Translation efficiency prediction method, screening method, model training method, and electronic device

Abstract

The invention relates to the technical field of bioinformatics, and relates in particular to a translation efficiency prediction method, a screening method, a model training method, and an electronic device. The translation efficiency prediction method is implemented on the basis of a translation efficiency prediction model, and the translation efficiency prediction model comprises an embedding module, an encoding module, and a prediction module. The method comprises: acquiring sequence data of a messenger ribonucleic acid (RNA); extracting features of the sequence data via the embedding module, so as to obtain a feature representation of the messenger RNA; inputting the feature representation into the encoding module and performing convolution processing on the input feature representation via the encoding module, encoding a convolution result on the basis of an attention mechanism, and outputting an updated feature representation on the basis of an encoding result; and performing translation efficiency prediction on the updated feature representation via the prediction module, so as to obtain a translation efficiency prediction result.

Description

Translation efficiency prediction method, screening method, model training method and electronic device

[0001] Related applications

[0002] This application claims priority to the Chinese invention patent application with application number 202410378367.0 filed on March 28, 2024, and cites the entire contents disclosed in the above patent application as part of this application. Technical Field

[0003] The present application relates to the field of bioinformatics, and in particular to a translation efficiency prediction method, a screening method, a model training method, and an electronic device. Background Art

[0004] Messenger RNA (mRNA) is a type of RNA that carries genetic information and serves as a template for protein synthesis. mRNA molecules are read by ribosomes within cells and, based on the base sequence they carry, guide the order in which amino acids are linked together to form proteins with specific functions.

[0005] The translation efficiency of mRNA (mRNA) refers to the rate and quality of translation of mRNA molecules into proteins by ribosomes within cells per unit time. Measuring mRNA translation efficiency is essential for drug development related to mRNA. Accurately measuring mRNA translation efficiency has a positive impact on drug development related to mRNA. In existing technologies, mRNA translation efficiency is generally measured through biological experiments. However, measuring mRNA translation efficiency through biological experiments is time-consuming and costly. Summary of the Invention

[0006] The embodiments of the present application provide a translation efficiency prediction method, a screening method, a model training method, and an electronic device, which use deep learning to predict the translation efficiency of messenger RNA, reduce the time consumption of translation efficiency measurement, and reduce costs.

[0007] The present application provides a method for predicting the translation efficiency of messenger RNA (mRNA), which is based on a translation efficiency prediction model. The translation efficiency prediction model includes an embedding module, an encoding module, and a prediction module. The method includes:

[0008] obtaining messenger RNA sequence data;

[0009] Extracting features of sequence data through the embedding module to obtain feature representation of messenger RNA;

[0010] Inputting the feature representation into the encoding module, performing convolution processing on the input feature representation through the encoding module, encoding the convolution result based on the attention mechanism, and outputting an updated feature representation according to the encoding result;

[0011] The prediction module is used to predict the translation efficiency of the updated feature representation to obtain a translation efficiency prediction result.

[0012] The present application provides a method for screening messenger RNA, which is implemented based on a translation efficiency prediction model, wherein the translation efficiency prediction model includes an embedding module, an encoding module, and a prediction module; the method includes:

[0013] obtaining sequence data of multiple messenger RNAs;

[0014] Extracting features of the sequence data through the embedding module to obtain a feature representation of the messenger RNA; inputting the feature representation into the encoding module to perform convolution processing on the input feature representation through the encoding module, encoding the convolution result based on the attention mechanism, and outputting an updated feature representation based on the encoding result; performing translation efficiency prediction on the updated feature representation through the prediction module to obtain a translation efficiency prediction result of the messenger RNA;

[0015] The plurality of messenger RNAs are screened according to the translation efficiency prediction results.

[0016] The present application provides a method for training a messenger RNA translation efficiency prediction model, wherein the translation efficiency prediction model includes an embedding module, an encoding module, and a prediction module. The method includes:

[0017] obtaining sequence data and tag data of messenger RNA, wherein the tag data represents translation efficiency of the messenger RNA;

[0018] Extracting features of the sequence data through the embedding module to obtain a feature representation of the messenger RNA; inputting the feature representation into the encoding module to perform convolution processing on the input feature representation through the encoding module, encoding the convolution result based on the attention mechanism, and outputting an updated feature representation based on the encoding result; performing translation efficiency prediction on the updated feature representation through the prediction module to obtain a translation efficiency prediction result of the messenger RNA;

[0019] The parameters of the translation efficiency prediction model are determined based on the translation efficiency prediction results and label data.

