Cross-species codon optimization method and system

Abstract

The invention provides a cross-species codon optimization method and system, electronic equipment, a computer readable storage medium and a computer program product, and relates to the technical field of sequence optimization, the method comprises the following steps: obtaining basic information, the basic information comprising an original amino acid sequence of an original species and category information of a target species; determining a current sequence position in the original amino acid sequence; constructing current input information based on the basic information and the prediction sequence set; predicting and generating a codon vector of a next sequence position in combination with the current input information and a preset prediction strategy; taking the next sequence position as a new current sequence position, adding the codon vector of the current sequence position into the prediction sequence set, and circularly executing the step until an output condition is met; on the basis of the prediction sequence set, a plurality of candidate nucleic acid sequences are generated, the optimization result corresponding to the target species is obtained, the biological correlation of the optimization result is improved, and the protein expression efficiency and the expression stability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of sequence optimization, and in particular to a cross-species codon optimization method, system, electronic device, computer-readable storage medium and computer program product. Background Art

[0002] Proteins are widely used in many fields such as medicine, industrial enzyme preparations and agriculture, and high-quality protein folding and modification are essential for their functions.

[0003] In order to improve the expression level of protein, nucleic acid sequences (mainly mRNA or DNA sequences) are usually optimized.

[0004] Traditional nucleic acid sequence optimization methods, such as replacing synonymous codons, aim to increase protein expression without changing the protein amino acid sequence. However, the above optimization methods have low biological relevance and are therefore not very applicable to cross-species nucleic acid sequence optimization.

[0005] Therefore, a cross-species codon optimization method, system, electronic device, computer-readable storage medium and computer program product need to be proposed. Summary of the invention

[0006] The present specification provides a cross-species codon optimization method, system, electronic device, computer-readable storage medium and computer program product, which can perform codon optimization based on the original amino acid sequence of the original species and the category information of the target species, generate other nucleic acid sequences corresponding to the same species, and improve protein expression efficiency.

[0007] The present application provides a cross-species codon optimization method using the following technical solutions, including:

[0008] Obtaining basic information, the basic information including: the original amino acid sequence of the original species and the category information of the target species;

[0009] Determine the current sequence position in the original amino acid sequence; construct current input information based on the basic information and the predicted sequence set; predict and generate a codon vector for the next sequence position in combination with the current input information and a preset prediction strategy; use the next sequence position as a new current sequence position, add the codon vector for the current sequence position to the predicted sequence set, and repeat this step until an output condition is met;

[0010] Based on the predicted sequence set, several candidate nucleic acid sequences are generated, and the candidate nucleic acid sequences are used as optimization results corresponding to the target species.

[0011] Optionally, constructing current input information based on the basic information and the prediction sequence set includes:

[0012] Searching for a corresponding species category unit based on the identification information of the target species;

[0013] The original amino acid sequence, the species category unit, the identification unit, and the predicted sequence set are sequentially combined to obtain the current input information.

[0014] Optionally, combining the current input information with a preset prediction strategy to predict and generate a codon vector for the next sequence position includes:

[0015] Constructing an input result of a current layer decoder based on the current input information and / or an output result of a previous layer decoder;

[0016] Based on the input result of the current layer decoder, generate an attention output matrix as the output result of the current layer decoder;

[0017] The next layer decoder is used as the new current layer decoder.

[0018] Optionally, generating an attention output matrix based on the input result of the current layer decoder as the output result of the current layer decoder includes:

[0019] Performing a linear transformation on each word embedding vector in the input result to obtain a transformation result; the transformation result includes: a query vector, a key vector, and a value vector;

[0020] Applying the rotation encoding matrix to the query vector and the key vector, performing rotation encoding on the query vector and the key vector, and obtaining a new query vector and a new key vector;

[0021] According to the query vector, key vector and value vector, the attention output matrix is ​​obtained;

[0022] The attention output matrices of all heads are concatenated and the concatenation result is linearly transformed to obtain the final multi-head attention output, which is used as the output result of the current layer decoder.

[0023] Optionally, the output condition includes: the current sequence position is the last sequence position of the original amino acid sequence.

[0024] Optionally, the method further comprises: screening the candidate sequence to determine the target codon sequence;

[0025] The step of screening the candidate sequence to determine the target codon sequence comprises:

[0026] Screening a number of candidate sequences according to sequence screening conditions to obtain a preferred codon sequence;

[0027] The protein expression level corresponding to the preferred codon sequence is determined, and the preferred codon sequence with the highest protein expression level is selected as the target codon sequence.

[0028] The cross-species codon optimization system provided in this application adopts the following technical solutions, including:

[0029] An acquisition module, used to acquire basic information, wherein the basic information includes: an original amino acid sequence of an original species and category information of a target species;

[0030] A prediction module, used to input the basic information into a cross-species codon optimization model to obtain a set of predicted sequences;

[0031] The optimization module generates a plurality of candidate nucleic acid sequences based on the predicted sequence set, and uses the candidate nucleic acid sequences as optimization results corresponding to the target species.

[0032] Optional, prediction module, including:

[0033] A position determination submodule, used to determine the current sequence position in the original amino acid sequence;

[0034] An information construction submodule, used to construct current input information based on the basic information and the prediction sequence set;

[0035] A prediction submodule, used to predict and generate a codon vector of the next sequence position by combining the current input information and a preset prediction strategy;

[0036] A loop submodule, used to take the next sequence position as the new current sequence position, add the codon vector of the current sequence position to the predicted sequence set, and execute this step cyclically until the output condition is met;

[0037] Optionally, the information construction submodule includes:

[0038] A first construction unit is used to search for a corresponding species category unit based on the identification information of the target species;

[0039] The second construction unit is used to sequentially combine the original amino acid sequence, the species category unit, the identification unit, and the predicted sequence set to obtain the current input information.

[0040] Optionally, the prediction submodule includes:

[0041] A first prediction unit, configured to construct an input result of a current layer decoder based on the current input information and / or an output result of a previous layer decoder;

[0042] A second prediction unit, configured to generate an attention output matrix based on an input result of the current layer decoder as an output result of the current layer decoder;

[0043] The next layer decoder is used as the new current layer decoder.

[0044] Optionally, the second prediction unit includes:

[0045] A transformation subunit, configured to perform a linear transformation on each word embedding vector in the input result to obtain a transformation result; the transformation result includes: a query vector, a key vector, and a value vector;

[0046] A rotation encoding subunit, used for applying the rotation encoding matrix to the query vector and the key vector, performing rotation encoding on the query vector and the key vector, and obtaining a new query vector and a new key vector;

[0047] The attention calculation subunit is used to obtain the attention output matrix according to the query vector, the key vector and the value vector;

[0048] The output subunit is used to concatenate the attention output matrices of all heads, perform linear transformation on the concatenation result, and obtain the final multi-head attention output as the output result of the current layer decoder.

