Method and system for optimizing mRNA (messenger ribonucleic acid) non-coding region sequence and electronic equipment

Abstract

The invention discloses an mRNA non-coding region sequence optimization method and system and electronic equipment, and the mRNA non-coding region sequence optimization method comprises the steps: constructing an initial candidate library according to a target protein; inputting the initial candidate library into a pre-trained mRNA sequence optimization model to obtain a prediction data set; performing multi-dimensional scoring and sequence optimization on the prediction data set to obtain a sequence recommendation group; performing biological verification on the sequence recommendation group to obtain an optimized mRNA sequence; wherein the prediction data set comprises a sequence ID, a sequence content, a prediction TE score and a confidence interval. According to the method, the translation efficiency of the mRNA sequence can be efficiently and accurately predicted, the candidate sequence with high expression potential is screened out, meanwhile, the consumption of computing resources is reduced, and the overall design cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioinformatics and computer cross, and particularly relates to an mRNA non-coding region sequence optimization method, an mRNA non-coding region sequence optimization system and an electronic device. BACKGROUND

[0002] mRNA therapy and vaccine show great potential in the fields of infectious disease prevention, tumor immunotherapy, etc. The non-coding region (Untranslated Region, UTR) of mRNA does not encode protein, but its sequence characteristics determine the secondary / tertiary structure of mRNA, which in turn deeply affects the stability, half-life and ribosome translation efficiency of mRNA. Among them, the computational design and optimization of mRNA sequence itself is the key step to determine the expression efficiency, stability and safety, and has important influence on the final effect of vaccine.

[0003] However, the mRNA sequence design in the related art has the following problems: 1. The existing sequence design often relies on the folding algorithm based on thermodynamics (such as RNAfold, Mfold). These methods mainly calculate the minimum free energy structure, have high computational complexity, are difficult to handle long sequences, and often ignore the complex biophysical environment in cells, resulting in poor correlation between the predicted structure and the real function in vivo. 2. The conventional CNN (Convolutional Neural Network, convolutional neural network) or RNN (Recurrent Neural Network, recurrent neural network) model lacks the ability to capture long-distance dependencies. For example, the RNA model based on BERT (Bidirectional Encoder Representations from Transformers, bidirectional Transformer encoder representation) usually regards RNA as a pure text sequence, ignoring the unique "base pairing" physical constraints of RNA. This leads to the sequence generated by the model to conform to the syntax, but it is difficult to fold into a stable functional structure in physics, and the prediction accuracy (especially the prediction of translation efficiency MRL (Mean Ribosome Load, average ribosome load)) is difficult to meet the industrial-level design requirements. SUMMARY

[0004] The present application is proposed to solve at least one of the above problems. According to a first aspect of the present application, there is provided an mRNA non-coding region sequence optimization method, comprising: constructing an initial candidate library according to a target protein; inputting the initial candidate library into a pre-trained mRNA sequence optimization model to obtain a prediction data set; performing multi-dimensional scoring and sequence optimization on the prediction data set to obtain a sequence recommendation group; and performing biological verification on the sequence recommendation group to obtain an optimized mRNA sequence; wherein the prediction data set comprises a sequence ID, sequence content, a predicted TE score and a confidence interval.

[0005] In an embodiment of the present application, the step of constructing an initial candidate library according to a target protein comprises: obtaining a number a of variant sequences according to the target protein by a random mutation strategy; obtaining a number b of optimized sequences according to the target protein by an evolutionary optimization strategy; merging and deduplicating the number a of variant sequences and the number b of optimized sequences to obtain a number c of unique sequences, and obtaining the initial candidate library according to the number c of unique sequences.

[0006] In an embodiment of the present application, the step of obtaining a number a of variant sequences according to the target protein by a random mutation strategy comprises: obtaining a reference sequence corresponding to the target protein, taking the reference sequence as a template, setting a mutation rate d, performing random base replacement for a first preset number of rounds to generate the number a of variant sequences; and the step of obtaining a number b of optimized sequences according to the target protein by an evolutionary optimization strategy comprises: selecting a number k of UTR sequences as an initial population according to the target protein, taking the pre-trained mRNA sequence optimization model as a fitness function to perform genetic algorithm iteration for a second preset number of rounds to generate the number b of optimized sequences.

