KR102412631B1 - A system for predicting base-editing efficiency and outcome product frequencies of base editors - Google Patents

Abstract

The present invention relates to a system for predicting nucleotide correction efficiency and result of nucleotide editing, and a method for predicting the efficiency and result of nucleotide correction using the system. Using the prediction system according to one aspect, it is possible to predict the efficiency and accuracy in a simple way without the need to individually manufacture and verify the gene scissors, so that it is possible to select the gene scissors that can be safely edited. Furthermore, among pathogenic/pseudopathogenic human point mutation diseases, it is possible to predict the efficiency and frequency of results in cases in which a disease can be made or corrected by nucleotide editing, so that a target disease for nucleotide correction can be selected.

Description

A system for predicting base-editing efficiency and outcome product frequencies of base editors

The present invention relates to a system for predicting nucleotide correction efficiency and result of nucleotide editing, and a method for predicting the efficiency and result of nucleotide correction using the system.

Point mutations are the most common form of pathogenic mutation, accounting for more than half of pathogenic or likely pathogenic mutations in humans, but this frequency may be biased due to the widespread use of short-read sequencing. have. The generation of these pathogenic point mutations in normal cells and organisms can lead to the development of relevant disease models. Conversely, correction of pathogenic point mutations in cells and organisms with the mutation can provide an isogenic control for studies of the effects of these point mutations. In addition, correction of these pathogenic point mutations can be a therapeutic modality for related diseases. In both the creation and correction of pathogenic point mutations, base editors convert one base pair to another in a targeted manner without generating double-strand breaks or requiring a donor DNA template. Direct conversion makes it a very attractive genome editing tool. Adenine base editors (ABEs) can convert A and T base pairs to G and C base pairs, and cytosine base editors (CBEs) convert G and C base pairs to A, T base pairs. can be converted to

However, if i) base editing efficiency is low and/or ii) unwanted concurrent mutations occur as a result of base editing, an editable window - i.e. a large number of target nucleotides in the base editing range If there is, it may be difficult to correct a disease model and pathogenic mutation induced by such nucleotide-correction gene scissors. Therefore, the efficiency and outcome of base editing is often measured experimentally before or during disease model generation, during therapeutic correction of pathogenic mutations. However, this experimental evaluation involves the preparation of single-guide RNA (sgRNA), delivery of ABE or CBE with the sgRNA, harvesting of cells containing these components, PCR amplification of the target sequence, and subsequent sequencing. It is a time-consuming, multi-step process that involves In addition, in order to use base editing as a tool for efficient high-throughput screening, it is necessary to know the efficiency and result of base editing in each target sequence. However, when thousands of target sequences are studied, the conventional evaluation method for evaluating the efficiency at each individual site as described above is not a practical approach. Also, such an assessment cannot be carried out if the cells derived from the patient contain the relevant mutation. In other words, there is no way to predict the frequency of various nucleotide correction results that nucleotide correction gene scissors can make.

Accordingly, the present inventors have tried to develop an in silico method that can predict the correction result through the nucleotide correction efficiency of nucleotide correction gene scissors and the nucleotide editing frequency for each position. As a result, the efficiency and frequency data of the correction result were produced in about 13,000 and 14,000 target sequences, respectively, for adenine nucleotide editing and cytosine nucleotide correction, respectively, and a target that significantly affects the efficiency of nucleotide correction By exploring the sequence context around the base, we develop a system that can simultaneously predict the efficiency and accuracy of base-correction gene scissors through the deep learning method based on the large-scale data, and furthermore, base correction for human point mutation diseases By predicting the efficiency of gene scissors, the present invention was completed by confirming that diseases that can be made with base-correction gene scissors and diseases that can be corrected can be selected.

the work aspect

a target sequence input unit for receiving a target sequence of the base editing gene scissors; and

By applying the target sequence input from the target sequence input unit to the efficiency prediction model and the correction result prediction model, respectively, the efficiency and correction result score of the nucleotide correction gene scissors are obtained, and the efficiency score and the correction result score are multiplied by the nucleotide correction gene scissors. An object of the present invention is to provide a nucleotide correction efficiency and result prediction system for nucleotide correction gene scissors including a result prediction unit that simultaneously predicts the efficiency and the result.

the other aspect

designing a target sequence of the base editing gene scissors; and

applying the designed target sequence to the base correction efficiency and result prediction system according to an aspect; It is to provide a method for predicting the nucleotide correction efficiency and result of nucleotide correction gene scissors including a.

Another aspect is

obtaining human point mutation data;

When the point mutation occurs from the human point mutation data, the normal base adenine (A) is changed to the abnormal base guanine (G); When the normal base guanine (G) is replaced with the abnormal base adenine (A); When the normal base cytosine (C) is replaced with the abnormal base thymine T); or first selecting data corresponding to a case in which the normal base thymine (T) is changed to the abnormal base cytosine (C);

Secondarily selecting data in which a point mutation exists at a position 3 to 10 bp from the 5' end of the protospacer region from among the firstly selected data;

thirdly selecting data corresponding to pathogenic or pseudopathogenic point mutations from among the secondarily selected data; and

It is to provide a method of providing information on human point mutation-related diseases that can use nucleotide correction gene scissors, including applying the tertiary selected data to a nucleotide correction efficiency and result prediction system according to an aspect.

Another aspect is to provide a computer-readable recording medium in which a program for executing the method by a computer is recorded.

Another aspect is

introducing base-correcting gene scissors into cells; A method of editing a target nucleotide in the genome of a cell comprising:

The base editing gene is (i) an RNA-guided nuclease or a gene encoding the same, (ii) a deaminase or a gene encoding the same, and (iii) a guide RNA capable of hybridizing with a target sequence or a guide RNA encoding the same contains a gene;

wherein the target sequence comprises a PAM sequence, a protospacer sequence, and a sequence complementary to a guide RNA,

The sequence complementary to the guide RNA is 5'-TAC-3', 5'-TAT-3', 5'-TAG-3', 5'-GAT-3', 5'-CAC-3', 5' -GAC-3', 5'-CAT-3', 5'-TAA-3', 5'-CAG-3', 5'-GAG-3', 5'-AAC-3', 5'-CAA -3', 5'-GAA-3', 5'-AAT-3', 5'-AAG-3', 5'-AAA-3', 5'-TCC-3', 5'-TCG-3 ', 5'-TCT-3', 5'-TCA-3', 5'-CCC-3', 5'-CCT-3', 5'-ACC-3', 5'-CCA-3', 5'-CCG-3', 5'-ACG-3', 5'-ACT-3', 5'-ACA-3', 5'-GCC-3', 5'-GCT-3', 5' -GCG-3', and 5'-GCA-3' comprising a sequence selected from the group consisting of,

The deaminase is to provide a method for editing a target nucleotide in the genome of a cell, characterized in that the deamination of adenine or cytosine in the target sequence.

Another aspect is

As a pharmaceutical composition for the prevention or treatment of human point mutation-related diseases comprising base-correcting gene scissors,

The base editing gene is (i) an RNA-guided nuclease or a gene encoding the same, (ii) a deaminase or a gene encoding the same, and (iii) a guide RNA capable of hybridizing with a target sequence or a guide RNA encoding the same contains a gene;

wherein the target sequence comprises a PAM sequence, a protospacer sequence, and a sequence complementary to a guide RNA,

The sequence complementary to the guide RNA is 5'-TAC-3', 5'-TAT-3', 5'-TAG-3', 5'-GAT-3', 5'-CAC-3', 5' -GAC-3', 5'-CAT-3', 5'-TAA-3', 5'-CAG-3', 5'-GAG-3', 5'-AAC-3', 5'-CAA -3', 5'-GAA-3', 5'-AAT-3', 5'-AAG-3', 5'-AAA-3', 5'-TCC-3', 5'-TCG-3 ', 5'-TCT-3', 5'-TCA-3', 5'-CCC-3', 5'-CCT-3', 5'-ACC-3', 5'-CCA-3', 5'-CCG-3', 5'-ACG-3', 5'-ACT-3', 5'-ACA-3', 5'-GCC-3', 5'-GCT-3', 5' -GCG-3', and 5'-GCA-3' comprising a sequence selected from the group consisting of,

The deaminase is to provide a pharmaceutical composition for preventing or treating a human point mutation-related disease, characterized in that the deamination of adenine or cytosine in the target sequence.

Other objects and advantages of the present application will become more apparent from the following detailed description in conjunction with the appended claims and drawings. Content not described in this specification will be omitted because it can be sufficiently recognized and inferred by those skilled in the technical field or similar technical field of the present application.

Each description and embodiment disclosed in this application is also applicable to each other description and embodiment. That is, all combinations of the various elements disclosed in the present application fall within the scope of the present application. In addition, it cannot be seen that the scope of the present application is limited by the detailed description described below.

the work aspect

a target sequence input unit for receiving a target sequence of the base editing gene scissors; and

By applying the target sequence input from the target sequence input unit to the efficiency prediction model and the correction result prediction model, respectively, the efficiency and correction result score of the nucleotide correction gene scissors are obtained, and the efficiency score and the correction result score are multiplied by the nucleotide correction gene scissors. To provide a system for predicting nucleotide correction efficiency and result of nucleotide correction gene scissors including a result prediction unit for simultaneously predicting efficiency and result.

As used herein, the term "base editor" is a new type of gene scissors derived from the CRISPR gene editing technique called the fourth-generation gene editing technique. Unlike the existing third-generation gene scissors that cut both strands of DNA, the base-correcting gene scissors works by replacing a single base. The base editing gene scissors consists of Nickase Cas9 (nCas9) that cuts one strand of DNA and a deaminase that breaks down adenine or cytosine. Specifically, dCas9 (“dead” Cas9) that removes the double-stranded DNA cutting function of CRISPR/Cas9 Alternatively, adenine base editor (ABE) and cytosine deaminase that can replace adenine (A) with guanine (G) by binding adenine deaminase to nCas9 There is Cytosine Base Editor (CBE), which can be combined to find only cytosine (C) in the DNA sequence and replace it with thymine (T). For example, in the case of CBE, when deaminase replaces cytosine (C) with uracil (U) in one strand of DNA cut with nCas9 or dCas9, the base changed to uracil (U) is converted to thymine (T) by the DNA repair process. It works on this principle. Using base-correcting gene scissors, a gene can be deleted or transformed into a desired trait by correcting or replacing a specific sequence.

From the 5' end of the protospacer region, 3 to 10 bp positions (20 bp positions are located immediately upstream of the PAM sequence (5'-NGG'-3') through high-throughput experiments. ), secure large-scale data on the activity confirmation and nucleotide correction results of nucleotide editing for 13,504 target sequences containing at least one target adenine and 14,157 target sequences containing at least one target cytosine, and construct a convolutional neural network By combining two models, the efficiency prediction model and the correction result prediction model built with deep learning used, the DeepABE and DeepCBE prediction system that can predict the nucleotide correction efficiency and all nucleotide editing results that nucleotide correction gene scissors can make. It was developed (DeepBaseEditor), and it was confirmed that the efficiency of the base editing gene scissors and the correction result could be predicted at the same time by verifying the accuracy of the prediction system.

The present inventors confirmed that the nucleotide correction efficiency and result prediction system of the constructed nucleotide correction gene scissors has superior performance compared to the traditional machine learning-based algorithm.

As used herein, the term "guide RNA (guide RNA)" refers to a target DNA-specific RNA, and all or part complementarily binds to the target sequence so that the adenine deaminase or cytosine deaminase of the base editing gene scissors is the target sequence In each of them, adenine (A) can be found and replaced with guanine (G), and cytosine (C) can be found and replaced with thymine (T).

Typically, the guide RNA comprises two RNAs, i.e., a crRNA (CRISPR RNA) and a tracrRNA (trans-activating crRNA) as a component; a dual RNA; Or it refers to a form comprising a first site comprising a sequence that is completely or partially complementary to a sequence in the target DNA and a second site comprising a sequence that interacts with an RNA-guided nuclease, but RNA- The guide nuclease may be included in the scope of the present invention without limitation as long as it has a form capable of having activity in the target sequence.

In addition, the guide RNA may include a scaffold sequence that helps the RNA-guided nuclease to be attached.

