Artificial intelligence-based molecular processing method and device, electronic device, and computer program

Abstract

The present application provides an artificial intelligence-based molecular processing method and its device, electronic device, computer-readable storage medium and computer program product, the method including: obtaining a three-dimensional structure of a target molecule; calling a neural network model to perform energy error prediction processing on the three-dimensional structure of the target molecule to obtain an energy error of the target molecule, where the neural network model is trained by fitting the energy error of the sample molecule, and the energy error of the sample molecule refers to the difference between the calculation results obtained when calculating the energy of the sample molecule according to two kinds of energy calculation mechanisms respectively; performing a first energy calculation processing on the three-dimensional structure of the target molecule to obtain a first energy of the target molecule; and performing an error correction processing on the first energy of the target molecule based on the energy error of the target molecule to obtain a second energy of the target molecule.

Description

[Technical field]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application is filed based on the Chinese patent application bearing application number 202210980553.2, filed with the China Patent Office on August 16, 2022, and claims priority to the Chinese patent application, the entire contents of which are incorporated herein by reference.

[0002] The present application relates to artificial intelligence technology, and in particular to an artificial intelligence-based molecular processing method and apparatus, electronic device, computer-readable storage medium, and computer program product. [Background technology]

[0003] Artificial intelligence (AI) is a comprehensive technology in computer science that endows machines with the capabilities of sensing, reasoning, and decision-making through the study of various design principles and implementation methods for intelligent machines. AI technology is a comprehensive academic field that includes a wide range of fields, including natural language processing technology and machine learning / deep learning. As technology develops, AI technology is being applied to more and more fields and is playing an increasingly important role.

[0004] In the new drug research and development process and material research and development process, as shown in Figure 1, after completing target identification and validation, screening needs to be performed on candidate drug compounds. In the screening process, it is usually necessary to calculate the energy of molecules in scenarios such as calculating molecular properties. The calculation results can help drug researchers analyze molecular properties, the binding ability of molecules to protein pockets, etc. Related technologies usually use computational chemistry to obtain molecular energy, but it is not possible to consider both the accuracy of obtaining molecular energy and the speed of obtaining molecular energy. Summary of the Invention [Problem to be solved by the invention]

[0005] The embodiments of the present application provide an artificial intelligence-based molecular processing method and an apparatus therefor, an electronic device, a computer-readable storage medium, and a computer program product, which can simultaneously improve the accuracy and speed of molecular energy calculation. [Means for solving the problem]

[0006] The technical solutions of the embodiments of the present application are realized as follows.

[0007] An embodiment of the present application provides an electronic device implemented artificial intelligence based molecular processing method, the method comprising: obtaining a three-dimensional structure of a target molecule; A step of calling a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule to obtain an energy error of the target molecule, wherein the neural network model is trained by fitting the energy error of the sample molecule, the energy error of the sample molecule refers to the difference between the calculation results obtained when calculating the energy of the sample molecule according to two kinds of energy calculation mechanisms, respectively, the two kinds of energy calculation mechanisms include a first energy calculation process and a second energy calculation process, the accuracy of the first energy calculation process is lower than the accuracy of the second energy calculation process, and the speed of the first energy calculation process is faster than the speed of the second energy calculation process; performing the first energy calculation process on the three-dimensional structure of the target molecule to obtain a first energy of the target molecule; and performing an error correction process on the first energy of the target molecule based on the energy error of the target molecule to obtain a second energy of the target molecule.

[0008] An embodiment of the present application provides an artificial intelligence based molecular processing apparatus, the apparatus comprising: an acquisition module configured to acquire a three-dimensional structure of the target molecule; A neural network module configured to call a neural network model to perform an energy error prediction process on a three-dimensional structure of the target molecule to obtain an energy error of the target molecule, wherein the neural network model is trained by fitting the energy error of the sample molecule, and the energy error of the sample molecule refers to the difference between the calculation results obtained when calculating the energy of the sample molecule according to two kinds of energy calculation mechanisms, respectively, the two kinds of energy calculation mechanisms include a first energy calculation process and a second energy calculation process, the accuracy of the first energy calculation process is lower than the accuracy of the second energy calculation process, and the speed of the first energy calculation process is faster than the speed of the second energy calculation process; and a calculation module configured to perform the first energy calculation process on a three-dimensional structure of the target molecule to obtain a first energy of the target molecule; and a correction module configured to perform an error correction process on the first energy of the target molecule based on an energy error of the target molecule to obtain a second energy of the target molecule.

