Method for predicting translation efficiency of messenger RNA vaccine and analysis device

Abstract

A method for predicting translation efficiency of an mRNA vaccine comprises the steps in which: an analysis device receives sequence information of a target mRNA vaccine; the analysis device extracts, from the sequence information, an input sequence including a 25 nt sequence to which a primer binds in a 5' untranslated region (UTR), a 50 nt sequence of the 5' UTR immediately before a coding sequence (CDS) site, and a 30 nt sequence after a start codon of a coding region; the analysis device calculates secondary structure information for the input sequence; and the analysis device inputs the input sequence and the secondary structure information into a pre-trained deep learning model so as to predict the translation efficiency of the corresponding mRNA vaccine.

Description

Method and analysis device for predicting the translation efficiency of messenger RNA vaccines

[0001] The technology described below relates to a method for predicting translation efficiency for mRNA (messenger RiboNucleic Acid) vaccines.

[0002] Traditionally, a vaccine is a biological product containing an antigen, either the pathogen itself, a portion of its components, or a toxin, which has been appropriately processed. Following COVID-19, mRNA vaccines, a next-generation vaccine, have been commercialized. mRNA vaccines consist of mRNA containing the antigen sequence and a lipid nanoparticle (LNP) carrier. The LNPs of mRNA vaccines, when injected into the body, deliver the mRNA containing the antigen sequence into cells, where they then produce the antigen protein. For an mRNA vaccine to elicit an immune response, sufficient antigen must be produced. Therefore, it is crucial for mRNA vaccines to exhibit high stability and translation efficiency after administration.

[0003] Conventional techniques predict translation efficiency based on the 5' UTR (Untranslated Region) sequence. Meanwhile, it is known that the secondary structure formed by mRNA is involved in translation efficiency. The technology described below aims to provide a method for predicting in vivo translation efficiency by reflecting various regions and secondary structures of mRNA.

[0004] In one aspect, a method for predicting the translation efficiency of an mRNA vaccine includes a step in which an analysis device receives sequence information of a target mRNA vaccine, a step in which the analysis device calculates some portions of a 5' UTR (Untranslated Region) among the sequence information and secondary structure information formed by the some portions, and a step in which the analysis device inputs the some portions and the secondary structure information into a pre-trained deep learning model to predict the translation efficiency of the target mRNA vaccine.

[0005] In another aspect, a method for predicting the translation efficiency of an mRNA vaccine includes a step of an analysis device receiving sequence information of a target mRNA vaccine, a step of the analysis device extracting an input sequence including a 25 nt sequence to which a primer binds in a 5' UTR (Untranslated Region) of the sequence information, a 50 nt sequence of the 5' UTR immediately before a CDS (Coding sequence) site, and a 30 nt sequence after the start codon of the coding region, a step of the analysis device calculating secondary structure information for the input sequence, and a step of the analysis device inputting the input sequence and the secondary structure information into a pre-trained deep learning model to predict the translation efficiency of the target mRNA vaccine.

[0006] An analysis device for predicting the translation efficiency of an mRNA vaccine includes an interface device for receiving sequence information of a target mRNA vaccine, a storage device for storing a pre-trained deep learning model, and a calculation device for predicting the translation efficiency of the target mRNA vaccine by inputting some portions of the 5' UTR (Untranslated Region) of the sequence information and secondary structure information formed by the some portions into the deep learning model.

[0007] The technology described below accurately predicts translation efficiency by further utilizing sequence information, including 5' UTR sequences, and sequence-to-sequence interaction information. This technology contributes to the design of high-performance mRNA.

[0008] Figure 1 is an example of mRNA structure.

[0009] Figure 2 is an example of a system for predicting the translation efficiency of an mRNA vaccine.

[0010] Figure 3 shows the relationship between the interaction between the 5' UTR and CDS sequences of mRNA and translation efficiency.

[0011] Figure 4 is an example illustrating the input data and encoding process of a deep learning model.

[0012] Figure 5 is an example of the process of building a deep learning model for predicting the translation efficiency of a target mRNA vaccine.

[0013] Figure 6 is an example of a deep learning model that predicts the translation efficiency of an mRNA vaccine.

[0014] Figure 7 shows the results of evaluating the translation efficiency prediction performance of the constructed deep learning model.

[0015] Figure 8 shows the results of comparing the performance of deep learning models using various input data.

