An interaction prediction method and system based on graph neural networks

Abstract

The application discloses an interaction prediction method and system based on a graph neural network, wherein the interaction prediction method of one embodiment comprises the following steps: receiving a relation network graph and inputting the graph neural network, the relation network graph comprising a plurality of nodes and edges connecting the nodes, the nodes representing entities and the edges representing relations between the connected entities; the graph neural network calculating difference information of each node and neighbor nodes of each node respectively, iteratively updating each node according to the difference information, and outputting a representation vector of each node; and a prediction module receiving the representation vector of each node, performing node pair prediction according to the representation vector, and outputting a prediction result. The interaction prediction method provided by the application updates the representation vector of each node by using difference information and similarity information of the current node and the neighbor nodes, can reduce iteration noise, accelerate iteration speed, and improve iteration efficiency.

Description

Technical Field

[0001] This invention relates to the field of graph neural network technology, and in particular to an interaction prediction method and system based on graph neural networks. Background Technology

[0002] Protein-protein interaction studies can reveal protein functions at the molecular level, helping to elucidate the patterns of cellular activities such as growth, development, metabolism, differentiation, and apoptosis. Identifying protein-protein interaction pairs across the entire genome is a crucial step in explaining cellular regulatory mechanisms. With advancements in protein-protein interaction experimental techniques, we can obtain vast amounts of protein-protein interaction data, even enabling genome-wide analysis. However, due to limitations in experimental techniques, traditional methods are not suitable for large-scale detection. Utilizing existing protein-protein interaction data, machine learning methods can facilitate large-scale detection.

[0003] Many current methods rely on graph neural networks and protein-protein interaction networks for prediction. In protein-protein interaction networks, vertices represent proteins, and pairs represent interactions between corresponding proteins. The prediction task is then accomplished using the graph's topology through methods such as graph neural networks or random walks. However, in practical applications, prediction accuracy has certain limitations. Summary of the Invention

[0004] To address at least one of the aforementioned problems, a first aspect of the present invention provides an interaction prediction method based on graph neural networks, comprising:

[0005] Receive a relational network graph and input it into a graph neural network. The relational network graph includes multiple nodes and edges connecting each node. The nodes represent entities, and the edges represent the relationships between the connected entities.

[0006] The graph neural network calculates the difference information of each node and its neighboring nodes, iteratively updates each node based on the difference information, and outputs the representation vector of each node.

[0007] The prediction module receives and performs node pair prediction based on the representation vectors of each node, and outputs the prediction results.

[0008] For example, in the interaction prediction method provided in some embodiments of this application, the graph neural network calculates the difference information of each node and its neighboring nodes, iteratively updates each node based on the difference information, and outputs the representation vector of each node. This further includes performing the following steps in each iteration:

[0009] The graph neural network calculates the difference information of each node and its neighboring nodes.

[0010] The message is reconstructed based on the difference information, and each node is iteratively updated through message passing to output the representation vector of each node.

[0011] For example, in the interaction prediction method provided in some embodiments of this application, the graph neural network further includes calculating the difference information of each node and its neighboring nodes:

[0012]

[0013] The relational network graph includes N nodes, h t ( i ) is node e i and all its neighboring nodes e j The difference information at time t, h t (e i ) represents node e at time t i The representation vector, h t (e j ) represents node e at time t. i Neighbor node e j The representation vector, N(i) represents node e. i The neighbor node number, α i,j For parameters.

[0014] For example, in the interaction prediction method provided in some embodiments of this application, the α i,j To ensure that all neighboring nodes of the current node best fit the parameters of the current node, the following condition must be met:

[0015]

[0016] Among them, h t (e i ) represents node e at time t. i The representation vector, h t (e j ) represents node e at time t i Neighbor node e j The representation vector.

[0017] For example, in the interaction prediction method provided in some embodiments of this application, the α i,j Further satisfy:

[0018]

[0019] Among them, h t (ei ) represents node e at time t i The representation vector, h t (e j ) represents node e at time t i Neighbor node e j The vector represents λ, where λ is a positive integer greater than 0. Regularization parameters to prevent overfitting.

