Candidate shRNA efficiency evaluation method and device for RNA interference

Abstract

The method for evaluating the efficiency of RNA interference is characterized by comprising: a step in which an evaluation device receives'sequence information of candidate guide RNA (guide RNA) for RNA interference 'and'information on the amount of pril-shRNA remaining after Drosha cleavage', and a step in which the evaluation device receives the sequence information of the candidate guide RNA (guide RNA) for RNA interference 'and the information on the amount of pril-shRNA remaining after Drosha cleavage; a step in which the evaluation device inputs the received'guide RNA sequence information 'and'pril-shRNA amount information' remaining after cleavage into an evaluation model for evaluating the efficiency of shRNA for RNA interference; and a step in which the evaluation device evaluates the efficiency of the shRNA for RNA interference on the basis of the value output by the evaluation model.

Description

Technical Field

[0001] The methods described below relate to apparatus and methods for evaluating the efficiency of candidate shRNAs for RNA interference. Background Art

[0002] RNA interference refers to the use of short RNA to artificially regulate the expression of specific genes. RNA interference can be performed using miRNA and siRNA.

[0003] Short hairpin RNA (shRNA) is one of the RNAs used for RNA interference. shRNA refers to the shRNA expressed in cells after transformation to perform RNA interference, so as to mimic the mature pathway of microRNA. Summary of the invention

[0004] Technical issues

[0005] In order to effectively regulate gene expression by the shRNA method, the design of the guide RNA (guide RNA) for shRNA is very important. As is well known, there are several existing guide RNA design methods. However, the problem with the existing known methods is that the data of siRNA is used instead of shRNA or only the guide RNA sequence information is used.

[0006] The invention described below aims to provide a method for evaluating the efficiency of candidate shRNA for RNA interference, which method uses the information of the guide RNA sequence for shRNA together with the information of the cleavage efficiency of Drosha obtained by pri-shRNA.

[0007] Solutions to the problem

[0008] The method for evaluating the efficiency of candidate shRNA for RNA interference includes: a step in which an evaluation device receives sequence information of a candidate guide RNA for RNA interference and information on the amount of pri-shRNA remaining after being cut by Drosha; a step in which the evaluation device inputs the received guide RNA sequence information and the information on the amount of pri-shRNA remaining after cutting into an evaluation model for evaluating the efficiency of shRNA for RNA interference; and a step in which the evaluation device evaluates the efficiency of shRNA for RNA interference based on the value output by the evaluation model.

[0009] The device for evaluating the efficiency of candidate shRNA for RNA interference includes: an input device for receiving input of candidate guide RNA sequence information for RNA interference and information on the amount of pri-shRNA remaining after being cut by Drosha; a storage device for storing an evaluation model for evaluating the efficiency of shRNA for RNA interference using the guide RNA sequence information received by the input device and the information on the amount of pri-shRNA remaining after cutting; and a computing device for inputting the guide RNA sequence information received by the input device and the information on the amount of pri-shRNA remaining after cutting into the evaluation model stored in the storage device, and then evaluating the efficiency of shRNA for RNA interference according to the value output by the evaluation model.

[0010] Effects of the Invention

[0011] Using the invention described below, not only can the guide RNA sequence information for shRNA be used, but also the cutting efficiency information of Drosha be used to grasp the efficiency of candidate shRNA for RNA interference. In addition, when using the technology described below, the flanking sequence information of the guide RNA can also be further used to grasp the efficiency of candidate shRNA. Thus, it can also contribute to the design of candidate shRNA for RNA interference.

[0012] When the invention described below is used, it can contribute to the development of therapeutic drugs for diseases using RNA interference methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is an example showing the overall process of judging the efficiency of shRNA for RNA interference.

[0014] Figure 2 This is an experimental result showing that the 'pri-shRNA amount' and 'shRNA efficiency' are negatively correlated.

[0015] Figure 3 This is an experimental result showing that the 'pri-shRNA amount' and the 'efficiency of pri-shRNA cleavage by drosha' are negatively correlated.

[0016] Figure 4 310 is an example showing the kind and amount of learning or test data used in evaluating the model 310 .

[0017] Figure 5 An embodiment of a structure of an evaluation model implemented as a learning model by a convolutional neural network (CNN) is shown.

[0018] Figure 6This is an experimental result of confirming the performance of the evaluation model 310 based on data that can be additionally input. Figure 7 An embodiment of the structure of a prediction model implemented as a learning model by a convolutional neural network (CNN) is shown.

[0019] Figure 8 are the experimental results for evaluating the performance of the prediction model.

[0020] Fig. 9 These are experimental results for evaluating the performance of shRNA efficiency prediction methods using the evaluation model and the prediction model.

[0021] Fig.10 are experimental results using other datasets instead of the 'TILE' dataset to evaluate the performance of the shRNA efficiency prediction method using the evaluation model and the prediction model.

[0022] Fig.11 These are experimental results that compare the performance of an evaluation device using an evaluation model and a prediction model with the performance of an existing evaluation device.

[0023] Fig.12 1 is an example showing the overall process of the evaluation apparatus 300 additionally receiving the flanking sequence information 140 to evaluate the efficiency of RNA interference.

[0024] Fig.13 The results of analyzing the efficiency of RNA interference according to the number of AU (adenine (A) and uracil (U)) in the downstream flanking 2nt.

[0025] Fig.14 are the experimental results where the performance of the enhanced evaluation model is compared with that of the evaluation model.

[0026] Fig.15 is the result of comparing the performance of the enhanced evaluation model with other evaluation models.