[0049] Optionally, the output condition includes: the current sequence position is the last sequence position of the original amino acid sequence.

[0050] Optionally, it also includes: a screening module, used to screen the candidate sequence to determine the target codon sequence;

[0051] Optionally, the screening module includes:

[0052] A first screening submodule, used to screen a plurality of candidate sequences according to a sequence screening condition to obtain a preferred codon sequence;

[0053] The second screening submodule is used to determine the protein expression level corresponding to the preferred codon sequence, and select the preferred codon sequence with the highest protein expression level as the target codon sequence.

[0054] This specification also provides an electronic device, wherein the electronic device includes:

[0055] processor; and,

[0056] A memory storing computer executable instructions, which when executed cause the processor to perform any of the above methods.

[0057] The present specification also provides a computer-readable storage medium, wherein the computer-readable storage medium stores one or more programs, and when the one or more programs are executed by a processor, any of the above methods is implemented.

[0058] This specification also provides a computer program product, wherein the computer program product includes: a computer program / instructions, and when the computer program / instructions are executed by a processor, any of the above methods is implemented.

[0059] In the present application, basic information is obtained, and the basic information includes: the original amino acid sequence of the original species and the category information of the target species; the current sequence position in the original amino acid sequence is determined; based on the basic information and the predicted sequence set, the current input information is constructed; in combination with the current input information and the preset prediction strategy, the codon vector of the next sequence position is predicted and generated; the next sequence position is used as the new current sequence position, and the codon vector of the current sequence position is added to the predicted sequence set, and this step is executed repeatedly until the output condition is met; based on the predicted sequence set, several candidate nucleic acid sequences are generated, and the candidate nucleic acid sequences are used as the optimization results corresponding to the target species, thereby improving the biological relevance of the optimization results, and improving the protein expression efficiency and the stability of expression. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of the principle of a cross-species codon optimization method provided in the embodiments of this specification;

[0061] Figure 2 A schematic diagram of a training sample for a cross-species codon optimization method provided in an embodiment of this specification;

[0062] Figure 3 A schematic diagram of a cross-species codon optimization method provided in the embodiments of this specification;

[0063] Figure 4 Model verification of a cross-species codon optimization method provided in the embodiments of this specification Figure 1 ;

[0064] Figure 5 Model verification of a cross-species codon optimization method provided in the embodiments of this specification Figure 2 ;

[0065] Figure 6 Model verification of a cross-species codon optimization method provided in the embodiments of this specification Figure 3 ;

[0066] Figure 7A schematic diagram of the structure of a cross-species codon optimization system provided in the embodiments of this specification;

[0067] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification;

[0068] Fig. 9 A schematic diagram of a computer-readable medium provided in accordance with an embodiment of the present specification. DETAILED DESCRIPTION

[0069] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0070] Exemplary embodiments of the present invention will now be described more fully with reference to the accompanying drawings. However, exemplary embodiments can be implemented in a variety of forms, and should not be construed as limiting the present invention to the embodiments set forth herein. On the contrary, providing these exemplary embodiments enables the present invention to be more comprehensive and complete, and is more convenient for fully conveying the inventive concept to those skilled in the art. The same reference numerals in the figures represent the same or similar elements, components or parts, and thus their repeated description will be omitted.

[0071] Under the premise of being consistent with the technical concept of the present invention, the features, structures, characteristics or other details described in a specific embodiment do not exclude that they can be combined in one or more other embodiments in a suitable manner.

[0072] In the description of specific embodiments, the features, structures, characteristics or other details described in the present invention are intended to enable those skilled in the art to fully understand the embodiments. However, it does not exclude that those skilled in the art can practice the technical solutions of the present invention without one or more of the specific features, structures, characteristics or other details.

[0073] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0074] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0075] The term "and / or" or "and / or" includes all combinations of any one or more of the associated listed items.

[0076] Figure 1 A schematic diagram of the principle of a cross-species codon optimization method provided in the embodiments of this specification, the method comprising:

[0077] S2 obtaining basic information, wherein the basic information includes: an original amino acid sequence of the original species and category information of the target species;

[0078] S3 determines the current sequence position in the original amino acid sequence; constructs the current input information based on the basic information and the predicted sequence set; predicts and generates the codon vector of the next sequence position in combination with the current input information and the preset prediction strategy; takes the next sequence position as the new current sequence position, adds the codon vector of the current sequence position to the predicted sequence set, and executes this step repeatedly until the output condition is met;

[0079] S4 generates a plurality of candidate nucleic acid sequences based on the predicted sequence set, and uses the candidate nucleic acid sequences as optimization results corresponding to the target species.

[0080] Proteins are widely used in many fields. In order to improve the application effect, it is crucial to increase the protein expression level. In the fields of biotechnology and genetic engineering, in order to improve the synthesis efficiency and quality of proteins, nucleic acid sequences (mainly mRNA or DNA sequences) are usually optimized.

[0081] Traditional nucleic acid sequence optimization methods mainly include optimization based on codon frequency, energy optimization, and optimization based on long short-term memory (LSTM) models. These methods improve protein expression levels by replacing synonymous codons (i.e., different codons encoding the same amino acid).

[0082] However, the traditional nucleic acid sequence optimization methods have the following main problems:

[0083] (1) The optimization method based on codon frequency fails to fully consider the interaction and context dependency between codons, which may result in that although a single codon appears to be optimized, the actual expression effect of the overall sequence in the organism is not as expected.

[0084] (2) Although the energy optimization method takes into account the physicochemical properties of the sequence, these properties do not always accurately reflect the actual expression mechanism of the biological system. Therefore, the effectiveness of this method in practical applications may be limited.

[0085] (3) Although the optimization method based on the LSTM model can capture long-range dependencies, it has low computational efficiency when processing large-scale data, and the model training is complex, making it difficult to apply in actual production.

[0086] In addition, the fact that a gene is highly expressed in one species or expression system does not mean that it will be so in another species or system. In other words, due to the differences in the actual expression mechanisms of various species in biological systems, it is possible that a gene that is highly expressed in one species may not be expressed well in another species.

[0087] Therefore, when performing cross-species nucleic acid sequence optimization, simply replacing synonymous codons may not effectively improve the expression level of the protein, and the optimization method has low applicability and accuracy in diverse biological systems. In order to comprehensively, efficiently and accurately predict and optimize nucleic acid sequences, the present invention provides a cross-species nucleic acid sequence optimization method, which specifically includes:

[0088] S1 builds and trains a cross-species codon optimization model;

[0089] S11 builds training samples;

[0090] S111 obtains a number of sequence data;

[0091] Determine several typical species; specifically, several species can be determined as typical species according to actual needs, and model training is performed based on sequence data of typical species to improve the fitness of the model.