[0020] For example, in some embodiments of the interaction prediction method provided in this application, the step of reconstructing the message based on the difference information and iteratively updating each node through message passing to output the representation vector of each node further includes:

[0021]

[0022] Among them, h t+1 (e i ) represents node e at time t+1. i The representation vector, σ is the nonlinear activation function, W1 and W2 are parameter matrices, h t ( i ) is node e i and all its neighboring nodes e j The difference information at time t, h t (e j ) represents node e at time t i Neighbor node e j The representation vector, N(i) represents node e. i The neighboring node numbers.

[0023] For example, in some embodiments of the interaction prediction method provided in this application, the prediction module receiving and performing node pair prediction based on the representation vectors of each node and outputting the prediction result further includes:

[0024] y = γ(W s [h 20 (pi),h 20 (pj),h 20 (pi)-h 20 (pj),h 20 (pi)·h 20 (pj)]+b s )

[0025] Where y represents the prediction result for the node pair of two nodes pi and pj, γ is the sigmoid function, and W s Let h be the parameter matrix. 20 (pi) is the representation vector of node pi, h 20(pj) is the representation vector of node pj, where nodes pi and pj are two nodes connected by an edge in the relational network graph, h 20 (pi)-h 20 (pj) is the difference between the representation vectors of the two nodes, h 20 (pi)·h 20 (pj) is the product of the representation vectors of the two nodes, [,] indicates concatenation, b s For bias.

[0026] A second aspect of the present invention provides an interaction prediction system based on a graph neural network, comprising an input unit, a graph neural network unit, and a prediction unit, wherein,

[0027] The input unit is configured to receive a relational network graph and input it into a graph neural network. The relational network graph includes multiple nodes and edges connecting the nodes. The nodes represent entities, and the edges represent the relationships between the connected entities.

[0028] The graph neural network unit is configured to calculate the difference information of each node and its neighboring nodes, iteratively update each node based on the difference information, and output the representation vector of each node.

[0029] The prediction unit is configured to receive and perform node pair prediction based on the representation vector of each node and output the prediction result.

[0030] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon.

[0031] When the program is executed by the processor, it implements the interaction prediction method as described in the first aspect.

[0032] A fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0033] When the processor executes the program, it implements the interaction prediction method as described in the first aspect.

[0034] The beneficial effects of this invention are as follows:

[0035] This invention addresses existing problems by providing an interaction prediction method and system based on graph neural networks. One embodiment of the interaction prediction method reconstructs the message aggregation method during iteration using the difference information between the current node and its neighboring nodes, iteratively updates each node through message passing, and then predicts node pairs using an interaction prediction module. This interaction prediction method utilizes the difference and similarity information between the current node and its neighboring nodes to update the representation vector of each node, thereby overcoming the problems existing in the prior art, reducing iteration noise, accelerating iteration speed, and improving iteration efficiency. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 A flowchart illustrating an embodiment of the interaction prediction method of the present invention is shown;

[0038] Figure 2 A block diagram of the interaction prediction system according to an embodiment of the present invention is shown;

[0039] Figure 3 A flowchart illustrating a protein-protein interaction prediction method according to an embodiment of the present invention is shown;

[0040] Figure 4 A flowchart illustrating a social relationship prediction method according to an embodiment of the present invention is shown;

[0041] Figure 5 A flowchart of a product recommendation prediction method according to an embodiment of the present invention is shown;

[0042] Figure 6 A schematic diagram of the structure of a computer device according to another embodiment of the present invention is shown. Detailed Implementation

[0043] To more clearly illustrate the present invention, the following description, in conjunction with preferred embodiments and accompanying drawings, further explains the invention. Similar components in the drawings are indicated by the same reference numerals. Those skilled in the art should understand that the specific description below is illustrative rather than restrictive and should not be construed as limiting the scope of protection of the present invention.

[0044] Graph Convolutional Networks (GCNs) are an application of neural networks to graph data. Traditional neural networks process images by constructing feature maps through weighted sums of the center pixel and its neighboring elements to extract spatial features. GCNs, on the other hand, are an effective means of extracting spatial information from graphs (topological graphs). By transforming nodes that originally only had their own information, GCNs aggregated neighbor information into the node itself, adding information from its neighbors. This technique can be used for node classification. After transformation by GCNs, the structure of local neighbor information can be known from a single node. In graph matching, it can be used for node similarity judgment as a constraint to prune the results.

