Method for detecting tumor derived mutation from cell-free DNA based on artificial intelligence and Method for early diagnosis of cancer using the same

Abstract

The present invention relates to an artificial intelligence-based method for early cancer diagnosis by detecting tumor-derived mutations in cell-free DNA, more specifically, to an artificial intelligence-based method for early cancer diagnosis by detecting tumor-derived mutations in cell-free DNA, using a method of obtaining sequence information from a biological sample, comparing the sequence information with a reference genome to detect mutations, and inputting and analyzing the detected mutation information into an artificial intelligence model trained to determine whether or not the mutation is tumor-derived. The method for detecting tumor-derived mutations in cell-free DNA and the method for early cancer diagnosis using the same according to the present invention diagnoses cancer early using both functional and sequence features of cancer based on artificial intelligence using Next Generation Sequencing (NGS), and has high accuracy and sensitivity and is commercially applicable, so that the method of the present invention is useful for early cancer diagnosis.

Description

[Technical field]

[0001] The present invention relates to an artificial intelligence-based method for early cancer diagnosis by detecting tumor-derived mutations in cell-free DNA, and more specifically, to a method for early cancer diagnosis by detecting tumor-derived mutations in cell-free DNA using a method of acquiring sequence information from a biological sample, comparing the sequence information with a reference genome to detect mutations, and inputting the detected mutation information into an artificial intelligence model trained to determine whether or not the mutation is tumor-derived for analysis. [Background technology]

[0002] The main goal of precision oncology is to improve cancer diagnosis and treatment. To this end, various genomic and other molecular analytical methods are applied to tumor material to identify known predictive markers for treatment selection, induce molecular subtype classification that can estimate prognosis, characterize somatic alterations associated with tumor progression, sense disrupted pathways, and identify molecular discriminators of metastatic disease. Although various next-generation sequencing (NGS)-based approaches have been used to characterize tumor genomes in detail, comprehensive multiparameter analysis can classify tumor types more accurately. For example, the TCGA (The Cancer Genome Atlas) research network has generated comprehensive molecular profiles at DNA, RNA, protein, and epigenetic levels for hundreds of tumors. Such multiparameter analysis has resulted in an improved understanding of the functional roles and molecular alterations for tumor types and identified new tumor subtypes. Importantly, such efforts have led to the identification of new drug targets, a prerequisite for realizing the potential of precision medicine. However, access to tumor material for molecular profiling is not generally possible and relies on invasive methods that are not compatible with serial monitoring of tumor genotype.

[0003] Therefore, precision oncology is increasingly focused on liquid biopsies, which are non-invasive, can be repeated many times, and allow easy disease monitoring. In fact, attempts are being made to use such liquid biopsies for early cancer detection. The term "liquid biopsy" was first used to describe a method in which the same diagnostic information can be derived from a blood sample derived from a tissue biopsy sample. In oncology, the term was used in a broad sense to mean the analysis and sampling of various biological fluids that are easily accessible, such as not only blood, but also urine, ascites, or pleura.

[0004] At this time, the analytes of the peripheral blood fluid include circulating tumor cells (CTCs), circulating cell-free DNA (cfDNA) and small RNAs of cancer patients containing circulating tumor DNA (ctDNA), as well as circulating cell-free RNA (cfRNA) containing mRNA, circulating extracellular vesicles (EVs) such as exosomes, tumor-educated platelets (TEPs), proteins and metabolites. Such analytes may provide information about the characteristics of the primary tumor or metastasis that are typically obtained by pathologists. In addition to the information about genomic mutations and copy number changes that are typically obtained from CTCs or ctDNA, liquid biopsies are generally used to generate information about transcriptomes, protoplasts, proteins and metabolites (Jacob J. Chabon et al., Nature, Vol. 580, pp. 245-25, 2020).

[0005] One type of liquid biopsy is cell free DNA (cfDNA), which is a method of analyzing small DNA fragments that are present in various bodily fluids, including blood. Research into early cancer diagnosis using cfDNA is actively progressing, but there are currently many areas that need improvement in research into accurate analysis of single nucleotide variants. The reason why cfDNA cancer research using single nucleotide variants is difficult is that most single nucleotide variants discovered through cfDNA are not derived from cancer. As the amount of single nucleotide variants derived from tumors is very low and distributed in the blood, it is difficult to accurately discover tumor-derived variants.

[0006] Therefore, much of the research being conducted has been limited to single gene mutations that are already well known to cause cancer, but the reality is that there are no specific mutations that are found repeatedly, and it is extremely rare for the same mutation to be found in multiple patients.

[0007] Under such technical background, the present inventors have made intensive efforts to develop an artificial intelligence-based method for discovering tumor-derived mutations in cell-free DNA and a method for early diagnosis of cancer using the same. As a result, they have confirmed that when mutations are detected from acquired sequence information and the detected mutation information is input into an artificial intelligence model trained to distinguish tumor-derived mutations, tumor-derived mutations in cell-free DNA can be detected with high sensitivity and accuracy, and cancer can be diagnosed early using the same, thereby completing the present invention. Summary of the Invention [Problem to be solved by the invention]

[0008] An object of the present invention is to provide an artificial intelligence-based method for detecting tumor-derived mutations in cell-free DNA.

[0009] Another object of the present invention is to provide a method for providing information for early diagnosis of cancer using the above method.

[0010] Another object of the present invention is to provide a method for early diagnosis of cancer using the above method.

[0011] Another object of the present invention is to provide an apparatus and a computer-readable storage medium for use in the method for providing information for early diagnosis of cancer.

[0012] Another object of the present invention is to provide an apparatus and a computer-readable storage medium for use in said method for early diagnosis of cancer. [Means for solving the problem]

[0013] To achieve the above object, the present invention comprises the steps of (a) extracting nucleic acid from a biological sample and obtaining sequence information; (b) aligning the sequence reads to a reference genome database; (c) detecting mutations based on the aligned sequence reads; and (d) inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations, and comparing the output value with a cut-off value to determine the presence or absence of a tumor-derived mutation; The artificial intelligence model in step (d) is trained to distinguish tumor-derived mutations based on one or more features selected from the group consisting of functional features of cancer, mutation patterns, and technical features.

[0014] The present invention also provides a method for providing information for the early diagnosis of cancer, comprising the steps of: (a) detecting tumor-derived mutations in cell-free DNA by the above method; and (b) determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0015] The present invention also provides an information providing device for early cancer diagnosis based on artificial intelligence, including a decoding unit that extracts nucleic acid from a biological sample and decodes sequence information; an alignment unit that aligns the decoded sequence to a standard chromosome sequence database; a mutation detection unit that detects mutations based on the aligned sequence information; a tumor-derived mutation detection unit that inputs the detected mutations into an artificial intelligence model trained to distinguish tumor-derived mutations and determines the presence or absence of a tumor-derived mutation; and a cancer diagnosis unit that determines the presence of cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0016] The present invention also provides a computer-readable storage medium, comprising instructions configured to be executed by a processor to provide information for the early diagnosis of cancer, the computer-readable storage medium ...

