Molecular property prediction method, apparatus, device, and readable storage medium

Abstract

This application discloses a method, apparatus, device, and readable storage medium for predicting molecular properties. The method includes: generating a graph structure of the target molecule based on its molecular structure; extracting feature vectors of atoms corresponding to nodes and feature vectors of chemical bonds corresponding to edges; encoding the feature vectors of atoms into a first quantum register to obtain the quantum states corresponding to atoms; encoding the feature vectors of chemical bonds into a second quantum register to obtain the quantum states corresponding to chemical bonds; performing cross-register entanglement operations on the first and second quantum registers through a target quantum entanglement circuit to fuse the information of the quantum states corresponding to atoms and the quantum states corresponding to chemical bonds to obtain the quantum states corresponding to the target molecule; and predicting the properties of the target molecule based on its quantum states to obtain predicted property values. This method improves the accuracy of molecular property prediction.

Description

Technical Field

[0001] This application belongs to the field of quantum computing technology, specifically relating to a method, apparatus, device, and readable storage medium for predicting molecular properties. Background Technology

[0002] To rapidly and accurately predict the physicochemical and biological activities of molecules, molecular property prediction is crucial. In this field, molecular structures are typically represented as graphs and learned using graph neural networks (GNNs). Here, atoms in the molecule are considered nodes in the graph, and chemical bonds are seen as edges connecting the nodes. GNNs use an iterative message-passing mechanism to aggregate information from neighboring nodes and edges to update the state of the central node, thereby learning the characterization of the entire molecule and using it to predict its physical, chemical, or biological activities. With the development of quantum computing technology, quantum circuits are beginning to replace some components in classical neural networks.

[0003] In related technologies, when predicting molecular properties, a unified target quantum circuit is typically used to simultaneously process node (atom) feature vectors and edge (chemical bond) feature vectors. These different types of feature vectors are integrated into a high-dimensional feature vector within the same quantum circuit, and then a feature fusion operation is performed to generate the final molecular characterization. This technique is prone to mutual interference between the two types of information, making it difficult to learn clear and independent characterizations, thus affecting the accuracy of the final molecular characterization and resulting in low accuracy in molecular property prediction. Summary of the Invention

[0004] This application provides a method, apparatus, device, and readable storage medium for predicting molecular properties, addressing the problem of low accuracy in molecular property prediction.

[0005] Firstly, a method for predicting molecular properties is provided, including:

[0006] Based on the molecular structure of the target molecule, a graph structure of the target molecule is generated, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure.

[0007] Extract the feature vectors of the atoms corresponding to the nodes and the feature vectors of the chemical bonds corresponding to the edges;

[0008] The feature vector of the atom is encoded into a first quantum register to obtain the quantum state corresponding to the atom. The number of qubits in the first quantum register matches the dimension of the feature vector of the atom.

[0009] The feature vector of the chemical bond is encoded into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond.

[0010] By performing a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, the information of the quantum state corresponding to the atom and the information of the quantum state corresponding to the chemical bond are fused to obtain the quantum state corresponding to the target molecule;

[0011] Based on the quantum state corresponding to the target molecule, the properties of the target molecule are predicted to obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

[0012] Secondly, this application provides a molecular property prediction device, comprising:

[0013] The first generation module is used to generate a graph structure of the target molecule based on the molecular structure of the target molecule, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure.

[0014] The first extraction module is used to extract the feature vectors of atoms corresponding to the nodes and the feature vectors of chemical bonds corresponding to the edges;

[0015] A first processing module is used to encode the feature vector of the atom into a first quantum register to obtain the quantum state corresponding to the atom, wherein the number of qubits of the first quantum register matches the dimension of the feature vector of the atom.

[0016] The second processing module is used to encode the feature vector of the chemical bond into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond.

[0017] The third processing module is used to perform a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, and to fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule;

[0018] The first prediction module is used to predict the properties of the target molecule based on the quantum state corresponding to the target molecule, and obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

[0019] Thirdly, embodiments of this application provide an electronic device, including a processor, the processor being used for:

[0020] Based on the molecular structure of the target molecule, a graph structure of the target molecule is generated, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure.

[0021] Extract the feature vectors of the atoms corresponding to the nodes and the feature vectors of the chemical bonds corresponding to the edges;

[0022] The feature vector of the atom is encoded into a first quantum register to obtain the quantum state corresponding to the atom. The number of qubits in the first quantum register matches the dimension of the feature vector of the atom.

[0023] The feature vector of the chemical bond is encoded into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond.