[0007] In an embodiment of the present application, the mRNA sequence optimization model adopts a deep neural network based on a Transformer architecture as a backbone, embeds a pair-wise position bias matrix into a multi-head self-attention mechanism, and injects structure prior into the model by when calculating attention scores; wherein, and respectively represent the position index of the th and the th nucleotide in the input mRNA sequence, represents the query vector of position in the th attention head, the key vector of position in the th attention head, represents the dimension of the key vector, Indicates pre-calculated, in position and The training process of the mRNA sequence optimization model includes: collecting multi-source transcriptome data, including ribosomal imprint data, synthetic UTR library, and viral mRNA; cleaning the multi-source transcriptome data; grouping the cleaned data according to the mRNA sequence length range and GC content range, and performing Z-score normalization on the translation efficiency index within each group to obtain a validation set; obtaining the total loss function of the current training cycle by weighted summation based on the main task loss of translation efficiency prediction, the auxiliary task loss of secondary structure feature prediction, and the auxiliary task loss of stability score prediction, and judging whether the preset training termination condition is met based on the performance of the validation set; if the training termination condition is not met, the gradient is calculated using the total loss function, the parameters of the pre-inserted low-rank adaptive matrix in the model are updated and adjusted, the main parameters of the model are frozen, and the model with adjusted parameters is used for training in the next training cycle; if the training termination condition is met, the early stopping mechanism is triggered to stop training, and the model parameter state corresponding to the optimal moment of the validation set evaluation index is determined as the trained mRNA sequence optimization model.

[0008] In one embodiment of the present invention, before inputting the initial candidate library into the pre-trained mRNA sequence optimization model, the method further includes: screening out target sequences in the initial candidate library with a length range of 30~100bp and containing only A, U, G, and C bases; converting the target sequences into Token IDs and packaging them according to Batch Size = m; and inputting the packaged data into the pre-trained mRNA sequence optimization model.

[0009] In one embodiment of the present invention, the step of performing multi-dimensional scoring and sequence selection on the prediction dataset includes: performing initial screening on the prediction dataset; performing comprehensive scoring on each remaining sequence after initial screening, with each remaining sequence corresponding to a score; sorting the scores of each remaining sequence from largest to smallest; and using the top n sorted sequences as the sequence recommendation group.

[0010] In one embodiment of the present invention, the sequence is comprehensively scored using the following formula:

[0011]

[0012] in, , , , , Indicates preset parameters. This indicates the predicted TE score; This represents the average normalized Hamming distance between the current sequence and the remaining sequences; This represents the normalized value calculated based on the minimum free energy; This indicates the degree of nonspecific matching between the current sequence and known key gene sequences in the human transcriptome; This indicates the difficulty level of chemical synthesis.

[0013] In one embodiment of the present invention, the step of biologically validating the recommended sequence set to obtain an optimized mRNA sequence includes: synthesizing a DNA template containing a T7 promoter, a target 5'-UTR, a protein CDS, and a universal 3'-UTR using the gBlock systemic chemical synthesis method based on each sequence in the recommended set; using the DNA template as a transcription template, preparing initial mRNA using T7 RNA polymerase, and performing capping and tailing treatments to obtain the target mRNA; transfecting the target mRNA into target cells and culturing for a preset time; and detecting the protein expression level of the target cells using flow cytometry.

[0014] According to a second aspect of the present invention, an mRNA non-coding region sequence optimization system is provided. The mRNA non-coding region sequence optimization system includes: a construction module for constructing an initial candidate library based on a target protein; an optimization module for inputting the initial candidate library into a pre-trained mRNA sequence optimization model to obtain a prediction dataset; a screening module for performing multi-dimensional scoring and sequence optimization on the prediction dataset to obtain a sequence recommendation group; and a validation module for performing biological validation on the sequence recommendation group to obtain optimized mRNA sequences; wherein the prediction dataset includes a sequence ID, sequence content, prediction TE score, and confidence interval.

[0015] According to a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, wherein when the computer program is executed by the processor, it implements any of the above-described mRNA non-coding region sequence optimization methods.

