CN106446600A - CRISPR/Cas9-based sgRNA design method - Google Patents

Abstract

The present invention relates to a method for designing sgRNA based on CRISPR/Cas9, which is characterized in that the method comprises the following steps: obtaining the value of the cleavage efficiency of sgRNA and corresponding Cas9; establishing a personalized sgRNA design model; using NDCG algorithm to measure and establish The quality of the personalized sgRNA design model and update the database; design sgRNA and give the evaluation value of each sgRNA. Compared with the prior art, the invention has the characteristics of high accuracy, complete features, wide application range and wide analysis data.

Description

A design method of sgRNA based on CRISPR/Cas9

technical field

The invention relates to the field of gene editing research, in particular to a method for designing sgRNA based on CRISPR/Cas9 gene editing technology.

Background technique

With the development of molecular biology, people have a deeper understanding of the constituent elements of life, but there are still many puzzles about the mechanism of life processes, especially the healing mechanism of certain diseases. The relationship between genes and phenotypes, and the mutual influence between genes, urgently require an engineering technology that can quickly knock out and insert genes in vivo. The CRISPR/Cas9 system appeared in time to meet the needs of scientific researchers.

The CRISPR/Cas9 system (Clustered regularly interspaced short palindromic repeats/CRISPR-associated protein 9) is a gene editing tool with simple operation and wide applicability. The whole system is mainly composed of a nuclease (Cas9) and an RNA (sgRNA) that guides recognition. sgRNA recognizes the target gene site through complementary base pairing, and then recruits Cas9 to perform enzyme cleavage and generate double-strand breaks, thereby realizing gene editing at the DNA level. Because of its wide applicability, convenience and time-saving, it can be quickly applied to various aspects, especially in the establishment of cancer models and the exploration of gene therapy, which has great advantages.

However, in the continuous exploration of scientists, it has been found that the cleavage efficiency of different sgRNAs designed for the same gene in the same cell is very different. If a high-efficiency sgRNA cannot be designed, it can only be compensated by increasing the concentration, which will give Cells bring a lot of gene garbage, and at the same time produce a high proportion of off-targets, which brings great inconvenience to researchers. Therefore, designing a sgRNA with high enzyme cutting efficiency is very important for gene research.

At present, there are nearly 30 kinds of sgRNA design software, which are mainly divided into two categories: one is to summarize some rules of sgRNA from experiments, for example, one end of the paired sgRNA sequence must contain a PAM sequence, and the 5' end should be GG, The GC content should be kept at about 60%, the seed sequence cannot tolerate mismatches, etc., and then directly screened by setting conditions; the other type mainly uses statistical methods to assign a weight to each base to calculate the specificity of sgRNA, such as CRISPR Design. Both types of software build a general model. However, due to the large heterogeneity between different species and different cells, the prediction performance of the existing software is not very good, and because different experimental conditions The heterogeneity of sgRNA has a certain impact on the digestion efficiency of sgRNA, and the general model evaluation accuracy is relatively low.

Therefore, considering the heterogeneity of species data from different platforms, it is extremely important for the study of the off-target problem of CRISPR/Cas9 system to establish a personalized model with data from different platforms or species to improve the specificity and efficiency of sgRNA.

Contents of the invention

The purpose of the present invention is to provide a CRISPR/Cas9-based sgRNA design method with high accuracy and wide application range for the above problems.

