Codon optimization

Abstract

A technique related to codon optimization is provided. In a method for optimizing nucleic acid sequences expressing proteins in a host, protein subsequence-nucleic acid subsequence pairs are obtained from collected highly expressed protein sequences and their encoding nucleic acid sequences to form a training set. And training a neural machine translation model by using the training set. The neural machine translation model realizes translation from an amino acid sequence to a codon sequence. A protein sequence needing codon optimization is cut into protein subsequences. And translating a protein subsequence from an amino acid sequence to a codon sequence by using the trained neural machine translation model. The translated subsequences are superimposed to merge into a full-length codon sequence. In the process of superimposing to synthesize the codon sequence, according to the frequency or number of synonymous codons corresponding to each amino acid position of the protein sequence, the synonymous codon with the highest occurrence frequency or number is selected as the optimal codon at the position.

Description

Codon optimization

[0001] Cross-references

[0002] This invention claims priority to Chinese patent application No. 202211485518.X filed on November 24, 2022, the entire contents of which are incorporated herein by reference. Technical Field

[0003] The present disclosure relates to protein expression in biotechnology, and more particularly to methods for optimizing nucleic acid sequences for protein expression in a host. Background Art

[0004] 1. Current Status of Codon Optimization Algorithms

[0005] Codon optimization is a technique that improves the efficiency of nucleic acid usage by selecting appropriate synonymous codons for a specific expression host system without changing the amino acid coding sequence. It is very important for increasing the expression of recombinant proteins and recombinant antibodies, as well as the effectiveness of DNA vaccines. These applications and their effects are described in the following documents: Mauro, VP and Chappell, SA (2014) A critical analysis of codon optimization in human therapeutics. Trends Mol. Med., 20, 604–613 and Angov, E. (2011) Codon usage: nature's roadmap to expression and folding of proteins. Biotechnol. J., 6, 650–659, the contents of which are hereby incorporated by reference into this disclosure.

[0006] With the rise of mRNA-based therapies and vaccines, codon optimization can also be used to improve the stability and expression of mRNA and reduce its immunogenicity. Relevant applications are recorded in the following documents: To, KK and Cho, WC (2021) An overview of rational design of mRNA-based therapeutics and vaccines. Expert Opin. Drug Discov., 16, 1307–1317 and Weissman, D. (2015) mRNA transcript therapy. Expert Rev. Vaccines, 14, 265–281, the contents of which are hereby incorporated into the present disclosure by reference.

[0007] Codon optimization algorithms are divided into two categories based on whether they are global optimizations: the first category is to optimize the entire protein sequence as a whole, which is called global optimization, such as GenScript's GenSmart codon optimization tool (see WO2020 / 024917A1, the contents of which are incorporated herein by reference), JCat (see Grote, A., Hiller, K., Scheer, M., Münch, R., B., Hempel, DC and Jahn, D. (2005) JCat: a novel tool to adapt codon usage of a target gene to its potential expression host. Nucleic Acids Res., 33, W526-531, the contents of which are incorporated herein by reference) and CHARMING (see Wright, G., Rodriguez, A., Li, J., Milenkovic, T., Emrich, SJ and Clark, PL (2022) CHARMING: Harmonizing synonymous codon usage to replicate a desired codon usage pattern. Protein Sci., 31, 221–231, the contents of which are incorporated herein by reference); the second category is to use a local fragment of the protein sequence as the optimization unit, That is, after the protein sequence is divided into subsequences by sequence slicing or sliding window, each subsequence is optimized first, and then the optimized subsequences are combined and spliced ​​into the full sequence, which is called local optimization. For example, GeneOptimizer of GeneArt (see Raab, D., Graf, M., Notka, F., T.and Wagner,R.(2010)The GeneOptimizer Algorithm:using a sliding window approach to cope with the vast sequencespace in multiparameter DNA sequence optimization.Syst.Synth.Biol.,4,215–225, the contents of which are incorporated herein by reference) and a codon optimization method based on an immune algorithm (see WO2019 / 020054A1, the contents of which are incorporated herein by reference).

[0008] Codon optimization algorithms can also be divided into two categories based on whether they use an end-to-end (or sequence-to-sequence) model framework. The first category is non-end-to-end optimization methods, which optimize one or more manually extracted feature parameters of the sequence through a combination of heuristic algorithms (such as genetic algorithms, particle swarm algorithms, simulated annealing) or exhaustive methods. The optimization process uses sequence feature parameters to associate protein expression, mRNA stability and other traits, and use this to evaluate the quality of the optimization output results. Common feature parameters include CAI (Codon Adaptation Index), Codon Context, CBI (Codon Bias Index), ENC (Effective Number of Codon), FOP (Frequency of Optimal Codons), CPP (Codon Preference Parameter), tAI (tRNA adaptation index), Hidden Stop Codon (Hidden Stop Codon), and more. Codon), GC content (the content of G bases and C bases), rare codon content, number of mRNA inhibitory regulatory motifs, mRNA secondary structure (mainly including hairpin structure and folding free energy), key codons and mathematical model scoring in machine learning, microRNA binding sites, G4 content (G-quadruplex content), etc. (see Gould, N., Hendy, O. and Papamichail, D. (2014) Computational tools and algorithms for designing customized synthetic genes. Front. Bioeng. Biotechnol., 2, 41, the contents of which are hereby incorporated by reference into the present disclosure); such software includes GenScript GenSmart codon optimization tool (see WO2020 / 024917A1), COOL (see Chin, JX, Chung, BK-S. and Lee, D.-Y. (2014) Codon Optimization OnLine (COOL): a web-based multi-objective optimization platform for synthetic genes design.Bioinformatics, 30, 2210–2212, the contents of which are hereby incorporated by reference into the present disclosure) and GeneArt GeneOptimizer (see Raab, D., Graf, M., Notka, F.,. T.and Wagner,R.(2010)The GeneOptimizer Algorithm:using a sliding window approach to cope with the vast sequence space in multiparameter DNA sequence optimization.Syst.Synth.Biol.,4,215–225)。The second category is an end-to-end optimization method, which is inspired by neural machine translation and does not require manual extraction of sequence features. Instead, it borrows the encoder-decoder algorithm framework of natural language processing to achieve direct corresponding conversion from protein sequence to nucleic acid sequence, that is, the input protein sequence is directly output as the corresponding nucleic acid sequence through the algorithm. Related optimization tools include RNN-based codon optimization (see Goulet, DR, Yan, Y., Agrawal, P., Waight, AB, Mak, AN and Zhu, Y. (2022) Codon Optimization Using a Recurrent Neural Network. J. Comput. Biol. and Jain, R., Jain, A., Mauro, E., LeShane, K. and Densmore, D. (2021) ICOR: Improving codon optimization with recurrent neural networks. bioRxiv, the contents of which are hereby incorporated into the present disclosure by reference), BiLSTM-based codon optimization (see Fu, H., Liang, Y., Zhong, X., Pan, Z., Huang, L., Zhang, H., Xu, Y., Zhou, W. and Liu, Z. (2020) Codon optimization with deep learning to enhance protein expression. Sci. Rep., 10, 1–9, the contents of which are incorporated herein by reference) and ANN-based codon optimization (see Tarakaram, Y., Mounika, Y., Prasanna, YL and Singh, T. (2021) Codon Optimization and Converting DNA Sequence into Protein Sequence using Deep Neural Networks. In. IEEE, pp. 1–5, the contents of which are incorporated herein by reference), etc.