[0027] Fig.16 The entirety of the figure shows an experimental process for confirming whether the enhanced evaluation model and the efficiency of RNA interference evaluated by the evaluation model are consistent with the experimental results in actual cells.

[0028] Fig.17 These are the results of analyzing the enhanced evaluation model and the evaluation model for evaluating the efficiency of RNA interference, as well as the knockdown (KD) results of each gene.

[0029] Fig.18 2 is an example showing the configuration of an evaluation device 600 for evaluating the efficiency of shRNA for RNA interference. DETAILED DESCRIPTION

[0030] The technology described below can be modified in various ways and has various embodiments, and specific embodiments are shown in the drawings and described in detail. However, this is not intended to limit the technology described below to specific embodiments, but should be understood to include all modifications, equivalents and even substitutes covered by the technical ideas and counting ranges expressed below.

[0031] The terms first, second, A, B, etc. may be used to describe various constituent elements, but these constituent elements are not limited by these terms and are only used to distinguish one constituent element from another constituent element. For example, without departing from the scope of the invention claims of the technology described below, the first constituent element may be named as the second constituent element, and similarly, the second constituent element may be named as the first constituent element. The term "and / or" includes a combination of multiple related described items or any one of multiple related described items.

[0032] It should be understood that, among the terms used in this specification, unless the context clearly provides otherwise, singular expressions include plural expressions, and terms such as "including" and "comprising" mean the existence of the described features, quantities, steps, operations, constituent elements, parts or combinations thereof, and do not exclude the existence or additional possibilities of one or more other features or quantities, steps, operations, parts or combinations thereof.

[0033] Before the detailed description of the drawings, it should be clear that the distinction between the constituent parts in this specification is based only on the main functions that each constituent part is responsible for. That is, two or more constituent parts described below are combined into one constituent part or one constituent part is divided into two or more constituent parts according to further subdivided functions. And, of course, each constituent part described below, in addition to the main functions it is responsible for, can also perform part or all of the functions that other constituent parts are responsible for, and part of the main functions that each constituent part is responsible for can also be performed exclusively by other constituent parts.

[0034] In addition, when performing a method or operating a method, unless the context clearly indicates a specific order, the various processes that constitute the method may occur in an order different from the order described. In other words, the various processes may be performed in the same order as the specified order, may be performed substantially simultaneously, or may be performed in the reverse order.

[0035] Before explaining the technology described below, necessary terms are explained.

[0036] In the techniques described below, RNA interference refers to the process by which RNA molecules block gene expression.

[0037] In the technology described below, short hairpin RNA (shRNA) is one of the methods of RNA interference. This method refers to transforming cells with plasmids to produce shRNA, and then the cells produce miRNAs on their own to perform RNA interference.

[0038] In the technology described below, pri-shRNA refers to the product first transcribed by RNA polymerase from transformed cells for RNA interference by the shRNA method. Pri-shRNA is cleaved into pre-shRNA in the nucleus together with Drosha and other proteins.

[0039] In the techniques described below, Drosha refers to an enzyme that prepares pre-shRNA by cleaving pri-shRNA in the nucleus.

[0040] In the technology described below, the learning model refers to a machine learning model, and the machine learning model can also be various types of models. For example, the machine learning model can also be a decision tree, a random forest (RF), a K-nearest neighbor (KNN), a naive Bayes, a support vector machine (SVM), an artificial neural network (ANN), etc.

[0041] The ANN may be a deep neural network (DNN), or may include a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), a deep belief network (DBN), a generative adversarial network (GAN), a relational network (RL), etc. In the following, for the sake of convenience, the CNN model is used as the center for explanation.

[0042] The CNN model may include a convolution layer, a pooling layer, a dropout layer, a batch normalization layer, a flatten layer, and a fully connected layer (or dense layer).

[0043] The convolution layer extracts features from the input information and creates a feature map from it. The pooling layer extracts the maximum value (Max Pooling layer) or the average value (Average Pooling Layer) from the value of the feature map in order to reduce the size of the feature map made in the convolution layer or to emphasize specific data. The dropout layer uses only a part of the neural network model during the training process to prevent overfitting in the deep learning model. The batch normalization layer standardizes each batch during the training process to prevent overfitting in the deep learning model. The flatten layer makes the features of the extracted data into one dimension. The fully connected layer connects the nodes of each layer and finally determines which category the input data belongs to. In this process, the SoftMAX function can also be used.

[0044] First, the overall process of determining the efficiency of shRNA for RNA interference will be described.

[0045] Figure 1 is an example showing the overall process of judging the efficiency of shRNA for RNA interference.

[0046] exist Figure 1 In the embodiment, the evaluation device 300 receives data through the input device 100, thereby evaluating the efficiency of shRNA used for RNA interference.

[0047] like Figure 1 As shown in (A), the evaluation device 300 may use only the evaluation model 310, or may use Figure 1 As shown in (B), both the evaluation model 310 and the prediction model 320 are used.

[0048] The efficiency of the shRNA for RNA interference evaluated by the evaluation device 300 may include information on the degree of RNA interference occurring by means of shRNA.

[0049] exist Figure 1 In the example, the data received by the evaluation device 300 are, as shown in (A), 'sequence information 110 of the guide RNA for RNA interference (hereinafter, guide RNA sequence information)' and 'the amount of pri-shRNA remaining after cleavage by Drosha 120 (hereinafter, information on the amount of pri-shRNA)'. Alternatively, as shown in (B), 'sequence information 130 of the Drosha cleavage site in the pri-shRNA (hereinafter, Drosha cleavage site sequence information)' may be received instead of 'the information on the amount of pri-shRNA 120'.