[0016] According to the mRNA non-coding region sequence optimization method, system, and electronic device provided in the embodiments of the present invention, the mRNA non-coding region sequence optimization method of the present invention first obtains a prediction dataset through a trained mRNA sequence optimization model, then selects from the prediction dataset to obtain a sequence recommendation group containing multiple sequences, and finally selects the required mRNA sequence through biological verification. It can efficiently and accurately predict the translation efficiency of mRNA sequences and screen out candidate sequences with high expression potential. At the same time, it reduces the consumption of computing resources and lowers the overall design cost. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for optimizing mRNA non-coding region sequences according to an embodiment of the present invention.

[0019] Figure 2 A schematic diagram showing the comparison of experimental results provided in an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of correlation analysis provided in an embodiment of the present invention;

[0021] Figure 4 This is a schematic diagram of the structure of an mRNA non-coding region sequence optimization system provided in an embodiment of the present invention;

[0022] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0024] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0025] It should be understood that the invention can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0026] To fully understand this invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by this invention. Optional embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.

[0027] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0028] To improve the accuracy of mRNA translation efficiency prediction and reduce computational resource consumption, the first aspect of this invention provides a method for optimizing mRNA non-coding region sequences, such as... Figure 1 As shown, methods for optimizing mRNA non-coding region sequences include:

[0029] S1. Construct an initial candidate library based on the target protein.

[0030] As an example, considering the characteristics and expression requirements of the target protein, a strategy combining random mutation and evolutionary search can be adopted to construct an initial candidate library based on known efficient motifs (such as Kozak sequences), and then iteratively optimize the sequence space using a genetic algorithm. A motif is a short, conserved sequence fragment or spatial structural unit with a specific function in biological macromolecules (nucleic acids, proteins).

[0031] S2. Input the initial candidate library into the pre-trained mRNA sequence optimization model to obtain the prediction dataset.

[0032] As an example, a high-quality validation set can be used to fine-tune the pre-trained model. To address the issues of high memory consumption and overfitting during full fine-tuning, a low-rank adaptation (LoRA) strategy can be employed.

[0033] The prediction dataset includes sequence IDs, sequence content, prediction TE (Translation Efficiency) scores, and confidence intervals.

[0034] S3. For the prediction dataset, perform multi-dimensional scoring and sequence optimization to obtain the sequence recommendation group.

[0035] As an example, a multi-dimensional scoring system and an optimal selection algorithm can be built to process the prediction dataset, selecting the top 50 sequences as the sequence recommendation group. The multi-dimensional scoring system can include dimensions such as translation efficiency, stability, and synthesis feasibility.

[0036] S4. Biological validation of the recommended sequence group was performed to obtain the optimized mRNA sequence.

[0037] For example, this can be verified through a wet test.

[0038] In addition, if the verification fails, the optimized mRNA sequence is fed back to the mRNA sequence optimization model for fine-tuning, and step S2 is repeated.

[0039] The mRNA non-coding region sequence optimization method of this invention first obtains a prediction dataset through a trained mRNA sequence optimization model, then selects from the prediction dataset to obtain a sequence recommendation group containing multiple sequences, and finally selects the required mRNA sequence through biological verification. This method can efficiently and accurately predict the translation efficiency of mRNA sequences and screen out candidate sequences with high expression potential. At the same time, it reduces the consumption of computing resources and lowers the overall design cost.

[0040] In some embodiments, the step of constructing an initial candidate library based on the target protein includes:

[0041] S11, based on the target protein, obtain a variant sequences through a random mutation strategy.

[0042] Specifically, the baseline sequence corresponding to the target protein can be obtained. Using the baseline sequence as a template, the mutation rate is set to d, and random base substitutions are performed in the first preset round to generate a variant sequences.

[0043] S12: Based on the target protein, b optimized sequences are obtained through an evolutionary optimization strategy.

[0044] Specifically, based on the target protein, k UTR sequences are selected as the initial population, and a pre-trained mRNA sequence optimization model is used as the fitness function for a second preset round of genetic algorithm iteration to generate b optimized sequences.

[0045] S13, merge and deduplicate a variant sequences and b optimized sequences to obtain c unique sequences, and obtain the initial candidate library based on c unique sequences.

[0046] In this embodiment, by combining extensive exploration of random mutations with AI-guided directed evolution, not only is sequence diversity guaranteed but optimization efficiency is also improved. At the same time, by using a deduplication strategy, redundant sequences are effectively avoided, reducing the waste of computing resources and ensuring the quality and coverage of the initial candidate library.