[0045] In related technologies, prediction is based on graph neural networks and protein-protein interaction networks. In the protein-protein interaction network, vertices represent proteins, and edges represent interactions between corresponding proteins. The prediction task is then accomplished using the graph's topology through methods such as graph neural networks or random walks. Specifically, the steps include: inputting a protein-protein interaction network, fusing a graph neural network representing differences, and a protein interaction prediction module. In the protein-protein interaction network, nodes represent proteins, and edges represent interactions between corresponding proteins. It includes N nodes {ei, i = 1, ..., N} and M edges {rj, j = 1, ..., M}. Graph neural networks can use message passing to update the node representation vectors, such as...

[0046]

[0047] Where σ is a nonlinear activation function, h t+1 (e i ) represents node e at time t+1. i The vector represents the node e. i The neighboring node numbers are given by W1 and W2, which are parameter matrices. However, in practical applications, it has been found that the representation vectors of different nodes tend to converge as the number of iterations increases during the iteration process.

[0048] To address the above situation, the inventors, through extensive research and experimentation, proposed that related technologies typically use information from neighboring nodes to update the current node at the next time step during iteration. Therefore, the reason why the representation vectors of different nodes tend to converge with the increase in the number of iterations is due to h... t (e i ) and h t (e jThere is redundant or identical information between nodes e, and after multiple iterations, node e... i and node e j The representation vectors between them converge, resulting in the loss of node-specific information.

[0049] Based on the above problems and the causes of these problems, such as Figure 1 As shown, one embodiment of the present invention provides an interaction prediction method based on graph neural networks, comprising:

[0050] Receive a relational network graph and input it into a graph neural network. The relational network graph includes multiple nodes and edges connecting each node. The nodes represent entities, and the edges represent the relationships between the connected entities.

[0051] The graph neural network calculates the difference information of each node and its neighboring nodes, iteratively updates each node based on the difference information, and outputs the representation vector of each node.

[0052] The prediction module receives and performs node pair prediction based on the representation vectors of each node, and outputs the prediction results.

[0053] In this embodiment, the message aggregation method is reconstructed using the difference information between the current node and its neighboring nodes during the iteration process. Then, each node is iteratively updated through message passing, and finally, node pair prediction is performed through the interaction prediction module. The interaction prediction method in this embodiment uses the difference and similarity information between the current node and its neighboring nodes to update the representation vector of each node, thereby overcoming the problems existing in the prior art, reducing iteration noise, accelerating iteration speed, and improving iteration efficiency.

[0054] In a specific example, protein-protein interaction prediction is used as an illustration, such as... Figure 1 As shown, it includes the following steps:

[0055] The first step is to receive a relational network graph and input it into a graph neural network. The relational network graph includes multiple nodes and edges connecting each node. The nodes represent entities, and the edges represent the relationships between the connected entities.

[0056] In this embodiment, a relational network graph is used as the input data for the graph neural network. Specifically, the relational network graph includes multiple nodes representing entities, and edges connecting every two nodes that represent the interaction between the two entities, thereby representing the relationship between the entities in the form of a graph structure.

[0057] It is worth noting that in this embodiment, before inputting the relationship network graph into the graph neural network, a relationship network graph is formed based on the protein-protein interaction network. Each node represents a protein entity including protein attributes, and the edge connecting two protein entities represents the relationship between the two protein entities. Thus, the relationships between proteins are represented using a graph structure, facilitating processing by the graph neural network.

[0058] The second step involves the graph neural network calculating the difference information of each node and its neighboring nodes, iteratively updating each node based on the difference information, and outputting the representation vector of each node.

[0059] In this embodiment, by acquiring the difference information between each node and its neighboring nodes, and iterating on different nodes based on the difference information and the information of each neighboring node, the convergence of the representation vectors of each node due to redundant or identical information during the iteration process is avoided. Specifically, this includes:

[0060] First, the graph neural network calculates the difference information of each node and its neighboring nodes.

[0061] In this embodiment, calculations are performed on each node to obtain the difference information between each node and its neighboring nodes, that is, the information that cannot be reconstructed from the neighboring nodes. In other words, the current node is fitted to the current node as closely as possible using each of its neighboring nodes, and the difference between the current node and the fitted node is the difference information between the current node and its neighboring nodes.