[0016] Figure 9 is an example of an analysis device for predicting the translation efficiency of an mRNA vaccine.

[0017] The technology described below is susceptible to various modifications and embodiments. Specific embodiments are illustrated and described in detail in the drawings. However, this does not limit the technology described below to specific embodiments, and it should be understood that all modifications, equivalents, and alternatives fall within the spirit and scope of the technology described below.

[0018] Terms such as first, second, A, and B may be used to describe various components, but these components are not limited by these terms and are used solely to distinguish one component from another. For example, without departing from the scope of the technology described below, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. The term "and / or" includes a combination of a plurality of related described items or any of a plurality of related described items.

[0019] As used herein, the singular expressions should be understood to include the plural expressions unless the context clearly dictates otherwise, and the term "comprises" and the like should be understood to mean the presence of a described feature, number, step, operation, component, part, or combination thereof, but not to exclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0020] Before going into a detailed description of the drawings, it should be made clear that the division of components in this specification is merely a division based on the main function of each component. In other words, two or more components described below may be combined into a single component, or a single component may be further subdivided into two or more components with more detailed functions. In addition to its own main function, each component described below may additionally perform some or all of the functions of other components, and of course, some of the main functions of each component may be exclusively performed by other components.

[0021] Additionally, in performing a method or method of operation, each process constituting the method may occur in a different order than the stated order, unless the context clearly indicates a specific order. That is, each process may occur in the same order as the stated order, may be performed substantially simultaneously, or may be performed in the opposite order.

[0022] The technology described below is a technique for predicting the translation efficiency of an mRNA vaccine composed of a specific sequence.

[0023] The following analysis device analyzes the sequence information of the mRNA vaccine being analyzed to predict its in vivo translation efficiency. The mRNA vaccine being analyzed is referred to as the "target mRNA vaccine." The target mRNA vaccine is designed with a specific sequence. The analysis device can be implemented in various forms, such as a computer, PC, smart device, or a server on a network.

[0024] The analysis device uses a learning model to predict the translation efficiency of the target mRNA vaccine.

[0025] A learning model refers to a machine learning model. A learning model encompasses various types of models. Examples include decision trees, random forests, XGBoost (eXtreme Gradient Boosting), LightGBM (Light Gradient Boosting Machine), CatBoost (Categorical Boosting), KNN (K-nearest neighbor), Naive Bayes, SVM (support vector machine), and artificial neural networks.

[0026] Deep neural networks (DNNs) are a representative example of artificial neural networks. DNNs are artificial neural network models comprised of multiple hidden layers between the input and output layers. Like typical artificial neural networks, DNNs can model complex non-linear relationships. Various types of DNN models have been studied, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), restricted Boltzmann machines (RBMs), deep belief networks (DBNs), generative adversarial networks (GANs), and relational networks (RNLs). Below, we focus on DNNs as a model for predicting the translation efficiency of target mRNA vaccines.

[0027] Figure 1 is an example of the mRNA structure. mRNA consists of a cap (CAP), 5' UTR, CDS (coding sequence), 3' UTR, and poly-A tail, depending on the sequence order. The cap is located at the 5' end and helps with protein production and protects the mRNA from destruction. The UTR is an untranslated region, with the 5' UTR and 3' UTR separated by the CDS. The poly-A tail is located at the 3' end and helps with protein production and maintains stability. The CDS is a sequence used for protein translation and consists of exons, including a start codon and a stop codon. The 5' UTR is known to have various translational control mechanisms. Therefore, previous mRNA vaccine research has primarily focused on optimizing the 5' UTR sequence.

[0028] Figure 2 is an example of a system (100) for predicting the replication efficiency of an mRNA vaccine. In Figure 1, the analysis device is illustrated as an example of a server (130) and a computer terminal (140, 150).

[0029] The mRNA design device (110) designs a specific sequence of a target mRNA vaccine. For example, a researcher can input sequence information for the mRNA vaccine into the mRNA design device (110). The mRNA design device (110) can store the sequence information for the mRNA vaccine in a database (DB, 120).

