Protein sequence prediction method, model training method and task platform

Abstract

The embodiment of the invention provides a protein sequence prediction method, a model training method and a task platform, and the protein sequence prediction method comprises the steps: obtaining to-be-processed protein structure information; generating a corresponding initial protein sequence according to the to-be-processed protein structure information; inputting the initial protein sequence into a sequence iteration denoising model based on preset prediction times, and obtaining at least one to-be-processed protein sequence output by the sequence iteration denoising model and a sequence prediction probability corresponding to each to-be-processed protein sequence; and determining a target protein sequence in the at least one to-be-processed protein sequence according to the sequence prediction probability corresponding to each to-be-processed protein sequence.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of computer technology, and in particular to a method for predicting protein sequences. This specification also relates to a model training method and a task platform. Background Technology

[0002] In the field of drug discovery, the discovery of macromolecular drugs has attracted increasing attention. Unlike small molecule drugs, macromolecular drugs are mainly protein-based drugs, making the design of a protein that meets the target function a crucial issue. The structure of a protein determines its function, and function determines the drug's properties. Therefore, given a protein structure, designing its sequence can help researchers quickly narrow down the design space while simultaneously fulfilling the corresponding function, namely protein refolding.

[0003] In the process of generating sequences from a given protein structure, if traditional discriminative methods are used for modeling, the generated sequence can only match one sequence in the protein sequence dataset. Furthermore, biases in the protein sequence data can lead to significant cumulative bias in the modeling, resulting in greater errors in subsequent protein sequence generation. Therefore, how to better design protein sequences that can be converted into a given protein structure has become a pressing problem for engineers. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a method for predicting protein sequences. One or more embodiments of this specification also relate to a model training method, a task platform, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a method for predicting protein sequences is provided, comprising: Obtain the structural information of the protein to be processed; Generate the corresponding initial protein sequence based on the structural information of the protein to be processed; The initial protein sequence is input into the sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model. The target protein sequence is determined from at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

[0006] According to a second aspect of the embodiments of this specification, a model training method is provided, applied to a cloud-side device, comprising: Obtain the sample protein structure information and the sample protein sequence corresponding to the sample protein structure information, and add noise to the sample protein sequence to generate a noisy protein sequence; Obtain the sample protein structure features corresponding to the sample protein structure information, and generate sample model update parameters based on the sample protein structure features and a preset number of time steps; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model. The model loss value is calculated based on the predicted protein sequence and the sample protein sequence; Adjust the model parameters of the sequence iterative denoising model based on the model loss value, and continue training the sequence iterative denoising model until the model training stopping condition is met; The model parameters of the sequence iterative denoising model are sent to the edge device.

[0007] According to a third aspect of the embodiments of this specification, a method for predicting protein sequences is provided, applied to cloud-side devices, comprising: The receiving end device sends the protein structure information to be processed; Generate the corresponding initial protein sequence based on the structural information of the protein to be processed; The initial protein sequence is input into the sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model. The target protein sequence is determined from at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed. The target protein sequence is sent to the end-side device.

[0008] According to a fourth aspect of the embodiments of this specification, a task platform is provided, including a request interface and a response unit; The request interface is used to receive the protein structure information to be processed sent by the end device; The response unit is configured to generate a corresponding initial protein sequence based on the structural information of the protein to be processed; input the initial protein sequence into a sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model; and determine the target protein sequence in at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

[0009] According to a fifth aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the above method.

[0010] According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0011] According to a seventh aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0012] The protein sequence prediction method provided in the embodiments of this specification obtains the initial protein sequence corresponding to the structural information of the protein to be processed. Then, it progressively denoises and predicts the initial protein sequence using an iterative sequence denoising model to determine the final target protein sequence. Iterative sequence denoising provides an iterative process with probabilistic interpretability. Through a pre-trained iterative sequence denoising model, the model learns the probabilistic dependence between protein structure distribution and protein sequence, acquiring the entire distribution information between the two. This leads to greater diversity in the generated protein sequences, resulting in better prediction performance in the application stage and contributing to the advancement of macromolecular drug discovery. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating a protein sequence prediction method provided in one embodiment of this specification; Figure 2 This is a schematic flowchart of a protein sequence prediction method provided in one embodiment of this specification; Figure 3 This is a flowchart illustrating a model training method for a sequence iterative denoising model, as provided in one embodiment of this specification. Figure 4 This is a schematic diagram of the model training structure of a sequence iterative denoising model provided in one embodiment of this specification; Figure 5 This is a flowchart illustrating a model training method provided in one embodiment of this specification; Figure 6 This is a schematic flowchart of a protein sequence prediction method provided in one embodiment of this specification; Figure 7 This is an architectural diagram of a protein sequence prediction system provided in one embodiment of this specification; Figure 8 This is a schematic diagram of a task platform provided in one embodiment of this specification; Figure 9 This is a schematic diagram of the structure of a protein sequence prediction device provided in one embodiment of this specification; Figure 10 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0014] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0015] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0016] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0017] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0018] In the field of drug discovery, the discovery of macromolecular drugs has attracted increasing attention. Unlike small molecule drugs, macromolecular drugs are mainly protein-based drugs, and how to design a protein that meets the target function has become an important issue.

[0019] Protein structure determines function, and function determines drug properties. Therefore, designing protein sequences based on a given protein structure can help us quickly narrow down the design space while still fulfilling the corresponding function; this is the problem of protein inversion. In the process of generating protein sequences from a given protein structure, if traditional modeling methods are used, the generated protein sequences can only fit a single sequence in the dataset, lacking novelty. Furthermore, data bias can lead to significant cumulative bias in the sequence generation model.