[0062] Based on this approach, in an optional embodiment, the difference information between each node and its neighboring nodes is as follows:

[0063]

[0064] The relationship network graph includes N nodes, hd t (e i ) is node e i and all its neighboring nodes e j The difference information at time t, h t (e i ) represents node e at time t i The representation vector, h t (e j ) represents node e at time t i Neighbor node e j The representation vector, N(i) represents node e. i The neighbor node number, α i, For parameters.

[0065] In this embodiment, ∑ j∈N() α i,j ×h t (e j ) is node e i Each neighbor node e j The current node that is fitted to the greatest extent possible, that is, the difference between the current node's representation vector and the current node fitted by each of its neighboring nodes, is the content that the current node cannot be reconstructed, that is, the difference information between the current node and its neighboring nodes.

[0066] In an optional embodiment, the α i, To ensure that all neighboring nodes of the current node best fit the parameters of the current node, the following condition must be met:

[0067]

[0068] Among them, h t (e i ) represents node e at time t i The representation vector, h t (e j ) represents node e at time t i Neighbor node e j The representation vector.

[0069] In this embodiment, α is the value that minimizes the formula. i,j That is, to maximize the fit of the parameters of the current node to each neighboring node.

[0070] Considering further improving α i, In another alternative embodiment, the accuracy of α i,j Further satisfy:

[0071]

[0072] Among them, h t (e i ) represents node e at time t i The representation vector, h t (e j ) represents node e at time t i Neighbor node e j The vector represents λ, where λ is a positive integer greater than 0. Regularization parameters to prevent overfitting.

[0073] In this embodiment, a regularization parameter is added to prevent overfitting. Further increase α i, This improves the accuracy of predictions, thereby further enhancing their accuracy.

[0074] Specifically, let Considering f i If it is a convex function, then let f i The minimum value of αi,j satisfies

[0075]

[0076] Then there is,

[0077]

[0078] Where the superscript T denotes transpose. The above equation can be expressed in matrix form:

[0079] Let matrix R be a 1×|N(i)| dimensional row vector, where |N(i)| represents node e. i The number of neighbors of the element in the j-th column, R,j = h. t (e i ) T ·h t (e j Let matrix U be a |N(i)|×|N(i)| matrix, where the element Uk,j = h in the k-th row and j-th column. t (e k ) T ·h t (e j If U = Q, then U = Q T Q. Let matrix A be a 1×|N(i)| dimensional vector, where the element in the j-th column is α. i, The above formula can then be expressed as:

[0080] R-AU-λA=0

[0081] => R = AU + λA

[0082] =A(U+λI)

[0083] => A = R(U + λI) -1

[0084] Where I is the identity matrix of |N(i)|×|N(i)|, and A is α i, Thus through α i, Get node e i Cannot be passed through neighbor node e j The reconstructed information, i.e.

[0085]

[0086] The above steps can obtain the result that makes the current node e i The representation vector h t (e i ), and its neighboring nodes e jThe difference between the largest current node and the vector fitted by the representation vector is the minimum α. i, .

[0087] Secondly, the message is reconstructed based on the difference information, and each node is iteratively updated through message passing to output the representation vector of each node.

[0088] Based on the difference information obtained in the above embodiments, the message is further reconstructed to iteratively update the representation vector of each node, as follows:

[0089]

[0090] Among them, h t+1 (e i ) represents node e at time t+1. i The representation vector, σ is the nonlinear activation function, W1 and W2 are parameter matrices, hd t (e i ) is node e i and all its neighboring nodes e j The difference information at time t, h t (e j ) represents node e at time t i Neighbor node e j The representation vector, N(i) represents node e. i The neighboring node numbers.

[0091] This embodiment initializes the parameters of the graph neural network and the initial vectors of each node by characterizing the graph structure of the protein-protein interaction network. Specifically, the maximum value of t is set to 20, that is, the graph neural network obtains the final vector representation of each node (i.e., protein) after 20 iterations, and the node vector size is 128 dimensions.

[0092] In this embodiment, the representation vector of the node is updated relative to the node representation vector during the iteration process in related technologies:

[0093]

[0094] The difference lies in the node e in the relevant technology. i The representation vector at time t is from W1h t (e i ) is changed to represent node e i W1hd, the difference information at time t t (e i This avoids redundant or identical information during the iteration process, which could cause the representation vectors of each node to converge as the number of iterations increases.