[0092] In one embodiment of the present specification, the types of typical species include, but are not limited to, mammals, Saccharomyces cerevisiae and Escherichia coli.

[0093] The categories of mammals include, but are not limited to, Chinese hamster (Cricetulus griseus), Homo sapiens / human (Homo sapiens), mouse (Mus musculus), and brown rat / rat (Rattus norvegicus).

[0094] Using the RefSeq database of NCBI as a data source, a large-scale collection of sequence data of multiple species was collected from the RefSeq genome data.

[0095] The sequence data includes: nucleic acid sequence, protein sequence encoded by nucleic acid sequence, species information of typical species corresponding to nucleic acid sequence. Among them, nucleic acid sequence is nucleotide sequence. Protein sequence is amino acid sequence. Species information includes: species name or species classification ID.

[0096] S112 performs data cleaning on the sequence data to obtain filtered sequence data.

[0097] The sequence data is cleaned to ensure the accuracy and consistency of the data.

[0098] S112-1 pre-processes the sequence data according to preset processing conditions, specifically including: removing non-text characters in the sequence data, and / or removing sensitive information in the sequence data.

[0099] S112-2: Filter the preprocessed sequence data according to preset filtering conditions to obtain filtered sequence data.

[0100] Specifically, the preset screening conditions include: a number of screening sub-conditions.

[0101] The sequence data that meets all the filter conditions is used as the filtered sequence data.

[0102] In one embodiment of the present specification, the screening sub-conditions include but are not limited to: length screening sub-conditions, end screening sub-conditions, and base screening sub-conditions.

[0103] Among them, the length screening condition includes: the sequence length of the nucleic acid sequence in the sequence data is a multiple of 3.

[0104] The end screening sub-condition includes: the nucleic acid sequence in the sequence data starts with a start codon, and the nucleic acid sequence in the sequence data ends with a stop codon.

[0105] The base filter condition includes: the nucleic acid sequence in the sequence data does not contain bases other than adenine (A), thymine (T), guanine (G) or cytosine (C).

[0106] By pre-setting screening conditions, the sequence data corresponding to nucleic acid sequences with unconventional lengths or forms are removed to improve the accuracy and efficiency of subsequent model training.

[0107] S113 adjusts the filtered sequence data according to the preset combination conditions to construct original sequence information;

[0108] S113-1 annotates the screened sequence data according to a preset labeling strategy to generate expression identification sequences, codon identification sequences, and species classification units;

[0109] The screened sequence data are labeled according to a preset labeling strategy to obtain a number of information units, and the types of information units include: codon units, amino acid units, and species units.

[0110] Specifically, the nucleic acid sequence, the protein sequence encoded by the nucleic acid sequence, and the species information corresponding to the nucleic acid sequence are extracted from the screened sequence data.

[0111] The codons in the nucleic acid sequence (each three bases in a group) are taken as a codon unit, and based on the order of the codons, a codon identification sequence corresponding to the nucleic acid sequence is generated from the start codon to the stop codon; the codon identification sequence includes: a plurality of ordered codon units. Among them, the arrangement order of the codon units in the codon identification sequence is consistent with the arrangement order of the codons in the nucleic acid sequence.

[0112] Each amino acid in the protein sequence is taken as an amino acid unit to generate an expression identifier sequence corresponding to the protein sequence; the expression identifier sequence includes a number of ordered amino acid units, wherein the arrangement order of the amino acid units in the expression identifier sequence is consistent with the arrangement order of the amino acids in the protein sequence.

[0113] Several species labels are preset, and each species information corresponds to a species label. The corresponding species label is matched based on the species information in the sequence data. A species label is a species category unit.

[0114] S113-2 arranges the expression identification sequence, codon identification sequence, and species category unit according to preset combination conditions to form a complete original sequence information.

[0115] Specifically, the expression identification sequence, the codon identification sequence, and the species classification unit are arranged in a combination order, wherein the combination order from left to right is: expression identification sequence, species classification unit, and codon identification sequence.

[0116] S114 collects identification units and original sequence information to construct training samples.

[0117] Set the identification unit, which is used to indicate the end position of the expression identification sequence and the species category unit. The information before the identification unit is the input information, and the information after the identification unit is the information (codon identification sequence) that the model wants to generate. As a preferred example, the identification unit is " <sos>” (start of sentence).

[0118] The identification unit is inserted between the species category unit and the codon identification sequence.

[0119] That is, from left to right, the expression identification sequence, species category unit, identification unit, and codon identification sequence are arranged into a complete training sample.

[0120] In one embodiment of the present specification, the screened sequence data includes:

[0121] Nucleic acid sequence: ATGCTGACGCTG…GAA.

[0122] Protein sequence encoded by nucleic acid sequence: MLTL…E.

[0123] Species information corresponding to the nucleic acid sequence: human.

[0124] Then, a codon identification sequence corresponding to the nucleic acid sequence is generated according to the codon in the nucleic acid sequence as a codon unit; the codon units in the codon identification sequence are "ATG", "CTG", "ACG", "CTG", ..., "GAA" in sequence.

[0125] Each amino acid in the protein sequence is taken as an amino acid unit to generate an expression marker sequence corresponding to the protein sequence; the amino acid units in the expression marker sequence are "M", "L", "T", "L", ... "E" in sequence.

[0126] Based on the species information "human", the corresponding species label "Human" is matched, and the species label "Human" is taken as a species type unit.

[0127] like Figure 2 As shown, the training samples include: ordered amino acid units in the expression identification sequence, species label unit "Human", identification unit <sos>, codons identify ordered codon units in a sequence.

[0128] S12 constructs a cross-species codon optimization model;

[0129] In one embodiment of the present specification, the cross-species codon optimization model is preferably an autoregressive language model, and GPT (Generative Pre-trained Transformer) is selected as the basic architecture. Utilizing the powerful data processing capabilities and large-scale training samples of GPT, the codons in the nucleic acid sequence corresponding to the target species can be more accurately predicted and optimized. Moreover, compared with the LSTM model, the model of the present invention has higher computational efficiency when processing large-scale data and complex sequences, and is suitable for efficient optimization needs in actual production.

[0130] The cross-species codon optimization model includes an embedding layer, a decoding layer, and an output layer.

[0131] When predicting each codon, it is necessary to convert the training samples and predicted codon vectors into numerical representations that the model can handle based on the embedding layer.