In order to realize the purpose of the present invention, the present invention provides a method for designing sgRNA based on CRISPR/Cas9, which method comprises the following steps:

1) Obtain the value of the cleavage efficiency of the sgRNA and the corresponding Cas9, specifically:

11) Obtain the value of the cleavage efficiency of sgRNA and corresponding Cas9 from the literature;

12) Obtain the sgRNA from the SRA database, and calculate and obtain the value of the enzyme cleavage efficiency of the corresponding Cas9;

13) Classify the data obtained in steps 11) and 12) into different reference genomes according to species, cell type and experimental conditions, and each reference genome lists a copy of the first column as the sgRNA name and the second column as The sgRNA sequence and the third column are tables of the corresponding Cas9 digestion efficiency;

2) Establish a personalized sgRNA design model, specifically:

21) Extract the sequence information of the sgRNA obtained in step 1) from the corresponding reference genome as required;

22) carry out binary coding to the sgRNA sequence information extracted in step 21) according to binary rules;

23) For the sgRNA obtained in step 21), judge the data type of the enzyme cutting efficiency of its Cas9, if it is a numerical type, then enter step 24), if it is a classification type, then enter step 25);

24) For the sgRNA sequence information encoded in step 22), use the Lasso model to perform feature extraction, and establish a personalized sgRNA design model according to standard linear regression;

25) For the encoded sgRNA sequence information in step 22), perform feature selection with L1 regularization in binary logistic regression, and then establish a personalized sgRNA design model according to L2 regularization in binary logistic regression;

3) Use the NDCG algorithm to measure the quality of the personalized sgRNA design model established in step 2) and update the SRA database, specifically:

31) Calculate the NDCG value of the personalized sgRNA design model established in step 2);

32) Judging whether there is a corresponding personalized sgRNA model in the existing SRA database, if otherwise it is added to the SRA database, and if so, enter step 33);

33) compare the personalized sgRNA model with the sgRNA model in the corresponding SRA database, and select the one with a large NDCG value to be stored in the SRA database;

4) Design sgRNA and give the evaluation value of each sgRNA, specifically:

41) According to the genome region given by the user, select a suitable reference genome from the SRA database, search for all sgRNAs that meet the design rules, and use them as the designed sgRNA;

42) For the sgRNA designed in step 41), use the personalized sgRNA model established in step 2) to evaluate.

Preferably, the value of the enzyme cleavage efficiency of the corresponding Cas9 calculated in the step 12) is specifically:

121) Aligning the read length of sgRNA and corresponding next-generation sequencing to the reference genome;

122) Take out the read length comprising sgRNA;

123) Judging whether the insertion or deletion on the DNA occurs at the cutting point and whether the insertion or deletion on the DNA is a frameshift mutation;

124) Count the frameshift mutation rate of each sgRNA, specifically:

125) Use the frameshift mutation rate calculated in step 124) as the value of the enzyme cutting efficiency of Cas9.

Preferably, the sequence information of the sgRNA in the step 21) includes the sgRNA sequence, the necessary marker fragment for sgRNA recognition of DNA, and the upstream and downstream bases of the spacer of the sgRNA, and the base length of the upstream and downstream of the spacer of the sgRNA is the platform default value or a value set by the user.

Preferably, the binary rules in step 22) are specifically: A corresponds to 1000, C corresponds to 0100, G corresponds to 0010, T corresponds to 0001, and N corresponds to 0000.

Preferably, performing feature extraction with the Lasso model in the step 24) is to select the feature vector by extracting non-zero weights, specifically:

Among them, w is the weight of the estimated feature vector, x is the feature vector of the selected sgRNA, n is the number of sgRNA, y is the value of the enzyme cutting efficiency of Cas9 corresponding to the sgRNA; α is a constant, ||w| | ₁ is a matrix of parameter vectors; the Lasso model solves this least-squares loss function by adding α||w|| ₁ , and features with non-zero weights are extracted by traversing the regularization matrix.

Preferably, the L1 regularization in the step 25) is specifically:

Among them, w and c are the weights and intercepts of the estimated features, X is the binary matrix of encoded sgRNAs, n is the number of sgRNAs, and y is the value of the enzyme cleavage efficiency of Cas9 corresponding to sgRNAs.

Preferably, the L2 regularization is specifically:

Preferably, the NDCG value of the personalized sgRNA design model calculated and established in the step 31) is specifically:

Among them, DCG is the numerical value calculated by predicted sorting, IDCG is the ideal DCG calculated by real sorting, and rel _i is the predicted sorting value of the i-th position.