As used herein, the term “target sequence” or “target sequence” refers to a nucleotide sequence expected to be targeted by the nucleotide-correcting gene scissors. Specifically, as a sequence that is expected to be targeted by the editing gene through the guide RNA, it may be a sequence known to exhibit activity, or a sequence to be analyzed by those skilled in the art using the system of the present invention. It may be an arbitrarily designed sequence, but any sequence to be analyzed because it has or is expected to have an activity of nucleotide editing may be included without limitation in the scope of the present invention.

Herein, the activity data of the base editing gene scissors can be obtained by introducing the base editing gene scissors into a cell library including an oligonucleotide including a nucleotide sequence encoding a guide RNA and a target sequence desired by the guide RNA. , but not limited thereto.

As used herein, the term "RNA-guided nuclease" refers to a nuclease capable of recognizing and cleaving a specific position on a desired genome, particularly a nuclease having target specificity by a guide RNA. The RNA-guided nuclease may include, but is not limited to, Cas9 (CRISPR-Associated Protein 9) and Cpf1.

As used herein, "Cas9 protein" is a major protein component of the CRISPR/Cas9 system, and forms a complex with crRNA (CRISPR RNA) and tracrRNA (trans-activating crRNA) to form an activated endonuclease or nickase ( nickase) is formed.

Cas9 protein or gene information can be obtained from a known database such as GenBank of the National Center for Biotechnology Information (NCBI), but any one capable of having target-specific nuclease activity together with guide RNA is included in the scope of the present invention. can In addition, the Cas9 protein may be linked to a protein transduction domain. The protein transduction domain may be polyarginine or HIV-derived TAT protein, but is not limited thereto. Furthermore, according to the purpose of the Cas9 protein, an additional domain may be appropriately linked by a person skilled in the art.

The Cas9 protein may include not only wild-type Cas9, but also inactivated Cas9 (dCas9), or a variant of Cas9 such as Cas9 nickase. The inactivated Cas9 may be an RNA-guided FokI nuclease (RFN) linking a FokI nuclease domain to dCas9, or a transcription activator or repressor domain linking a dCas9 to dCas9, The Cas9 kinase may be D10A Cas9 or H840A Cas9, but is not limited thereto.

The Cas9 protein is not limited to its origin. For example, the Cas9 protein may include Streptococcus pyogenes, Francisella novicida, Streptococcus thermophilus, Legionella pneumophila, Listeria innocua). Or Streptococcus mutans (Streptococcus mutans) may be derived.

Herein, in order to express Cas9 in a viral vector, a vector capable of expressing a part of Cas9 was constructed. That is, the Cas9 protein was divided into sizes capable of packaging in a viral vector and expressed in each vector. The Cas9 protein produced in the above manner is called split-Cas9, and split-Cas9 divides the Cas9 protein, which has not been packaged through a viral vector, etc., due to its large size, into packagingable sizes and expresses each of them through a vector. It is characterized in that it does not lose its function in the cell.

Cas9 protein herein may preferably be any one selected from the group consisting of dCas9, nCas9 and SpCas9. In one embodiment, intein-mediated split-Cas9-based ABE and CBE were used.

In one embodiment, the efficiency prediction model is

receiving activity data of base-correcting gene scissors through an information input unit; and

It may be generated through the step of generating an efficiency prediction model by performing deep learning based on a convolutional neural network (CNN) using the data received from the information input unit.

In one embodiment, the calibration result prediction model is

Receiving an input through an information input unit for receiving correction result data of the base correction gene scissors; and

It may be generated through the step of performing deep learning based on a convolutional neural network (CNN) using the data input from the information input unit to generate a correction result prediction model.

As used herein, the term "activity" of nucleotide editing is an activity in which a single base is replaced by nucleotide editing, that is, an RNA-guided nuclease, specifically Cas9, cleaves a gene in a target sequence, and a deaminase It refers to the activity of converting adenine (A) to guanine (G) or cytosine (C) to thymine (T). As used herein, the term "activity data" corresponds to data that can extract and learn the relationship between a specific target sequence and the base-correcting gene scissors, and the system of the present invention can generate an efficiency prediction model using the activity data. have.

Specifically, the activity data of the base-correcting gene scissors can be obtained by sequencing the bases of the target sequence. For example, data may be obtained by performing deep sequencing or RNAseq, but as long as activity data of base-correcting gene scissors can be obtained through detection of an edited base, it is not limited to a specific method.

The form, type, size, etc. of data are not limited as long as the base-editing gene scissors can exhibit the activity shown in the target sequence.

The activity data of the base editing gene scissors may be existing known activity data, or may be activity data directly obtained by any method that can be appropriately adopted by those skilled in the art. For the purposes of the present invention, the base editing gene scissors activity data The method by which the data is obtained is not limited, as long as it is data capable of generating an activity prediction model capable of predicting the activity of

In one embodiment, the activity data of the base editing gene scissors may be in consideration of the sequence context around the target nucleotide of the base editing gene scissors target.

As used herein, the term “sequence context” refers to sequence information around a target nucleotide for which base editing is to be performed.

In one embodiment, the activity data of the base editing gene is

introducing nucleotide-correcting gene scissors into a cell library comprising a nucleotide sequence encoding a guide RNA and an oligonucleotide comprising a target sequence desired by the guide RNA;

performing deep sequencing using the DNA isolated from the cell library into which the base correction gene scissors have been introduced; and

It can be obtained through the step of detecting whether the target nucleotide is converted within the base correction range from the sequence data obtained from the deep sequencing.

As used herein, the term "base correction range" or "editable window" refers to a nucleotide correction range in which nucleotide editing in a target sequence exhibits activity. In one embodiment, the "base correction range" means a range between positions 3 and 10 bp in the 5' to 3' direction among 20 positions in the protospacer targeted by the guide RNA, which is referred to as a wide editable window. do. In one embodiment, in the case of a narrow editable window, it means a range between positions 4 and 8 bp in the 5' to 3' direction out of 20 positions in the protospacer that the guide RNA targets.

As used herein, the term "correction result" or "editing result" of nucleotide editing refers to an editing product made as a result of nucleotide editing activity for a target sequence. On the other hand, if there are a plurality of editable target nucleotides within the base editing range (editable window), undesired bases may be edited. " means the frequency of each product produced as a result of the activity of base-correcting gene scissors. Many of the human point mutation-related diseases have multiple identical bases within the range of base editing, so it is important to predict the editing frequency for each location in advance to safely use base editing scissors.

In one embodiment, the data of the correction result of the base correction gene scissors is

introducing nucleotide-correcting gene scissors into a cell library comprising a nucleotide sequence encoding a guide RNA and an oligonucleotide comprising a target sequence desired by the guide RNA;

performing deep sequencing using the DNA isolated from the cell library into which the base correction gene scissors have been introduced;

It can be obtained through the step of detecting the target nucleotide conversion frequency within the range of base correction from the sequence data obtained from the deep sequencing.

The calibration result data of the base editing gene scissors may be known data, or may be activity data directly obtained by any method that can be appropriately adopted by those skilled in the art. For the purpose of the present invention, the base editing gene The method by which the data is obtained is not limited as long as it is data capable of generating a correction result prediction model capable of predicting the correction result of scissors.

In the present application, each of the efficiency prediction model and the correction result prediction model uses a known database in which activity data or correction result data of base correction gene scissors is stored, and is to be generated through deep learning technology using large-scale data input from the database. can That is, the activity data or calibration result data of the base-correction gene scissors can be obtained from known databases, literature, etc., in addition to those obtained by direct measurement, or obtained by secondary processing of data obtained from the database or literature, Any data capable of predicting the efficiency and editing results of nucleotide editing for any target sequence by extracting the relationship between the sequence and the nucleotide editing efficiency or nucleotide editing frequency and combining the extracted features can be used without limitation.

As used herein, the term “oligonucleotide” refers to a material in which several to hundreds of nucleotides are linked by a phosphodiester bond, and for the purposes of the present invention, the oligonucleotide may be double-stranded DNA. As used herein, the oligonucleotide may have a length of 10 to 300 bp, preferably 50 to 200 bp, more preferably 100 to 180 bp, but is not limited thereto, depending on the purpose of analysis, etc. It can be appropriately adjusted by those skilled in the art.

The oligonucleotide herein includes a guide RNA coding nucleotide sequence and a target sequence. In addition, the oligonucleotide may include an additional sequence to which a primer can be bound so that it can be PCR amplified.

The target sequence may have a length of 10 to 100 bp, preferably 20 to 50 bp, more preferably 20 to 30 bp, and most preferably 24 to 26 bp, but is not particularly limited thereto.

In addition, the guide RNA coding sequence may have a length of 10 to 100 bp, preferably 15 to 50 bp, more preferably 20 to 30 bp, but is not particularly limited thereto.

The oligonucleotide may further include a barcode sequence.

The barcode sequence refers to a nucleotide sequence for identifying each oligonucleotide. The barcode sequence herein may not include two or more repeating nucleotides (AA, TT, CC, GG), but is not particularly limited thereto as long as it is designed to identify each oligonucleotide. In the plurality of oligonucleotides, the barcode sequence may be designed so that at least two bases are different so that each oligonucleotide can be identified. The barcode sequence may have a length of 5 to 50 bp, but is not particularly limited thereto.

The oligonucleotide may be introduced into a cell and integrated into a chromosome.

As used herein, the term “library” refers to a pool or population including two or more types of the same material having different characteristics. Accordingly, the oligonucleotide library may be a population comprising two or more types of oligonucleotides having different base sequences, such as guide RNA, and/or two types of oligonucleotides having different target sequences, and the cell library is two or more types of cells with different characteristics. , Specifically, for the purpose of the present invention, the oligonucleotides included in each cell may be different, for example, an introduced guide RNA, and/or a target sequence, or a population of cells different from each other.

As used herein, the term "vector" refers to a medium, such as a genetic construct, that enables the delivery of the oligonucleotide into a cell, and the vector herein is an oligonucleotide comprising each guide RNA coding sequence and a target sequence. may include The vector may be a viral vector or a plasmid vector, and the viral vector may specifically be a lentiviral vector or a retroviral vector, but is not limited thereto and a vector known to those skilled in the art as long as it can achieve the object of the present invention. can be used freely.

Specifically, the vector may include an essential regulatory element operably linked to the insert, ie, the oligonucleotide, so that the insert can be expressed when present in a cell of an individual.

The vector can be prepared and purified using standard recombinant DNA techniques. The type of the vector is not particularly limited as long as it can act in target cells such as prokaryotic cells and eukaryotic cells. A vector may include a promoter, an initiation codon, and a stop codon terminator. In addition, DNA encoding the signal peptide, and/or enhancer sequence, and/or the untranslated region on the 5' side and 3' side of the desired gene, and/or a selectable marker region, and/or a replicable unit, etc. are appropriately added may include

A method of delivering the vector to a cell for preparing a library can be accomplished using various methods known in the art. For example, calcium phosphate-DNA co-precipitation method, DEAE-dextran-mediated transfection method, polybrene-mediated transfection method, electroshock method, microinjection method, liposome fusion method, lipofectamine and protoplast fusion method, etc. It can be carried out by a number of known methods. In addition, in the case of using a viral vector, a target object, that is, the vector can be delivered into a cell using viral particles by means of infection. In addition, the vector can be introduced into the cell by gene bambadment or the like.

The introduced vector may exist as a vector itself in a cell or may be integrated into a chromosome, but is not particularly limited thereto.

The cell library prepared herein refers to a cell population into which an oligonucleotide comprising a guide RNA-target sequence has been introduced. In this case, each cell may be introduced with a vector, specifically, a different type and/or number of viruses.

The type of cell into which the vector can be introduced may be appropriately selected by those skilled in the art depending on the type of vector and/or the type of target cell, for example, bacterial cells such as Escherichia coli, Streptomyces, Salmonella typhimurium; yeast cells; Fungal cells such as Pichia pastoris; insect cells such as Drosophila and Spodoptera Sf9 cells; CHO (chinese hamster ovary cells), SP2/0 (mouse myeloma), human lymphoblastoid, COS, NSO (mouse myeloma), 293T, Bow melanoma cells, HT-1080, BHK ( animal cells such as baby hamster kidney cells, HEK (human embryonic kidney cells), and PERC.6 (human retinal cells); or plant cells.

As used herein, the term "information input unit" refers to a component that receives the above-described activity data or correction result data of the base-correction gene scissors, and the information input unit is directly inserted into the base-correction gene scissors by the user of the prediction system according to an embodiment. It may be to receive related data or to receive pre-stored data, but is not limited thereto.

