Molecular graph classification method, system, storage medium and terminal based on quantum convolution

Abstract

The present invention discloses a molecular graph classification method, system, storage medium and terminal based on quantum convolution, which pools the molecular graph structure data of biological data sets through a pooling module FPOOL; quantum encodes the molecular graph node features after pooling to a quantum system; realizes node feature transfer through cross-qubit gate operation to obtain updated molecular graph node features, wherein the node feature transfer includes self-node feature cycle and inter-node feature transfer; performs quantum measurement according to the updated molecular graph node features, and predicts the molecular graph attributes according to the measurement results. The present invention uses a distinguishable pooling module FPOOL to pool protein molecular graphs, which can greatly retain the original graph features while greatly reducing the number of quantum-encoded graph nodes and node feature dimensions, thereby reducing the number of quantum bits to realize quantum graph convolutional neural networks, which can greatly promote the research of graph classification tasks in the field of biochemistry.

Description

Technical Field

[0001] The present invention relates to the technical field of molecular graph analysis, and in particular to a molecular graph classification method, system, storage medium and terminal based on quantum convolution. Background Art

[0002] In the field of biochemistry, research focuses on modeling molecular topological structures, while many molecular structures and properties are reflected in the structural characteristics of the molecular graph itself. The use of graph convolutional neural network (GCN) methods to characterize these molecular structures can significantly help tasks including new drug discovery and drug classification. Graph convolutional neural networks predict graph attributes by learning effective node features and have strong feature extraction and aggregation capabilities when processing graph structure data. As a convenient and general analysis tool, graph convolutional neural networks have been successfully applied to graph attribute classification tasks in the field of biochemistry.

[0003] Quantum computing uses quantum bits as basic units and uses principles such as quantum superposition and interference to achieve parallel computing. It can provide exponential acceleration on certain computationally difficult problems. Quantum neural networks can achieve faster speeds than classical algorithms and are an important direction for the leapfrog development of computing power in the future. However, there are few attempts to use the advantages of quantum computing to improve the machine learning tasks of graph convolutional neural networks for large-scale data sets, and the application of quantum graph neural networks for graph attribute classification in practical applications has not been fully explored. Most of the existing quantum convolutional neural network algorithms are applied to two-dimensional image recognition or processing, or are just simulations of classical algorithms, and cannot be directly applied to the study of molecular graph attribute classification tasks for biological data sets. Therefore, the study of the application of quantum graph convolutional neural networks to molecular graph attribute classification of biological data sets has extremely strong practical significance. Summary of the invention

[0004] The purpose of the present invention is to overcome the problems existing in the prior art, and provide a molecular graph classification method, system, storage medium and terminal based on quantum convolution, which utilizes a series of quantum logic gates to realize the feature cycle of its own nodes and feature learning between adjacent nodes, and performs pooling operations on graph data to reduce the number of nodes and feature dimensions. Finally, the quantum measurement results are combined with the classical node features as graph classification features to reduce the impact of quantum device noise, and finally a quantum graph convolutional neural network is constructed for the classification of graph attributes of biological data sets.

[0005] The objective of the present invention is achieved through the following technical solutions:

[0006] In a first aspect, a molecular graph classification method based on quantum convolution is provided, the method comprising the following steps:

[0007] S1, pooling the molecular graph structure data of the biological dataset through the pooling module FPOOL;

[0008] S2, quantum encode the pooled molecular graph node features into the quantum system;

[0009] S3. Implement node feature transfer through cross-qubit gate operation to obtain updated molecular graph node features, wherein the node feature transfer includes self-node feature loop and inter-node feature transfer;

[0010] S4. Perform quantum measurement based on the updated molecular graph node features, and predict the molecular graph properties based on the measurement results.

[0011] As a preferred option, the molecular graph classification method based on quantum convolution uses the gradient descent method to continuously optimize the pooling result in step S1.

[0012] As a preferred option, the molecular graph classification method based on quantum convolution further includes the steps of:

[0013] S5. Encapsulate steps S1-S4 and construct a quantum graph convolutional neural network for molecular graph prediction of biological datasets.