[0020] An embodiment of the present application further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above method when executing the computer program.

[0021] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0022] An embodiment of the present application further provides a computer program product, which includes a computer program, and implements the above method when the computer program is executed by a processor.

[0023] The technical solution of the embodiment of the present application can predict the translation efficiency of messenger RNA based on a translation efficiency prediction model. The translation efficiency prediction model includes an embedding module, an encoding module and a prediction module. The sequence data of the messenger RNA can be obtained; the features of the sequence data can be extracted by the embedding module to obtain a feature representation of the messenger RNA; the feature representation can be input into the encoding module, so that the input feature representation is convoluted by the encoding module, the convolution result is encoded based on the attention mechanism, and the updated feature representation is output according to the encoding result; the translation efficiency of the updated feature representation can be predicted by the prediction module to obtain a translation efficiency prediction result. In this way, a deep learning-based approach can accurately and high-throughput predict the translation efficiency of messenger RNA, reduce measurement time and reduce costs. In addition, the encoding module can use convolution and attention mechanisms to capture the dependency relationship between local features and global features in the sequence data. The encoding module's ability to understand the input data is improved, thereby improving the accuracy of translation efficiency prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. The drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0025] FIG1 is a schematic flow chart of a method for predicting translation efficiency in an embodiment of the present application;

[0026] FIG2 is a schematic diagram of the structure of a translation efficiency prediction model in an embodiment of the present application;

[0027] FIG3 is a flow chart of a method for training a translation efficiency prediction model in an embodiment of the present application;

[0028] FIG4 is a schematic flow chart of the method for screening messenger RNA in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The specific embodiments described herein are only used to explain the present disclosure, rather than to limit the present disclosure. Based on the described embodiments of the present disclosure, all other embodiments obtained by those of ordinary skill in the art fall within the scope of protection of the present disclosure. In addition, relational terms such as "first" and "second" are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0030] Messenger RNA (mRNA) therapy has emerged as a highly promising platform for disease treatment, enabling modular assembly of sequence elements and, in theory, the delivery and translation of any protein. Compared to recombinant proteins expressed in mammalian cell lines, in vitro transcription of mRNA is faster and more flexible, enabling faster responses to public health emergencies. For example, mRNA-based vaccines demonstrate the strong clinical and commercial value of mRNA therapies.

[0031] The untranslated region (UTR) is a segment of an mRNA molecule located upstream of the start codon and downstream of the stop codon. It is a sequence that is not translated. Untranslated regions can be transcribed along with the coding region and are present in mature mRNA. The UTR upstream of the start codon of an mRNA is called the 5' untranslated region (5'UTR). The 5' UTR is a key factor influencing the efficiency of mRNA translation. Existing methods for determining the translation efficiency of the 5' UTR typically rely on experimental measurements, which have drawbacks such as time-consuming, low-throughput, and high cost. Deep learning is a data-driven modeling approach that has demonstrated excellent performance in many fields, including image recognition, natural language processing, and computational biology. Convolutional neural networks excel at motif recognition, offering better performance than previous non-deep learning methods. They can also capture dependencies between data, which is crucial for tasks involving nucleic acid sequences. Convolutional neural networks can effectively predict the function, origin, and properties of nucleic acid sequences. However, methods such as convolutional neural networks, which rely on sequential computation, are difficult to parallelize and suffer from the vanishing gradient problem.

[0032] The present invention provides a method for predicting the translation efficiency of messenger RNA (mRNA). The method can accurately predict the translation efficiency of mRNA by leveraging the local feature extraction capabilities of convolution combined with the global feature extraction capabilities of the self-attention mechanism. This reduces the time required to measure translation efficiency and lowers costs. The method can be applied to processors in computer devices such as laptops, desktops, and servers. The method can be implemented by running a translation efficiency prediction model on the processor. The translation efficiency prediction model can include an embedding module, an encoding module, and a prediction module.

[0033] Please refer to Figure 1. The method may include the following steps.

[0034] Step 11: Obtain messenger RNA sequence data.

[0035] Step 12: Extract the features of the sequence data through the embedding module to obtain the feature representation of messenger RNA.

[0036] Step 13: Input the feature representation into the encoding module to perform convolution processing on the input feature representation through the encoding module, encode the convolution result based on the attention mechanism, and output the updated feature representation according to the encoding result.

[0037] Step 14: Using the prediction module, the translation efficiency of the updated feature representation is predicted to obtain the translation efficiency prediction result of the messenger RNA.