[0047] In some embodiments, the mRNA sequence optimization model employs a deep neural network based on the Transformer architecture as its backbone, embedding pairwise position bias matrices (PPBs) into a multi-head self-attention mechanism, and calculating the attention score using... This involves injecting structural priors into the model.

[0048] It should be noted that the PPB matrix is ​​pre-calculated based on RNA base pairing rules (Watson-Crick pairing and Wobble pairing), and the matrix dimensions can be represented as (seq_len, seq_len, num_heads).

[0049] It is worth mentioning that in the calculation of attention scores, through Introducing structural priors in this way can assign higher attention weights to potential paired bases (such as AU, GC), enabling the model to perceive the non-Euclidean spatial structural features of mRNA molecules.

[0050] in, and These represent the nth mRNA sequence in the input mRNA sequence. The and the first A position index of nucleotides, Indicates the first A position in the attention. The query vector, In the A position in the attention. The key vector, This represents the dimension of the key vector. Indicates pre-calculated, in position and The pairwise positional offset values ​​between them.

[0051] In this embodiment, by embedding prior knowledge of RNA secondary structure into the Transformer model, the model's ability to perceive the spatial conformation of mRNA molecules is significantly improved; and through the pairwise position bias mechanism, the model can accurately identify functional base pairing patterns, greatly improving the accuracy of translation efficiency prediction and biological interpretability.

[0052] In some embodiments, the training process of the mRNA sequence optimization model includes:

[0053] S211 collects multi-source transcriptome data, including ribosomal imprint data, synthetic UTR libraries, and viral mRNA.

[0054] S212 is used to clean multi-source transcriptome data.

[0055] As an example, data cleaning includes removing sequences containing non-standard bases, abnormal lengths (<30bp or >500bp), and low complexity (such as high entropy).

[0056] S213. The cleaned data were grouped according to the mRNA sequence length range and GC content range, and the translation efficiency index within each group was normalized by Z-score to obtain the validation set.

[0057] Specifically, a stratified standardization strategy can be adopted for data from different species and experimental batches. This involves grouping data based on mRNA sequence length ranges and GC (Guanine-Cytosine) content ranges, and then performing Z-score normalization on the translation efficiency index within each group to eliminate the confounding effects of length and GC content on the translation efficiency baseline.

[0058] S214. Based on the main task loss predicted by translation efficiency, the auxiliary task loss predicted by secondary structure features, and the auxiliary task loss predicted by stability score, the total loss function of the current training cycle is obtained by weighted summation, and the training termination condition is determined based on the performance on the validation set.

[0059] S215 If the training termination condition is not met, the gradient is calculated using the total loss function, the parameters of the pre-inserted low-rank adaptive matrix in the model are updated and adjusted, the main parameters of the model are kept frozen, and the model with adjusted parameters is used for training in the next training cycle.

[0060] It should be noted that the low-rank adaptive strategy freezes the main parameters of the pre-trained model and only introduces and updates the low-rank decomposition matrix (e.g., Rank=8) in the Q, K, and V projection matrices of the Transformer layer and the feedforward network, reducing the number of trainable parameters to about 2% of the original model.

[0061] S216 If the training termination condition is met, the early stopping mechanism is triggered to stop training, and the model parameter state corresponding to the optimal moment of the validation set evaluation index is determined as the trained mRNA sequence optimization model.

[0062] In this embodiment, the training process eliminates the confusion caused by sequence length and GC content on translation efficiency through hierarchical standardization of multi-source transcriptome data, thereby improving data quality. Furthermore, by combining multi-task learning with low-rank adaptive fine-tuning strategies, the training cost is reduced while maintaining the stability of the main structure of the model. This enables the model to comprehensively learn multi-dimensional biological characteristics such as translation efficiency, secondary structure, and stability, thereby obtaining an optimized model with high generalization and high prediction accuracy.

[0063] In some embodiments, before inputting the initial candidate library into the pre-trained mRNA sequence optimization model, the mRNA non-coding region sequence optimization method further includes:

[0064] S221, select target sequences from the initial candidate library that are 30~100bp in length and contain only A, U, G, and C bases.

[0065] S222, convert the target sequence into a Token ID and pack it according to Batch Size = m.

