An optimization method, system, device and storage medium of AAV capsid protein

Abstract

This application discloses an optimization method, system, device, and storage medium for AAV capsid proteins. The method involves acquiring the first amino acid sequence data and target performance score data of the initial AAV capsid protein; inputting the first amino acid sequence data into a trained scoring model to extract a first feature vector and predict the first score data; calculating the difference between the target score data and the first score data; updating the first feature vector using gradient backpropagation if the difference value is greater than a preset threshold; and determining the second amino acid sequence data based on the current first feature vector using a trained sequence decoding model, and obtaining the target AAV capsid protein based on the second amino acid sequence data. This method can conveniently and efficiently find AAV variants that meet the requirements, is simple to apply, and has low implementation cost. This application can be widely used in the field of bioinformatics.

Description

Technical Field

[0001] This application relates to the field of bioinformatics, and in particular to a method, system, device, and storage medium for optimizing an AAV capsid protein. Background Technology

[0002] Adeno-associated virus (AAV) is a type of single-stranded linear DNA defective virus. Due to its low genotoxicity in humans, it is one of the most promising gene therapy vectors. However, wild-type AAV has drawbacks such as insufficient tissue-specific targeting and low yield, and new AAVs are usually obtained by "capsid engineering".

[0003] In related technologies, when modifying new AAVs using capsid engineering, DNA libraries are typically constructed through random peptide display and family DNA shuffling. Natural selection is then simulated by targeted screening for specific characteristics to identify AAVs with those characteristics. However, to obtain AAV mutants that meet the target criteria, each DNA library requires multiple rounds of screening. This process demands high library capacity and sophisticated experimental techniques, is complex, and incurs high labor costs. Furthermore, since the library capacity often fails to cover all DNA mutation types, the resulting AAVs may not meet the intended requirements. Summary of the Invention

[0004] The purpose of this application is to at least partially solve one of the technical problems existing in the related art.

[0005] Therefore, one object of the embodiments of this application is to provide a method, system, device and storage medium for optimizing AAV capsid proteins.

[0006] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of this application include:

[0007] On one hand, embodiments of this application provide a method for optimizing AAV capsid proteins, the method comprising:

[0008] Obtain the first amino acid sequence data of the initial AAV capsid protein and the target performance score data for the predetermined performance;

[0009] The first amino acid sequence data is input into the trained scoring model, the first feature vector of the first amino acid sequence data is extracted, and the first score data is predicted based on the first feature vector.

[0010] Calculate the difference between the target score data and the first score data;

[0011] If the difference value is greater than a preset threshold, the first feature vector corresponding to the first amino acid sequence data is updated using gradient backpropagation technology;

[0012] Alternatively, if the difference value is less than or equal to the preset threshold, the second amino acid sequence data is determined using the trained sequence decoding model based on the current first feature vector, and the target AAV capsid protein is obtained based on the second amino acid sequence data.

[0013] In addition, the optimization method for AAV capsid protein according to the above embodiments of this application may also have the following additional technical features:

[0014] Furthermore, in one embodiment of this application, the predetermined performance includes at least one of yield performance, various tissue targeting performance, or immune performance.

[0015] Furthermore, in one embodiment of this application, the scoring model includes a sequence feature extraction model and a regression prediction module; the step of extracting a first feature vector from the first amino acid sequence data and predicting the first scoring data based on the first feature vector includes:

[0016] The first feature vector is obtained by extracting the sequence feature extraction model.

[0017] The regression prediction module performs regression prediction on the first feature vector to obtain the first score data.

[0018] Furthermore, in one embodiment of this application, the scoring model is trained through the following steps:

[0019] Obtain the first training dataset; the first training dataset includes the third amino acid sequence data of AAV capsid proteins of several samples and the label score data of the predetermined performance;

[0020] The third amino acid sequence data is input into the initialized scoring model to obtain the second scoring data output by the scoring model;

[0021] Based on the second scoring data and the label scoring data, determine the first loss value for training;

[0022] Based on the first loss value, the parameters of the scoring model are updated to obtain the trained scoring model.

[0023] Furthermore, in one embodiment of this application, the sequence decoding model is trained through the following steps:

[0024] Obtain a second training dataset; the second training dataset includes the fourth amino acid sequence data of several samples of AAV capsid protein;

[0025] The fourth amino acid sequence data is input into the sequence feature extraction model in the scoring model to extract the second feature vector corresponding to the fourth amino acid sequence data.