[0038] The embodiment of the present application can implement a method for predicting the translation efficiency of messenger RNA based on a processor to achieve accurate and high-throughput prediction of the translation efficiency of messenger RNA, thereby improving screening efficiency and reducing screening costs.

[0039] In some embodiments, sequence data may include sequence data consisting of nucleotides on a messenger RNA molecule arranged in a specific order. Nucleotides may include ribonucleotides, such as adenine ribonucleotides, uracil ribonucleotides, cytosine ribonucleotides, guanine ribonucleotides, etc. Nucleotides may include ribose, nitrogenous bases, and phosphate groups, etc. In some scenario examples, sequence data may include sequence data of untranslated regions, such as 5'UTR sequence data. Specifically, the sequence data of messenger RNA can be directly obtained, for example, sequence data input by a user or sent by other devices can be received. Alternatively, a data file of messenger RNA can be obtained, for example, a data file input by a user or sent by other devices can be received. Thus, the data file can be parsed to obtain the sequence data of messenger RNA. The data file may include fasta files, fastq files, etc. Furthermore, the sequence data may be obtained by measuring using a measurement device for measuring sequence data, the measurement device being communicatively connected to a server. When the measurement device obtains the sequence data of the messenger RNA, the measurement device transmits the sequence data of the messenger RNA to the server, so that the server can subsequently process the sequence data of the messenger RNA. It should be noted that the measurement device may be any device capable of measuring and obtaining sequence data, and this application is not limited thereto.

[0040] In some embodiments, the translation efficiency prediction model is used to predict the translation efficiency of messenger RNA, such as the translation efficiency of the 5' untranslated region. Please refer to Figure 2. The translation efficiency prediction model may include an embedding module, an encoding module and a prediction module. The embedding module is used to extract the features of the sequence data and obtain the feature representation of the messenger RNA. The encoding module is used to update the feature representation of the messenger RNA. The encoding module specifically includes N encoders, where N is a positive integer greater than or equal to 1. The N encoders can have the same or similar structure and are stacked sequentially in the encoding module so that the feature representation can be updated in sequence. Each encoder can perform convolution processing on the input feature representation, encode the convolution result based on the attention mechanism, and output the updated feature representation according to the encoding result. The prediction module is used to predict the translation efficiency of the updated feature representation to obtain the translation efficiency prediction result of the messenger RNA.

[0041] In some embodiments, the sequence data of the mRNA can be directly input into the embedding module to obtain a characteristic representation of the mRNA. Alternatively, the encoding of the nucleotides in the mRNA can be determined according to a set encoding method; the encoding of the sequence data can be generated based on the encoding of the nucleotides; thereby, the encoding of the sequence data can be input into the embedding module to obtain a characteristic representation of the mRNA. For example, if the mRNA may include multiple nucleotides, the encoding of each nucleotide can be determined separately according to the set encoding method. The set encoding method can include a one-hot encoding method. Specifically, for example, if the mRNA may include multiple nucleotides such as adenine ribonucleotides, uracil ribonucleotides, cytosine ribonucleotides, and guanine ribonucleotides, the encodings of adenine ribonucleotides, uracil ribonucleotides, cytosine ribonucleotides, and guanine ribonucleotides can be determined to be 1000, 0100, 0010, and 0001, respectively. Thus, the encoding of the sequence data can be generated based on the encoding of the nucleotides. The encoding of the sequence data can be a characteristic matrix. For example, messenger RNA may include four nucleotides: adenine ribonucleotide, uracil ribonucleotide, cytosine ribonucleotide, and guanine ribonucleotide. The length of the sequence data may be l, and the encoding of the sequence data may be a l×4 feature matrix. In some scenario examples, the embedding module includes an embedding layer and a linear layer. The encoding of the sequence data can be mapped to a vector through the embedding layer, providing a usable sequence representation for the model. The embedding layer can map the encoding of the sequence data to an initial latent vector of a fixed size. The linear layer can transform the dimension of the vector to obtain a feature representation. The linear layer can project the initial latent vector through the dimension to adapt to the input requirements of the encoding module. In this embodiment, the feature representation of messenger RNA may include a vector. For example, the encoding of the sequence data may be a l×4 feature matrix, and the feature representation of the messenger RNA may be a l×d latent vector, where d is the dimension size that meets the model requirements after projection by the linear layer.

[0042] In some embodiments, the feature representation of the mRNA can be input into an encoding module. The encoding module can perform convolution on the input feature representation, encode the convolution result based on an attention mechanism, and output an updated feature representation based on the encoding result. This can leverage the local feature extraction capability of convolution and combine it with the global feature extraction capability of the self-attention mechanism to accurately obtain the feature representation of the mRNA, thereby improving the ability to understand the input data.

