A method and device for evaluating the safety of an mRNA sequence, an electronic device, and a storage medium

Abstract

The present application relates to the technical field of mRNA sequence evaluation, and discloses a kind of mRNA sequence safety evaluation method, device, electronic equipment and storage medium, the method comprises: obtaining target mRNA sequence;Extract the biophysical characteristics of target mRNA sequence;Biophysical characteristics include: primary structure characteristics, secondary structure characteristics and functional characteristics;Biophysical characteristics are input into the mRNA sequence safety evaluation model constructed in advance, and safety evaluation result is obtained.The primary structure characteristics, secondary structure characteristics and functional characteristics of mRNA sequence are considered as comprehensive factors and input into the mRNA sequence safety evaluation model, the internal correlation between sequence structure characteristics, spatial structure characteristics, translation function characteristics and safety can be accurately captured, and the accuracy and reliability of the model for mRNA sequence safety evaluation are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of mRNA sequence evaluation technology, and specifically to a method, apparatus, electronic device, and storage medium for evaluating the safety of mRNA sequences. Background Technology

[0002] Safety assessment of mRNA sequences is a crucial step in ensuring the safety and efficacy of mRNA vaccines or drug treatments. However, current methods for assessing the safety of mRNA sequences face certain technical limitations. Traditional rule-based methods rely heavily on manually defined sequence features, such as GC content, U content, and specific motifs, resulting in limited feature coverage and difficulty in capturing complex sequence patterns. Furthermore, the discrimination thresholds of these methods often depend on subjective settings, lacking sufficient support from biological theory and exhibiting poor predictive ability for novel mRNA sequences.

[0003] Although current research has attempted to use deep learning for mRNA sequence safety assessment, most existing models use convolutional neural networks or recurrent neural networks to process sequence data, which do not fully consider the unique biophysical characteristics of mRNA, such as secondary structure and translation dynamics. This makes it difficult to accurately capture the intrinsic relationship between sequence structure and safety, resulting in problems such as misjudgment of safety level and bias in the location of risk regions, ultimately leading to insufficient overall prediction accuracy of the model. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for assessing the safety of mRNA sequences, in order to solve the problem that existing technologies consider only a single factor and the assessment results are unreliable when assessing the safety of mRNA sequences.

[0005] In a first aspect, the present invention provides a method for assessing the safety of an mRNA sequence, the method comprising:

[0006] Obtain the target mRNA sequence;

[0007] Extract the biophysical characteristics of the target mRNA sequence; biophysical characteristics include: primary structure characteristics, secondary structure characteristics, and functional characteristics;

[0008] Biophysical features are input into a pre-constructed mRNA sequence safety assessment model to obtain safety assessment results. The mRNA sequence safety assessment model includes a codon-aware dynamic convolution module, a structure-aware graph neural network module, and a ribosome translation simulation module arranged in parallel. The codon-aware dynamic convolution module is used to extract global long-range features and multi-scale local features. The structure-aware graph neural network module is used to extract global structural features that integrate the local and global spatial relationships of secondary structures. The ribosome translation simulation module is used to simulate the ribosome movement process, quantify the translation difficulty, and predict translation pause sites.

[0009] This invention incorporates the primary, secondary, and functional characteristics of mRNA sequences as comprehensive factors into the mRNA sequence safety assessment model. This approach can accurately capture the intrinsic correlation between sequence structural features, spatial structural features, translational functional features, and safety, effectively improving the accuracy and reliability of the model in assessing mRNA sequence safety.

[0010] In one alternative embodiment, the primary structural features include: base composition frequency, GC content, uridine content, dinucleotide frequency, and trinucleotide frequency.

[0011] Secondary structure features include: minimum free energy calculated using the RNA folding algorithm, base pairing probability matrix, structure accessibility score, and pseudoknot detection features;

[0012] Functional features include: immune stimulation motif matching count, codon fitness index, translation initiation site strength score, and polyadenylate signal detection features.

[0013] In one optional implementation, the mRNA sequence safety assessment model further includes:

[0014] The feature concatenation unit is used to concatenate the feature vectors output by the codon-aware dynamic convolution module, the structure-aware graph neural network module, and the ribosome translation simulation module to obtain concatenated features.

[0015] The uncertainty quantification unit is used to predict the probability value of each base position of the target mRNA sequence under each risk type based on splicing features, by enabling a random inactivation mechanism and performing multiple forward propagations, and to obtain probability distribution parameters.

[0016] The safety assessment output unit is used to determine the safety assessment results based on probability distribution parameters. The safety assessment results include: safety level classification results, risk area location results, and biological interpretation results.