[0050] The 'guide RNA sequence information 110 for RNA interference' used as the input data represents the guide RNA sequence information for RNA interference. The sequence information may be information having a length of 22 nt (nucleotide).

[0051] The 'pri-shRNA amount remaining after cleavage by Drosha 120' used as the input data may also mean the amount remaining after cleavage by Drosha among the pri-shRNA amount produced from the transformed cells. This may be done by using the actually measured value as shown in (A) or by using the value predicted by the prediction model as shown in (B).

[0052] The 'Drosha cleavage site sequence information 130' used as the input data may include sequence information near the site actually cleaved by drosha in the pri-shRNA. This may also include information that in the pri-shRNA, a sequence from 27 nt (nucleotides) upstream to 29 nt downstream based on the 5p Drosha cleavage site (5p Drosha cleavage site sequence) is connected to a sequence from 31 nt upstream to 25 nt downstream based on the 3p Drosha cleavage site (3p Drosha cleavage site sequence) and information that the 3p Drosha cleavage site sequence is flipped and connected to the 5p cleavage site sequence.

[0053] The 'amount of pri-shRNA remaining after being cleaved by Drosha 120' or the 'sequence near the cleavage site of Drosha in pri-shRNA 130' used as the input data may be related to the efficiency of shRNA. Specifically, the amount of pri-shRNA may also be related to the cleavage efficiency of Drosha. The specific content is described below.

[0054] exist Figure 1 In the example, the evaluation model 310 uses the received 'guide RNA sequence information 110' and 'pri-shRNA amount information 120' as data basis to output a result value for evaluating the efficiency of shRNA for RNA interference.

[0055] The evaluation model may be a learning model that is learned by learning existing data to evaluate the efficiency of shRNA for RNA interference.

[0056] exist Figure 1 In the example, the prediction model 320 receives the 'Drosha cleavage site sequence information' 110 and outputs a result value for predicting the 'pri-shRNA amount remaining after the pri-shRNA is cleaved by Drosha'.

[0057] The prediction model may be a learning model that predicts the amount of pri-shRNA remaining after pri-shRNA is cleaved by Drosha by learning existing data.

[0058] The prediction model may be stored in the same evaluation device as the evaluation model and used, or may be stored in a prediction device different from the evaluation device and used (not shown).

[0059] In the shRNA efficiency analysis method using the evaluation model, the 'amount of pri-shRNA remaining after cleavage by Drosha' is used because the 'amount of pri-shRNA remaining after cleavage by Drosha' shows the cleavage efficiency of Drosha, which is negatively correlated with the 'shRNA efficiency'.

[0060] The following is an experimental result confirming that the amount of pri-shRNA is negatively correlated with the efficiency of shRNA ( Figure 2 , Figure 3 ) for explanation.

[0061] Figure 2 This is an experimental result showing that the 'pri-shRNA amount' and 'shRNA efficiency' are negatively correlated.

[0062] Figure 2The figure is a figure analyzed using RNA sequencing data of 'TILE' data among the data types described below.

[0063] exist Figure 2 In the graph, the X-axis shows the relative amount of pir-shRNA in the cells, and the Y-axis shows the result value of measuring the efficiency of RNA interference.

[0064] like Figure 2 As shown in the figure, it can be confirmed that as the amount of pri-shRNA increases, the efficiency of RNA interference decreases. Conversely, it can be seen that as the amount of pri-shRNA decreases, the efficiency of RNA interference increases.

[0065] By analyzing Figure 2 The result value can confirm that the amount of pri-shRNA is negatively correlated with the efficiency of RNA interference.

[0066] Figure 3 This is an experimental result showing that the 'pri-shRNA amount' and the 'efficiency of pri-shRNA cleavage by drosha' are negatively correlated.

[0067] Figure 3 The X-axis in Figure (A) and Figure (B) shows the numerical values ​​of AU (adenine and uracil) in the guide RNA and pre-shRNA, respectively, the Y-axis in Figure (A) shows the relative amount of pri-shRNA, and Figure (B) shows the degree of pri-shRNA cleavage by Drosha.

[0068] from Figure 3 As can be seen from Figure (A), the relationship between the 'number of AUs in the guide RNA' and the amount of pri-shRNA is in a [flipped "U"] relationship.

[0069] from Figure 3 As can be seen from the figure (B), it is shown that the relationship between the number of AUs in pre-miR-30 and the cleavage efficiency of Drosha is in a [perfect "U" shape] relationship. The pre-miR-30 refers to one of the pre-shRNAs, using miR-30 as the backbone.

[0070] Analysis of the Figure 3 The results of (A) and (B) show that the amount of pri-shRNA is inversely related to the cleavage efficiency of Drosha. As the amount of Pri-shRNA cleaved by Drosha increases, the amount of pre-shRNA and Mature-shRNA increases, which confirms that the efficiency of RNA interference is higher.

[0071] Said Figure 3 The conclusions of the analysis and Figure 2 This is consistent with the conclusion that 'the amount of pri-shRNA is negatively correlated with the efficiency of RNA interference'.

[0072] The researchers used the fact that the amount of 'pri-shRNA remaining after being cleaved by Drosha' is negatively correlated with 'shRNA efficiency' to implement an evaluation model 310 for evaluating shRNA efficiency.

[0073] The following describes the data used to construct the evaluation model 310. Figure 4 ) and the structure of the model ( Figure 5 ) and performance evaluation ( Figure 6 ).

[0074] Figure 4 310 is an example showing the types and amounts of learning or test data used in implementing the estimation model 310 .

[0075] The data may also be used to implement the prediction model 320 described below.

