Amino acid sequence three-dimensional structure inference method, device, equipment and medium

Abstract

The application discloses a three-dimensional structure reasoning method, device and equipment of an amino acid sequence and a medium, relates to the technical field of deep learning, and is applied to a reasoning device and comprises the following steps: splitting a reasoning model into a first sub-model and a second sub-model; performing operation on a to-be-reasoned amino acid sequence by using the first sub-model through a first processor; splitting the second sub-model, weight data of the second sub-model and an operation result by using a splitting unit determined based on dimension information through a second processor, compiling the split sub-model into an executable file, performing operation on the split operation result by using the split weight data and the executable file, and obtaining a plurality of operation results; splicing the plurality of operation results through the second processor, and determining the three-dimensional structure of the to-be-reasoned amino acid sequence by using the splicing result. The application solves the problems that the memory of a high-performance computing unit is limited and cannot well process dynamic reasoning operation.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and in particular to a method, apparatus, device, and medium for three-dimensional structural reasoning of amino acid sequences. Background Technology

[0002] The Alphafold model uses deep learning methods to infer the three-dimensional structure of proteins based solely on their amino acid sequences, which has potential value for biological and medical research.

[0003] However, the alphafold model requires a large amount of computation and is slow to infer on a central processing unit (CPU), making it difficult to be widely used in practice. To address this, researchers have tried to perform inference on high-performance computing units such as GPGPU (General Purpose Computing on GPU) and artificial intelligence accelerators. However, high-performance computing units often have memory limitations, and the three-dimensional structure of proteins is highly dynamic during inference, which high-performance computing units cannot handle well.

[0004] Therefore, the aforementioned technical problems urgently need to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for three-dimensional structural reasoning of amino acid sequences, which can solve the problems of limited memory in high-performance computing units and their inability to handle highly dynamic reasoning operations. The specific solution is as follows:

[0006] In a first aspect, this application discloses a three-dimensional structure reasoning method for amino acid sequences, applied to a reasoning device, comprising:

[0007] Determine the target dimension information based on the length information of the amino acid sequence to be inferred;

[0008] The target inference model is split into a first sub-model adapted to the computing characteristics of the local first processor and a second sub-model adapted to the computing characteristics of the local second processor; the second processor is a processor of a different type than the first processor and meets preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the inference device through a preset connection channel;

[0009] The first sub-model and the second sub-model are respectively assigned to the first processor and the second processor;

[0010] The first processor uses the first sub-model to perform calculations on the amino acid sequence to be inferred, and obtains a first calculation result.

[0011] The second processor uses the splitting unit determined based on the target dimension information to split the second sub-model, the weight data of the second sub-model, and the first calculation result. Then, the split sub-model is compiled into an executable file, and the split weight data and the executable file are used to perform calculations on the split calculation result to obtain multiple second calculation results.

[0012] The second processor splices together multiple second operation results, and uses the processing result obtained after processing the spliced ​​results to determine the three-dimensional structure of the amino acid sequence to be inferred.

[0013] Optionally, before the second processor splits the second sub-model, the weight data of the second sub-model, and the first calculation result using the splitting unit determined based on the target dimension information, the method further includes:

[0014] The weight data of the second sub-model stored in the device memory is copied to the memory unit through the preset connection channel so as to obtain the weight data of the second sub-model from the memory unit.

[0015] Optionally, the step of using the first processor and the first sub-model to perform calculations on the amino acid sequence to be inferred to obtain a first calculation result includes:

[0016] The first processor uses the first sub-model to compare the amino acid sequence to be inferred with the amino acid sequences in a preset gene database to obtain a first multiple sequence alignment matrix.

[0017] The first processor uses the first sub-model to compare the amino acid sequence to be reasoned with the structural data in the preset structure database to obtain the first structure comparison matrix.

[0018] Accordingly, the second processor splits the first calculation result, including:

[0019] The second processor splits the first multi-sequence alignment matrix and the first structure alignment matrix to obtain the split operation result containing the split first multi-sequence alignment matrix and the split first structure alignment matrix;

[0020] Furthermore, the executable file includes multiple neural network components that share the split weight data; correspondingly, the step of using the split weight data and the executable file to perform calculations on the split calculation results to obtain multiple second calculation results includes:

[0021] The split weight data and the neural network component are used to perform operations on the split first multi-sequence alignment matrix and the split first structure alignment matrix to obtain multiple second operation results corresponding to the split first multi-sequence alignment matrix and multiple second operation results corresponding to the split first structure alignment matrix, respectively.