[0095] In this embodiment, the graph neural network calculates the difference information of each node and its neighboring nodes, reconstructs the message based on the difference information, and iteratively updates each node through message passing, outputting the representation vector of each node. In other words, this embodiment utilizes the difference information W1h between the current node and its neighboring nodes. t (e i ) and similarity information ∑ j∈N() W2h t (e j Iterative updates of the representation vector of each node effectively reduce the impact of redundant information.

[0096] The third step is for the prediction module to receive and perform node pair prediction based on the representation vectors of each node, and then output the prediction results.

[0097] In this embodiment, a prediction module further predicts the node pairs and outputs the predicted relationship between the two nodes in the node pair. Specifically, this includes:

[0098] In an optional embodiment, the prediction module receiving and performing node pair prediction based on the representation vectors of each node and outputting the prediction result further includes:

[0099] y = γ(W s [h 20 (pi),h 20 (pj),h 20 (pi)-h 20 (pj),h 20 (pi)·h 20 (pj)]+b s )

[0100] Where y represents the prediction result for the node pair of two nodes pi and pj, γ is the sigmoid function, and W s Let h be the parameter matrix. 20 (pi) is the representation vector of node pi, h 20 (pj) is the representation vector of node pj, where nodes pi and pj are two nodes connected by an edge in the relational network graph, h 20 (pi)-h 20 (pj) is the difference between the representation vectors of the two nodes, h 20 (pi)·h 20 (pj) is the product of the representation vectors of the two nodes, [,] indicates concatenation, b s For bias.

[0101] In this embodiment, the prediction result of the interaction between the two nodes is obtained by concatenating the representation vectors of the two nodes in the node pair, the difference between them, and the product between them.

[0102] Thus, the prediction of protein-protein interactions is complete. This embodiment reconstructs the message aggregation method during iteration using the difference information between the current node and its neighboring nodes, then iteratively updates each node through message passing, and finally predicts node pairs using the interaction prediction module. The interaction prediction method in this embodiment utilizes the difference and similarity information between the current node and its neighboring nodes to update the representation vector of each node, thereby overcoming the problems existing in the prior art, reducing iteration noise, accelerating iteration speed, and improving iteration efficiency.

[0103] Corresponding to the interaction prediction method provided in the above embodiments, one embodiment of this application also provides an interaction prediction system that applies the above interaction prediction method, such as... Figure 2 As shown, it includes an input unit, a graph neural network unit, and a prediction unit, wherein,

[0104] The input unit is configured to receive a relational network graph and input it into a graph neural network. The relational network graph includes multiple nodes and edges connecting the nodes. The nodes represent entities, and the edges represent the relationships between the connected entities.

[0105] The graph neural network unit is configured to calculate the difference information of each node and its neighboring nodes, iteratively update each node based on the difference information, and output the representation vector of each node.

[0106] The prediction unit is configured to receive and perform node pair prediction based on the representation vector of each node and output the prediction result.

[0107] In this embodiment, the receiving relationship network graph is obtained through the input unit. During the iteration process, the graph neural network unit reconstructs the message aggregation method using the difference information between the current node and its neighboring nodes. Then, each node is iteratively updated through message passing. Finally, the prediction unit performs node pair prediction. This embodiment's interaction prediction system updates the representation vector of each node using the difference and similarity information between the current node and its neighboring nodes, thereby overcoming the problems existing in the prior art, reducing iteration noise, accelerating iteration speed, and improving iteration efficiency.

[0108] Since the interaction prediction system provided in this application corresponds to the interaction prediction methods provided in the above embodiments, the previous implementation methods are also applicable to the interaction prediction system provided in this embodiment, and will not be described in detail in this embodiment.

[0109] Based on the above-described interaction prediction method, one embodiment of this application also provides a protein-protein interaction prediction method, such as... Figure 3As shown,

[0110] A relational network graph is formed based on the protein-protein interaction network, where each node represents a protein entity that includes protein properties, and the edge connecting two protein entities represents the relationship between the two protein entities.

[0111] Prediction is made using interaction prediction methods.