[0043] In some embodiments, the encoding module may include N encoders stacked in sequence. Thus, the feature representation of the messenger RNA can be input into the first encoder of the N encoders, so that the feature representation can be updated in sequence through the N encoders. In this embodiment, each encoder can perform convolution processing on the input feature representation, can encode the convolution result based on the attention mechanism, and can output the updated feature representation based on the encoding result. In this way, each encoder can utilize the local feature extraction capability of convolution, combined with the global feature extraction capability of the self-attention mechanism, to accurately obtain the feature representation of the messenger RNA. The N encoders can be in a series form. Each encoder from the second encoder can receive the feature representation output by the previous encoder, and then output the updated feature representation after processing by the encoder. In this way, the feature representation is iteratively updated by the N encoders, thereby improving the accuracy of the translation efficiency prediction.

[0044] In some scenario examples, each encoder may include a convolutional layer, an attention layer, a position-wise feedforward network (FFN), and a deconvolutional layer. The input feature representation can be convolved by the convolutional layer to effectively extract local features of the sequence data using the convolution operation. For example, the formula Perform a one-dimensional convolution operation on the input feature representation. C i represents the convolution result of the i-th data element in the feature representation, k represents the size of the convolution kernel corresponding to the convolution layer, S i+k represents the i+kth data element in the feature representation, w k Represents the kth data element in the convolution kernel corresponding to the convolution layer. The convolution result can be encoded by the attention layer to effectively extract the global features of the sequence data using the attention mechanism. For example, the attention layer can include a multi-head self-attention layer. The self-attention mechanism enables each local feature to pay attention to all local features (including itself), so that global dependencies can be obtained between local features of any distance. The input of the multi-head self-attention layer can include queries, keys, values, etc., and their dimensions can be recorded as d k d k d v To facilitate calculations, queries, keys, and values ​​can be packaged into matrices, denoted as Q, K, and V. Calculations can then be performed using the attention function. The attention function can be expressed as where K T Represents the transpose of matrix K. The multi-head attention function is based on the attention function and is divided into h heads, each with a dimension of d model / h, and calculate the attention through the above attention function respectively. After the calculation is completed, the calculation results of each head are spliced ​​on the feature dimension and mapped to the corresponding dimension. The multi-head attention function can be expressed as MultiHead(Q,K,V)=Concat(head1,…,head h )W O In this embodiment, i ranges from 1 to h. Respectively represent mapping the matrices Q, K, and V of the i-th attention head into dimensions d model / h parameter matrix. W O It is a parameter matrix used to map the dimensions of each head after splicing to the initial dimension d model . The encoding result can be nonlinearly transformed through a position feedforward neural network. The position feedforward neural network can independently update the state of each position in the sequence, enabling the model to better capture the position information in the sequence. It can also perform nonlinear transformations on the output from the attention layer, thereby enhancing the learning ability of the model. The position feedforward neural network can be expressed as FFN(x) = ReLU(xW1+b1)W2+b2. Among them, W1 and W2 are linear transformation matrices, and b1 and b2 are the bias matrices of the two linear transformations respectively. The position feedforward network contains two linear transformations, and the ReLU (rectified linear unit) activation function is applied in the middle. The ReLU activation function can be expressed as ReLU(x) = max(0,x). The nonlinear transformation result can be deconvolved through a deconvolution layer to obtain an updated feature representation. The deconvolution layer and the convolution layer can have the same convolution kernel. The convolution layer can use the convolution kernel for convolution processing, and the deconvolution layer can use the convolution kernel for deconvolution processing.

[0045] The attention layer and positional feedforward neural network can be those found in the Transformer model. The convolutional layer can be that found in the CNN model. This allows the encoder to fuse the Transformer and CNN models, leveraging the CNN model's ability to effectively extract local features while combining the Transformer model's self-attention mechanism's ability to capture the global dependencies of local features, thereby improving the encoder's ability to understand data.

[0046] Furthermore, each encoder may also include a normalization layer. The normalization layer is used to normalize the data, making the data distribution more stable, thereby increasing the model's generalization ability and improving model performance. The normalization layer may specifically include a first normalization layer, a second normalization layer, a third normalization layer, and a fourth normalization layer. The first normalization layer is used to normalize the output of the convolutional layer, and the attention layer can encode the output of the first normalization layer. The second normalization layer is used to perform residual connections on the input and output of the attention layer and normalize the connection results. The position feedforward neural network can perform a nonlinear transformation on the output of the second normalization layer. The third normalization layer is used to perform residual connections on the input and output of the position feedforward neural network and normalize the connection results. The deconvolution layer can perform deconvolution on the output of the third normalization layer. The fourth normalization layer is used to perform residual connections on the input and output of the deconvolution layer and normalize the connection results, thereby obtaining an updated feature representation.