[0030] The server (130) can receive sequence information of the target mRNA from the mRNA design device (110) or DB (120). The server (130) can predict translation efficiency based on the sequence information of the target mRNA. (i) The server (130) can extract specific sequence information from the sequence information of the target mRNA. In addition, the server (130) can predict secondary structure using at least a part of the extracted sequence information. The secondary structure may be a structure formed by a region other than the 5' UTR. (ii) The server (130) can consistently encode the extracted sequence information and secondary structure information and input them into a pre-trained deep learning model. (iii) The server (130) can predict translation efficiency of the target mRNA vaccine based on the value output by the deep learning model. The input data and model construction process used by the deep learning model will be described later. A user (10) can access the server (130) through a user terminal (PC, smartphone, etc.) and check the analysis results performed by the server (130).

[0031] The computer terminal (140) can receive sequence information of the target mRNA from the mRNA design device (110) or DB (120). The computer terminal (140) can predict translation efficiency based on the sequence information of the target mRNA. (i) The computer terminal (140) can extract specific sequence information from the sequence information of the target mRNA. In addition, the computer terminal (140) can predict secondary structure using at least a part of the extracted sequence information. The secondary structure may be a structure formed by a region other than the 5' UTR. (ii) The computer terminal (140) can consistently encode the extracted sequence information and secondary structure information and input them into a pre-trained deep learning model. (iii) The computer terminal (140) can predict translation efficiency of the target mRNA vaccine based on the value output by the deep learning model. The input data and model construction process used by the deep learning model will be described later. The user (20) can check the analysis results through the computer terminal (140) that he or she uses.

[0032] The computer terminal (150) is an analysis device and an mRNA design device at the same time. The computer terminal (150) receives the sequence of the target mRNA vaccine from the user (30). The computer terminal (150) can predict the translation efficiency based on the sequence information of the target mRNA. (i) The computer terminal (150) can extract specific sequence information from the sequence information of the target mRNA. In addition, the computer terminal (150) can predict the secondary structure using at least a part of the extracted sequence information. The secondary structure may be a structure formed by a region other than the 5' UTR. (ii) The computer terminal (150) can consistently encode the extracted sequence information and secondary structure information and input them into a pre-trained deep learning model. (iii) The computer terminal (150) can predict the translation efficiency of the target mRNA vaccine based on the value output by the deep learning model. The input data and model construction process used by the deep learning model will be described later. The user (30) can check the analysis results through the computer terminal (150) he or she uses.

[0033] The researchers built the aforementioned deep learning model using a dataset previously used in research on models predicting mRNA translation efficiency. The previous model is Moderna's Optimus 5-Prime (see Paul J. Sample et al., Human 5' UTR design and variant effect prediction from a massively parallel translation assay, Nature Biotechnology volume 37, pp. 803-809, 2019). Optimus 5-Prime predicts the translation efficiency of a given mRNA based on the 5' UTR sequence. Optimus 5-Prime is a model created using randomly generated 50-nt 5' UTR sequences and the Mean Ribosome Load (MRL) scores of the sequences as training data. The researchers built the model using 280,000 data points previously used in the development of Optimus 5-Prime. The researchers used the top 20,000 lead counts out of 280,000 data as validation data and 260,000 data as training data.

[0034] The deep learning model developed by the researchers predicts translation efficiency using information significant for translation efficiency prediction in addition to 5' UTR sequence information. Specifically, the deep learning model utilizes the secondary structure formed by the 5' UTR (or a portion of the 5' UTR sequence) and a portion of the CDS to predict mRNA translation efficiency.

[0035] Figure 3 shows the relationship between the interactions between the 5' UTR and CDS sequences of mRNA and translation efficiency. Figure 3 is the result of analyzing data used in Optimus 5-Prime. Figure 3 depicts the relationship between the number of base pairs between the 5' UTR of mRNA and the 30 nt region following the start codon in the CDS and the degree of translation efficiency (MRL score). Looking at Figure 3, we can see that there is a certain correlation between the binding relationship between the 5' UTR and the 30 nt region following the start codon in the CDS and the translation efficiency. Considering this, the deep learning model uses the secondary structure information formed between the 5' UTR of mRNA and the 30 nt region following the start codon in the CDS as input data.

[0036] Deep learning models can predict the translation efficiency of mRNA vaccines using input data such as Table 1 below.