[0022] Optionally, after performing operations on the split first multi-sequence alignment matrix and the split first structure alignment matrix based on the split weight data and the neural network component, the method further includes:

[0023] Release the weight data of the second sub-model in the memory unit.

[0024] Optionally, the step of concatenating multiple second operation results by the second processor includes:

[0025] The second processor concatenates multiple second operation results corresponding to the split first multi-sequence alignment matrix to obtain a second multi-sequence alignment matrix.

[0026] The second processor concatenates multiple second operation results corresponding to the split first structure alignment matrix to obtain a second structure alignment matrix.

[0027] Optionally, the process of determining the three-dimensional structure of the amino acid sequence to be deduced using the processing result obtained after processing the splicing result includes:

[0028] The second multiple sequence alignment matrix and the second structure alignment matrix are processed based on a preset attention mechanism to obtain a target weight matrix; wherein, the target weight matrix contains weight values ​​of multiple target amino acid sequences that match the amino acid sequence to be inferred.

[0029] Optionally, determining the three-dimensional structure of the amino acid sequence to be deduced using the processing result obtained after processing the splicing result includes:

[0030] The target weight matrix and the first multiple sequence alignment matrix are subjected to an outer product operation, and the characteristics of the amino acid sequence to be inferred are determined based on the average value of the outer product operation results, so as to obtain the three-dimensional structure of the amino acid sequence to be inferred based on the characteristics of the amino acid sequence to be inferred.

[0031] Secondly, this application discloses a three-dimensional structure inference device for amino acid sequences, applied to inference equipment, comprising:

[0032] The dimension determination module is used to determine the target dimension information based on the length information of the amino acid sequence to be inferred.

[0033] The first splitting module is used to split the target inference model into a first sub-model adapted to the computing characteristics of the local first processor and a second sub-model adapted to the computing characteristics of the local second processor; the second processor is a processor of a different type than the first processor and meets preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the inference device through a preset connection channel;

[0034] A model allocation module is used to allocate the first sub-model and the second sub-model to the first processor and the second processor respectively;

[0035] The first calculation module is used to perform calculations on the amino acid sequence to be inferred using the first sub-model through the first processor to obtain a first calculation result;

[0036] The second splitting module is used to split the second sub-model, the weight data of the second sub-model, and the first calculation result by the second processor using a splitting unit determined based on the target dimension information;

[0037] The second calculation module is used to compile the split sub-models into executable files, and to use the split weight data and the executable files to perform calculations on the split calculation results to obtain multiple second calculation results;

[0038] The three-dimensional structure determination module is used to splice multiple second operation results through the second processor, and to determine the three-dimensional structure of the amino acid sequence to be inferred by using the processing result obtained after processing the splicing result.

[0039] Thirdly, this application discloses an electronic device, including:

[0040] Memory, used to store computer programs;

[0041] A processor is used to execute the computer program to implement the aforementioned method for three-dimensional structural reasoning of amino acid sequences.

[0042] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned method for three-dimensional structural reasoning of amino acid sequences.

[0043] As can be seen, this application proposes a three-dimensional structure reasoning method for amino acid sequences, applied to a reasoning device, comprising: determining target dimension information based on the length information of the amino acid sequence to be reasoned; splitting the target reasoning model into a first sub-model adapted to the computational characteristics of a local first processor and a second sub-model adapted to the computational characteristics of a local second processor; the second processor is a processor of a different type from the first processor and meets preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the reasoning device through a preset connection channel; allocating the first sub-model and the second sub-model to the first processor and the second processor respectively; performing computation on the amino acid sequence to be reasoned using the first sub-model by the first processor to obtain a first computation result; splitting the second sub-model, the weight data of the second sub-model, and the first computation result by the second processor using the splitting unit determined based on the target dimension information, then compiling the split sub-model into an executable file, and performing computation on the split computation result using the split weight data and the executable file to obtain multiple second computation results; splicing the multiple second computation results by the second processor, and determining the three-dimensional structure of the amino acid sequence to be reasoned using the processing result obtained after processing the spliced ​​result.

