Multi-dimensional data feature extraction method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a multi-dimensional data feature extraction method and device, equipment and a storage medium, and relates to the technical field of data processing. The method comprises the following steps: acquiring discrete protein residue data and continuous protein structure data, then determining discrete network parameters and continuous network parameters, and inputting the protein residue data and the protein structure data as initial input data into a continuous discrete convolutional neural network model; and utilizing a continuous discrete convolution layer to update convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and performing feature extraction on the initial input data based on the updated convolution kernel parameters to obtain multi-dimensional representation output data. The continuous discrete convolution layer is used for fusing the feature extraction process of the continuous convolution and the feature extraction process of the discrete convolution according to the discrete network parameters and the continuous network parameters, the updated convolution kernel parameters are used for carrying out feature extraction on the multi-dimensional input data, and the accuracy of the feature extraction result of the multi-dimensional data is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, equipment and storage medium for extracting multidimensional data features. Background Art

[0002] Through multi-layer convolution and pooling operations, convolutional neural networks can effectively extract local features of data and gradually build high-level feature representations. However, in practical applications, many data sets do not contain only single-dimensional information, but are composed of information from multiple dimensions. These dimensions may include discrete features and continuous features, so feature extraction of multi-dimensional data needs to be considered in the data representation process based on convolutional neural networks.

[0003] In related technologies, discrete data usually has clear category boundaries, while continuous data may exhibit more complex distribution characteristics. When processing such multi-dimensional mixed data, convolutional neural networks often find it difficult to fully exploit the comprehensive features of discrete data and continuous data. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a multidimensional data feature extraction method, device, equipment and storage medium to improve the accuracy of feature extraction of multidimensional data by convolutional neural networks.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a multidimensional data feature extraction method, which is applied to a continuous discrete convolutional neural network model, wherein the continuous discrete convolutional neural network model includes at least a continuous discrete convolutional layer, and the method includes:

[0006] Obtain discrete protein residue data and continuous protein structure data;

[0007] determining discrete network parameters from at least one discrete convolutional network based on the protein residue data, and determining continuous network parameters based on the protein structure data;

[0008] The protein residue data and the protein structure data are input as initial input data into the continuous discrete convolutional neural network model, the continuous discrete convolution layer is used to update the convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and the initial input data is feature extracted based on the updated convolution kernel parameters to obtain multi-dimensional representation output data.

[0009] In one embodiment, the determining of discrete network parameters from at least one discrete convolutional network based on the protein residue data, and the determining of continuous network parameters based on the protein structure data, comprises:

[0010] Selecting a one-dimensional convolutional network from at least one discrete convolutional network according to the protein residue data, and obtaining a convolution kernel size parameter of the one-dimensional convolutional network as the discrete network parameter;

[0011] The convolution kernel space radius parameter of the continuous convolution network is selected as the continuous network parameter according to the protein structure data.

[0012] In one embodiment, the obtaining of discrete protein residue data and continuous protein structure data comprises:

[0013] Acquire initial protein data corresponding to at least one protein, each of the initial protein data comprising a protein identifier;

[0014] Obtaining initial residue information and / or initial structure information from the initial protein data;

[0015] Using the protein identifier, pairing the residue initial information and the structure initial information belonging to the same initial protein data to obtain at least one data pair;

[0016] In each of the data pairs, the residue initial information is encoded to obtain the protein residue data, and the structure initial information is encoded to obtain the protein structure data.

[0017] In one embodiment, encoding the residue initial information to obtain the protein residue data comprises:

[0018] Obtaining the residue type of each protein main chain atom in the initial residue information;

[0019] Integer-encode the residue type to obtain the residue type code corresponding to each atom of the protein main chain;

[0020] The protein residue data is obtained based on the residue type coding, and the protein residue data is discrete one-dimensional data.

[0021] In one embodiment, encoding the initial structure information to obtain the protein structure data includes:

[0022] Obtaining the three-dimensional coordinates of each protein main chain atom from the initial structural information;

[0023] The protein structure data is obtained according to the three-dimensional coordinates, and the protein structure data is continuous three-dimensional data.

[0024] In one embodiment, the updating of convolution kernel parameters using the continuous discrete convolution layer according to the discrete network parameters and the continuous network parameters includes:

[0025] Determine at least one three-dimensional offset according to the convolution kernel space radius parameter, and determine at least one one-dimensional offset according to the convolution kernel size parameter;

[0026] Calculate the sum of the three-dimensional offset and the convolution kernel space radius parameter to obtain a first intermediate value, and calculate the reciprocal of the first intermediate value as an offset parameter;

[0027] Obtaining a model parameter of the continuous discrete convolutional layer corresponding to the one-dimensional offset;

[0028] Obtaining an updated convolution kernel according to the product of the offset parameter and the model parameter;

[0029] The values ​​of the one-dimensional offset and / or the three-dimensional offset are changed one by one to obtain a plurality of offset combinations, and the convolution kernel parameters corresponding to each of the offset combinations are calculated.

[0030] In one embodiment, the feature extraction of the initial input data based on the updated convolution kernel to obtain multi-dimensional representation output data includes:

[0031] Determine the three-dimensional coordinates of the main chain atoms of the protein one by one from the initial input data as current coordinates;

[0032] Acquire the three-dimensional offsets one by one as current three-dimensional offsets, obtain neighboring coordinates according to the current coordinates and the current three-dimensional offsets, and acquire the residue type corresponding to the neighboring coordinates as the current residue type;

[0033] Obtaining at least one of the convolution kernel parameters corresponding to the current three-dimensional offset, multiplying the corresponding current residue type and the convolution kernel parameter to obtain a second intermediate value, and accumulating the second intermediate values ​​corresponding to all the three-dimensional offsets to obtain the characterization data corresponding to the current coordinates;

[0034] The multi-dimensional representation output data is obtained according to all the representation data.