[0017] The present invention also provides a method for early diagnosis of cancer, comprising: (a) detecting a tumor-derived mutation in cell-free DNA by the above method; and (b) determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0018] The present invention also provides an artificial intelligence-based early cancer diagnosis device, including a decoding unit that extracts nucleic acid from a biological sample and decodes sequence information; an alignment unit that aligns the decoded sequence to a standard chromosome sequence database; a mutation detection unit that detects mutations based on the aligned sequence information; a tumor-derived mutation detection unit that inputs the detected mutations into an artificial intelligence model trained to distinguish tumor-derived mutations to determine the presence or absence of a tumor-derived mutation; and a cancer diagnosis unit that determines the presence of cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0019] The present invention also provides a computer-readable storage medium including instructions configured to be executed by a processor for performing early cancer diagnosis, the ... [Brief description of the drawings]

[0020] [Figure 1] 1 is an overall flowchart for detecting tumor-derived mutations in cell-free DNA based on artificial intelligence according to the present invention. [Diagram 2] 1 is an overall flowchart of an artificial intelligence-based early cancer diagnosis method according to the present invention. [Diagram 3] The results show the analysis results of the characteristics of single gene mutations in cell-free DNA according to the origin, which were discovered according to one embodiment of the present invention. The upper panel shows the mutational signatures of single gene mutations in cell-free DNA of breast cancer patients according to the origin, which were analyzed according to one embodiment, and the lower panel shows the mutational signatures in cancer tissues of cancer patients by cancer type, which were developed in a large-scale cancer gene project called Pan-cancer Analysis of Whole Genomes (PCAWG). The mutational signature is the concept that there is a specific pattern in the types of single gene mutations that occur in a specific type of cancer. [Figure 4] These are the results of confirming the distribution of biological features in breast cancer derived from cfDNA of breast cancer patients. (A) shows the results of confirming the replication score, H3K9me3, and gene expression level, and (B) shows the accumulation pattern of single gene mutations (Regional Mutation Density, RMD). [Diagram 5] 1 shows the results of confirming the performance of a learning model for detecting single gene mutations derived from breast cancer constructed according to one embodiment of the present invention, in which (A) shows the performance of the classification model using sensitivity and specificity in an ROC curve, and (B) shows the performance of the classification model using precision and recall in a PR curve. [Figure 6] 1 shows the results of evaluating the importance of individual features used in a learning model constructed according to one embodiment of the present invention. [Figure 7] 13 is a result of comparing a mutational signature predicted using a learning model constructed according to an embodiment of the present invention with an actual result. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention belongs. Generally, the nomenclature used herein and the laboratory procedures described below are those well known and commonly employed in the art.

[0022] Terms such as first, second, A, B, etc. may be used to describe various components, but the components are not limited by the terms and are used only to distinguish one component from another. For example, a first component may be named a second component, and similarly, a second component may be named a first component, without departing from the scope of the technology described below. The term "and / or" includes a combination of multiple related listed items or any of multiple related listed items.

[0023] In the terms used in this specification, singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as "comprise" should be understood to mean the presence of a stated feature, number, step, operation, component, part, or combination thereof, but not to exclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0024] Before describing the drawings in detail, it should be made clear that the division of components in this specification is merely a division according to the main function of each component. That is, two or more components described below may be combined into one component, or one component may be divided into two or more components according to more detailed functions. Furthermore, each component described below may additionally perform some or all of the functions of other components in addition to its own main function, and some of the main functions of each component may be exclusively performed by other components.

[0025] Additionally, in performing a method or method of operation, the steps constituting the method may be performed out of the order stated, unless the context clearly dictates a particular order, i.e., the steps may be performed in the same order as stated, substantially simultaneously, or in the opposite order.

[0026] In the present invention, it was demonstrated that tumor-derived mutations in cell-free DNA can be detected with high sensitivity and accuracy by aligning sequence analysis data obtained from a sample to a reference genome, detecting mutations from the aligned nucleic acid fragments, and inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations.

[0027] In one embodiment of the present invention, a learning model capable of discovering tumor-derived mutations using 48 features including cancer functional features and sequencing quality features was constructed, and the performance was tested using cfDNA, tumor, and WBC liquid biopsy genome data from 38 breast cancer patients, confirming that the model performed well (Figure 5).

[0028] In the present invention, the term "reads" refers to a single nucleic acid fragment whose sequence information has been analyzed using various methods known in the art. Therefore, in this specification, the terms "sequence information" and "read" have the same meaning in that they are the result of obtaining sequence information through a sequencing process.

[0029] In the present invention, the term "tumor-derived mutation" refers to a mutation that originates from a cancer cell.

[0030] Therefore, in one aspect, the present invention comprises: The method for detecting tumor-derived mutations in cell-free DNA based on artificial intelligence includes the following steps: (a) extracting nucleic acid from a biological sample and obtaining sequence information; (b) aligning the sequence reads to a reference genome database; (c) detecting mutations based on the aligned sequence reads; and (d) inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations, and comparing the output value with a cut-off value to determine the presence or absence of a tumor-derived mutation; The artificial intelligence model in step (d) is characterized in that it is trained to distinguish tumor-derived mutations based on one or more features selected from the group consisting of functional features of cancer, mutation patterns, and technical features.

[0031] In the present invention, the step of obtaining sequence information in step (a) may be characterized as being carried out by a method including the following steps: (ai) obtaining nucleic acid from the biological sample; (a-ii) removing proteins, lipids, and other residues from the collected nucleic acid by a salting-out method, a column chromatography method, or a beads method to obtain purified nucleic acid; (a-iii) preparing a single-end sequencing or pair-end sequencing library from the purified nucleic acids or the nucleic acids randomly fragmented by enzymatic cleavage, crushing, or hydroshear method; (a-iv) subjecting the prepared library to a next-generation sequencer; and (av) A step of acquiring sequence information (reads) of the nucleic acid using a next-generation gene sequencer.

[0032] In the present invention, the step of obtaining sequence information in step (a) may be characterized by obtaining the sequence information through whole genome sequencing of the isolated cell-free DNA at a depth of 1 million to 100 million reads.

[0033] In the present invention, the biological sample means any substance, biological fluid, tissue, or cell obtained from or derived from an individual, and includes, for example, whole blood, leukocytes, peripheral blood mononuclear cells, buffy coat, blood (including plasma and serum), sputum, tears, mucus, nasal washes, nasal aspirate, breath, urine, semen, saliva, peritoneal washings, pelvic fluids, cystic fluid, meningeal fluid, amniotic fluid, glandular fluid, pancreatic juice, and the like. The fluids may include, but are not limited to, lymph fluid, pleural fluid, nipple aspirate, bronchial aspirate, synovial fluid, joint aspirate, organ secretions, cells, cell extracts, hair, buccal cells, placental cells, cerebrospinal fluid, and mixtures thereof.