[0076] The data used in the previously published papers are used. Among the data, shERWOOD can be confirmed in "Knott, Simon RV, et al." A computational algorithm to predict shRNA potency." Molecular cell 56.6 (2014): 796-807.", TILE can be confirmed in "Fellmann, Christof, et al." Functional identification of optimized RNAi triggers using a massively parallel sensor", Ras can be confirmed in "Yuan, Tina L., et al." Development of siRNA Payloads to Target KRAS-Mutant Cancer RNAi Therapy for KRAS-Mutant Cancer." Cancer discovery 4.10 (2014): 1182-1197.", and M1 / mirR-E can be confirmed in "Pelossof, Raphael, et al." Prediction of potent shRNAs with a sequential classification algorithm." Nature biotechnology 35.4 (2017): 350-353.".

[0077] The types of data sets that can be used as the learning data include shERWOD (S), TILE (T), and M1 (M). In this case, 200,000 shERWOODs, 16,000 TILEs, and 18,000 M1s can be used as the learning data.

[0078] The types of data sets that can be used as the test data include shERWOOD (S), TILE (T), M1 (M), Ras shRNA, miR-E shRNA, and UltramiR (U). At this time, 39844 shERWOODs, 2564 TILEs, 2290 M1s, 9804 Ras, 311 miR-Es, and 780 Ultramirs can be used as test data.

[0079] The learning data and test data use miR-30 as the backbone. Among them, shERWOOD, TILE, M1 and Ras use miR-30 without CNNC sequence information, but miR-E and UltramiR use miR-30 with CNNC sequence information.

[0080] In the CNNC, C represents a cytosine residue, and N represents any nucleotide such as A, T, C, or G. CNNC refers to a sequence that, together with the UG motif, helps shRNA and SRSF3 (serine / arginine-rich splicing factor 3) to adhere to pre-shRNA to promote the cleavage of pri-shRNA into pre-shRNA.

[0081] Figure 5 An embodiment of the structure of the evaluation model 310 as a learning model is shown by using a convolution neural network (CNN).

[0082] The evaluation model 310 may be composed of a convolution layer and a fully connected layer. The convolution layer of the evaluation model may also include three convolution layers, a maximum pooling layer between the first convolution layer and the second convolution layer, and an average pooling layer after the last convolution layer. The fully connected layer of the evaluation model may include a flattening layer and two densely connected layers.

[0083] Figure 5 The sequence of the guide RNA received by the evaluation model 310 consists of a series of nucleotides. The evaluation model receives the sequence of the guide RNA in the form of one-hot vector encoding according to the type of nucleotide (A / T / G / C). A 22×4 matrix of the one-hot vector encoding of the guide RNA of 22 nt length is constructed and input into the evaluation model.

[0084] The estimation model 310 may also receive additional input data using a concatenation layer.

[0085] Figure 5The output values ​​of the convolutional layer can be batch normalized, and then, the exponential linear unit (ELU) or the rectified exponential linear unit (RELU) can be used for activation.

[0086] The size of the filter used in the first convolutional layer can be 3×4, 5×4 or 7×4, and the number of filters can be 64, 128 or 256.

[0087] The size of the filters used in the second convolution layer and the third convolution layer can be 3×1, 5×1 or 7×1, and the number of filters can be 64, 128 or 256.

[0088] Figure 5 The filter size used in the maximum pooling layer can be 2×1 and the stride can be 2.

[0089] Figure 5 The sizes of the first dense layer and the second dense layer may also be 64 or 128. After the second dense layer, in order to output the value of the efficiency of the shRNA for RNA interference, a layer may appear to output the regression result value by performing a linear transformation on the previous result value.

[0090] Figure 6 This is an experimental result of confirming the performance of the evaluation model 310 based on data that can be additionally input.

[0091] exist Figure 6 In the figure, the "T" on the X-axis is the information used for testing the evaluation model and represents the "TILE" data set, and the Y-axis represents the Spearman correlation coefficient.

[0092] like Figure 6 As shown in the figure, it can be confirmed that the performance is higher when the information of the amount of pri-shRNA (22nt+pri-shRNA, the second from the left in the figure, (3)) is used together with the guide RNA sequence information, compared with the case where only the guide RNA sequence information (22nt, the leftmost (1) in the figure) is used as the input data (input feature).

[0093] In addition, it was confirmed that the model using the amount of pre-shRNA or mature-shRNA instead of the amount of pri-shRNA as input data (22nt+pre-sRNA (3) or 22nt+mature-shRNA (4), the third and fourth from the left) showed almost no difference in performance compared to the model using only guide RNA.

[0094] according to Figure 6 The experimental results confirmed that the evaluation model that utilizes not only the guide RNA sequence information but also the pri-shRNA quantity information has the best performance.

[0095] The information of the shRNA to be evaluated may not include information on the 'actually measured amount of pri-shRNA.' Therefore, the researchers constructed a prediction model 320 for predicting the 'pri-shRNA amount' based on the given information.

[0096] The following describes the structure of the prediction model 320 ( Figure 7 ) and its performance ( Figure 8 ).

[0097] Figure 7 An embodiment of the structure of the prediction model 320 as a learning model is shown by using a convolution neural network (CNN).

[0098] The prediction model 320 may be composed of convolutional layers and fully connected layers. The convolutional layers of the evaluation model may include three convolutional layers, a maximum pooling layer between the first convolution and the second convolution, and an average pooling layer after the last convolutional layer. The fully connected layers of the prediction model may include a flattening layer and two dense layers.

[0099] Figure 7 The prediction model 320 receives a Drosha cleavage site sequence.