[0044] Beneficial Effects: In the inference architecture of this application, the memory and the unit memory of the computing unit are interconnected. This allows the computing unit to perform inference operations using the memory interconnected with the unit memory, solving the problem of limited unit memory in high-performance computing units during inference. Furthermore, this application splits the target inference model into a first sub-model and a second sub-model, allocating the first sub-model to the central processing unit (CPU) and the second sub-model to the computing unit. This allows the CPU and the high-performance computing unit to perform their respective inference operations, reducing the workload of the high-performance computing unit and further improving the problem of limited unit memory during inference. In addition, this application has the high-performance computing unit perform the operations on the second sub-model, which involves a large amount of computation, improving computational performance while solving the memory limitation problem. Furthermore, this application proposes compiling the split sub-models into executable files, thus converting dynamic inference operations into a static computation graph, solving the problem that high-performance computing units cannot effectively handle highly dynamic inference operations. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0046] Figure 1 This is a flowchart of a three-dimensional structure reasoning method for an amino acid sequence disclosed in this application;

[0047] Figure 2 This is a schematic diagram of a reasoning device disclosed in this application;

[0048] Figure 3 This is a schematic diagram of the structure of a three-dimensional structure inference device for an amino acid sequence disclosed in this application;

[0049] Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] High-performance computing units (HPUs) often suffer from memory constraints, and the three-dimensional structure of proteins is highly dynamic during inference, which HPUs cannot effectively handle. Therefore, this application proposes a three-dimensional structure inference scheme for amino acid sequences, which can solve the problems of memory constraints and the inability of HPUs to handle highly dynamic inference operations.

[0052] To address this, this application proposes a three-dimensional structure reasoning scheme for amino acid sequences, which solves the problems of limited memory in high-performance computing units and their inability to handle highly dynamic reasoning operations effectively.

[0053] This application discloses a method for three-dimensional structural reasoning of amino acid sequences. See [link to relevant documentation]. Figure 1 As shown, the method includes:

[0054] Step S11: Determine the target dimension information based on the length information of the amino acid sequence to be inferred.

[0055] In this embodiment, the input dimension information includes target dimension information, which varies with the length of the amino acid sequence. Therefore, this embodiment should first determine the length of the amino acid sequence to be inferred, and then determine the target dimension information based on the length of the amino acid sequence to be inferred, so as to determine the splitting unit based on the target dimension information. For example, the size of the splitting unit is 4. It is understood that the size of the splitting unit in this application depends on the actual situation and is not limited here.

[0056] Step S12: The target inference model is split into a first sub-model adapted to the computing characteristics of the local first processor and a second sub-model adapted to the computing characteristics of the local second processor; the second processor is a processor of a different type than the first processor and meets the preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the inference device through a preset connection channel.

[0057] Figure 2 A schematic diagram of an inference device is provided. In this inference device, a first processor is connected to device memory, and a memory unit in a second processor accesses the device memory of the inference device through a preset connection channel. Since the second processor can directly interact with the memory unit, and the first processor can directly interact with the device memory, this application connects the device memory and the unit memory, solving the problems of the second processor not being able to directly interact with the device memory and the first processor not being able to interact with the second processor. For example, the first processor includes a central processing unit, and the second processor includes a high-performance computing unit.

[0058] In this embodiment, the target inference model is divided into a first sub-model adapted to the computing characteristics of the local first processor and a second sub-model adapted to the computing characteristics of the local second processor. It should be noted that computing characteristics refer to the different computing characteristics that different processors are suitable for performing different computing operations. For example, the central processing unit is suitable for processing conditions and branch statements, while the high-performance computing unit is suitable for processing matrix multiplication and convolution.

[0059] Step S13: Assign the first sub-model and the second sub-model to the first processor and the second processor respectively.

[0060] In this embodiment, the first sub-model and the second sub-model are respectively assigned to the first processor and the second processor, so that the first processor can perform corresponding calculations based on the first sub-model, and the second processor can perform corresponding calculations based on the second sub-model.