[0112] In this embodiment, proteins are used as nodes in a relational network graph, including proteins and their attributes. The relationship between two proteins is represented by edges connecting them, thus forming a graph-structured protein interaction network. The interaction prediction method described above is then used to predict the relationship between any two proteins. Specific implementation details are provided in the foregoing embodiments and will not be repeated here.

[0113] Based on the above interaction prediction method, one embodiment of this application also provides a social relationship prediction method, such as... Figure 4 As shown,

[0114] A relationship network graph is formed based on social networks, where each node represents a natural person entity that includes social attributes, and the edge connecting two natural person entities represents the relationship between the two natural person entities.

[0115] Prediction is made using interaction prediction methods.

[0116] In this embodiment, each natural person in the social network is treated as a node in the relational network graph, including the natural person and the natural person's attributes. The relationship between two natural persons is represented by the edge connecting them, thereby forming a graph-structured interaction network between natural persons. The interaction prediction method described above is used to make predictions, thereby obtaining the prediction results of the relationship between any two natural persons.

[0117] Specifically, in a social network, network nodes represent individuals, and edges represent friendships between the corresponding individuals. For example, in a social network with three individuals A, B, and C, where A and B are friends, and B and C are friends, there is an edge between A and B, and another edge between B and C. The social relationship prediction method in this embodiment forms a relationship network graph, representing A, B, and C as vectors (i.e., h...). T (A), h T (B), h T (C) is then input into a graph neural network, and after multiple iterations, the probability of A and C becoming friends is predicted.

[0118] It's worth noting that considering information differences in social relationship prediction significantly impacts prediction accuracy. For example, suppose A's interests are travel and basketball, B's interests are basketball and food, and C's interests are food and reading. A and B become friends because they both enjoy sports, and B and C become friends because they both enjoy food. Without considering these differences, through a graph neural network, A's information can be entirely transmitted from B to C, and vice versa. Therefore, the probability of A and C becoming friends would be high. However, in reality, their probability of being friends is relatively low because A is active while C is more sedentary. By considering these differences—specifically, the difference between A and B being travel, and between B and C being reading—the amount of information transmitted between A and C decreases, meaning their correlation decreases, and the probability of them becoming friends decreases, thus effectively improving prediction accuracy.

[0119] Based on the above interaction prediction method, one embodiment of this application also provides a product recommendation prediction method, such as... Figure 5 As shown,

[0120] A relationship network diagram is formed based on user purchases of goods, where each node represents a user entity or a product entity, and the edge connecting two nodes represents the relationship between the user entity and the product entity.

[0121] Prediction is made using interaction prediction methods.

[0122] In this embodiment, each user and each product in the user-product network is treated as a node in the relational network graph, including users and user attributes, and products and product attributes. The relationship between the two nodes is represented by the edges connecting them, thereby forming a graph-structured interaction network between users and products. The interaction prediction method described above is used to make predictions, thereby obtaining the prediction results of the relationship between any user and product.

[0123] Specifically, if a user has purchased a certain product, then an edge exists between them. The product recommendation prediction method in this embodiment represents users and products as representation vectors, and the prediction module predicts the probability of a user purchasing a certain product, thereby achieving accurate product recommendations. This method has broad application prospects. For specific implementation details, please refer to the foregoing embodiments, which will not be repeated here.

[0124] Another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements: an interaction prediction method based on a graph neural network, or a protein-protein interaction prediction method, or a social relationship prediction method, or a product recommendation prediction method.

[0125] In practical applications, the computer-readable storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0126] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0127] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0128] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0129] like Figure 6As shown, another embodiment of the present invention provides a structural schematic diagram of a computer device. Figure 6 The computer device 12 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0130] like Figure 6 As shown, the computer device 12 is represented in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and a bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0131] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0132] Computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0133] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 6 Not shown; usually referred to as a "hard drive"). Although Figure 6 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0134] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0135] Computer device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with the computer device 12, and / or with any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 22. Furthermore, computer device 12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 20. Figure 6 As shown, network adapter 20 communicates with other modules of computer device 12 via bus 18. It should be understood that, although... Figure 6 As not shown, it can be used in conjunction with computer device 12 with other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0136] The processor unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing an interaction prediction method based on a graph neural network, a protein-protein interaction prediction method, a social relationship prediction method, or a product recommendation prediction method provided in the embodiments of the present invention.