[0017] In this embodiment, a feature splicing unit fuses multi-dimensional features of mRNA, an uncertainty quantification unit performs probabilistic modeling and uncertainty quantification of the prediction results, and a safety assessment output unit outputs multi-dimensional safety assessment results, comprehensively covering the structural, functional, and translational risks of the mRNA sequence. While achieving precise safety grading and accurate location of risk sites, it quantifies prediction uncertainties and provides interpretable biological evidence, effectively improving the accuracy, reliability, and interpretability of mRNA safety assessment, and providing strong technical support for mRNA drug development and safety screening.

[0018] In one alternative implementation, the codon-aware dynamic convolution module includes:

[0019] The global feature extraction submodule is used to extract global long-range features of the target mRNA sequence based on biophysical characteristics;

[0020] The local feature extraction submodule includes multiple convolutional units of different scales, used to extract local sequence features at different scales;

[0021] The codon position attention mechanism submodule is used to generate dynamic attention weights based on position encoding;

[0022] The weighted splicing submodule is used to weightedly fuse dynamic attention weights with local sequence features to obtain weighted splicing features;

[0023] The adaptive fusion gating mechanism submodule is used to receive global long-range features and weighted concatenated features, and dynamically learn the feature fusion weights through the gating mechanism to achieve adaptive fusion of cross-scale output.

[0024] In this implementation, features are extracted in parallel through global and local branches. Simultaneously, positional encoding embedding and an attention generation network are combined to generate dynamic attention weights, and local features at various scales are weighted to enhance key local information. Subsequently, the global features and weighted local features are concatenated, and an adaptive fusion gating mechanism dynamically learns the fusion weights of global and local features, ultimately outputting a hybrid feature vector that integrates global long-range information and local multi-scale information. This codon-aware dynamic convolution module can effectively characterize the sequence features of mRNA, providing more effective feature input for mRNA sequence security assessment.

[0025] In one alternative implementation, the structure-aware graph neural network module includes:

[0026] The graph construction unit is used to construct graph structure data that characterizes the secondary structure and spatial adjacency of mRNA, and generate an initial structure graph feature vector; where nodes represent bases and edges represent base pairing relationships or spatial proximity relationships;

[0027] The graph convolutional unit is used to generate node features that fuse the structural associations between mRNA bases based on the initial structure graph feature vector through iterative operations of message function, aggregation function and update function.

[0028] The graph pooling unit is used to calculate attention weights based on node features. After attention-weighted aggregation of node features, global structural features that integrate the local and global spatial relationships of the secondary structure are obtained.

[0029] In this embodiment, key biological information such as the secondary structure and spatial folding of mRNA is explicitly modeled in the form of graph topology, and multi-layer graph neural network message passing is performed. Combined with attention pooling to extract global structural features, it can accurately capture long-range structural dependencies between bases and provide more effective feature input for mRNA sequence safety assessment.

[0030] In one alternative implementation, the ribosomal translation simulation module includes:

[0031] The sliding window feature extraction unit is used to select local features in the ribosome-covered region;

[0032] Bidirectional long short-term memory network units are used to capture contextual temporal dependencies in the translation process based on local features of the ribosome-covered region;

[0033] The pause site prediction head unit is used to predict the translation difficulty probability value for each position;

[0034] The output unit is used to output the probability of translation pauses at each base position of the target mRNA sequence.

[0035] In this embodiment, the entire ribosomal translation process is accurately simulated through the collaborative design of sliding window feature extraction, Bi-LSTM temporal modeling, and pause site prediction head. The translation pause sites are quantified, significantly improving the accuracy and sensitivity of mRNA translation pause risk prediction, and providing key dynamic feature support for sequence optimization, expression efficiency improvement, and safety assessment.

[0036] In one optional implementation, the security assessment output unit includes:

[0037] The risk area visualization sub-unit is used to generate a sequence of visual heatmaps that label high-risk areas;

[0038] The natural language interpretation subunit is used to convert safety assessment results into biologically interpretable text;

[0039] The sequence optimization suggestion subunit is used to provide optimization suggestions for target mRNA sequences that are classified as high-risk in the safety assessment results.

[0040] In this embodiment, the extracted multi-scale biophysical features are input into the model for end-to-end prediction, and the output includes a safety level, potential risk areas, and targeted optimization suggestions with biological interpretability, thereby achieving a comprehensive, rapid, and reliable assessment of mRNA sequence safety.