[0034] In the present invention, the term "reference population" refers to a reference population that can be compared with a standard base sequence database, and means a group of people who are currently free of a specific disease or illness. In the present invention, the standard base sequence in the standard chromosome sequence database of the reference population may be a reference chromosome registered in a public health institution such as NCBI.

[0035] In the present invention, the nucleic acid in step (a) may be cell-free DNA, and more preferably, may be circulating tumor DNA, but is not limited thereto.

[0036] In the present invention, the next-generation sequencer can be used with any sequencing method known in the art. Sequencing of the nucleic acids separated by the selection method is typically performed using next-generation sequencing (NGS). Next-generation sequencing includes any sequencing method that determines the nucleotide sequence of an individual nucleic acid molecule or one of its clonally extended proxies for an individual nucleic acid molecule in a highly similar manner (e.g., 105 or more molecules are sequenced simultaneously). In one embodiment, the relative abundance of a nucleic acid species in a library can be estimated by measuring the relative occurrence of its cognate sequence in the data generated by the sequencing experiment. Next-generation sequencing methods are known in the art and are described, for example, in Metzker, M. (2010) Nature Biotechnology Reviews 11:31-46, which is incorporated herein by reference.

[0037] In one embodiment, next generation sequencing is performed to determine the nucleotide sequence of individual nucleic acid molecules (e.g., the HeliScope Gene Sequencing system from Helicos BioSciences and the PacBio RS system from Pacific Biosciences). In other embodiments, sequencing, e.g., massively parallel short-read sequencing (e.g., Solexa sequencer, Illumina Inc., San Diego, Calif.) methods that generate more bases of sequence per sequencing unit than other sequencing methods that generate fewer but longer reads, determine the nucleotide sequence of clonally extended proxies for individual nucleic acid molecules (e.g., Solexa sequencer, Illumina Inc., San Diego, Calif.; 454 Life Sciences (Branford, Conn.) and Ion Torrent). Other methods or machines for next generation sequencing include, but are not limited to, those provided by 454 Life Sciences (Branford, Connecticut), Applied Biosystems (Foster City, California; SOLiD Sequencer), Helicos Biosciences Corporation (Cambridge, Massachusetts), and emulsion and microfluidic sequencing techniques such as nano-infusion (e.g., GnuBio infusion).

[0038] Platforms for next generation sequencing include, but are not limited to, the Roche / 454 Genome Sequencer (GS) FLX system, the Illumina / Solexa Genome Analyzer (GA), the Life / APG Support Oligonucleotide Ligation Detection (SOLiD) system, Polonator's G.007 system, Helicos BioSciences' HeliScope Gene Sequencing system and Pacific Biosciences' PacBio RS system.

[0039] In the present invention, the alignment step of step (b) may be performed using, but is not limited to, the BWA algorithm and the Hg19 sequence.

[0040] In the present invention, the BWA algorithm can include, but is not limited to, BWA-ALN, BWA-SW, or Bowtie2, etc.

[0041] In the present invention, the method may further include a step of selecting reads having a mapping quality score of the aligned nucleic acid fragments equal to or greater than a reference value before performing step (c). The reference value may be used without limitation as long as it is a value that can confirm the quality of the aligned nucleic acid fragments, and may be preferably 50 to 70 points, more preferably 60 points, but is not limited thereto.

[0042] In the present invention, the step of detecting a mutation in step (c) may be characterized by comprising the following steps: (ci) selecting from the aligned reads sequences that differ from a reference gene; and (c-ii) storing the selected base sequence information;

[0043] In the present invention, the (ci) step may use any method known to a person skilled in the art that can detect mutations, and preferably may be, but is not limited to, Mutect2, LoFreq, Delly2, etc.

[0044] In the present invention, step (c-ii) may be stored in a specific file format, or the mutation information detected in step (ci) may be used as is.

[0045] In the present invention, the functional characteristic of a cancer may be any genetic, epigenetic or transcriptional characteristic that affects the occurrence of a single gene mutation for each cancer type, and may be characterized as one or more selected from the group consisting of regional mutation density (RMD), replication time, H3K4Me1, H3K4Me3, H3K9Me3, H3K27Me3, H3K36Me3, Dnase I hypersensitive site (DHS), gene expression level of protein binding site (footprint) in DHS, cancer positive selection score, and cancer negative selection score, but is not limited thereto.

[0046] In the present invention, the regional mutation density (RMD) is used in a similar sense to the background mutation rate, and means that the whole genome is divided into certain intervals to calculate the mutation frequency.

[0047] In the present invention, the accumulation pattern of single gene mutations by cancer type is a quantitative value for whether the area of ​​the cancer has many mutations or few mutations. Single gene mutations in cancer are not uniformly distributed in the human genome. The amount of single gene mutations accumulated varies depending on the entire genome region, and the accumulation pattern by cancer type also differs greatly. In addition, epigenetic characteristics (histone modification, replication time) are the main cause of the accumulation pattern of single gene mutations by cancer type, and the accumulation pattern of single gene mutations contains the epigenetic characteristics of the cancer type.

[0048] The accumulation pattern of single gene mutations varies by whole genome region and by cancer type, so it can be a useful indicator for discovering tumor-derived mutations. The accumulation pattern of single gene mutations can be used to determine whether the discovered mutation is located in an area with a high probability of occurrence in the cancer. Mutations discovered in an area with a high probability of occurrence in the cancer are likely to be actual tumor-derived mutations, not cfDNA artifacts. In addition, the accumulation pattern of single gene mutations also includes epigenetic features, so epigenetic features can also be taken into consideration when discovering tumor-derived mutations.

[0049] In addition, the accumulation patterns of haematopoiesis mutations can be used to identify regions where mutations frequently occur in blood cells, the accumulation patterns of acellular mutations in normal humans can be used to identify regions where cfDNA artifacts are frequently found, and the accumulation patterns of germline mutations in normal humans can be used to identify regions where germline mutations frequently occur.

[0050] To calculate the accumulation pattern of single-gene mutations, WGS with a sufficient number of samples from a large cohort is required. All mutations discovered from the samples are combined to calculate the accumulation pattern of single-gene mutations.

[0051] The whole genome is divided into certain intervals, for example, 10 kb or 1 Mb, and the mutation frequency is calculated. The number of mutations in each interval is normalized by dividing it by the number of mutations discovered in the whole genome.

[0052] It is important to set appropriate intervals, because short intervals across the genome (e.g., 1 kb units) can result in regions that are too small to see patterns, while long intervals (e.g., 10 Mb) can result in localized patterns being aggregated.

[0053] In the present invention, the mutation pattern may be any mutation in which a normal base is mutated to another base to cause a functional abnormality of a gene, and may be characterized as being preferably one or more selected from the group consisting of C->A, C->G, C->T, T->A, T->C, and T->G, but is not limited thereto.