[0020] Based on this, this specification provides a method for predicting protein sequences. This specification also relates to a model training method, a task platform, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0021] See Figure 1 , Figure 1 A flowchart of a protein sequence prediction method according to an embodiment of this specification is shown, specifically including the following steps.

[0022] Step 102: Obtain the structural information of the protein to be processed.

[0023] Protein structure refers to the spatial structure of a protein molecule. Protein structure information can be understood as the electronic information of the protein structure. In the methods provided in the embodiments of this specification, protein structure information, protein sequence, etc. are all information used for computer simulation calculation. They are electronic information of real protein structures and protein sequences after being electronically virtualized and stored in the computer for subsequent virtual processing.

[0024] Obtaining protein structure information involves technicians collecting and digitizing real protein structures, then storing this information in a computer. This mapping of real-world protein structures onto the computer allows technicians to perform subsequent processing and analysis. A protein's structure determines its function; in the medical and pharmaceutical fields, it can influence drug properties. Once a protein structure is known, it is defolded to predict its corresponding protein sequence. The protein structure information to be processed is the protein structure information that needs to be defolded to predict its corresponding protein sequence.

[0025] In practical applications, the protein structure information to be processed can be pre-saved to a specific storage location on the terminal. Upon receiving an information retrieval command, the terminal retrieves the protein structure information from that storage location. Alternatively, the protein structure information can be uploaded by a technician during the current processing phase. For example, a technician sends a protein structure information processing command to the terminal, which includes the protein structure information to be processed. The methods provided in the embodiments of this specification do not limit the specific method of obtaining the protein structure information to be processed; the actual application shall prevail. Obtaining the protein structure information to be processed provides a basis for data processing in subsequent processes.

[0026] Step 104: Generate the corresponding initial protein sequence based on the structural information of the protein to be processed.

[0027] The initial protein sequence can be understood as the predicted protein sequence after simple processing of the protein structure information to be processed. At this time, the initial protein sequence is an inaccurate protein sequence and needs to be further denoised to eliminate noise in the initial protein sequence.

[0028] In the method provided in the embodiments of this specification, the purpose is to predict the protein sequence corresponding to the structural information of the protein to be processed. However, the prediction process is relatively complex. During the data processing, the data deviation accumulates, which leads to a large cumulative deviation in the generation of the protein sequence. Therefore, the initial protein sequence will have a lot of noise. In the subsequent processing, a generation method based on diffusion bridge theory can be used to gradually remove the noise in the initial protein and generate a higher quality protein sequence.

[0029] In one specific embodiment provided in this specification, generating a corresponding initial protein sequence based on the structural information of the protein to be processed includes: The structural information of the protein to be processed is input into the structural encoder; Obtain the initial protein sequence output by the structure encoder.

[0030] In practical applications, based on the initial protein sequence corresponding to the protein structure information generator, some known structure encoders on the market can be used to input the protein structure information into the structure encoder for encoding processing, and obtain the initial protein sequence output by the structure encoder.

[0031] The structural encoder is used to extract feature information of the protein structure to be processed and generate the initial protein sequence. See [link to documentation]. Figure 2 , Figure 2 The diagram illustrates a flowchart of a protein sequence prediction method provided in one embodiment of this specification.Figure 2 As shown, s represents the structural information of the protein to be processed. This structural information is input into a structure encoder for processing. The structure encoder classifies each amino acid to predict the initial protein sequence (e.g., ...). Figure 2 (x in the text).

[0032] Step 106: Input the initial protein sequence into the sequence iterative denoising model based on the preset number of predictions, and obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model.

[0033] In the specific embodiments provided in this specification, the preset prediction count can be understood as the number of times the initial protein sequence is denoised and predicted. In this embodiment, the initial protein sequence needs to be denoised and predicted multiple times. Each denoising and prediction involves multiple time steps. After each denoising and prediction, a protein sequence to be processed and its corresponding sequence prediction probability are generated. In practical applications, if the preset prediction count is 5 times, the initial protein sequence will be denoised and predicted 5 times, ultimately generating 5 protein sequences to be processed and their corresponding sequence prediction probabilities. Figure 2 The preset number of predictions shown is 3. Figure 2 In this context, y1, y2, and y3 represent the sequence prediction probabilities.

[0034] The sequence iterative denoising model is used to perform denoising prediction based on an initial protein sequence, thereby generating the corresponding protein sequence to be processed and the sequence prediction probability corresponding to the protein sequence to be processed. In a specific embodiment provided in this specification, the sequence iterative denoising model can be a protein backfolding prediction model based on a diffusion bridge generation method.

[0035] like Figure 2 As shown, taking a preset prediction count of 3 as an example, the initial protein sequence undergoes 3 denoising predictions, resulting in 3 protein sequences to be processed and the corresponding sequence prediction probability for each sequence. In each denoising prediction, the sequence undergoes T denoising processes, where t is the time step and t is less than or equal to T.

[0036] The following explanation uses the example of performing a single denoising prediction on the initial protein sequence. In practical applications, the initial protein sequence can be denoised a corresponding number of times based on a preset number of predictions.

[0037] In one specific embodiment provided in this specification, the sequence iterative denoising model includes an encoder and a classifier; The initial protein sequence is input into a sequence iterative denoising model to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed, as output by the sequence iterative denoising model, including: The initial protein sequence is input into the encoder to obtain the protein coding information output by the encoder; The protein coding information is input into the classifier to obtain the protein sequence to be processed and the sequence prediction probability corresponding to the protein sequence to be processed output by the classifier.

[0038] In one specific embodiment provided in this specification, the sequence iterative denoising model includes an encoder and a classifier, wherein the encoder is used to extract sequence feature information from the initial protein sequence, and the classifier is used to classify according to the information output by the encoder, thereby generating the final protein sequence to be processed and the sequence prediction probability.