[0026] The second feature vector is used as the input to the sequence decoding model, and the second loss value for training is determined based on the output value of the sequence decoding model and the fourth amino acid sequence.

[0027] Based on the second loss value, the parameters of the sequence decoding model are updated to obtain the trained sequence decoding model.

[0028] Furthermore, in one embodiment of this application, the step of updating the gradient of the first feature vector using gradient backpropagation technology includes:

[0029] Calculate the third loss value based on the first scoring data and the target scoring data;

[0030] Calculate the gradient of the third loss value with respect to the first feature vector;

[0031] The first feature vector is updated in the direction of the gradient descent.

[0032] On the other hand, embodiments of this application provide an optimization system for AAV capsid proteins, the system comprising:

[0033] The acquisition unit is used to acquire the first amino acid sequence data of the initial AAV capsid protein and the target score data of the predetermined performance.

[0034] The prediction unit is used to input the first amino acid sequence data into the trained scoring model, extract the first feature vector of the first amino acid sequence data, and predict the first score data based on the first feature vector.

[0035] A calculation unit is used to calculate the difference between the target score data and the first score data;

[0036] The first processing unit is configured to update the first feature vector corresponding to the first amino acid sequence data using gradient backpropagation technology if the difference value is greater than a preset threshold.

[0037] The second processing unit is used to determine the second amino acid sequence data based on the current first feature vector and the trained sequence decoding model if the difference value is less than or equal to the preset threshold, and to obtain the target AAV capsid protein based on the second amino acid sequence data.

[0038] Furthermore, in one embodiment of this application, the first processing unit is specifically used for:

[0039] Calculate the third loss value based on the first scoring data and the target scoring data;

[0040] Calculate the gradient of the third loss value with respect to the first feature vector;

[0041] The first feature vector is updated in the direction of the gradient descent.

[0042] On the other hand, embodiments of this application provide a terminal device, including:

[0043] At least one processor;

[0044] At least one memory for storing at least one program;

[0045] When the at least one program is executed by the at least one processor, the at least one processor implements the above-described optimization method for AAV capsid proteins.

[0046] On the other hand, embodiments of this application also provide a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to implement the above-described optimization method for an AAV capsid protein.

[0047] The advantages and beneficial effects of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application:

[0048] This application discloses an optimization method for AAV capsid proteins. The method includes: acquiring the first amino acid sequence data and target performance score data of the initial AAV capsid protein; inputting the first amino acid sequence data into a trained scoring model, extracting a first feature vector, and predicting the first score data; calculating the difference between the target score data and the first score data; if the difference value is greater than a preset threshold, updating the first feature vector using gradient backpropagation technology; if the difference value is less than or equal to the preset threshold, determining the second amino acid sequence data using a trained sequence decoding model based on the current first feature vector, and obtaining the target AAV capsid protein based on the second amino acid sequence data. This method can conveniently and efficiently find AAV variants that meet the requirements, greatly reducing the burden on staff. Furthermore, this method is compatible with any deep learning network, is simple to apply, and has a low implementation cost. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of this application or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0050] Figure 1 This is a flowchart illustrating an optimization method for AAV capsid protein provided in an embodiment of this application.

[0051] Figure 2 This is a schematic diagram illustrating the working principle of a scoring model provided in the embodiments of this application;

[0052] Figure 3 This is a schematic diagram illustrating the working principle of a sequence decoding model provided in the embodiments of this application;

[0053] Figure 4 This is a schematic diagram illustrating the working principle of an AAV capsid protein optimization algorithm provided in this application embodiment;

[0054] Figure 5 This is a schematic diagram illustrating the optimization of amino acid sequence data for an AAV capsid protein provided in the embodiments of this application;

[0055] Figure 6 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Detailed Implementation

[0056] The present application will be further described below with reference to the accompanying drawings and specific embodiments. The described embodiments should not be considered as limitations on the present application, and all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present application.

[0057] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0059] Before introducing the methods provided in the embodiments of this application, a brief description of some background technologies involved in this application will be given first.