[0037] Input data type Region Content Sequence information 5' UTR Fixed 25 nt 5' UTR Random 50 nt CDS 30 nt after start codon Secondary structure information 5' UTR and CDS Binding relationship between (i) 25 nt + 50 nt of 5' UTR and (ii) 30 nt of CDS

[0038] The input data includes at least one of the following: ① a fixed 25 nt sequence of the 5' UTR (primer binding site), ② the last 50 nt of the 5' UTR (50 nt immediately before the CDS), ③ 30 nt after the start codon of the CDS, and ④ secondary structure information formed by the 5' UTR and CDS regions. The secondary structure may be a secondary structure formed by the 5' UTR (①+②) and the CDS (③). Furthermore, the secondary structure may be a structure formed by a portion of the 5' UTR sequence and 30 nt of the CDS.

[0039] Figure 4 is an example illustrating the input data and encoding process of a deep learning model. (1) The sequence information includes a fixed 25nt sequence (primer binding site) of the 5' UTR, an arbitrary 50nt of the 5' UTR, and 30nt after the start codon of the CDS. Therefore, the sequence information can be a total of 105nt sequence. (2) The secondary structure information includes binding information formed by the sequence of the 5' UTR and the 30nt of the CDS. The secondary structure information can be predicted using a commercial program. For example, the analysis device can use a program such as RNAfold to generate secondary structure information for the input sequence (sequence information of Figure 4). The secondary structure information includes the sequence positions that bind to each other in the entire sequence. The corresponding parentheses (" and ")" in the secondary structure information indicate bases that bind to each other. The analysis device consistently encodes the input data and inputs it to the deep learning model. The bottom of Figure 4 shows an example of one-hot vector encoding for the input data. Therefore, the deep learning model can receive a matrix of size 105 × 7 as input.

[0040] Figure 5 illustrates an example of the process (200) for building a deep learning model for predicting the translation efficiency of a target mRNA vaccine. The deep learning model is a model that predicts the translation efficiency of an mRNA vaccine. The learning process is described as being performed by a learning device. The learning device refers to a computer device capable of performing data processing and learning model training.

[0041] The DB may include a data set consisting of full sequence information of an mRNA and translation efficiency information of the mRNA (e.g., MRL score).

[0042] The learning device can extract data necessary for model building or verification from the database (210). The deep learning model uses the 5' UTR of mRNA as input data, as well as various regions' sequences and secondary structures formed by some of the sequences. Therefore, the learning device can extract a fixed 25 nt from the 5' UTR, a random 50 nt from the 5' UTR, 30 nt after the start codon in the CDS, and the MRL score (= label value) from the data in the database.

[0043] The learning device can divide the entire collected data into learning data and verification data (220).

[0044] The learning device performs learning of a deep learning model using learning data (220).

[0045] The learning device extracts a single input data from the learning data and performs the learning process. The learning device can predict the secondary structure information formed in the body by the 5' UTR sequence (25 nt + 50 nt) and the CDS sequence (30 nt). The learning device can predict the secondary structure of the input sequence using a program such as RNAfold. As shown in Figure 4, the learning device can encode mRNA sequence information and secondary structure information into a one-hot vector. The learning device can input the encoded input data into a deep learning model to predict translation efficiency.

[0046] The learning device compares the predicted values ​​output by the deep learning model with the label value (e.g., the MRL score for the sequence) and updates the model's parameters so that the deep learning model outputs values ​​closer to the correct answer. The learning device repeats the learning process using data from the training data.

[0047] Once the learning process is complete, the learning device can perform verification on the trained deep learning model (230). The learning device extracts one of the verification data sets and performs the verification process. The learning device inputs the selected data into the deep learning model to predict translation efficiency, and performs verification by comparing the predicted value with the correct value.

[0048] Figure 6 is an example of a deep learning model that predicts the translation efficiency of an mRNA vaccine. The deep learning model may include a feature extraction layer that extracts features from encoded input data and a classification layer that receives the feature map output from the feature extraction layer and predicts the translation efficiency. The feature extraction layer may include multiple convolutional layers and a max pooling layer. The classification layer may include an average pooling layer, a dense layer, and a final output layer. The output layer may output a final prediction value using an activation function. In addition, the deep learning model may include batch normalization layers, as shown in Figure 6.

[0049] The researchers evaluated the prediction accuracy of the trained model using 20,000 5' UTR sequences.

[0050] Figure 7 shows the results of evaluating the translation efficiency prediction performance of the constructed deep learning model. Figure 7 shows the translation efficiency prediction performance of the conventional model, Optimus 5-prime (denoted as Optimus) and the constructed deep learning model (denoted as Ours). Translation efficiency performance was evaluated using the coefficient of determination (r square). The Optimus 5-prime model had an R square value of 0.93, while the constructed model (Ours) showed an R square value of 0.943.