[0039] Specifically, the initial protein sequence is input into the encoder for processing to obtain the protein coding information output by the encoder. The protein coding information can be understood as the feature information extracted after the initial protein sequence is processed by the encoder. In the method provided in the embodiments of this specification, since the protein sequence is sequence data, the encoder can use a network structure of the Transformer model that is more suitable for sequence processing.

[0040] The protein-coding information output by the encoder is input into the classifier, where it is further decoded to output the corresponding protein sequence to be processed and the sequence prediction probability of the protein sequence to be processed.

[0041] Furthermore, the encoder comprises n sequentially connected coding layers, where n is a positive integer greater than 1; The initial protein sequence is input into the encoder to obtain the protein coding information output by the encoder, including: The initial protein sequence is input into the first coding layer to obtain the first coding vector output by the first coding layer. Input the (i-1)th encoding vector into the ith encoding layer to obtain the ith encoding vector output by the ith encoding layer, where 2≤i≤n; Increment i by 1 and check if i is greater than n. If yes, use the i-th encoding vector as the protein encoding information. If no, continue to execute the operation of inputting the (i-1)-th encoding vector into the i-th encoding layer to obtain the i-th encoding vector output by the i-th encoding layer.

[0042] As described above, the encoder uses a Transformer network structure. The encoder provided in this embodiment also includes n sequentially connected coding layers, where n is a positive integer greater than 1. For example, the encoder includes six sequentially connected coding layers. Sequentially connected coding layers mean that the output of one coding layer is the input of the next.

[0043] During the encoder's processing of the initial protein sequence, the initial protein sequence first undergoes embedding processing in an embedding layer. Embedding can be understood as adapting natural language into machine language that can be recognized by the machine. After embedding processing, the initial protein sequence generates protein sequence embedding features. These protein sequence embedding features are then input into the first coding layer to obtain the first coding vector output by the first coding layer.

[0044] In all coding layers except the first, the output of the previous coding layer is used as the input of the current coding layer. After processing by the current coding layer, a coding vector is output, and this vector is then input to the next coding layer. This process continues until the last coding layer processes the coding vector from the penultimate coding layer, at which point the protein-coding information output by the encoder is obtained. In practical applications, the coding vector output by the last coding layer can be directly used as the protein-coding information output by the encoder; alternatively, the coding vectors output by each coding layer can be weighted and then used as the protein-coding information output by the encoder. The specific implementation provided in this specification does not limit this approach.

[0045] In one specific embodiment provided in this specification, the coding layer includes a coding unit and a conversion unit; The (i-1)th encoded vector is input into the ith encoded layer to obtain the ith encoded vector output by the ith encoded layer, including: Model update parameters are generated based on a preset number of time steps, and the model parameters of the coding units in the i-th coding layer are updated based on the model update parameters. The (i-1)th encoding vector is input into the encoding unit to obtain the (i-1)th reference vector; The (i-1)th reference vector is input to the conversion unit to generate the i-th coding vector corresponding to the i-th coding layer.

[0046] In practical applications, each coding layer also includes a coding unit and a conversion unit. In this embodiment, the data processing flow in a certain coding layer is used as an example for explanation.

[0047] The method provided in this application denoises the initial protein sequence step by step over T time steps. In order to ensure the stability of the model parameters in the coding layer, the method provided in this specification generates model update parameters based on a preset number of time steps. These model update parameters are then combined with the original model parameters in the coding layer to obtain the current model parameters corresponding to the current time step. The data input to the coding layer is then processed based on the current model parameters.

[0048] For example, when time step t=0, the model update parameter p0 is obtained through the multilayer perceptron model, and combined with the model parameter p in the encoding layer, the model parameter p' = p + p0 at time step t=0 is obtained; when time step t=1, the model update parameter p1 is obtained through the multilayer perceptron model, and combined with the model parameter p in the encoding layer, the model parameter p' = p + p1 at time step t=1 is obtained; ... when time step t=T, the model update parameter pT is obtained through the multilayer perceptron model, and combined with the model parameter p in the encoding layer, the model parameter p' = p + pT at time step t=T is obtained.

[0049] In the method provided in the embodiments of this specification, at each time step, the model parameters of the coding units in each coding layer are updated using model update parameters. The coding vector output by the previous coding layer is input into the current coding unit of the current coding layer to obtain the reference vector output by the current coding unit. Then, the reference vector is input into the current transformation unit of the current coding layer to obtain the coding vector output by the current transformation unit.

[0050] Step 108: Determine the target protein sequence in at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

[0051] After obtaining multiple protein sequences to be processed and their corresponding sequence prediction probabilities, a target protein sequence can be determined from at least one protein sequence to be processed based on the prediction probabilities. The target protein sequence can be understood as the final protein sequence predicted by the method provided in the embodiments of this specification.

[0052] Specifically, the protein sequences to be processed can be sorted in descending order of their predicted probabilities, and the protein sequence with the highest predicted probability can be selected as the target protein sequence based on the sorting results.

[0053] The protein sequence prediction method provided in the embodiments of this specification proposes a protein reverse-folding method based on Markov bridges. This method specifically includes a structure encoder and an iterative sequence denoising model. The structure encoder obtains the initial protein sequence corresponding to the structural information of the protein to be processed. The iterative sequence denoising model progressively denoises and predicts the initial protein sequence to determine the final target protein sequence. The Markov bridge model based on diffusion bridge theory provides a probabilistically interpretable iterative process for sequence denoising. Through pre-trained iterative sequence denoising model, the model learns the probabilistic dependence between protein structure distribution and protein sequence, acquiring the entire distribution information between the two. This brings greater diversity to the generated protein sequences and mitigates error accumulation bias, resulting in better prediction performance in the application stage and contributing to the advancement of macromolecular drug discovery.