[0024] By performing a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, the information of the quantum state corresponding to the atom and the information of the quantum state corresponding to the chemical bond are fused to obtain the quantum state corresponding to the target molecule;

[0025] Based on the quantum state corresponding to the target molecule, the properties of the target molecule are predicted to obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

[0026] Fourthly, embodiments of this application provide an electronic device including a processor and a memory, wherein the memory stores a program or instructions executable on the processor, and the program or instructions, when executed by the processor, implement the steps of the molecular property prediction method as described in the first aspect.

[0027] Fifthly, embodiments of this application provide a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the molecular property prediction method as described in the first aspect.

[0028] Sixthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the molecular property prediction method as described above.

[0029] In this embodiment, a graph structure of the target molecule is generated based on its molecular structure, wherein atoms of the molecular structure correspond to nodes of the graph structure, and chemical bonds of the molecular structure correspond to edges of the graph structure. Feature vectors of atoms corresponding to nodes and feature vectors of chemical bonds corresponding to edges are extracted. The feature vectors of atoms are encoded into a first quantum register to obtain the quantum state corresponding to the atom, the number of qubits in the first quantum register matching the dimension of the feature vector of the atom. The feature vectors of chemical bonds are encoded into a second quantum register to obtain the quantum state corresponding to the chemical bond, the number of qubits in the second quantum register matching the dimension of the feature vector of the chemical bond. A cross-register entanglement operation is performed on the first and second quantum registers through a target quantum entanglement circuit to fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond, obtaining the quantum state corresponding to the target molecule. Based on the quantum state corresponding to the target molecule, property prediction is performed on the target molecule to obtain property prediction values, which are used to characterize the property prediction information of the target molecule. This embodiment encodes the feature vectors of atoms and molecules onto two separate sets of quantum registers, thereby avoiding mutual interference between information of different dimensions from the source, ensuring the purity and integrity of information encoding, which in turn is beneficial to the accuracy of the final molecular characterization and thus improves the accuracy of molecular property prediction. Attached Figure Description

[0030] Figure 1 A schematic flowchart illustrating the molecular property prediction method provided in the embodiments of this application;

[0031] Figure 2 A schematic diagram of the dual-channel parallel quantum coding architecture provided in the embodiments of this application;

[0032] Figure 3 A schematic diagram of the target quantum entanglement circuit provided in an embodiment of this application;

[0033] Figure 4 A schematic diagram of the molecular property prediction system provided in the embodiments of this application;

[0034] Figure 5 This is a schematic diagram of the structure of the energy-saving control device provided in the embodiments of this application;

[0035] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0036] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and are not used to describe a specified order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, a first object can be one or more. Furthermore, in the specification and claims, "and" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0038] It is worth noting that the technologies described in this application are not limited to Long Term Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA), and other systems. The terms "system" and "network" in this application are often used interchangeably, and the described technologies can be used with the systems and radio technologies mentioned above, as well as with other systems and radio technologies. However, the following description describes New Radio (NR) systems for illustrative purposes, and NR terminology is used in most of the following description. These technologies can also be applied to applications beyond NR systems, such as 6th Generation (6G) communication systems.

[0039] The following description, in conjunction with the accompanying drawings, further illustrates the molecular property prediction method, apparatus, device, storage medium, and program product proposed in the embodiments of the application.

[0040] Please see Figure 1 , Figure 1 A flowchart illustrating a molecular property prediction method provided in this application embodiment is shown in the figure. The method includes the following steps:

[0041] Step 101: Generate a graph structure of the target molecule based on its molecular structure, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure.

[0042] In this step, the microscopic structure of chemical molecules can be converted into a graph data structure that can be processed by a computer, facilitating subsequent computation using quantum circuits or graph neural networks. Atoms in the molecular structure correspond to nodes in the graph structure, and chemical bonds correspond to edges. For example, the system receives a molecular representation in a standard format, such as the SMILES (Simplified Linear Input Canonical) string (e.g., SMILES for benzene is c1ccccc1). The system uses a cheminformatics toolkit (such as RDKit) to parse the SMILES string into an undirected graph structure G=(V,E), where V is the set of atoms (nodes) and E is the set of chemical bonds (edges).

[0043] Step 102: Extract the feature vectors of the atoms corresponding to the nodes and the feature vectors of the chemical bonds corresponding to the edges;

[0044] In this step, for each atom v∈V corresponding to each node in the graph structure, the system extracts a d v 3D eigenvector h v This feature vector contains various physicochemical properties of the atom, such as: atomic number, period and group, valence, formal charge, hybridization type (sp, sp2, sp3), whether it is an aromatic atom, and whether it is a chiral center.