[0132] The decoding layer includes 36 layers of the same decoder. Each layer of decoders is responsible for further refining the generated codon sequence to improve its accuracy and biological functionality. Among them, the last layer of decoders obtains context information, which is passed through the output layer to obtain the probability distribution of the codon at the corresponding position.

[0133] Each decoder layer includes a multi-head self-attention module (Scaled Dot-Product Attention) and a feed forward network (FFN) to capture the complex dependencies in the sequence.

[0134] The multi-head self-attention module is used to process different positions in the sequence in parallel to capture long-range dependencies within the sequence. The multi-head design increases the representation capacity of the model, allowing it to focus on multiple aspects of the sequence simultaneously.

[0135] The feedforward neural network is located after the multi-head self-attention module. The feedforward neural network includes a feedforward layer (FF layer) and a residual connection and normalization layer (Add&Norm layer). Among them, the feedforward layer (FF layer) is used for nonlinear transformation and feature enhancement.

[0136] The Add&Norm layer is responsible for accelerating training and improving model stability through residual connections and layer normalization. Specifically, the Add&Norm layer combines residual connections and layer normalization. Residual connections help alleviate the gradient vanishing or gradient exploding problems in deep network training by adding the input of the current layer directly to the output of the current layer. Layer normalization normalizes the activation value of each sample to accelerate the training process and improve model stability.

[0137] To alleviate the problem of slow decoder speed and high memory requirements, a flash attention mechanism is introduced to reduce the computational complexity and memory consumption by reducing the number of positions that need to be calculated for attention or adopting a more efficient attention calculation method. In one embodiment of the present specification, a local sensitive hashing (LSH) attention or a block-based sparse attention mechanism is used to reduce the amount of calculation by focusing only on the key parts or patterns in the sequence.

[0138] S13 uses training samples to train cross-species codon optimization models;

[0139] S131 initializes the cross-species codon optimization model.

[0140] In one embodiment of the present specification, the model parameters are initialized using Xavier initialization or random initialization, which helps the learning process of the model by keeping the variance of the input and output consistent.

[0141] Define the forward propagation function and implement the calculation of the Transformer encoder and decoder.

[0142] A loss function is defined, wherein the loss function is preferably a cross entropy loss function.

[0143] Define an optimizer, preferably the Adam optimizer.

[0144] Considering that in the early stage of the training process, the model weights usually start from random initialization, and the model does not have a good understanding of the distribution of the original sequence information and the relationship between the features. If you start training directly from a larger learning rate, it may cause large fluctuations in the model parameters during the update process, and may even cause the model to fall into a local optimal solution or cause training instability. Therefore, based on the warm-up strategy, a smaller learning rate is used in the early stage of training, and then the learning rate is gradually increased to the preset value.

[0145] In one embodiment of the present specification, a learning rate scheduler is customized. Specifically, at the beginning of the training process, a codon optimization model is used to process one million codon tags for training to achieve warm-up and then a linear decrease in the learning rate.

[0146] S132 uses training samples to pre-train cross-species codon optimization models;

[0147] Through pre-training, the codon optimization model learns the basic patterns and structures in nucleic acid sequences.

[0148] Specifically, through autoregressive training, the codon optimization model attempts to predict the next codon in a given context. Through multiple rounds of training, the codon optimization model gradually learns the rules and potential semantics in nucleic acid sequences, protein sequences and corresponding species.

[0149] In one embodiment of the present specification, the learning rate of the codon optimization model is preferably 1e-4; the batch size is preferably 70; and the number of training rounds is preferably 100.

[0150] The following combination Figure 3 Briefly describe the model training process:

[0151] In the process of codon optimization, it is necessary to optimize the codons corresponding to the amino acids at each sequence position in turn according to the arrangement order of the amino acids in the expression marker sequence;

[0152] 1. Obtain training samples and construct input sequences based on the current sequence position;

[0153] The input sequence includes: expression identification sequence, species category unit, identification unit ( <sos>), and each predicted codon vector. An expression identifier sequence is a unit of information, a species category unit is a unit of information, an identifier unit is a unit of information, and each predicted codon vector is a unit of information.

[0154] That is to say, when the codon at the first sequence position needs to be predicted, since there is no predicted codon vector (the number of predicted codon vectors is 0), the input sequence only includes 3 units of information, namely, the expression identifier sequence, the species category unit, and the identification unit. When the codon at the second sequence position needs to be predicted, 1 codon vector has been predicted, and the input sequence includes 4 units of information, namely, the expression identifier sequence, the species category unit, the identification unit, and the predicted codon vector at the first sequence position. When the codon at the nth sequence position needs to be predicted, n-1 codon vectors have been predicted, and there are 3+(n-1) units of information in the input sequence, namely, the expression identifier sequence, the species category unit, the identification unit, the predicted codon vector at the first sequence position to the predicted codon vector at the n-1th sequence position.

[0155] In one embodiment of the present specification, one unit of information is one token.

[0156] 2. Transform the input sequence through the embedding layer to obtain several word embedding vectors.

[0157] According to the dimension of the word embedding vector, each unit information in the input sequence is converted into a fixed-size, continuous d-dimensional word embedding vector. Among them, d is a hyperparameter that is adjusted according to the actual task and dataset. As a preferred option, d = 768.

[0158] 2. The codon vector of the next sequence position is predicted through the decoder layer.

[0159] The input sequence is processed step by step through multiple decoders in the decoding layer, and the last decoding layer predicts the embedding vector of the next word based on the context;

[0160] The steps for each decoder are as follows:

[0161] 2.1 Perform linear transformation on the word embedding vector to obtain the query vector Q, key vector K and value vector V.

[0162] In order to handle the problem of sequence order, the position information of the word embedding vector is added by rotating the encoding matrix.

[0163] 2.2 Generate a rotation encoding matrix and perform rotation encoding (RoPE) on the query vector Q and the key vector K. Specifically, apply the rotation encoding matrix to the query vector Q and the key vector K respectively to obtain a new query vector Q and a new key vector K.

[0164] 2.3 According to the query vector Q, key vector K and value vector V, the attention output matrix is ​​obtained, where the attention output matrix

[0165] 2.4 Concatenate the attention output matrices of all heads, perform linear transformation on the concatenated result, and obtain the final multi-head attention output as the output result of the current layer decoder.

[0166] MultiHead(Q,K,V)=Concat(head 1 ,......,head h )W O ;

[0167] Where head i =Attention(QW i Q , KW i K , VW i V );

[0168] Preferably, h=8, d=k=64.

[0169] 2.5 Optimization through Feed Forward Network (FFN).

[0170] Feedforward neural network, including a FF layer and an Add&Norm layer.

[0171] In this step, the final multi-head attention output in 2.4 can be transformed linearly, nonlinearly (such as ReLU), and linearly in turn: For example, FFN(x) = ReLU(xW 1 +b 1 )W 2 +b 2 .