[0041] In a second aspect, the present invention provides an mRNA sequence safety assessment device, the device comprising:

[0042] The acquisition module is used to acquire the target mRNA sequence;

[0043] The extraction module is used to extract the biophysical features of the target mRNA sequence; the biophysical features include: primary structure features, secondary structure features, and functional features;

[0044] The evaluation module is used to input biophysical features into a pre-constructed mRNA sequence safety evaluation model to obtain safety evaluation results. The mRNA sequence safety evaluation model includes a codon-aware dynamic convolution module, a structure-aware graph neural network module, and a ribosome translation simulation module arranged in parallel. The codon-aware dynamic convolution module is used to extract global long-range features and multi-scale local features. The structure-aware graph neural network module is used to extract global structural features that integrate the spatial relationships between local and global secondary structures. The ribosome translation simulation module is used to simulate the ribosome movement process, quantify the translation difficulty, and predict translation pause sites.

[0045] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the mRNA sequence safety assessment method of the first aspect or any corresponding embodiment described above.

[0046] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the mRNA sequence safety assessment method of the first aspect or any corresponding embodiment described above.

[0047] It should be noted that the mRNA sequence safety assessment device, electronic device, and computer-readable storage medium provided by this invention correspond to the mRNA sequence safety assessment method described above. Therefore, for the beneficial effects of the mRNA sequence safety assessment device, electronic device, and computer-readable storage medium, please refer to the description of the corresponding beneficial effects of the mRNA sequence safety assessment method above, and will not be repeated here. Attached Figure Description

[0048] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0049] Figure 1 This is a schematic flowchart of an mRNA sequence safety assessment method according to an embodiment of the present invention;

[0050] Figure 2 This is another flowchart illustrating the mRNA sequence safety assessment method according to an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of a codon-aware dynamic convolution module according to an embodiment of the present invention;

[0052] Figure 4 This is a schematic diagram of a structure-aware graph neural network module according to an embodiment of the present invention;

[0053] Figure 5 This is a schematic diagram of a ribosome translation simulation module according to an embodiment of the present invention;

[0054] Figure 6 This is a schematic diagram of the confidence level prediction results of the risk area location map according to an embodiment of the present invention;

[0055] Figure 7 This is a schematic diagram of the loss descent curve during the model training process according to an embodiment of the present invention;

[0056] Figure 8 This is a schematic diagram of the ROC curve for the accuracy of the model results according to an embodiment of the present invention;

[0057] Figure 9 This is a structural block diagram of an mRNA sequence safety assessment device according to an embodiment of the present invention;

[0058] Figure 10 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0061] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0062] According to an embodiment of the present invention, an embodiment of an mRNA sequence safety assessment method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0063] This embodiment provides a method for assessing the safety of mRNA sequences, which can be used on servers, terminals, mobile terminals, etc. Figure 1 This is a flowchart of an mRNA sequence safety assessment method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0064] Step S101: Obtain the target mRNA sequence.

[0065] The target mRNA sequence in this embodiment can be a candidate mRNA sequence designed through preliminary calculations, or an mRNA sequence obtained from biological databases or relevant literature. It may contain potential safety hazards such as immunogenicity and expression risks caused by the sequence's own structure or codon settings.

[0066] Before feature extraction, the mRNA sequence is usually converted into a discrete representation by an encoding module and combined with positional encoding to form an initial feature representation. That is, the raw mRNA nucleotide sequence of pure characters is converted into a discrete numerical representation that the model can recognize (such as one-hot encoding, embedding vector), and combined with positional encoding to form the basic numerical features of the sequence for subsequent biophysical feature extraction.

[0067] Step S102: Extract the biophysical features of the target mRNA sequence; the biophysical features include: primary structure features, secondary structure features, and functional features.

[0068] The biophysical features in this embodiment are multi-scale biophysical features, which contain feature representations with multi-dimensional information, that is, they cover features at multiple levels, from primary structure and secondary structure to function.

[0069] In some optional implementations, primary structural features include: base composition frequency, GC content, uridine content, dinucleotide frequency, and trinucleotide frequency; secondary structural features include: minimum free energy calculated by RNA folding algorithm, base pairing probability matrix, structural accessibility score, and pseudoknot detection feature; functional features include: immune stimulation motif matching count, codon fitness index, translation initiation site strength score, and polyadenylate signal detection feature.

[0070] The three types of features extracted above will be spliced ​​and fused into multimodal input data, providing a foundation for subsequent model predictions.

[0071] Step S103: Input biophysical features into a pre-constructed mRNA sequence safety assessment model to obtain safety assessment results; the mRNA sequence safety assessment model includes a codon-aware dynamic convolution module, a structure-aware graph neural network module, and a ribosome translation simulation module set up in parallel; the codon-aware dynamic convolution module is used to extract global long-range features and multi-scale local features; the structure-aware graph neural network module is used to extract global structural features that integrate the local and global spatial relationships of secondary structures; the ribosome translation simulation module is used to simulate the ribosome movement process, quantify the translation difficulty, and predict translation pause sites.