[0054] In the present invention, C→A means to confirm whether the detected mutation is a mutation of normal base C to mutant base A, C→G means to confirm whether the detected mutation is a mutation of normal base C to mutant base G, and the others have the same meaning.

[0055] In the present invention, the technical feature of the mutation may be any feature of sequence information extracted from sequence information (reads) aligned to the position of a single gene mutation, and may be any of the following: average read depth of the reference allele in the mutation region, average mapping quality, average base quality, average number of mismatches, average position of the reference allele, average total base quality of mismatches, number or position of bases with a phread quality of 2 at the 3' end, average clipped read length, average position of the read from the 3' end, ratio of plus strand reads, DNA fragment length; The average read depth of the variant allele in the variant region, the average mapping quality, the average base quality, the average number of mismatches, the average position of the reference allele, the average total base quality of the mismatches, the number or position of bases with a phread quality of 2 at the 3' end, the average clipped read length, the ratio of plus strand reads, the DNA fragment length, the DNA fragment, and The present invention can be characterized by including one or more features selected from the group consisting of, but not limited to, MUT.notBoth (the number of DNA fragments that have no overlap at the mutation position in forward and reverse reads + the number of fragments that have overlaps but do not have matching mutations).

[0056] In the present invention, the features of step (d) may include the features listed in Table 1 below.

[0057] [Table 1] TIFF0007634714000002.tif243170TIFF0007634714000003.tif242170TIFF0007634714000004.tif74170

[0058] In the present invention, the artificial intelligence model in step (d) may be any learning model that learns to distinguish whether or not a mutation is tumor-derived, and may be characterized as being one or more selected from the group consisting of random forest, XGboost, and deep neural network, but is not limited thereto.

[0059] In the present invention, the reference value in step (d) may be any value that can distinguish whether the detected mutation is tumor-derived or not, and may be, but is not limited to, 0.5. If the reference value is 0.5, it may be characterized in that if the reference value is 0.5 or more, it is determined that the mutation is tumor-derived.

[0060] In the present invention, the artificial intelligence model is trained so that when there is a tumor-derived mutation, the output result is close to 1, and when there is no tumor-derived mutation, the output result is close to 0. Using 0.5 as the standard, if the output is 0.5 or higher, it is determined that there is cancer, and if the output is 0.5 or lower, it is determined that there is no cancer.

[0061] It is obvious to any ordinary technician that the standard value of 0.5 is a value that can change at any time. For example, if you want to reduce false positives, you can set a standard value higher than 0.5 to set a stricter standard for determining whether cancer is present, and if you want to reduce false negatives, you can measure a lower standard and set a slightly looser standard for determining whether cancer is present.

[0062] In the present invention, 38 breast cancer patients were divided in a ratio of 8:2 and used as a train set (30 patients) and a test set (8 patients) for learning and performance evaluation of the artificial intelligence model. 2418 mutations and 8749 artifacts derived from acellular tumors of 30 breast cancer patients were used for learning, and 1159 mutations and 2441 artifacts derived from acellular tumors of 8 breast cancer patients were used as a test set. In addition, the train set (30 patients) was divided in a ratio of 3:1 and used as a train set and a valid set for learning and performance evaluation of the DNN model.

[0063] In the present invention, when the artificial intelligence model is a random forest, the loss function can be characterized by being expressed by the following Equation 1 or 2.

[0064]

number

[0065] Let τ be the set of all possible cases of the parameter θ of the node splitting function. Then, τ at the training stage of the jth node i A subset τ where ⊂ τ i The optimal value of the parameter θ j * is τ i It is calculated as the value that maximizes the objective function (loss function) defined by the amount of information gained.

[0066]

number

[0067] I is the amount of information gained, and S is the set of data arriving at a node. i denotes the data set that will be included in the child node i ∈{L,R} to the left or right of this node. In addition, |·| and H(S) denote the number of data items belonging to the data set and the Shannon entropy, respectively.

[0068] In the present invention, when the artificial intelligence model is XGBoost, the loss function can be characterized by being expressed by the following Equation 3.

[0069]

number

[0070] l was predicted using a differentiable convex loss function. Calculate the difference between TIFF0007634714000008.tif6128 and the actual y. Ω gives a penalty for the complexity of the model. f k means an independent tree structure.

[0071] In the present invention, when the artificial intelligence model is a deep neural network, the loss function can be characterized by being expressed by the following Equation 4.

[0072]

number

[0073] The loss function is binary cross entropy, N is the total number of samples, TIFF0007634714000010.tif6128 is the probability that the model predicts that the i-th input value is close to class 1, and y i is the actual class of the i-th input value.

[0074] In the present invention, when the artificial intelligence model is a DNN, the learning may be characterized by including the following steps: i) classifying the detected mutation data into training, validation, and test data; In this case, the training data is used to learn the model, the validation data is used to verify hyper-parameter tuning, and the test data is used to evaluate the performance of the optimal model after it is produced. ii) constructing an optimal model through hyper-parameter tuning and learning process; iii) Comparing the performance of various models obtained through hyper-parameter tuning using validation data, and determining the model with the best performance on the validation data as the optimal model;

[0075] In the present invention, the hyper-parameter tuning process is a process of optimizing values ​​of various parameters (number of convolution layers, number of dense layers, number of convolution filters, etc.) constituting a DNN model, and the hyper-parameter tuning process can be characterized by using Bayesian optimization and grid search techniques.

[0076] In the present invention, the learning process may be characterized in that the internal parameters (weights) of the DNN model are optimized using the determined hyper-parameters, and when the validation loss starts to increase compared to the training loss, it is determined that the model is overfitting, and model learning is interrupted before that happens.

[0077] In the present invention, the result value of the analysis of the mutation data input to the artificial intelligence model in step d) may be any specific score or real number, and is preferably characterized as being a real number, but is not limited thereto.

[0078] In the present invention, when the artificial intelligence model is a DNN, a real value means a value expressed as a probability value by adjusting the output of the artificial intelligence model on a scale of 0 to 1 using a sigmoid function or a softmax function in the last layer.

[0079] From another viewpoint, the present invention provides How to provide information for early cancer diagnosis, including the following steps: (a) detecting tumor-derived mutations in cell-free DNA by the method described above; and (b) determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0080] From another viewpoint, the present invention provides A sequencing unit that extracts nucleic acids from biological samples and decodes sequence information; an alignment section that aligns the decoded sequences to a standard chromosome sequence database; a mutation detection unit that detects mutations based on the aligned sequence information; a tumor-derived mutation detector that inputs the detected mutations into an artificial intelligence model trained to distinguish tumor-derived mutations to determine the presence or absence of tumor-derived mutations; and A cancer diagnostic section that determines the presence of cancer or minimal residual cancer if a tumor-derived mutation is detected; The present invention relates to an information providing device for early diagnosis of cancer based on artificial intelligence.