[0014] As a preferred option, in the molecular graph classification method based on quantum convolution, the step S1 specifically includes:

[0015] S11, constructing a pooling matrix according to the parameter matrix and the molecular graph node feature matrix;

[0016] S12. Update the protein molecular graph node adjacency matrix and the molecular graph node feature matrix according to the pooling matrix.

[0017] As a preferred option, in the molecular graph classification method based on quantum convolution, the step S2 specifically includes:

[0018] S21, normalizing the molecular graph node feature information, and then filling the molecular graph node features with a dimension ranging from 0 to an integer logarithm;

[0019] S22. Perform amplitude encoding on the molecular graph node features.

[0020] As a preferred option, in the molecular graph classification method based on quantum convolution, the step S3 specifically includes:

[0021] S31. Use a single qubit to record the connection between a pair of nodes. When there is a connection, set the qubit state to |1>, otherwise to |0>;

[0022] S32, introduce U1 quantum gate to realize the feature loop of its own node and CU1 quantum gate to realize the feature transfer between node pairs;

[0023] S33. For all quantum bits, the node feature loop and node pair feature transfer quantum gate are applied to realize the feature update between the pooled graph node pairs.

[0024] As a preferred option, in the molecular graph classification method based on quantum convolution, the step S4 specifically includes:

[0025] S41. Use the Pauli-Z gate to measure the expected value of each qubit.

[0026] S42. Use quantum measurement results for molecular graph label prediction.

[0027] In a second aspect, a molecular graph classification system based on quantum convolution is provided, the system comprising:

[0028] A molecular graph pooling module, configured to pool the molecular graph structure data of the biological dataset through a pooling module FPOOL;

[0029] A molecular graph quantum encoding module, configured to quantum encode the pooled molecular graph node features onto a quantum system;

[0030] A node feature transfer module is configured to implement node feature transfer through cross-qubit gate operations to obtain updated molecular graph node features, wherein the node feature transfer includes self-node feature loop and inter-node feature transfer;

[0031] The molecular graph classification module is configured to perform quantum measurement based on the updated molecular graph node features and predict the molecular graph properties based on the measurement results.

[0032] In a third aspect, a computer storage medium is provided, on which computer instructions are stored, and when the computer instructions are executed, the relevant steps in any one of the molecular graph classification methods based on quantum convolution are executed.

[0033] In a fourth aspect, a terminal is provided, comprising a memory and a processor, wherein the memory stores computer instructions executable on the processor, and the processor executes relevant steps in any one of the molecular graph classification methods based on quantum convolution when executing the computer instructions.

[0034] It should be further explained that the technical features corresponding to the above options can be combined or replaced with each other to form a new technical solution if there is no conflict.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] (1) The present invention uses a distinguishable pooling module FPOOL to pool the protein molecule graph, which can greatly retain the original graph features while greatly reducing the number of graph nodes and node feature dimensions of quantum coding, thereby reducing the number of quantum bits to realize quantum graph convolutional neural network.

[0037] (2) In one example, compared with other quantum coding methods, the present invention adopts amplitude coding, which can encode node features more intuitively and requires fewer quantum bits for quantum coding.

[0038] (3) In one example, since classical features are insensitive to quantum noise, the quantum measurement results are connected with the input classical features as the input of the graph prediction label. This can greatly reduce the impact of quantum circuit noise on the graph classification results while retaining the molecular feature information.