[0035] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a multidimensional data feature extraction device, which is applied to a continuous discrete convolutional neural network model, wherein the continuous discrete convolutional neural network model includes at least a continuous discrete convolutional layer, and the device includes:

[0036] Data acquisition module: used to obtain discrete protein residue data and continuous protein structure data;

[0037] A model parameter acquisition module: used to determine discrete network parameters from at least one discrete convolutional network according to the protein residue data, and to determine continuous network parameters according to the protein structure data;

[0038] Continuous discrete convolution module: used to input the protein residue data and the protein structure data as initial input data into the continuous discrete convolutional neural network model, use the continuous discrete convolution layer to update the convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and extract features of the initial input data based on the updated convolution kernel parameters to obtain multi-dimensional representation output data.

[0039] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0040] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a storage medium, which is a storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0041] The multidimensional data feature extraction method, device, equipment and storage medium proposed in the embodiment of the present application obtains discrete protein residue data and continuous protein structure data, determines discrete network parameters from at least one discrete convolution network according to the protein residue data, and determines continuous network parameters according to the protein structure data, inputs the protein residue data and protein structure data as initial input data into the continuous discrete convolution neural network model, updates the convolution kernel parameters according to the discrete network parameters and the continuous network parameters using the continuous discrete convolution layer, and extracts features from the initial input data based on the updated convolution kernel parameters to obtain multidimensional characterization output data. After obtaining the discrete network parameters and the continuous network parameters in the embodiment of the present application, the feature extraction process of the continuous convolution and the discrete convolution is integrated according to the two network parameters using the continuous discrete convolution layer. By updating the convolution kernel parameters and extracting features from the multidimensional input data using the updated convolution kernel parameters, the complex features in the multidimensional data can be understood and characterized more comprehensively. This processing process converts the data into an efficient and information-rich representation form, which can not only improve the accuracy of the feature data obtained by feature extraction, but also improve the performance of the feature data in various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a structural diagram of the continuous discrete convolutional neural network model provided in the embodiment of the present application.

[0043] Figure 2 It is a flow chart of the multidimensional data feature extraction method provided in an embodiment of the present application.

[0044] Figure 3It is a flowchart for obtaining discrete protein residue data and continuous protein structure data provided in an embodiment of the present application.

[0045] Figure 4 It is a flow chart of encoding the initial residue information to obtain protein residue data provided in an embodiment of the present application.

[0046] Figure 5 It is a flowchart of encoding initial structural information to obtain protein structural data provided by an embodiment of the present application.

[0047] Figure 6 It is a schematic diagram of obtaining discrete protein residue data and continuous protein structure data provided in an embodiment of the present application.

[0048] Figure 7 It is a schematic diagram of determining network parameters provided in an embodiment of the present application.

[0049] Figure 8 It is a flowchart provided in an embodiment of the present application for determining discrete network parameters from at least one discrete convolutional network based on protein residue data and determining continuous network parameters based on protein structure data.

[0050] Fig. 9 It is a schematic diagram of the principle of a one-dimensional discrete convolutional network provided in an embodiment of the present application.

[0051] Fig.10 It is a schematic diagram of the principle of the continuous convolutional network provided in an embodiment of the present application.

[0052] Fig.11 This is a flowchart of updating convolution kernel parameters according to discrete network parameters and continuous network parameters using a continuous discrete convolution layer provided in an embodiment of the present application.

[0053] Fig.12 It is a flowchart provided in an embodiment of the present application for extracting features from initial input data based on an updated convolution kernel to obtain multi-dimensional representation output data.

[0054] Fig.13 It is a schematic diagram of multi-dimensional characterization provided in an embodiment of the present application.

[0055] Fig.14 This is a structural block diagram of a multidimensional data feature extraction device provided in yet another embodiment of the present application.

[0056] Fig.15 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0058] It should be noted that although the functional modules are divided in the device schematic and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0060] First, some nouns involved in this application are analyzed:

[0061] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0062] Convolutional neural network is an algorithm that can automatically extract useful features from data. It is widely used in image processing, speech recognition and other fields. It can extract important information from data through "convolution". Through multi-layer convolution and pooling operations, convolutional neural network can effectively extract local features of data and gradually build high-level feature representation. However, in practical applications, many data sets do not contain only single-dimensional information, but are composed of information from multiple dimensions. These dimensions may include discrete features and continuous features. Therefore, in the process of data representation based on convolutional neural network, it is necessary to consider the feature extraction of multi-dimensional data.

[0063] In related technologies, discrete data usually has clear category boundaries, while continuous data may exhibit more complex distribution characteristics. When processing such multi-dimensional mixed data, convolutional neural networks often find it difficult to fully exploit the comprehensive features of discrete data and continuous data.

[0064] Based on this, the embodiments of the present application provide a method, apparatus, device and storage medium for extracting features from multidimensional data. After obtaining discrete network parameters and continuous network parameters, the feature extraction processes of continuous convolution and discrete convolution are integrated using a continuous discrete convolution layer according to the two network parameters. By updating the convolution kernel parameters and extracting features from multidimensional input data using the updated convolution kernel parameters, the complex features in the multidimensional data can be more comprehensively understood and characterized. This processing converts the data into an efficient and information-rich representation, which can not only improve the accuracy of the feature data obtained by feature extraction, but also improve the performance of the feature data in various application scenarios.