[0072] The safety assessment results include a safety level classification (high, medium, and low risk areas), a risk area location map, and biologically based interpretations and optimization suggestions. These results can effectively support subsequent experimental design and decision-making in vaccine design, gene therapy, and other fields.

[0073] In this embodiment, the primary, secondary, and functional features of the mRNA sequence are comprehensively considered and input into the mRNA sequence safety assessment model. This accurately captures the intrinsic correlation between sequence structure features, spatial structure features, translational function features, and safety, effectively improving the accuracy and reliability of the model's mRNA sequence safety assessment. Furthermore, the mRNA sequence safety assessment model in this embodiment achieves prediction AUCs of 0.85, 0.81, and 0.82 for high, medium, and low risk, respectively, significantly outperforming traditional rule-based or single machine learning-based assessment methods.

[0074] In some alternative implementations, refer to Figure 2 As shown, the mRNA sequence safety assessment model also includes:

[0075] The feature concatenation unit is used to concatenate the feature vectors output by the codon-aware dynamic convolution module, the structure-aware graph neural network module, and the ribosome translation simulation module to obtain concatenated features.

[0076] The uncertainty quantification unit is used to predict the probability value of each base position of the target mRNA sequence under each risk type based on splicing features, by enabling the random inactivation (Dropout) mechanism and performing multiple forward propagations, and obtain the probability distribution parameters.

[0077] The uncertainty quantization unit is primarily used for prediction calibration and feature selection to output a high-confidence feature set. The input to the uncertainty quantization unit is the total fused feature vector generated by pooling and concatenating three parallel modules, i.e., the concatenated feature. Through the Dropout mechanism, the uncertainty quantization unit forces the use of the Dropout layer from the model training phase during the inference phase, inputting the total fused feature vector into the model for N independent forward propagations. The mean and standard deviation of the prediction distribution are calculated as a measure of prediction uncertainty, generating a set of independent prediction results. Finally, a set of N prediction results is obtained, forming the prediction probability distribution for each base position and risk type of the mRNA sequence. This completes the probabilistic modeling and confidence selection of the prediction results, ultimately outputting a structured high-confidence feature set, providing a unique input for the downstream output module.

[0078] The model can not only predict the security level, but also output the network's uncertainty and calibrate the confidence level.

[0079] The safety assessment output unit is used to determine the safety assessment results based on probability distribution parameters. The safety assessment results include: safety level classification results, risk area location results, and biological interpretation results.

[0080] The safety assessment output unit is mainly used for feature translation and biological interpretation. This safety assessment output unit includes four sub-units: a classification unit, a risk region visualization unit, a natural language interpretation unit, and a sequence optimization suggestion unit. The four sub-units share the high-confidence feature set of the upstream input and independently process and generate corresponding results based on the well-known biological knowledge that each biophysical feature is associated with mRNA safety (e.g., high GC content → unstable secondary structure → degradation risk; multiple CpG motifs → high immunogenicity → inflammatory response). Finally, they are integrated into an end-to-end assessment report.

[0081] In this embodiment, a feature splicing unit fuses multi-dimensional features of mRNA, an uncertainty quantification unit enables probabilistic modeling and uncertainty quantification of prediction results, and a safety assessment output unit outputs multi-dimensional safety assessment results. This comprehensively covers the structural, functional, and translational risks of the mRNA sequence, achieving precise safety grading and accurate location of risk sites while quantifying prediction uncertainties and providing interpretable biological evidence. This effectively improves the accuracy, reliability, and interpretability of mRNA safety assessment, providing strong technical support for mRNA drug development and safety screening.

[0082] In some alternative implementations, refer to Figure 3 As shown, the codon-aware dynamic convolution module includes:

[0083] The global feature extraction submodule is used to extract global long-range features of the target mRNA sequence based on biophysical characteristics. This submodule incorporates a Transformer mechanism, which combines multi-head self-attention with a feedforward neural network, to capture long-range dependencies in the mRNA sequence.

[0084] The local feature extraction submodule includes multiple convolutional units of different scales, used to extract local sequence features at different scales. In this embodiment, multi-scale convolutional kernels with parallel processing are set, such as 3nt, 6nt, 9nt, and 12nt kernels, to extract local features at different scales from the mRNA sequence.

[0085] The codon position attention mechanism submodule is used to generate dynamic attention weights based on position encoding. It can obtain the relative position index of each nucleotide in the sequence within the codon, convert it into a discrete representation, input it into a multilayer perceptron, and output the attention weights corresponding to each convolutional kernel. That is, the feature weights can be adaptively adjusted according to biological key sites (codon positions).