[0047] In some embodiments, the convolution kernel sizes of the N encoders tend to decrease in a specified direction, and the specified direction is the input to output direction of the encoding module. This allows the convolution layer to extract the local features of the messenger RNA motif more and more finely, so that the attention layer can understand the dependencies between local features of different sizes more and more finely. And the approach from coarse-grained to fine-grained also helps to reduce the amount of computation. The convolution kernel can be a matrix, and the size of the convolution kernel can include the number of rows and / or columns of the matrix. The convolution kernel sizes of the N encoders can decrease monotonically in a specified direction. For example, the convolution kernel size of the first encoder among the N encoders can be K. The difference in the convolution kernel sizes between adjacent encoders can be equal. Specifically, for example, the convolution kernel sizes of the N encoders can be K, K-1, K-2, ..., K-N+1, respectively. Alternatively, the difference in the convolution kernel sizes between adjacent encoders can also be unequal. Specifically, the difference in the convolution kernel sizes between adjacent encoders can decrease. Specifically, for example, the convolution kernel size of the encoder can be Ki 2 +1, i represents the position of the encoder in the N encoders. The position of the first encoder can be 1, and the position of the Nth encoder can be N. Alternatively, the difference in the convolution kernel size between adjacent encoders can also be increased. Specifically, for example, the convolution kernel size of the encoder can be i represents the position of the encoder among N encoders.

[0048] In some embodiments, the prediction module may include a pooling layer, a linear layer, a nonlinear layer, a fully connected layer, and an output layer. The updated feature representation may be pooled using the pooling layer. The pooling result may be subjected to dimensionality reduction using the linear layer. The dimensionality reduction result may be subjected to a nonlinear transformation using the nonlinear layer. The nonlinear transformation result may be mapped to a label space using the fully connected layer. The label space is the set of all potential output values ​​that the model can produce. The label space covers all possible outputs of the model. The output layer may perform translation efficiency prediction on the output of the fully connected layer, thereby obtaining a translation efficiency prediction result for the messenger RNA. The pooling layer may further include a first pooling layer and a second pooling layer. The updated feature representation may be pooled using the first pooling layer and the second pooling layer, respectively. The pooling results of the first pooling layer and the second pooling layer may be aggregated. The aggregated result may then be processed using the linear layer. The first pooling layer may perform maximum pooling on the updated feature representation. The second pooling layer may perform mean pooling on the updated feature representation. By concatenating the pooling results of the first and second pooling layers, a richer feature representation can be formed. The concatenated feature representation is then passed through a linear layer for dimensionality reduction and a nonlinear layer for nonlinear transformation, before passing through a fully connected layer and an output layer to obtain the final prediction result.

[0049] In certain embodiments, the translation efficiency prediction result of messenger RNA can include the average ribosome load (MRL) of the 5' untranslated region for the screening of messenger RNA. The average ribosome load is used to represent the binding ability of messenger RNA to ribosomes, thereby being able to indicate the translation efficiency of messenger RNA. Of course, the translation efficiency prediction result of messenger RNA can also include other numerical values, such as TE (Translation Efficiency), the half-life of mRNA, etc. TE can be the ratio of protein expression to messenger RNA expression.

[0050] The translation efficiency prediction method of the embodiment of the present application can predict the translation efficiency of messenger RNA based on a translation efficiency prediction model. The translation efficiency prediction model includes an embedding module, an encoding module, and a prediction module. The sequence data of the messenger RNA can be obtained; the features of the sequence data can be extracted by the embedding module to obtain a feature representation of the messenger RNA; the feature representation can be input into the encoding module, and the input feature representation can be convolved by the encoding module. The convolution result is encoded based on the attention mechanism, and the updated feature representation is output based on the encoding result; the translation efficiency of the updated feature representation can be predicted by the prediction module to obtain the translation efficiency prediction result of the messenger RNA. This deep learning-based approach can accurately and high-throughput predict the translation efficiency of messenger RNA, reduce the time consumption of translation efficiency measurement, and reduce costs. In addition, the encoding module can use convolution and attention mechanisms to capture the dependency relationship between local features and global features in the sequence data. This improves the encoding module's ability to understand the input data. As a result, the accuracy of translation efficiency prediction is improved.