[0054] See Figure 3 , Figure 3 A flowchart is shown of a model training method for a sequence iterative denoising model according to an embodiment of this specification, which specifically includes the following steps.

[0055] Step 302: Obtain the sample protein structure information and the sample protein sequence corresponding to the sample protein structure information, and add noise to the sample protein sequence to generate a noisy protein sequence.

[0056] In this embodiment, the training method of the sequence iterative denoising model is further explained. The training process of the sequence iterative denoising model is supervised training, and the sample data used in the training process includes sample protein structure information and the sample protein sequences corresponding to the sample protein structure information.

[0057] In the method provided in the embodiments of this specification, during the inference phase of the model, iterative denoising of the noisy protein sequence is required. Therefore, during the training phase of the sequence iterative denoising model, it is necessary to enable the model to learn the probabilistic dependence between protein structure distribution and protein sequence. That is, during the model training phase of the sequence iterative denoising model provided in the embodiments of this specification, some noise needs to be pre-added to the sample protein sequence to generate a noisy protein sequence. Then, the sequence iterative denoising model is used to denoise the noisy protein sequence, making the denoised content closer to the sample protein sequence.

[0058] The noise protein sequence is the protein sequence after noise has been added to the sample protein sequence. The method for adding noise to the sample protein sequence can refer to currently available noise addition methods. In the methods provided in the embodiments of this specification, the specific method for adding noise to the sample protein sequence is not limited, and the actual application shall prevail.

[0059] Step 304: Obtain the sample protein structure features corresponding to the sample protein structure information, and generate sample model update parameters based on the sample protein structure features and a preset time step.

[0060] In this embodiment, the corresponding sample protein structural features are obtained based on the sample protein structural information. In a specific embodiment provided in this specification, obtaining the sample protein structural features corresponding to the sample protein structural information includes: The protein structure information of the sample is input into the structure encoder; Obtain the sample protein structural features output by the structural encoder.

[0061] Similar to the model inference stage, the corresponding sample protein structure features are generated based on the sample protein structure information. This can be achieved through a pre-trained structure encoder. In the structure encoder, the representation information of each node (amino acid) in the protein structure is learned through information transmission, and the amino acid type is predicted based on a linear classification head to obtain the prior distribution, thereby generating the sample protein structure features.

[0062] In the training method provided in the embodiments of this specification, sample model update parameters are generated based on the protein structure characteristics of the samples and a preset number of time steps. In the method provided in the embodiments of this specification, to ensure the training speed and effectiveness of the sequence iterative denoising model, some modules in the sequence iterative denoising model (such as the encoding units in the encoder) are pre-trained. During the continued training of the sequence iterative denoising model, the model parameters of these modules are retained without adjustment. Instead, different model parameters are obtained by combining the sample model update parameters with the original model parameters for subsequent processing of the sequence iterative denoising model.

[0063] In one specific embodiment provided in this specification, sample model update parameters are generated based on the sample protein structural characteristics and a preset number of time steps, including: The sample protein structural features and the preset time steps are input into the multilayer perceptron model; Obtain the sample model update parameters output by the multilayer perceptron model.

[0064] In this embodiment, after obtaining the structural features of the sample protein, pooling is performed on these features. The pooling result and a preset time step t are input into a multilayer perceptron model. The multilayer perceptron model is used to generate sample model update parameters that adjust the fixed model parameters in the iterative denoising model. During subsequent iterative processing, the sample model update parameters can be combined with the fixed parameters of the model itself to obtain the model parameters corresponding to the current time step, and then used for subsequent data processing.

[0065] Step 306: Input the noisy protein sequence, the sample model update parameters, and the sample protein structural features into the sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model.

[0066] After preprocessing the sample data, we can obtain the noisy protein sequences, sample model update parameters, and sample protein structural features. This information is then input into an iterative sequence denoising model to obtain the predicted protein sequences output by the model.

[0067] The following section, using the model structure of the sequence iterative denoising model as an example, further explains the data processing flow of the sequence iterative denoising model.

[0068] In one specific embodiment provided in this specification, the sequence iterative denoising model includes an encoder and a classifier; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into a sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model, including: The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the encoder to obtain the predicted protein coding information output by the encoder. The predicted protein coding information is input into the classifier to obtain the predicted protein sequence output by the classifier and the sequence prediction probability corresponding to the predicted protein sequence.

[0069] In this embodiment, the model structure of the sequence iterative denoising model is consistent with the model parameter structure of the application stage of the above-mentioned sequence iterative denoising model, including an encoder and a classifier. The encoder is used to extract the sequence feature information of the input noisy protein sequence, and the classifier is used to classify according to the sequence feature information and output the predicted protein sequence and the sequence prediction probability.

[0070] Specifically, the noisy protein sequence, sample model update parameters, and sample protein structural features are input into the encoder. The encoder processes these three pieces of information to generate predicted protein coding information. This predicted protein coding information is then processed by a classifier to obtain the predicted protein sequence and its corresponding sequence prediction probability.

[0071] In one specific embodiment provided in this specification, the encoder includes at least one encoding layer, and the encoding layer includes an encoding unit and a conversion unit; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the encoder to obtain the predicted protein coding information output by the encoder, including: Adjust the model parameters of each coding layer according to the sample model update parameters; Accordingly, for any coding layer, the method includes: The information from the input coding layer is input into the coding unit to obtain a reference vector. The information from the input coding layer includes the noisy protein sequence or the coding vector output by the previous coding layer. The reference vector and the sample protein structural features are input into the conversion unit to obtain the coding vector corresponding to the current coding layer; The coding vector output from the last coding layer is used as the predicted protein coding information output by the encoder.