[0039] (4) In one example, a quantum graph convolutional neural network based on molecular graph classification of biological datasets was constructed, which can greatly promote the research of graph classification tasks in the field of biochemistry. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of a molecular graph classification method based on quantum convolution according to an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of a pooling process according to an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of a classic graph convolution shown in an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of feature transfer in quantum graph convolution according to an embodiment of the present invention;

[0044] Figure 5 This is a schematic diagram of a process for training a neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0047] Reference Figure 1In an exemplary embodiment, a molecular graph classification method based on quantum convolution is provided, the method comprising the following steps:

[0048] S1, pooling the molecular graph structure data of the biological dataset through the pooling module FPOOL;

[0049] S2, quantum encode the pooled molecular graph node features into the quantum system;

[0050] S3. Implement node feature transfer through cross-qubit gate operation to obtain updated molecular graph node features, wherein the node feature transfer includes self-node feature loop and inter-node feature transfer;

[0051] S4. Perform quantum measurement based on the updated molecular graph node features, and predict the molecular graph properties based on the measurement results.

[0052] Specifically, due to the limited number of quantum bits of existing quantum machines, it is difficult to encode a large number of graph nodes. Therefore, the present invention adopts a distinguishable pooling module FPOOL to pool the protein molecule graph, which can greatly retain the original graph features while greatly reducing the number of quantum-encoded graph nodes and node feature dimensions, thereby reducing the number of quantum bits to realize quantum graph convolutional neural networks.

[0053] Furthermore, the step S1 specifically includes:

[0054] S11. Construct a pooling matrix S based on the parameter matrix and the molecular graph node feature matrix. A row at the pooling matrix S represents the probability of a node being clustered into the pooling graph of q nodes. Specifically, due to the limited number of quantum bits in existing quantum machines, it is difficult to encode the node features of large-scale data sets. By constructing a pooling module to pool the molecular graph, the original graph information can be retained to the greatest extent and the original data feature information can be retained. Suppose the weight matrix W∈R d×n1 , the weight matrix B∈R n1 , d is the node feature dimension, n1 is the number of nodes in the pooled graph, and the pooling matrix S = softmax(X.W+B) is constructed. X is the initial node feature matrix, and the softmax function is calculated for each row, thereby obtaining the pooling matrix S∈R n×n1 ,s ij represents the probability of clustering the i-th node to the j-th node.

[0055] S12, update the protein molecule graph node adjacency matrix and the molecule graph node feature matrix according to the pooling matrix. Specifically, the pooling graph adjacency matrix A′ and the node feature matrix X′ are A′=S T AS∈R q×q , X′=S T X∈Rq×d , where q represents the number of nodes in the graph after pooling, that is, the original n nodes become q nodes. The pooling process is as follows Figure 2 shown.

[0056] Furthermore, the step S2 specifically includes:

[0057] S21, normalizing the molecular graph node feature information, and then filling the molecular graph node features with a dimension ranging from 0 to an integer logarithm;

[0058] S22. Perform amplitude encoding on the molecular graph node features.

[0059] Specifically, the molecular graph node features obtained by clustering the pooling module are normalized: ∑ j |x j | 2 = 1. And fill the dimension to the dimension from 0 to the integer logarithm, and use amplitude coding to encode the node feature information into the quantum circuit using n quantum bits. The encoding process is as follows:

[0060]

[0061] Among them, the node feature is reflected in the amplitude size, and each graph node is encoded using the above method.

[0062] Furthermore, the step S3 specifically includes:

[0063] S31. Use a single qubit to record the connection between a pair of nodes. When there is a connection, set the qubit state to |1>, otherwise to |0>;

[0064] S32, introduce U1 quantum gate to realize the feature loop of its own node and CU1 quantum gate to realize the feature transfer between node pairs;

[0065] S33. For all quantum bits, the node feature loop and node pair feature transfer quantum gate are applied to realize the feature update between the pooled graph node pairs.