[0051] Figure 8 shows the results of comparing the performance of deep learning models using various input data. The performance of translation efficiency is measured by the coefficient of determination (r 2) were evaluated. The researchers evaluated the performance of (i) Model 1, which used only a random 50 nt of the 5' UTR as input data, (ii) Model 2, which used a random 50 nt of the 5' UTR and the secondary structure information formed by the 50 nt region as input data, (iii) Model 3, which used a random 50 nt of the 5' UTR and the secondary structure information formed by the 105 nt region of the 5' UTR and CDS as input data, and (iv) Model 4, which used a 105 nt region of the 5' UTR and CDS and the secondary structure information formed by the 105 nt region as input data. As shown in Fig. 8, Model 3 showed higher performance than Model 1 or Model 2, and Model 4 showed the highest performance. In summary, the deep learning model that uses the secondary structure information together with the sequence of the 5' UTR showed higher performance than the model that did not use the secondary structure. Additionally, deep learning models that use additional sequences, such as fixed 25 nt and CDS 30 nt, in addition to the 50 nt of the 5' UTR, outperformed models that use only 50 nt.

[0052] Figure 9 illustrates an example of an analysis device (300) for predicting the translation efficiency of an mRNA vaccine. The analysis device (300) corresponds to the aforementioned analysis devices (130, 140, 150 of Figure 2). The analysis device (300) can be physically implemented in various forms. For example, the analysis device (300) can take the form of a computer device such as a PC, a network server, a data processing chipset, or the like.

[0053] The analysis device (300) may include a storage device (310), a memory (320), a computing device (330), an interface device (340), a communication device (350), and an output device (360).

[0054] The storage device (310) can store a deep learning model that predicts the translation efficiency of an mRNA vaccine. In this case, the deep learning model is a pre-trained model.

[0055] The storage device (310) can store a program or code that controls the process of predicting the translation efficiency of the target mRNA vaccine.

[0056] The storage device (310) can store sequence information of the target mRNA vaccine. At this time, the sequence information may be the entire sequence of the target mRNA.

[0057] The storage device (310) can store a program that receives an RNA sequence and predicts the secondary structure formed by the sequence.

[0058] The storage device (310) can store the analysis results.

[0059] The memory (320) can store data and information generated during the process of the analysis device (300) predicting the translation efficiency of the target mRNA vaccine.

[0060] The interface device (340) is a device that receives certain commands and data from the outside.

[0061] The interface device (340) can receive sequence information of the target mRNA vaccine from a physically connected input device or an external storage device.

[0062] The interface device (340) can also transmit the analysis results to an external object.

[0063] A communication device (350) refers to a configuration that receives and transmits certain information through a wired or wireless network.

[0064] The communication device (350) can receive sequence information of the target mRNA vaccine from an external object.

[0065] The communication device (350) may also transmit the analysis results to an external object such as a user terminal.

[0066] The interface device (340) also includes a configuration that transmits data received by the communication device (350) to the inside of the analysis device (300).

[0067] An output device (360) is a device that outputs certain information. The output device (360) can output interfaces, analysis results, etc. required for the data processing process.

[0068] The computing device (330) can extract input data from the sequence information of the target mRNA vaccine. The input data can include at least one of ① a fixed 25 nt sequence of the 5' UTR (primer binding site), ② the last 50 nt of the 5' UTR (50 nt immediately before the CDS), and ③ 30 nt after the start codon of the CDS.

[0069] The computing device (330) can predict secondary structure information composed of a sequence extracted as input data. The computing device (330) can predict secondary structure information by inputting input data into a secondary structure prediction program. For example, the computing device (330) can predict secondary structure information of 105 nt composed of a fixed 25 nt sequence of the 5' UTR, an arbitrary 50 nt of the 5' UTR, and 30 nt after the start codon of the CDS.

[0070] The computational device (330) performs one-hot vector encoding on the sequence and secondary structure information, which are input data, as shown in FIG. 4. Through this, the computational device (330) can produce a vector matrix for the input data.

[0071] The computational unit (330) inputs encoded input data (or vector matrix) into a trained deep learning model to produce a predicted value for translation efficiency. This allows the computational unit (330) to predict the translation efficiency of a target mRNA vaccine. The computational unit (330) can also predict the effectiveness of the current target mRNA vaccine.