Preferably, the design rules in the step 41) are specifically:

20bp+PAM

Among them, bp is the unit indicating the length of DNA, and PAM is the necessary marker fragment for sgRNA to recognize DNA.

Compared with the prior art, the present invention has the following beneficial effects:

(1) For different types of cells of different species, personalized strategies are used, and data-driven machine learning algorithms are used for modeling, and the evaluation accuracy is greatly improved.

(2) Use new coding rules to make the found features more complete, not only between PAM and spacer.

(3) The user is given the process of constructing the model by himself, which makes the application scope wider, not limited to only some species in the database.

(4) Using the OTF rate of NGS data as the enzyme digestion rate expands the range of data that can be analyzed;

(5) Users can upload their own data to expand the database, which accelerates the accumulation of data and helps to solve the current dilemma of not being able to design the optimal sgRNA due to insufficient data.

Description of drawings

Fig. 1 is the method flowchart of establishing individualized sgRNA model and model assessment;

Figure 2 is a flowchart of the method for designing and evaluating sgRNAs.

detailed description

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is carried out on the premise of the technical solution of the present invention, and detailed implementation and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

Explanation of acronyms:

CRISPR: Clustered regularly interspaced short palindromic repeats

clustered regularly interspaced small palindromic repeats

Cas9: an enzyme related to the CRISPR type II system

NGS: Next Generation Sequencing, next generation sequencing

PAM: Protospacer-adjacent motif, a marker fragment necessary for sgRNA to recognize DNA

sgRNA: RNA that acts as a guide in the CRISPR/Cas9 system

indel: insertions and deletions on DNA caused by CRISPR/Cas9 editing

spacer: About 20 bases in the sgRNA that start complementary base pairing

OTF: out of frame, frame shift mutation.

Read: read length, which is the sequencing sequence obtained in one reaction in high-throughput sequencing.

This embodiment provides a method for designing sgRNA based on CRISPR/Cas9. The process of establishing a personalized sgRNA design model for different types of cells of different species can be established according to different requirements and design sgRNA, specifically including the following four steps:

(1) Data collection: The receipts collected from the literature are generally of two types: sgRNA and the corresponding enzyme digestion efficiency numerical type or sgRNA and the corresponding enzyme digestion efficiency type (such as valid or invalid); from SRA The NGS downloaded in the database only has a numerical type. Because the flow of NGS data after OTF rate statistics is consistent with the numerical type collected in the literature, this example only elaborates on the literature classification and NGS data.

Classified data: For the classified data collected from the literature, this embodiment stipulates that valid is 1, and invalid is 0, and it is sorted into a format such as Table 1.

Table 1

sgID sequence Score sgRNA_1 CGCAACCTGCTCCAGCGCCTACGG 1 sgRNA_2 CAGTCTACATAACACGCCCATGG 1 sgRNA_3 CGCAACCTGCTCCAGCGCCTACGG 1 ... ... ... sgRNA_1_1 GGCAACCGTGGCGGCAATCGAGG 0 sgRNA_2_2 CTTCTCGGAATTCGGTGAAGGTGG 0 sgRNA_3_3 AACCTCCCGGCTTCTCGGAATTCGG 0 ... ... ...

Numerical data: For NGS numerical data, first compare the sgRNA sequence and NGS reads to the human reference genome through BWA, take out the reads containing sgRNA, and judge whether indels are generated at the cutting point and whether the indels are OTF , and then count the OTF rate of each sgRNA (OTF rate=the total number of reads containing the sgRNA and OTF divided by the total number of reads containing the sgRNA). Finally, it is organized into the format shown in Table 2.