[0051] The present application also provides a method for training a messenger RNA translation efficiency prediction model. This method can be applied to processors in computing devices such as laptops, desktops, and servers. The method can be implemented by running a translation efficiency prediction model on the processor. The translation efficiency prediction model can include an embedding module, an encoding module, and a prediction module.

[0052] Please refer to Figure 3. The method may include the following steps.

[0053] Step 21: Obtain the sequence data and tag data of the messenger RNA.

[0054] Step 22: Input the sequence data into the translation efficiency prediction model to obtain the translation efficiency prediction result of messenger RNA.

[0055] Step 23: Determine the parameters of the translation efficiency prediction model based on the translation efficiency prediction results and label data.

[0056] In some embodiments, sequence data and label data of multiple messenger RNAs can be obtained. For the acquisition of sequence data, please refer to the relevant explanations of the aforementioned embodiments. The sequence data of multiple messenger RNAs can come from an open source public genome database. Of course, other methods can also be used to obtain the sequence data of multiple messenger RNAs. Label data is used to represent the actual translation efficiency of messenger RNA. Label data can include the average ribosome load of the 5' untranslated region, TE, half-life of mRNA, etc. The label data can be obtained by biological experiment measurement using a measuring instrument. It should be noted that the measuring instrument can be any instrument that can determine the translation efficiency of messenger RNA, and different biological experiments can use different measuring instruments, which is not limited in this application.

[0057] In some embodiments, the features of the sequence data can be extracted through the embedding module to obtain a feature representation of the messenger RNA; the feature representation can be input into the encoding module, the encoding module performs convolution processing on the input feature representation, the convolution result is encoded based on the attention mechanism, and the updated feature representation is output according to the encoding result; the translation efficiency of the updated feature representation is predicted by the prediction module to obtain a translation efficiency prediction result of the messenger RNA.

[0058] In certain embodiments, the sequence data of each messenger RNA can be input into the translation efficiency prediction model respectively to obtain the translation efficiency prediction result of the messenger RNA. The translation efficiency prediction result of the messenger RNA can include the average ribosome load (Mean Ribosome Load, MRL) of the 5' non-translated region. Average ribosome load is used to represent the binding ability of messenger RNA and ribosome, so as to indicate the translation efficiency of the messenger RNA. Of course, the translation efficiency prediction result of the messenger RNA can also include other numerical values, such as TE, the half-life of mRNA, etc. TE can be the ratio of protein expression to messenger RNA expression. Specific process can be seen in the aforementioned embodiment, and will not be repeated here.

[0059] In some embodiments, loss information can be calculated using a loss function based on the translation efficiency prediction results and label data of messenger RNA; parameters of the translation efficiency prediction model can be determined based on the loss information: specifically, for example, the parameters of the translation efficiency prediction model can be optimized based on the loss information to obtain the parameters of the optimized translation efficiency prediction model, and a trained translation efficiency prediction model can be constructed based on the parameters of the optimized translation efficiency prediction model to perform translation efficiency prediction. The loss function can include a mean square error loss function, an absolute error loss function, etc. For example, the mean square error loss function can be expressed as Where n represents the number of messenger RNA, y i represents the translation efficiency prediction result of the i-th messenger RNA, represents the label data for the i-th mRNA. Based on the loss information, the parameters of the translation efficiency prediction model can be calculated using methods such as gradient descent. In practical applications, the parameters of the translation efficiency prediction model can be obtained by iterating steps 21-23 multiple times based on multiple sets of mRNA sequence data and label data.

[0060] The training method of the embodiment of the present application can obtain the sequence data and label data of the messenger RNA, and the label data represents the actual translation efficiency of the messenger RNA; the sequence data can be input into the translation efficiency prediction model to obtain the translation efficiency prediction result of the messenger RNA; the parameters of the translation efficiency prediction model can be determined based on the translation efficiency prediction result and the label data. In this way, the training of the translation efficiency prediction model can be realized, thereby facilitating the prediction of the translation efficiency of the messenger RNA using a deep learning method, reducing the time consumption of measuring the translation efficiency, and reducing the cost of translation efficiency.

[0061] The present application also provides a method for screening messenger RNA (mRNA). This method can be applied to computers such as laptops, desktops, and servers. The method can be implemented based on a translation efficiency prediction model, which can include an embedding module, an encoding module, and a prediction module.

[0062] Please refer to Figure 4. The method may include the following steps.

[0063] Step 31: Obtain sequence data of multiple messenger RNAs.

[0064] Step 32: Input the sequence data into the translation efficiency prediction model to obtain the translation efficiency prediction result of the messenger RNA.