[0009] The embodiment of the present application is a memory storing computer executable instructions; and a processor configured to execute computer-executable instructions stored in the memory to implement an artificial intelligence based molecular processing method according to an embodiment of the present application.

[0010] An embodiment of the present application provides a computer-readable storage medium having stored thereon computer-executable instructions that, when executed by a processor, cause the processor to implement an artificial intelligence based molecular processing method according to an embodiment of the present application.

[0011] An embodiment of the present application provides a computer program product comprising a computer program or computer executable instructions which, when executed by a processor, causes the processor to implement an artificial intelligence based molecular processing method according to an embodiment of the present application. Effect of the Invention

[0012] The embodiments of the present application have the following beneficial effects:

[0013] A first energy calculation process is performed on the three-dimensional structure of the target molecule to obtain a first energy of the target molecule. Since the speed of the first energy calculation process is faster than that of the second energy calculation process, the speed of the energy calculation process is increased, and an energy error prediction process is performed on the three-dimensional structure of the target molecule using a neural network model to obtain an energy error of the target molecule, and an error correction process is performed on the first energy calculated based on the energy error to obtain a second energy of the target molecule. Since the energy error can represent the difference in the calculation result between the highly accurate second energy calculation process and the less accurate first energy calculation process, the accuracy of the second energy can be improved by correcting the first energy calculated using the energy error predicted by deep learning. [Brief description of the drawings]

[0014] [Figure 1] 1 is an exemplary flow chart of pharmaceutical research and development according to an embodiment of the present application. [Diagram 2] 1 is a schematic diagram of the architecture of an artificial intelligence based molecular processing system according to an embodiment of the present application; [Diagram 3] FIG. 2 is an exemplary structural diagram of an electronic device according to an embodiment of the present application. [Figure 4A] 1 is an exemplary flow chart of an artificial intelligence based molecular processing method according to an embodiment of the present application. [Figure 4B] 1 is an exemplary flow chart of an artificial intelligence based molecular processing method according to an embodiment of the present application. [Figure 4C]1 is an exemplary flow chart of an artificial intelligence based molecular processing method according to an embodiment of the present application. [Diagram 5] FIG. 2 is an exemplary structural diagram of a deep quantum chemical model according to an embodiment of the present application. [Figure 6] FIG. 11 is a comparison diagram of the effects of the first data set according to an embodiment of the present application. [Figure 7] FIG. 11 is a comparison diagram of the effect of the second data set according to an embodiment of the present application. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in more detail hereinafter with reference to the drawings, and the described embodiments should not be understood as limitations to the present application, and all other embodiments obtained by those skilled in the art without creative efforts shall be included in the protection scope of the present invention.

[0016] In the following description, the term "some embodiments" refers to a subset of all possible embodiments, and it is understood that the terms "some embodiments" may refer to the same or different subsets of all possible embodiments, which may be combined with each other without conflict.

[0017] The terms "first / second / third" referred to in the following description are merely intended to distinguish between similar objects and do not represent a particular order of the objects, and it is to be understood that "first / second / third" may, in some cases, be interchangeable with a particular order or precedence, whereby the embodiments of the present application described herein may be performed in an order other than that illustrated or described.

[0018] Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to be limiting of the present application.

[0019] Before describing the embodiments of the present application in more detail, the nouns and terms referred to in the embodiments of the present application will be explained. The nouns and terms referred to in the embodiments of the present application have the following interpretations.

[0020] 1) Computational Chemistry is a branch of theoretical chemistry that aims primarily at calculating molecular properties, such as total energy, dipole moment, quadrupole moment, vibrational frequencies, reactivity, etc., using valid mathematical approximations and computer programs, and is used to interpret some specific chemical problems.

[0021] 2) Molecular energy includes the kinetic energy and potential energy of a molecule. The kinetic energy of a molecule refers to the energy due to the movement of a molecule, and the potential energy of a molecule refers to the potential energy that is generated by the molecular forces possessed by a molecule and is determined by its relative position.

[0022] 3) Semi-empirical quantum mechanical methods: Theoretically, quantum mechanical methods have high accuracy, but when faced with biological macromolecular systems, the computational cost is very high and cannot meet the needs of practical applications. Semi-empirical quantum mechanical methods that introduce empirical parameters are approximate methods that trade off time and accuracy, and can be used to score and estimate the affinity between ligands and proteins, which is of great significance in computer-aided drug design.