[0061] Step S14: The first processor performs calculations on the amino acid sequence to be inferred using the first sub-model to obtain the first calculation result.

[0062] In this embodiment, the first processor uses the first sub-model to compare the amino acid sequence to be inferred with the amino acid sequences in a preset gene database to obtain a first multiple sequence alignment matrix. The first processor also uses the first sub-model to compare the amino acid sequence to be inferred with the structural data in a preset structure database to obtain a first structure alignment matrix.

[0063] This embodiment completes the search of the gene database and the structure database through the first sub-model. The gene database contains amino acid sequences such as those of fish, rabbits, and humans, while the structure database contains the three-dimensional structural information of the corresponding amino acid sequences. Since this part mainly involves database query operations, the computation is performed on the central processing unit (CPU). In this way, this application reduces the workload of the high-performance computing unit by having the CPU handle a portion of the computation, resulting in a more balanced use of computing resources.

[0064] Step S15: The second processor splits the second sub-model, the weight data of the second sub-model, and the first calculation result using the splitting unit determined based on the target dimension information. Then, the split sub-model is compiled into an executable file, and the split weight data and the executable file are used to perform calculations on the split calculation result to obtain multiple second calculation results.

[0065] Considering the limitations of high-performance computing units, this embodiment requires the second processor to split the second sub-model, the weight data of the second sub-model, and the first operation result using a splitting unit determined based on the target dimension information. Specifically, the splitting of the first operation result by the second processor includes: splitting the first multi-sequence alignment matrix and the first structure alignment matrix by the second processor to obtain a split operation result containing the split first multi-sequence alignment matrix and the split first structure alignment matrix. Considering that high-performance computing units cannot handle highly dynamic inference operations well, this embodiment compiles the split sub-model into an executable file, realizing the conversion of dynamic inference operations into a static computation graph, and uses the split weight data and the executable file to perform operations on the split operation result to obtain multiple second operation results. Specifically, operations are performed on the split first multi-sequence alignment matrix and the split first structure alignment matrix based on the split weight data and the neural network component to obtain multiple second operation results corresponding to the split first multi-sequence alignment matrix and multiple second operation results corresponding to the split first structure alignment matrix, respectively. It should be noted that when using neural network components for computation, the results need to be processed using Dropout (random deactivation) to prevent overfitting.

[0066] In this embodiment, due to limitations in high-performance computing units, the weight data of the second sub-model is stored in the device memory and copied from the device memory to the memory unit when needed. Specifically, the weight data of the second sub-model stored in the device memory is copied to the memory unit through the preset connection channel so that the weight data of the second sub-model can be retrieved from the memory unit.

[0067] In other words, weight data is copied from device memory to the memory unit only when it is needed. Most of the time, the weight data remains in device memory, significantly saving the unit memory of the high-performance computing unit and avoiding the problem of limited unit memory. Furthermore, after the computation is completed, the weight data of the second sub-model in the memory unit is released to further save unit memory and avoid the problem of limited unit memory.

[0068] Step S16: The second processor splices together multiple second operation results, and the processing result obtained after processing the splicing results is used to determine the three-dimensional structure of the amino acid sequence to be inferred.

[0069] In this embodiment, the second processor concatenates multiple second operation results corresponding to the split first multi-sequence alignment matrix to obtain a second multi-sequence alignment matrix. Similarly, the second processor concatenates multiple second operation results corresponding to the split first structure alignment matrix to obtain a second structure alignment matrix. Further, the second multi-sequence alignment matrix and the second structure alignment matrix are processed based on a preset attention mechanism to obtain a target weight matrix. The target weight matrix contains weight values ​​for multiple target amino acid sequences that match the amino acid sequence to be inferred. An outer product operation is performed between the target weight matrix and the first multi-sequence alignment matrix. The features of the amino acid sequence to be inferred are determined based on the average value of the outer product result, so that the three-dimensional structure of the amino acid sequence to be inferred can be obtained based on these features.