[0065] Step 33: Screen multiple messenger RNAs based on the translation efficiency prediction results.

[0066] In some embodiments, for sequence data, please refer to the relevant explanations of the aforementioned embodiments, which will not be repeated here.

[0067] In some embodiments, the features of the sequence data can be extracted through the embedding module to obtain a feature representation of the messenger RNA; the feature representation can be input into the encoding module, the encoding module performs convolution processing on the input feature representation, the convolution result is encoded based on the attention mechanism, and the updated feature representation is output according to the encoding result; the translation efficiency of the updated feature representation is predicted by the prediction module to obtain a translation efficiency prediction result of the messenger RNA.

[0068] In certain embodiments, the sequence data of each messenger RNA can be input into the translation efficiency prediction model respectively to obtain the translation efficiency prediction result of the messenger RNA. The translation efficiency prediction result of the messenger RNA can include the average ribosome load (Mean Ribosome Load, MRL) of the 5' non-translated region. Average ribosome load is used to represent the binding ability of messenger RNA and ribosome, so as to indicate the translation efficiency of the messenger RNA. Of course, the translation efficiency prediction result of the messenger RNA can also include other numerical values, such as TE, the half-life of mRNA, etc. TE can be the ratio of protein expression to messenger RNA expression. Specific process can be seen in the aforementioned embodiment, and will not be repeated here.

[0069] In some embodiments, multiple mRNAs can be screened based on the translation efficiency prediction results. For example, mRNAs whose translation efficiency meets set conditions can be selected from multiple mRNAs. The set conditions can include a translation efficiency greater than or equal to a set threshold. Specifically, for example, mRNAs whose mean ribosome load (MRL) is greater than or equal to a set threshold can be selected from multiple mRNAs.

[0070] Multiple messenger RNAs can be screened based solely on the translation efficiency prediction results. Of course, multiple messenger RNAs can also be screened based on the translation efficiency prediction results in combination with one or more other indicator data. The indicator data include, but are not limited to: the ratio of guanine ribonucleotides to cytosine ribonucleotides, the ratio of guanine ribonucleotides to cytosine ribonucleotides, the ratio of adenine ribonucleotides to uracil ribonucleotides, the ratio of adenine ribonucleotides to uracil ribonucleotides, the minimum folding free energy (MFE), etc. For example, the translation efficiency can correspond to a first set condition, and the indicator data can correspond to a second set condition. Then, the translation efficiency prediction results of multiple messenger RNAs and one or more other indicator data can be obtained; messenger RNAs whose translation efficiency meets the first set condition and each other indicator data meets the corresponding second set condition can be selected from the multiple messenger RNAs.

[0071] The selected mRNAs meet business needs and can be used in drug development, accelerating R&D and reducing costs. For example, high-throughput, high-translation-efficiency 5'UTRs can be screened to provide technical support for mRNA therapies.

[0072] The screening method of the present invention can obtain sequence data of multiple mRNAs; input the sequence data into a translation efficiency prediction model to obtain a translation efficiency prediction result of the mRNA; and screen the multiple mRNAs based on the translation efficiency prediction result. In this way, mRNAs that meet business needs can be obtained.

[0073] An embodiment of the present application also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for predicting the translation efficiency of messenger RNA, the method for screening messenger RNA, and the method for training a messenger RNA translation efficiency prediction model are implemented.

[0074] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned messenger RNA translation efficiency prediction method, messenger RNA screening method, and messenger RNA translation efficiency prediction model training method.

[0075] An embodiment of the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the above-mentioned messenger RNA translation efficiency prediction method, messenger RNA screening method, and messenger RNA translation efficiency prediction model training method.

[0076] Those skilled in the art will appreciate that the present application can be provided as a method, system, or computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. The computer can be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0078] The various functional units in the embodiments of the present application may be integrated into one processing unit, or each functional unit may exist physically separately, or two or more functional units may be integrated into one processing unit.

[0079] Those skilled in the art will understand that the descriptions of the various embodiments in this application each have their own focus. For portions not described in detail in a particular embodiment, reference can be made to the relevant descriptions of other embodiments. In addition, it is understood that after reading this application document, those skilled in the art may, without inventive effort, conceive of any combination of some or all of the embodiments listed in this application, and such combinations are also within the scope of disclosure and protection of this application.

[0080] Although the present application has been described through examples, those skilled in the art will appreciate that the above examples are merely intended to help understand the core concept of the present application. Those skilled in the art will appreciate that the present application is susceptible to numerous variations and modifications. It is intended that the appended claims encompass such variations and modifications without departing from the spirit of the present application.