[0065] The embodiments of the present application provide a multidimensional data feature extraction method, apparatus, device and storage medium, which are specifically described through the following embodiments. First, the multidimensional data feature extraction method in the embodiments of the present application is described.

[0066] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that the machines have the functions of perception, reasoning and decision-making.

[0067] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0068] The multidimensional data feature extraction method provided in the embodiment of the present application relates to the field of data processing technology. The multidimensional data feature extraction method provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server side, and can also be a computer program running in a terminal or a server side. For example, a computer program can be a native program or software module in an operating system; it can be a local (Native) application (Application, APP), that is, a program that needs to be installed in the operating system to run, such as a client that supports multidimensional data feature extraction, that is, a program that can be run only by downloading it to a browser environment; it can also be a small program that can be embedded in any APP. In short, the above-mentioned computer program can be an application, module or plug-in in any form. Among them, the terminal communicates with the server via a network. The multidimensional data feature extraction method can be executed by a terminal or a server, or by a terminal and a server in collaboration.

[0069] In some embodiments, the terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, or a smart watch. The server may be an independent server, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms; or a service node in a blockchain system, wherein each service node in the blockchain system forms a peer-to-peer (P2P) network, and the P2P protocol is an application layer protocol running on the Transmission Control Protocol (TCP) protocol. The terminal and the server may be connected via a communication connection such as Bluetooth, a Universal Serial Bus (USB) or a network, which is not limited in this embodiment.

[0070] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0071] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0072] First, the continuous discrete convolutional neural network model in the embodiment of the present application is described. Figure 1 , Figure 1 It is a structural diagram of the continuous discrete convolutional neural network model provided in the embodiment of the present application.

[0073] Reference Figure 1In the continuous discrete convolutional neural network model, the input data is sequentially sent to the batch normalization layer and the first linear layer, and then after being processed by the activation function, it is sent to the continuous discrete convolutional layer for multi-dimensional characterization to obtain multi-dimensional characterization output data, and then the multi-dimensional characterization output data is sent to the second linear layer to obtain the final output. Among them, the batch normalization layer normalizes the mean and variance of the input data according to the batch to reduce the offset of the internal covariate of the data. The data after batch normalization is sent to the first linear layer, which can be a fully connected layer. In this layer, each input node is connected to the output node through a weight to form a linear combination to capture the linear relationship in the data. The data output from the first linear layer is introduced through an activation function, such as ReLU, Sigmoid or Tanh, to introduce nonlinearity, so that the continuous discrete convolutional neural network model can learn and represent complex nonlinear relationships. Next, the continuous discrete convolutional layer performs a multi-dimensional characterization process to obtain multi-dimensional characterization output data, which contains the feature information of the input data in different dimensions. The multi-dimensional representation output data is then fed into the second linear layer for further feature integration and transformation to obtain the final output data. This output can be a classification result, regression value, or other form of prediction result, depending on the design goals and tasks of the continuous discrete convolutional neural network model.

[0074] In one embodiment, the batch normalization layer, the first linear layer, and the continuous discrete convolution layer are used as a feature extraction structure, and multiple feature extraction structures are cascaded in a cascade manner to further extract multi-dimensional features of the input data. For example, the number of cascades can be 8, which is not specifically limited in this embodiment.

[0075] Combine the following Figure 1 The multidimensional data feature extraction method in the embodiment of the present application is described.

[0076] Figure 2 is an optional flow chart of the multidimensional data feature extraction method provided in the embodiment of the present application. Figure 2 The method may include but is not limited to steps 110 to 130. It can also be understood that this embodiment Figure 2 The order of step 110 to step 130 is not specifically limited, and the order of steps can be adjusted or some steps can be reduced or added according to actual needs.

[0077] Step 110: Obtain discrete protein residue data and continuous protein structure data.

[0078] In one embodiment, the multidimensional data feature extraction method of the embodiment of the present application can be used for data in different fields. Characterizing multidimensional data can analyze and process multidimensional complex data and convert it into a more useful form so that it can be applied to tasks such as classification, prediction or optimization. The specific application scenarios are wide. For example, in the field of image processing, the spatial structure (discrete) of the image and the continuous motion of the object can be analyzed at the same time; in time series analysis, the changing trend of stock prices or meteorological data can be processed; in the field of biological data processing, such as the analysis of protein sequences and structures, accuracy and efficiency can also be improved. Through this method, complex data from different dimensions can be efficiently processed and converted into useful information, thereby improving the accuracy and efficiency of various tasks. The analysis process of protein sequences and structures is used as an example below for explanation, which does not limit specific data types and application scenarios.

[0079] In one embodiment, referring to Figure 3 , Figure 3 This is a flow chart of obtaining discrete protein residue data and continuous protein structure data provided by an embodiment of the present application, which specifically includes the following steps:

[0080] Step 310: Obtain initial protein data corresponding to at least one protein.

[0081] In one embodiment, protein-related data may be obtained from a database, wherein the protein-related data includes initial protein data corresponding to a plurality of proteins. In order to distinguish the data, a unique protein identifier is generated for each initial protein data.

[0082] Step 320: Obtaining initial residue information and / or initial structure information from the initial protein data.