[0070] In this embodiment, during the process of processing the second multiple sequence alignment matrix and the second structure alignment matrix based on the preset attention mechanism to obtain the target weight matrix, specifically, the column vectors of the second multiple sequence alignment matrix and the second structure alignment matrix are processed based on the attention mechanism to obtain the specific amino acid sequences, and the row vectors of the second multiple sequence alignment matrix and the second structure alignment matrix are processed based on the attention mechanism to obtain the species information of the amino acid sequences. That is, the target weight may contain three amino acid sequences, which are the amino acid sequences of humans, fish, and birds, respectively, with probabilities of 0.8, 0.1, and 0.1, respectively. It should be noted that since this submodule involves many vector calculations, this module is computed on a high-performance computing unit to obtain better performance.

[0071] In summary, this application determines the dimensionality of the tensors in the inference model based on the length of the amino acid sequence to be inferred, and segments the inference process. This allows the resulting sub-models to be executed on either a high-performance computing unit (HPU) or a central processing unit (CPU). Furthermore, for sub-models executable on a computer board, they are compiled into HPU-executable files according to a static computation graph. In this application, the target dimension of each sub-model is consistent with the segmentation unit; therefore, the segmented sub-models are structurally repetitive. This allows for the generation of executable files with a single compilation operation, significantly reducing compilation time and saving computational resources. Furthermore, not all sub-models are executed by the HPU. Sub-models suitable for CPU processing are computed on the CPU, improving overall performance and addressing the memory limitations of HPUs.

[0072] For example, a certain company's T100 board is a high-performance computing unit. Early versions of this board only supported static graph computation. Using the method described in this application, the task of performing alphafold model inference in cooperation between the T100 and the central processing unit was completed, achieving higher performance than the pure central processing unit inference scheme.

[0073] Furthermore, based on the method described in this application, this application can be further improved. For example, for the GFPGAN model for face reconstruction, the GFPGAN model can be split into multiple sub-models, and the split sub-models can be allocated to the central processing unit and the high-performance computing unit. Then, the central processing unit can complete the corresponding reconstruction operation. Furthermore, the operation result of the central processing unit can be sent to the sub-model on the high-performance computing unit so that the sub-model on the high-performance computing unit can complete the final operation. In this way, the application of the GFPGAN model in the face blur reconstruction scenario can be promoted.

[0074] Accordingly, embodiments of this application also disclose a three-dimensional structure inference device for amino acid sequences, see [link to relevant documentation]. Figure 3 As shown, the device includes:

[0075] The dimension determination module 11 is used to determine the target dimension information based on the length information of the amino acid sequence to be inferred.

[0076] The first splitting module 12 is used to split the target inference model into a first sub-model adapted to the computing characteristics of the local first processor and a second sub-model adapted to the computing characteristics of the local second processor; the second processor is a processor of a different type than the first processor and meets preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the inference device through a preset connection channel;

[0077] Model allocation module 13 is used to allocate the first sub-model and the second sub-model to the first processor and the second processor respectively;

[0078] The first calculation module 14 is used to perform calculations on the amino acid sequence to be reasoned using the first sub-model through the first processor to obtain a first calculation result;

[0079] The second splitting module 15 is used to split the second sub-model, the weight data of the second sub-model, and the first calculation result by the second processor using a splitting unit determined based on the target dimension information;

[0080] The second calculation module 16 is used to compile the split sub-models into executable files, and to use the split weight data and the executable files to perform calculations on the split calculation results to obtain multiple second calculation results.

[0081] The three-dimensional structure determination module 17 is used to splice multiple second operation results through the second processor, and use the processing result obtained after processing the splicing result to determine the three-dimensional structure of the amino acid sequence to be inferred.

[0082] For more detailed information on the working process of each of the above modules, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.

[0083] As can be seen, this application proposes a three-dimensional structure reasoning method for amino acid sequences, applied to a reasoning device, comprising: determining target dimension information based on the length information of the amino acid sequence to be reasoned; splitting the target reasoning model into a first sub-model adapted to the computational characteristics of a local first processor and a second sub-model adapted to the computational characteristics of a local second processor; the second processor is a processor of a different type from the first processor and meets preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the reasoning device through a preset connection channel; allocating the first sub-model and the second sub-model to the first processor and the second processor respectively; performing computation on the amino acid sequence to be reasoned using the first sub-model by the first processor to obtain a first computation result; splitting the second sub-model, the weight data of the second sub-model, and the first computation result by the second processor using the splitting unit determined based on the target dimension information, then compiling the split sub-model into an executable file, and performing computation on the split computation result using the split weight data and the executable file to obtain multiple second computation results; splicing the multiple second computation results by the second processor, and determining the three-dimensional structure of the amino acid sequence to be reasoned using the processing result obtained after processing the spliced ​​result.