[0072] In practical applications, an encoder specifically includes at least one coding layer, and each coding layer includes a coding unit and a conversion unit.

[0073] The noisy protein sequence, sample model update parameters, and sample protein structural features are input into the encoder for processing. First, the model parameters in each coding layer need to be updated using the sample model update parameters; specifically, the model parameters of each coding unit need to be updated. The method for updating the model parameters of each coding layer based on the sample model update parameters is the same as the model parameter update method in the above-described model application implementation, and will not be repeated here.

[0074] The information input to the current coding layer is fed into the current coding unit corresponding to the current layer. The model parameters of the current coding unit are the updated model parameters. When the current coding layer is the first coding layer, the information input to the current coding layer is the noisy protein sequence; when the current coding layer is any other coding layer besides the first coding layer, the information input to the current coding layer is the coding vector output by the previous coding layer.

[0075] The coding unit of the current coding layer processes the input information to obtain a reference vector, which is then input into the transformation unit of the current coding layer. In the transformation unit, the reference vector is combined with the sample protein structural features to generate the coding vector corresponding to the current coding layer. After processing by each coding layer, the coding vector output by the last coding layer can be used as the predicted protein coding information output by the encoder.

[0076] After obtaining the predicted protein coding information, it is input into the classifier for processing, and the predicted protein sequence output by the classifier and the sequence prediction probability corresponding to the predicted protein sequence can be obtained.

[0077] Step 308: Calculate the model loss value based on the predicted protein sequence and the sample protein sequence.

[0078] The method provided in the embodiments of this specification employs a supervised model training approach. After obtaining the predicted protein sequence, the model loss value can be calculated based on the predicted protein sequence and the sample protein sequence. Many functions can be used to calculate the model loss value in the method provided in this specification, such as the cross-entropy loss function, the maximum loss function, the average loss function, etc. This specification does not limit the specific method of using the loss function; the appropriate method should be determined based on the actual application.

[0079] Step 310: Adjust the model parameters of the sequence iterative denoising model according to the model loss value, and continue training the sequence iterative denoising model until the model training stopping condition is met.

[0080] Once the model loss value is obtained, the model parameters of the sequence iterative denoising model can be further adjusted using this value. Specifically, the model loss value is backpropagated within the sequence iterative denoising model to adjust the model parameters, and the model is trained continuously until the training stops. The trained sequence iterative denoising model can then be used to iteratively denoise and predict initial protein sequences based on protein structure predictions, thereby generating more suitable target protein sequences.

[0081] In one specific embodiment provided in this specification, adjusting the model parameters of the sequence iterative denoising model based on the model loss value includes: The model parameters of each transformation unit in the sequence iterative denoising model are adjusted based on the model loss value.

[0082] In practical applications, some model parameters in sequence iterative denoising models are pre-trained and fixed, such as the model parameters of the encoding units in each encoding layer. Therefore, adjusting the model parameters of a sequence iterative denoising model based on the model loss value specifically refers to adjusting the model parameters of each transformation unit in the encoder of the sequence iterative denoising model based on the model loss value.

[0083] In another specific embodiment provided in this specification, the method further includes: The model parameters of the multilayer perceptron model are adjusted based on the model loss value.

[0084] In practical applications, the model parameters of the multilayer perceptron (MLP) model can be further adjusted based on the model loss value. The MLP model is used to generate sample model update parameters to adjust the fixed model parameters in the sequence iterative denoising model. With the encoding units in the encoder pre-trained and the model parameters fixed, adjusting the MLP model parameters can bring changes to the encoder's encoding process, thereby obtaining richer encoded features.

[0085] See Figure 4 , Figure 4 A schematic diagram of the model training structure of a sequence iterative denoising model provided in one embodiment of this specification is shown. Figure 4 As shown, the sequence iterative denoising model only reflects the structure of the encoder; the classifier of the sequence iterative denoising model is not shown.

[0086] like Figure 4 As shown, the sample protein structure information is first input into the structure encoder to obtain the sample protein structure features. The sample protein structure features and the preset time steps are then input into the multilayer perceptron model, which generates sample model update parameters.

[0087] During the processing at each time step, the updated model parameters generated using the sample at the current time step are combined with the original model parameters of the coding units in the model to obtain the updated model parameters for each coding unit. The updated model parameters are then used to process the input noisy protein sequence to obtain the reference vector output by the coding unit.

[0088] The reference vector output by the coding unit and the structural features of the sample protein are fused in the transformation unit. After downsampling and upsampling, the predicted protein coding information is finally output.

[0089] During the adjustment of model parameters, the model parameters of the conversion unit and the multilayer perceptron model are adjusted, but the model parameters of the coding unit and the structural encoder are not adjusted.

[0090] The model training method provided in the embodiments of this specification enables the pre-trained protein language model to effectively capture time step and structural information. Simultaneously, it ensures that the model parameters of the sequence iterative denoising model remain compatible with the pre-trained protein language model. The trained sequence iterative denoising model is used for iterative denoising of the initial protein sequence.

[0091] See Figure 5 , Figure 5 This is a flowchart illustrating a model training method provided in one embodiment of this specification. The method is applied to cloud-side devices and specifically includes the following steps: Step 502: Obtain the sample protein structure information and the sample protein sequence corresponding to the sample protein structure information, and add noise to the sample protein sequence to generate a noisy protein sequence.

[0092] Step 504: Obtain the sample protein structure features corresponding to the sample protein structure information, and generate sample model update parameters based on the sample protein structure features and a preset time step.