[0172] Of course, residual connection and layer normalization can also be added to the results of the above transformation to stabilize the training process and accelerate convergence: specifically, Norm(x)=LayerNorm(x+FFN(x)).

[0173] In each decoder, the output of the decoder of the current layer is passed to the decoder of the next layer and used as the input of the multi-head self-attention module and feed-forward neural network of the next layer. In this way, the decoder of each subsequent layer can build a higher-level representation based on the output of the decoder of the previous layer.

[0174] Generate predictions based on the last decoder layer.

[0175] The prediction result is a context vector calculated based on the output of the entire decoder stack and the attention mechanism. That is, although each layer generates a context vector, only the output of the last layer is used to generate the prediction. This is done through a linear layer and a softmax function to convert the output of the model into a prediction vocabulary (for cDNA, the vocabulary contains 64 codons and the end of the expression sequence). <eos>, a total of 65 units of information) in the probability distribution of each codon.

[0176] In biology, a specific combination of three nucleotides (called codons) can encode an amino acid. However, due to the degeneracy of codons, given an amino acid sequence, its corresponding nucleic acid sequence may not be unique. Therefore, the prediction results include the predicted probability distribution of the codons corresponding to each codon unit.

[0177] 3. The codon with the highest predicted probability distribution is used as the predicted codon of the corresponding position, the predicted codon of the corresponding position is compared with the actual codon, and the loss (such as cross entropy loss) is calculated, and the calculated loss is used to update the parameters of the model.

[0178] exist Figure 3 In the experiment, the prediction results output by the model are consistent with the input results, that is, they perfectly reproduce the known natural cDNA sequences, indicating the accuracy of the model.

[0179] S14 Experimental validation of codon optimization models across species.

[0180] Obtain the protein sequence (first sequence) and the category of the target species as input sequences;

[0181] The input sequence is input into the codon optimization model to generate several optimized mRNA sequences for monoclonal antibodies and interleukins, and the optimized mRNA sequences are used as the second sequence. In one embodiment of the present specification, the number of the second sequence is 100-200.

[0182] For each second sequence, the Codon Adaptation Index (CAI), normalized minimum folding energy (Normalized MFE) and guanine-cytosine (GC) content of the second sequence are calculated to ensure that the generated second sequences are similar to natural mRNA in biophysical properties.

[0183] Screening a plurality of second sequences according to a sequence screening condition, where the sequence screening condition includes a plurality of sequence screening sub-conditions;

[0184] Sequence screening sub-conditions included: CAI of the second sequence was higher than 0.8;

[0185] and / or, the sequence screening sub-condition includes: the proportion of the first 30 bases in the second sequence participating in the secondary structure formation is less than 0.5;

[0186] And / or, the sequence screening sub-conditions include: the GC content of the second sequence is between 0.5 and 0.6;

[0187] And / or, the sequence screening sub-condition includes: the second sequence has a difference in MFE;

[0188] And / or, the sequence screening sub-condition includes: there is a difference of at least N bases between any two second sequences, and preferably, N=5.

[0189] The second sequences that meet all the sequence screening sub-conditions are retained, and a preset number of sequences that best meet the criteria are selected from the retained second sequences as the third sequence, where the preset number of sequences is 5-10.

[0190] The third sequence is cloned into a plasmid vector; the plasmid vector is preferably a pTT5 plasmid vector, and the plasmid vector contains functional elements such as a promoter, an antibiotic resistance gene, and an Fc tag.

[0191] The plasmid vector was transfected into cell lines such as Chinese hamster ovary cells (e.g., CHO-S) and human kidney epithelial cells (e.g., HEK-293F) for experiments.

[0192] Specifically, a liposome transfection reagent was used to introduce the plasmid into the cells. Each time 3 mL of cell suspension was transfected, 3 μg of plasmid (antibody: 1.5 μg for each chain) was cultured in a 37°C, 8% CO2 incubator. The culture medium was collected on the 5th or 6th day after transfection, and the protein concentration was analyzed using Biacore T200;

[0193] Each construct was transfected three times, and the average and standard deviation were calculated. The results showed that each third sequence showed a high protein expression level in CHO-S and HEK-293F cells.

[0194] like Figure 4-6 The experimental verification results are shown. Figure 4 It shows: the distribution of key indexes related to the expression of human natural gene sequences, sequences obtained based on OptimumGene, sequences obtained based on GeneOptimizer, sequences obtained based on LinearDesign, and sequences obtained based on the model of the present invention (CodonGPT) in monoclonal antibodies and interleukins. Among them, the distribution of key indexes of the sequences obtained by the model of the present invention is similar to that of the human natural gene sequences.

[0195] Figure 5 It shows: the distribution of key indexes related to the expression of Chinese hamster natural gene sequence, sequence obtained based on OptimumGene, sequence obtained based on GeneOptimizer, sequence obtained based on LinearDesign, sequence obtained based on JCB_RNN, and sequence obtained based on the model of the present invention (CodonGPT) in monoclonal antibody and interleukin. Among them, the distribution of key indexes of the sequence obtained by the model of the present invention is similar to that of the Chinese hamster natural gene sequence.

[0196] Figure 6 It represents: the natural gene sequence obtained based on the wild-type gene sequence, the sequence obtained based on OptimumGene, the sequence obtained based on GeneOptimizer, the sequence obtained based on LinearDesign, the sequence obtained based on JCB_RNN, the sequence obtained based on the model of the present invention (CodonGPT), etc., and the expression levels of three monoclonal antibodies (mAb) and three interleukins in two systems (CHO and HEK293).

[0197] Among them, OptimumGene refers to the codon optimization technology independently developed by GenScript, and GeneOptimizer refers to the gene optimization method independently developed by Thermo Fisher. LinearDesign is a computational method for optimizing mRNA sequences, mainly used to improve the stability and immunogenicity of sequences. JCB_RNN refers to a RNN-based codon optimization technology published by Dennis R. Goulet et al. in the Journal of Computational Biology, and WT is the wild-type gene sequence.

[0198] The cross-species codon optimization model of the present invention utilizes the powerful data processing capabilities and context-dependent modeling capabilities of GPT to implement an efficient, accurate and biologically relevant method for optimizing codons in nucleic acid sequences to improve the expression level of the target protein in a specific expression system.

[0199] S2 obtaining basic information, wherein the basic information includes: an original amino acid sequence of the original species and category information of the target species;

[0200] In one embodiment of the present specification, the original nucleic acid sequence is obtained, and based on the codon encoding rules, the original nucleic acid sequence is converted into a corresponding amino acid sequence. In another embodiment of the present specification, the amino acid sequence is directly obtained.