[0081] In the present invention, the decoding unit may include a nucleic acid injection unit that injects nucleic acid extracted from an independent device; and a sequence information analysis unit that analyzes sequence information of the injected nucleic acid, and may preferably be an NGS analysis device, but is not limited thereto.

[0082] In the present invention, the decoding unit may receive and decode sequence information data generated by an independent device.

[0083] From another viewpoint, the present invention provides 1. A computer-readable storage medium, comprising instructions configured to be executed by a processor to provide information for early diagnosis of cancer; (a) extracting nucleic acid from a biological sample and obtaining sequence information; (b) aligning the sequence reads to a reference genome database; (c) detecting mutations based on the aligned sequence reads; (d) inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations, and comparing the output value with a cut-off value to determine the presence or absence of a tumor-derived mutation; The artificial intelligence model in step (d) is trained to distinguish tumor-derived mutations based on one or more features selected from the group consisting of functional features, mutation patterns, and technical features of cancer; and (e) A computer-readable storage medium comprising instructions configured to be executed by a processor to provide information for early diagnosis of cancer by determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0084] In another aspect, the method of the present application may be implemented using a computer. In one embodiment, the computer includes one or more processors coupled to a chipset. The chipset may also include memory, storage, a keyboard, a graphics adapter, a pointing device, and a network adapter. In one embodiment, the chipset functionality is enabled by a memory controller hub and an I / O controller hub. In another embodiment, the memory may be directly coupled to the processor instead of the chipset. The storage is any device capable of maintaining data, including a hard drive, a compact disk read-only memory (CD-ROM), a DVD, or other memory device. The memory is responsible for the data and instructions used by the processor. The pointing device may be a mouse, a track ball, or other type of pointing device, and is used in combination with a keyboard to transfer input data to the computer system. The graphics adapter displays images and other information on a display. The network adapter is coupled to the computer system by a local or long-distance communication network. The computer used in the present application is not, however, limited to the above configurations, and may include none or additional configurations, and may be part of a Storage Area Network (SAN), and the computer of the present application may be configured to execute modules in a program for performing the method of the present application.

[0085] In the present application, a module may refer to a functional and structural combination of hardware for performing the technical idea of ​​the present application and software for driving the hardware. For example, the module may refer to a logical unit of a certain code and a hardware resource for performing the certain code, and it is obvious to those skilled in the art that the module does not necessarily refer to physically connected code or one type of hardware.

[0086] In another aspect, the present invention relates to a method for early diagnosis of cancer, comprising: (a) detecting tumor-derived mutations in cell-free DNA by the above-mentioned method; and (b) determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0087] In another aspect, the present invention relates to a method for treating a cancer patient, comprising: (a) detecting tumor-derived mutations in cell-free DNA by the above-mentioned method; (b) determining that a patient has cancer or microscopic residual cancer if a tumor-derived mutation is detected; and (c) treating the patient determined to have cancer or microscopic residual cancer.

[0088] In the present invention, the cancer therapeutic agent can be used without limitation as long as it is a method capable of treating cancer or microscopic residual cancer, and is preferably characterized by being performed by any one or more methods selected from the group consisting of surgery, adjuvant chemotherapy, neoadjuvant chemotherapy, radiation therapy, hormone therapy, cytotoxic therapy, immunotherapy, adoptive T cell therapy, targeted therapy, and combinations thereof, more preferably, by administering a cancer therapeutic agent, and most preferably, by administering any one or more anticancer agents selected from the group consisting of chemical anticancer agents, targeted anticancer agents, and immune anticancer agents, but is not limited thereto.

[0089] From another perspective, the present invention relates to an artificial intelligence-based early cancer diagnosis device, which includes a decoding unit that extracts nucleic acid from a biological sample and decodes sequence information; an alignment unit that aligns the decoded sequence to a standard chromosome sequence database; a mutation detection unit that detects mutations based on the aligned sequence information; a tumor-derived mutation detection unit that inputs the detected mutations into an artificial intelligence model trained to distinguish tumor-derived mutations and determines the presence or absence of a tumor-derived mutation; and a cancer diagnosis unit that determines the presence of cancer or microscopic residual cancer if a tumor-derived mutation is detected.

[0090] In another aspect, the present invention relates to a computer-readable storage medium including instructions configured to be executed by a processor for performing early diagnosis of cancer, the computer-readable storage medium including instructions configured to be executed by a processor for performing early diagnosis of cancer, the computer-readable storage medium including instructions configured to be executed by a processor for performing early diagnosis of cancer, the computer-readable storage medium including instructions configured to be executed by a processor for performing early diagnosis of cancer, the computer-readable storage medium comprising: (a) a step of extracting nucleic acid from a biological sample to obtain sequence information; (b) a step of aligning the obtained sequence information (reads) with a reference genome database; (c) a step of detecting mutations based on the aligned sequence information (reads); (d) a step of inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations and comparing an output value with a cut-off value to determine the presence or absence of a tumor-derived mutation, the artificial intelligence model in the step (d) being trained to distinguish tumor-derived mutations based on one or more features selected from the group consisting of a functional feature, a mutation pattern, and a technical feature of cancer; and (e) a step of determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected. EXAMPLES

[0091] The present invention will be described in more detail with reference to the following examples. It will be obvious to those skilled in the art that these examples are merely for the purpose of illustrating the present invention and are not intended to limit the scope of the present invention.

[0092] Example 1. Obtaining tumor, WBC, and cfDNA WGS data and confirming the origin of mutations discovered from cfDNA

[0093] To confirm whether the single gene mutations extracted from cfDNA are tumor-derived, haematopoiesis, or artifacts, whole genome sequencing genomic data from tumor tissue, plasma-depleted whole blood cells (WBC), and cfDNA from each patient is required. After obtaining WGS samples of tumor tissue, WBC, and cfDNA from cancer patients, processing was performed using the GATK pipeline. To obtain single gene mutation profiles derived from each patient's tumor, tumor, haematopoiesis, and cfDNA mutations were extracted using Mutect2.

[0094] The data used for mining were whole exome sequencing data for tumor tissue, WBC, and cfDNA for 38 metastatic breast cancer patients, and were the phs001417.v1.p1 data registered in the dbGaP database in Adalsteinsson, VA et al. Nat. Commun. 8, 1324 (2017).

[0095] Specifically, we created the obtained sequence information (reads) into a BAM file format that can be used for variant discovery. A BAM file is a binary file that contains information on the alignment of sequence information (reads) to a reference genome database. Genome Analysis Tool Kit (GATK) provides tools and standard analysis pipelines for NGS data analysis, and we used the data preprocessing pipeline for variant discovery provided by GATK (see: https: / / gatk.broadinstitute.org / hc / en-us / articles / 360035535912-Data-pre-processing-for-variant-discovery). The preprocessing process is roughly divided into three steps. First, the obtained sequence information (reads) is aligned to a reference genome database. Second, the duplicated sequence information (reads) generated by PCR during the sequence information (reads) production process is displayed. The third step is to recalculate and adjust the base quality of the sequence information (reads) (Base Quality Score Recalibration).