[0083] In one embodiment, residue refers to a part of the amino acid molecule that constitutes a protein. When amino acids are linked to form a polypeptide chain through peptide bonds, each amino acid, except for the amino group (-NH2) and carboxyl group (-COOH) on its α-carbon atom that participate in the formation of peptide bonds, the rest (including the side chain R group) is called a residue. Residues are the basis of protein function, because different amino acid residues have different chemical properties and spatial structures, which directly affect the folding, stability and interaction of proteins with other molecules. The type of residue is determined by its side chain R group, and there are 20 standard amino acid residues, each with a unique single-letter or three-letter code. The structure of a protein usually refers to its three-dimensional spatial conformation, that is, the relative position and arrangement of all atoms in the protein molecule.

[0084] In one embodiment, when data is separated from the initial protein data, since the initial protein data may be incomplete, for example, some initial protein data can only obtain one of the residue initial information and the structure initial information, data separation is performed from the initial protein data to obtain the residue initial information, the structure initial information, or the residue initial information + the structure initial information.

[0085] Step 330: Use the protein identifier to pair the residue initial information and the structure initial information belonging to the same initial protein data to obtain at least one data pair.

[0086] In one embodiment, therefore, in the embodiment of the present application, both the residue initial information and the structure initial information are labeled accordingly according to the protein identifier corresponding to the initial protein data from which they are derived, that is, the residue initial information or the structure initial information from the same initial protein data has the same protein identifier. Therefore, the embodiment of the present application uses the protein identifier to pair the residue initial information and the structure initial information belonging to the same initial protein data to obtain at least one data pair, each of which contains the residue initial information and the structure initial information from the same initial protein data.

[0087] Step 340: In each data pair, the residue initial information is encoded to obtain protein residue data, and the structure initial information is encoded to obtain protein structure data.

[0088] In one embodiment, after the pairing is successful, the initial residue information needs to be encoded to obtain protein residue data and the initial structure information needs to be encoded to obtain protein structure data. Figure 4 , Figure 4 The flowchart of encoding the initial residue information to obtain protein residue data provided by the embodiment of the present application specifically includes the following steps:

[0089] Step 410: Obtain the residue type of each protein main chain atom in the initial residue information.

[0090] In one embodiment, residue type refers to the type of amino acid residues that make up the backbone of a protein. Each amino acid has a specific side chain (R group) that determines the chemical properties and functions of the amino acid. Therefore, when referring to residue type, it refers to these different types of amino acids, which are linked together by peptide bonds to form polypeptide chains, which are then folded into proteins with specific functions.

[0091] Step 420: integer-encode the residue type to obtain the residue type code corresponding to each protein main chain atom.

[0092] In one embodiment, the amino acid at each position in the protein sequence is represented by its corresponding single-letter code or three-letter code, for example, alanine can be represented as "A" or "Ala". These codes are standardized and are used to represent the 20 common amino acids and an additional symbol "X" to represent unknown or unspecified amino acid residues. Therefore, the present embodiment represents the residue type as p i (1≤i≤20,i∈N * ), N * Integer, use p 21 Indicates unknown residue types. The residue types of each protein main chain atom in the residue initial information are integer-encoded to obtain the residue type code corresponding to each protein main chain atom. For example, the residue type of a protein main chain atom is p i , then according to p i →i is mapped and represented as an integer i between 1 and 21, and the integer i is the corresponding residue type code.

[0093] Step 430: Obtain protein residue data based on the residue type encoding.

[0094] In one embodiment, all the residue types obtained are encoded as sequences, that is, the protein residue data P is obtained. seq , denoted as P seq =[q 1 ,q 2 ,…,q l ], where q∈{i|1≤i≤21,i∈N *}, l is the length of the residue sequence.

[0095] As can be seen from the above, protein residue data is discrete one-dimensional data, which describes the order of amino acids corresponding to the main chain atoms of the protein. Each residue type code in the protein residue data can be regarded as an independent unit, so the protein residue data is one-dimensional discrete data.

[0096] Next, refer to Figure 5 , Figure 5 The flowchart of encoding the initial structure information to obtain protein structure data provided by the embodiment of the present application specifically includes the following steps:

[0097] Step 510: Obtain the three-dimensional coordinates of each protein main chain atom from the initial structure information.

[0098] In one embodiment, the three-dimensional coordinates of the i-th protein backbone atom are represented by s i=(x,y,z)∈R3, the three-dimensional coordinates of the protein main chain atoms provide the conformational information of the protein in space. These coordinates are continuous because they are used to describe the specific positions of the protein main chain atoms in three-dimensional space. These positions can be any real numbers, not just integers or discrete values.

[0099] Step 520: Obtain protein structure data according to the three-dimensional coordinates.

[0100] In one embodiment, all three-dimensional coordinates are collected to obtain protein structure data P str , denoted as P str =[s 1 ,s 2 ,…,s l ], where l is the protein main chain atom C α The corresponding number of atoms is equivalent to the length of the residue sequence mentioned above. It can be seen that the protein structure data is continuous three-dimensional data.

[0101] In one embodiment, referring to Figure 6 , Figure 6 Schematic diagram of obtaining discrete protein residue data and continuous protein structure data provided by the embodiment of the present application. Figure 6 , obtain the initial protein data corresponding to at least one protein from the protein database, determine whether there is initial structural information in the initial protein data, if not, discard the corresponding data, if yes, decompose the initial protein data to obtain the initial residue information and the initial structural information. Then encode the initial residue information to obtain discrete protein residue data, and encode the initial structural information to obtain continuous protein structure data.

[0102] Step 120: Determine discrete network parameters from at least one discrete convolutional network based on the protein residue data, and determine continuous network parameters based on the protein structure data.