[0201] In order to facilitate the subsequent optimization of codons, the obtained amino acid sequence is preprocessed according to preset processing conditions to obtain the original amino acid sequence;

[0202] The preset processing conditions include: removing non-text characters in the amino acid sequence; and / or removing sensitive information in the amino acid sequence. The non-text characters include spaces, tabs, etc.

[0203] In one embodiment of the present specification, each character in the amino acid sequence is traversed to remove all non-standard amino acid characters (such as characters outside the alphabet, spaces, line breaks, etc.) Any sensitive information that may be contained in the amino acid sequence is identified and processed or deleted.

[0204] Check whether the preprocessed amino acid sequence still contains a valid amino acid sequence to verify the sequence integrity. If the verification passes, the preprocessed amino acid sequence is used as the original amino acid sequence.

[0205] S3 inputs the basic information into a cross-species codon optimization model to obtain a predicted sequence set;

[0206] S31 determines the current sequence position m in the original amino acid sequence;

[0207] According to the position of each amino acid in the original amino acid sequence, starting from the first sequence position, the codon for the next sequence position is predicted in sequence.

[0208] S32 constructs current input information based on the basic information and the prediction sequence set;

[0209] Find the corresponding species category unit based on the identification information of the target species;

[0210] The original amino acid sequence, species category unit, identification unit, and predicted sequence set are sequentially combined to obtain the current input information. According to the dimension of the word embedding vector, each unit information in the input result is converted into a fixed-size, continuous word embedding vector.

[0211] Among them, the original amino acid sequence is taken as a unit information, the species category unit is taken as a unit information, the identification unit is taken as a unit information, and each codon vector in the predicted sequence set is a unit information. It is easy to understand that when there are 0 codon vectors in the predicted sequence set, the current input information includes 3 unit information. Among them, the codon vector prediction of the first sequence position is mainly predicted based on the original amino acid sequence, the category identification of the target species, and the identification unit.

[0212] The predicted sequence set includes the current sequence position and all codon vectors corresponding to the position before the current sequence position.

[0213] S33 predicts and generates a codon vector of the next sequence position by combining the current input information and a preset prediction strategy;

[0214] S331 constructs an input result of a current layer decoder based on the current input information and / or an output result of a previous layer decoder;

[0215] If the current layer decoder is the first layer decoder, the output result of the previous layer decoder is empty, that is, the input result of the current layer decoder contains the current input information.

[0216] If the current layer decoder is not the first layer decoder, the input result of the current layer decoder is the output result of the previous layer decoder.

[0217] S332 generates an attention output matrix based on the input result of the current layer decoder as the output result of the current layer decoder;

[0218] Specifically, S332-1 performs a linear transformation on each word embedding vector in the input result to obtain a transformation result; the transformation result includes: a query vector, a key vector, and a value vector;

[0219] S332-2 applies the rotation coding matrix to the query vector Q and the key vector K, performs rotation coding on them, and obtains a new query vector Q and a new key vector K;

[0220] Specifically, a rotation encoding matrix containing position information is generated according to the number h of unit information in the input result.

[0221] The rotation encoding matrix is ​​applied to the query vector Q and the key vector K respectively to obtain a new query vector Q and a new key vector K.

[0222] The rotation encoding matrix is ​​applied to the query vector and the key vector to update the query vector and the key vector.

[0223] Specifically, the rotation encoding corresponding to the i-th word embedding vector in the rotation encoding matrix is ​​applied to q i and k i In one embodiment of the present specification, the rotation code corresponding to the i-th word embedding vector in the rotation coding matrix is ​​applied to the query vector q corresponding to the i-th word embedding vector i , for the query vector q i Update; apply the rotation code corresponding to the i-th word embedding vector in the rotation coding matrix to the key vector k corresponding to the i-th word embedding vector i , for the key vector k i to update.

[0224] In one embodiment of the present specification, rotational position encoding implements relative position encoding by rotating a vector, so that the model can better capture the relative position information in the sequence.

[0225] S332-3 obtains an attention output matrix according to the query vector Q, the key vector K and the value vector V;

[0226] Among them, the attention output matrix

[0227] In one embodiment of the present specification, the dot product of the query vector corresponding to the current word embedding vector and each key vector in the key vector matrix K is calculated, and the initial matching result is obtained by summarizing. Specifically, the key vector matrix K is constructed, and the key vector matrix K includes: the key vector k corresponding to the current word embedding vector and each previous word embedding vector j , 1≤j≤i. Calculate the query vector q corresponding to the current word embedding vector i and all key vectors k in the key vector matrix K j The dot product of is summed up to get the matching dot product result.

[0228] The initial matching result is scaled based on the scaling factor to obtain a scaled matching result, which is used as the target matching result. Specifically, the initial matching result is divided by a scaling factor. Scaling is performed to obtain the attention output matrix; where d k Indicates the dimension of the key vector.

[0229] In another embodiment of the present specification, the query vector q corresponding to the current word embedding vector is calculated. i and all key vectors k j′ The dot product of is summarized to get the first matching result.

[0230] Scaling the first matching result based on the scaling factor to obtain a scaled matching result;

[0231] The scaled matching results are adjusted to obtain the attention output matrix. Among them, the key vector k corresponding to the i+1th word embedding vector is calculated based on the Mask mechanism. i+1 Masking is performed to avoid the query vector q corresponding to the i-th word embedding vector i Interact with the key vectors corresponding to the i+1th word embedding vector and subsequent word embedding vectors.

[0232] Specifically, a lower triangular matrix (or equivalent mask) is applied to the scaled dot product result to set future positions to negative infinity (before sof1max), so that the values ​​of these positions will be close to 0 after sof1max. That is, q i ×k i+1 ,q i ×k i+2 , …, q i ×k h Both are 0.

[0233] S332-4 concatenates the attention output matrices of all heads, performs linear transformation on the concatenated result, obtains the final multi-head attention output, and uses it as the output result of the current layer decoder.

[0234] S333 uses the next layer decoder as the new current layer decoder and repeats S332 until the output result of the last layer decoder is output.

[0235] The output result of the last layer decoder is used as the codon vector of the next sequence position, and the codon vector is added to the prediction vector set.

[0236] The present invention captures contextual dependencies in nucleic acid sequences based on an attention mechanism, thereby improving the accuracy and biological relevance of optimization results.

[0237] S34 takes the next sequence position as the new current sequence position, adds the codon vector x[m] of the current sequence position to the predicted sequence set, and executes step S3 repeatedly until the output condition is met;

[0238] The output conditions include: the current sequence position is the last sequence position of the original amino acid sequence.

[0239] In this loop, the codon vector corresponding to each sequence position is generated in turn.