[0060] Artificial Intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0061] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence. Its applications span all areas of artificial intelligence. Machine learning (deep learning) typically includes techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0062] Adeno-associated virus (AAV) is a type of single-stranded linear DNA defective virus. Due to its low genotoxicity in humans, it is one of the most promising gene therapy vectors. However, wild-type AAV has drawbacks such as insufficient tissue-specific targeting and low yield, and new AAVs are usually obtained by "capsid engineering".

[0063] In related technologies, when modifying new AAVs using capsid engineering, DNA libraries are typically constructed through random peptide display and family DNA shuffling. Natural selection is then simulated by targeted screening for specific characteristics to identify AAVs with those characteristics. However, to obtain AAV mutants that meet the target criteria, each DNA library requires multiple rounds of screening. This process demands high library capacity and sophisticated experimental techniques, is complex, and incurs high labor costs. Furthermore, since the library capacity often fails to cover all DNA mutation types, the resulting AAVs may not meet the intended requirements.

[0064] In view of this, this application provides an optimization method for AAV capsid proteins. This method, based on artificial intelligence and other technologies, trains a scoring model capable of evaluating the performance of AAV capsid proteins and a sequence decoding model capable of decoding feature vectors to obtain amino acid sequence data. Then, based on the target performance score data and the predicted first score data of the current AAV capsid protein, the amino acid sequence of the AAV capsid protein is optimized using gradient backpropagation technology, making its corresponding first score data approximate the target score data, thereby obtaining second amino acid sequence data that meets the performance requirements. Based on this data, the optimized target AAV capsid protein can be determined. The method provided in this application can conveniently and efficiently find AAV variants that meet the requirements, greatly reducing the burden on staff. Furthermore, this method is compatible with any deep learning network, is simple to apply, and has low implementation costs.

[0065] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for optimizing AAV capsid proteins according to an embodiment of this application. (Refer to...) Figure 1 The optimization method for this AAV capsid protein includes, but is not limited to:

[0066] Step 110: Obtain the first amino acid sequence data of the initial AAV capsid protein and the target score data for the predetermined performance;

[0067] In this step, when optimizing the AAV capsid protein, the desired target performance can generally be designed in advance. Based on the target performance, the target score data for the target AAV capsid protein in each predetermined performance aspect can be determined. Here, the types of predetermined performance can be flexibly set as needed, for example, they may include at least one of yield performance, various tissue targeting performance, or immune performance; this application does not limit this. For each performance, the score range can also be flexibly set, for example, it can be set to 1 to 10 points, with higher scores indicating better performance.

[0068] After designing the target score data for the desired AAV capsid protein, in this embodiment, a basic AAV capsid protein can be selected as the starting point for optimization, and this AAV capsid protein is denoted as the initial AAV capsid protein. It should be noted that, to improve the efficiency of finding the target AAV capsid protein, in this embodiment, the initial AAV capsid protein can be selected as closely as possible to the target score data; this application does not limit its specific type. Once the initial AAV capsid protein is selected, its amino acid sequence data can be obtained and denoted as the first amino acid sequence data.

[0069] Step 120: Input the first amino acid sequence data into the trained scoring model, extract the first feature vector of the first amino acid sequence data, and predict the first score data based on the first feature vector;

[0070] In this step, after obtaining the first amino acid sequence data of the initial AAV capsid protein, it can be input into the trained scoring model. In this embodiment, the scoring model is mainly used to predict the performance of the AAV capsid protein, providing score data for one or more specified performance aspects. Specifically, in this embodiment, the scoring model can be built using any machine learning algorithm, and this application does not limit its specific structure. Generally, the scoring model can include an input layer, a hidden layer, and an output layer. The input layer is used to input the amino acid sequence data, the hidden layer is used to extract the feature vector of the amino acid sequence data, and the output layer is used to output the score data corresponding to the AAV capsid protein. In this embodiment, the feature vector corresponding to the first amino acid sequence data is denoted as the first feature vector; the score data corresponding to the initial AAV capsid protein is denoted as the first score data. The first score data can include score data for one or more performance aspects, and this application does not limit this.