[0045] For each edge in the graph structure corresponding to a chemical bond e∈E, the system extracts a d e 3D eigenvector h e This vector contains various properties of chemical bonds, such as: bond type (single, double, triple, aromatic), whether it is conjugated, whether it is in a ring structure, bond length, etc.

[0046] Step 103: Encode the feature vector of the atom into the first quantum register to obtain the quantum state corresponding to the atom. The number of qubits in the first quantum register matches the dimension of the feature vector of the atom.

[0047] In this step, to meet the requirements of quantum encoding, the feature vectors h of all extracted atoms can be... vand chemical bond eigenvector h e The values ​​in the array are all normalized to a specific range, such as [0, 2π], to serve as the angle parameters for the quantum rotation gate. The normalized atomic feature vectors are encoded into the first quantum register, thus obtaining the quantum state corresponding to the atom. The number of qubits in the first quantum register is not arbitrarily set, but needs to be strictly matched with the dimension of the atomic feature vector to ensure that each feature component can be stably encoded through its corresponding qubit. Specifically, a single-qubit rotation gate can be used to load the normalized feature values ​​as rotation angles onto the qubits, transforming classical atomic features into quantum state information that can participate in quantum computing. Throughout the process, atomic information is encoded independently, without mixing with chemical bond information, laying the foundation for subsequent precise information fusion.

[0048] Step 104: Encode the feature vector of the chemical bond into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond.

[0049] In this step, the normalized chemical bond feature vector is encoded into the second quantum register, thereby obtaining the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register also needs to match the dimension of the chemical bond feature vector to ensure that the inherent features of the chemical bond itself, such as bond type, bond length, conjugation, and ring structure, can be completely and losslessly encoded into the quantum state. The encoding process uses quantum circuits and unitary operators that are independent of the atomic encoding. After the encoding is completed, the two sets of registers are in a tensor product state, without entanglement or interference between them, thus fully preserving the independent information of the chemical bond.

[0050] Step 105: Perform cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, and fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule;

[0051] In this step, a pre-constructed target quantum entanglement circuit is used to perform cross-register entanglement operations on the first and second quantum registers, deeply fusing the information carried by the atomic quantum state and the chemical bond quantum state, ultimately obtaining a molecular quantum state that can completely represent the overall information of the target molecule. This target entanglement circuit can be composed of multiple layers of trainable parameterized single-qubit gates and fixed two-qubit entanglement gates. Quantum correlations are established between the two sets of registers through entanglement gates such as CNOT or CZ, enabling efficient interaction and fusion of atomic and chemical bond features, and accurately capturing the quantum mechanical correlations between atoms and chemical bonds within the molecule.

[0052] Step 106: Based on the quantum state corresponding to the target molecule, perform property prediction on the target molecule to obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

[0053] In this step, quantum measurements can be performed on the fused molecular quantum states, the Pauli Z operator expectation value of each quantum bit can be calculated, the measurement results can be combined into a classical molecular characterization vector of fixed dimension, and then input into a shallow classical neural network to obtain the final property prediction value.

[0054] In one embodiment, a graph structure of the target molecule is generated based on its molecular structure, wherein atoms of the molecular structure correspond to nodes in the graph structure, and chemical bonds of the molecular structure correspond to edges in the graph structure; feature vectors of atoms corresponding to nodes and feature vectors of chemical bonds corresponding to edges are extracted; the feature vectors of atoms are encoded into a first quantum register to obtain the quantum state corresponding to the atom, wherein the number of qubits in the first quantum register matches the dimension of the feature vector of the atom; the feature vectors of chemical bonds are encoded into a second quantum register to obtain the quantum state corresponding to the chemical bond, wherein the number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond; a cross-register entanglement operation is performed on the first and second quantum registers through a target quantum entanglement circuit to fuse the information of the quantum state corresponding to the atom and the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule; based on the quantum state corresponding to the target molecule, the properties of the target molecule are predicted to obtain the predicted property value of the target molecule, wherein the predicted property value is used to characterize the predicted property information of the target molecule.

[0055] In this implementation, the feature vectors of atoms and molecules are encoded into two separate sets of quantum registers, which avoids mutual interference between information of different dimensions from the source, ensures the purity and integrity of information encoding, and thus improves the accuracy of the final molecular characterization and the accuracy of molecular property prediction.