[0240] S4 generating a plurality of candidate nucleic acid sequences based on the predicted sequence set, and using the candidate nucleic acid sequences as optimization results corresponding to the target species;

[0241] The predicted sequence set includes several codon vectors, and the codon vectors are arranged in a preset order, which can be an ascending order of the generation time of the codon vectors or an ascending order of the corresponding sequence positions.

[0242] Based on all the codon vectors in the predicted sequence set, several candidate nucleic acid sequences are generated, and the candidate nucleic acid sequences are taken as preferred results.

[0243] The codon vector is mapped to a specific output space, and the prediction result corresponding to the codon vector is determined. Based on the prediction result, several candidate nucleic acid sequences are generated as the preferred result.

[0244] The prediction results include the probability distribution of the codons corresponding to each amino acid.

[0245] In one embodiment of the present specification, a sequence set of candidate sequences is constructed based on a beam search algorithm. Initially, the sequence set of candidate sequences is empty. Then, the candidate sequences are iteratively expanded based on the beam search algorithm by adding the next possible codon and selecting the most likely expansion based on the prediction result.

[0246] At each expansion, the candidate sequences in the candidate set are reordered and pruned according to the scoring function to keep the number of candidate sequences in line with the preset candidate requirement. The scoring function includes: the sum of the logarithmic probabilities of the sequences. The preset candidate requirement is preferably 20. That is, the number of candidate sequences in the sequence set is preferably 20.

[0247] Furthermore, when constructing a candidate sequence based on the beam search algorithm, multiple different temperature values ​​are determined, and for each temperature value, a sequence set is obtained.

[0248] All candidate sequences in all sequence sets are taken as the preferred results.

[0249] S5 screens the candidate sequences to determine the target codon sequence;

[0250] For each candidate sequence, the preference index value, frequency index value and content index value of the candidate sequence are calculated.

[0251] Among them, the preference index value is Codon Adaptation Index (CAI). Among them, CAI is an indicator to measure the degree of match between the codon usage preference in a gene sequence and the optimal codon usage frequency in its host organism.

[0252] The stability index value is the normalized minimum folding energy (Normalized MFE). Among them, nMFE is an indicator for evaluating the stability of RNA sequences.

[0253] The content index value is the guanine-cytosine (GC) content. The GC content refers to the ratio of guanine (G) and cytosine (C) bases in a DNA or RNA sequence.

[0254] S51 screening the candidate sequences according to the sequence screening conditions to obtain a preferred codon sequence;

[0255] The sequence screening condition includes several sequence screening sub-conditions;

[0256] Sequence screening sub-conditions include: the preference index value of the candidate sequence is higher than 0.8;

[0257] and / or, the sequence screening sub-conditions include: the proportion of the first 30 bases in the candidate sequence participating in the secondary structure formation is less than 0.5;

[0258] And / or, the sequence screening sub-conditions include: the content index value of the candidate sequence is between 0.5 and 0.6;

[0259] and / or, the sequence screening sub-conditions include: the candidate sequences have differences in stability index values;

[0260] And / or, the sequence screening sub-condition includes: there is a difference of at least N bases between any two candidate sequences, and preferably, N=5.

[0261] The candidate sequences that meet all the sequence screening conditions are retained, and a preset number of sequences that best meet the criteria are selected from the retained candidate sequences as preferred codon sequences, and the preset number of sequences is preferably 5-10.

[0262] In this specification, bioinformatics tools are used to perform secondary structure prediction and energy calculation on multiple candidate nucleic acid sequences to further screen candidate nucleic acid sequences with better performance.

[0263] S52 determines the protein expression level corresponding to the preferred codon sequence, and selects the preferred codon sequence with the highest protein expression level as the target codon sequence.

[0264] The preferred codon sequence is synthesized in vitro and transferred into an appropriate expression system to measure its protein expression level. According to the experimental results, the preferred codon sequence with the highest protein expression level is selected as the target codon sequence.

[0265] Figure 7 A schematic diagram of a cross-species codon optimization system provided in an embodiment of this specification, the system comprising:

[0266] An acquisition module 720 is used to acquire basic information, wherein the basic information includes: an original amino acid sequence of an original species and category information of a target species;

[0267] A prediction module 730 is used to input the basic information into a cross-species codon optimization model to obtain a set of predicted sequences;

[0268] The optimization module 740 generates a plurality of candidate nucleic acid sequences based on the predicted sequence set, and uses the candidate nucleic acid sequences as optimization results corresponding to the target species.

[0269] Optionally, the prediction module 730 includes:

[0270] A position determination submodule, used to determine the current sequence position in the original amino acid sequence;

[0271] An information construction submodule, used to construct current input information based on the basic information and the prediction sequence set;

[0272] A prediction submodule, used to predict and generate a codon vector of the next sequence position by combining the current input information and a preset prediction strategy;

[0273] A loop submodule, used to take the next sequence position as the new current sequence position, add the codon vector of the current sequence position to the predicted sequence set, and execute this step cyclically until the output condition is met;

[0274] Optionally, the information construction submodule includes:

[0275] A first construction unit is used to search for a corresponding species category unit based on the identification information of the target species;

[0276] The second construction unit is used to sequentially combine the original amino acid sequence, the species category unit, the identification unit, and the predicted sequence set to obtain the current input information.

[0277] Optionally, the prediction submodule includes:

[0278] A first prediction unit, configured to construct an input result of a current layer decoder based on the current input information and / or an output result of a previous layer decoder;

[0279] A second prediction unit, configured to generate an attention output matrix based on an input result of the current layer decoder as an output result of the current layer decoder;

[0280] The next layer decoder is used as the new current layer decoder.

[0281] Optionally, the second prediction unit includes:

[0282] A transformation subunit, configured to perform a linear transformation on each word embedding vector in the input result to obtain a transformation result; the transformation result includes: a query vector, a key vector, and a value vector;

[0283] A rotation encoding subunit, used for applying the rotation encoding matrix to the query vector and the key vector, performing rotation encoding on the query vector and the key vector, and obtaining a new query vector and a new key vector;

[0284] The attention calculation subunit is used to obtain the attention output matrix according to the query vector, the key vector and the value vector;

[0285] The output subunit is used to concatenate the attention output matrices of all heads, perform linear transformation on the concatenation result, and obtain the final multi-head attention output as the output result of the current layer decoder.

[0286] Optionally, the output condition includes: the current sequence position is the last sequence position of the original amino acid sequence.

[0287] Optionally, it also includes: a screening module, used to screen the candidate sequence to determine the target codon sequence;

[0288] Optionally, the screening module includes:

[0289] A first screening submodule, used to screen a plurality of candidate sequences according to a sequence screening condition to obtain a preferred codon sequence;

[0290] The second screening submodule is used to determine the protein expression level corresponding to the preferred codon sequence, and select the preferred codon sequence with the highest protein expression level as the target codon sequence.