Table 2

sgID sequence Score sgRNA_1 CGCAACCTGCTCCAGCGCCTACGG 0.2345 sgRNA_2 CAGTCTACATAACACGCCCATGG 0.7846 sgRNA_3 CGCAACCTGCTCCAGCGCCTACGG 0.2367 ... ... ...

(2) Modeling: As shown in Figure 1, the sequence information of the collected sgRNA is extracted from the corresponding reference genome. Assuming that the upstream and downstream sequences are set to be 35 and 32 bases respectively, the extracted sequence is 90 (35+20+3+32) bases. CACCTGGTAT GTTCGTATCG GGCAGAATATCGCAACCTGC TCAGCGCC TA CGGTCCATCT CGCTCAGGTACGACTGACCGACCCAGTCTA.

The extracted sgRNA information is binary coded, and the rules are shown in Table 3.

table 3

Then the above 90 bases can be coded as:

0100 1000 0100 0100 0001 0010 0010 0001 1000 0001

0010 0001 0001 0100 0010 0001 1000 0001 0100 0010

0010 0010 0100 1000 0010 1000 1000 0001 1000 0001

0100 0010 0100 1000 1000 0100 0100 0001 0010 0100

0001 0100 1000 0010 0100 0010 0100 0100 0001 1000

0100 0010 0010 0001 0100 0100 1000 0001 0100 0001

0100 0010 0100 0001 0100 1000 0010 0010 0001 1000

0100 0010 1000 0100 0001 0010 1000 0100 0100 0010

1000 0100 0100 0100 1000 0010 0001 0100 0001 1000

Use machine learning methods to extract features and build a personalized sgRNA design model.

For categorical data, logistic regression is used to select features and build predictive models. The binary classification logistic regression has two optional regularizations. The present invention uses L1 regularization for feature selection, and L2 regularization for model building.

L1 regularized logistic regression solves the following optimization problems for sparse feature selection:

Among them, w and c are the weights and intercepts of the estimated features, X is the feature representation of the training samples, n is the number of training samples, and y is the enzyme cutting efficiency value corresponding to the sgRNA.

Solve the minimized value function with L2 penalized logistic regression:

For numerical data, the Lasso model is used for feature selection, and the standard linear regression is used to establish a predictive model. Lasso is a linear model that estimates sparse correlation coefficients, mainly by extracting non-zero weights to select feature vectors. The objective function to minimize is:

Among them, w is the weight of the estimated feature vector, x is the feature vector of the selected sgRNA, n is the number of training samples, y is the enzyme cutting efficiency value corresponding to the sgRNA; α is a constant, ||w|| ₁ is a matrix of parameter vectors; the Lasso model solves this least squares loss function by adding α||w|| ₁ , and by traversing the regularization matrix, features with non-zero weights are extracted, which are considered to be important influences Elements of sgRNA cleavage efficiency.

After selecting these features, a standard linear regression is then used to build an evaluation model.

Two files are generated for both numerical and subtype modeling results: one is an xml file, which contains selected features and cross-validation results; the other file is a pkl file, which contains the established predictive model and a binary file.

The content of the xml file is as follows:

(3) Evaluation model: The NDCG algorithm is used to measure the quality of the prediction model. NDCG (Normalized Discounted Cumulative Gain, normalized discounted cumulative gain) is mainly used to measure the performance of a sorting model, and its value represents the predicted sorting results and The similarity between the actual rankings ranges between 0 and 1, 1 means complete agreement, and the larger the value, the better the model. The specific formula is as follows:

DCG (Discounted Cumulative Gain, discounted cumulative gain) is a value calculated by predictive sorting, and IDCG (ideal DCG) is an ideal DCG calculated by real sorting. The mathematical definition of DCG is as follows:

where rel _i is the ranking value predicted at the i-th position.

As shown in the table below, sgID is the name of the sgRNA, seq is the spacer sequence of the sgRNA, Benchmark Score is the benchmark score, BS_rank is the ranking of the Benchmark Score, Cage is the score of the prediction model evaluation of the present invention, and C_rank is the ranking of the Cage as shown in Table 4 Show.