[0084] Beneficial Effects: In the inference architecture of this application, the memory and the unit memory of the computing unit are interconnected. This allows the computing unit to perform inference operations using the memory interconnected with the unit memory, solving the problem of limited unit memory in high-performance computing units during inference. Furthermore, this application splits the target inference model into a first sub-model and a second sub-model, allocating the first sub-model to the central processing unit (CPU) and the second sub-model to the computing unit. This allows the CPU and the high-performance computing unit to perform their respective inference operations, reducing the workload of the high-performance computing unit and further improving the problem of limited unit memory during inference. In addition, this application has the high-performance computing unit perform the operations on the second sub-model, which involves a large amount of computation, improving computational performance while solving the memory limitation problem. Furthermore, this application proposes compiling the split sub-models into executable files, thus converting dynamic inference operations into a static computation graph, solving the problem that high-performance computing units cannot effectively handle highly dynamic inference operations.

[0085] Furthermore, embodiments of this application also provide an electronic device. Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0086] Figure 4This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a display screen 23, an input / output interface 24, a communication interface 25, a power supply 26, and a communication bus 27. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the three-dimensional structure reasoning method for amino acid sequences disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0087] In this embodiment, the power supply 26 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 25 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 24 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0088] Furthermore, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored thereon may include computer programs 221, and the storage method may be temporary storage or permanent storage. In addition to including a computer program capable of performing the three-dimensional structure reasoning method of amino acid sequences executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 221 may further include computer programs capable of performing other specific tasks.

[0089] Furthermore, embodiments of this application also disclose a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned method for three-dimensional structural reasoning of amino acid sequences.

[0090] For the specific steps of this method, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.

[0091] Furthermore, embodiments of this application also disclose a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the three-dimensional structure reasoning method for amino acid sequences as disclosed above.

[0092] The various embodiments in this application are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. For the same or similar parts between the various embodiments, refer to each other. As for the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and relevant parts can be referred to in the method section.

[0093] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0094] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0095] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0096] The above provides a detailed description of the method, apparatus, device, and storage medium for three-dimensional structural reasoning of amino acid sequences provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for three-dimensional structural deduction of an amino acid sequence, characterized in that, Applied to inference devices, including: Determine the target dimension information based on the length information of the amino acid sequence to be inferred; The target inference model is split into a first sub-model adapted to the computing characteristics of the local first processor and a second sub-model adapted to the computing characteristics of the local second processor; the second processor is a processor of a different type than the first processor and meets preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the inference device through a preset connection channel; The first sub-model and the second sub-model are respectively assigned to the first processor and the second processor; The first processor uses the first sub-model to perform calculations on the amino acid sequence to be inferred, and obtains a first calculation result. The second processor uses the splitting unit determined based on the target dimension information to split the second sub-model, the weight data of the second sub-model, and the first calculation result. Then, the split sub-model is compiled into an executable file, and the split weight data and the executable file are used to perform calculations on the split calculation result to obtain multiple second calculation results. The second processor splices together multiple second operation results, and uses the processing result obtained after processing the spliced ​​results to determine the three-dimensional structure of the amino acid sequence to be inferred.

2. The method for three-dimensional structural reasoning of amino acid sequences according to claim 1, characterized in that, Before the second processor splits the second sub-model, the weight data of the second sub-model, and the first calculation result using the splitting unit determined based on the target dimension information, the process further includes: The weight data of the second sub-model stored in the device memory is copied to the memory unit through the preset connection channel so as to obtain the weight data of the second sub-model from the memory unit.

