained by requesting the Leewon Medical Foundation of Korea) to determine the sensitivity in diagnosing or predicting cancer occurrence. was confirmed, and the results are shown in FIG. 7 .

As can be seen in FIG. 7 , when the diagnosis or prediction method of the present invention was used for STAGE I early cancer, it was confirmed that the sensitivity was higher than when only Grail's liquid biopsy and CNV single assay were used.

Claims (15)

1. acquiring first analysis data for cell-free DNA (cfDNA) concentration, second analysis data for copy number variation (CNV), and third analysis data for tumor marker expression level; and

   A method of diagnosing or predicting cancer, comprising: a prediction unit diagnosing or predicting whether or not cancer occurs by analyzing the analysis data using a machine learning model trained to calculate a cancer occurrence probability through operation on the analysis data.
2. According to claim 1,

   The first analysis data is,

   extracting blood from the subject; and

   A method for diagnosing or predicting cancer, which is obtained by a method comprising; measuring the amount of cfDNA detected per ml of the extracted blood.
3. According to claim 1,

   The second analysis data is,

   a) inputting sequence information of test nucleotide sequence fragments of cfDNA (cell-free DNA) from blood isolated from a subject;

   b) arranging homology positions by comparing the sequence information of the test nucleotide sequence fragments with a human reference genome database;

   c) segmenting a target chromosome among the test nucleotide sequence fragments to a predetermined size and increasing the segment size to generate a normalized two-dimensional matrix in which rows and columns represent segment sizes and positions, respectively;

   d) forming a two-dimensional matrix of Z-score values by calculating a Z-score value of the generated two-dimensional matrix;

   e) selecting the lowest Z-score value from among the Z-score values lower than the Z-score value calculated from the target chromosome of the reference genome sequence fragment; and

   f) Gene copy number variation (CNV) is determined by checking the position of the segment where the Z-score value gradually increases from the row and column to which the lowest Z-score value belongs to the row with the smallest segment size. A method of diagnosing or predicting cancer that is obtained by a method comprising; confirming the location and size of the occurrence.
4. According to claim 1,

   Step a) is

   a-1) centrifuging the blood collected from the subject to separate plasma;

   a-2) extracting cfDNA (cell-free DNA) from the separated plasma;

   a-3) preparing a library using the extracted cfDNA; and

   a-4) pooling the library and then decoding the nucleotide sequence by next generation sequencing (NGS);
5. According to claim 1,

   Step c) is a method of diagnosing or predicting cancer, wherein n is 1 to 24 repeatedly performing; step c)-n of generating an n-th row by segmenting the chromosome to a size of 0.5 Mb * n.
6. According to claim 1,

   In step d), the Z-score values for all segment sizes of the two-dimensional matrix are calculated using the mean and standard deviation (SD) values calculated by normalizing samples of the control group having normal chromosomes. Cancer of diagnostic or predictive methods.
7. According to claim 1,

   In step e), the portion showing the lowest Z-score value is selected among the values having the Z-score value lower than the reference value, and in step f), the segment size is the highest from the row and column to which the lowest Z-score value belongs. A method for diagnosing or predicting cancer, wherein the copy number variation is confirmed by checking the location of the segment where the Z-score value gradually increases up to a small row.
8. According to claim 1,

   The Z-score value is calculated according to Equation 1 below using the mean and standard deviation (SD) of the position (xy) of the two-dimensional matrix calculated from the target chromosome of the reference genome sequence fragment. A method for diagnosing or predicting cancer.

   [Equation 1]

   

   (where Z=Z-score, M=matrix, cor.gc=correctable gc percent, xy=location of matrix)
9. According to claim 1,

   The third analysis data is

   obtaining a biological sample from the subject; and

   A method for diagnosing or predicting cancer obtained by a method comprising; measuring the concentration of a tumor marker in the biological sample.
10. 10. The method of claim 9,

    The tumor markers are Cyfra 21-1, CA 15-3, AFP, CEA and CA 19-9. A method for diagnosing or predicting cancer.
11. 10. The method of claim 9,

    The method for diagnosing or predicting cancer, wherein the biological sample is isolated from blood, plasma or serum.
12. According to claim 1,

    The step of diagnosing or predicting

    calculating, by the machine learning model, a plurality of calculations to which weights between a plurality of layers of the machine learning model are applied on the analyzed data to calculate an output value indicating the probability of cancer in a patient to be examined; and

    estimating, by the predicting unit, whether cancer occurs according to the output value;

    A method for diagnosing or predicting cancer comprising a.
13. According to claim 1,

    Before the obtaining step,

    preparing, by the learning unit, learning data, which is analysis data of a patient whose cancer occurrence is known;

    setting, by the learning unit, a label according to whether or not cancer occurs in the learning data;

    calculating, by the machine learning model before the learning is completed, an output value indicating the probability of cancer occurrence by performing a plurality of calculations to which a weight between a plurality of layers of the machine learning model is applied to the learning data;

    calculating, by the learning unit, a loss that is a difference between an output value and a label; and

    performing, by the learning unit, optimization of modifying the weight w of the machine learning model (MLM) using a backpropagation algorithm so that the loss is minimized;

    A method of diagnosing or predicting cancer further comprising a.
14. 14. The method of claim 13,

    The step of calculating the loss is

    the learning department

    loss function

    

    The loss is calculated according to

    L means loss,

    The i is an index corresponding to the training data,

    Where x _i is training data that is an input to the machine learning model,

    Wherein y _i is a label indicating an expected value for the training data,

    y _i is the label corresponding to the i-th training data,

    Wherein f (x _i ) is an output value calculated by the machine learning model with respect to the i-th learning data. A method for diagnosing or predicting cancer.
15. According to claim 1,

    The cancer is lung cancer, head and neck cancer, breast cancer, ovarian cancer, liver cancer, testicular cancer, colorectal cancer, thyroid cancer, pancreatic cancer, cervical cancer, bladder cancer, digestive cancer and gallbladder cancer at least one selected from the group consisting of a diagnosis or prediction method of cancer.

Images