The system of the present invention may further include, but is not limited to, a storage unit for storing previously obtained data on nucleotide-correction gene scissors or known data on nucleotide-correction gene scissors. When the storage unit is included, the information input unit of the system of the present invention may receive data of a size or range set from the storage unit, and may be used to predict the activity of the base editing gene scissors or the correction result.

In one embodiment, the "efficiency prediction model" and the "correction result prediction model" use the data about the nucleotide correction gene inputted through the information input unit, and use the target sequence, the nucleotide editing result, and the frequency of the nucleotide editing result. By extracting and combining the relationship, it can be generated through the step of generating a predictive model capable of learning the relationship between the target sequence and the base-correcting gene scissors. The efficiency prediction model and the correction result prediction model are generated using deep learning technology based on the learned information, and the user of the prediction system according to an embodiment can check the efficiency and correction result of the base correction gene scissors through the prediction model predictable.

Specifically, the prediction model of the present application is based on a convolutional neural network (CNN) to perform deep-learning to learn the relationship between the target sequence and the base editing result, and the frequency of the base editing result. It is not limited thereto.

As used herein, the term "Deep Learning" is an artificial intelligence (AI) technology that enables computers to think and learn like humans. it is technology By using the deep learning technology, a computer can recognize, reason, and judge by itself without a person setting all judgment criteria, and it is possible to use it extensively for voice image recognition and photo analysis.

In other words, deep learning is a machine learning that attempts high-level abstractions (summarizing core contents or functions in large amounts of data or complex data) through a combination of several nonlinear transformation methods. learning) can be defined as a set of algorithms.

As used herein, the term "convolutional neural networks (CNN)" refers to a technique for extracting a feature representing a part of provided information and generalizing the information through layering.

The present inventors produce a large amount of data on the efficiency and calibration results of gene scissors using a mass measurement method of gene scissors activity, and actual experimental results based on a deep learning framework using powerful convolutional neural networks (CNNs) Develop an active prediction model with high reliability in which the correlation between the value and the predicted value suggested by artificial intelligence converges to 0.69 to 0.79, and a correction result prediction model with high reliability in which the correlation reaches 0.91 to 0.93, and By combining the two models, a prediction system named DeepABE and DeepCBE was developed. Furthermore, the accuracy was verified in biologically replicated samples and human induced pluripotent stem cells.

As used herein, the term “result prediction unit” refers to the efficiency prediction model and the correction result prediction model constructed by the above-described method by applying the target sequence input through the target sequence input unit to predict the nucleotide correction efficiency and result of the nucleotide correction gene scissors. is a configuration that In one embodiment, the result prediction unit can predict the nucleotide correction efficiency and result of nucleotide correction from the target sequence information, but factors for increasing the accuracy of prediction, for example, the sequence context or chromatin around the target nucleotide Accessibility may be additionally considered.

Specifically, the result prediction unit may predict whether or not the base editing of the target sequence or the frequency of the base editing by the base editing gene scissors by a preset method, but is not limited thereto. If the result prediction unit is an indicator capable of predicting the activity of other nucleotide-correcting gene scissors in addition to the presence of nucleotide editing or nucleotide-editing frequency, it can be used to predict the activity of nucleotide-correction gene scissors irrespective of the type, form, or prediction method.

In one embodiment, the result prediction unit applies the target sequence input from the target sequence input unit to the efficiency prediction model and the correction result prediction model, respectively, to obtain the efficiency and correction result score of the base correction gene scissors, and the efficiency score and the correction result By multiplying the score, the nucleotide correction efficiency and results of nucleotide correction gene scissors can be predicted.

The efficiency score may be calculated using the following [Equation 1] for each position of the target sequence.

[Equation 1]

In addition, the calibration result score may be calculated using the following [Equation 2].

[Equation 2]

The prediction system according to an aspect further includes a fine-tuning unit for optimizing (fine-tuning) the activity of the base-correcting gene scissors predicted in the result prediction unit using the sequence context around the target nucleotide or chromatin accessibility information of the target sequence can do.

The "fine adjustment unit" of the present invention provides not only the sequence information of the input target sequence, but also the sequence context around the target nucleotide or the target sequence of the nucleotide editing gene scissors, in order to increase the nucleotide correction efficiency of nucleotide correction and the accuracy of the result prediction system. It refers to a configuration that optimizes the activity of the base-correcting gene scissors predicted in the efficiency prediction model by considering the accessibility of chromatin to

The chromatin accessibility information may be obtained from known databases, literature, etc., or may be directly measured, and specifically may be calculated from the sensitivity of the target sequence to DNase I, but is not limited thereto.

In one embodiment, the nucleotide correction efficiency and result prediction system of the nucleotide correction gene scissors may further include an output unit for outputting the efficiency and result of the nucleotide correction gene scissors predicted by the result prediction unit.

The information on the nucleotide correction efficiency and result of the nucleotide correction gene scissors output by the output unit is not limited in the form or type of the output signal.

Another aspect includes the steps of designing a target sequence of the base editing gene; and

It provides a nucleotide correction efficiency and result prediction method of nucleotide correction gene scissors, comprising applying the designed target sequence to a nucleotide correction efficiency and result prediction system according to an aspect.

According to the method according to one aspect, by predicting the base editing efficiency and the frequency of the editing result through the prediction model in which the correlation between the actual experimental result value and the prediction value presented by the prediction model converges to 0.50 to 0.95, safe correction is possible It is possible to select the gene scissors and provide disease information capable of making or correcting a disease model with the gene scissors.

the other aspect

obtaining human point mutation data;

When the point mutation occurs from the human point mutation data, the normal base adenine (A) is changed to the abnormal base guanine (G); When the normal base guanine (G) is replaced with the abnormal base adenine (A); When the normal base cytosine (C) is replaced with the abnormal base thymine T); or first selecting data corresponding to a case in which the normal base thymine (T) is changed to the abnormal base cytosine (C);

Secondarily selecting data in which a point mutation exists at a position 3 to 10 bp from the 5' end of the protospacer region from among the firstly selected data;

thirdly selecting data corresponding to pathogenic or pseudopathogenic point mutations from among the secondarily selected data; and

It provides a method of providing information on human point mutation-related diseases that can use nucleotide correction gene scissors, including applying the thirdly selected data to a nucleotide correction efficiency and result prediction system according to an aspect.

The human point mutation-associated disease has a point mutation within a range of 3 to 10 bp from the 5' end of the protospacer (herein, "base correction range"), and/or a PAM sequence downstream of the protospacer. In some cases, it occurs because a normal base (A or C) is changed to an abnormal base (G or T) within the base correction range; Alternatively, any disease caused by changing a normal base (G or T) within the base correction range to an abnormal base (A or C) may be included without limitation.

In one embodiment, the human point mutation-related disease is Usher syndrome, tumor necrosis factor receptor-associated periodic syndrome (TNF receptor-associated periodic syndrome: TRAPS), Marfan syndrome (marfan syndrome), type 3 adolescence onset Diabetes (Type 3 form of Maturity-Onset Diabetes of the Young: MODY3), congenital stationary night blindness type 1F, familial hypercholesterolemia, congenital myasthenic syndrome (CMS) ), Lynch syndrome, etc. were confirmed, and in the case of CBE, Loeys-Dietz syndrome (LDS), retinitis pigmentosa, leptin deficiency or dysfunction, family The group consisting of Family hypercholesterolemia, autosomal recessive deafness, cholesterol monooxygenase (side-chain-cleaving) deficiency and progressive myoclonus epilepsy. It may be any one selected from, but is not limited thereto.

Another aspect provides a computer-readable recording medium in which a program for executing the method by a computer is recorded.

The program may be an embodiment of the nucleotide correction efficiency and result prediction system of nucleotide correction gene scissors or the nucleotide correction efficiency and result prediction method of nucleotide correction gene scissors of an aspect in a computer programming language, and the nucleotide correction efficiency of nucleotide correction gene scissors and predicting outcomes.

Computer programming languages that can implement the program of the present invention include, but are not limited to, Python, C, C++, Java, Fortran, Visual Basic, and the like. The program may be stored in a recording medium such as a USB memory, compact disc read only memory (CDROM), hard disk, magnetic diskette, or similar medium or device, and may be connected to an internal or external network system. For example, the computer system accesses a sequence database such as GenBank (http://www.ncbi.nlm.nih.gov/nucleotide) using HTTP, HTTPS, or XML protocol to access a target gene and a regulatory region of the gene. of the nucleic acid sequence can be searched.

The program may be provided online or offline, and it may be provided in the form of a computer program stored in a recording medium in combination with an electronic device implemented by a computer to execute a system for predicting the nucleotide correction efficiency and result of the nucleotide correction gene scissors. .

Another aspect is a method of editing a target nucleotide in the genome of a cell comprising the step of introducing a base editing gene into the cell,

The base editing gene is (i) an RNA-guided nuclease or a gene encoding the same, (ii) a deaminase or a gene encoding the same, and (iii) a guide RNA capable of hybridizing with a target sequence or a guide RNA encoding the same contains a gene;

wherein the target sequence comprises a PAM sequence, a protospacer sequence, and a sequence complementary to a guide RNA,

The sequence complementary to the guide RNA is 5'-TAC-3', 5'-TAT-3', 5'-TAG-3', 5'-GAT-3', 5'-CAC-3', 5' -GAC-3', 5'-CAT-3', 5'-TAA-3', 5'-CAG-3', 5'-GAG-3', 5'-AAC-3', 5'-CAA -3', 5'-GAA-3', 5'-AAT-3', 5'-AAG-3', 5'-AAA-3', 5'-TCC-3', 5'-TCG-3 ', 5'-TCT-3', 5'-TCA-3', 5'-CCC-3', 5'-CCT-3', 5'-ACC-3', 5'-CCA-3', 5'-CCG-3', 5'-ACG-3', 5'-ACT-3', 5'-ACA-3', 5'-GCC-3', 5'-GCT-3', 5' -GCG-3', and 5'-GCA-3' comprising a sequence selected from the group consisting of,

The deaminase provides a method for editing a target nucleotide in the genome of a cell, characterized in that it deaminates adenine or cytosine in the target sequence.

In one embodiment, the RNA-guided nuclease may be selected from the group consisting of SpCas9, nCas9, and dCas9.

In one embodiment, the target nucleotide may be present at a position 3 to 10 bp from the 5' end of the protospacer region.

Another aspect is a pharmaceutical composition for preventing or treating a human point mutation-related disease comprising a base-correcting gene,

The base editing gene is (i) an RNA-guided nuclease or a gene encoding the same, (ii) a deaminase or a gene encoding the same, and (iii) a guide RNA capable of hybridizing with a target sequence or a guide RNA encoding the same contains a gene;

wherein the target sequence comprises a PAM sequence, a protospacer sequence, and a sequence complementary to a guide RNA,

The sequence complementary to the guide RNA is 5'-TAC-3', 5'-TAT-3', 5'-TAG-3', 5'-GAT-3', 5'-CAC-3', 5' -GAC-3', 5'-CAT-3', 5'-TAA-3', 5'-CAG-3', 5'-GAG-3', 5'-AAC-3', 5'-CAA -3', 5'-GAA-3', 5'-AAT-3', 5'-AAG-3', 5'-AAA-3', 5'-TCC-3', 5'-TCG-3 ', 5'-TCT-3', 5'-TCA-3', 5'-CCC-3', 5'-CCT-3', 5'-ACC-3', 5'-CCA-3', 5'-CCG-3', 5'-ACG-3', 5'-ACT-3', 5'-ACA-3', 5'-GCC-3', 5'-GCT-3', 5' -GCG-3', and 5'-GCA-3' comprising a sequence selected from the group consisting of,

The deaminase provides a pharmaceutical composition for preventing or treating a human point mutation-related disease, characterized in that the deamination of adenine or cytosine in a target sequence.

Using the prediction system according to one aspect, it is possible to predict the efficiency and accuracy in a simple way without the need to individually manufacture and verify the gene scissors, so that it is possible to select the gene scissors that can be safely edited. Furthermore, among pathogenic/pseudopathogenic human point mutation diseases, it is possible to predict the efficiency and frequency of results in cases in which a disease can be made or corrected by nucleotide editing, so that a target disease for nucleotide correction can be selected.