[0009] 2. Neural Machine Translation

[0010] Neural Machine Translation (NMT) is a type of machine translation that uses artificial neural networks for translation. Its performance significantly outperforms statistical machine translation in many translation applications. The prototype of this method was conceived by the IBM TJ Watson Research Center in the late 1980s and was implemented in the first academic paper in 2013. At present, neural machine translation models generally consist of an encoder-decoder framework. Compared with statistical machine translation, the neural machine translation model has a simple structure, does not require feature engineering and translation rule extraction, solves the problems of long-distance memory and complex bi-text word alignment, and the model takes up less memory and supports incremental training (see Mohamed, SA, Elsayed, AA, Hassan, Y. and Abdou, MA (2021) Neural machine translation: past, present, and future. Neural Comput. Appl., 33, 15919–15931, the contents of which are incorporated into this disclosure by reference). The encoder-decoder framework of a neural machine translation model can generally be composed of CNN, RNN, GRU, LSTM, BiLSTM, or it can also integrate an attention mechanism or pre-trained models such as BERT, RoBERTa, XLNet, GPT2, or use the Transformer framework.Commonly used toolboxes for building neural machine translation include OpenNMT, Nmt, XNMT, Nematus, SOCKEYE, Marian, Fairseq, Neural monkey, Tensor2tensot, NMT-KERAS, THUMT, etc. (see Tan, Z., Wang, S., Yang, Z., Chen, G., Huang, X., Sun, M. and Liu, Y. (2020) Neural machine translation: A review of methods, resources, and tools. AI Open, 1, 5–21, Singh, S. P., Kumar, A., Darbari, H., Singh, L., Rastogi, A. and Jain, S. (2017) Machine translation using deep learning: An overview. In. IEEE, pp. 162–167 and Zhu, J., Xia, Y., Wu, L., He, D., Qin, T., Zhou, W., Li, H. and Liu, T.-Y. (2020) Incorporating bert into neural machine translation. ArXiv Prepr. ArXiv200206823, the contents of these documents are hereby incorporated into the present disclosure by reference).

[0011] 3. Ribosome Profiling

[0012] Ribosome profiling (Ribo-seq) technology was invented in 2009. It is a technology that studies the binding of ribosomes to mRNA by using deep sequencing technology. It can be used to detect and comprehensively monitor the process of protein translation in vivo, such as the discovery of ribosome stalling sites, ORF verification, and the study of translation elongation and termination mechanisms (see Brar, GA and Weissman, JS (2015) Ribosome profiling reveals the what, when, where and how of protein synthesis. Nat. Rev. Mol. Cell Biol., 16, 651–664, the contents of which are hereby incorporated by reference into the present disclosure). During the library construction process of Ribo-seq sequencing samples, ribosome-bound fragments (protected mRNA fragments) are generated. During deep sequencing, the main peak of the fragments is generally around 30 nt, with slight differences between different species. For example, the main peak of Ribo-seq fragments in S. cerevisiae is 28-29 nt, while the main peak of Ribo-seq fragments in mammalian cells is 30-31 nt. Some species also have main peaks of Ribo-seq fragments between 15-65 nt (see Jackson, R. and Standart, N. (2015) The awesome power of ribosome profiling. Rna, 21, 652-654, the contents of which are hereby incorporated by reference into this disclosure). This finding shows that the length of mRNA fragments directly occupied (bound) by ribosomal large and small subunits or intact ribosomes is mainly concentrated between 15-66 nt.

[0013] The basic logic behind the neural machine translation model for solving codon optimization problems is to treat protein sequences as one natural language and the encoded nucleic acid sequence as another natural language, and then use the neural machine translation model to complete the end-to-end translation process. However, there are currently several problems:

[0014] (1) Compared with traditional natural language translation, the structural units of protein and DNA sequences are less complex (i.e., there are only 20 common amino acids and only 4 nucleotides) and the sequences are continuous; there is a one-to-many mapping relationship between amino acids and synonymous codons; and there are no complex grammatical rules.

[0015] (2) For a single protein expression host, the corresponding training set (corpus) is small, especially for lower organisms. For example, even if all protein-coding genes of E. coli K12 substr.MG1655 are used, there are only more than 4,300 proteins, which is insufficient as a training set (or corpus) for deep networks.

[0016] (3) The length of protein sequences varies greatly. For example, the smallest protein of E. coli is the K+ transport (KdpF) protein containing only 29 amino acids, while the largest protein is the RNase T protein containing 1538 amino acids. The neural machine translation methods currently used for codon optimization directly use the full protein sequence as input for global optimization. However, the long-term memory effect of the globally optimized neural machine translation model is at risk of gradually decreasing as the length of the protein sequence increases.

[0017] (4) It is not combined with existing biological research results and has no clear biological logical support.

[0018] The accumulation of the above problems has resulted in the existing codon optimization methods based on neural machine translation models not achieving improved effectiveness. For details, see the evaluation results of RNN-based codon optimization methods and the more widely used centralized codon optimization tools in the industry - GenScript's GenSmart codon optimization tool and GeneArt's GeneOptimizer (see Goulet, DR, Yan, Y., Agrawal, P., Waight, AB, Mak, AN and Zhu, Y. (2022) Codon Optimization Using a Recurrent Neural Network. J. Comput. Biol.).