[0096] Using the previously constructed patient-specific tumor and haematopoiesis single gene mutation profile, we confirmed which of the mutations found in cfDNA were single gene mutations derived from tumors.As a result of checking with breast cancer patient samples, we confirmed that an average of 15.6% (97 mutations) of the single gene mutations found in cfDNA were tumor-derived mutations, and the artifact rate was very high at 84%.

[0097] Example 2. Extraction of functional features of cancer that can uncover tumor-derived mutations 2-1. Extraction of transcriptome data, gene expression levels, and selection scores Repli-seq, Dnase-seq, and ChIP-seq (H3K4me1, H3K4me3, H3K9me3, H3K27me3, H3K36me3) were obtained from ENCODE and preprocessed, and RNA-seq data from cancer-specific patients from TCGA was used for transcriptome data. In addition, positive selection and negative selection score data for each cancer type were also used as features for the model to be developed.

[0098] First, genomic and epigenetic data for MCF7, a breast cancer cell line, was collected using ENCODE. Repli-seq, Dnase-seq, and ChIP-seq (H3K4me1, H3K4me3, H3K9me3, H3K27me3, H3K36me3) for the MCF7 cell line were obtained from ENCODE. Transcript data from 1,099 TCGA breast cancer patients was used for the transcriptome data from the Toil database. The Toil database is a large-scale transcriptome database that was created by unifying data from large-scale transcriptome cohorts through the same preprocessing process. The expression levels of each gene in 1,099 breast cancer patients were averaged, and the average expression levels of each gene in breast cancer patients were calculated and used as features for the artificial intelligence model.

[0099] Quantitative values ​​for genes that undergo positive or negative selection as breast cancer progresses and are prone or unlikely to undergo mutations were used as features of the artificial intelligence model. The positive selection quantitative value was the average of the breast cancer positive selection quantitative values ​​from two papers. The negative section quantitative value was collected from one paper. (Source: ENCODE: https: / / www.encodeproject.org / ) (Source: Toil: https: / / doi.org / 10.1038 / nbt.3772) (Source: Positive selection: 10.1016 / j.cell.2017.09.042, 10.1038 / ng.3987) (Source: Negative selection: 10.1016 / j.cell.2017.09.042)

[0100] 2-2. Extraction of mutation signature It has already been shown that tumor-derived mutations and haematopoiesis mutations have different molecular characteristics. Recently, it has been reported that single gene mutations derived from tumors and single gene mutations in haematopoiesis have different mutational signatures (Jacom J. Chabon et al., Nature, Vol. 580, pp. 245-251, 2020). Therefore, the types of single gene mutations used when calculating the mutational signature using the data of Example 1 were broadly divided into six types (T>G, T>C, T>A, C>T, C>G, C>A), and the distribution characteristics of the mutations confirmed in the liquid biopsy according to their origin (tumor, haematopoiesis, artifact) were analyzed. As a result of the analysis, it was confirmed that tumor-derived mutations, haematopoiesis mutations, and artifacts show different mutational signature patterns. Since the mutations confirmed in the liquid biopsy show different mutational signature patterns according to their origin, the mutational signature was used as a feature of the algorithm.

[0101] The mutational signature was calculated using a program called bedtools and a python script. bedtools is a command-line program that supports rapid mutual calculation of genome data including one-dimensional coordinate systems such as BED, GFF3, and VCF. The nucleobase of the reference genome sequence at the excavated mutation position and the base of the excavated mutation were confirmed to determine which base the mutation occurred from.

[0102] 2-3. Extraction of regional mutation density (RMD) patterns of single gene mutations The mechanism by which single gene mutations occur varies by cancer type, and the manner in which mutations accumulate also differ. In particular, the pattern of passenger mutation accumulation varies greatly by cancer type, and there are previous studies that use such characteristics to classify cancer types by passenger mutations. Therefore, the accumulation pattern of single gene mutations (regional mutation density) for such cancer types was used as a feature of the tumor-derived mutation mining algorithm. The accumulation pattern of haematopoiesis mutations, the accumulation pattern of acellular mutations in normal humans, and the accumulation pattern of germline mutations in normal humans were also used as features of the artificial intelligence model. In the examples, the accumulation pattern of single gene mutations in breast cancer, the accumulation pattern of haematopoiesis mutations, the accumulation pattern of acellular mutations in normal humans, and the accumulation pattern of germline mutations in normal humans were used as features of the artificial intelligence model.

[0103] The mutation accumulation patterns were calculated as follows.

[0104] The entire genome was divided into intervals of a certain length, and the number of mutations that occurred in each interval (1 Mb or 10 kb) was added up to calculate the number of mutations in each interval. The number of mutations in each interval was then divided by the total number of mutations to perform normalization.

[0105] Accumulation patterns of single gene mutations by cancer type were constructed using WGS produced by the international cancer genome project called PCAWG (Pan-cancer analysis of whole genomes, Campbell, PJ, Getz, G. et al., Nature 578, 82-93, 2020).

[0106] Haematopoiesis mutation accumulation patterns were constructed using blood WGS of PCAWG ovarian cancer patients.

[0107] The cell-free mutation accumulation patterns of normal humans were constructed using cell-free WGS of 100 normal humans from the Green Cross Genome.

[0108] Normal human germline mutation accumulation patterns were constructed using large-scale WGS of The Genome Aggregation Database (gnomAD, Karczewski, KJ et al., Nature 581, 434-443, 2020).

[0109] Example 3. Training an artificial intelligence algorithm to mine single gene mutations from tumors To develop an algorithm for discovering single gene mutations derived from tumors in cfDNA, 22 functional features for each cancer type obtained through the previous analysis and 26 sequencing quality features extracted from the genome data of the patient were used together. The genome data of the patient was the genome data of Example 1. For the sequencing quality features, the genome data of each patient's liquid biopsy was preprocessed through the gatk pipeline, and then 26 sequencing quality features for the position of the single gene mutation were extracted using a tool called bamcount. A total of 48 features extracted in this way were used to develop an algorithm for discovering single gene mutations derived from tumors.

[0110] The 48 extracted features are as shown in Table 1 above.

[0111] The AI ​​algorithm built a binary classification model to distinguish between tumor-related mutations and the rest of the single gene mutations extracted from cfDNA. Three AI models, Random Forest, XGBoost, and Deep Neural Network, were used for model training.

[0112] To optimize the Random Forest and XGBoost models, the training data was repeatedly classified into training and validation data using 5-fold cross validation, and the process of hyper-parameter tuning was carried out. To optimize the Deep Neural Network, the detected mutation data was classified into training, validation, and test data, and then hyper-parameter tuning was carried out.