[0103] In one embodiment, referring to Figure 7 , Figure 7 : This is a schematic diagram of network parameter determination provided by an embodiment of the present application. Corresponding convolutional networks are selected for discrete data and continuous data respectively. For example, for discrete data, a corresponding network is selected from three discrete convolutional networks, such as a one-dimensional discrete convolutional network, a two-dimensional discrete convolutional network, or a three-dimensional discrete convolutional network, according to the corresponding data dimension. For continuous data, regardless of the data dimension, a continuous convolutional network is determined.

[0104] The following is an explanation of the application scenarios of protein sequences and structures in the embodiments of the present application. Specifically, discrete network parameters are determined from at least one discrete convolutional network based on protein residue data, and continuous network parameters are determined based on protein structure data. Figure 8 , Figure 8 The present invention provides a flowchart of determining discrete network parameters from at least one discrete convolutional network according to protein residue data and determining continuous network parameters according to protein structure data, which specifically includes the following steps:

[0105] Step 810: Select a one-dimensional convolutional network from at least one discrete convolutional network according to the protein residue data, and obtain a convolution kernel size parameter of the one-dimensional convolutional network as a discrete network parameter.

[0106] In one embodiment, since the protein residue data is one-dimensional discrete data, Figure 7 A one-dimensional discrete convolutional network is selected from the three discrete convolutional networks shown. At this time, the convolution kernel size parameters of the one-dimensional convolutional network are obtained as discrete network parameters.

[0107] Reference Fig. 9 , Fig. 9 This is a schematic diagram of the principle of a one-dimensional discrete convolutional network provided in an embodiment of the present application. For a one-dimensional discrete convolutional network, its convolution operation slides the input data through the convolution kernel and calculates the local features of each position. The formula is as follows:

[0108]

[0109] Among them, x(t) is the input data, w1(Δ) is the convolution kernel with a convolution kernel size parameter of k, and y(t) is the output data, which represents the convolution result of the input and the convolution kernel at time t. During the convolution calculation process, the convolution kernel w1 slides on the input data x, and the sliding range is determined by the convolution kernel size parameter k. The k shown in the figure is 3, and the dot product is calculated during the sliding process.

[0110] Taking the protein residue data in the embodiment of the present application as an example, if a one-dimensional discrete convolutional network is used for representation, it is expressed as:

[0111]

[0112] Here, q(t) is the residue type at the tth position of the input data.

[0113] In one embodiment, the convolution kernel size parameter k in the above formula is used as a discrete network parameter.

[0114] Step 820: Select the convolution kernel space radius parameter of the continuous convolution network as the continuous network parameter according to the protein structure data.

[0115] In one embodiment, the protein structure data is three-dimensional continuous data, so a continuous convolutional network is selected, and the convolution kernel space radius parameters of the continuous convolutional network are obtained as continuous network parameters.

[0116] Reference Fig.10 , Fig.10 : is a schematic diagram of the principle of the continuous convolutional network provided in the embodiment of the present application. For the continuous convolutional network, the kernel function w2 is parameterized using the displacement weight θ, and the formula is as follows:

[0117]

[0118] Among them, x(s) is the input data, w2(δ;θ) is The parameterized convolution kernel in the range, θ represents the model parameters of the continuous convolution network, y(s) is the output data, which represents the convolution result of the input data and the parameterized convolution kernel at the three-dimensional coordinate s, and δ represents the three-dimensional offset from the point to the three-dimensional coordinate s.

[0119] For the protein structure data in the embodiment of the present application, δ is a three-dimensional vector. The continuous convolution process is expressed as:

[0120]

[0121] Among them, q(s) is the residue type of the input data at the three-dimensional coordinate s, and r represents the radius parameter of the convolution kernel space, which is used to limit the convolution kernel space. the specific scope of The range is the spherical space shown in the figure, which contains multiple protein main chain atoms, and ||δ||≤r means that The offset corresponding to the point in the range.

[0122] From the above, we can see that the convolution kernel size parameter k is selected for discrete network parameters, and the convolution kernel space radius parameter r is selected for continuous network parameters. These two parameters can reflect the computational nature of discrete convolution and continuous convolution.

[0123] Step 130: Input the protein residue data and protein structure data as initial input data into the continuous discrete convolutional neural network model, use the continuous discrete convolution layer to update the convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and extract features of the initial input data based on the updated convolution kernel parameters to obtain multi-dimensional representation output data.

[0124] In one embodiment, referring to Fig.11 , Fig.11It is a flowchart of updating convolution kernel parameters according to discrete network parameters and continuous network parameters using a continuous discrete convolution layer provided in an embodiment of the present application, which specifically includes the following steps:

[0125] Step 1110: Determine at least one three-dimensional offset based on the convolution kernel space radius parameter, and determine at least one one-dimensional offset based on the convolution kernel size parameter.

[0126] In one embodiment, the three-dimensional offset δ is set to represent the parameter change of continuous convolution during the continuous discrete convolution process. Under the restriction of the convolution kernel space radius parameter r, the numerical range of the three-dimensional offset is expressed as: ||δ||≤r.

[0127] In addition, a one-dimensional offset Δ is set to represent the parameter change of discrete convolution during continuous discrete convolution. Under the restriction of the convolution kernel size parameter k, the numerical range of the one-dimensional offset is expressed as: 0≤Δ≤k-1.

[0128] Step 1120: Calculate the sum of the three-dimensional offset and the convolution kernel space radius parameter to obtain a first intermediate value, and calculate the reciprocal of the first intermediate value as the offset parameter.

[0129] In one embodiment, the first intermediate value is expressed as:

[0130] ||δ||+r

[0131] The offset parameter is expressed as:

[0132]

[0133] Step 1130: Obtain model parameters of the continuous discrete convolutional layer corresponding to the one-dimensional offset.