[0019] Therefore, there is a need for a codon optimization technology that can utilize the development results of codon optimization algorithms to overcome the various defects pointed out above.

[0020] Summary of the Invention

[0021] In order to improve the use effect of neural machine translation in codon optimization, the present invention proposes a codon optimization method based on a neural machine translation model and local optimization of a sliding window. The method sets the sliding window size according to the main peak size of the ribosome-bound fragment (protected mRNA fragments) in ribosome profiling / Ribo-seq, slides the sliding window along the protein sequence to generate sub-fragments, and then uses a transformer model containing an attention mechanism (Self-Attention) and positional encoding (Positional Encoding) (see Lin, T., Wang, Y., Liu, X. and Qiu, X. (2021) A survey of transformers. ArXiv Prepr. ArXiv210604554 and Tay, Y., Dehghani, M., Bahri, D. and Metzler, D. (2020) Efficient transformers: A survey. ACM Comput.Surv.CSUR, the contents of which are hereby incorporated by reference into the present disclosure), perform sequence-to-sequence translation of the protein sequence sub-fragments, then superimpose the translated sub-fragment sequences, and based on the synonymous codon frequency corresponding to each amino acid position, preferentially select the high-frequency synonymous codon at the amino acid position, and finally merge the subsequences into the full-length sequence.

[0022] According to the first aspect of the present disclosure, a computer-implemented method for optimizing nucleic acid sequences for expressing proteins in a host is provided. The method may include: obtaining protein subsequence-nucleic acid subsequence pairs based on collected highly expressed protein sequences and their encoding nucleic acid sequences to form a training set; using the training set to train a neural machine translation model, the neural machine translation model implementing translation from amino acid sequences to codon sequences; cutting protein sequences that require codon optimization into protein subsequences; using the trained neural machine translation model to translate the protein subsequences from amino acid sequences to codon sequences; and superimposing the translated subsequences to merge into a full-length codon sequence. Based on the frequency or number of synonymous codons corresponding to each amino acid position in the protein sequence, the synonymous codon with the highest frequency or number is selected as the optimal codon for that position.

[0023] In the method according to the first aspect of the present disclosure, the step of forming a training set may further include: adding padding amino acid sequences at both ends of the protein sequence; adding corresponding padding codons at both ends of the coding nucleic acid sequence; using a sliding window to extract subsequences of the protein sequence and its coding nucleic acid sequence; segmenting the protein subsequences with amino acids as the smallest unit; segmenting the nucleic acid subsequences with codons as the smallest unit; and adding sentence start markers and sentence end markers before and after each protein subsequence and nucleic acid subsequence, respectively, to form a training set.

[0024] In the above steps, preferably, the filling amino acid sequence is N amino acids; the length of the sliding window on the protein sequence is L1 amino acids and L1=N+1, and the sliding step is 1 amino acid; the length of the sliding window on the coding nucleic acid sequence is L2 nucleotides and L2=3×L1, and the sliding step is 3 nucleotides.

[0025] Preferably, the filler amino acid sequence may be N consecutive methionines, or N consecutive tryptophans.

[0026] When the filling amino acid is methionine, the corresponding filling codon is ATG. When the filling amino acid is tryptophan, the corresponding filling codon is TGG.

[0027] Preferably, the length L1 of the sliding window on the protein sequence is 5 to 22 amino acids, and the length L2 of the sliding window on the coding nucleic acid sequence is 15 to 66 nucleotides.

[0028] In the method according to the first aspect of the present disclosure, the neural machine translation model may be a Transformer model.

[0029] In addition, the neural machine translation model can be an artificial neural network machine translation model containing an attention mechanism and position encoding.

[0030] In the method according to the first aspect of the present disclosure, the step of cutting the protein sequence that needs to be codon-optimized into protein subsequences further includes: adding filler amino acid sequences to both ends of the protein sequence that needs to be codon-optimized; and cutting the protein sequence into protein subsequences using a sliding window.

[0031] In the above steps, preferably, the filler amino acid sequence consists of N amino acids; the sliding window is L1 amino acids long, where L1 = N + 1. By adding the filler amino acid sequence, each amino acid in the protein sequence is scanned the same number of times by the sliding window, and each amino acid is included in the L1 protein subsequences after segmentation. During the step of translating the protein subsequences from amino acid sequences to codon sequences, the amino acid at each position in the protein sequence is translated L1 times by the trained neural machine translation model.

[0032] Preferably, the filler amino acid sequence is N consecutive methionines, or N consecutive tryptophans.

[0033] Similar to the step of forming a training set, in the step of cutting the protein sequence to be codon-optimized into subsequences, preferably, the length L1 of the sliding window is 5 to 22 amino acids.

[0034] In the method according to the first aspect of the present disclosure, in the step of superimposing the translated subsequences to merge into a full-length codon sequence, the synonymous codon with the highest frequency or number of occurrence for each amino acid in the protein sequence after L1 translations in the L1 subsequences is selected as the optimal codon for the amino acid position, thereby a sequence consisting of the optimal codons for all amino acids is used as a synthesized full-length codon sequence.

[0035] In the method according to the first aspect of the present disclosure, for an amino acid position, when two or more synonymous codons appear with the same frequency or number, the optimal codon for the position is selected based on at least one of the following criteria: codon bias, GC%, and folding energy ranking of nucleic acid subfragments.

[0036] The method according to the first aspect of the present disclosure may further comprise: performing synonymous codon replacement on harmful characteristic sequences in the expression host, enzyme cutting sites in molecular cloning operations, or regions that may produce variable splicing sites.

[0037] According to a second aspect of the present disclosure, a non-transitory computer-readable storage medium is provided for storing a computer program, wherein the computer program includes instructions that, when executed by a processor of an electronic device, cause the electronic device to implement the method according to the first aspect of the present disclosure.

[0038] According to a third aspect of the present disclosure, a system for optimizing a nucleic acid sequence for protein expression in a host is provided. The system comprises: a processor; a memory; and a computer program. The computer program is stored in the memory and configured to be executed by the processor, the computer program comprising instructions for implementing the method according to the first aspect of the present disclosure.

[0039] According to a fourth aspect of the present disclosure, an electronic device for optimizing a nucleic acid sequence for expressing a protein in a host is provided, characterized in that the device comprises a tool for implementing the method according to the first aspect of the present disclosure.