[0113] Experimental example 1. Analysis of the origin of single gene mutations extracted from cfDNA Using genetic data from cfDNA, tumor, and WBC liquid biopsies of 38 breast cancer patients in Example 1, we analyzed the characteristics of tumor-derived mutations extracted from cfDNA and conducted learning and performance evaluation of the tumor-derived mutation extraction algorithm.

[0114] By dividing single gene mutations found in cfDNA from breast cancer patients by origin and comparing the mutational signatures, it was confirmed that tumor-derived mutations frequently had C>T and C>G mutations, while artifacts frequently had C>A mutations. It was confirmed that the mutations found in cfDNA had different characteristics depending on their origin (Figure 3).

[0115] In addition, we investigated the distribution of biological features in breast cancer derived from cfDNA of breast cancer patients. Early replication timing regions have relatively few SNVs, while late replication timing regions are known to have many mutations due to poor repair mechanisms.

[0116] As a result, as shown in Figure 4A, the lower the replication score, the later the replication region, and we confirmed that the replication score was low in tumor mutations in cfDNA, which is consistent with the previously known mechanism. We also confirmed that tumor mutations occurred frequently in breast cancer heterochromatin with high H3K9me3 values, which is also consistent with the previously known biological mechanism. We confirmed that gene expression levels were low in tumor mutations, which is consistent with the characteristic that highly expressed genes are less likely to undergo mutations, and confirmed that biological features are important factors for distinguishing tumor-derived mutations from artifacts and blood.

[0117] In addition, we compared the RMD values ​​of cfDNA origins in breast cancer patients and confirmed that the largest difference was observed among biological features according to origin, as shown in Figure 4B. In other words, tumor-derived mutations tended to have higher PCAWG breast cancer RMD than cfDNA artifacts, while cfDNA hematopoiesis mutations showed higher cfDNA RMD in PCAWG blood, gnomAD, and normal humans.

[0118] Experimental example 2. Learning and performance evaluation of an artificial intelligence algorithm for discovering single gene mutations from tumors For the training and performance evaluation of the tumor-derived single gene mutation mining algorithm, 38 patients were divided into 30 training data and 8 performance evaluation data. As a result of the performance evaluation after constructing the tumor-derived single gene mutation mining algorithm, it was confirmed that Random Forest and DNN showed excellent performance with ROC AUC of 0.922 and 0.864. In addition, it was confirmed that Average Precision also showed excellent performance with 0.585 (Figure 5).

[0119] Experimental example 3. Key feature analysis of the mining algorithm for single gene mutations derived from breast cancer tumors We analyzed which of the 48 features used to train the algorithm were important in discovering tumor-derived mutations.

[0120] An analysis was conducted to determine which of the 22 cancer functional features used to train the algorithm were important in uncovering tumor-derived mutations. The cancer functional features were subdivided into six features related to mutational signatures and 16 biological features, and the analysis was conducted afterwards. Feature importance was measured using the degree to which the performance of the learning model (F1 score) decreased when feature values ​​were randomly shuffled (random shuffling). The process of randomly shuffling and measuring model performance was performed a total of 100 times, and the average degree of decrease in model performance was measured.

[0121] As a result, as shown in Figure 6, it can be seen that the mutation accumulation pattern (regional mutation density) plays the most important role in uncovering tumor-derived mutations. Three mutation accumulation pattern features are ranked 1st, 2nd, and 3rd in biological feature importance, and among them, the breast cancer mutation accumulation pattern (pcawg_tumor_rmd) plays the most important role. It was also confirmed that the histone modification marker H3K27me3 and DNA replication timing play an important role.

[0122] Experimental example 4. Prediction of mutational signatures using the developed algorithm We verified whether the algorithm developed in this study can actually predict mutational signature patterns in cancer. We compared the mutational signatures in tumors from eight breast cancer patients with the results of mutational signature analysis using tumor-derived mutations predicted by the algorithm developed in this study. The analysis results confirmed that the algorithm developed in this study can predict mutational signatures in tumors (Figure 7).

[0123] Although the specific parts of the present invention have been described in detail above, it is clear to those skilled in the art that such specific techniques are merely preferred embodiments and do not limit the scope of the present invention. Therefore, the substantial scope of the present invention is defined by the appended claims and their equivalents. [Industrial Applicability]

[0124] The method for detecting tumor-derived mutations in cell-free DNA and the method for early cancer diagnosis using the same according to the present invention diagnoses cancer early using both functional and sequence characteristics of cancer based on artificial intelligence using next generation sequencing (NGS), and has high accuracy and sensitivity and is commercially applicable, making the method of the present invention useful for early cancer diagnosis.

Claims

1. A method for detecting tumor-derived mutations in cell-free DNA based on artificial intelligence, comprising the following steps: (a) extracting nucleic acid from a biological sample and obtaining sequence information; (b) aligning the obtained sequence information (reads) to a reference genome database; (c) detecting mutations based on the aligned sequence reads; and (d) inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations, and comparing the output value with a cut-off value to determine the presence or absence of a tumor-derived mutation; The artificial intelligence model in step (d) is trained to distinguish tumor-derived mutations based on features consisting of functional features, mutation patterns, and technical features of cancer; The functional characteristics of the cancer at stage (d) include: (i) The accumulation pattern of single gene mutations (regional mutation density, RMD) and (ii) characterized in that it includes one or more features selected from the group consisting of replication time, H3K4Me1, H3K4Me3, H3K9Me3, H3K27Me3, H3K36Me3, Dnase I hypersensitive site (DHS), protein binding site (footprint) gene expression level in DHS, positive selection score, and negative selection score; The mutation pattern in step (d) is at least one selected from the group consisting of C->A, C->G, C->T, T->A, T->C, and T->G; The technical characteristics of the mutation in step (d) include the average read depth of the reference allele in the mutation region, the average mapping quality, the average base quality, the average number of mismatches, the average position of the reference allele, the average total base quality of the mismatches, the number or position of bases with a phread quality of 2 at the 3' end, the average clipped read length, the average position of the read from the 3' end, the ratio of plus strand reads, the length of the DNA fragment, and the average position of the read from the 3' end. length); Average read depth of variant alleles in the variant region, average mapping quality, average base quality, average number of mismatches, average position of reference allele, average total base quality of mismatches, number or position of bases with phread quality of 2 at the 3' end, average clipped read length, average position of reads from the 3' end, ratio of plus strand reads, DNA fragment length; DNA fragment length; Fragment ratio; and MUT.notBoth (the number of DNA fragments that do not overlap at the mutation position in forward and reverse reads + the number of fragments that have overlaps but do not have the same mutation) is one or more features selected from the group consisting of.