[0086] The weighted concatenation submodule is used to weight and fuse dynamic attention weights with local sequence features to obtain weighted concatenated features. After dynamically weighting and extracting local features through multi-scale convolutional kernels and codon position attention mechanism, the calculated attention weights are multiplied element-wise with the corresponding convolutional layer outputs. All weighted convolutional kernel outputs are concatenated along the channel dimension, and the sequence dimension is restored through transpose operation to output the final mixed feature vector, i.e., the weighted concatenated features.

[0087] The adaptive fusion gating mechanism submodule is used to receive global long-range features and weighted concatenated features, and dynamically learn the feature fusion weights through the gating mechanism to achieve adaptive fusion of cross-scale output.

[0088] In this embodiment, features are extracted in parallel through global and local branches. Simultaneously, positional encoding embedding and an attention generation network are combined to generate dynamic attention weights, and local features at various scales are weighted to enhance key local information. Subsequently, the global features and weighted local features are concatenated, and an adaptive fusion gating mechanism dynamically learns the fusion weights of global and local features, ultimately outputting a hybrid feature vector that integrates global long-range information and local multi-scale information. This codon-aware dynamic convolution module can effectively characterize the sequence features of mRNA, providing more effective feature input for mRNA sequence security assessment.

[0089] Codon phase awareness and multi-scale convolution improve the faithful representation of biological semantics; structural vectorization and efficient computation achieve better prediction efficiency; the introduction of translation process features enhances sensitivity to immunogenicity and expression risks. Through multi-layer information collaborative processing, and by utilizing local patterns, global dependencies, structural relationships, and translation dynamics information, the model's prediction accuracy and robustness, as well as its interpretability and operability, are effectively improved.

[0090] In an optional implementation, refer to Figure 4 As shown, the structure-aware graph neural network module includes:

[0091] The graph construction unit is used to construct graph structure data that characterizes the secondary structure and spatial adjacency of mRNA, and generate an initial structure graph feature vector. Nodes represent bases, and edges represent base pairing or spatial proximity relationships. Combined with normalized distances between nodes and pairing types, edge feature vectors are formed.

[0092] Graph convolutional units, based on initial structural graph feature vectors, iteratively operate through message functions, aggregation functions, and update functions. This allows each base node to aggregate structural association information from neighboring nodes and update its own node state. After a preset number of iterations, node features fusing structural associations between mRNA bases are generated. In other words, multi-layer graph convolution operations are performed, concatenating source node features with edge features and generating messages through linear transformation. Then, a scatter-point addition aggregation operation is used to aggregate all messages pointing to the same target node in parallel, updating the target node features to learn a structure-aware sequence representation.

[0093] The graph pooling unit is used to calculate attention weights based on node features. After attention-weighted aggregation of node features, global structural features that integrate the local and global spatial relationships of the secondary structure are obtained.

[0094] Among them, the graph pooling unit is based on the node features that are updated iteratively by graph convolution and integrate the local spatial relationships of the secondary structure. It forms a global structural feature that integrates the local and global spatial relationships of the secondary structure through attention-weighted aggregation.

[0095] In this embodiment, key biological information such as the secondary structure and spatial folding of mRNA is explicitly modeled in the form of graph topology. Multi-layer graph neural network message passing is executed, and attention pooling is used to extract global structural features. This accurately captures long-range structural dependencies between bases, providing more effective feature input for mRNA sequence safety assessment. Furthermore, vectorized edge construction and aggregation can effectively reduce computational overhead.

[0096] In some alternative implementations, refer to Figure 5 As shown, the ribosomal translation simulation module includes:

[0097] The sliding window feature extraction unit is used to select local features in the ribosome-covered region. A one-dimensional convolutional layer can be used as a sliding window extractor to model local features of the sequence in a sliding window (approximately 30 nt in size of a ribosome), predict translation difficulty, and identify potential pause sites. Translation dynamics features are introduced into the model to enhance sensitivity to expression risks.

[0098] Bidirectional Long Short-Term Memory (Bi-LSTM) units are used to capture contextual temporal dependencies in the translation process based on local features of the ribosome-covered region.

[0099] The pause site prediction head unit is used to predict the translation difficulty probability value at each position. Specifically, the output of the Bi-LSTM is passed through a fully connected layer and a sigmoid activation function to predict the translation difficulty probability at each position. A threshold (0.7) can be set to mark the center positions of windows with difficulty probabilities exceeding this threshold as potential ribosomal pause sites.

[0100] The output unit is used to output the probability of translation pause sites at each base position of the target mRNA sequence.