1 is a diagram analyzing the characteristics of adenine and cytosine base-correcting gene scissors based on large-scale activity data.
(a, b) the relationship between the location of editable (a) adenine or (b) cytosine in the protospacer and the frequency of base editing measured in a high-throughput manner in the integrated target sequences. Base editing frequencies measured at positions 1 to 20 of the protospacer region in the integrated target sequence are shown. Position 20 is located just upstream of the PAM (NGG). The number of target sequences analyzed (n) was as follows: (a) n = 2,427 to 2,898 for ABE; (b) n = 2,847 to 3,858 for CBE. The top, center, and bottom lines in the box represent the 25th, 50th, and 75th percentiles, respectively. Whiskers represent the 1st and 99th percentile values, respectively.
(c, d) SpCas9-induced indel frequencies and (c) ABE- or (d) CBE-induced base conversion efficiencies were determined for the same integrated target sequence. In order to exclude the effect of nucleotide positions in the protospacer region on the nucleotide editing efficiency, only target sequences having bases corresponding to positions in the same protospacer region were compared. Heat color indicates the number of target sequences in hexagonal bins. The number of analyzed target sequences (n) was as follows: (c) n = 2,172 to 2,307 for ABE; (d) n = 2,746 to 2,964 for CBE. Spearman's correlation coefficient (R) and Pearson's (Pearson's) correlation coefficient (r) are indicated.
(e, f) Effect of sequence context around the target base (red) on (e) ABE- and (f) CBE-directed base editing frequencies at positions 4-8 bp from the 4' end of the protospacer region . The target base conversion frequency was normalized to the median frequency of the sequence motif representing the highest median editing frequency at each position, yielding a relative frequency. The number of target motifs analyzed (n) was as follows: (e) n = 383 to 1,413 for ABE; (f) n = 498 to 1,110 for CBE. The top, center, and bottom lines in the box represent the 25th, 50th, and 75th percentiles, respectively. The whiskers represent the 1st and 99th percentile values, respectively. Subsets of context sequences with no statistically significant differences in target base conversion frequencies were denoted by letters such as a, b, c, ... and h (P<0.05; one-way ANOVA and Tukey's post hoc). determined by the assay).
2 is a diagram showing the effect of sequence context around a target adenine (left) or cytosine (right) on base editing at each position. Target bases are shown in red, and the relative base editing frequencies were normalized to the median frequencies of sequence motifs representing the highest median editing frequencies at each position. The number of analyzed target sequences (n) is as follows:
ABE at position 4: n = 159 (TAC), 106 (TAT), 99 (TAG), 126 (GAT), 143 (GAC), 176 (CAT), 219 (CAC), n = 78 (TAA), n = 169 (GAG), n = 267 (CAG), n = 146 (AAC), n = 144 (GAA), n = 158 (CAA), n = 123 (AAT); n = 166 (AAG), n = 154 (AAA);
ABE at position 5: n = 153 (TAC), 97 (TAT), 70 (TAG), 155 (GAT), 161 (GAC), 205 (CAT), 252 (CAC), 2 (TAA), 165 ( GAG), 268 (CAG), 140 (AAC), 181 (GAA), 189 (CAA), 123 (AAT), 129 (AAG), 142 (AAA);
ABE at position 6: n = 186 (TAC), 103 (TAT), 117 (TAG), 163 (GAT), 188 (GAC), 168 (CAT), 226 (CAC), 80 (TAA), 194 ( GAG), 306 (CAG), 135 (AAC), 165 (GAA), 117 (CAA), 130 (AAT), 176 (AAG), 143 (AAA);
ABE at position 7: n = 168 (TAC), 108 (TAT), 119 (TAG), 127 (GAT), 155 (GAC), 169 (CAT), 235 (CAC), 78 (TAA), 174 ( GAG), 289 (CAG), 125 (AAC), 178 (GAA), 134 (CAA), 97 (AAT), 162 (AAG), 121 (AAA);
ABE at position 8: n = 170 (TAC), 105 (TAT), 76 (TAG), 148 (GAT), 169 (GAC), 221 (CAT), 240 (CAC), (TAA), 165 (GAG) ), 282 (CAG), 118 (AAC), 194 (GAA), 170 (CAA), 125 (AAT), 134 (AAG), 134 (AAA);
CBE at position 4: n = 186 (TCC), 101 (TCG), 176 (TCT), 211 (TCA), 230 (CCC), 219 (CCT), 183 (ACC), 243 (CCA), 183 ( CCG), 115 (ACG), 153 (ACT), 177 (ACA), 203 (GCC), 220 (GCT), 170 (GCG), 194 (GCA);
CBE at position 5: n = 173 (TCC), 93 (TCG), 173 (TCT), 173 (TCA), 182 (CCC), 231 (CCT), 181 (ACC), 219 (CCA), 170 ( CCG), 91 (ACG), 178 (ACT), 157 (ACA), 222 (GCC), 223 (GCT), 148 (GCG), 84 (GCA);
CBE at position 6: n = 198 (TCC), 108 (TCG), 157 (TCT), 174 (TCA), 192 (CCC), 195 (CCT), 193 (ACC), 195 (CCA), 176 ( CCG), 104 (ACG), 152 (ACT), 161 (ACA), 228 (GCC), 159 (GCT), 154 (GCG), 200 (GCA);
CBE at position 7: n = 195 (TCC), 125 (TCG), 171 (TCT), 232 (TCA), 180 (CCC), 208 (CCT), 198 (ACC), 248 (CCA), 175 ( CCG), 118 (ACG), 165 (ACT), 165 (ACA), 213 (GCC), 213 (GCT), 158 (GCG), 173 (GCA);
CBE at position 8: n = 163 (TCC), 93 (TCG), 152 (TCT), 199 (TCA), 177 (CCC), 233 (CCT), 211 (ACC), 205 (CCA), 171 ( CCG), 70 (ACG), 126 (ACT), 193 (ACA), 227 (GCC), 225 (GCT), 172 (GCG), 200 (GCA).
The top, center, and bottom lines in the box represent the 25th, 50th, and 75th percentiles, respectively. The whiskers represent the 1st and 99th percentile values, respectively. Subsets of context sequences without statistically significant differences in base editing and indel frequencies were denoted by letters such as a, b, c, ... and j (P<0.05; one-way ANOVA and Tukey's determined by post hoc tests).
3 is a diagram showing the effect of the sequence context around the target base (red) on the frequency of SpCas9-induced indels at positions 4 to 8 bp from the 5' end of the protospacer region. The target sequences of (a) and (b) are the same as those of (e, f) of FIG. 1 , respectively. Indel frequencies were normalized to the median frequencies of sequence motifs representing the highest median editing frequency at each position. The number of analyzed target sequences per 3 nucleotide motif (n) is:
(a) n = 807 (TAC), 478 (TAT), 464 (TAG), 685 (GAT), 1,142 (GAC), 787 (CAT), 903 (CAC), 350 (TAA), 832 (GAG), 1,379 (CAG), 638 (AAC), 823 (GAA), 738 (CAA), 558 (AAT), 735 (AAG), 620 (AAA);
(b) n = 915 (TCC), 520 (TCG), 829 (TCT), 989 (TCA), 961 (CCC), 1,086 (CCT), 966 (ACC), 1,110 (CCA), 875 (CCG), 498 (ACG), 774 (ACT), 853 (ACA), 1,093 (GCC), 1,040 (GCT), 802 (GCG), 951 (GCA).
The top, center, and bottom lines in the box represent the 25th, 50th, and 75th percentiles, respectively. The whiskers represent the 1st and 99th percentile values, respectively. Subsets of context sequences without statistically significant differences in base editing and indel frequencies were denoted by letters a, b, c, ... and f (P<0.05; one-way ANOVA and Tukey's determined by post hoc tests).
4 is a schematic diagram of a predictive model for predicting the nucleotide correction efficiency of nucleotide correction gene scissors and the frequency of possible correction results. Three adenines (indicated in red) within editable windows (bold and underlined) are shown as examples, and computationally predicted and experimentally determined frequencies for ABE-edited results are shown. A protospacer adjacent motif (PAM) is shown in blue, and base-edited nucleotides are shown in lower case letters.
5 is a diagram illustrating a process of determining the number of hidden layers and the length of an input sequence using cross-validation in the development of ABE_efficiency, CBE_efficiency, ABE_proportion and CBE_proportion models. The heat map shows the mean (a) Spearman correlation coefficient, and (b) Kullback-Leibler divergence values of 10-fold cross-validation (n=10).
6 is a diagram illustrating the development and evaluation of a predictive model for predicting the efficiency and results of aBe- and CBe-induced base conversion in a given target sequence.
(a) Performance evaluation of ABE_efficiency, ABE_proportion and DeepABE at integrated and endogenous sites. The number of target sequences analyzed, the frequency of calibration results and the efficiency (n) of each result were respectively as follows: n = 438, n = 1,976 and n = 2,124 for integrated positions; For endogenous locations, n = 94, n = 435 and n = 462. Spearman correlation coefficients (R) and Pearson correlation coefficients (r) are shown.
(b) Performance evaluation of CBE_efficiency, CBE_proportion, and DeepCBE at the integration and endogenous locations. The number of target sequences analyzed, the frequency of calibration results and the efficiency (n) of each result were respectively as follows: n = 482, n = 2,978 and n = 3,107 for integrated sites; n = 10; n = 522 and n = 553 for endogenous sites.
(c) Predicted base-editing correction frequencies (by ABE_proportion or CBE_proportion) vs. endogenous sites for identical and random target pairs. Performance evaluation of ABE_proportion and CBE_proportion using the symmetrized KL divergence value (Kullback-Leibler divergence value) between the measured base editing correction results. Integration site (HT_ABE_Test (corresponding to HT_ABE in drawing) or HT_CBE_Test (corresponding to HT_CBE in drawing)) vs. The KL divergence values between the frequencies of the base editing corrections measured at the endogenous sites (Endo_ABE_HEK293T or Endo_CBE_HEK293T) are shown as reference comparisons. The number of target sequences analyzed (n) is as follows: (from left to right) n = 59, n = 62, n = 269, n = 62, n = 52, n = 65, n = 290 and n = 65.
(d) Performance comparison of DeepABE/CBE and DeepABE/CBE-CA (chromatin accessibility). Each point represents the Spearman correlation coefficient between the measured indel frequency and the predicted activity. Fine-tuning and subsequent test results over a total of 10 runs (n = 2 Х 5) were shown (NS: not significant).
7 is a diagram showing the outline of DeepABE and DeepCBE development.
8 and 9 are diagrams illustrating performance evaluation results of ABE_efficiency, ABE_proportion and DeepABE in endogenous regions of HEK293T (ac), HCT116 (d) and U2OS (e, f) cells. The number of analyzed target sequences, the results, and the efficiency (n) of each result were respectively as follows: n = 94; n = 435; HEK293T cells, n = 462 for replicate 1; n = 87; n = 353; HEK293T, n = 379 for replicate 2; n = 75; n = 316; HEK293T, n = 337 for replicate 3; n = 41; n = 213; n = 244 for HCT116 cells; n = 24; n = 100; U2OS cells, n = 124 for replicate 1; n = 25; n = 91; For U2OS cells, replicate 2, n = 116. Spearman correlation coefficients (R) and Pearson correlation coefficients (r) are indicated.
10 and 11 are diagrams illustrating performance evaluation results of CBE_efficiency, CBE_proportion and DeepCBE in endogenous regions of HEK293T (ac), HCT116 (d) and U2OS (e, f) cells. The number of analyzed target sequences, the results, and the efficiency (n) of each result were respectively as follows: n = 102; n = 522; HEK293T, for replicate 1, n = 553; n = 95; n = 531; HEK293T, n = 559 for replicate 2; n = 83; n = 413; HEK293T, n = 436 for replicate 3; n = 36; n = 193; n = 203 for HCT116 cells; n = 28; n = 149; U2OS cells, n = 170 for replicate 1; n = 23; n = 136; U2OS cells, n = 159 for replicate 2. Spearman correlation coefficients (R) and Pearson correlation coefficients (r) are indicated.
12 is a prediction result for ABE and CBE-induced modeling and correction of disease-related human point mutations. Said pathogenic or pseudopathogenic point mutations are associated with PAM (NGG) at an appropriate distance and are in principle generated from wild-type sequences or wild-type sequences by ABE or CBE using an editable window from positions 3 to 10. may have been converted to
(a) Number of disease-associated point mutations that could in principle be generated using ABE (green) or CBE (orange).
(b) The number of disease-associated point mutations that can in principle be corrected using ABE (green) or CBE (orange).
(c, d) Performance evaluation of (c) ABE_proportion and CBE_proportion and (d) DeepABE and DeepCBE on the modeling of disease-associated point mutations in human iPSCs. Modeling was performed by introducing pathogenic/pseudopathogenic mutations in normal human iPSCs by ABE or CBE. Spearman correlation coefficient (R) and Pearson correlation coefficient (r) are indicated. The number of results in (c) is n = 465 (for ABE) and 767 (for CBE). In (d), the number of pathogenic/pseudopathogenic mutation sites is n = 31 (for ABE) and 49 (for CBE). )to be.
13 is a diagram showing predictions of ABE- and CBE-induced modeling and disease-related human point mutation correction results.
(a) Distribution diagram showing the results of in silico experiments on ABE- and CBE-induced modeling and disease-related human point mutation correction. A pie chart shows the number of pathogenic and pseudopathogenic point mutations that can be generated or corrected with efficiencies ≥ 5% (red) or <5% (blue). Point mutations with a single A or C within the base correction range are indicated in light red or light blue, and point mutations with two or more A or C within the base correction range are indicated in dark red or dark blue. The area of each pie is proportional to the number of corresponding point mutations.
Specifically, the chart on the left of (a) shows the number of human diseases that can be made with nucleotide editing, and the chart on the top right of (a) shows the number of human diseases that can be used without additional mutations with nucleotide correction. In the pie chart, the efficiency of base-correction gene scissors was evaluated as a value obtained by multiplying the "efficiency score" of the present application and the "correction result score". In the case of the lower right chart of (a), it shows the number of human diseases that can be used with additional non-dangerous mutations with base-correction gene scissors. It was calculated by adding the correction results that were predicted to be absent.
(b) A table showing examples of correction results that can be made with base-correction gene scissors.