2. 2. The method of claim 1, wherein step (a) is performed by a method comprising the steps of: (ai) obtaining nucleic acid from a biological sample; (a-ii) removing proteins, fats, and other residues from the collected nucleic acids using a salting-out method, a column chromatography method, or a beads method to obtain purified nucleic acids; (a-iii) preparing a single-end sequencing or pair-end sequencing library from the purified nucleic acids or the nucleic acids randomly fragmented by enzymatic cleavage, crushing, or hydroshear method; (a-iv) reacting the prepared library with a next-generation gene sequencer; and (a-v) Steps for obtaining nucleic acid sequence information (reads) using a next-generation gene sequencer.

3. 2. The method of claim 1, further comprising the step of selecting reads having a mapping quality score of aligned nucleic acid fragments equal to or greater than a reference value before performing step (c).

4. The method for detecting tumor-derived mutations in cell-free DNA based on artificial intelligence according to claim 3, wherein the reference value is 50 to 70 points.

5. 2. The method of claim 1, wherein the step of detecting mutations in step (c) comprises the steps of: (c-i) selecting a base sequence that differs from a reference gene from the aligned reads; and (c-ii) storing the selected base sequence information.

6. The method of claim 1 , wherein the features of step (d) include the features listed in Table 1 below. 【Table 1】

7. 2. The method of claim 1, wherein the artificial intelligence model in step (d) is trained to distinguish whether or not the mutation is a tumor-derived mutation.

8. 8. The method of claim 7, wherein the artificial intelligence model is at least one selected from the group consisting of random forest, XGboost, and deep neural network.

9. 1. A computer-implemented method for providing information for early diagnosis of cancer, the method comprising the steps of: (a) detecting tumor-derived mutations in cell-free DNA by the method of any one of claims 1 to 8; and (b) determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected.

10. a decoding unit for extracting nucleic acid from a biological sample and decoding sequence information; an alignment section that aligns the decoded sequence to a standard chromosomal sequence database; a mutation detection unit that detects mutations based on the aligned sequence information; A tumor-derived mutation detection unit that inputs the detected mutations into an artificial intelligence model trained to distinguish tumor-derived mutations and determines the presence or absence of tumor-derived mutations; and a cancer diagnostic component that determines that there is cancer or minimal residual cancer if a tumor-derived mutation is detected; An information providing device for early cancer diagnosis based on artificial intelligence, comprising: The providing device may perform an artificial intelligence-based method for detecting tumor-derived mutations in cell-free DNA, the method comprising the steps of: (a) extracting nucleic acid from a biological sample and obtaining sequence information; (b) aligning the obtained sequence information (reads) to a reference genome database; (c) detecting mutations based on the aligned sequence reads; and (d) inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations, and comparing the output value with a cut-off value to determine the presence or absence of a tumor-derived mutation; The artificial intelligence model in step (d) is trained to distinguish tumor-derived mutations based on features consisting of functional features, mutation patterns, and technical features of cancer; The functional characteristics of the cancer at stage (d) include: (i) The accumulation pattern of single gene mutations (regional mutation density, RMD) and (ii) characterized in that it includes one or more features selected from the group consisting of replication time, H3K4Me1, H3K4Me3, H3K9Me3, H3K27Me3, H3K36Me3, Dnase I hypersensitive site (DHS), protein binding site (footprint) gene expression level in DHS, positive selection score, and negative selection score; The mutation pattern in step (d) is at least one selected from the group consisting of C->A, C->G, C->T, T->A, T->C, and T->G; The technical characteristics of the mutation in step (d) include the average read depth of the reference allele in the mutation region, the average mapping quality, the average base quality, the average number of mismatches, the average position of the reference allele, the average total base quality of the mismatches, the number or position of bases with a phread quality of 2 at the 3' end, the average clipped read length, the average position of the read from the 3' end, the ratio of plus strand reads, the length of the DNA fragment, and the average position of the read from the 3' end. length); Average read depth of variant alleles in the variant region, average mapping quality, average base quality, average number of mismatches, average position of reference allele, average total base quality of mismatches, number or position of bases with phread quality of 2 at the 3' end, average clipped read length, average position of reads from the 3' end, ratio of plus strand reads, DNA fragment length; DNA fragment length; Fragment ratio; and MUT.notBoth (the number of DNA fragments that do not overlap at the mutation position in forward and reverse reads + the number of fragments that have overlaps but do not have the same mutation) is one or more features selected from the group consisting of.

11. 1. A computer-readable storage medium, comprising instructions configured to be executed by a processor to provide information for early diagnosis of cancer; (a) extracting nucleic acid from a biological sample and obtaining sequence information; (b) aligning the obtained sequence information (reads) to a reference genome database; (c) detecting mutations based on the aligned sequence reads; (d) inputting the detected mutation information into an artificial intelligence model trained to distinguish tumor-derived mutations, and comparing the output value with a cut-off value to determine the presence or absence of a tumor-derived mutation; The artificial intelligence model in step (d) is characterized in that it is trained to distinguish tumor-derived mutations based on features consisting of functional features, mutation patterns, and technical features of cancer; and (e) a computer-readable storage medium comprising instructions configured to be executed by a processor to provide information for early diagnosis of cancer by determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected, the tumor-derived mutation is a tumor-derived mutation in cell-free DNA; The functional characteristics of the cancer at stage (d) include: (i) The accumulation pattern of single gene mutations (regional mutation density, RMD) and (ii) characterized in that it includes one or more features selected from the group consisting of replication time, H3K4Me1, H3K4Me3, H3K9Me3, H3K27Me3, H3K36Me3, Dnase I hypersensitive site (DHS), protein binding site (footprint) gene expression level in DHS, positive selection score, and negative selection score; The mutation pattern in step (d) is at least one selected from the group consisting of C->A, C->G, C->T, T->A, T->C, and T->G; The technical characteristics of the mutation in step (d) include the average read depth of the reference allele in the mutation region, the average mapping quality, the average base quality, the average number of mismatches, the average position of the reference allele, the average total base quality of the mismatches, the number or position of bases with a phread quality of 2 at the 3' end, the average clipped read length, the average position of the read from the 3' end, the ratio of plus strand reads, the length of the DNA fragment, and the average position of the read from the 3' end. length); Average read depth of variant alleles in the variant region, average mapping quality, average base quality, average number of mismatches, average position of reference allele, average total base quality of mismatches, number or position of bases with phread quality of 2 at the 3' end, average clipped read length, average position of reads from the 3' end, ratio of plus strand reads, DNA fragment length; DNA fragment length; Fragment ratio; and MUT.notBoth (the number of DNA fragments that do not overlap at the mutation position in forward and reverse reads + the number of fragments that have overlaps but do not have the same mutation) is one or more features selected from the group consisting of.

12. 1. A computer-implemented method for early diagnosis of cancer, comprising the steps of: (a) detecting tumor-derived mutations in cell-free DNA by the method of any one of claims 1 to 8; and (b) determining that there is cancer or microscopic residual cancer if a tumor-derived mutation is detected.

Images