[0072] The computing device (330) may be a device such as a processor, AP, or chip embedded with a program that processes data and performs certain operations.

[0073] Additionally, the method for predicting the translation efficiency of an mRNA vaccine or the method for designing an mRNA vaccine, as described above, may be implemented as a program (or application) comprising an executable algorithm that can be executed on a computer. The program may be stored and provided on a temporary or non-transitory computer-readable medium.

[0074] A non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, the various applications or programs described above may be stored and provided on a non-transitory readable medium, such as a CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM (read-only memory), PROM (programmable read only memory), EPROM (Erasable PROM, EPROM), EEPROM (Electrically EPROM), or flash memory.

[0075] Temporarily readable media refers to various types of RAM, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchronous DRAM (Synclink DRAM, SLDRAM), and Direct Rambus RAM (DRRAM).

[0076] The present embodiment and the drawings attached to the present specification only clearly illustrate a part of the technical idea included in the above-described technology, and it is obvious that all modified examples and specific embodiments that can be easily inferred by a person skilled in the art within the scope of the technical idea included in the specification and drawings of the above-described technology are included in the scope of the rights of the above-described technology.

Claims

1. A step in which the analysis device receives sequence information of the target mRNA vaccine; A step in which the above analysis device calculates some portions of the 5' UTR (Untranslated Region) among the above sequence information and secondary structure information formed by the some portions; and A method for predicting the translation efficiency of an mRNA vaccine, comprising a step of inputting the partial portion and the secondary structure information into a pre-learned deep learning model by the above analysis device to predict the translation efficiency of the target mRNA vaccine.

2. In paragraph 1, A method for predicting the translation efficiency of an mRNA vaccine comprising a 25 nt sequence to which a primer binds and a 50 nt sequence of the 5' UTR immediately before the CDS (coding sequence) region.

3. In paragraph 1, The above analysis device is a method for predicting the translation efficiency of an mRNA vaccine by further inputting a 30 nt sequence after the start codon of the coding region into the deep learning model.

4. In paragraph 1, A method for predicting the translation efficiency of an mRNA vaccine, wherein the secondary structure information is structural information composed of a 25 nt sequence to which a primer binds in the 5' UTR, an arbitrary 50 nt sequence selected from the 5' UTR, and a 30 nt sequence after the start codon of the coding region.

5. In paragraph 1, The above analysis device is a method for predicting the translation efficiency of an mRNA vaccine by encoding the above-mentioned partial portion and the above-mentioned secondary structure information into a one-hot vector and inputting it into the above-mentioned deep learning model.

6. Step in which the analysis device receives sequence information of the target mRNA vaccine; A step in which the above analysis device extracts an input sequence including a 25 nt sequence to which a primer binds in the 5' UTR (Untranslated Region) of the above sequence information, a 50 nt sequence of the 5' UTR immediately before the CDS (Coding sequence) site, and a 30 nt sequence after the start codon of the coding region; a step in which the above analysis device produces secondary structure information for the input sequence; and A method for predicting the translation efficiency of an mRNA vaccine, comprising a step of predicting the translation efficiency of the target mRNA vaccine by inputting the input sequence and the secondary structure information into a pre-learned deep learning model by the above analysis device.

7. Interface device for receiving sequence information of target mRNA vaccine; A storage device for storing a pre-trained deep learning model; and An analysis device for predicting the translation efficiency of an mRNA vaccine, including a computational device that inputs some portions of the 5' UTR (Untranslated Region) of the above sequence information and secondary structure information formed by the some portions into the deep learning model to predict the translation efficiency of the target mRNA vaccine.

8. In paragraph 7, An analysis device for predicting the translation efficiency of an mRNA vaccine, wherein the above-mentioned portion includes a 25-nt sequence to which a primer binds and a 50-nt sequence of the 5' UTR immediately before the CDS (coding sequence) portion.

9. In paragraph 7, The above-mentioned calculation device is an analysis device that predicts the translation efficiency of an mRNA vaccine by inputting a 30 nt sequence after the start codon of the coding region into the above-mentioned deep learning model.

10. In paragraph 7, The above secondary structure information is an analysis device for predicting the translation efficiency of an mRNA vaccine, which is structural information composed of a 25 nt sequence to which a primer binds in the 5' UTR, an arbitrary 50 nt sequence selected from the 5' UTR, and a 30 nt sequence after the start codon of the coding region.

Images