[0137] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. An interaction prediction method based on graph neural networks, characterized in that, include: The system receives a relational network graph and inputs it into a graph neural network. The relational network graph is formed based on a protein-protein interaction network and includes multiple nodes and edges connecting each node. The nodes represent protein entities that contain protein attributes, and the edges represent the interaction relationships between the connected protein entities. The graph neural network calculates the difference information between each node and its neighboring nodes, iteratively updates each node based on the difference information, and outputs the representation vector of each node. The prediction module receives and performs node pair prediction based on the representation vector of each node and outputs the prediction results. The prediction results are the prediction results of the interactions between the protein entities, which are used to reveal the cellular regulatory mechanism. The graph neural network calculates the difference information between each node and its neighboring nodes, iteratively updates each node based on the difference information, and outputs the representation vector of each node. This further includes performing the following steps in each iteration: The graph neural network calculates the differences between each node and its neighboring nodes. The message is reconstructed based on the difference information, and each node is iteratively updated through message passing to output the representation vector of each node. The graph neural network further includes calculating the differences between each node and its neighboring nodes, including: The relationship network graph includes N nodes. For nodes and all its neighboring nodes The difference information at time t, Represents the node at time t The representation vector, Represents the node at time t neighboring nodes The representation vector, Represents a node The neighbor node numbers, For parameters; The To ensure that all neighboring nodes of the current node best fit the parameters of the current node, the following condition must be met: in, Represents the node at time t The representation vector, Represents the node at time t neighboring nodes The representation vector.

2. The interaction prediction method according to claim 1, characterized in that, The Further satisfy: in, Represents the node at time t The representation vector, Represents the node at time t neighboring nodes The representation vector, It is a positive integer greater than 0. Regularization parameters to prevent overfitting.

3. The interaction prediction method according to claim 1, characterized in that, The step of reconstructing the message based on the difference information and iteratively updating each node through message passing to output the representation vector of each node further includes: in, The node at time t+1 The vector represents σ, where σ is the nonlinear activation function, and W1 and W2 are parameter matrices. For nodes and all its neighboring nodes The difference information at time t, Represents the node at time t neighboring nodes The representation vector, Represents a node The neighboring node numbers.

4. The interaction prediction method according to claim 1, characterized in that, The prediction module receives and performs node pair prediction based on the representation vectors of each node, and outputs the prediction results, further including: Where y represents two nodes and The node pairs prediction results It is the sigmoid function. For the parameter matrix, Let be the representation vector of node pi. Let pj be the representation vector of the node, and the node... and Let be the two nodes connected by an edge in the relational network graph. The difference between the representation vectors of the two nodes. It is the product of the representation vectors of the two nodes, and [,] indicates concatenation. For bias.

5. An interaction prediction system based on graph neural networks, characterized in that, It includes an input unit, a graph neural network unit, and a prediction unit, wherein, The input unit is configured to receive a relational network graph and input it into a graph neural network. The relational network graph is formed based on a protein-protein interaction network and includes multiple nodes and edges connecting each node. The nodes represent protein entities containing protein attributes, and the edges represent the interaction relationships between the connected protein entities. The graph neural network unit is configured to calculate the difference information between each node and its neighboring nodes, iteratively update each node based on the difference information, and output the representation vector of each node. The configuration further includes: the graph neural network calculating the difference information between each node and its neighboring nodes, which further includes: The relationship network graph includes N nodes. For nodes and all its neighboring nodes The difference information at time t, Represents the node at time t The representation vector, Represents the node at time t neighboring nodes The representation vector, Represents a node The neighbor node numbers, For parameters; the To ensure that all neighboring nodes of the current node best fit the parameters of the current node, the following condition must be met: in, Represents the node at time t The representation vector, Represents the node at time t neighboring nodes The representation vector of each node is obtained; the message is reconstructed based on the difference information, and the nodes are iteratively updated through message passing to output the representation vector of each node. The prediction unit is configured to receive and perform node pair prediction based on the representation vector of each node and output the prediction result, which is the prediction result of the interaction between the protein entities, and is used to reveal the cellular regulatory mechanism.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the interaction prediction method as described in any one of claims 1-4.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the interaction prediction method as described in any one of claims 1-4.

Images