[0100] The Drosha cleavage site sequence includes: in the pri-shRNA, information of connecting the sequence from upstream 27nt (nucleotide) to downstream 29nt (5p Drosha cleavage site sequence) based on the Drosha cleavage site of 5p and the sequence from upstream 31nt to downstream 25nt (3p Drosha cleavage site sequence) based on the Drosha cleavage site of 3p, as well as information of flipping the 3p Drosha cleavage site sequence and connecting it to the 5p cleavage site sequence.

[0101] The Drosha cleavage site sequence is composed of a series of nucleotides. According to the types of nucleotides (A / T / G / C), the Drosha cleavage site sequence is encoded by one-hot vector.

[0102] Figure 7 The output values ​​of the convolutional layer can be batch normalized, and then, the exponential linear unit (ELU) or the rectified exponential linear unit (RELU) can be used for activation.

[0103] The size of the filter used in the first convolutional layer may be 9×9, and the number of filters may be 128, 256, or 512.

[0104] The size of the filters used in the second convolution layer and the third convolution layer may be 3×1, 5×1, or 7×1, and the number of filters may be 128, 256, or 512.

[0105] Figure 7 The filter size used in the max pooling layer can be 2×1 and the stride can be 2.

[0106] Figure 7 The size of the first dense layer and the second dense layer can also be 64 or 128.

[0107] Figure 7 The output values ​​of the densely connected layer can be batch normalized, and then, the exponential linear unit (ELU) or the rectified exponential linear unit (RELU) can be used for activation.

[0108] After the second dense layer, in order to output the predicted value of the amount of pir-shRNA remaining after being cut by Drosha, a layer may also appear for outputting a regression result value by performing a linear transformation on the result value of the previous layer.

[0109] Figure 8 It is an experimental result for evaluating the performance of the constructed prediction model 320.

[0110] exist Figure 8In the figure, the X-axis represents the actual “amount of pri-shRNA remaining after cleavage by Drosha”, and the Y-axis represents the “amount of pri-shRNA remaining after cleavage by Drosha” predicted by the prediction model.

[0111] analyze Figure 8 The result value confirms that the correlation coefficient (R value) between the actual remaining amount and the amount predicted by the prediction model is 0.9. Therefore, it is shown that the predicted value of the prediction model 320 is very consistent with the actual value.

[0112] The researchers used evaluation model 310 and prediction model 320 to confirm the performance of the method for predicting shRNA efficiency.

[0113] Below, we describe the experimental results of evaluating the performance ( Fig. 9 , Fig.10 , Fig.11 ).

[0114] Fig. 9 is an evaluation result of the performance of the shRNA efficiency prediction method using the evaluation model 310 and the prediction model 320 .

[0115] exist Fig. 9 In the figure, the “T” on the X-axis is for the test data type and represents the TILE dataset, and the Y-axis represents the Spearman correlation coefficient.

[0116] like Fig. 9 As shown, it can be seen that the performance of the model using only guide RNA sequence information as input data (input feature) (22nt, the far left in the figure, (1)) is the lowest. On the contrary, it can be confirmed that the model using the actual measured pri-shRNA amount of guide RNA sequence information as input data (22nt+observed Drosha processing surplus, (3), the far right in the figure) has the best performance. At the same time, the model using the value of pri-shRNA predicted by the prediction model as input data instead of the actual measured value (22nt+predicted Drosha processing surplus, (2), in the middle of the figure) performs better than the model using only guide RNA sequence information.

[0117] analyze Fig. 9 The result values ​​showed that the evaluation method of shRNA efficiency using both the evaluation model and the prediction model had higher performance than the case using only the guide RNA sequence information, and there was no significant difference compared to the case using the actually measured pri-shRNA amount.

[0118] Fig.10are the results of evaluating the performance of the shRNA efficiency prediction method using the evaluation model 310 and the prediction model 320 using other datasets instead of the 'TILE' dataset.

[0119] exist Fig.10 In the example, the X-axis represents the type of data used in the test (holdout, independent), and the Y-axis represents the Spearman correlation coefficient. The S, T, M, R, E, and U marked on the X-axis represent Figure 4 The data types shown in .

[0120] like Fig.10 As shown, it can be confirmed that the model that inputs not only the guide RNA sequence information but also the values ​​predicted by the prediction model (right side of each figure) has better performance than the model that uses only the guide RNA sequence information as input data (left side of each figure).

[0121] analyze Fig.10 The result value shows that when other test data are used as test data instead of the 'TILE' dataset, the performance of the evaluation device for judging the efficiency of shRNA for RNA interference by using both the evaluation model and the prediction model is also superior to the performance of the evaluation device for judging the efficiency of shRNA for RNA interference using only the guide RNA sequence information.

[0122] Fig.11 These are experimental results comparing the performance of an estimation device using both estimation model 310 and prediction model 320 with the performance of a conventional estimation device.

[0123] exist Fig.11 In the figure, shRNAI on the left side refers to an evaluation device using both the evaluation model and the prediction model, and SplashRNA on the right side refers to an existing evaluation device using only guide RNA sequence information.

[0124] Fig.11 The X-axis of each figure represents the efficiency of RNA interference actually observed, and the Y-axis represents the efficiency of RNA interference predicted by the model.

[0125] from Fig.11 It can be seen that the correlation coefficient of the evaluation device using both the evaluation model and the prediction model is 0.403, but the correlation coefficient of the existing evaluation device is 0.370.

[0126] analyze Fig.11 From the result values, it can be confirmed that the performance of the evaluation device using both the evaluation model and the prediction model is higher than that of the existing evaluation device by about 9%.

[0127] The evaluation device 300 described above can evaluate the efficiency of RNA interference by additionally receiving flanking sequence information together with the guide RNA sequence information.