[0040] According to a fifth aspect of the present disclosure, a computer program product stored on a recordable medium for optimizing a nucleic acid sequence for protein expression in a host is provided, the computer program product comprising computer software for implementing the method according to the first aspect of the present disclosure.

[0041] According to a sixth aspect of the present disclosure, an isolated nucleic acid molecule is provided, which comprises the optimized nucleic acid sequence obtained by the method according to the first aspect of the present disclosure.

[0042] According to a seventh aspect of the present disclosure, a vector is provided, comprising the isolated nucleic acid molecule according to the sixth aspect of the present disclosure.

[0043] According to an eighth aspect of the present disclosure, a recombinant host cell is provided, comprising the isolated nucleic acid molecule according to the sixth aspect of the present disclosure or the vector according to the seventh aspect of the present disclosure.

[0044] According to a ninth aspect of the present disclosure, a method for expressing a protein in a host cell is provided. The method comprises: obtaining an optimized nucleic acid sequence for expressing the protein in the host cell using the method according to the first aspect of the present disclosure; synthesizing a nucleic acid molecule comprising the optimized nucleic acid sequence; introducing the nucleic acid molecule into the host cell to obtain a recombinant host cell; and culturing the recombinant host cell under conditions that allow expression of the protein from the optimized nucleic acid sequence.

[0045] According to the codon optimization technology disclosed in the present invention, the complexity of neural machine translation is reduced and the effectiveness is improved through continuous local optimization. For example, the neural machine translation model of the transformer or its variant uses an attention mechanism and position encoding to improve the translation effect. In the process of local optimization, the size of the sliding window is formulated based on the conclusions of Ribo-Seq, the interaction between ribosomes and mRNA in the biological process of protein synthesis is fitted, and a sliding window is used to generate subsequences. Finally, the optimal codon is selected based on the number / frequency of synonymous codons appearing at a certain position. This step essentially integrates the process of local folding and local unfolding of mRNA, as well as the codon usage preference, double codon usage preference and other features involved in the process. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention will be more fully understood from the following detailed description taken in conjunction with the accompanying drawings, in which like elements are numbered in a similar manner, and in which:

[0047] FIG1 is a flow chart of a method for optimizing a nucleic acid sequence for expressing a protein in a host according to an embodiment of the present invention.

[0048] FIG2 is a schematic diagram of an embodiment of the method of FIG1 during implementation. When the sliding window is 3 amino acids, 2 M's are added at both ends of the sequence.

[0049] FIG3 is a schematic diagram of an implementation method used in the implementation of the method of FIG1 , where when the sliding window is 3 amino acids, the sliding step is 1, the sequence is read sequentially, and each amino acid is read 3 times.

[0050] FIG4 is a schematic diagram of the operation of the converter framework.

[0051] FIG5 is a graph showing the expression levels of the M2-1 protein obtained by different codon optimization methods. DETAILED DESCRIPTION

[0052] Unless otherwise defined, technical and scientific terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0053] The technical solutions of the present invention will be further described in detail below by way of examples and in conjunction with the accompanying drawings. Unless otherwise stated, the methods and materials of the embodiments described below are all conventional products that can be purchased on the market. Those skilled in the art will appreciate that the methods and materials described below are merely exemplary and should not be construed as limiting the scope of the present invention.

[0054] The present invention provides a codon optimization method based on a neural machine translation model and local optimization of a sliding window.

[0055] FIG1 is a flow chart of a method for optimizing a nucleic acid sequence for expressing a protein in a host according to an embodiment of the present invention.

[0056] As shown in FIG1 , a method 100 for optimizing a nucleic acid sequence for protein expression in a host according to an embodiment of the present invention begins at step S110. In step S110, protein subsequence-nucleic acid subsequence pairs are obtained based on the collected highly expressed protein sequences and their encoding nucleic acid sequences to form a training set. This step corresponds to the stage of preparing a training set (i.e., a corpus) for implementing the method of the present invention, which will be described in detail below.

[0057] Stage 1: Prepare the training set (i.e. corpus)

[0058] First, collect highly expressed protein sequences and their encoding nucleic acid sequences. For example, you can collect highly expressed protein sequences and their encoding nucleic acid sequences based on the proteome, transcriptome, and genome analysis results and annotation information of the protein expression host. For each highly expressed protein sequence and its encoding nucleic acid sequence, perform the following processing.

[0059] (1) Add padding amino acid sequences at both ends of the protein sequence, and add corresponding padding codons at both ends of the coding nucleic acid sequence.

[0060] Specifically, N padding amino acids can be added at both ends of the protein sequence. The padding amino acid sequence can be randomly composed of N amino acids, generally N consecutive M (methionine / methionine) or W (tryptophan). Here, those skilled in the art should recognize that the purpose of padding amino acids at the beginning and end of the protein sequence is to allow the first and last amino acids in the sequence to be read multiple times (the same number of times as other amino acids in the sequence). M or W is filled because it corresponds to the start codon and can promote high expression. Accordingly, corresponding padding codons are added at both ends of the encoding nucleic acid sequence, that is, M corresponds to ATG and W corresponds to TGG.

[0061] (2) Use sliding to extract subsequences of protein sequences and their encoding nucleic acid sequences.

[0062] Subsequence extraction of protein sequences is performed using a sliding window. The length of the sliding window on the protein sequence is L1, typically 5-22 amino acids. It is important to note that under natural conditions, the ribosome sliding window length is 5. A training set can be generated if the sliding window length is greater than 2, preferably greater than 7. The sliding window length of 5-22 is chosen based on the number of amino acids corresponding to the main length of the ribosome-bound fragment in Ribo-seq from different species. The relationship between the sliding window length and the number of padded amino acids is L1 = N + 1. The sliding step length for subsequence extraction of protein sequences is 1 amino acid. Simultaneously, subsequence extraction of coding nucleic acid sequences is performed using a sliding window. The sliding window length on the coding nucleic acid sequence is L2 nucleotides, typically 15-66 nucleotides, which is three times the sliding window size L1 of the protein sequence, i.e., L2 = 3 × L1. The sliding step length for subsequence extraction of coding nucleic acid sequences is 3 nucleotides.