[0134] In one embodiment, the model parameter of the continuous discrete convolution layer corresponding to the one-dimensional offset is expressed as θ Δ , that is, the discrete convolution process slides on the sequence corresponding to the protein residue data, and the one-dimensional offset Δ can determine the sliding range. The trained continuous discrete convolution layer will have corresponding model parameters θ under different one-dimensional offsets Δ. Δ , model parameters mainly refer to the weights of the convolution kernel.

[0135] Step 1140: Obtain an updated convolution kernel according to the product of the offset parameter and the model parameter.

[0136] In one embodiment, for a continuous convolution process, the updated convolution kernel is expressed as:

[0137]

[0138] Step 1150: Change the values ​​of the one-dimensional offset and / or the three-dimensional offset one by one to obtain multiple offset combinations, and calculate the convolution kernel parameters corresponding to each offset combination.

[0139] In one embodiment, within the range of ||δ||≤r,0≤Δ≤k-1, the values ​​of the one-dimensional offset and / or the three-dimensional offset are changed one by one to obtain different (δ,Δ) combinations as offset combinations. Then, the convolution kernel parameter w(δ,Δ) corresponding to each offset combination is calculated.

[0140] Next, refer to Fig.12 , Fig.12 The present invention provides a flowchart of extracting features from initial input data based on an updated convolution kernel to obtain multi-dimensional representation output data, which specifically includes the following steps:

[0141] Step 1210: Determine the three-dimensional coordinates of the main chain atoms of the protein one by one from the initial input data as the current coordinates.

[0142] In one embodiment, referring to Fig.13 , Fig.13 is a schematic diagram of multi-dimensional characterization provided in the embodiment of the present application. Assuming that the i-th protein main chain atom is selected in the figure, its three-dimensional coordinate s i as the current coordinates.

[0143] Step 1220: Obtain the three-dimensional offsets one by one as the current three-dimensional offsets, obtain the neighboring coordinates according to the current coordinates and the current three-dimensional offsets, and obtain the residue type corresponding to the neighboring coordinates as the current residue type.

[0144] In one embodiment, referring to Fig.13 According to the current coordinates and the 3D offset, multiple neighboring coordinates can be determined within the range of the convolution kernel space radius parameter r, as shown in the filled circle in the figure. Each 3D offset corresponds to a neighboring coordinate. At this time, assuming that the jth 3D offset is selected, the current 3D offset is δ j , get the residue type of the protein main chain atom corresponding to the neighboring coordinates as the current residue type.

[0145] Step 1230: Obtain at least one convolution kernel parameter corresponding to the current three-dimensional offset, multiply the corresponding current residue type and the convolution kernel parameter to obtain a second intermediate value, and accumulate the second intermediate values ​​corresponding to all three-dimensional offsets to obtain the characterization data corresponding to the current three-dimensional offset.

[0146] In one embodiment, Fig.13In the figure, the black dotted line is used to represent the one-dimensional offset. It can be seen that the three-dimensional offset will correspond to multiple one-dimensional offsets. For example, in the figure, k = 6, and the protein main chain atom of the current coordinate is taken as the center. Three protein main chain atoms are obtained before and after it, corresponding to Δ = 0, Δ = 1, Δ = 2, Δ = 3, Δ = 4, and Δ = 5, respectively. At this time, the convolution kernel parameters corresponding to each corresponding one-dimensional offset are obtained, that is, when ||δ|| = δ j ,0≤Δ≤k-1 range (δ j ,Δ) combination, then the corresponding current residue type and convolution kernel parameter are multiplied to obtain the second intermediate value.

[0147] For example, the mth one-dimensional offset Δ m , the corresponding second intermediate value is expressed as:

[0148] q(s i +δ j )·w(δ j ,Δ m )

[0149] Next, the second intermediate value corresponding to the current three-dimensional offset is accumulated to obtain the intermediate representation data y(s) corresponding to the current three-dimensional offset. i , δ j ), expressed as:

[0150]

[0151] By accumulating the intermediate characterization data corresponding to all three-dimensional offsets, the characterization data corresponding to the main chain atoms of the protein at the current coordinates can be obtained, which is expressed as:

[0152]

[0153] From the above, it can be seen that the calculation process of the characterization data corresponding to each protein main chain atom combines continuous convolution and discrete convolution, so the obtained characterization data is more accurate.

[0154] Step 1240: Obtain multi-dimensional representation output data according to all the representation data.

[0155] In one embodiment, the above process is performed for each protein main chain atom to obtain corresponding characterization data, and finally the characterization data corresponding to all protein main chain atoms are output in the form of a sequence, so that multidimensional characterization output data corresponding to the initial input data can be obtained, which is expressed as: [y(s 1 ),y(s 2 ),…,y(s l )].

[0156] In one embodiment, a training process of a continuous discrete convolutional neural network model is described.

[0157] Different loss functions are selected according to different training tasks, such as cross-entropy loss (Cross-Entropy Loss) in classification tasks and mean squared error (MSE) in regression tasks. The following introduces these two loss functions respectively.

[0158] If the network's task is to classify proteins, such as predicting the functional category of proteins, then the cross entropy loss function is a common choice. For classification tasks, the cross entropy loss measures the gap between the predicted probability distribution of the network output and the probability distribution of the true category. The formula is as follows:

[0159]

[0160] Where C is the number of categories, y i is the true label, which is 1 for the correct class and 0 for other classes. is the probability of the i-th category predicted by the model. The purpose of the cross entropy loss is to maximize the probability output corresponding to the true category and reduce the output probability of other categories. For multi-class classification tasks, the continuous discrete convolutional neural network model is guided by back propagation to update the weights, thereby gradually improving the prediction of the correct category.