Hereinafter, an aspect will be described in more detail through Examples and Experimental Examples. However, these Examples and Experimental Examples are for illustrative purposes of an aspect, and the scope of an aspect is not limited to these Examples and Experimental Examples, and the Examples and Experimental Examples of an aspect are average knowledge in the art. It is provided to more completely explain an aspect to those with

Experimental method

1. Construction of oligonucleotide library and plasmid library

Twist Bioscience Co. was commissioned to prepare a library of a total of 17,840 oligonucleotide pools. As the target sequence for the oligonucleotide pool, any synthetic sequence was generated without any information about the sgRNA or whether it contained A or C within an editable window, etc. The oligonucleotide pool was selected from 9,824 target sequences randomly selected from the GeCKOv1 library, 1,804 target sequences selected from genes encoding cell surface markers, and genes associated with resistance to vemurafenib, selumetinib and 6-thioguanine. 546 target sequences from selected 2,484 target sequences, 998 input sequences containing guide sequences with extremely low or high GC content (≤20% or ≥80%) and human coding and non-coding genes associated with the gene of interest were included. . In the case of the 546 target sequences, five barcodes per target sequence were used to generate 5-fold coverage for each target site. Taken together, the 17,840 oligonucleotide set consists of 9,824+1,804+2,484+998+(546Х5) oligonucleotides and contains 9,824+1,804+2,484+998+546=15,656 pairs of sgRNA-coding and target sequences. Using this, the HT_ABE_Train and HT_CBE_Train data sets were generated, of which 546 pairs were used to generate the HT_ABE_Test and HT_CBE_Test data sets.

A plasmid library containing the guide RNA and the corresponding target sequence pair was constructed with a two-step cloning process of Gibson assembly followed by restriction enzyme cutting and ligation.

2. Preparation of plasmid vectors

For the preparation of individual split-ABE plasmids, fragments of the ABE7.10-coding sequence (Addgene, no. 102919) and intein-mediated split-Cas9-coding sequence were PCR amplified and lentiCas9-Blast (Addgene, no. 52962) or pX601 (Addgene, no. 61591) plasmid. The resulting plasmids were named Lenti_Split-ABE-N-Blast, Lenti_SplitABE-C-Hygro-eGFP, AAV_Split-ABE-N and AAV_Split-ABE-C.

For the preparation of individual split-BE4 plasmids, fragments of the BE4-coding sequence (Addgene, no. 100802) and the intein-mediated split-Cas9-coding sequence were PCR amplified and lentiCas9-Blast (Addgene, no. 52962) or pX601 (Addgene, no. 61591) was cloned into a plasmid. The resulting plasmids were named Lenti_Split-BE4-N-Blast, Lenti_Split-BE4-C-Hygro, AAV_Split-BE4-N and AAV_Split-BE4-C.

To prepare the lentiviral vector encoding the full-length ABE7.10, the pLenti6/V5-GW/LacZ plasmid was purchased (ThermoFisher) and modified for subsequent replication: the LacZ fragment was digested with EcoRV restriction enzyme (NEB) and Wood The sequence of the posttranscriptional regulatory element of woodchuck hepatitis was inserted into the KpnI enzyme-recognition site. The full-length ABE-coding sequence from pCMV-ABE7.10 (Addgene, no. 107723) was cloned into the modified pLenti6/V5-GW/LacZ plasmid and the resulting plasmid was named Lenti-ABE-Blast. The pcDNA-BSD plasmid was prepared by amplifying the BSD gene by PCR, digesting it with Kpnl and EcoRI, and cloning the pcDNA 3.1(+) vector (Invitrogen).

3. Production of Lentiviruses

For lentiviral library production, HEK293T cells (ATCC), which are human embryonic kidney cells, were prepared. The three transfer plasmids containing the gene of interest, psPAX2 and pMD2.G, were mixed in a weight ratio of 4:3:1 to generate a plasmid mixture of a total of 20 μg, which was transfected into HEK293T cells using Lipofectamine 2000 (Invitrogen). infected. A fresh medium was added to the cells 12 hours after transfection, and a supernatant containing virus was obtained 36 hours after transfection. The obtained supernatant was filtered through a Millex-HV 0.45 μm low-protein binding membrane (Millipore), and the aliquot was stored at -80°C until use. Virus yield was verified by measuring with Lenti-X p24 Rapid Titer Kit (Clontech). For virus titer calculation, serially diluted virus aliquots were transduced into HEK293T cells in the presence of 8 μg ml ^-1 polybrene, and 2 μg ml ^-1 puromycin or 20 μg ml ^-1 blasticidin It was calculated by culturing in the presence of (blasticidin) S (InvivoGen).

For transduction of the prepared lentiviral library, HEK293T cells (9.0Х10 ⁶ ) were cultured overnight in a culture dish. A lentiviral library with a multiplicity of infection (MOI) of 0.3 was transduced into HEK293T cells in the presence of 8 μg ml ⁻¹ polybrene, and the cells were cultured for 15 to 18 hours. Cells were cultured in the presence of 2 μgml ⁻¹ puromycin to remove non-transduced cells, and the cell library was maintained at an amount of 9.0Х10 ⁶ cells.

4. Delivery of ABE, CBE to Cell Libraries

For ABE, Lenti_Split-ABE-N-Blast and Lenti_Split-ABE-C-Hygro-Egfp were mixed in a weight ratio of 1:1 to produce a total of 240 μg of a plasmid mixture, which was 5.2Х10 ⁷ using Lipofectamine 2000 Canine sheep cells were transferred to the library. For CBE, Lenti_Split-BE4-N-Blast, Lenti_Split-BE4-C-Hygro, and pcDNA-BSD were mixed in a weight ratio of 9:9:2 to generate a total of 20 μg of plasmid mixture, followed by a neon transfection system (ThermoFisher Scientific). ) was used to electroporate 2Х10 ⁶ library cells. The next day, the culture medium was replaced with DMEM supplemented with 10% FBS, 40 μg ml ^-1 blasticidin S (InvivoGen) and 80 μg ml ^-1 hygromycin B gold (InvivoGen), and cultured 5 days after transfection. Water was collected and used.

5. Determination of base editing frequency at endogenous sites

For evaluation of ABE and CBE activity at endogenous sites, a total of 153 target sites out of 546 endogenous targets were selected (70 DHS sites, 83 non-DHS sites). 100 ng of plasmid encoding sgRNA in HEK293T cells (pRG2; Addgene no. 104174) and full-length ABE7.10 (Lenti-ABE-Blast) or split-BE4 (lenti-BE4-N-Blast, lenti-BE4-C- Each activity was measured by transfection with a mixture of 100 ng of plasmids encoding Hygro; 1:1 ratio). HCT116 cells were treated with 100 ng of plasmid encoding sgRNA (pRG2), split-ABE (AAV-Split-ABE7.10-N, AAV-Split-ABE7.10-C; 1:1 ratio) or split-BE4 (AAV-Split) -BE4-N, AAV-Split-BE4-C; 1:1 ratio) and 50 ng of plasmids encoding eGFP and puromycin N-acetyltransferase. HEK293T or HCT116 cells were transfected at a density of 1.0Х10 ⁵ or 4.0Х10 ⁴ per well. U2OS cells consisted of 1 μg of plasmid encoding sgRNA (pRG2, Addgene no. 104174), 500 ng of plasmid encoding eGFP and puromycin N-acetyltransferase (pEGFP-Puro, Addgene no. 45561) and full-length ABE7.10 (Lenti- ABE-Blast) or split-BE4 (Lenti-BE4-N-Blast, Lenti-BE4-C-Hygro; 1:1 ratio) were transfected with a mixture of 1 μg of plasmids. The plasmid mixture was electroporated into 1x10 ⁶ U2OS cells using a neon transfection system (ThermoFisher Scientific). After overnight incubation, the culture medium was replaced with DMEM supplemented with 10% FBS and 2 μg ml ^-1 of puromycin (InvivoGen). Cells transfected with ABE7.10 or BE4 were collected and deep sequencing after 5 days (HEK293T cells and U2OS cells) or 3.5 days (HCT116 cells).

6. Disease Modeling and Correction of Disease-Associated Mutations

Disease modeling and correction of disease-related mutations that can be made with base-correcting gene scissors in human induced pluripotent stem cells were confirmed. First, a total of 95 mutations containing multiple (at least 3 or more) targets A or C among the disease-associated point mutations in the ClinVar database and located in a narrow editable window (positions 4 to 8) were selected. Normal human iPSCs were cultured in Essential 8 medium (ThermoFisher Scientific). To induce disease-associated point mutations, human iPSCs were subjected to split-ABE (AAV-Split-ABE7.10-N, AAV-Split-ABE7.10-C; 1:1 ratio) or split-BE4 (AAV-Split) -BE4-N, AAV-Split-BE4-C; 1:1 ratio), 150 ng of plasmid encoding sgRNA (pRG2, Addgene no. 104174) and eGFP with puromycin N-acetyltransferase (pEGFP-Puro, Addgene no. 45561) and transfected with 500 ng of a mixture of 100 ng of plasmid. To measure the efficiency of ABE and CBE-mediated disease-associated point mutation correction, synthetic target sequences containing pathogenic point mutations were delivered as lentiviruses into normal human iPSCs. Non-transduced cells were removed by adding 4 μg ml ^-1 blasticidin to the culture medium. Next, using a lipofectamine reagent (ThermoFisher Scientific), iPSCs into which the target sequence was introduced were transfected with a mixture of the three types of plasmids. After overnight incubation after transfection, the culture medium was replaced with Essential 8 medium supplemented with 10 μM Y-27632 (Sigma-Aldrich) and 1 μg ml ⁻¹ puromycin (Gibco). 24 hours after selection with puromycin, the medium was removed, and Essential 8 medium supplemented with 10 μM Y-27632 per well was added to the cells. Three days after transfection, cells were collected and genomic DNA was deep-sequenced to measure the efficiency of gene scissors and the result of base editing.

7. Deep sequencing

Genomic DNA was isolated from cells using the Wizard Genomic DNA Purification Kit (Promega). PCR was performed using 2Х pfu PCR Smart mix (Solgent). For high-throughput experiments, a total of 264 μg of genomic DNA was used for each cell library in the first PCR to achieve a coverage of 1,400x or greater for the library. The resulting PCR products were combined into a single pool and purified with a MEGAquick-spin total fragment DNA purification kit (iNtRON Biotechnology). A sample of 20 ng of the purified product was amplified by secondary PCR using primers containing the Illumina adapter and barcode sequence.