[0063] More generally, the sliding window size (i.e., length) is consistent with the main length of the ribosome-bound fragment in ribosome profiling (Ribo-seq), and is also the size of the main region where the ribosome interacts with mRNA in protein biosynthesis; accordingly, the sliding step size is consistent with the protein biosynthesis process in which the ribosome slides along the mRNA for a single codon length during protein translation. Here, the sliding of the sliding window is actually used to imitate the process of ribosomes sliding along mRNA during protein translation. Therefore, a sliding window of a similar ribosome occupancy size is selected, which is more biomimetic. The AI ​​algorithm of the present invention has thus achieved better performance than other AI algorithms.

[0064] (3) Segment the protein subsequences with amino acids as the smallest unit and segment the nucleic acid subsequences with codons as the smallest unit.

[0065] (4) A sentence start marker and a sentence end marker are added before and after each protein subsequence and nucleic acid subsequence to form a training set (i.e., corpus).

[0066] Figure 2 illustrates the operations of the above-mentioned stages. Specifically, Figure 2 is a schematic diagram of the implementation method used in the implementation process of the method of Figure 1. When the length of the sliding window on the protein sequence is 3 amino acids, 2 M are added to both ends of the sequence. Assuming that a protein expression host has a highly expressed protein, its protein sequence is ESDNTE, and its encoding nucleic acid sequence is TCATTCCTTTTTCTGTCA, let N = 2, the padding amino acid is M, the length of the sliding window on the protein sequence is L1 = 3, the length on the encoding nucleic acid sequence is L2 = 9, the sentence start mark is "^", and the sentence end mark is "$", after sequence padding, sliding window scanning, word segmentation, and special mark addition, 8 protein subsequence-nucleic acid subsequence pairs can be obtained and added to the training set (corpus).

[0067] Next, still referring to FIG. 1 , the method proceeds to step S120 . In step S120 , the neural machine translation model is trained using the training set. The neural machine translation model translates from amino acid sequences to codon sequences. Specifically, see the following detailed description of model training and end-to-end translation.

[0068] Phase 2: Model training and end-to-end translation

[0069] According to a preferred embodiment of the present invention, in the process of model training and subsequent model application, i.e., translating protein sequences, the neural machine translation model can use a Transformer model. Transformer model (see Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, AN, Kaiser, and Polosukhin, I. (2017) Attention is all you need. Adv. Neural Inf. Process. Syst., 30, the contents of which are hereby incorporated into the present disclosure by reference) and its variants have been widely used in the fields of computer vision, video, text, and speech (see Xu, Y., Wei, H., Lin, M., Deng, Y., Sheng, K., Zhang, M., Tang, F., Dong, W., Huang, F. and Xu, C. (2022) Transformers in computational visual media: A survey. Comput. Vis. Media, 8, 33–62, AMand Andonie,R.(2020)Visualizing transformers for nlp:a brief survey.In.IEEE,pp.270–279 and Khan,S.,Naseer,M.,Hayat,M.,Zamir,SW,Khan,FSand Shah,M.(2021)Transformers in vision:A survey.ACM Comput.Surv.CSUR, the contents of these documents are incorporated into the present disclosure by reference), the operation of this stage can directly use the classic transformer model containing multi-head attention mechanism and positional encoding (see Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,AN,Kaiser, and Polosukhin, I. (2017) Attention is all you need. Adv. Neural Inf. Process. Syst., 30) trains the protein subsequence-nucleic acid subsequence corpus (training set) prepared in stage one to achieve direct end-to-end (end-to-end) or sequence-to-sequence translation of amino acid sequence to codon sequence. This step can also be integrated with pre-trained models such as Bert, GPT / GPT-2 / GPT-3 or modeled using transformer variants (X-formers) (see Lin, T., Wang, Y., Liu, X. and Qiu, X. (2021) A survey of transformers. ArXiv Prepr. ArXiv210604554). More generally, the neural machine translation model can be an artificial neural network machine translation model containing an attention mechanism and positional encoding. The concepts of fusion and pre-training models are mentioned here. Regarding fusion and pre-trained models, for example, the encoders of the protein model ProtBert (see link: https: / / pubmed.ncbi.nlm.nih.gov / 34232869 / ) and the DNA model DNABert (see link: https: / / academic.oup.com / bioinformatics / article / 37 / 15 / 2112 / 6128680) can be used to extract features from the input and output, respectively, and use them as input and output embeddings. For example, see the introduction in the following link: https: / / arxiv.org / abs / 2002.06823. The contents of these introductions and knowledge are hereby incorporated into this disclosure by reference.

[0070] Next, continue to refer to Figure 1, the method proceeds to steps S130, S140 and S150. In step S130, the protein sequence that needs to be codon optimized is cut into protein subsequences. After the translation tool (trained neural machine translation model) is prepared in the previous stage, in step S140, the trained neural machine translation model is utilized to translate the protein subsequence from amino acid sequence to codon sequence. In step S150, the translated subsequences are superimposed to merge into a full-length codon sequence. In the process of superimposing the subsequences to merge into a codon sequence, according to the frequency or number of synonymous codons corresponding to each amino acid position of the protein sequence, the synonymous codon with the highest frequency or number is selected as the optimal codon for the position. Specifically, referring to the following detailed description of the translation process of the protein sequence.

[0071] Stage 3: Protein sequence translation

[0072] (1) Add filler amino acid sequences at both ends of the protein sequence that needs to be codon optimized.

[0073] N padding amino acids are added to each end of the protein sequence to be codon-optimized. The padding amino acid sequence can be a random sequence of N amino acids. As in the training process, this can be a continuous sequence of N M (methionine / methionine) or W (tryptophan). The reason for choosing M or W is that most highly expressed protein sequences contain M or W.

[0074] (2) Use a sliding window to cut the protein sequence into protein subsequences.

[0075] During the cleavage process, generally, N=L1-1, i.e., the length of the sliding window is L1 amino acids and L1=N+1. The sliding step is still 1 amino acid. The length of the sliding window L1 can be 5 to 22 amino acids.

[0076] FIG3 is a schematic diagram of an embodiment of the method of FIG1 during implementation. When the sliding window is 3 amino acids long, the sliding step is 1, and the sequence is read sequentially, with each amino acid read 3 times. That is, as shown in FIG3 , N = 2 and L1 = 3 amino acids. In the same manner as the training corpus processing, the protein sequence of length n is cut into protein subsequences by the sliding window. By padding the amino acid sequence, each amino acid in the protein sequence can be scanned the same number of times by the sliding window, that is, each amino acid will be included in the L1 cut protein subsequences.