[0071] Step 130: Calculate the difference between the target score data and the first score data;

[0072] In this step, after obtaining the first score data corresponding to the initial AAV capsid protein, the difference value between the target score data and the first score data can be calculated. Here, the difference value is used to characterize the magnitude of the difference between the target score data and the first score data. It can be measured by any index. For example, in some embodiments, the difference value can be the absolute value of the difference between the two; in other embodiments, the difference value can be the proportion of the difference between the two to the target score data or the first score data; in still other embodiments, the difference value can be the mean square error of the two, etc. This application does not limit this. It is understood that in the embodiments of this application, the larger the difference value between the target score data and the first score data, the less the score data obtained by the current first amino acid sequence data is close to the pre-designed target score data. That is, the difference between the initial AAV capsid protein and the target AAV capsid protein is still large, so further optimization of the initial AAV capsid protein is needed. Conversely, the smaller the difference value between the target score data and the first score data, the closer the score data obtained by the current first amino acid sequence data is to the pre-designed target score data, and the closer the optimization process of the initial AAV capsid protein is to completion.

[0073] Step 140: If the difference value is greater than a preset threshold, update the gradient of the first feature vector using gradient backpropagation technology;

[0074] In this step, a preset threshold can be set in advance to measure whether the optimization process has achieved the expected results. If the difference value obtained earlier is greater than the preset threshold, it indicates that the current first score data has not yet met the requirements of the target score data. At this time, gradient backpropagation technology can be used to update the gradient of the first feature vector, that is, modify the first feature vector, and then return the obtained first feature vector to the aforementioned scoring model to re-predict the first score data based on the first feature vector. In this embodiment, gradient backpropagation technology calculates the gradient of the difference value with respect to the first feature vector, modifies and updates the first feature vector in the direction of gradient descent, thereby obtaining a new first feature vector. The first score data re-predicted based on the new first feature vector can reduce the difference value, that is, the score data of the amino acid sequence data corresponding to the newly obtained first feature vector can be closer to the target score data. In this way, the amino acid sequence data of the AAV capsid protein can be optimized once. In some embodiments, the difference value can be determined by a loss function. For example, a third loss value can be calculated based on the first score data and the target score data, and the third loss value is used as the difference value.

[0075] It should be noted that, in the embodiments of this application, the amino acid sequence data of the AAV capsid protein can be optimized multiple times, that is, the step of returning the first feature vector to predict the first score data each time can be executed multiple times. The optimization process continues until the difference value is less than or equal to a preset threshold, at which point the process exits.

[0076] Step 150: If the difference value is less than or equal to the preset threshold, determine the second amino acid sequence data using the trained sequence decoding model based on the current first feature vector, and obtain the target AAV capsid protein based on the second amino acid sequence data.

[0077] In this step, when the difference value is less than or equal to a preset threshold, it indicates that the score data of the amino acid sequence data corresponding to the current first feature vector is very close to the target score data, basically meeting the pre-designed requirements. Therefore, at this point, the amino acid sequence that meets the design requirements can be determined by reverse engineering using the sequence decoding model based on the current first feature vector, and recorded as the second amino acid sequence data. Based on the second amino acid sequence data, the optimized target AAV capsid protein can be determined.

[0078] It is understood that the AAV capsid protein optimization method provided in this application embodiment, based on artificial intelligence and other technologies, trains a scoring model that can score the performance of the AAV capsid protein and a sequence decoding model that can decode feature vectors to obtain amino acid sequence data. Then, based on the target score data of the predetermined performance and the first score data predicted for the current AAV capsid protein, the amino acid sequence of the AAV capsid protein is optimized using gradient backpropagation technology, so that the corresponding first score data approximates the target score data, thereby obtaining the second amino acid sequence data that meets the performance requirements. Based on this data, the optimized target AAV capsid protein can be determined. The method provided in this application embodiment can conveniently and efficiently find AAV variants that meet the requirements, greatly reducing the burden on staff. Moreover, this method is compatible with any deep learning network, is simple to apply, and has low implementation cost.

[0079] In some embodiments, refer to Figure 2 The scoring model includes a sequence feature extraction model and a regression prediction module. The sequence feature extraction model is a BERT network, and the regression prediction module consists of a bidirectional long short-term memory network and a fully connected layer connected in series. Extracting the first feature vector from the first amino acid sequence data and predicting the first scoring data based on the first feature vector includes:

[0080] The first feature vector is obtained by extracting the sequence feature extraction model.

[0081] The regression prediction module performs regression prediction on the first feature vector to obtain the first score data.