[0056] Optionally, encoding the feature vector of the atom into a first quantum register to obtain the quantum state corresponding to the atom includes:

[0057] The qubits of the first quantum register are initialized to obtain the processed first quantum register;

[0058] The eigenvector of the atom is loaded into the processed first quantum register by using the first encoded unitary operator to obtain the quantum state corresponding to the atom.

[0059] In one implementation, see Figure 2 The first quantum register (i.e. Figure 2 All n in the atomic registers) v Each qubit is initialized to the |0> state. Then, it is encoded using a unitary operator U. v (h v )Action on QR v Above, the eigenvector h of the atom v Loaded into a quantum state. This process can be represented as: v .

[0060] in, It typically consists of a series of single-bit rotation gates, for example The physical meaning of this formula lies in transforming the classical eigenvector h... v The i-th component (e.g., the normalized atomic number) is directly used as the rotation angle of the i-th qubit, thereby "encoding" the classical numerical information into the probability amplitude of the quantum state.

[0061] This method of initialization followed by encoding ensures that atomic features enter the quantum state independently, completely, and accurately, avoiding crosstalk with other information. It provides a reliable foundation for subsequent dual-channel parallel processing and cross-register entanglement fusion, and improves the stability and accuracy of quantum embedding and molecular property prediction.

[0062] Optionally, encoding the feature vector of the chemical bond into a second quantum register to obtain the quantum state corresponding to the chemical bond includes:

[0063] The qubits of the second quantum register are initialized and processed to obtain the processed second quantum register;

[0064] The feature vector of the chemical bond is loaded into the processed second quantum register by using the second encoded unitary operator to obtain the quantum state corresponding to the chemical bond.

[0065] In one implementation, see further. Figure 2 The second quantum register (i.e. Figure 2 All n in the chemical bond register e Each qubit is also initialized to the |0> state. This is achieved through a separate encoded unitary operator U. e (h e )Action on QR e Above, the eigenvector h of the chemical bond e Loading in:

[0066] e

[0067] This formula encodes the characteristics of chemical bonds (such as bond type) into separate chemical bond registers in the same way. After the encoding phase is complete, the total number of substates of the system is the tensor product of the two substates: Tensor product This clearly indicates that there is no interaction between the two sets of qubits at this point, and the information is completely separated, laying the foundation for subsequent controllable information fusion steps.

[0068] In this implementation, the second quantum register is first initialized, ensuring all qubits are uniformly in a standard |0> initial state, eliminating interference from residual quantum states, guaranteeing a consistent encoding environment, and ensuring reproducible results. Then, chemical bond feature vectors are loaded using an independent second encoding unitary operator, accurately and completely encoding inherent chemical information such as bond type, bond length, and conjugation into quantum states. This method provides chemical bond features with a dedicated, pure encoding channel, completely independent of the atom encoding channel, avoiding mutual interference between features, preserving key chemical bond information, and providing a stable and reliable input for subsequent cross-register quantum entanglement fusion, significantly improving molecular quantum embedding quality and property prediction accuracy.

[0069] Optionally, the target quantum entanglement circuit is a parameterized quantum entanglement circuit, which includes multiple layers of trainable units, each of which includes a single-bit parameterized rotation gate layer and a fixed entanglement gate layer.

[0070] The step of performing a cross-register entanglement operation on the first and second quantum registers through a target quantum entanglement circuit to fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule includes:

[0071] After performing parameterized rotation on the qubits of the first quantum register and the qubits of the second quantum register through the single-bit parameterized rotation gate of each layer of trainable units, a two-bit entanglement operation is performed between the first quantum register and the second quantum register through the fixed entanglement gate of each layer of trainable units, thereby obtaining the quantum correlation between the features of the atom and the features of the chemical bond.

[0072] Based on the quantum correlation, the information of the quantum state corresponding to the atom is fused with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule.

[0073] In one implementation, see Figure 3 To enable models to learn the complex relationships between atoms and chemical bonds, parameterized quantum entanglement circuits (or "quantum neural networks") can be used. Acting on Above. This can be expressed by the formula:

[0074] in, It is a unitary operator containing L layers. This is the set of all trainable parameters. This formula describes how to fuse separate atomic and chemical bond information through an optimizable quantum process. The parameterized quantum entangled circuit comprises multiple layers of trainable units, each layer including a single-bit parameterized rotation gate layer and a fixed entanglement gate layer, wherein:

[0075] Parameterized single-qubit gate layers are used for all qubits (including QR codes). v and QR e In this process, trainable single-qubit rotation gates are applied to achieve fine-grained manipulation and feature optimization of quantum states. Fixed entanglement gate layers are used to apply a series of fixed two-qubit entanglement gates (such as CNOT gates or CZ gates). Crucially, these entanglement gates must act between atomic registers and chemical bond registers to enable information exchange between the two types of quantum states. Taking a typical CNOT gate as an example, it can transform the qubit q representing atomic aromaticity into a single-qubit rotation gate. v As a control bit, the quantum bit q, which characterizes the aromaticity of the corresponding chemical bond, will be used. e As controlled qubits, quantum correlations between atomic properties and chemical bond properties are realized. Through a large number of such cross-register entanglement operations, the model can effectively learn and simulate the intrinsic quantum correlations between atomic properties and their chemical bonding environment, improving the accuracy of molecular characterization.

[0076] Optionally, the step of predicting the properties of the target molecule based on its corresponding quantum state to obtain predicted property values ​​for the target molecule includes:

[0077] Calculate the expected value of the Pauli Z operator for each qubit in the quantum state corresponding to the target molecule;

[0078] The feature vector of the target molecule is generated based on the expected values ​​of all qubits in the quantum state corresponding to the target molecule;

[0079] Based on the feature vector of the target molecule, the properties of the target molecule are predicted to obtain the predicted property values ​​of the target molecule.

[0080] In one implementation, the quantum state |ψ| corresponding to the target molecule molecule >Measurements are performed. Typically, this is done by calculating the Pauli Z operator σ for each qubit qk. z It is achieved by the expected value. The specific calculation formula is:

[0081] in, This indicates that the Pauli Z operator operates on the k-th qubit. Due to the aforementioned entanglement operation, each measurement M...K All measurements contain a fusion of atomic and chemical bond information, rather than merely reflecting a single initial feature. All measurement results M K They are collected to form an n v +n e The eigenvector F of the target molecule in dimension molecule This vector represents the high-quality, quantum-enhanced feature representation of the target molecule obtained after processing using the method described in this application. Then, based on the feature vector of the target molecule, its properties are predicted.

[0082] In this implementation, the expectation value of the Pauli Z operator for each qubit is first calculated. This transforms the atomic and chemical bond fusion information contained in the quantum state into stable, computable classical values, avoiding interference from quantum randomness. A molecular feature vector is generated based on all expectation values, which fully preserves the high-dimensional correlation information after quantum entanglement fusion and better reflects the microscopic quantum nature of molecules than traditional features. Using this vector for property prediction accurately reflects the interactions between atoms and chemical bonds, significantly improving prediction accuracy and robustness.

[0083] Optionally, the step of predicting the properties of the target molecule based on its feature vector to obtain the predicted property value of the target molecule includes:

[0084] The feature vector of the target molecule is input into the target neural network, and the target neural network predicts the properties of the target molecule to obtain the predicted property value of the target molecule. The target neural network includes a target loss function, which is constructed based on the predicted property value and the true property value of the target molecule. The target neural network is obtained by iteratively updating the training parameters in the target quantum entanglement circuit to minimize the target loss function.

[0085] In one implementation, the feature vector of the target molecule is input into a target neural network. The target neural network can be a shallow classical neural network (e.g., a fully connected layer). The target neural network predicts the properties of the target molecule, obtaining predicted property values. The target neural network includes a target loss function, constructed based on the predicted and actual property values ​​of the target molecule. This target loss function is obtained by iteratively updating the training parameters in the target quantum entangled circuit to minimize the target loss function. Specifically, a classical optimizer (such as Adam) can be used, employing gradient descent to iteratively update the trainable parameters in the quantum circuit until the model converges. The gradient calculation here uses a quantum analytical gradient method, such as the parameter-shift rule. This method accurately calculates the gradient of the loss function with respect to the parameter by performing small forward and reverse shifts on the trainable parameters and calculating the expected values ​​of the quantum circuit twice. This allows for end-to-end optimization of the entire hybrid model using the classical gradient descent algorithm.

[0086] In this implementation, constructing the target loss function using predicted and true values ​​clarifies the direction of model optimization and ensures accurate and reliable prediction results. By iteratively updating the training parameters of the quantum entangled circuit to minimize the loss, end-to-end joint optimization of quantum circuits and classical networks is achieved, making quantum encoding and information fusion more closely aligned with molecular characteristics.

[0087] In one implementation, see Figure 4 The aforementioned molecular property prediction methods can be used to Figure 4 The system executes the commands. This system mainly includes the following modules:

[0088] Information Processing Module: A classic processor, physically which can be a server or computer. This module is responsible for receiving molecular map data, parsing the molecular structure, separating and generating feature vectors for atoms and chemical bonds, and passing these feature vectors to the quantum control hardware. Simultaneously, this module is also responsible for executing the optimization loop of the target neural network, calculating the loss function, and updating the parameters.