Table 4

sgID seq Benchmark Score BS_rank Cage Score C_rank sg1000 GCAGGTACCCTGCAACGTCGCGG 0.789456865 1 0.6905 1 sg1001 CTCCACTAGTCCCCGCGCCGCGG 0.506422166 2 0.6026 2 sg1 GTAATGGCTTCCTCGTGAGTTGG 0.325738326 3 0.5548 3 sg1002 GACTCCGTTGGGATCCGCGCCGG 0.092078991 4 0.5095 4 sg10 ATCTTAAGCAAACGCTTACCAGG 0.072255575 5 0.4959 5 sg1003 CCCGAAACGGTTGACTCCGTTGG 0.037552375 6 0.4473 6 sg1004 AGGCGCGCGATCCAGGTAGCTGG 0.019922477 7 0.3281 8 sg100 AAAAAGCTGATGAAGTTGTTTGG 0.017296539 8 0.3357 7 sg1005 CGGGGCCACCGCGACGTTGCAGG 0.002206787 9 0.3056 9 ... ... ... ... ... ...

TOP50 NDCG＝0.876322904

TOP 10% NDCG = 0.84340749

If there is no such model in the database, it will be updated to the database, otherwise, the NDCG values of the two groups will be calculated for comparison, and if the new model is greater than the NDCG value of the existing model, it can be updated to the database.

(4) Design and evaluation: As shown in Figure 2, to evaluate the sgRNA designed by the user or to design the sgRNA for the genomic region given by the user (such as chromosome 1, 1,000,000 to 1,002,000, hg19), first determine the The species or cell type of the sgRNA to be evaluated, and then select a suitable model for evaluation. If there is no suitable model, a similar model can be selected. This example provides 10 models involving 3 species and 8 types of cells for selection. . The resulting output is shown in Table 5.

table 5

At this point, users can choose the sgRNA that suits their needs for further research.

Claims (9)

1. the method for designing based on the sgRNA of CRISPR/Cas9, it is characterised in that the method comprises the following steps：

1) obtain the value of the digesting efficiency of sgRNA and corresponding Cas9, be specially：

11) from document, obtain the value of the digesting efficiency of sgRNA and corresponding Cas9；

12) from SRA database, obtain sgRNA, calculate the value of the digesting efficiency obtaining corresponding Cas9；

13) according to species, cell type and experiment condition by step 11) and 12) in the data that get be categorized into different ginsengs Examining genome, each a first is classified as sgRNA title, second is classified as sgRNA sequence and the with reference to listing in genome The form of three digesting efficiency being classified as corresponding Cas9；

2) set up personalized sgRNA to design a model, be specially：

21) according to demand from corresponding with reference to genome, extraction step 1) in the sequence information of sgRNA that obtains；

22) to step 21) in extract sgRNA sequence information carry out binary coding according to binary rules；

23) to step 21) the middle sgRNA obtaining, it is judged that the data type of the digesting efficiency of its Cas9, if numeric type then enters Step 24), if classifying type then enters step 25)；

24) to step 22) in coding after sgRNA sequence information, carry out feature extraction with Lasso model, according to normal linearity Return the personalized sgRNA of foundation to design a model；

25) to step 22) in coding after sgRNA sequence information, with two sorted logics recurrence in L1 regularization carry out feature Selecting, the L2 regularization in returning further according to two sorted logics is set up personalized sgRNA and is designed a model；

3) use NDCG algorithm to weigh step 2) in the quality that designs a model of personalized sgRNA set up update SRA database, It is specially：

31) calculation procedure 2) in the NDCG value that designs a model of personalized sgRNA set up；

32) judge whether existing SRA database has corresponding personalized sgRNA model, if being otherwise added to SRA data Storehouse, if then entering step 33)；

33) the sgRNA model in this personalization sgRNA model and corresponding SRA database is compared, select that NDCG value is big one It is stored in SRA database；

4) design sgRNA the assessed value providing each sgRNA, be specially：

41) genome area being given according to user, it is suitable with reference to genome to choose from SRA database, therefrom searches for institute There is the sgRNA meeting design rule, as the sgRNA of design；

42) to step 41) in design sgRNA, use step 2) in set up personalized sgRNA model be estimated.

2. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 1, it is characterised in that described step 12) value of the digesting efficiency being calculated corresponding Cas9 in is specially：

121) the long comparison of reading of sgRNA and corresponding two generations order-checking to reference on genome；

122) reading comprising sgRNA is taken out long；

123) judge whether the insertion in the insertion whether cut point produces on DNA or deletion and DNA or deletion are frameshit Sudden change；

124) add up the frameshift mutation rate of each sgRNA, be specially：

125) using step 124) in calculated frameshift mutation rate as the value of the digesting efficiency of Cas9.

3. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 1, it is characterised in that described step 21) in, the sequence information of sgRNA includes that sgRNA sequence, sgRNA identify the spacer of the required mark fragment of DNA and sgRNA The base of upstream and downstream, the bases longs of the upstream and downstream of the spacer of described sgRNA is the value of platform default value or user setup.

4. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 1, it is characterised in that described step 22) binary rules in is specially：A correspondence 1000, C correspondence 0100, G correspondence 0010, T correspondence 0001, N correspondence 0000.

5. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 1, it is characterised in that described step 24) carrying out feature extraction with Lasso model in is to select characteristic vector by extracting non-zero weight, is specially：

$\underset{w}{min}\frac{1}{2n}\left|\right|xw-y|{|}_2^2+\mathrm{\α}|\left|w\right|{|}_1$

Wherein, w is the weight of estimative characteristic vector, and x is the characteristic vector of selected sgRNA, and n is the quantity of sgRNA, Y is the value of the digesting efficiency of the corresponding Cas9 of sgRNA；α is a constant, | | w | |₁It is the matrix of parameter vector；Lasso model By increasing α | | w | |₁Solving this least square loss function, by traversal regularization matrix, the feature of non-zero weight is carried Take out.

6. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 1, it is characterised in that described step 25) the L1 regularization in is specially：

$\underset{w,c}{min}\frac{1}{2}\left|\right|w|{|}_1+{\mathrm{C\Σ}}_{i=1}^{n}log\left(\exp \right(-y_i\left(X_i^{T}w+c\right))+1)$

Wherein, w and c is weight and the intercept of estimative feature, and X is the binary matrix of the sgRNA of coding, and n is sgRNA Quantity, y is the value of the digesting efficiency of the corresponding Cas9 of sgRNA.

7. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 6, it is characterised in that described L2 is just Then change and be specially：

$\underset{w,c}{min}\frac{1}{2}{w}^{T}w+{\mathrm{C\Σ}}_{i=1}^{n}log\left(\exp \right(-y_i\left(X_i^{T}w+c\right))+1).$

8. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 1, it is characterised in that described step 31) the NDCG value that the personalized sgRNA that in, calculating is set up designs a model is specially：

$NDCG=\frac{DCG}{IDCG}$

$DCG={\mathrm{rel}}_1+\underset{i=2}{\overset{n}{\Σ}}\frac{{\mathrm{rel}}_i}{{\log }_{2}i}$

Wherein, DCG is the numerical value calculating with prediction sequence, and IDCG is the preferable DCG, rel calculating gained with true sequence_iIt is The ranking value of the i-th position prediction.

9. the method for designing of the sgRNA based on CRISPR/Cas9 according to claim 1, it is characterised in that described step 41) in, design rule is specially：

20bp+PAM

Wherein, bp is for representing the unit of DNA length, and PAM is that sgRNA identifies the required mark fragment of DNA.