[0101] In this embodiment, the entire ribosomal translation process is accurately simulated through the collaborative design of sliding window feature extraction, Bi-LSTM temporal modeling, and pause site prediction head. The translation pause sites are quantified, significantly improving the accuracy and sensitivity of mRNA translation pause risk prediction, and providing key dynamic feature support for sequence optimization, expression efficiency improvement, and safety assessment.

[0102] In some optional implementations, the security assessment output unit includes:

[0103] The risk area visualization subunit is used to generate visualized sequence heatmaps that label high-risk areas. Ultimately, it can generate and output the risk area location map prediction confidence level in the form of a heatmap. For details, please refer to [link / reference]. Figure 6 As shown.

[0104] The Natural Language Interpretation Subunit is used to convert security assessment results into biologically interpretable text.

[0105] The sequence optimization suggestion subunit is used to provide optimization suggestions for target mRNA sequences that are classified as high-risk in the safety assessment results.

[0106] Based on the high-risk regions located (such as high-content CpG islands, high-dimensional codon stop sites, etc.), it can automatically generate matching biological explanatory texts, potential mechanism analyses, and modification suggestions for high-risk sequences (such as replacing them with synonymous codons, optimizing local secondary structures, etc.).

[0107] This embodiment provides a biophysical feature fusion-based mRNA sequence safety assessment method, which extracts multi-scale biophysical features from the mRNA sequence. The extracted multi-scale biophysical features are input into a model for end-to-end prediction, outputting a biologically interpretable safety level, potential risk regions, and targeted modification suggestions, thus achieving a comprehensive, rapid, and reliable assessment of mRNA sequence safety.

[0108] In the early stages of training the mRNA sequence safety assessment model based on a multi-task contrastive learning training framework, the loss descent curve during training was as follows: Figure 7 As shown in the figure. Specifically, it includes a joint optimization task for safety classification, a structural consistency prediction task, and a sequence comparison learning task. Furthermore, a biophysical constraint term is introduced for regularization. Biological constraint regularization is incorporated into the loss function to ensure that the model learning conforms to biophysical principles. The ROC curve of the model results is shown in the figure. Figure 8 As shown, the model not only provides a safety level and risk location, but also outputs interpretable text and targeted modification suggestions. Furthermore, it incorporates uncertainty quantification during the inference phase to enhance the credibility of the results, thus meeting the stringent requirements of speed, reliability, and interpretability in applications such as vaccine design and gene therapy.

[0109] This embodiment combines dynamic convolution, graph neural networks, and a translation simulator in a multi-dimensional fusion to accurately capture the synergistic relationships between local mRNA patterns, global structural dependencies, and dynamic translation features. The network's unique characteristics of structured feature extraction, uncertainty hierarchy evaluation, and lightweight deployment not only significantly accelerate problem identification in the target region but also effectively avoid the limitations of traditional single numerical prediction through end-to-end interpretable output and automatic suggestion generation. This provides safer, more systematic, and effective intelligent design guidance for the precise development of nucleic acid drugs and vaccines.

[0110] This embodiment also provides an mRNA sequence safety assessment device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0111] This embodiment provides an mRNA sequence safety assessment device, such as... Figure 9 As shown, it includes:

[0112] Acquisition module 201 is used to acquire the target mRNA sequence;

[0113] Extraction module 202 is used to extract the biophysical features of the target mRNA sequence; the biophysical features include: primary structure features, secondary structure features and functional features;

[0114] Evaluation module 203 is used to input biophysical features into a pre-constructed mRNA sequence safety evaluation model to obtain safety evaluation results. The mRNA sequence safety evaluation model includes a codon-aware dynamic convolution module, a structure-aware graph neural network module, and a ribosome translation simulation module arranged in parallel. The codon-aware dynamic convolution module is used to extract global long-range features and multi-scale local features. The structure-aware graph neural network module is used to extract global structural features that integrate the local and global spatial relationships of secondary structures. The ribosome translation simulation module is used to simulate the ribosome movement process, quantify the translation difficulty, and predict translation pause sites.

[0115] In one alternative embodiment, the primary structural features include: base composition frequency, GC content, uridine content, dinucleotide frequency, and trinucleotide frequency.

[0116] Secondary structure features include: minimum free energy, base pairing probability matrix, structure accessibility score, and pseudoknot detection features obtained by RNA folding algorithm;

[0117] Functional features include: immune stimulation motif matching count, codon fitness index, translation initiation site strength score, and polyadenylate signal detection features.

[0118] In one optional implementation, the mRNA sequence safety assessment model further includes:

[0119] The feature concatenation unit is used to concatenate the feature vectors output by the codon-aware dynamic convolution module, the structure-aware graph neural network module, and the ribosome translation simulation module to obtain concatenated features.