[0077] As with the training process, the sliding window length is generally consistent with the main length of the ribosome-bound segment in ribosome profiling. Correspondingly, the sliding step size of the sliding window is consistent with the protein biosynthesis process during protein translation, where the ribosome slides along the mRNA by a single codon.

[0078] (3) Using the trained neural machine translation model, the protein subsequence is translated from amino acid sequence to codon sequence.

[0079] Since each amino acid will be included in L1 protein subsequences after cutting, the amino acid at each position in the protein sequence will be translated L1 times by the neural machine translation model trained in step S120.

[0080] (4) Superimpose the translated subsequences to merge into the full-length codon sequence.

[0081] Finally, after L1 translations of an amino acid, the synonymous codon with the highest frequency (or number) is selected as the optimal codon for that position. In other words, based on the frequency or number of synonymous codons corresponding to each amino acid position in the protein sequence, the synonymous codon with the highest frequency or number is selected as the optimal codon for that position.

[0082] That is, in the process of superimposing the translated subsequences to merge into the full-length codon sequence, the synonymous codon with the highest frequency or number of occurrences for each amino acid in the protein sequence after L1 translations in the L1 subsequences is selected as the optimal codon for the amino acid position, thereby taking the sequence composed of the optimal codons for all amino acids as the synthesized full-length codon sequence.

[0083] For a particular amino acid position, when two or more synonymous codons occur with the same frequency or number, selection can be based on other screening criteria, such as codon bias, GC%, folding energy ranking of nucleic acid fragments, etc. For example, the downstream application can be considered: if an mRNA vaccine is to be made, a protein with fewer Ts can be selected, or a protein with fewer or more CGs can be considered in conjunction with upstream and downstream proteins.

[0084] In addition, according to a preferred embodiment of the present invention, the above method may further include the following preferred processing mode: that is, for harmful characteristic sequences in the expression host, restriction sites of molecular cloning operations, or regions that may produce variable splicing sites, synonymous codon replacement is performed.

[0085] Stage 4: Filtering of harmful sites and variable splicing sites

[0086] For harmful characteristic sequences (motifs) reported in the expression host, restriction sites for molecular cloning operations, and regions that may produce variable splicing sites, potential risks are reduced or deleted by replacing synonymous codons in such regions of the optimized sequence.

[0087] Beneficial technical effects

[0088] In addition to the beneficial effects mentioned above, through the teachings and examples of this disclosure, those skilled in the art should recognize that: based on the size of the sliding window formulated according to Ribo-Seq, the interaction between ribosomes and mRNA in the biological process of protein synthesis is fitted, subsequences are generated using the sliding window, and the complexity of neural machine translation is reduced and the effectiveness is improved through continuous local optimization. At the same time, the transformer or its variant uses an attention mechanism to improve the translation effect. Finally, the optimal codon is selected based on the number / frequency of synonymous codons appearing at a certain position. This step essentially integrates the process of local folding and local unfolding of mRNA, as well as the codon usage preference, double codon usage preference and other features involved in the process.

[0089] Evaluate

[0090] Data show that the GenSmart Codon Optimizer tool is more effective than other commonly used tools, and the GenSmart Codon Optimizer tool (see WO 2020 / 024917 A1) has performed very well in multiple evaluations of codon optimization algorithms in different host expression systems and different application fields (such as gene cell therapy). For example, see the following documents, the contents of which are incorporated into this disclosure by reference:

[0091] Goulet, DR, Yan, Y., Agrawal, P., Waight, AB, Mak, AN and Zhu, Y. (2022) Codon Optimization Using a Recurrent Neural Network. J. Comput. Biol., Ranaghan, MJ, Li, JJ, Laprise, DM and Garvie, CW (2021) Assessing optimal: inequalities in codon optimization algorithms. BMC Biol., 19, 1–13

[0092] Gao,Z.,Ravendran,S.,Mikkelsen,NS,Haldrup,J.,Cai,H.,Ding,X.,Paludan,SR,Thomsen,MK,Mikkelsen,JGand Bak,RO(2022)A truncated reverse transcriptase enhances prime editing by split AAV vectors.Mol.Ther.

[0093] Dreismann, AK, McClements, ME, Barnard, AR, Orhan, E., Hughes, JP, Lachmann, PJ and MacLaren, RE (2021) Functional expression of complement factor I following AAV-mediated gene delivery in the retina of mice and human cells. Gene Ther., 28, 265–276

[0094] Karaiyan,P.,Chang,CCH,Chan,E.-S.,Tey,BT,Ramanan,RNand Ooi,CW(2022)In silico screening and heterologous expression of soluble dimethyl sulfide monooxygenases of microbial origin in Escherichia coli.Appl.Microbiol.Biotechnol.,106,4523–4537

[0095] Therefore, in this section, E. coli BL-21 (DE3) was used as the expression system, and the expression of protein M2-1 (NCBI Accession No. Q2Y2M2.1, sequence shown in SEQ ID NO: 1, and containing a His tag sequence at the C-terminus) was used as a test case. The sequence optimized using the GenSmart codon optimization tool is shown in SEQ ID NO: 2, and the sequence optimized using the method of the present invention is shown in SEQ ID NO: 3. Protein expression was induced by IPTG at 15°C for 16 hours using both methods, and the codon optimization effects of the GenSmart codon optimization tool and the optimization process of the present invention were compared. Sequence information for the sequences SEQ ID NO: 1, SEQ ID NO: 2, and SEQ ID NO: 3 described herein is provided below.