[0093] Step 506: Input the noisy protein sequence, the sample model update parameters, and the sample protein structural features into the sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model.

[0094] Step 508: Calculate the model loss value based on the predicted protein sequence and the sample protein sequence.

[0095] Step 510: Adjust the model parameters of the sequence iterative denoising model according to the model loss value, and continue training the sequence iterative denoising model until the model training stopping condition is met.

[0096] Step 512: Send the model parameters of the sequence iterative denoising model to the end device.

[0097] It should be noted that steps 502-510 are implemented in the same way as steps 302-310 above, and will not be described again in the embodiments of this specification.

[0098] In practical applications, training a model requires a large amount of data and significant computing resources, which edge devices may lack. Therefore, model training can be performed on cloud devices. After obtaining the model parameters of the sequence iterative denoising model, the cloud device can send these parameters to the edge device. The edge device then constructs a sequence iterative denoising model locally based on these parameters and further utilizes the model for protein sequence prediction.

[0099] The model training method provided in the embodiments of this specification enables the pre-trained protein language model to effectively capture time step and structural information. Simultaneously, it ensures that the model parameters of the sequence iterative denoising model remain compatible with the pre-trained protein language model. The trained sequence iterative denoising model is used for iterative denoising of the initial protein sequence.

[0100] See Figure 6 , Figure 6 This is a flowchart illustrating a protein sequence prediction method provided in one embodiment of this specification. The method is applied to cloud-based devices and specifically includes the following steps: Step 602: Receive the protein structure information to be processed sent by the receiving end-side device.

[0101] Step 604: Generate the corresponding initial protein sequence based on the structural information of the protein to be processed.

[0102] Step 606: Input the initial protein sequence into the sequence iterative denoising model based on the preset number of predictions, and obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model.

[0103] Step 608: Determine the target protein sequence in at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

[0104] Step 610: Send the target protein sequence to the end-side device.

[0105] See Figure 7 , Figure 7 This specification illustrates an architecture diagram of a protein sequence prediction system provided in one embodiment. The protein sequence prediction system may include a client 100 and a server 200. Client 100 is used to send the protein structure information to be processed to server 200; Server 200 is used to receive the protein structure information to be processed sent by the end device; generate a corresponding initial protein sequence based on the protein structure information to be processed; input the initial protein sequence into a sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model; determine the target protein sequence from at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed; and send the target protein sequence to client 100. Client 100 is also used to receive the target protein sequence sent by server 200.

[0106] A protein sequence prediction system may include multiple clients 100 and a server 200. Clients 100 can be referred to as edge devices, and server 200 can be referred to as cloud devices. Multiple clients 100 can establish communication connections through server 200. In the protein sequence prediction scenario, server 200 is used to provide protein sequence prediction services among multiple clients 100. Each client 100 can act as either a sender or a receiver, communicating through server 200.

[0107] Users can interact with server 200 through client 100 to receive data sent by other clients 100, or send data to other clients 100, etc. In the scenario of protein sequence prediction, users can publish data streams to server 200 through client 100, server 200 can generate target protein sequences based on the data streams, and push the target protein sequences to other clients that have established communication.

[0108] In this system, client 100 and server 200 establish a connection via a network. The network provides the medium for communication between client 100 and server 200. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. Data transmitted by client 100 may need to undergo encoding, transcoding, compression, or other processing before being published to server 200.

[0109] Client 100 can be a browser, an app (application), a web application such as an H5 (HyperText Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. Client 100 can be developed based on the software development kit (SDK) of the corresponding service provided by server 200, such as a real-time communication (RTC) SDK. Client 100 can be deployed on electronic devices and depends on the device or certain apps on the device to run. Electronic devices may have displays and support information browsing, such as personal mobile terminals like mobile phones, tablets, and personal computers. Various other types of applications can also be configured on electronic devices, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platform software.

[0110] Server 200 may include servers providing various services, such as servers providing communication services to multiple clients, servers supporting backend training of models used on clients, and servers processing data sent by clients. It should be noted that server 200 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server integrated with blockchain. The server can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0111] It is worth noting that the protein sequence prediction method provided in the embodiments of this specification is generally executed by the server. However, in other embodiments of this specification, the client may also have similar functions to the server, thereby executing the protein sequence prediction method provided in the embodiments of this specification. In other embodiments, the protein sequence prediction method provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0112] Figure 8 A schematic diagram of a task platform provided in an embodiment of this specification is shown. The task platform includes a request interface 802 and a response unit 804, wherein: The request interface 802 is used to receive the protein structure information to be processed sent by the end device; The response unit 804 is used to generate a corresponding initial protein sequence based on the structural information of the protein to be processed; input the initial protein sequence into a sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model; and determine the target protein sequence in at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

[0113] Corresponding to the above method embodiments, this specification also provides embodiments of a protein sequence prediction device. Figure 9 A schematic diagram of a protein sequence prediction device provided in one embodiment of this specification is shown. Figure 9 As shown, the device includes: The acquisition module 902 is configured to acquire the structural information of the protein to be processed. The generation module 904 is configured to generate a corresponding initial protein sequence based on the structural information of the protein to be processed. The prediction module 906 is configured to input the initial protein sequence into the sequence iterative denoising model based on a preset number of predictions, and obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model. The determination module 908 is configured to determine the target protein sequence in at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

[0114] Optionally, the generation module 904 is further configured to: The structural information of the protein to be processed is input into the structural encoder; Obtain the initial protein sequence output by the structure encoder.

[0115] Optionally, the sequence iterative denoising model includes an encoder and a classifier; The prediction module 906 is further configured as follows: The initial protein sequence is input into the encoder to obtain the protein coding information output by the encoder; The protein coding information is input into the classifier to obtain the protein sequence to be processed and the sequence prediction probability corresponding to the protein sequence to be processed output by the classifier.