[0089] Quantum encoding module: This module contains a quantum processor that integrates a specific dual-channel quantum circuit architecture. It receives classical feature data and current trainable parameters from the information processing module and, based on this data, controls the parameters of quantum gates (such as the phase and amplitude of microwave pulses acting on qubits) to perform quantum encoding, variational entanglement, and information fusion.

[0090] Joint processing and measurement module: Physically, this is part of the quantum processor. This module is responsible for executing entanglement circuits, generating the final molecular quantum state, reading the measurement results through measurement devices (such as resonant cavities), and returning the feature vector of the target molecule to the information processing module.

[0091] Please see Figure 5 , Figure 5 This is a schematic diagram of a molecular property prediction device provided in an embodiment of this application. As shown in the figure, the device 500 includes:

[0092] The first generation module 501 is used to generate a graph structure of the target molecule based on the molecular structure of the target molecule, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure.

[0093] The first extraction module 502 is used to extract the feature vectors of atoms corresponding to the nodes and the feature vectors of chemical bonds corresponding to the edges;

[0094] The first processing module 503 is used to encode the feature vector of the atom into a first quantum register to obtain the quantum state corresponding to the atom, wherein the number of qubits of the first quantum register matches the dimension of the feature vector of the atom.

[0095] The second processing module 504 is used to encode the feature vector of the chemical bond into a second quantum register to obtain the quantum state corresponding to the chemical bond, wherein the number of qubits of the second quantum register matches the dimension of the feature vector of the chemical bond;

[0096] The third processing module 505 is used to perform a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, and fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule;

[0097] The first prediction module 506 is used to predict the properties of the target molecule based on the quantum state corresponding to the target molecule, and obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

[0098] Optionally, the first processing module includes:

[0099] The first processing unit is used to initialize the qubits of the first quantum register to obtain the processed first quantum register.

[0100] The first loading unit is used to load the feature vector of the atom into the processed first quantum register through the first encoded unitary operator to obtain the quantum state corresponding to the atom.

[0101] Optionally, the second processing module includes:

[0102] The second processing unit is used to initialize and process the qubits of the second quantum register to obtain the processed second quantum register.

[0103] The second loading unit is used to load the feature vector of the chemical bond into the processed second quantum register through the second encoded unitary operator to obtain the quantum state corresponding to the chemical bond.

[0104] Optionally, the target quantum entanglement circuit is a parameterized quantum entanglement circuit, which includes multiple layers of trainable units, each of which includes a single-bit parameterized rotation gate layer and a fixed entanglement gate layer.

[0105] The third processing module includes:

[0106] The third processing unit is used to perform parameterized rotation on the qubits of the first quantum register and the qubits of the second quantum register through the single-bit parameterized rotation gate of each layer of trainable units, and then perform a two-bit entanglement operation between the first quantum register and the second quantum register through the fixed entanglement gate of each layer of trainable units to obtain the quantum correlation between the features of the atom and the features of the chemical bond.

[0107] The first fusion unit is used to fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond according to the quantum correlation, so as to obtain the quantum state corresponding to the target molecule.

[0108] Optionally, the first prediction module includes:

[0109] The first computing unit is used to calculate the expected value of the Pauli Z operator for each qubit in the quantum state corresponding to the target molecule;

[0110] The first generation unit is used to generate the feature vector of the target molecule based on the expected values ​​of all qubits in the quantum state corresponding to the target molecule;

[0111] The first prediction unit is used to predict the properties of the target molecule based on the feature vector of the target molecule, and obtain the predicted property value of the target molecule.

[0112] Optionally, the step of predicting the properties of the target molecule based on its feature vector to obtain the predicted property value of the target molecule includes:

[0113] The feature vector of the target molecule is input into the target neural network, and the target neural network predicts the properties of the target molecule to obtain the predicted property value of the target molecule. The target neural network includes a target loss function, which is constructed based on the predicted property value and the true property value of the target molecule. The target neural network is obtained by iteratively updating the training parameters in the target quantum entanglement circuit to minimize the target loss function.

[0114] The molecular property prediction device provided in this application embodiment can achieve... Figure 1 The various processes implemented in the method embodiments shown achieve the same technical effects, and will not be described again here to avoid repetition.