[0066] Specifically, in the classic graph convolutional neural network, the features of the nodes are updated by retaining some of their own features and aggregating the features of adjacent nodes. As defined, N(i) is the node v i The adjacent neighbor set; a ij is the connection v in the matrix A i and v j The edge weight, Aggregate represents the aggregation function, W (l) ∈R d×d is a trainable mapping matrix. The classic graph convolution process is as follows Figure 3 In quantum graph convolution, we also refer to the idea of ​​updating node features in classical graph convolution. We use the U1 quantum gate to realize the node feature transfer between itself and itself. U1(x i ,[i]) is the quantum gate U1 applied to quantum bit i, with parameter x i . Use CU1 quantum gate to transfer features between nodes, CU1(x j ,[j,i]) is the target bit for quantum bit i, the control bit for quantum bit j, and the parameter is x j , set a separate quantum bit to save the node topology information. If there is an edge between the node pairs, the quantum bit is set to |1>, otherwise it is |0>. If the quantum bit is |1>, the CU1 gate works, otherwise it does not work. Figure 4 As shown in the figure, the convolution layer of the quantum graph convolutional neural network is established. After k layers of information transmission, each node integrates the features of other nodes in the graph according to the graph structure, and can be used for molecular graph attribute prediction of biological data sets based on the updated node features. Among them, the graph convolution aggregates the information of adjacent nodes through convolution operation. Each time the information is transmitted, a graph convolution operation is performed. After k times of graph convolution operation, each node integrates the information of k-hop neighbors.

[0067] Furthermore, the step S4 specifically includes:

[0068] S41. Use the Pauli-Z gate to measure the expected value of each qubit.

[0069] S42, using quantum measurement results for molecular graph label prediction. Specifically, the quantum bits that have passed through k convolutional layers are measured, and the quantum measurement results are connected with the classical node features as the node feature representation of the corresponding graph node.

[0070] Use the aggregation function to aggregate the graph node features as the graph feature representation. The aggregation function can be Mean, Max, Sum, etc. The graph feature representation vector y G The calculation formula is:

[0071]

[0072] The graph feature representation vector is passed to an MLP multi-layer perceptron to obtain the graph classification output, and the parameters of the MLP are subsequently adjusted through training.

[0073] Furthermore, the molecular graph classification method based on quantum convolution also includes the steps of:

[0074] S5. Encapsulate steps S1-S4 and construct a quantum graph convolutional neural network for molecular graph prediction of biological datasets, which can greatly promote the research of graph classification tasks in the field of biochemistry.

[0075] Further, refer to Figure 5 , a molecular graph classification method based on quantum convolution, using the gradient descent method to continuously optimize the pooling results in step S1. Specifically, the biological data set is divided into a training set and a test set, and the training set is continuously trained with quantum graph convolution, and the weight matrix W and the bias matrix B are optimized and adjusted using the gradient optimization model, that is, by adjusting the parameters in the quantum convolutional neural network, the quantum convolutional neural network is trained by repeating steps S1-S4, and finally the trained quantum bits are measured to obtain the final measurement results of each node, and the classical node features are connected to reduce quantum noise and improve the accuracy of molecular type prediction. Thus, a quantum graph convolutional neural network for molecular prediction of biological data sets is constructed.

[0076] In another exemplary embodiment, a molecular graph classification system based on quantum convolution is provided, the system comprising:

[0077] A molecular graph pooling module, configured to pool the molecular graph structure data of the biological dataset through a pooling module FPOOL;

[0078] A molecular graph quantum encoding module, configured to quantum encode the pooled molecular graph node features onto a quantum system;

[0079] A node feature transfer module is configured to implement node feature transfer through cross-qubit gate operations to obtain updated molecular graph node features, wherein the node feature transfer includes self-node feature loop and inter-node feature transfer;

[0080] The molecular graph classification module is configured to perform quantum measurement based on the updated molecular graph node features and predict the molecular graph properties based on the measurement results.

[0081] In another exemplary embodiment, the present invention provides a computer storage medium having computer instructions stored thereon, which, when executed, execute relevant steps of the molecular graph classification method based on quantum convolution.

[0082] Based on this understanding, the technical solution of this embodiment, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0083] In another exemplary embodiment, the present invention provides a terminal including a memory and a processor, wherein the memory stores computer instructions executable on the processor, and the processor executes relevant steps of the molecular graph classification method based on quantum convolution when executing the computer instructions.

[0084] The processor may be a single-core or multi-core central processing unit or a specific integrated circuit, or one or more integrated circuits configured to implement the present invention.