[0082] This application embodiment provides a specific structure for a scoring model, which may include a sequence feature extraction model, a bidirectional long short-term memory network (BSSN), and a fully connected layer. The sequence feature extraction model extracts feature vectors corresponding to amino acid sequence data. The BSSN performs deep feature extraction on the feature vectors. The fully connected layer acts as a regressor for regression prediction, outputting scoring data. In terms of parameter settings, the hidden layer size of the sequence feature extraction model (BERT) can be set to 768, with a total of 4 hidden layers. There are 8 attention heads. The intermediate layer size of the Transformer's encoding layer is set to 512, and the maximum embedding position is 15 / 50 (AAV9 / AAV2). The optimizer is Adam W, and the learning decay is cosine annealing (LR) with an initial learning rate of 0.00002. The states of the last two hidden layers of the BSSN are merged and used as the input to the regressor.

[0083] In some embodiments, the scoring model is trained through the following steps:

[0084] Obtain the first training dataset; the first training dataset includes the third amino acid sequence data of AAV capsid proteins of several samples and the label score data of the predetermined performance;

[0085] The third amino acid sequence data is input into the initialized scoring model to obtain the second scoring data output by the scoring model;

[0086] Based on the second scoring data and the label scoring data, determine the training loss value;

[0087] Based on the loss value, the parameters of the scoring model are updated to obtain the trained scoring model.

[0088] It should be noted that, in the embodiments of this application, the scoring model needs to be trained before it is put into use to adjust its internal parameters, thereby achieving better prediction results. Specifically, refer to... Figure 4 When training the model, the third amino acid sequence data corresponding to the AAV capsid protein of a batch of samples and the tag score data corresponding to the third amino acid sequence data can be obtained. The tag score data is used to characterize the true performance score of the AAV capsid protein of the samples, and this data can be obtained through experimental testing.

[0089] Then, each third amino acid sequence data and its corresponding tag score data can be used as a set of training data. The input data to the model is the third amino acid sequence data. The model predicts the third amino acid sequence data, and the output data of the model is the second score data. After obtaining the second score data output by the model, the accuracy of the model prediction can be evaluated based on the second score data and the tag score data, thereby updating the model parameters. The prediction process in the above procedure has been similarly described in the previous embodiments and will not be repeated here.

[0090] Specifically, for machine learning models, the accuracy of the output can be measured by a loss function. The loss function is defined on a single training data point and measures the prediction error of that data point. Specifically, the loss value is determined by the label of that single training data point and the model's output on that data point. However, in actual training, a training dataset contains many data points. Therefore, a cost function is generally used to measure the overall error of the training dataset. The cost function is defined on the entire training dataset and calculates the average prediction error of all training data points, providing a better measure of the model's prediction performance. For a typical machine learning model, the aforementioned cost function, plus a regularization term to measure model complexity, forms the training objective function. Based on this objective function, the loss value for the entire training dataset can be calculated.

[0091] Specifically, in related fields, there are many commonly used loss functions, such as 0-1 loss function, squared loss function, absolute loss function, log loss function, and cross-entropy loss function, all of which can be used as loss functions for machine learning models, and will not be elaborated on here. In the embodiments of this application, any one of these loss functions can be selected to determine the first loss value for training, such as the mean squared error loss function. Based on the first loss value, the parameters of the scoring model are updated using the backpropagation algorithm, and the trained scoring model can be obtained after several iterations. The specific number of iterations can be preset, or training can be considered complete when the accuracy requirement on the test set is met.

[0092] In some embodiments, the sequence decoding model is trained through the following steps:

[0093] Obtain a second training dataset; the second training dataset includes the fourth amino acid sequence data of several samples of AAV capsid protein;

[0094] The fourth amino acid sequence data is input into the sequence feature extraction model in the scoring model to extract the second feature vector corresponding to the fourth amino acid sequence data.

[0095] The second feature vector is used as the input to the sequence decoding model, and the second loss value for training is determined based on the output value of the sequence decoding model and the fourth amino acid sequence.

[0096] Based on the second loss value, the parameters of the sequence decoding model are updated to obtain the trained sequence decoding model.

[0097] In the embodiments of this application, reference is made to Figure 3 The role of the sequence decoding model is to decode feature vectors into amino acid sequence data. During training, a second training dataset can be acquired, which may include amino acid sequence data of several AAV capsid proteins, denoted as the fourth amino acid sequence data. This fourth amino acid sequence data can then be input into the sequence feature extraction model within the scoring model to extract the corresponding second feature vector. This second feature vector is then used as input to the sequence decoding model to decode and obtain the output value, which represents the amino acid sequence. Based on the output value and the fourth amino acid sequence, the accuracy of the sequence decoding model can be measured. Therefore, a second loss value can be determined using the output value and the fourth amino acid sequence, and this second loss value can be used for backpropagation to update the parameters of the sequence decoding model, resulting in a well-trained sequence decoding model.