[0128] Fig.12 1 is an example showing the overall process of the evaluation apparatus 300 additionally receiving the flanking sequence information 140 to evaluate the efficiency of RNA interference.

[0129] The evaluation device 300 can not only receive the guide RNA sequence information 110 and the pri-shRNA amount information 120, but also can additionally receive the flanking sequence information 140 ( Fig.12 (A)). Similarly, the evaluation device 300 can use the prediction model 320 ( Fig.12 (B)).

[0130] In order to utilize the flanking sequence information 140, the evaluation device 300 may use an enhanced evaluation model 330. The enhanced evaluation model 330 may be a model for evaluating the efficiency of RNA interference by additionally receiving the flanking sequence information 140. The enhanced evaluation model 330 may be a model based on the evaluation model 310. As an embodiment, the enhanced evaluation model 330 may be a model constructed by fine-tuning the evaluation model 310.

[0131] The additionally received flanking sequence information 140 may include the upstream and downstream sequence information of the guide RNA. As an embodiment, the flanking sequence information 140 may include the sequence information of the upstream 14nt and the downstream 14nt of the guide RNA. That is, the evaluation device 300 may evaluate the interference efficiency of the RNA based on the sequence information of 50nt including the guide RNA sequence information 22nt and the flanking sequence information 28nt (14nt×2).

[0132] Below, experimental results evaluating the process by which researchers constructed the enhanced evaluation model 330 and the performance of the constructed enhanced evaluation model 330 are discussed.

[0133] Fig.13 The results of the analysis of the efficiency of RNA interference according to the number of AUs in the downstream flanking 2nt are shown in Figure 2. The dataset used for the analysis is Figure 4 The datasets in (shERWOOD(S), TILE(T), M1(M) and Ras(R)).

[0134] analyze Fig.13The results of , it can be confirmed that the efficiency of RNA interference increases with the increase of the number of AU in the downstream flanking 2nt. This can be confirmed from each of all data sets (S, T, M, R).

[0135] pass Fig.13 It can be confirmed that RNA interference efficiency is affected not only by guide RNA sequence information but also by flanking sequence information. That is, when not only guide RNA sequence information but also flanking sequence information is used, RNA interference efficiency can be evaluated more accurately.

[0136] Based on these facts, the researchers constructed a model that can utilize not only the guidRNA sequence information but also the flanking sequence information. Specifically, based on the evaluation model 310 that receives the guide RNA sequence information and the information of the pri-shRNA quantity, the researchers constructed an enhanced evaluation model 330 that can additionally receive the flanking sequence information. Specifically, the enhanced evaluation model 330 is a model that adjusts the parameters of the evaluation model 310.

[0137] The enhanced evaluation model can receive not only the guide RNA sequence information, but also the flanking sequence information. Specifically, the enhanced evaluation model can receive information on 22nt as the guide RNA sequence information and information on 28nt (14nt×2) as the flanking sequence information. That is, the enhanced evaluation model can also receive 50nt base sequence information.

[0138] Fig.14 are the experimental results comparing the performance of the enhanced evaluation model with that of the evaluation model.

[0139] exist Fig.14 The bar on the left side of each figure is an evaluation model that uses only the guide RNA sequence information and the information on the predicted pri-shRNA amount (shRNAI S+M+P ) performance evaluation results. That is, Figure 1 Evaluation model 310 in.

[0140] exist Fig.14 The bars on the right side of each figure are the performance evaluation results of the enhanced evaluation model (shRNAI) using the flanking sequence information as additional input data. Fig.12 The enhanced evaluation model 330 in .

[0141] from Fig.14From the results, it can be confirmed that the performance of the model using the target sequence information (guide RNA sequence information + flanking sequence information) and the information on the predicted pri-shRNA amount is better than the performance of the evaluation model using only the guide RNA sequence information and the information on the predicted pri-shRNA amount. In other words, it can be confirmed that the performance of the enhanced evaluation model 330 that additionally uses the flanking sequence information is better than the performance of the evaluation model 310.

[0142] Fig.15 is the comparison result of the performance of the enhanced evaluation model with other evaluation models. Specifically, the enhanced evaluation model (shRNAI, (1)) and the evaluation model (shRNAI S+M+P , (2)) performance compared to the control group (SplashRNA T+Mpos (3), DSIR (4) and SeqScore (5). The experiments were conducted using multiple datasets. The backbones and cells used in the datasets were different.

[0143] from Fig.15 From the results, it can be confirmed that the overall enhanced evaluation model (shRNAI) and evaluation model (shRNAI S+M+P ) outperformed the control group (SplashRNA T+Mpos , DSIR, and SeqScore). In addition, it can be confirmed in some datasets (Huang 2014, Banito 2018, and David 2016) that the performance of the enhanced evaluation model (shRNAI) is better than that of the evaluation model (shRNAI S+M+P ) has higher performance.

[0144] Fig.16 The entirety of the figure shows an experimental process for confirming whether the enhanced evaluation model and the efficiency of RNA interference evaluated by the evaluation model are consistent with the experimental results in actual cells.

[0145] For evaluation of the model (shRNAI S+M+P The target genes for evaluating the performance of the enhanced evaluation model (shRNAI) were PTEN, BAP1, NF2, AXN1, PBPM1, and RELA. The target gene for evaluating the performance of the enhanced evaluation model (shRNAI) was UPF1. The vector used for transformation was mlr-E (LT3GEPIR), and qRT-PCR was used to confirm whether RNA interference was performed.

[0146] Fig.17 is to use Fig.16The experimental method is to enhance the evaluation model and the evaluation model to evaluate the efficiency of RNA interference and analyze the knockout (Knock-Down, KD) results of each gene.