[0291] The functions of the system of the embodiment of the present invention have been described in the above method embodiment, so for details not fully described in this embodiment, please refer to the relevant description in the above embodiment, and no further description will be given here.

[0292] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0293] The electronic device embodiment of the present invention is described below, and the electronic device can be regarded as a physical implementation of the method and device embodiments of the present invention described above. The details described in the electronic device embodiment of the present invention should be regarded as a supplement to the above method or device embodiments; details not disclosed in the electronic device embodiment of the present invention can be implemented with reference to the above method or device embodiments.

[0294] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification. Figure 8 The computer device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0295] like Figure 8 As shown, the computer device 800 of this exemplary embodiment is in the form of a general data processing device. The components of the computer device 800 may include but are not limited to: at least one processor 810, at least one memory 820, a network interface 830, a display unit 840, an input component 850, etc.

[0296] The memory 820 stores a computer-readable program, which may be a source program or a code of a read-only program. The program may be executed by the processor 810, so that the processor 810 performs the steps of various embodiments of the present invention. For example, the processor 810 may perform the following steps: Figure 1 Steps shown.

[0297] The memory 820 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) and / or a cache memory unit, and may further include a read-only memory unit (ROM). The memory 820 may also include a program / utility having a set (at least one) of program modules, such program modules including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include the implementation of a network environment.

[0298] Also included is a bus (not shown) which may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0299] The computer device 800 may also communicate with one or more external devices (e.g., keyboard, display, network device, Bluetooth device, etc.) so that a user can interact with the computer device 800 via these external devices, and / or so that the computer device 800 can communicate with one or more other data processing devices (e.g., routers, modems, etc.). Such communication may be performed through the network interface 830, and may also be performed through a network adapter with one or more networks (e.g., local area network (LAN), wide area network (WAN) and / or public network, such as the Internet). The network adapter may communicate with other modules of the computer device 800 via a bus. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in the computer device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0300] Fig. 9 Schematic diagram of a computer readable medium embodiment of the present invention. Fig. 9 As shown, the computer program can be stored on one or more computer-readable media. The computer-readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. When the computer program is executed by one or more data processing devices, the computer-readable medium is enabled to implement the above method of the present invention.

[0301] Through the description of the above implementation modes, it is easy for those skilled in the art to understand that the exemplary embodiments described in the present invention can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation mode of the present invention can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a data processing device (which can be a personal computer, a server, or a network device, etc.) to perform the above method according to the present invention.

[0302] The computer readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by an instruction execution system, an apparatus, or a device or used in combination with it. The program code contained on the readable storage medium may be transmitted with any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0303] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0304] In summary, the present invention can be implemented by a method, apparatus, electronic device or computer-readable medium that executes a computer program. In practice, a general data processing device such as a microprocessor or a digital signal processor (DSP) can be used to implement some or all functions of the present invention.

[0305] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the present invention is not inherently related to any specific computer, virtual device or electronic device, and various general devices can also implement the present invention. The above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.< / eos> < / sos> < / sos> < / sos>

Claims

1. A cross-species codon optimization method, characterized in that: include: Obtaining basic information, the basic information including: the original amino acid sequence of the original species and the category information of the target species; Determine the current sequence position in the original amino acid sequence; construct current input information based on the basic information and the predicted sequence set; predict and generate a codon vector for the next sequence position in combination with the current input information and a preset prediction strategy; use the next sequence position as a new current sequence position, add the codon vector for the current sequence position to the predicted sequence set, and repeat this step until an output condition is met; Based on the predicted sequence set, several candidate nucleic acid sequences are generated, and the candidate nucleic acid sequences are used as optimization results corresponding to the target species.

2. A cross-species codon optimization method according to claim 1, characterized in that: The constructing current input information based on the basic information and the prediction sequence set includes: Searching for a corresponding species category unit based on the identification information of the target species; The original amino acid sequence, the species category unit, the identification unit, and the predicted sequence set are sequentially combined to obtain the current input information.

3. A cross-species codon optimization method according to claim 1, characterized in that: The step of combining the current input information with a preset prediction strategy to predict and generate a codon vector for the next sequence position includes: Constructing an input result of a current layer decoder based on the current input information and / or an output result of a previous layer decoder; Based on the input result of the current layer decoder, generate an attention output matrix as the output result of the current layer decoder; The next layer decoder is used as the new current layer decoder.

4. A cross-species codon optimization method as claimed in claim 3, characterized in that: The step of generating an attention output matrix based on the input result of the current layer decoder as the output result of the current layer decoder includes: Performing a linear transformation on each word embedding vector in the input result to obtain a transformation result; the transformation result includes: a query vector, a key vector, and a value vector; Applying the rotation encoding matrix to the query vector and the key vector, performing rotation encoding on the query vector and the key vector, and obtaining a new query vector and a new key vector; According to the query vector, key vector and value vector, the attention output matrix is ​​obtained; The attention output matrices of all heads are concatenated and the concatenation result is linearly transformed to obtain the final multi-head attention output, which is used as the output result of the current layer decoder.

5. A cross-species codon optimization method according to claim 1, characterized in that: The output conditions include: the current sequence position is the last sequence position of the original amino acid sequence.

6. A cross-species codon optimization method as claimed in claim 1, characterized in that: Also includes: Screening the candidate sequences to determine the target codon sequence; The step of screening the candidate sequence to determine the target codon sequence comprises: Screening a number of candidate sequences according to sequence screening conditions to obtain a preferred codon sequence; The protein expression level corresponding to the preferred codon sequence is determined, and the preferred codon sequence with the highest protein expression level is selected as the target codon sequence.

7. A cross-species codon optimization system, characterized in that: include: An acquisition module, used to acquire basic information, wherein the basic information includes: an original amino acid sequence of an original species and category information of a target species; A prediction module is used to determine the current sequence position in the original amino acid sequence; construct the current input information based on the basic information and the predicted sequence set; predict and generate the codon vector of the next sequence position in combination with the current input information and a preset prediction strategy; take the next sequence position as the new current sequence position, add the codon vector of the current sequence position to the predicted sequence set, and execute this step repeatedly until the output condition is met; The optimization module generates a plurality of candidate nucleic acid sequences based on the predicted sequence set, and uses the candidate nucleic acid sequences as optimization results corresponding to the target species.

8. An electronic device, wherein: The electronic device includes: processor; and, A memory storing computer executable instructions which, when executed, cause the processor to perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, wherein: The computer-readable storage medium stores one or more programs, and when the one or more programs are executed by a processor, the method of any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images