[0098] Reference Figure 5 In this embodiment, the aforementioned method is used to optimize an initial AAV capsid protein. The target score for the predetermined performance before optimization is 2, and the score for this performance of the initial AAV capsid protein is approximately -2.25. The amino acid sequence data is then optimized and adjusted. Figure 5 The inner frame is divided into the differences between the original amino acid sequence and the optimized amino acid sequence. The final score of this performance of the target AAV capsid protein is about 1.993, which basically meets the design requirements. It can be seen that this method can conveniently and efficiently find AAV variants that meet the requirements.

[0099] This application embodiment also provides an optimization system for AAV capsid proteins, the system comprising:

[0100] The acquisition unit is used to acquire the first amino acid sequence data of the initial AAV capsid protein and the target score data of the predetermined performance.

[0101] The prediction unit is used to input the first amino acid sequence data into the trained scoring model, extract the first feature vector of the first amino acid sequence data, and predict the first score data based on the first feature vector.

[0102] A calculation unit is used to calculate the difference between the target score data and the first score data;

[0103] The first processing unit is used to update the first feature vector corresponding to the first amino acid sequence data using gradient backpropagation technology;

[0104] The second processing unit is used to determine the second amino acid sequence data based on the current first feature vector and the trained sequence decoding model if the difference value is less than or equal to the preset threshold, and to obtain the target AAV capsid protein based on the second amino acid sequence data.

[0105] Further, the first processing unit is specifically used for:

[0106] Calculate the third loss value based on the first scoring data and the target scoring data;

[0107] Calculate the gradient of the third loss value with respect to the first feature vector;

[0108] The first feature vector is updated in the direction of the gradient descent.

[0109] Understandable, Figure 1 The content of the AAV capsid protein optimization method embodiment shown is applicable to the AAV capsid protein optimization system embodiment. The specific functions implemented by the AAV capsid protein optimization system embodiment are the same as those in this embodiment. Figure 1 The optimization method for AAV capsid protein shown in this embodiment is the same, and the beneficial effects achieved are the same. Figure 1 The beneficial effects achieved by the example of the optimization method for AAV capsid protein shown are also the same.

[0110] Reference Figure 6 This application also discloses a terminal device, including:

[0111] At least one processor 201;

[0112] At least one memory 202 is used to store at least one program;

[0113] When at least one program is executed by at least one processor 201, such that at least one processor 201 performs as follows: Figure 1 An embodiment of an optimization method for AAV capsid protein is shown.

[0114] It is understandable that, such as Figure 1 The content of the AAV capsid protein optimization method embodiment shown is applicable to the embodiment of this terminal device. The specific functions implemented by the embodiment of this terminal device are the same as those shown below. Figure 1The optimization method for AAV capsid protein shown in the example is the same, and the beneficial effects achieved are the same as those described above. Figure 1 The beneficial effects achieved by the example of the optimization method for AAV capsid protein shown are also the same.

[0115] This application also discloses a computer-readable storage medium storing a processor-executable program, which, when executed by a processor, is used to implement, for example... Figure 1 An embodiment of an optimization method for AAV capsid protein is shown.

[0116] It is understandable that, such as Figure 1 The content of the AAV capsid protein optimization method embodiment shown is applicable to the embodiments of this computer-readable storage medium. The specific functions implemented in the embodiments of this computer-readable storage medium are the same as those shown below. Figure 1 The optimization method for AAV capsid protein shown in the example is the same, and the beneficial effects achieved are the same as those described above. Figure 1 The beneficial effects achieved by the example of the optimization method for AAV capsid protein shown are also the same.

[0117] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0118] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical system and / or software module, or one or more functions and / or features may be implemented in a separate physical system or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the system disclosed herein, the actual implementation of the module will be understood within the scope of ordinary skill of an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0119] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0120] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, system, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, system, or device). For the purposes of this specification, "computer-readable medium" can mean any system that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, system, or device.