[0120] The uncertainty quantification unit is used to predict the probability value of each base position of the target mRNA sequence under each risk type based on splicing features, by enabling a random inactivation mechanism and performing multiple forward propagations, and to obtain probability distribution parameters.

[0121] The safety assessment output unit is used to determine the safety assessment results based on probability distribution parameters. The safety assessment results include: safety level classification results, risk area location results, and biological interpretation results.

[0122] In one alternative implementation, the codon-aware dynamic convolution module includes:

[0123] The global feature extraction submodule is used to extract global long-range features of the target mRNA sequence based on biophysical characteristics;

[0124] The local feature extraction submodule includes multiple convolutional units of different scales, used to extract local sequence features at different scales;

[0125] The codon position attention mechanism submodule is used to generate dynamic attention weights based on position encoding;

[0126] The weighted splicing submodule is used to weightedly fuse dynamic attention weights with local sequence features to obtain weighted splicing features;

[0127] The adaptive fusion gating mechanism submodule is used to receive global long-range features and weighted concatenated features, and dynamically learn the feature fusion weights through the gating mechanism to achieve adaptive fusion of cross-scale output.

[0128] In one alternative implementation, the structure-aware graph neural network module includes:

[0129] The graph construction unit is used to construct graph structure data that characterizes the secondary structure and spatial adjacency of mRNA, and generate an initial structure graph feature vector; where nodes represent bases and edges represent base pairing relationships or spatial proximity relationships;

[0130] The graph convolutional unit is used to generate node features that fuse the structural associations between mRNA bases based on the initial structure graph feature vector through iterative operations of message function, aggregation function and update function.

[0131] The graph pooling unit is used to calculate attention weights based on node features. After attention-weighted aggregation of node features, global structural features that integrate the local and global spatial relationships of the secondary structure are obtained.

[0132] In one alternative implementation, the ribosomal translation simulation module includes:

[0133] The sliding window feature extraction unit is used to select local features in the ribosome-covered region;

[0134] Bidirectional long short-term memory network units are used to capture contextual temporal dependencies in the translation process based on local features of the ribosome-covered region;

[0135] The pause site prediction head unit is used to predict the translation difficulty probability value for each position;

[0136] The output unit is used to output the probability of translation pause sites at each base position of the target mRNA sequence.

[0137] In one optional implementation, the security assessment output unit includes:

[0138] The risk area visualization sub-unit is used to generate a sequence of visual heatmaps that label high-risk areas;

[0139] The natural language interpretation subunit is used to convert safety assessment results into biologically interpretable text;

[0140] The sequence optimization suggestion subunit is used to provide optimization suggestions for target mRNA sequences that are classified as high-risk in the safety assessment results.

[0141] The mRNA sequence safety assessment device provided in this embodiment of the invention can execute the mRNA sequence safety assessment method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0142] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0143] The following is a detailed reference. Figure 10 The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from memory 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0144] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 10 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0145] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a memory 308, or installed from a ROM 302. When the computer program is executed by the processor 301, it performs the functions defined in the mRNA sequence safety assessment method of the embodiments of the present invention.

[0146] Figure 10 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0147] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the mRNA sequence safety assessment method shown in the above embodiments is implemented.

[0148] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0149] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for assessing the safety of an mRNA sequence, wherein the method is used for non-therapeutic and non-diagnostic purposes, characterized in that, The method includes: Obtain the target mRNA sequence; Biophysical features of the target mRNA sequence are extracted; these biophysical features include: primary structure features, secondary structure features, and functional features; the primary structure features include: base composition frequency, GC content, uridine content, dinucleotide frequency, and trinucleotide frequency; the secondary structure features include: minimum free energy, base pairing probability matrix, structural accessibility score, and false knot detection features obtained by RNA folding algorithm; the functional features include: immune stimulation motif matching count, codon fitness index, translation initiation site strength score, and polyadenylate signal detection features. The biophysical characteristics are input into a pre-constructed mRNA sequence safety assessment model to obtain a safety assessment result. The mRNA sequence safety assessment model includes a codon-aware dynamic convolutional module, a structure-aware graph neural network module, and a ribosome translation simulation module arranged in parallel. The mRNA sequence safety assessment model further includes: a feature splicing unit, used to splice the feature vectors output by the codon-aware dynamic convolutional module, the structure-aware graph neural network module, and the ribosome translation simulation module to obtain spliced ​​features; an uncertainty quantification unit, used to predict the probability value of each base position of the target mRNA sequence under each risk type based on the spliced ​​features by enabling a random inactivation mechanism and performing multiple forward propagations to obtain probability distribution parameters; and a safety assessment output unit, used to determine the safety assessment result based on the probability distribution parameters. The codon-aware dynamic convolution module is used to extract global long-range features and multi-scale local features; the structure-aware graph neural network module is used to extract global structural features that fuse the local and global spatial relationships of secondary structures; the ribosome translation simulation module is used to simulate the ribosome movement process, quantify the translation difficulty, and predict translation pause sites; the security assessment results include: security level classification results, risk area location results, and biological interpretation results.