[0161] In the prediction of continuous variables (such as predicting the physical properties or activity of proteins), the mean square error is usually used. The mean square error calculates the average of the squared differences between the predicted value and the true value. The formula is as follows:

[0162]

[0163] Where N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample. MSE penalizes the square of the gap between the predicted value and the true value. The larger the gap, the greater the loss. Through back propagation, the continuous discrete convolutional neural network model will try to minimize this gap so that the predicted value is closer to the true value.

[0164] In addition, during the training of the continuous discrete convolutional neural network model, the RAdam optimizer was used to adjust the network parameters, and the initial learning rate was 1.0×10 -4 , the total training round is 150 epochs; at the same time, the learning rate adjustment strategy is introduced. When the training loss stagnates, the learning rate is appropriately reduced to ensure that the continuous discrete convolutional neural network model can continue to learn effectively. Through the above steps, the learning efficiency of the entire continuous discrete convolutional neural network model training process is improved, the convergence is accelerated, and the ideal performance can be achieved faster.

[0165] The embodiment of the present application utilizes a continuous discrete convolution layer to integrate discrete and continuous data to perform a unified feature extraction process. By performing a joint convolution operation on discrete data and continuous data, the feature representation of the data is extracted from multiple dimensions, such as one-dimensional sequences and three-dimensional coordinates, to form a unified high-dimensional feature representation, thereby improving the accuracy of multi-dimensional representation.

[0166] The technical solution provided in the embodiment of the present application obtains discrete protein residue data and continuous protein structure data, determines discrete network parameters from at least one discrete convolutional network according to the protein residue data, and determines continuous network parameters according to the protein structure data, inputs the protein residue data and protein structure data as initial input data into the continuous discrete convolutional neural network model, uses the continuous discrete convolution layer to update the convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and extracts features from the initial input data based on the updated convolution kernel parameters to obtain multidimensional characterization output data. After obtaining the discrete network parameters and the continuous network parameters in the embodiment of the present application, the feature extraction process of the continuous convolution and the discrete convolution is integrated according to the two network parameters using the continuous discrete convolution layer. By updating the convolution kernel parameters and extracting features from the multidimensional input data using the updated convolution kernel parameters, the complex features in the multidimensional data can be more comprehensively understood and characterized. This processing process converts the data into an efficient and information-rich representation form, which can not only improve the accuracy of the feature data obtained by feature extraction, but also improve the performance of the feature data in various application scenarios.

[0167] The present application also provides a multidimensional data feature extraction device, which can implement the multidimensional data feature extraction method. Fig.14 , the device comprises:

[0168] Data acquisition module 1410: used to acquire discrete protein residue data and continuous protein structure data.

[0169] Model parameter acquisition module 1420: used to determine discrete network parameters from at least one discrete convolutional network based on protein residue data, and determine continuous network parameters based on protein structure data.

[0170] Continuous discrete convolution module 1430: used to input protein residue data and protein structure data as initial input data into the continuous discrete convolution neural network model, use the continuous discrete convolution layer to update the convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and extract features of the initial input data based on the updated convolution kernel parameters to obtain multi-dimensional representation output data.

[0171] The specific implementation of the multidimensional data feature extraction device of this embodiment is basically the same as the specific implementation of the multidimensional data feature extraction method described above, and will not be repeated here.

[0172] The present application also provides an electronic device, including:

[0173] at least one memory;

[0174] at least one processor;

[0175] at least one program;

[0176] The program is stored in the memory, and the processor executes the at least one program to implement the multidimensional data feature extraction method implemented in the present application. The electronic device can be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a car computer, etc.

[0177] See also Fig.15 , Fig.15 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0178] The processor 1501 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0179] The memory 1502 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1502 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1502, and the processor 1501 calls and executes the multidimensional data feature extraction method of the embodiment of this application;

[0180] Input / output interface 1503, used to implement information input and output;

[0181] Communication interface 1504, used to realize communication interaction between the device and other devices, which can be realized by wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.); and

[0182] A bus 1505 that transmits information between the various components of the device (e.g., the processor 1501, the memory 1502, the input / output interface 1503, and the communication interface 1504);

[0183] The processor 1501 , the memory 1502 , the input / output interface 1503 and the communication interface 1504 are connected to each other in communication within the device via the bus 1505 .

[0184] An embodiment of the present application further provides a storage medium, which is a storage medium storing a computer program, and the computer program implements the above-mentioned multidimensional data feature extraction method when executed by a processor.

[0185] As a non-transitory storage medium, the memory can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0186] The multidimensional data feature extraction method, device, equipment and storage medium proposed in the embodiment of the present application obtains discrete protein residue data and continuous protein structure data, determines discrete network parameters from at least one discrete convolution network according to the protein residue data, and determines continuous network parameters according to the protein structure data, inputs the protein residue data and protein structure data as initial input data into the continuous discrete convolution neural network model, updates the convolution kernel parameters according to the discrete network parameters and the continuous network parameters using the continuous discrete convolution layer, and extracts features from the initial input data based on the updated convolution kernel parameters to obtain multidimensional characterization output data. After obtaining the discrete network parameters and the continuous network parameters in the embodiment of the present application, the feature extraction process of the continuous convolution and the discrete convolution is integrated according to the two network parameters using the continuous discrete convolution layer. By updating the convolution kernel parameters and extracting features from the multidimensional input data using the updated convolution kernel parameters, the complex features in the multidimensional data can be understood and characterized more comprehensively. This processing process converts the data into an efficient and information-rich representation form, which can not only improve the accuracy of the feature data obtained by feature extraction, but also improve the performance of the feature data in various application scenarios.