[0116] Optionally, the encoder includes n sequentially connected coding layers, where n is a positive integer greater than 1; The prediction module 906 is further configured as follows: The initial protein sequence is input into the first coding layer to obtain the first coding vector output by the first coding layer. Input the (i-1)th encoding vector into the ith encoding layer to obtain the ith encoding vector output by the ith encoding layer, where 2≤i≤n; Increment i by 1 and check if i is greater than n. If yes, use the i-th encoding vector as the protein encoding information. If no, continue to execute the operation of inputting the (i-1)-th encoding vector into the i-th encoding layer to obtain the i-th encoding vector output by the i-th encoding layer.

[0117] Optionally, the coding layer includes coding units and conversion units; The prediction module 906 is further configured as follows: Model update parameters are generated based on a preset number of time steps, and the model parameters of the coding units in the i-th coding layer are updated based on the model update parameters. The (i-1)th encoding vector is input into the encoding unit to obtain the (i-1)th reference vector; The (i-1)th reference vector is input to the conversion unit to generate the i-th coding vector corresponding to the i-th coding layer.

[0118] Optionally, the device further includes a training module configured to: Obtain the sample protein structure information and the sample protein sequence corresponding to the sample protein structure information, and add noise to the sample protein sequence to generate a noisy protein sequence; Obtain the sample protein structure features corresponding to the sample protein structure information, and generate sample model update parameters based on the sample protein structure features and a preset number of time steps; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model. The model loss value is calculated based on the predicted protein sequence and the sample protein sequence; The model parameters of the sequence iterative denoising model are adjusted according to the model loss value, and the sequence iterative denoising model is trained until the model training stops.

[0119] Optionally, the training module is further configured to: The protein structure information of the sample is input into the structure encoder; Obtain the sample protein structural features output by the structural encoder.

[0120] Optionally, the sequence iterative denoising model includes an encoder and a classifier; The training module is further configured as follows: The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the encoder to obtain the predicted protein coding information output by the encoder. The predicted protein coding information is input into the classifier to obtain the predicted protein sequence output by the classifier and the sequence prediction probability corresponding to the predicted protein sequence.

[0121] Optionally, the encoder includes at least one encoding layer, and the encoding layer includes encoding units and conversion units; The training module is further configured as follows: Adjust the model parameters of each coding layer according to the sample model update parameters; The information from the input coding layer is input into the coding unit to obtain a reference vector. The information from the input coding layer includes the noisy protein sequence or the coding vector output by the previous coding layer. The reference vector and the sample protein structural features are input into the conversion unit to obtain the coding vector corresponding to the current coding layer; The coding vector output from the last coding layer is used as the predicted protein coding information output by the encoder.

[0122] Optionally, the training module is further configured to: The model parameters of each transformation unit in the sequence iterative denoising model are adjusted based on the model loss value.

[0123] Optionally, the training module is further configured to: The sample protein structural features and the preset time steps are input into the multilayer perceptron model; Obtain the sample model update parameters output by the multilayer perceptron model.

[0124] Optionally, the training module is further configured to: The model parameters of the multilayer perceptron model are adjusted based on the model loss value.

[0125] The above is a schematic scheme of a protein sequence prediction device according to this embodiment. It should be noted that the technical solution of this protein sequence prediction device and the technical solution of the protein sequence prediction method described above belong to the same concept. For details not described in detail in the technical solution of the protein sequence prediction device, please refer to the description of the technical solution of the protein sequence prediction method described above.

[0126] Figure 10 A structural block diagram of a computing device 1000 according to one embodiment of this specification is shown. The components of the computing device 1000 include, but are not limited to, a memory 1010 and a processor 1020. The processor 1020 is connected to the memory 1010 via a bus 1030, and a database 1050 is used to store data.

[0127] The computing device 1000 also includes an access device 1040, which enables the computing device 1000 to communicate via one or more networks 1060. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 1040 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0128] In one embodiment of this specification, the above-described components of the computing device 1000 and Figure 10 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 10 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0129] The computing device 1000 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 1000 can also be a mobile or stationary server.

[0130] The processor 1020 is used to execute the following computer program / instructions, which, when executed by the processor, implement the steps of the above-mentioned protein sequence prediction method or model training method.

[0131] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, the computing device embodiments are relatively simple in description because they are fundamentally similar to the protein sequence prediction method or model training method embodiments; relevant parts can be referred to the descriptions of the protein sequence prediction method or model training method embodiments.

[0132] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the protein sequence prediction method or model training method described above.

[0133] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the computer-readable storage medium embodiments are described simply because they are substantially similar to the protein sequence prediction method or model training method embodiments; relevant parts can be referred to the descriptions of the protein sequence prediction method or model training method embodiments.

[0134] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described protein sequence prediction method or model training method.

[0135] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product belongs to the same concept as the technical solution of the protein sequence prediction method or model training method described above. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the protein sequence prediction method or model training method described above.

[0136] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0137] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0138] It should be noted that the above description describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this specification.

[0139] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0140] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predicting protein sequences, comprising: Obtain the structural information of the protein to be processed; Generate the corresponding initial protein sequence based on the structural information of the protein to be processed; The initial protein sequence is input into the sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model. The target protein sequence is determined from at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

2. The method as described in claim 1, wherein generating a corresponding initial protein sequence based on the structural information of the protein to be processed comprises: The structural information of the protein to be processed is input into the structural encoder; Obtain the initial protein sequence output by the structure encoder.