[0115] Specifically, see Figure 6 As shown, this application embodiment also provides an electronic device, including a bus 601, an antenna 603, a bus interface 604, a processor 605, and a memory 606. The processor 605 is used for:

[0116] Based on the molecular structure of the target molecule, a graph structure of the target molecule is generated, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure.

[0117] Extract the feature vectors of the atoms corresponding to the nodes and the feature vectors of the chemical bonds corresponding to the edges;

[0118] The feature vector of the atom is encoded into a first quantum register to obtain the quantum state corresponding to the atom. The number of qubits in the first quantum register matches the dimension of the feature vector of the atom.

[0119] The feature vector of the chemical bond is encoded into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond.

[0120] By performing a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, the information of the quantum state corresponding to the atom and the information of the quantum state corresponding to the chemical bond are fused to obtain the quantum state corresponding to the target molecule;

[0121] Based on the quantum state corresponding to the target molecule, the properties of the target molecule are predicted to obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

[0122] Optionally, the processor 605 is specifically used for:

[0123] The qubits of the first quantum register are initialized to obtain the processed first quantum register;

[0124] The eigenvector of the atom is loaded into the processed first quantum register by using the first encoded unitary operator to obtain the quantum state corresponding to the atom.

[0125] Optionally, the processor 605 is specifically used for:

[0126] The qubits of the second quantum register are initialized and processed to obtain the processed second quantum register;

[0127] The feature vector of the chemical bond is loaded into the processed second quantum register by using the second encoded unitary operator to obtain the quantum state corresponding to the chemical bond.

[0128] Optionally, the target quantum entanglement circuit is a parameterized quantum entanglement circuit, which includes multiple layers of trainable units, each of which includes a single-bit parameterized rotation gate layer and a fixed entanglement gate layer.

[0129] The processor 605 is specifically used for:

[0130] After performing parameterized rotation on the qubits of the first quantum register and the qubits of the second quantum register through the single-bit parameterized rotation gate of each layer of trainable units, a two-bit entanglement operation is performed between the first quantum register and the second quantum register through the fixed entanglement gate of each layer of trainable units, thereby obtaining the quantum correlation between the features of the atom and the features of the chemical bond.

[0131] Based on the quantum correlation, the information of the quantum state corresponding to the atom is fused with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule.

[0132] Optionally, the processor 605 is specifically used for:

[0133] Calculate the expected value of the Pauli Z operator for each qubit in the quantum state corresponding to the target molecule;

[0134] The feature vector of the target molecule is generated based on the expected values ​​of all qubits in the quantum state corresponding to the target molecule;

[0135] Based on the feature vector of the target molecule, the properties of the target molecule are predicted.

[0136] Optionally, the processor 605 is specifically used for:

[0137] The feature vector of the target molecule is input into the target neural network, and the target neural network predicts the properties of the target molecule to obtain the predicted property value of the target molecule. The target neural network includes a target loss function, which is constructed based on the predicted property value and the true property value of the target molecule. The target neural network is obtained by iteratively updating the training parameters in the target quantum entanglement circuit to minimize the target loss function.

[0138] This application also provides an electronic device, including: a processor, a memory, and a program stored in the memory and executable on the processor. When the program is executed by the processor, it implements the various processes of the above-described molecular property prediction method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0139] This application provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the molecular property prediction method embodiments described above and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0140] The processor mentioned above is the processor in the terminal described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk. In some examples, the readable storage medium may be a non-transient readable storage medium.

[0141] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above-described... Figure 1 The various processes of the method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0142] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method of predicting molecular properties, characterized by, The method includes: Based on the molecular structure of the target molecule, a graph structure of the target molecule is generated, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure. Extract the feature vectors of the atoms corresponding to the nodes and the feature vectors of the chemical bonds corresponding to the edges; The feature vector of the atom is encoded into a first quantum register to obtain the quantum state corresponding to the atom. The number of qubits in the first quantum register matches the dimension of the feature vector of the atom. The feature vector of the chemical bond is encoded into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond. By performing a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, the information of the quantum state corresponding to the atom and the information of the quantum state corresponding to the chemical bond are fused to obtain the quantum state corresponding to the target molecule; Based on the quantum state corresponding to the target molecule, the properties of the target molecule are predicted to obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

2. The method of predicting molecular properties according to claim 1, wherein Encoding the feature vector of the atom into a first quantum register to obtain the quantum state corresponding to the atom includes: The qubits of the first quantum register are initialized to obtain the processed first quantum register; The eigenvector of the atom is loaded into the processed first quantum register by using the first encoded unitary operator to obtain the quantum state corresponding to the atom.