3. The method for three-dimensional structural reasoning of amino acid sequences according to claim 2, characterized in that, The step of performing calculations on the amino acid sequence to be reasoned using the first sub-model through the first processor to obtain a first calculation result includes: The first processor uses the first sub-model to compare the amino acid sequence to be inferred with the amino acid sequences in a preset gene database to obtain a first multiple sequence alignment matrix. The first processor uses the first sub-model to compare the amino acid sequence to be reasoned with the structural data in the preset structure database to obtain the first structure comparison matrix. Accordingly, the second processor splits the first calculation result, including: The second processor splits the first multi-sequence alignment matrix and the first structure alignment matrix to obtain the split operation result containing the split first multi-sequence alignment matrix and the split first structure alignment matrix; Furthermore, the executable file includes multiple neural network components that share the split weight data; correspondingly, the step of using the split weight data and the executable file to perform calculations on the split calculation results to obtain multiple second calculation results includes: The split weight data and the neural network component are used to perform operations on the split first multi-sequence alignment matrix and the split first structure alignment matrix to obtain multiple second operation results corresponding to the split first multi-sequence alignment matrix and multiple second operation results corresponding to the split first structure alignment matrix, respectively.

4. The method for three-dimensional structural reasoning of amino acid sequences according to claim 3, characterized in that, After performing calculations on the split first multi-sequence alignment matrix and the split first structure alignment matrix based on the split weight data and the neural network component, the method further includes: Release the weight data of the second sub-model in the memory unit.

5. The method for three-dimensional structural reasoning of amino acid sequences according to claim 3, characterized in that, The step of concatenating multiple second operation results through the second processor includes: The second processor concatenates multiple second operation results corresponding to the split first multi-sequence alignment matrix to obtain a second multi-sequence alignment matrix. The second processor concatenates multiple second operation results corresponding to the split first structure alignment matrix to obtain a second structure alignment matrix.

6. The method for three-dimensional structural reasoning of amino acid sequences according to claim 5, characterized in that, The process of determining the three-dimensional structure of the amino acid sequence to be deduced using the processing results obtained after processing the splicing results includes: The second multiple sequence alignment matrix and the second structure alignment matrix are processed based on a preset attention mechanism to obtain a target weight matrix; wherein, the target weight matrix contains weight values ​​of multiple target amino acid sequences that match the amino acid sequence to be inferred.

7. The method for three-dimensional structural reasoning of amino acid sequences according to claim 6, characterized in that, The process of determining the three-dimensional structure of the amino acid sequence to be deduced using the processing result obtained after processing the splicing result includes: The target weight matrix and the first multiple sequence alignment matrix are subjected to an outer product operation, and the characteristics of the amino acid sequence to be inferred are determined based on the average value of the outer product operation results, so as to obtain the three-dimensional structure of the amino acid sequence to be inferred based on the characteristics of the amino acid sequence to be inferred.

8. A three-dimensional structure inference device for an amino acid sequence, characterized in that, Applied to inference devices, including: The dimension determination module is used to determine the target dimension information based on the length information of the amino acid sequence to be inferred. The first splitting module is used to split the target inference model into a first sub-model adapted to the computing characteristics of the local first processor and a second sub-model adapted to the computing characteristics of the local second processor; the second processor is a processor of a different type than the first processor and meets preset high-performance conditions, and the memory unit in the second processor accesses the device memory of the inference device through a preset connection channel; A model allocation module is used to allocate the first sub-model and the second sub-model to the first processor and the second processor respectively; The first calculation module is used to perform calculations on the amino acid sequence to be inferred using the first sub-model through the first processor to obtain a first calculation result; The second splitting module is used to split the second sub-model, the weight data of the second sub-model, and the first calculation result by the second processor using a splitting unit determined based on the target dimension information; The second calculation module is used to compile the split sub-models into executable files, and to use the split weight data and the executable files to perform calculations on the split calculation results to obtain multiple second calculation results; The three-dimensional structure determination module is used to splice multiple second operation results through the second processor, and to determine the three-dimensional structure of the amino acid sequence to be inferred by using the processing result obtained after processing the splicing result.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the three-dimensional structure reasoning method for amino acid sequences as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the three-dimensional structure reasoning method for amino acid sequences as described in any one of claims 1 to 7.

Images