[0147] Use (SplashRNA T+Mpos ) as an enhanced evaluation model (shRNAI) and an evaluation model (shRNAI S+M+P ) control group.

[0148] It can be confirmed that the control group (SplashRNA T+Mpos ) 2 is 0.061, whereas the enhanced evaluation model (shRNAI) and the evaluation model (shRNAI S+M+P ) 2 In particular, it can be confirmed that the r of the enhanced evaluation model (shRNAI) 2 Highest.

[0149] pass Fig.17 , it can be confirmed that the enhanced evaluation model (shRNAI) and the evaluation model (shRNAI S+M+P ) outperformed the control group (SplashRNA T+Mpos ) is higher. In particular, it can be confirmed that the performance of the enhanced evaluation model (shRNAI) is higher than that of the evaluation model (shRNAI S+M+P ) has higher performance.

[0150] Below, through Fig.18 A device for evaluating the efficiency of shRNA for RNA interference is described.

[0151] Fig.18 2 is an example showing the configuration of an evaluation device 600 for evaluating the efficiency of shRNA for RNA interference.

[0152] The evaluation device 600 corresponds to the evaluation device ( Figure 1 of 300).

[0153] The evaluation device 600 can be physically implemented in various forms, such as a PC, a notebook computer, a smart device, a server, or a dedicated chipset for data processing.

[0154] The evaluation device 600 may further include an input device 610 , a storage device 620 , and a computing device 630 .

[0155] The evaluation device may further include an output device 640 .

[0156] Fig.18The input device 610 may also include an interface device (keyboard, mouse, touch screen, etc.) for receiving input of predetermined instructions or data.

[0157] The input device 610 corresponds to the input device ( Figure 1 of 100).

[0158] The input device 610 may further include a communication device for receiving and transmitting predetermined information through a wired network or a wireless network.

[0159] The input device 610 may also receive information input configuration via a separate storage device (USB, CD, hard disk, etc.).

[0160] The input device 610 may receive input data through a separate measurement device or a separate DB.

[0161] The input device 610 may also receive sequence information of the guide RNA and the amount of the pri-shRNA remaining after being cleaved by Drosha.

[0162] The input device 610 may receive 'sequence information flanking the Drosha cleavage site in the pri-shRNA' instead of the information on the amount of 'pri-shRNA remaining after being cleaved by Drosha'.

[0163] The input device 610 can receive flanking sequence information.

[0164] Fig.18 The storage device 620 may also store information from the input device 610 .

[0165] The storage device 620 may also store an evaluation model for analyzing the efficiency of shRNA for RNA interference ( Figure 1 310) or a prediction model for predicting the amount of pri-shRNA ( Figure 1 Alternatively, the storage device 620 may also store the enhanced evaluation model ( Fig.12 330 in ).

[0166] The storage device 620 may also store the result values ​​output by the evaluation model and the prediction model.

[0167] The storage device 620 may also store learning data used by the learning model.

[0168] Fig.18The computing device 630 inputs the 'guide RNA sequence information' and the 'pri-shRNA amount remaining after being cleaved by Drosha' received from the input device 610 into the evaluation model, and analyzes the efficiency of the shRNA used for RNA interference based on the output value thereof.

[0169] The computing device 630 may also input the 'predicted amount of pir-shRNA' output by the prediction model instead of the 'amount of pri-shRNA remaining after being cleaved by Drosha' into the evaluation model to analyze the efficiency of shRNA used for RNA interference.

[0170] After the computing device 630 adds the 'flanking sequence information' to the evaluation model, it can evaluate the efficiency of the shRNA for RNA interference according to the output value of the evaluation model. That is, the computing device 630 can evaluate the efficiency of the shRNA for RNA interference using the enhanced evaluation model.

[0171] Fig.18 The output device 640 may be a device that outputs predetermined information.

[0172] The output device 640 can output the interface, input data, analysis results, etc. required in the data process.

[0173] The output device 640 may also be physically implemented in various forms, such as a display, a device for outputting documents, a communication device, etc.

[0174] In addition, the method for evaluating the efficiency of candidate shRNA for RNA interference as described above can be implemented by a program (or application) including an executable algorithm that can be run on a computer, and the program can be provided by being stored in a transient or non-transitory computer readable medium.

[0175] Non-temporary readable media refers to media that semi-permanently stores data and is readable by a device (reading) rather than registers, caches, memories, etc., which store data for a short period of time. Specifically, the various applications or programs mentioned above can be stored in non-temporary readable media such as CDs, DVDs, hard disks, Blu-ray discs, USBs, memory cards, read-only memory (ROM), programmable read only memory (PROM), EPROM (ErasablePROM, EPROM) or EEPROM (Electrically EPROM) or flash memory.

[0176] Temporary readable media refers to various types of RAM, including static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous dynamic random access memory (Synclink DRAM, SLDRAM) and direct random access memory (DirectRambus RAM, DRRAM).

[0177] This embodiment and the drawings attached to this specification clearly illustrate part of the technical ideas contained in the technology described above, and it is obvious that within the scope of the technical ideas contained in the description and drawings of the technology described above, the modified examples and specific embodiments that can be easily inferred by ordinary technicians in the technical field to which the present invention belongs are all included in the scope of protection of the invention claimed for the technology described above.

Claims

1. A method for evaluating the efficiency of candidate shRNA for RNA interference, wherein: include: The step of receiving the evaluation device with 'sequence information of candidate guide RNA for RNA interference' and 'information on the amount of pri-shRNA remaining after being cleaved by Drosha'; The evaluation device inputs the received 'guide RNA sequence information' and 'pri-shRNA amount information remaining after cleavage' into an evaluation model for evaluating the efficiency of shRNA for RNA interference; and The step of the evaluation means evaluating the efficiency of the shRNA for RNA interference based on the value output by the evaluation model.