2. The method according to claim 1, characterized in that, The codon-aware dynamic convolution module includes: A global feature extraction submodule is used to extract global long-range features of the target mRNA sequence based on the biophysical features; The local feature extraction submodule includes multiple convolutional units of different scales, used to extract local sequence features at different scales; The codon position attention mechanism submodule is used to generate dynamic attention weights based on position encoding; The weighted splicing submodule is used to weightedly fuse the dynamic attention weights with the local sequence features to obtain weighted splicing features; The adaptive fusion gating mechanism submodule is used to receive the global long-range features and the weighted concatenation features, and dynamically learn the feature fusion weights through the gating mechanism to achieve adaptive fusion of cross-scale output.

3. The method according to claim 1, characterized in that, The structure-aware graph neural network module includes: The graph construction unit is used to construct graph structure data that characterizes the secondary structure and spatial adjacency of mRNA, and generate an initial structure graph feature vector; where nodes represent bases and edges represent base pairing relationships or spatial proximity relationships; The graph convolutional unit is used to generate node features that fuse the structural associations between mRNA bases based on the initial structure graph feature vector through iterative operations of message function, aggregation function and update function. The graph pooling unit is used to calculate attention weights based on the node features, and after performing attention-weighted aggregation on the node features, obtain global structural features that fuse the local and global spatial relationships of the secondary structure.

4. The method according to claim 1, characterized in that, The ribosome translation simulation module includes: The sliding window feature extraction unit is used to select local features in the ribosome-covered region; A bidirectional long short-term memory network unit is used to capture contextual temporal dependencies in the translation process based on local features of the ribosome-covered region; The pause site prediction head unit is used to predict the translation difficulty probability value for each position; The output unit is used to output the probability of translation pause sites at each base position of the target mRNA sequence.

5. The method according to claim 1, characterized in that, The security assessment output unit includes: The risk area visualization sub-unit is used to generate a sequence of visual heatmaps that label high-risk areas; The natural language interpretation subunit is used to convert the security assessment results into biological interpretation text; The sequence optimization suggestion subunit is used to provide optimization suggestions for the target mRNA sequence whose safety level classification result is high-risk in the safety assessment results.

6. An mRNA sequence safety assessment device, characterized in that, The device includes: The acquisition module is used to acquire the target mRNA sequence; An extraction module is used to extract the biophysical features of the target mRNA sequence. The biophysical features include: primary structure features, secondary structure features, and functional features. The primary structure features include: base composition frequencies, GC content, uridine content, dinucleotide frequencies, and trinucleotide frequencies. The secondary structure features include: minimum free energy calculated using an RNA folding algorithm, base pairing probability matrix, structural accessibility score, and false knot detection features. The functional features include: immune stimulation motif matching count, codon fitness index, translation initiation site strength score, and polyadenylate signal detection features. An evaluation module is used to input the biophysical characteristics into a pre-constructed mRNA sequence safety evaluation model to obtain a safety evaluation result. The mRNA sequence safety evaluation model includes a codon-aware dynamic convolution module, a structure-aware graph neural network module, and a ribosome translation simulation module arranged in parallel. The mRNA sequence safety evaluation model further includes: a feature splicing unit, used to splice the feature vectors output by the codon-aware dynamic convolution module, the structure-aware graph neural network module, and the ribosome translation simulation module to obtain spliced ​​features; an uncertainty quantification unit, used to predict the probability value of each base position of the target mRNA sequence under each risk type based on the spliced ​​features by enabling a random inactivation mechanism and performing multiple forward propagations to obtain probability distribution parameters; and a safety evaluation output unit, used to determine the safety evaluation result based on the probability distribution parameters. The codon-aware dynamic convolution module is used to extract global long-range features and multi-scale local features; the structure-aware graph neural network module is used to extract global structural features that fuse the local and global spatial relationships of secondary structures; the ribosome translation simulation module is used to simulate the ribosome movement process, quantify the translation difficulty, and predict translation pause sites; the security assessment results include: security level classification results, risk area location results, and biological interpretation results.

7. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the mRNA sequence safety assessment method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a computer to perform the mRNA sequence safety assessment method according to any one of claims 1 to 5.

Images