[0121] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections with one or more wires (electronic systems), portable computer disk drives (magnetic systems), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic systems, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0122] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0123] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0124] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0125] The foregoing has provided a detailed description of the preferred embodiments of this application. However, this application is not limited to these embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

[0126] In the description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0127] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for optimizing AAV capsid proteins, characterized in that, The method includes: Obtain the first amino acid sequence data of the initial AAV capsid protein and the target performance score data for the predetermined performance; The first amino acid sequence data is input into the trained scoring model, the first feature vector of the first amino acid sequence data is extracted, and the first score data is predicted based on the first feature vector. Calculate the difference between the target score data and the first score data; If the difference value is greater than a preset threshold, the first feature vector corresponding to the first amino acid sequence data is updated using gradient backpropagation technology, and the process returns to the step of predicting the first score data based on the first feature vector. If the difference value is less than or equal to the preset threshold, the second amino acid sequence data is determined using the trained sequence decoding model based on the current first feature vector, and the target AAV capsid protein is obtained based on the second amino acid sequence data. The step of updating the gradient of the first feature vector using gradient backpropagation technology includes: Calculate the third loss value based on the first scoring data and the target scoring data; Calculate the gradient of the third loss value with respect to the first feature vector; The first feature vector is updated in the direction of the gradient descent.

2. The method for optimizing AAV capsid protein according to claim 1, characterized in that, The predetermined performance includes at least one of yield performance, various tissue targeting performance, or immune performance.

3. The method for optimizing AAV capsid protein according to claim 1, characterized in that, The scoring model includes a sequence feature extraction model and a regression prediction module; the step of extracting a first feature vector from the first amino acid sequence data and predicting the first scoring data based on the first feature vector includes: The first feature vector is obtained by extracting the sequence feature extraction model. The regression prediction module performs regression prediction on the first feature vector to obtain the first score data.

4. The method for optimizing AAV capsid protein according to claim 1, characterized in that, The scoring model is trained through the following steps: Obtain the first training dataset; the first training dataset includes the third amino acid sequence data of AAV capsid proteins of several samples and the label score data of the predetermined performance; The third amino acid sequence data is input into the initialized scoring model to obtain the second scoring data output by the scoring model; Based on the second scoring data and the label scoring data, determine the first loss value for training; Based on the first loss value, the parameters of the scoring model are updated to obtain the trained scoring model.

5. The method for optimizing AAV capsid protein according to claim 3, characterized in that, The sequence decoding model is trained through the following steps: Obtain a second training dataset; the second training dataset includes the fourth amino acid sequence data of several samples of AAV capsid protein; The fourth amino acid sequence data is input into the sequence feature extraction model in the scoring model to extract the second feature vector corresponding to the fourth amino acid sequence data. The second feature vector is used as the input to the sequence decoding model, and the second loss value for training is determined based on the output value of the sequence decoding model and the fourth amino acid sequence. Based on the second loss value, the parameters of the sequence decoding model are updated to obtain the trained sequence decoding model.

6. An optimization system for AAV capsid proteins, characterized in that, The system includes: The acquisition unit is used to acquire the first amino acid sequence data of the initial AAV capsid protein and the target score data of the predetermined performance. The prediction unit is used to input the first amino acid sequence data into the trained scoring model, extract the first feature vector of the first amino acid sequence data, and predict the first score data based on the first feature vector. A calculation unit is used to calculate the difference between the target score data and the first score data; The first processing unit is configured to update the first feature vector corresponding to the first amino acid sequence data using gradient backpropagation technology if the difference value is greater than a preset threshold, and return to execute the step of predicting the first score data based on the first feature vector. The second processing unit is used to determine the second amino acid sequence data based on the current first feature vector and the trained sequence decoding model if the difference value is less than or equal to the preset threshold, and to obtain the target AAV capsid protein based on the second amino acid sequence data. The first processing unit is specifically used for: Calculate the third loss value based on the first scoring data and the target scoring data; Calculate the gradient of the third loss value with respect to the first feature vector; The first feature vector is updated in the direction of the gradient descent.

7. A terminal device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements an optimization method for an AAV capsid protein as described in any one of claims 1-5.

8. A computer-readable storage medium storing a processor-executable program, characterized in that: The processor-executable program, when executed by the processor, is used to implement an optimization method for an AAV capsid protein as described in any one of claims 1-5.

Images