[0066] It should be noted that Token ID refers to converting each base (A, U, G, C) in the mRNA sequence into a corresponding numerical identifier, enabling the model to process sequence data; Batch Size indicates the number of sequences input into the model for processing each time.

[0067] S223, input the packaged data into the pre-trained mRNA sequence optimization model.

[0068] In this embodiment, by preprocessing the initial candidate library, the data processing speed can be improved and the computational efficiency can be increased.

[0069] In some embodiments, the steps of performing multi-dimensional scoring and sequence optimization on the prediction dataset include:

[0070] S31, perform initial screening and filtering on the prediction dataset.

[0071] S32, perform a comprehensive score on each of the remaining sequences after the initial screening, with each remaining sequence corresponding to a score.

[0072] As an example, a comprehensive scoring system can be established, which, in addition to the core translation efficiency score, can also introduce sequence diversity to avoid homogenization; and introduce structural stability and off-target risk as penalty items.

[0073] Specifically, the sequence is comprehensively scored using the following formula:

[0074]

[0075] in, , , , , Indicates preset parameters. This indicates the predicted TE score; This represents the average normalized Hamming distance between the current sequence and the remaining sequences; This represents the normalized value calculated based on the minimum free energy; This indicates the degree of nonspecific matching between the current sequence and known key gene sequences in the human transcriptome; This indicates the difficulty level of chemical synthesis.

[0076] S33, sort the scores of each remaining sequence in descending order.

[0077] S34: Select the first n sorted sequences as the recommended sequence group.

[0078] As an example, the top-N preferred sequences that combine high translation efficiency and high stability can be output for experimental verification.

[0079] In this embodiment, the multi-dimensional scoring system avoids the limitations of single-index optimization by balancing key factors such as translation efficiency, sequence diversity, and structural stability; and by introducing penalties such as off-target risk and synthesis difficulty, it improves the experimental feasibility of the recommended sequences.

[0080] In some embodiments, the step of biologically validating the sequence recommendation group to obtain the optimized mRNA sequence includes:

[0081] S41. Based on the sequences in the recommended group, use the gBlock systemic chemical synthesis to synthesize a DNA template containing the T7 promoter, target 5'-UTR, protein CDS, and universal 3'-UTR.

[0082] S42 uses a DNA template as a transcription template, prepares initial mRNA using T7 RNA polymerase, and performs capping and tailing treatments to obtain the target mRNA.

[0083] S43, transfect the target mRNA into the target cells and culture for the preset time.

[0084] S44, the protein expression level of the target cells was detected by flow cytometry.

[0085] Next, this invention will use the optimization design of the 5'-UTR of the mRNA sequence targeting a certain viral Spike protein as an example to illustrate the operation process of this invention:

[0086] Preparation Phase: Model Deployment and Benchmark Establishment. First, the optimization objective was established as maximizing the translation efficiency of the SARS-CoV-2 Spike protein. Next, for the benchmark sequence, the 5'-UTR sequence of the existing BNT162b2 (Pfizer / BioNTech) vaccine was selected as the control group (sequence fragment: GCCGCCGCCGCCACCAUG...), which is known to have approximately 1.5 times the translation efficiency of the wild type. For the coding region sequence, the full-length coding sequence of the Spike protein (3822 bp, encoding 1274 amino acids) was selected. For the computing environment, a server with a single NVIDIA A100 GPU (40GB VRAM) was configured, running the PyTorch 2.0 deep learning framework. For model loading, the pre-trained mRNA sequence optimization model (parameters frozen) was loaded, along with the LoRA weight matrix (Rank=8) and regression prediction head fine-tuned on the mRNA translation efficiency dataset. The model's coefficient of determination in TE prediction was validated on the test set. The value reached 0.725.

[0087] Step A1: High-throughput generation of candidate sequences.

[0088] To explore the vast sequence space, this embodiment combines two strategies—random mutation and evolutionary algorithms—to generate a candidate library.

[0089] Strategy 1 (random mutation): Using a baseline sequence (75bp in length) as a template, a mutation rate of 0.2 is set, and 40 rounds of random base substitutions are performed to generate approximately 400,000 variant sequences.

[0090] Strategy 2 (Evolutionary Optimization): Select 100 known efficient UTR sequences as the initial population, and use the optimization model of this invention as the fitness function to perform 50 rounds of genetic algorithm iteration.