The primers used in the experiment are as follows (5'-3').

Primers for oligonucleotide pool amplification

- Forward: TTGAAAGTATTTCGATTTCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACC (SEQ ID NO: 1)

-reverse: GAGTAAGCTGACCGCTGAAGTACAAGTGGTAGAGTAGAGATCTAGTTACGCCAAGCT (SEQ ID NO: 2)

Primers for the first PCR reaction

- Forward: ACACTCTTTCCCTACACGACGCTCTTCCGATCTCTTGAAAAAGTGGCACCGAGTCG (SEQ ID NO: 3)

ACACTCTTTCCCTACACGACGCTCTTCCGATCTTCTTGAAAAAGTGGCACCGAGTCG (SEQ ID NO: 4)

ACACTCTTTCCCTACACGACGCTCTTCCGATCTCGCTTGAAAAAGTGGCACCGAGTCG (SEQ ID NO: 5)

- Reverse: GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTTTAAGTCGAGTAAGCTGACCGCTGAAG (SEQ ID NO: 6)

GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTATTAAGTCGAGTAAGCTGACCGCTGAAG (SEQ ID NO: 7)

GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTTATTAAGTCGAGTAAGCTGACCGCTGAAG (SEQ ID NO: 8)

Primer for secondary PCR reaction (lowercase letters in the center of the base sequence mean 8bp barcode sequence)

- Forward: AATGATACGGCGACCACCGAGATCTACACtatagcctACACTCTTTCCCTACACGAC (SEQ ID NO: 9)

AATGATACGGCGACCACCGAGATCTACACatagaggcACACTCTTTCCCTACACGAC (SEQ ID NO: 10)

AATGATACGGCGACCACCGAGATCTACACcctatcctACACTCTTTCCCTACACGAC (SEQ ID NO: 11)

AATGATACGGCGACCACCGAGATCTACACggctctgaACACTCTTTCCCTACACGAC (SEQ ID NO: 12)

AATGATACGGCGACCACCGAGATCTACACaggcgaagACACTCTTTCCCTACACGAC (SEQ ID NO: 13)

AATGATACGGCGACCACCGAGATCTACACtaatcttaACACTCTTTCCCTACACGAC (SEQ ID NO: 14)

AATGATACGGCGACCACCGAGATCTACACcaggacgtACACTCTTTCCCTACACGAC (SEQ ID NO: 15)

AATGATACGGCGACCACCGAGATCTACACgtactgacACACTCTTTCCCTACACGAC (SEQ ID NO: 16)

- reverse: CAAGCAGAAGACGGCATACGAGATcgagtaatGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 17)

CAAGCAGAAGACGGCATACGAGATtctccggaGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 18)

CAAGCAGAAGACGGCATACGAGATaatgagcgGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 19)

CAAGCAGAAGACGGCATACGAGATggaatctcGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 20)

CAAGCAGAAGACGGCATACGAGATttctgaatGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 21)

CAAGCAGAAGACGGCATACGAGATacgaattcGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 22)

CAAGCAGAAGACGGCATACGAGATAgcttcagGTGACTGGAGTTCAGACGTGT (SEQ ID NO:23)

CAAGCAGAAGACGGCATACGAGATgcgcattaGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 24)

CAAGCAGAAGACGGCATACGAGATcatagccgGTGACTGGAGTTCAGACGTGT (SEQ ID NO:25)

CAAGCAGAAGACGGCATACGAGATttcgcggaGTGACTGGAGTTTCAGACGTGT (SEQ ID NO: 26)

CAAGCAGAAGACGGCATACGAGATgcgcgagaGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 27)

CAAGCAGAAGACGGCATACGAGATctatcgctGTGACTGGAGTTCAGACGTGT (SEQ ID NO: 28)

8. Analysis of base editing efficiency and frequency of calibration results

Sequence data was analyzed by modifying the Python scripts program for the analysis of base editing efficiency and proofreading results. Each guide RNA and target sequence pair was identified using a unique 15-nt barcode sequence located upstream of the target sequence. Insertions or deletions around the expected cleavage site (8-nt region in the middle of the cleavage site) for indels were considered nuclease-induced indels. To analyze the efficiency and results of base editing, reads sorted by a unique barcode sequence are aligned with a Python script, and reads containing indels are compared with a reference sequence using the Needleman-Wunsch algorithm. were classified. When the target nucleotides of ABE and CBE within the editable window were converted to T or G, respectively, they were counted as base edited. Then, the base editing efficiency at each position was calculated according to the following equation.

The total number of reads is the sum of all base calls including indels. To increase the accuracy of the analysis for base editing frequency, target sequences with a total deep sequencing read count of less than 100 in the deep sequencing data were excluded.

When there is only one target nucleotide within the editable window of positions 3 to 10, the base editing efficiency is equal to the conversion efficiency at the target nucleotide, and the result of base editing converges in two cases: edited or not. . However, when the target sequence composition is random, the probability of having only one target nucleotide in the editable window is 13%, and in the remaining 87%, one or more target nucleotides exist, resulting in complicated results after base editing. Accordingly, to analyze the frequency of the calibration results, the sorted reads were recalculated according to the calibration results in the base editing window using a Python script. Since the target nucleotide conversion efficiency is not affected by the position of the nucleotide within the editable window of positions 3 to 10, the case where the target nucleotide was converted to an unintended nucleotide was excluded. The frequency of each base editing result was calculated according to the following equation.

To increase the accuracy of the analysis of the base editing results, target sequences with a modification count of less than 100 (or less than 200 in the case of experiments using iPSCs) were filtered out, and reads containing unintended nucleotide conversions were excluded. The absolute frequency of the base editing result can be calculated by multiplying the base editing result frequency by the base editing efficiency, and can be calculated according to the following equation:

9. Consideration of chromatin accessibility

DNase-seq narrow peak data obtained from ENCODE42 was used to consider chromatin accessibility. For each target position, the 23 base PAM + protospacer sequence was aligned to the hg19 human reference genome using bowtie43. The target site overlapping the DNase-seq narrow peak was considered as the DHS site.

10. Deep Learning Using Convolutional Neural Networks

A model was developed that performed deep learning on a data set generated using a convolutional neural network (CNN) without a pooling layer and predicted the efficiency and correction result of base-correction gene scissors. To prevent overfitting the model to the training data set, early stopping was used based on the validation score, and a dropout rate of 0.3 was used in each layer. All models have a filter shape (1,3); channel order, 4; The stride (1,1) and convolution layer did not apply padding, and 150, 60, 60 and 150 filters were used for the ABE_proportion, ABE_efficiency, CBE_proportion and CBE_efficiency models, respectively. After the convolution layer, one (ABE_proportion, ABE_efficiency) or two (CBE_proportion, CBE_efficiency) fully connected layers were used. /256 and 500/50. After the convolution operation, the ReLU activation function was applied.

11. Symmetrized KL divergence

Symmetrical Kullback-Leibler divergence was used to calculate the similarity of the distribution of base editing results. To avoid division by zero, false counts of 0.001 in the training data set and 0.5 in the test data set were added, and then the KL divergence was calculated by the following equation. P _i and Q _i mean predicted and evaluated frequencies, respectively.

12. ClinVar Data Analysis

To validate base editing efficiency and results for modeling and therapeutic correction of human disease-associated mutations, the publicly available ClinVar (v. clinvar_20190219_hg38) data set was filtered and used. Filtering involves first selecting point mutations with an A or C within an editable window of positions 3 to 10 while at an appropriate distance from the PAM (NGG) sequence, and then selecting target point mutations marked as pathogenic or pseudopathogenic on the ClinVar database. This was done in two steps.

13. Statistical Significance

A two-tailed Student's t-test was used to compare the nucleotide editing efficiency between the DHS site and the non-DHS site. Steiger's test was used to compare the performance between DeepABE/CBE and DeepABE-CA/CBE-CA (a model considering chromatin accessibility). One-way ANOVA and Tukey's post hoc test were used to determine the effect of sequence context around the target base on the base editing frequency. Statistical significance was determined using PASW Statistics (v.18.0, IBM) and Microsoft Excel (v.16.0, Microsoft Corporation).

14. Python script and deep sequencing data

In one embodiment, the Python scripts of DeepABE, DeepCBE, and DeepCBE-CA are provided on github (https://github.com/MyungjaeSong/Paired-Library, https://github.com/CRISPRJWCHOI/BaseEditing_tool ), and in one embodiment The deep sequencing data generated in the example is accessible from the NCBI Sequence Read Archive (SRA; http://www.ncbi.nlm.nih.gov/sra/ ) with accession number SRP150719 (PRJNA476544).

Example

Example 1. Genetic scissors efficiency and calibration result data set production

1-1. production of data sets

For mass validation of gene scissors, a lentiviral library of 15,656 guide RNA coding and target sequence pairs used in a previous study (Kim et al, Nat Methods, 2017) was used to produce a data set of the efficiency and proofreading result of the gene scissors. . Of the total 15,656 sequence pairs, 13,504 at positions 3 to 10 bp from the 5' end of the protospacer region (position 20 is immediately upstream of PAM(NGG)) contained at least one target adenine, and 14,157 contains at least one target cytosine. The cell library was transfected with a plasmid encoding ABE7.10 or BE4, and base conversion efficiency and base editing results at target nucleotides were evaluated 5 days after transfection. In this case, since the plasmid vector encoding ABE or CBE has a relatively large size, intein-mediated split-Cas9-based ABE and CBE were used together. As a result, four independent data sets were generated and named HT_ABE_Train, HT_ABE_Test, HT_CBE_Train, and HT_CBE_Test, respectively. In addition, data sets on ABE and CBE activities at 96 and 102 endogenous target sites were generated, and were named Endo_ABE and Endo_CBE, respectively. As a result, the ABE and CBE activities at positions 3 to 10 of the target site corresponding to the DNase I hypersensitive region (DHS) were 1.1-fold (P=0.55) or 1.8-fold (P=0.0077) higher than the non-DHS site, respectively. Appearance was confirmed. Through the above results, it was confirmed that CBE activity was increased at a site with high chromatin accessibility than at a site with low chromatin accessibility, and ABE activity was hardly affected by chromatin accessibility.

1-2. Searching for factors affecting the activity of base-editing gene editing

Through high-throughput evaluation, ABE- and CBE-directed base editing occurs in a wide window corresponding to positions 3 to 10, but a relatively high level of base editing is observed in both ABE and CBE positions 4 to 8. It was confirmed that it can be achieved in a narrow window (FIG. 1 a, b).

When the activity of Cas9 and the activity of ABE or CBE were compared in the same target sequence, in most cases when the activity of ABE or CBE was high, the activity of Cas9 was also high, confirming a non-significant positive correlation (FIG. 1 c, d). However, when Cas9 activity is high, ABE or CBE activity is sometimes high and sometimes low, and as a result, ABE, CBE, and SpCas9 nuclease activity exhibit an asymmetric correlation. Furthermore, the nucleotide preference associated with high activity of ABE or CBE or high activity of Cas9 at each position of the target sequence is determined, and the asymmetric correlation is different from the nucleotide preference of SpCas9 and ABE/CBE at a specific position of the target sequence. It was confirmed that because Specifically, in the case of ABE/CBE, a strong nucleotide preference was observed at the nucleotide immediately adjacent to the target nucleotide (±1 bp of the target nucleotide), and this preference was preserved regardless of the position of the target nucleotide.

Furthermore, as a result of analyzing the base editing efficiency of ABE in all 16 possible NAN sequences to analyze the effect of the sequence context surrounding on base editing on ABE and CBE activity, the base conversion efficiency was determined by TAB (especially TAY, Y=C or T), and it was confirmed that A-rich contexts such as AAC, GAA, CAA, AAT, AAG and AAA were the lowest in A-rich contexts (FIG. 1 e). In the case of TAA, the effect of T and the effect of A at the 3' position were offset, so the base editing efficiency was not high. Similarly, as a result of analyzing the base editing efficiency of CBE in all possible NCN sequences, similar to ABE, the base conversion efficiency is highest when the 5' position is T (i.e., TCN), and when the 5' position is G (e.g. For example, GCC, GCT, GCG and GCA) were confirmed to be the lowest (FIG. 1 f). In contrast to ABE, when the target nucleotide is repeated at one or two adjacent positions, such as TCC, CCC, CCT, ACC, CCA, CCG and GCC, it was confirmed that the conversion efficiency of C base is slightly higher.