3. The method as described in claim 1, wherein the sequence iterative denoising model comprises an encoder and a classifier; The initial protein sequence is input into a sequence iterative denoising model to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed, as output by the sequence iterative denoising model, including: The initial protein sequence is input into the encoder to obtain the protein coding information output by the encoder; The protein coding information is input into the classifier to obtain the protein sequence to be processed and the sequence prediction probability corresponding to the protein sequence to be processed output by the classifier.

4. The method as described in claim 3, wherein the encoder comprises n sequentially connected coding layers, where n is a positive integer greater than 1; The initial protein sequence is input into the encoder to obtain the protein coding information output by the encoder, including: The initial protein sequence is input into the first coding layer to obtain the first coding vector output by the first coding layer. Input the (i-1)th encoding vector into the ith encoding layer to obtain the ith encoding vector output by the ith encoding layer, where 2≤i≤n; Increment i by 1 and check if i is greater than n. If yes, use the i-th encoding vector as the protein encoding information. If not, continue to execute the operation of inputting the (i-1)-th encoding vector into the i-th encoding layer to obtain the i-th encoding vector output by the i-th encoding layer.

5. The method as described in claim 4, wherein the coding layer comprises a coding unit and a conversion unit; The (i-1)th encoded vector is input into the ith encoded layer to obtain the ith encoded vector output by the ith encoded layer, including: Model update parameters are generated based on a preset number of time steps, and the model parameters of the coding units in the i-th coding layer are updated based on the model update parameters. The (i-1)th encoding vector is input into the encoding unit to obtain the (i-1)th reference vector; The (i-1)th reference vector is input to the conversion unit to generate the i-th coding vector corresponding to the i-th coding layer.

6. The method of claim 1, wherein the sequence iterative denoising model is obtained by training through the following steps: Obtain the sample protein structure information and the sample protein sequence corresponding to the sample protein structure information, and add noise to the sample protein sequence to generate a noisy protein sequence; Obtain the sample protein structure features corresponding to the sample protein structure information, and generate sample model update parameters based on the sample protein structure features and a preset number of time steps; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model. The model loss value is calculated based on the predicted protein sequence and the sample protein sequence; The model parameters of the sequence iterative denoising model are adjusted according to the model loss value, and the sequence iterative denoising model is trained until the model training stops.

7. The method of claim 6, wherein obtaining the sample protein structural features corresponding to the sample protein structural information comprises: The protein structure information of the sample is input into the structure encoder; Obtain the sample protein structural features output by the structural encoder.

8. The method of claim 6, wherein the sequence iterative denoising model comprises an encoder and a classifier; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into a sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model, including: The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the encoder to obtain the predicted protein coding information output by the encoder. The predicted protein coding information is input into the classifier to obtain the predicted protein sequence output by the classifier and the sequence prediction probability corresponding to the predicted protein sequence.

9. The method of claim 8, wherein the encoder comprises at least one encoding layer, the encoding layer comprising encoding units and conversion units; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the encoder to obtain the predicted protein coding information output by the encoder, including: Adjust the model parameters of each coding layer according to the sample model update parameters; Accordingly, for any coding layer, the method includes: The information from the input coding layer is input into the coding unit to obtain a reference vector. The information from the input coding layer includes the noisy protein sequence or the coding vector output by the previous coding layer. The reference vector and the sample protein structural features are input into the conversion unit to obtain the coding vector corresponding to the current coding layer; The coding vector output from the last coding layer is used as the predicted protein coding information output by the encoder.

10. The method of claim 9, wherein adjusting the model parameters of the sequence iterative denoising model according to the model loss value comprises: The model parameters of each transformation unit in the sequence iterative denoising model are adjusted based on the model loss value.

11. The method of claim 6, wherein generating sample model update parameters based on the sample protein structural features and a preset number of time steps includes: The sample protein structural features and the preset time steps are input into the multilayer perceptron model; Obtain the sample model update parameters output by the multilayer perceptron model.

12. The method of claim 11, further comprising: The model parameters of the multilayer perceptron model are adjusted based on the model loss value.

13. A model training method applied to cloud-side devices, comprising: Obtain the sample protein structure information and the sample protein sequence corresponding to the sample protein structure information, and add noise to the sample protein sequence to generate a noisy protein sequence; Obtain the sample protein structure features corresponding to the sample protein structure information, and generate sample model update parameters based on the sample protein structure features and a preset number of time steps; The noisy protein sequence, the sample model update parameters, and the sample protein structural features are input into the sequence iterative denoising model to obtain the predicted protein sequence output by the sequence iterative denoising model. The model loss value is calculated based on the predicted protein sequence and the sample protein sequence; Adjust the model parameters of the sequence iterative denoising model based on the model loss value, and continue training the sequence iterative denoising model until the model training stopping condition is met; The model parameters of the sequence iterative denoising model are sent to the edge device.

14. A method for predicting protein sequences, applied to cloud-side devices, comprising: The receiving end device sends the protein structure information to be processed; Generate the corresponding initial protein sequence based on the structural information of the protein to be processed; The initial protein sequence is input into the sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model. The target protein sequence is determined from at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed. The target protein sequence is sent to the end-side device.

15. A task platform, comprising a request interface and a response unit; The request interface is used to receive the protein structure information to be processed sent by the end device; The response unit is used to generate a corresponding initial protein sequence based on the structural information of the protein to be processed. The initial protein sequence is input into the sequence iterative denoising model based on a preset number of predictions to obtain at least one protein sequence to be processed and the sequence prediction probability corresponding to each protein sequence to be processed output by the sequence iterative denoising model. The target protein sequence is determined from at least one protein sequence to be processed based on the sequence prediction probability corresponding to each protein sequence to be processed.

16. A computing device, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 14.

17. A computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 14.

18. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 14.

Images