[0096] The structure of the transformer model is shown in Figure 4, which is consistent with the literature. Its specific description can also be found in the relevant literature cited in the Transformer model above, and will not be repeated here. In the above example, the size of the nucleic acid sliding window is 30 nt, and the size of the protein sliding window is 10 amino acids (for protein sequences). The nucleic acid sliding window slides from the 5' end to the 3' end in steps of 3 nt each time, and the protein moving window moves one amino acid each time from the N-terminus (amino terminus) to the C-terminus (carboxyl terminus); the left and right ends of the subsequence are: protein subsequences are filled with 9 M, and nucleic acid subsequences are filled with 9 ATG. The multi-head module of the attention mechanism of the transformer model is 8 heads, the encoding layer and the decoding layer are both 6 layers, the batch size of each batch during training is 128, the training period (epoch) is 500, the word embedding vector length is 512, the drop ratio is 0.1, the size of the fully connected layer is 2048, the activation function is Relu, the eps of the layer normalization is 0.00001, the word tokens of the protein sequence are the abbreviations of 20 amino acids, asterisks (which can represent termination codes) and sentence termination and start markers, and the word tokens of the nucleic acid sequence are 64 codons and sentence termination and start markers. The above are common hyperparameters in AI training models, which will have an impact on the final effect of the model. If the choice is not good, the model architecture may not have a good performance. These parameters are hyperparameters obtained after learning based on the training set. In fact, different parameters can be obtained by inputting different training sets. This is a very good parameter used in this application, but this application is not limited to this parameter.

[0097] After expression, the whole bacterial sample was detected by SDS-PAGE. The results are shown in Figure 5 and the expression levels are shown in the table below.

[0098] Consistent with the results of this example, the algorithm of the present invention has an overall improvement in protein expression compared to the GenSmart codon optimization algorithm in experiments with more than 100 protein expressions.

[0099] Since the protein sequences corresponding to the two plasmids are the same, they are just different coding sequences obtained through different optimization algorithms. Therefore, the improvement in expression level has nothing to do with the protein sequence itself. As mentioned in the background technology, the RNN-based AI codon optimization is only comparable to the effect of GenSmart. According to the above evaluation results, the effect of the present invention has surpassed GenSmart, which indirectly shows that the present invention is better than the RNN-based AI codon optimization.

[0100] The specific expression evaluation experimental process is as follows:

[0101] 1) Conversion

[0102] Remove BL21(DE3) competent cells from the ultra-low temperature freezer and thaw on ice. Add 100 ng of plasmid to the competent cells and mix thoroughly by pipetting gently. Transform two tubes of cells with the M2-1_GS plasmid and the other two tubes with the M2-1_ZQ plasmid. Place on ice for 30 minutes, then heat shock in a water bath at 42°C for 90 seconds. After placing on ice for 3 minutes, add 100 μl of room-temperature LB liquid medium. Incubate on a shaker at 37°C, 200 rpm, for 60 minutes. Mix thoroughly and plate onto kanamycin-resistant plates. Invert the plates and incubate overnight at 37°C.

[0103] 2) Experiment with expression

[0104] Two single clones were selected and inoculated into 4 ml LB tubes containing 50 μg / ml kanamycin. The cells were shaken at 37°C and 200 rpm. When the OD600 reached 0.6-0.8, 0.5 mM IPTG was added to each of the two tubes (one transformed with the M2-1_GS plasmid and the other with the M2-1_ZQ plasmid) (their expression results correspond to lanes 1 and 2 in Figure 4, respectively). The cells were cultured at 15°C for 16 hours. Two tubes (one transformed with the M2-1_GS plasmid and the other with the M2-1_ZQ plasmid) were not induced with IPTG and served as negative controls (their expression results correspond to lanes NC1 and NC2 in Figure 4, respectively).

[0105] 3) Sample preparation and SDS

[0106] After centrifugation, remove 450 μl of the pellet from the culture medium and resuspend it in 300 μl of lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 5% glycerol, pH 8.0). Ultrasonicate for 1 minute. Mix 100 μl of the lysate with 50 μl of 5x loading buffer, heat at 100°C for 10 minutes, and centrifuge at 15,000 rpm for 5 minutes. Protein expression was then analyzed by SDS-PAGE (see Figure 5).

[0107] In addition, those skilled in the art will recognize that the methods disclosed herein can be implemented as computer programs. As described above in conjunction with the accompanying drawings, the methods of the above embodiments are executed by one or more programs, and the instructions in the programs cause a computer or processor to execute the algorithms described in conjunction with the accompanying drawings. These programs can be stored and provided to a computer or processor using various types of non-transitory computer-readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (such as floppy disks, magnetic tapes, and hard disk drives), magneto-optical recording media (such as magneto-optical disks), CD-ROMs (compact disk read-only memories), CD-Rs, CD-R / Ws, and semiconductor memories (such as ROMs, PROMs (programmable ROMs), EPROMs (erasable PROMs), flash ROMs, and RAMs (random access memories)). Furthermore, these programs can be provided to a computer using various types of transient computer-readable media. Examples of transient computer-readable media include electrical signals, optical signals, and electromagnetic waves. Transient computer-readable media can be used to provide programs to a computer via wired communication paths such as wires and optical fibers or wireless communication paths.

[0108] For example, according to one embodiment of the present disclosure, a non-transitory computer-readable storage medium may be provided for storing a computer program, wherein the computer program includes instructions that, when executed by a processor of an electronic device, enable the electronic device to implement the method described above.

[0109] In addition, according to the present disclosure, a system for optimizing a nucleic acid sequence for protein expression in a host may be proposed, the system comprising: a processor; a memory; and a computer program. The computer program is stored in the memory and configured to be executed by the processor. The computer program includes instructions for implementing the method described above.

[0110] In addition, the present invention also relates to an electronic device for optimizing a nucleic acid sequence for protein expression in a host, characterized in that the device comprises means for implementing the method described above.

[0111] It is also possible to propose a computer program product stored on a recordable medium for optimizing a nucleic acid sequence for expression of a protein in a host, said computer program product comprising computer software for implementing the method as described above.

[0112] On the other hand, according to the above disclosure, the present invention may also have the following applications: For example, an isolated nucleic acid molecule may be proposed, which comprises the optimized nucleic acid sequence obtained by the above method.

[0113] A vector may also be proposed, comprising an isolated nucleic acid molecule as described above.

[0114] Furthermore, a recombinant host cell is proposed, which comprises an isolated nucleic acid molecule as described above or a vector as described above.

[0115] In addition, according to one embodiment of the present invention, a method for expressing a protein in a host cell is proposed, the method comprising: (a) obtaining an optimized nucleic acid sequence for expressing the protein in the host cell using the method as described above; (b) synthesizing a nucleic acid molecule comprising the optimized nucleic acid sequence; (c) introducing the nucleic acid molecule into the host cell to obtain a recombinant host cell; and (d) culturing the recombinant host cell under conditions allowing the protein to be expressed from the optimized nucleic acid sequence.