[0091] During the iteration process, the average TE score of the population rapidly increased from 1.2 to 1.6 in the first 10 rounds; it reached 1.9 in the 30th round; and converged to 1.92 in the 50th round. This process generated approximately 50,000 pre-optimized sequences.

[0092] Merging: The sequences generated by the two strategies above are merged and deduplicated to obtain an initial candidate library containing approximately 450,000 unique sequences.

[0093] Step A2: Perform model inference and obtain the prediction dataset.

[0094] Batch translation efficiency prediction was performed on 450,000 unique sequences in the initial candidate library.

[0095] Preprocessing: Verify that all sequences are between 30-100 bp in length and contain only A, U, G, and C bases. Convert the sequences into token IDs and package them according to Batch Size = 512.

[0096] Inference process: Input the data from the initial candidate library into the model. The model automatically calculates the pairwise position bias matrix, extracts features using the LoRA-fine-tuned Transformer layer, and outputs the predicted TE score through the regression head.

[0097] Efficiency statistics: In a single-card A100 environment, processing 450,000 sequences took approximately 53 seconds (an average of approximately 0.06 seconds per batch). Compared to the traditional RNAfold algorithm, this method achieves an order-of-magnitude speedup.

[0098] Output: Obtain the prediction dataset containing sequence ID, sequence content, prediction TE score, and confidence interval.

[0099] Step A3: Multi-dimensional scoring and sequence optimization.

[0100] To ensure that the selected sequences not only have high translation efficiency but also experimental feasibility and safety, a multi-dimensional scoring formula is used for sorting.

[0101] Initial screening: Sequences with a TE score <1.5 (below the baseline), a structural stability score <0.5, or containing highly complex homopolymers (such as AAAAA...) were removed. Approximately 68,000 sequences remained after filtering.

[0102] Composite Score Calculation: Calculate the composite score for the remaining sequences. ):

[0103]

[0104] Final recommendation: Select the Top-100 sequences as the recommendation group based on descending order of CS scores.

[0105] Step A4: Wet experiment verification and result analysis.

[0106] The top-10 sequences with the highest CS scores (numbered Seq#1 to Seq#10) in the sequence recommendation group were selected for biological validation.

[0107] If the verification passes, the optimized mRNA sequence is obtained. If the verification fails, the optimized mRNA sequence is fed back to step A2 to fine-tune the mRNA sequence optimization model.

[0108] Experimental methods:

[0109] i. Sequence synthesis: DNA templates containing the T7 promoter, target 5'-UTR, Spike protein CDS, and universal 3'-UTR were chemically synthesized using the gBlock system.

[0110] ii. In vitro transcription: mRNA was prepared using T7 RNA polymerase and then capped and tailed.

[0111] iii. Cell transfection: mRNA was transfected into HEK293T cells and cultured for 72 hours.

[0112] iv. Phenotypic detection: Spike protein expression levels were detected using flow cytometry.

[0113] Experimental results, such as Figure 2 As shown:

[0114] Control group (Ref): The expression level of the existing vaccine sequence was set at 100%.

[0115] Experimental group:

[0116] Seq#1: The expression level was 135% ± 8% (p<0.05), which was significantly better than that of the control group.

[0117] Seq#3: The expression level was 142% ± 7% (p<0.05), which was the highest in the group and consistent with the highest TE score predicted by the model (1.92).

[0118] Seq#5: The expression level was 95% ± 6%, which was not significantly improved, but it was consistent with the model's low prediction score (1.22).

[0119] Correlation analysis: such as Figure 3 As shown, the model prediction score is highly positively correlated with the experimentally measured protein expression level, with a Pearson correlation coefficient ρ = 0.69.

[0120] In the above embodiments, the computation time from data preparation to obtaining the Top-100 sequences is only 3 days, and the total cycle, combined with wet experimental verification, is controlled within 4 weeks, while traditional high-throughput screening typically takes 3-6 months. Furthermore, 8 of the recommended Top-10 sequences outperform or equal the benchmark sequence, with the optimal sequence achieving a 42% increase in expression levels. Compared to full-library synthesis screening, this method only requires the synthesis of a small number of AI-optimized sequences, reducing development costs by approximately 70%.

[0121] In addition, the present invention also provides an mRNA non-coding region sequence optimization system, such as... Figure 4 As shown, the mRNA non-coding region sequence optimization system includes: a construction module 10, an optimization module 20, a screening module 30, and a validation module 40.