2. The method for evaluating the efficiency of candidate shRNA for RNA interference according to claim 1, wherein: The evaluation model is a learning model that is learned using learning data to evaluate the efficiency of shRNA for RNA interference. The learning data includes 'guide RNA sequence information for RNA interference' and 'pri-shRNA amount information remaining after being cleaved by Drosha'.

3. The method for evaluating the efficiency of candidate shRNA for RNA interference according to claim 1, wherein: The information of the pri-shRNA remaining after being cut by Drosha in the cell is the information of the output value from a separate prediction model. The prediction model is a model that receives 'sequence information of Drosha cleavage sites in pri-shRNA' as input data and predicts how much of the pri-shRNA remains after being cleaved by Drosha.

4. The method for evaluating the efficiency of candidate shRNA for RNA interference according to claim 3, wherein: The cleavage site sequence information input into the prediction model includes: In the pri-shRNA, the information of the sequence from 27 nt (nucleotide) upstream to 29 nt downstream based on the Drosha cleavage site of 5p (5p Drosha cleavage site sequence) and the sequence from 31 nt upstream to 25 nt downstream based on the Drosha cleavage site of 3p (3p Drosha cleavage site sequence) are connected; and The 3p Drosha cleavage site sequence was flipped and ligated to the information of the 5p cleavage site sequence.

5. The method for evaluating the efficiency of candidate shRNA for RNA interference according to claim 3, wherein: The prediction model is a learning model that uses learning data to learn to predict the amount of pri-shRNA remaining after being cleaved by Drosha. The learning data includes 'sequence information of Drosha cleavage sites within pri-shRNA'.

6. The method for evaluating the efficiency of candidate shRNA for RNA interference according to claim 1, wherein: The step of accepting input includes the step of additionally receiving input of 'flanking sequence information', The step of inputting the 'flanking sequence information' into the evaluation model includes the step of additionally inputting the 'flanking sequence information' into the evaluation model, The flanking sequence information includes upstream and downstream sequence information of the guide RNA sequence information.

7. The method for evaluating the efficiency of candidate shRNA for RNA interference according to claim 6, wherein: The flanking sequence information includes the upstream 14nt sequence information and the downstream 14nt sequence information of the guide RNA sequence information.

8. A device for evaluating the efficiency of candidate shRNA for RNA interference, wherein: include: An input device receives input of candidate guide RNA sequence information for RNA interference and information on the amount of pri-shRNA remaining after being cleaved by Drosha; a storage device for storing an evaluation model for evaluating the efficiency of shRNA for RNA interference using the guide RNA sequence information received by the input device and the information on the amount of pri-shRNA remaining after cleavage; and The computing device inputs the guide RNA sequence information and the pri-shRNA amount information remaining after cutting received by the input device into the evaluation model stored in the storage device, and then evaluates the efficiency of the shRNA used for RNA interference according to the value output by the evaluation model.

9. The candidate shRNA efficiency evaluation device for RNA interference according to claim 8, wherein: The evaluation model is a learning model that is learned using learning data to evaluate the efficiency of shRNA for RNA interference, The learning data includes 'guide RNA sequence information for RNA interference' and 'pri-shRNA amount information remaining after being cleaved by Drosha'.

10. The candidate shRNA efficiency evaluation device for RNA interference according to claim 8, wherein: The information of the pri-shRNA remaining after being cut by Drosha in the cell is the information of the output value from a separate prediction model. The prediction model is a model that receives the sequence information of the Drosha cleavage site in the pri-shRNA as input data and predicts how much of the pri-shRNA will remain after being cleaved by Drosha.

11. The device for evaluating the efficiency of candidate shRNA for RNA interference according to claim 10, wherein: The cleavage site sequence information input into the prediction model includes: In the pri-shRNA, the sequence from 27 nt (nucleotide) upstream to 29 nt downstream (5p Drosha cleavage site sequence) based on the Drosha cleavage site of 5p is connected with the sequence from 31 nt upstream to 25 nt downstream (3p Drosha cleavage site sequence) based on the Drosha cleavage site of 3p; and The 3p Drosha cleavage site sequence was flipped and ligated to the information of the 5p cleavage site sequence.

12. The device for evaluating the efficiency of candidate shRNA for RNA interference according to claim 10, wherein: The prediction model is a learning model that uses learning data to learn to predict the amount of pri-shRNA remaining after being cleaved by Drosha. The learning data includes 'sequence information of Drosha cleavage sites within pri-shRNA'.

13. The candidate shRNA efficiency evaluation device for RNA interference according to claim 8, wherein: The input device also receives flanking sequence information, The evaluation model is a model for evaluating the efficiency of shRNA for RNA interference using the guide RNA sequence information, the information on the amount of pri-shRNA remaining after cleavage, and the flanking sequence information. After further inputting the 'flanking sequence information' into the evaluation model, the computing device evaluates the efficiency of the shRNA for RNA interference according to the output value of the evaluation model, The flanking sequence information includes upstream and downstream sequence information of the guide RNA sequence information.

14. The device for evaluating the efficiency of candidate shRNA for RNA interference according to claim 13, wherein: The flanking sequence information includes the upstream 14nt sequence information and the downstream 14nt sequence information of the guide RNA sequence information.

15. A computer-readable storage medium, wherein: The storage medium stores a program for executing the method for evaluating the efficiency of candidate shRNA for RNA interference according to claim 1 in a computer.

Images