[0085] Embodiments of the subject matter and functional operations described in this specification may be implemented in: tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more of them. Embodiments of the subject matter described in this specification may be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier to be executed by a data processing device or to control the operation of the data processing device. Alternatively or additionally, the program instructions may be encoded on an artificially generated propagated signal, such as a machine-generated electrical, optical, or electromagnetic signal, which is generated to encode and transmit information to a suitable receiver device for execution by the data processing device.

[0086] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuits, such as FPGAs (field programmable gate arrays) or ASICs (application-specific integrated circuits), and the apparatus can also be implemented as special purpose logic circuits.

[0087] Processors suitable for executing computer programs include, for example, general and / or special microprocessors, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read-only memory and / or a random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, the computer will also include one or more large-capacity storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operably coupled to this large-capacity storage device to receive data from it or to transmit data to it, or both. However, the computer does not necessarily have such a device. In addition, the computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, just to name a few.

[0088] Although this specification includes many specific implementation details, these should not be interpreted as limiting the scope of any invention or the scope of protection claimed, but are mainly used to describe the features of the specific embodiments of specific inventions. Certain features described in multiple embodiments in this specification may also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination. In addition, although features may work in certain combinations as described above and even initially claim protection, one or more features from the claimed combination may be removed from the combination in some cases, and the claimed combination may point to a sub-combination or a variation of a sub-combination.

[0089] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or requiring that all illustrated operations be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0090] The above specific implementation methods are detailed descriptions of the present invention. It cannot be determined that the specific implementation methods of the present invention are limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions and substitutions can be made without departing from the concept of the present invention, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A molecular graph classification method based on quantum convolution, characterized in that: The method comprises the following steps: S1, pooling the molecular graph structure data of the biological data set through the pooling module FPOOL; the step S1 specifically includes: S11, constructing a pooling matrix according to the parameter matrix and the molecular graph node feature matrix; S12, updating the protein molecular graph node adjacency matrix and the molecular graph node feature matrix according to the pooling matrix; S2, quantum encoding the pooled molecular graph node features into a quantum system; the step S2 specifically includes: S21, normalizing the molecular graph node feature information, and then filling the molecular graph node features with a dimension ranging from 0 to an integer logarithm; S22, performing amplitude coding on the node features of the molecular graph; specifically, using amplitude coding to encode the node feature information into the quantum circuit using n quantum bits; S3, implementing node feature transfer through cross-qubit gate operation to obtain updated molecular graph node features, wherein the node feature transfer includes self-node feature loop and inter-node feature transfer; the step S3 specifically includes: S31. Use a single qubit to record the connection between a pair of nodes. When there is a connection, set the qubit state to |1>, otherwise to |0>; S32, introduce U1 quantum gate to realize the feature loop of its own node and CU1 quantum gate to realize the feature transfer between node pairs; specifically, define CU1(x j ,[j,i]) is the target bit for quantum bit i, the control bit for quantum bit j, and the parameter is x j , set a separate quantum bit to save the node topology information. If there is an edge between the node pairs, the quantum bit is set to |1>, otherwise it is |0>. If the quantum bit is |1>, the CU1 gate works, otherwise it does not work, so as to establish the convolution layer of the quantum graph convolutional neural network; S33, for all quantum bits, apply the node feature loop and node pair feature transfer quantum gate to realize the feature update between the pooled graph node pairs; specifically, after the convolution layer passes k layers of information, each node fuses the features of other nodes in the graph according to the graph structure, and can be used for the prediction of the molecular graph attributes of the biological data set according to the updated node features, wherein the graph convolution aggregates the adjacent node information through the convolution operation, and each information transmission is a graph convolution operation, and after k graph convolution operations, each node fuses the information of k-hop neighbors; S4. Perform quantum measurement based on the updated molecular graph node features, and predict the molecular graph attributes based on the measurement results. Specifically, measure the quantum bits that have passed through k convolutional layers, and connect the quantum measurement results with the classical node features as the node feature representation of the corresponding graph node. Use the aggregation function to aggregate the graph node features as the graph feature representation. Pass the graph feature representation vector to an MLP multi-layer perceptron to obtain the graph classification output, and then adjust the parameters of the MLP through training.