[0122] Module 10 is used to build an initial candidate library based on the target protein.

[0123] The optimization module 20 is used to input the initial candidate library into the pre-trained mRNA sequence optimization model to obtain the prediction dataset.

[0124] The filtering module 30 is used to perform multi-dimensional scoring and sequence optimization on the prediction dataset to obtain the recommended sequence group.

[0125] The validation module 40 is used to perform biological validation on the sequence recommendation group to obtain the optimized mRNA sequence.

[0126] The prediction dataset includes sequence ID, sequence content, prediction TE score, and confidence interval.

[0127] Other specific implementations of the mRNA non-coding region sequence optimization system of the present invention can be found in the specific implementations of the mRNA non-coding region sequence optimization method of the above embodiments of the present invention.

[0128] In addition, the present invention also provides an electronic device, such as Figure 5 As shown, the controller 500 includes a processor 501 and a memory 503. The processor 501 and the memory 503 are connected, for example, via a bus 502.

[0129] Optionally, the controller 500 may also include a transceiver 504. It should be noted that in practical applications, the transceiver 504 is not limited to one, and the structure of the controller 500 does not constitute a limitation on the embodiments of the present invention.

[0130] Processor 501 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 501 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0131] Bus 502 may include a pathway for transmitting information between the aforementioned components. Bus 502 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 502 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0132] The memory 503 stores a computer program corresponding to the mRNA non-coding region sequence optimization method of the above embodiments of the present invention. This computer program is executed under the control of the processor 501. The processor 501 executes the computer program stored in the memory 503 to implement the content shown in the aforementioned method embodiments.

[0133] The controller 500 includes, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 5 The controller 500 shown is merely an example and should not be construed as limiting the functionality and scope of use of embodiments of the present invention.

[0134] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of this application. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0135] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0136] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0137] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0138] Similarly, it should be understood that, in order to streamline this application and aid in understanding one or more of the various inventive aspects, features of this application may sometimes be grouped together in a single embodiment, figure, or description thereof in the description of exemplary embodiments of this application. However, this approach should not be construed as reflecting an intention that the claimed application requires more features than are expressly recited in each claim. Rather, as reflected in the corresponding claims, its inventive point lies in solving the corresponding technical problem with features fewer than all features of a single disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this application.

[0139] Those skilled in the art will understand that, apart from the mutual exclusion of features, all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or elements of any method or apparatus so disclosed may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0140] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0141] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some modules according to the embodiments of this application. This application can also be implemented as an apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such an implementation of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0142] It should be noted that the above embodiments are illustrative of this application and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0143] The above description is merely a specific embodiment or illustration of the embodiments of this application. The scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method for optimizing mRNA non-coding region sequences, characterized in that, The method includes: Based on the target protein, construct an initial candidate library; The initial candidate library is input into a pre-trained mRNA sequence optimization model to obtain a prediction dataset; The predicted dataset is then subjected to multi-dimensional scoring and sequence optimization to obtain a sequence recommendation group; Biological validation was performed on the recommended sequence group to obtain the optimized mRNA sequence; The prediction dataset includes sequence ID, sequence content, prediction TE score, and confidence interval.

2. The method for optimizing mRNA non-coding region sequences according to claim 1, characterized in that, The step of constructing an initial candidate library based on the target protein includes: Based on the target protein, a variant sequences are obtained through a random mutation strategy; Based on the target protein, b optimized sequences were obtained through an evolutionary optimization strategy; The variant sequences described in section a and the optimized sequences described in section b are merged and deduplicated to obtain unique sequences in section c. The initial candidate library is obtained based on the unique sequences in section c.

3. The method for optimizing mRNA non-coding region sequences according to claim 2, characterized in that, The step of obtaining a variant sequences based on the target protein using a random mutation strategy includes: Obtain the baseline sequence corresponding to the target protein, use the baseline sequence as a template, set the mutation rate to d, perform a first preset round of random base substitutions to generate a variant sequences; The step of obtaining b optimized sequences based on the target protein through an evolutionary optimization strategy includes: Based on the target protein, k UTR sequences are selected as the initial population, and the pre-trained mRNA sequence optimization model is used as the fitness function for a second preset round of genetic algorithm iteration to generate b optimized sequences.