[0116] The embodiments of the present invention are not limited to the above embodiments. Without departing from the spirit and scope of the present invention, ordinary technicians in this field can make various changes and improvements to the present invention in form and details, and these are all considered to fall within the scope of protection of the present invention.

[0117] The sequence information is as follows:

[0118] SEQ ID NO: 1 M2-1 protein

[0119] SEQ ID NO:2 GenSmart optimized sequence

[0120] SEQ ID NO:3 Sequence optimized by the method of the present invention

Claims

1. A computer-implemented method for optimizing a nucleic acid sequence for protein expression in a host, characterized in that: The method comprises: Based on the collected highly expressed protein sequences and their encoding nucleic acid sequences, protein subsequence-nucleic acid subsequence pairs are obtained to form a training set; Using the training set to train a neural machine translation model, wherein the neural machine translation model implements translation from an amino acid sequence to a codon sequence; Cutting the protein sequence that needs to be codon-optimized into protein subsequences; Using the trained neural machine translation model, the protein subsequence is translated from amino acid sequence to codon sequence; The translated subsequences are superimposed to merge into a full-length codon sequence, wherein, based on the frequency or number of synonymous codons corresponding to each amino acid position in the protein sequence, the synonymous codon with the highest frequency or number is selected as the optimal codon for that position.

2. The method according to claim 1, characterized in that The step of forming a training set further includes: Add padding amino acid sequences at both ends of the protein sequence; Add corresponding filling codons at both ends of the coding nucleic acid sequence; Use sliding windows to extract subsequences from protein sequences and their coding nucleic acid sequences; Segment the protein subsequences using amino acids as the smallest unit; Segment the nucleic acid subsequence using codon as the smallest unit; A sentence start marker and a sentence end marker are added before and after each protein subsequence and nucleic acid subsequence to form a training set.

3. The method according to claim 2, characterized in that The filling amino acid sequence is N amino acids, The length of the sliding window on the protein sequence is L1 amino acids and L1=N+1, and the sliding step is 1 amino acid. The length of the sliding window on the coding nucleic acid sequence is L2 nucleotides and L2=3×L1, and the sliding step is 3 nucleotides.

4. The method according to claim 3, characterized in that The filling amino acid sequence is N consecutive methionines or N consecutive tryptophans.

5. The method according to claim 4, characterized in that When the filling amino acid is methionine, the corresponding filling codon is ATG; when the filling amino acid is tryptophan, the corresponding filling codon is TGG.

6. The method according to any one of claims 3 to 5, characterized in that The length L1 of the sliding window on the protein sequence is 5 to 22 amino acids, and the length L2 of the sliding window on the coding nucleic acid sequence is 15 to 66 nucleotides.

7. The method according to any one of claims 1 to 6, characterized in that The neural machine translation model is a transformer model.

8. The method according to any one of claims 1 to 7, characterized in that The neural machine translation model is an artificial neural network machine translation model containing an attention mechanism and position encoding.

9. The method according to any one of claims 1 to 8, characterized in that The step of cutting the protein sequence that needs to be codon-optimized into protein subsequences further includes: Add filler amino acid sequences to both ends of the protein sequence that needs to be codon-optimized; The protein sequence is cut into protein subsequences using a sliding window.

10. The method according to claim 9, characterized in that The filling amino acid sequence is N amino acids, The length of the sliding window is L1 amino acids and L1=N+1, and the sliding step is 1 amino acid. By adding the padding amino acid sequence, each amino acid in the protein sequence is scanned the same number of times by the sliding window, and each amino acid will be included in the L1 protein subsequences after cutting. In the step of translating the protein subsequence from the amino acid sequence to the codon sequence, the amino acid at each position of the protein sequence is translated L1 times by the trained neural machine translation model.

11. The method according to claim 10, characterized in that The filling amino acid sequence is N consecutive methionines or N consecutive tryptophans.

12. The method according to claim 10 or 11, characterized in that The length L1 of the sliding window is 5 to 22 amino acids.

13. The method according to any one of claims 10 to 12, characterized in that In the step of superimposing the translated subsequences to merge into a full-length codon sequence, the synonymous codon with the highest frequency or number of occurrence for each amino acid in the protein sequence after L1 translations in the L1 subsequences is selected as the optimal codon for the amino acid position, thereby a sequence consisting of the optimal codons for all amino acids is used as the synthesized full-length codon sequence.

14. The method according to any one of claims 1 to 13, characterized in that For an amino acid position, when two or more synonymous codons appear at the same frequency or number, the optimal codon for that position is selected based on at least one of the following criteria: codon bias, GC%, and folding energy ranking of nucleic acid subfragments.

15. The method according to any one of claims 1 to 14, characterized in that The method further comprises: Synonymous codon replacement is performed for harmful characteristic sequences in the expression host, restriction sites in molecular cloning operations, or regional sequences that may produce variable splicing sites. 16 . A non-transitory computer-readable storage medium for storing a computer program, the computer program comprising instructions that, when executed by a processor of an electronic device, cause the electronic device to implement the method of claim 1 .

17. A system for optimizing a nucleic acid sequence for protein expression in a host, the system comprising: processor; Memory; and A computer program, wherein the computer program is stored in the memory and configured to be executed by the processor, the computer program comprising instructions for implementing the method of any one of claims 1 to 15.

18. An electronic device for optimizing a nucleic acid sequence for protein expression in a host, characterized in that: The apparatus comprises means for carrying out the method of any one of claims 1-15.

19. A computer program product stored on a recordable medium for optimizing a nucleic acid sequence for protein expression in a host, the computer program product comprising computer software for implementing the method of any one of claims 1 to 15.

20. An isolated nucleic acid molecule comprising the optimized nucleic acid sequence obtained by the method of any one of claims 1-15.

21. A vector comprising the isolated nucleic acid molecule of claim 20.

22. A recombinant host cell comprising the isolated nucleic acid molecule of claim 20 or the vector of claim 21.

23. A method for expressing a protein in a host cell, the method comprising: (a) obtaining an optimized nucleic acid sequence for expressing the protein in the host cell using the method of any one of claims 1 to 15; (b) synthesizing a nucleic acid molecule comprising the optimized nucleic acid sequence; (c) introducing the nucleic acid molecule into the host cell to obtain a recombinant host cell; and (d) culturing the recombinant host cell under conditions that allow expression of the protein from the optimized nucleic acid sequence.

Images