Claims

1. A method for predicting the translation efficiency of messenger RNA, characterized in that: The method is implemented based on a translation efficiency prediction model, which includes an embedding module, an encoding module, and a prediction module; the method includes: obtaining messenger RNA sequence data; Extracting features of sequence data through the embedding module to obtain feature representation of messenger RNA; Inputting the feature representation into the encoding module, performing convolution processing on the input feature representation through the encoding module, encoding the convolution result based on the attention mechanism, and outputting an updated feature representation according to the encoding result; and The prediction module performs translation efficiency prediction on the updated feature representation to obtain a translation efficiency prediction result.

2. The method according to claim 1, characterized in that The sequence data includes sequence data of the 5' untranslated region in the messenger RNA, and the translation efficiency prediction result includes the average ribosome load of the 5' untranslated region.

3. The method according to claim 1, characterized in that The method further comprises: Determine the coding of nucleotides in messenger RNA according to the set coding method; generating a code for the sequence data according to the code of the nucleotides; The embedding module includes an embedding layer and a linear layer, and the step of extracting features of sequence data includes: Mapping the encoding of the sequence data into a vector through the embedding layer; and The dimension of the vector is transformed through the linear layer to obtain a feature representation.

4. The method according to claim 1, wherein The encoding module includes N encoders stacked in sequence, where N is a positive integer greater than 1; the step of inputting the feature representation into the encoding module includes: The feature representation is input into the first encoder of the N encoders to update the feature representation in sequence through the N encoders; wherein each of the encoders is used to perform convolution processing on the input feature representation, encode the convolution result based on the attention mechanism, and output the updated feature representation according to the encoding result.

5. The method according to claim 4, characterized in that Each of the encoders includes a convolution kernel, and the sizes of the convolution kernels of the N encoders show a decreasing trend in a specified direction, and the specified direction is the input to output direction of the encoding module.

6. The method according to claim 1, characterized in that The encoder includes a convolutional layer, an attention layer, a position feedforward neural network, and a deconvolutional layer; and the step of updating the feature representation of messenger RNA includes: Performing convolution processing on the input feature representation through the convolution layer; Encode the convolution result through the attention layer; Performing nonlinear transformation on the encoding result by using the position feedforward neural network; and The nonlinear transformation result is deconvolved by the deconvolution layer to obtain an updated feature representation.

7. The method according to claim 1, characterized in that The prediction module includes a pooling layer, a linear layer, a nonlinear layer, a fully connected layer and an output layer; The steps of predicting translation efficiency for the updated feature representation include: Performing pooling processing on the updated feature representation through the pooling layer; Performing dimensionality reduction processing on the pooling processing result through the linear layer; Performing nonlinear transformation on the dimensionality reduction processing result through the nonlinear layer; Mapping the nonlinear transformation result to the label space through the fully connected layer; and The output layer predicts the translation efficiency of the output of the fully connected layer.

8. A method for screening messenger RNA, characterized in that: The method is implemented based on a translation efficiency prediction model, which includes an embedding module, an encoding module, and a prediction module; the method includes: obtaining sequence data of multiple messenger RNAs; Extracting features of the sequence data through the embedding module to obtain a feature representation of the messenger RNA; inputting the feature representation into the encoding module to perform convolution processing on the input feature representation through the encoding module, encoding the convolution result based on the attention mechanism, and outputting an updated feature representation according to the encoding result; performing translation efficiency prediction on the updated feature representation through the prediction module to obtain a translation efficiency prediction result of the messenger RNA; and The plurality of messenger RNAs are screened according to the translation efficiency prediction results.

9. A method for training a messenger RNA translation efficiency prediction model, characterized in that: The translation efficiency prediction model includes an embedding module, an encoding module and a prediction module, and the method includes: Obtaining sequence data and tag data of the messenger RNA, wherein the tag data represents the actual translation efficiency of the messenger RNA; Extracting features of the sequence data through the embedding module to obtain a feature representation of the messenger RNA; inputting the feature representation into the encoding module to perform convolution processing on the input feature representation through the encoding module, encoding the convolution result based on the attention mechanism, and outputting an updated feature representation according to the encoding result; performing translation efficiency prediction on the updated feature representation through the prediction module to obtain a translation efficiency prediction result of the messenger RNA; and The parameters of the translation efficiency prediction model are determined based on the translation efficiency prediction results and label data.

10. An electronic device, characterized in that: include: at least one processor; A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions comprising instructions for executing the method according to any one of claims 1-9.

Images