4. The method for optimizing mRNA non-coding region sequences according to claim 3, characterized in that, The mRNA sequence optimization model uses a deep neural network based on the Transformer architecture as its backbone, embedding pairwise position bias matrices into a multi-head self-attention mechanism, and calculating the attention score through... This involves injecting structural priors into the model; in, and These represent the nth mRNA sequence in the input mRNA sequence. The and the first A position index of nucleotides, Indicates the first A position in the attention. The query vector, In the A position in the attention. The key vector, The dimension of the key vector. Indicates pre-calculated, in position and Pairwise positional offset values ​​between; The training process of the mRNA sequence optimization model includes: Collect multi-source transcriptome data, including ribosomal imprinting data, synthetic UTR libraries, and viral mRNA; The multi-source transcriptome data was cleaned; The cleaned data were grouped according to the mRNA sequence length range and GC content range, and the translation efficiency index within each group was normalized by Z-score to obtain the validation set. The total loss function for the current training cycle is obtained by weighted summation based on the main task loss predicted by translation efficiency, the auxiliary task loss predicted by secondary structure features, and the auxiliary task loss predicted by stability score. The preset training termination condition is then determined based on the performance of the validation set. If the training termination condition is not met, the gradient is calculated using the total loss function, the parameters of the pre-inserted low-rank adaptive matrix in the model are updated and adjusted, the main parameters of the model are kept frozen, and the model with adjusted parameters is used for training in the next training cycle. If the training termination condition is met, the early stopping mechanism is triggered to stop training, and the model parameter state corresponding to the optimal moment of the validation set evaluation index is determined as the trained mRNA sequence optimization model.

5. The method for optimizing mRNA non-coding region sequences according to claim 1, characterized in that, Before inputting the initial candidate library into the pre-trained mRNA sequence optimization model, the method further includes: Target sequences in the initial candidate library with a length range of 30~100bp and containing only A, U, G, and C bases were selected. The target sequence is converted into a Token ID and packaged according to Batch Size = m; The packaged data is input into the pre-trained mRNA sequence optimization model.

6. The method for optimizing mRNA non-coding region sequences according to claim 1, characterized in that, The steps of performing multi-dimensional scoring and sequence optimization on the predicted dataset include: The predicted dataset is then subjected to initial screening and filtering. The remaining sequences after the initial screening are comprehensively scored, and each remaining sequence corresponds to a score. Sort the scores of the remaining sequences in descending order; The first n sorted sequences are used as the recommended sequence group.

7. The method for optimizing mRNA non-coding region sequences according to claim 6, characterized in that, The sequence is comprehensively scored using the following formula: ； in, , , , , Indicates preset parameters. This indicates the predicted TE score; This represents the average normalized Hamming distance between the current sequence and the remaining sequences; This represents the normalized value calculated based on the minimum free energy; This indicates the degree of nonspecific matching between the current sequence and known key gene sequences in the human transcriptome; This indicates the difficulty level of chemical synthesis.

8. The method for optimizing mRNA non-coding region sequences according to claim 1, characterized in that, The step of performing biological validation on the recommended sequence group to obtain the optimized mRNA sequence includes: Based on the sequences in the recommended group, DNA templates containing the T7 promoter, target 5'-UTR, protein CDS, and universal 3'-UTR were chemically synthesized using the gBlock system. The DNA template was used as a transcription template, and initial mRNA was prepared using T7 RNA polymerase. Capping and tailing were then performed to obtain the target mRNA. The target mRNA was transfected into target cells and cultured for a predetermined time. The protein expression levels of the target cells were detected by flow cytometry.

9. A system for optimizing mRNA non-coding region sequences, characterized in that, The system includes: The building block is used to construct an initial candidate library based on the target protein; The optimization module is used to input the initial candidate library into a pre-trained mRNA sequence optimization model to obtain a prediction dataset; The filtering module is used to perform multi-dimensional scoring and sequence optimization on the prediction dataset to obtain a sequence recommendation group; The verification module is used to perform biological verification on the recommended sequence group to obtain the optimized mRNA sequence; The prediction dataset includes sequence ID, sequence content, prediction TE score, and confidence interval.

10. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory, which, when executed by the processor, implements the mRNA non-coding region sequence optimization method as described in any one of claims 1-8.

Images