[0187] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0188] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0189] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0190] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0191] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0192] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0193] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0194] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0195] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0196] If the integrated unit is implemented in the form of 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 the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store programs.

[0197] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for extracting features from multidimensional data, characterized in that: Applied to a continuous discrete convolutional neural network model, the continuous discrete convolutional neural network model includes at least a continuous discrete convolutional layer, and the method includes: Obtain discrete protein residue data and continuous protein structure data; determining discrete network parameters from at least one discrete convolutional network based on the protein residue data, and determining continuous network parameters based on the protein structure data; The protein residue data and the protein structure data are input as initial input data into the continuous discrete convolutional neural network model, the continuous discrete convolution layer is used to update the convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and the initial input data is feature extracted based on the updated convolution kernel parameters to obtain multi-dimensional representation output data.

2. The multidimensional data feature extraction method according to claim 1, characterized in that: Determining discrete network parameters from at least one discrete convolutional network according to the protein residue data, and determining continuous network parameters according to the protein structure data, comprises: Selecting a one-dimensional convolutional network from at least one discrete convolutional network according to the protein residue data, and obtaining a convolution kernel size parameter of the one-dimensional convolutional network as the discrete network parameter; The convolution kernel space radius parameter of the continuous convolution network is selected as the continuous network parameter according to the protein structure data.

3. The multidimensional data feature extraction method according to claim 2, characterized in that: The step of obtaining discrete protein residue data and continuous protein structure data comprises: Acquire initial protein data corresponding to at least one protein, each of the initial protein data comprising a protein identifier; Obtaining initial residue information and / or initial structure information from the initial protein data; Using the protein identifier, pairing the residue initial information and the structure initial information belonging to the same initial protein data to obtain at least one data pair; In each of the data pairs, the residue initial information is encoded to obtain the protein residue data, and the structure initial information is encoded to obtain the protein structure data.

4. The multidimensional data feature extraction method according to claim 3, characterized in that: The step of encoding the initial residue information to obtain the protein residue data comprises: Obtaining the residue type of each protein main chain atom in the initial residue information; Integer-encode the residue type to obtain the residue type code corresponding to each atom of the protein main chain; The protein residue data is obtained based on the residue type coding, and the protein residue data is discrete one-dimensional data.

5. The multidimensional data feature extraction method according to claim 4, characterized in that: The step of encoding the initial structure information to obtain the protein structure data comprises: Obtaining the three-dimensional coordinates of each protein main chain atom from the initial structural information; The protein structure data is obtained according to the three-dimensional coordinates, and the protein structure data is continuous three-dimensional data.

6. The multidimensional data feature extraction method according to claim 5, characterized in that: The updating of convolution kernel parameters according to the discrete network parameters and the continuous network parameters by using the continuous discrete convolution layer includes: Determine at least one three-dimensional offset according to the convolution kernel space radius parameter, and determine at least one one-dimensional offset according to the convolution kernel size parameter; Calculate the sum of the three-dimensional offset and the convolution kernel space radius parameter to obtain a first intermediate value, and calculate the reciprocal of the first intermediate value as an offset parameter; Obtaining a model parameter of the continuous discrete convolutional layer corresponding to the one-dimensional offset; Obtaining an updated convolution kernel according to the product of the offset parameter and the model parameter; The values ​​of the one-dimensional offset and / or the three-dimensional offset are changed one by one to obtain a plurality of offset combinations, and the convolution kernel parameters corresponding to each of the offset combinations are calculated.

7. The multidimensional data feature extraction method according to claim 6, characterized in that: The extracting features of the initial input data based on the updated convolution kernel to obtain multi-dimensional representation output data includes: Determine the three-dimensional coordinates of the main chain atoms of the protein one by one from the initial input data as current coordinates; Acquire the three-dimensional offsets one by one as current three-dimensional offsets, obtain neighboring coordinates according to the current coordinates and the current three-dimensional offsets, and acquire the residue type corresponding to the neighboring coordinates as the current residue type; Obtaining at least one of the convolution kernel parameters corresponding to the current three-dimensional offset, multiplying the corresponding current residue type and the convolution kernel parameter to obtain a second intermediate value, and accumulating the second intermediate values ​​corresponding to all the three-dimensional offsets to obtain the characterization data corresponding to the current coordinates; The multi-dimensional representation output data is obtained according to all the representation data.

8. A multidimensional data feature extraction device, characterized in that: Applied to a continuous discrete convolutional neural network model, the continuous discrete convolutional neural network model includes at least a continuous discrete convolutional layer, and the device includes: Data acquisition module: used to obtain discrete protein residue data and continuous protein structure data; A model parameter acquisition module: used to determine discrete network parameters from at least one discrete convolutional network according to the protein residue data, and to determine continuous network parameters according to the protein structure data; Continuous discrete convolution module: used to input the protein residue data and the protein structure data as initial input data into the continuous discrete convolutional neural network model, use the continuous discrete convolution layer to update the convolution kernel parameters according to the discrete network parameters and the continuous network parameters, and extract features of the initial input data based on the updated convolution kernel parameters to obtain multi-dimensional representation output data.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the multidimensional data feature extraction method according to any one of claims 1 to 7 when executing the computer program.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the multidimensional data feature extraction method according to any one of claims 1 to 7 is implemented.

Images