2. The molecular graph classification method based on quantum convolution according to claim 1, characterized in that: The pooling result in step S1 is continuously optimized using the gradient descent method.

3. The molecular graph classification method based on quantum convolution according to claim 1, characterized in that: Also includes the steps: S5. Encapsulate steps S1-S4 and construct a quantum graph convolutional neural network for molecular graph prediction of biological datasets.

4. The molecular graph classification method based on quantum convolution according to claim 1, characterized in that: The step S4 specifically includes: S41. Use the Pauli-Z gate to measure the expected value of each qubit. S42. Use quantum measurement results for molecular graph label prediction.

5. Molecular graph classification system based on quantum convolution, characterized in that: The system comprises: The molecular graph pooling module is configured to pool the molecular graph structure data of the biological data set through the pooling module FPOOL; the pooling specifically includes: Construct a pooling matrix based on the parameter matrix and the molecular graph node feature matrix; Update the protein molecular graph node adjacency matrix and the molecular graph node feature matrix according to the pooling matrix; The molecular graph quantum coding module is configured to quantum encode the pooled molecular graph node features into a quantum system; the encoding process specifically includes: Normalize the molecular graph node feature information, and then fill the molecular graph node features into dimensions ranging from 0 to integer logarithms; Amplitude encoding is performed on the node features of the molecular graph; specifically, the node feature information is encoded into the quantum circuit using n quantum bits using amplitude encoding; The node feature transfer module is configured to implement node feature transfer through cross-qubit gate operations to obtain updated molecular graph node features, wherein the node feature transfer includes self-node feature loop and inter-node feature transfer; the node transfer specifically includes: Use a single qubit to record the connection between a pair of nodes. When there is a connection, the qubit state is set to |1>, otherwise it is |0>; The U1 quantum gate is introduced to realize the feature loop of its own node and the CU1 quantum gate is introduced to realize the feature transfer between node pairs; specifically, CU1(x j ,[j,i]) is the target bit for quantum bit i, the control bit for quantum bit j, and the parameter is x j , set a separate quantum bit to save the node topology information. If there is an edge between the node pairs, the quantum bit is set to |1>, otherwise it is |0>. If the quantum bit is |1>, the CU1 gate works, otherwise it does not work, so as to establish the convolution layer of the quantum graph convolutional neural network; For all quantum bits, the node feature loop and node pair feature transfer quantum gates are applied to realize the feature update between the pooled graph node pairs; specifically, after the convolution layer passes k layers of information, each node fuses the features of other nodes in the graph according to the graph structure, and can be used for the prediction of the molecular graph attributes of the biological data set according to the updated node features, wherein the graph convolution aggregates the adjacent node information through the convolution operation, and each information transfer is a graph convolution operation. After k graph convolution operations, each node fuses the information of k-hop neighbors; The molecular graph classification module is configured to perform quantum measurements based on the updated molecular graph node features and predict molecular graph attributes based on the measurement results. Specifically, the quantum bits that have passed through k convolutional layers are measured, and the quantum measurement results are connected with the classical node features as the node feature representation of the corresponding graph node. The graph node features are aggregated using an aggregation function as a graph feature representation. The graph feature representation vector is passed to an MLP multi-layer perceptron to obtain the graph classification output, and the parameters of the MLP are subsequently adjusted through training.

6. A computer storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed, the relevant steps in the molecular graph classification method based on quantum convolution described in any one of claims 1 to 4 are executed.

7. A terminal comprising a memory and a processor, wherein the memory stores computer instructions that can be executed on the processor, characterized in that: When the processor runs the computer instructions, it executes the relevant steps in the molecular graph classification method based on quantum convolution described in any one of claims 1-4.

Images