This context preference was observed at all positions (positions 4 to 8) of the narrow window, confirming that the nucleotide context had a strong effect on the base editing efficiency (FIG. 2). Specifically, it was confirmed that the maximum-fold difference in the conversion efficiency of the target nucleotide related to the sequence context was 45-fold and 13-fold for ABE and CBE, respectively.

On the other hand, in the case of indel generation by SpCas9, such context preference was hardly observed (FIG. 3).

Example 2. Construction of a base-correction gene scissors activity and result prediction model

2-1. Development of Efficiency Prediction Model

A predictive model was constructed using the efficiency and calibration result data of the gene scissors produced in Example 1 and deep learning technology. First, using a deep learning framework and HT_ABE_Train and HT_CBE_Train training data sets, a nucleotide editing efficiency prediction model named ABE_efficiency and CBE_efficiency was developed (FIG. 4). At this time, the base-editing efficiency was determined by analyzing the base-edited sequences within a wide editable window (positions 3 to 10 in the case of ABE and CBE) regardless of the number of base-edited nucleotides among the total DNA copies analyzed. Refers to the percentage of DNA copies possessed, and a neural network architecture is used as a deep learning technique. The number of hidden layers and the length of the input sequence were determined using 10-fold cross validation. In the case of ABE_efficiency, the two hidden layer models and the three hidden layer models that showed the highest performance showed similar performance (0.776 vs. 0.778, respectively), so two hidden layer models were selected (FIG. 5).

As a result, the ABE_efficiency model reached correlation coefficients of Spearman R=0.72, Pearson r=0.70 (HT_ABE_Test) and Spearman R=0.76, and Pearson r=0.70 (Endo_ABE_HEK293T), respectively, with HT_ABE_Test and Endo_ABE_HEK293T as test data sets, reaching excellent performance correlation coefficients. (a of FIG. 6), and similarly, the CBE_efficiency model also reached a correlation coefficient of Spearman R=0.79, Pearson r=0.78 (HT_CBE_Test) and Spearman R=0.69, Pearson r=0.60 (Endo_CBE_HEK293T) when using HT_CBE_Test and Endo_CBE_HEK293T. It was confirmed that excellent performance was obtained (FIG. 6 b).

2-2. Development of calibration result prediction model

Since different sequences are generated as a result of base editing when there is more than one target nucleotide in the editable window, another deep learning framework and the HT_ABE_Train and HT_CBE_Train training datasets were used to predict the relative frequency of these corrections using the ABE_proportion and A correction result prediction model named CBE_proportion was developed. In addition to the Spearman correlation coefficient, symmetric Kullback-Leibler (KL) divergence was used together to reflect the similarity of the calibration results.

As a result, the ABE_proportion model showed high performance of Pearson r = 0.95 (HT_ABE_Test) and Pearson r = 0.93 (Endo_ABE_HEK293T), respectively (Fig. 6 a), and similarly, the CBE_proportion model also Pearson r = 0.95 (HT_CBE_Test) and Pearson r = 0.91 (Endo_CBE_HEK293T) showed a high performance (Fig. 6 b). Furthermore, the symmetric KL divergence between the calibration result frequency of Endo_ABE_HEK293T and the predicted value from ABE_proportion was similarly low (median KL = 0.11) and Endo_CBE_HEK293T between Endo_ABE_HEK293T and HT_ABE_Test (median KL = 0.10) in the same target sequence. It was confirmed that the symmetric KL divergence between the frequency and the predicted value from CBE_proportion was similarly low (median KL = 0.36) (median KL = 0.36) between Endo_CBE_HEK293T and HT_CBE_Test in the same target sequence (median KL = 0.18). ).

Specifically, the results of predicting the nucleotide editing efficiency and correction results of ABE and CBE nucleotide editing are shown in Tables 1 and 3 below, and examples of human point mutation diseases that can be made or corrected with the nucleotide correction gene scissors are shown in Table 2 and 4 .

[Table 1]

ABE

[Table 2]

ABE

[Table 3]

CBE

[Table 4]

CBE

As shown in Tables 2 and 4, in the case of ABE as an example of a human point mutation disease that can be made or corrected with base-correcting gene scissors, Usher syndrome, tumor necrosis factor receptor-related periodic syndrome (TNF receptor-associated periodic) syndrome: TRAPS, marfan syndrome, Type 3 form of Maturity-Onset Diabetes of the Young: MODY3, Congenital stationary night blindness type 1F, familial high Familiar hypercholesterolemia, congenital myasthenic syndrome (CMS), Lynch syndrome, etc. were confirmed, and in the case of CBE, Loeys-Dietz syndrome (LDS), retinitis pigmentosa (retinitis pigmentosa), leptin deficiency or dysfunction, familial hypercholesterolemia, autosomal recessive deafness, cholesterol monooxygenase (side-chain-) cleaving deficiency), progressive myoclonus epilepsy, etc. were confirmed.

From the above results, it was confirmed that the developed ABE_proportion and CBE_proportion models can predict the result frequency of base editing with high accuracy. On the other hand, this effect is not confirmed when using traditional machine learning or shallow neural networks (e.g., AdaBoost, Boosted RT, SVM, Ridge, Lasso, ElasticNet, Random Forest, etc.), so the prediction system according to an aspect is deep It was found that it was possible to predict the nucleotide correction efficiency and correction result of nucleotide correction gene scissors with very high accuracy by using running.

On the other hand, the present inventors confirmed that the prediction of Cas12a efficiency in the endogenous site through previous studies can be improved by considering chromatin accessibility. Accordingly, in one embodiment, it was confirmed that CBE was affected by chromatin accessibility, but ABE was not, and additional tests were performed to determine whether consideration of chromatin accessibility could improve the prediction of nucleotide editing efficiency. First, the Endo_CBE data set was divided into paired subsets through stratified random sampling in the hierarchy of DNase I hypersensitive (DHS) and non-DHS (non-DHS) regions, and DHS of similar proportions /Non-DHS sites were assigned to each subset and were named Endo_CBE_1A and Endo_CBE_1B, respectively. By repeating the random sampling to generate additional four data sets named Endo_CBE_2A, Endo_CBE_2B, etc., and fine-tuning DeepCBE using the binary chromatin accessibility information obtained from the Endo_CBE_1A data set and ENCODE (Encyclopedia of DNA elements), We developed DeepCBE-CA (Chromatin Accessibility)-1A, a CBE-based gene editing efficiency and outcome prediction model based on both target sequence information and chromatin accessibility.

Next, DeepCBE-CA-1A was evaluated using another data set (Endo_CBE_1B) as the test data set. Repeat the tuning and subsequent tests by swapping the test and training datasets (e.g., Endo_CBE_1B as the training dataset for tuning and Endo_CBE_1A as the test dataset), then use the other 4 pairs of data This was repeated using the set. As a result of a total of 10 fine-tuning and subsequent tests, the Spearman correlation of these fine-tuned models was similar to that of DeepCBE (Fig. 6d), so fine-tuning considering chromatin accessibility information did not improve the accuracy of DeepCBE. confirmed that it is not. As for ABE, a total of 10 tests were performed equally using DeepABE and Endo_ABE. As a result, consideration of chromatin accessibility information did not improve the prediction accuracy for ABE.

2-3. Development of DeepABE and DeepCBE

Next, ABE_efficiency and ABE_proportion developed in Examples 2-1 and 2-2 were combined to predict the absolute frequency of the base-edited result, and CBE_efficiency and CBE_proportion were combined to improve base-editing efficiency and all the nucleotide-editing gene scissors that can be made. DeepABE and DeepCBE models capable of predicting base editing results were created.

As a result, when tested with HT_ABE_Test and Endo_ABE_HEK293T, DeepABE reached high correlation coefficients of Spearman R=0.90, Pearson r=0.92 (HT_ABE_Test) and Spearman R=0.86, Pearson r=0.80 (Endo_ABE_HEK293T) to predict the frequency of results of base editing. showed excellent performance in (a of FIG. 6). Similarly, DeepCBE also reached a high correlation coefficient of Spearman R=0.86, Pearson r=0.87 (HT_CBE_Test) and Spearman R=0.83, Pearson r=0.71 (Endo_CBE_HEK293T), confirming that it exhibited very good performance (Fig. 6b) ).

On the other hand, the ABE_efficiency, CBE_efficiency, ABE_proportion, and CBE_proportion models are based on a deep learning framework using powerful convolutional neural networks (CNNs), and were specifically developed as follows: (1) a target sequence and converting an input sequence containing a neighboring sequence into a four-dimensional binary matrix; (2) a 3-nt long filter is moved through a 4-dimensional binary matrix to determine position weight matrices; (3) in fully connected layers, the extracted features are summed according to a weighted sum, rectified linear unit (ReLU) activation function; (4) In the output layer, perform linear regression and predict the activity score or result frequency for each target sequence. DeepABE and DeepCBE scores were obtained simply by multiplying the scores of ABE_proportion and ABE_efficiency or CBE_proportion and CBE_efficiency respectively ( FIG. 7 ).

2-4. Verification of Accuracy of DeepABE and DeepCBE

The accuracy of the prediction models ABE_efficiency, ABE_proportion, DeepABE, CBE_efficiency, CBE_proportion and DeepCBE constructed in Example 2 was verified using biological replicates. Specifically, the performance of ABE_efficiency, ABE_proportion, DeepABE, CBE_efficiency, CBE_proportion and DeepCBE in endogenous sites of HCT116 cells, U2OS cells and HEK293T cells was evaluated. As a result, it was confirmed that the ABE_proportion and CBE_proportion as well as the DeepABE and DeepCBE models showed excellent performance across all cell types ( FIGS. 8 to 11 ).

Example 3. Application of predictive model to human point mutation disease

3-1. Application to human point mutation disease data

Using the developed predictive model and the human point mutation disease data reported in ClinVar4, we predicted the nucleotide editing efficiency and results for disease modeling and therapeutic correction of human point mutations in nucleotide editing. For disease modeling, pathogenic and pseudopathogenic point mutations that can be generated using editable windows of positions 3 to 10 with PAM (NGG) at an appropriate distance were explored. As a result, they could theoretically be generated using ABE and CBE. It was confirmed that there were 2,917 and 8,759 possible point mutations, respectively. Among these point mutations, 24% of ABEs (691 of 2,917) and 13% of CBEs (1,113 of 8,759) occurred in an editable window with only one A or C among these point mutations. , and the remaining 76% and 87% of the point mutations contained at least one A or C, respectively (FIG. 12 a).

Furthermore, as a result of searching for pathogenic and pseudopathogenic point mutations that can be corrected using nucleotide editing, it was confirmed that, in principle, 8,930 and 2,873 mutations can be converted into wild-type sequences using ABE and CBE, respectively. Of these mutations, 21% of ABEs (1,834 of 8,930) and 12% of CBEs (336 of 2,873) had only one A or C in the editable window, resulting in the majority It was confirmed that the base-editable mutation of ' occurred in a window with a large number of target nucleotides (FIG. 12 b).

Through the above results, the number of human diseases that can be made with ABE is a total of 2,917, the number of human diseases that can be made with CBE is a total of 8,759, the number of human diseases that can be corrected using ABE is a total of 8,930, It was confirmed that the number of human diseases that can be corrected using CBE is a total of 2,873.

3-2. Application to human induced pluripotent stem cells

Using human induced pluripotent stem cells (iPSCs) and the developed predictive model, the nucleotide editing efficiency and results were predicted for disease modeling and therapeutic correction of human point mutations in nucleotide editing. As a result, it was confirmed that there was a significant correlation between the predicted correction efficiency and the measured correction result frequency in both disease modeling and treatment (FIG. 12 c, d).