3. The method of predicting molecular properties according to claim 1, wherein Encoding the feature vector of the chemical bond into a second quantum register to obtain the quantum state corresponding to the chemical bond includes: The qubits of the second quantum register are initialized and processed to obtain the processed second quantum register; The feature vector of the chemical bond is loaded into the processed second quantum register by using the second encoded unitary operator to obtain the quantum state corresponding to the chemical bond.

4. The method of predicting molecular properties according to claim 1, wherein The target quantum entanglement circuit is a parameterized quantum entanglement circuit, which includes multiple layers of trainable units. Each layer of trainable units includes a single-bit parameterized rotation gate layer and a fixed entanglement gate layer. The step of performing a cross-register entanglement operation on the first and second quantum registers through a target quantum entanglement circuit to fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule includes: After performing parameterized rotation on the qubits of the first quantum register and the qubits of the second quantum register through the single-bit parameterized rotation gate of each layer of trainable units, a two-bit entanglement operation is performed between the first quantum register and the second quantum register through the fixed entanglement gate of each layer of trainable units, thereby obtaining the quantum correlation between the features of the atom and the features of the chemical bond. Based on the quantum correlation, the information of the quantum state corresponding to the atom is fused with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule.

5. The method of predicting molecular properties according to claim 1, wherein The process of predicting the properties of the target molecule based on its corresponding quantum state to obtain predicted property values ​​includes: Calculate the expected value of the Pauli Z operator for each qubit in the quantum state corresponding to the target molecule; The feature vector of the target molecule is generated based on the expected values ​​of all qubits in the quantum state corresponding to the target molecule; Based on the feature vector of the target molecule, the properties of the target molecule are predicted to obtain the predicted property values ​​of the target molecule.

6. The method of predicting molecular properties according to claim 5, wherein The step of predicting the properties of the target molecule based on its feature vector to obtain the predicted property value of the target molecule includes: The feature vector of the target molecule is input into the target neural network, and the target neural network predicts the properties of the target molecule to obtain the predicted property value of the target molecule. The target neural network includes a target loss function, which is constructed based on the predicted property value and the true property value of the target molecule. The target neural network is obtained by iteratively updating the training parameters in the target quantum entanglement circuit to minimize the target loss function.

7. A molecular property prediction apparatus characterized by comprising: The device includes: The first generation module is used to generate a graph structure of the target molecule based on the molecular structure of the target molecule, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure. The first extraction module is used to extract the feature vectors of atoms corresponding to the nodes and the feature vectors of chemical bonds corresponding to the edges; A first processing module is used to encode the feature vector of the atom into a first quantum register to obtain the quantum state corresponding to the atom, wherein the number of qubits of the first quantum register matches the dimension of the feature vector of the atom. The second processing module is used to encode the feature vector of the chemical bond into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond. The third processing module is used to perform a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, and to fuse the information of the quantum state corresponding to the atom with the information of the quantum state corresponding to the chemical bond to obtain the quantum state corresponding to the target molecule; The first prediction module is used to predict the properties of the target molecule based on the quantum state corresponding to the target molecule, and obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

8. An electronic device, comprising: The electronic device includes a processor, the processor being used for: Based on the molecular structure of the target molecule, a graph structure of the target molecule is generated, wherein the atoms of the molecular structure correspond to the nodes of the graph structure, and the chemical bonds of the molecular structure correspond to the edges of the graph structure. Extract the feature vectors of the atoms corresponding to the nodes and the feature vectors of the chemical bonds corresponding to the edges; The feature vector of the atom is encoded into a first quantum register to obtain the quantum state corresponding to the atom. The number of qubits in the first quantum register matches the dimension of the feature vector of the atom. The feature vector of the chemical bond is encoded into a second quantum register to obtain the quantum state corresponding to the chemical bond. The number of qubits in the second quantum register matches the dimension of the feature vector of the chemical bond. By performing a cross-register entanglement operation on the first quantum register and the second quantum register through the target quantum entanglement circuit, the information of the quantum state corresponding to the atom and the information of the quantum state corresponding to the chemical bond are fused to obtain the quantum state corresponding to the target molecule; Based on the quantum state corresponding to the target molecule, the properties of the target molecule are predicted to obtain the property prediction value of the target molecule. The property prediction value is used to characterize the property prediction information of the target molecule.

9. An electronic device, comprising: It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the molecular property prediction method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the molecular property prediction method as described in any one of claims 1 to 6.

11. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the molecular property prediction method as described in any one of claims 1 to 6.

Images