3-3. Efficiency prediction of base-correcting gene scissors

Based on the modeling performed in Examples 3-1 and 3-2, the nucleotide-correction efficiency of nucleotide-correction gene scissors for human point mutation diseases was predicted, and the distribution according to the efficiency was confirmed (FIG. 13). When the frequency of base editing results containing the mutation of interest is higher than 5%, if we define the modeling as "efficient", then 639 out of 691 mutations in single A (92%) within the editable window (92%); It was found that 1,225 (55%) of 2,226 mutations in A were efficiently modeled. In the case of correctable point mutations, if the modeling is defined as "efficient" when the frequency of the wild-type sequence is higher than 5% as a result of base editing, 1,728 out of 1,834 mutations in single A within the editable window (94 %), it was confirmed that 4,038 (57%) of 7,096 mutations in A were converted to wild type and efficiently modeled.

Through the above results, using the prediction system according to an aspect, it is possible to predict both the nucleotide correction efficiency of nucleotide correction for human point mutation disease and the frequency of possible correction results, so that the optimal nucleotide correction It was confirmed that the selected base-correcting gene scissors can provide a primary result regarding whether or not a therapeutic effect will be shown in the disease.

In particular, in the case of nucleotide-correction gene scissors, unlike the existing CRISPR gene scissors, efficiency prediction alone is insufficient for gene scissors selection and it is necessary to predict not only efficiency but also the nucleotide correction result. It is expected that by predicting the efficiency of the editing gene and the frequency of various nucleotide correction results, it is possible to select the gene scissors that can be safely edited, and to design an appropriate guide RNA accordingly, which is expected to be usefully used in the treatment of genetic diseases.

The above description is for illustration, and those of ordinary skill in the art to which the present invention pertains will understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive.

<110> Industry-Academic Cooperation Foundation, Yonsei University <120> A system for predicting base-editing efficiency and outcome product frequencies of base editors <130> PN134603KR <160> 28 <170> KoPatentIn 3.0 <210> 1 <211> 55 <212> DNA <213> Artificial Sequence <220> <223> primer_F for oligonucleotide pool amplification <400> 1 ttgaaagtat ttcgatttct tggctttata tatcttgtgg aaaggacgaa acacc 55 <210> 2 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_R for oligonucleotide pool amplification <400> 2 gagtaagctg accgctgaag tacaagtggt agagtagaga tctagttacg ccaagct 57 <210> 3 <211> 56 <212> DNA <213> Artificial Sequence <220> <223> primer_F_1 for 1st PCR <400> 3 acactctttc cctacacgac gctcttccga tctcttgaaa aagtggcacc gagtcg 56 <210> 4 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_2 for 1st PCR <400> 4 acactctttc cctacacgac gctcttccga tcttcttgaa aaagtggcac cgagtcg 57 <210> 5 <211> 58 <212> DNA <213> Artificial Sequence <220> <223> primer_F_3 for 1st PCR <400> 5 acactctttc cctacacgac gctcttccga tctcgcttga aaaagtggca ccgagtcg 58 <210> 6 <211> 61 <212> DNA <213> Artificial Sequence <220> <223> primer_R_1 for 1st PCR <400> 6 gtgactggag ttcagacgtg tgctcttccg atctttaagt cgagtaagct gaccgctgaa 60 g 61 <210> 7 <211> 62 <212> DNA <213> Artificial Sequence <220> <223> primer_R_2 for 1st PCR <400> 7 gtgactggag ttcagacgtg tgctcttccg atctattaag tcgagtaagc tgaccgctga 60 ag 62 <210> 8 <211> 63 <212> DNA <213> Artificial Sequence <220> <223> primer_R_3 for 1st PCR <400> 8 gtgactggag ttcagacgtg tgctcttccg atcttattaa gtcgagtaag ctgaccgctg 60 aag 63 <210> 9 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_1 for 2nd PCR <400> 9 aatgatacgg cgaccaccga gatctacact atagcctaca ctctttccct acacgac 57 <210> 10 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_2 for 2nd PCR <400> 10 aatgatacgg cgaccaccga gatctacaca tagaggcaca ctctttccct acacgac 57 <210> 11 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_3 for 2nd PCR <400> 11 aatgatacgg cgaccaccga gatctacacc ctatcctaca ctctttccct acacgac 57 <210> 12 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_4 for 2nd PCR <400> 12 aatgatacgg cgaccaccga gatctacacg gctctgaaca ctctttccct acacgac 57 <210> 13 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_5 for 2nd PCR <400> 13 aatgatacgg cgaccaccga gatctacaca ggcgaagaca ctctttccct acacgac 57 <210> 14 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_6 for 2nd PCR <400> 14 aatgatacgg cgaccaccga gatctacact aatcttaaca ctctttccct acacgac 57 <210> 15 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_7 for 2nd PCR <400> 15 aatgatacgg cgaccaccga gatctacacc aggacgtaca ctctttccct acacgac 57 <210> 16 <211> 57 <212> DNA <213> Artificial Sequence <220> <223> primer_F_8 for 2nd PCR <400> 16 aatgatacgg cgaccaccga gatctacacg tactgacaca ctctttccct acacgac 57 <210> 17 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_1 for 2nd PCR <400> 17 caagcagaag acggcatacg agatcgagta atgtgactgg agttcagacg tgt 53 <210> 18 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_2 for 2nd PCR <400> 18 caagcagaag acggcatacg agattctccg gagtgactgg agttcagacg tgt 53 <210> 19 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_3 for 2nd PCR <400> 19 caagcagaag acggcatacg agataatgag cggtgactgg agttcagacg tgt 53 <210> 20 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_4 for 2nd PCR <400> 20 caagcagaag acggcatacg agatggaatc tcgtgactgg agttcagacg tgt 53 <210> 21 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_5 for 2nd PCR <400> 21 caagcagaag acggcatacg agatttctga atgtgactgg agttcagacg tgt 53 <210> 22 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_6 for 2nd PCR <400> 22 caagcagaag acggcatacg agatacgaat tcgtgactgg agttcagacg tgt 53 <210> 23 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_7 for 2nd PCR <400> 23 caagcagaag acggcatacg agatagcttc aggtgactgg agttcagacg tgt 53 <210> 24 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_8 for 2nd PCR <400> 24 caagcagaag acggcatacg agatgcgcat tagtgactgg agttcagacg tgt 53 <210> 25 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_9 for 2nd PCR <400> 25 caagcagaag acggcatacg agatcatagc cggtgactgg agttcagacg tgt 53 <210> 26 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_10 for 2nd PCR <400> 26 caagcagaag acggcatacg agatttcgcg gagtgactgg agttcagacg tgt 53 <210> 27 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_11 for 2nd PCR <400> 27 caagcagaag acggcatacg agatgcgcga gagtgactgg agttcagacg tgt 53 <210> 28 <211> 53 <212> DNA <213> Artificial Sequence <220> <223> primer_R_12 for 2nd PCR <400> 28 caagcagaag acggcatacg agatctatcg ctgtgactgg agttcagacg tgt 53

Claims (21)

a target sequence input unit for receiving a target sequence of the base editing gene scissors; and
By applying the target sequence input from the target sequence input unit to the efficiency prediction model and the correction result prediction model, respectively, the efficiency and correction result score of the nucleotide correction gene scissors are obtained, and the efficiency score and the correction result score are multiplied by the nucleotide correction gene scissors. A nucleotide correction efficiency and result prediction system of nucleotide correction gene scissors, comprising a result prediction unit for simultaneously predicting efficiency and result.
The method according to claim 1,
The efficiency prediction model is
receiving activity data of base-correcting gene scissors through an information input unit; and
Base correction of base correction gene scissors, which is generated through the step of generating an efficiency prediction model by performing deep learning based on a convolutional neural network (CNN) using the data received from the information input unit Efficiency and outcome prediction system.
The method according to claim 1,
The calibration result prediction model is
Receiving an input through an information input unit for receiving the correction result data of the base correction gene scissors; and
The base of the base correction gene scissors, which is generated by performing deep learning based on a convolutional neural network (CNN) using the data input from the information input unit and generating a prediction model for the correction result Calibration Efficiency and Outcome Prediction System.
The method according to claim 1,
The base editing gene scissors is an adenine base editing gene scissors (Adenine Base Editor: ABE) or a cytosine base editing gene scissors (Cytosine Base Editor: CBE), the base editing efficiency and result prediction system of the base editing gene scissors.
3. The method according to claim 2,
The base editing efficiency and result prediction system of the base editing gene scissors, wherein the activity data of the base editing gene scissors is a sequence context (context) around the target nucleotide of the base editing gene scissors target.
3. The method according to claim 2,
The activity data of the base-correcting gene scissors is
introducing nucleotide-correcting gene scissors into a cell library comprising a nucleotide sequence encoding a guide RNA and an oligonucleotide comprising a target sequence desired by the guide RNA;
performing deep sequencing using the DNA isolated from the cell library into which the base correction gene scissors have been introduced; and
It is obtained through the step of detecting whether the target nucleotide is converted within the range of nucleotide correction from the sequence data obtained from the deep sequencing, nucleotide correction efficiency and result prediction system of nucleotide correction gene scissors.
4. The method according to claim 3,
The data of the correction result of the base correction gene scissors is
introducing nucleotide-correcting gene scissors into a cell library comprising a nucleotide sequence encoding a guide RNA and an oligonucleotide comprising a target sequence desired by the guide RNA;
performing deep sequencing using the DNA isolated from the cell library into which the base correction gene scissors have been introduced;
It is obtained through the step of detecting a target nucleotide conversion frequency within the range of nucleotide correction from the sequence data obtained from the deep sequencing, nucleotide correction efficiency and result prediction system of nucleotide correction gene scissors.
The method according to claim 1,
The efficiency score is calculated using the following [Equation 1] for each position of the target sequence, the nucleotide correction efficiency and result prediction system of the nucleotide correction gene scissors.
[Equation 1]

The method according to claim 1,
The correction result score is calculated using the following [Equation 2], the nucleotide correction efficiency and result prediction system of the nucleotide correction gene scissors.
[Equation 2]

The method according to claim 1,
The target sequence is composed of 24 to 26 nucleotides, the nucleotide correction efficiency and result prediction system of the nucleotide correction gene scissors.
The method according to claim 1,
The target sequence includes a PAM sequence and a protospacer sequence, a system for predicting base editing efficiency and results of base editing gene scissors.
The method according to claim 1,
The nucleotide correction efficiency and result prediction system of the nucleotide correction gene scissors further comprises an output unit for outputting the efficiency and result of the nucleotide correction gene scissors predicted by the result prediction unit, the nucleotide correction efficiency and result of the nucleotide correction gene scissors prediction system.
designing a target sequence of the base editing gene scissors; and
A method for predicting nucleotide correction efficiency and result of nucleotide correction gene scissors, comprising the step of applying the designed target sequence to the nucleotide correction efficiency and result prediction system according to claim 1.
obtaining human point mutation data;
first selecting data corresponding to pathogenic or pseudopathogenic point mutations from among the human point mutation data;
a case in which a point mutation occurs by changing the normal base adenine (A) to the abnormal base guanine (G) in the firstly selected data; When the normal base guanine (G) is replaced with the abnormal base adenine (A); When the normal base cytosine (C) is replaced with the abnormal base thymine T); or secondarily selecting data corresponding to a case in which the normal base thymine (T) is changed to the abnormal base cytosine (C);
a third step of selecting data in which a point mutation exists at a position 3 to 10 bp from the 5' end of the protospacer region among the secondarily selected data;
and
A method of providing information on human point mutation-related diseases that can use nucleotide correction gene scissors, comprising the step of applying the tertiary selected data to the nucleotide correction efficiency and result prediction system according to claim 1.
15. The method of claim 14,
The human point mutation-related diseases include Usher syndrome, TNF receptor-associated periodic syndrome (TRAPS), marfan syndrome, and type 3 adolescent onset diabetes (Type 3 form of Maturity-Onset Diabetes of the Young: MODY3), congenital stationary night blindness type 1F, familial hypercholesterolemia, congenital myasthenic syndrome (CMS), Lynch syndrome syndrome) were confirmed, and in the case of CBE, Loeys-Dietz syndrome (LDS), retinitis pigmentosa, leptin deficiency or dysfunction, familial hypercholesterolemia hypercholesterolemia), autosomal recessive deafness, cholesterol monooxygenase (side-chain-cleaving) deficiency, and progressive myoclonus epilepsy, any one selected from the group consisting of A method of providing information on human point mutation-related diseases that can use base-correction gene scissors.
A computer-readable recording medium in which a program for executing the method according to any one of claims 13 to 15 by a computer is recorded.
delete delete delete delete delete

Images