[0023] 4) Delta Learning refers to the ability of a learning system to continually learn new knowledge from new examples and retain most of the knowledge it has previously learned. Delta Learning is very similar to humans' own learning mode.

[0024] 5) The first-principles method (a method of calculating from scratch) is a quantum chemical calculation method that directly solves the Schrödinger equation based on the basic principles of quantum mechanics. The first-principles method is characterized by the lack of empirical parameters and the fact that the system is not overly simplified. Calculations are performed for a variety of different chemical systems using essentially the same method.

[0025] 6) Density Functional Theory is a method for studying the electronic structure of many-body electron systems. Density Functional Theory has a wide range of applications in physics and chemistry, and is particularly used to study the properties of molecules and aggregate states. It is one of the most used methods in the fields of aggregate state physics, materials science, and computational chemistry.

[0026] Related molecular energy calculation methods include two types: a computational chemistry-based method and a deep learning-based method. Computational chemistry methods include quantum mechanics methods and molecular mechanics (MM) methods.

[0027] Quantum mechanical methods include the first-principles method and the semi-empirical quantum mechanical methods, and the first-principles method is a method that solves the Schrödinger equation using rigorous approximations based on the first principles of quantum chemistry. The first-principles method includes a method based on the wave function and a method based on the density functional, and a representative method based on the wave function is based on the Hartree-Fock equation, and a representative method based on the density functional is based on the density functional theory.

[0028] Molecular mechanics methods typically use a molecular force field to describe the effect of various forms of interaction forces on the potential energy of a molecule, which does not calculate electronic interactions and is a simplified model of molecular structure.

[0029] In the deep learning method, a deep learning model is mainly used to predict molecular energy, and a small data set is used as a set of training samples to train the deep learning model based on the set of training samples, so that the deep learning model can learn the features of the atomic structure and predict the molecular energy based on the features obtained by extraction.

[0030] The first-principles method of quantum mechanics has high calculation accuracy, but requires a large amount of calculation, which takes a long time to calculate and makes it difficult to perform large-scale calculations using the first-principles method. The semi-empirical quantum mechanics method of quantum mechanics uses semi-empirical parameters instead of molecular integrals, which can speed up calculations but sacrifices calculation accuracy.

[0031] Methods based on molecular mechanics force fields can achieve faster computation speeds, but are also the least accurate.

[0032] Deep learning methods do not fully utilize the features of atoms, support few atom types, have low ability to capture long-range interactions of molecules, and do not explore predictions for the energy of different molecular conformations. Simple deep learning methods need to improve the accuracy of molecular energy prediction.

[0033] In order to solve the above problems, the embodiments of the present application provide an artificial intelligence-based molecular processing method and its device, electronic device, computer-readable storage medium, and computer program product, which can simultaneously improve the accuracy and speed of molecular energy calculation.

[0034] The artificial intelligence-based molecular processing method according to the embodiment of the present application may be implemented by the terminal / server alone, or may be implemented by the terminal and the server together. For example, the terminal independently implements the artificial intelligence-based molecular processing method described below, or the terminal transmits an energy evaluation request for the target molecule to the server, and the server executes the artificial intelligence-based molecular processing method based on the received energy evaluation request for the target molecule to obtain the three-dimensional structure of the target molecule, invokes a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule, obtains the energy error of the target molecule, performs a first energy calculation process on the three-dimensional structure of the target molecule, obtains the first energy of the target molecule, and performs an error correction process on the first energy of the target molecule based on the energy error of the target molecule to obtain the second energy of the target molecule. Thus, the researcher can perform subsequent analysis research based on the second energy of the target molecule, for example, determine the binding ability of the target molecule to a protein pocket based on the second energy of the target molecule, and perform screening for candidate pharmaceutical compounds through the binding ability of the target molecule to the protein pocket.

[0035] The electronic device for molecular processing according to the embodiment of the present application may be various types of terminal devices or servers, where the server may be an independent physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services, and the terminal may be, but is not limited to, a smartphone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc. The terminal and the server may be directly or indirectly connected by a wired or wireless communication method, and are not limited in the present application.

[0036] Taking a server as an example, it may be, for example, a server cluster deployed in the cloud, and opens artificial intelligence services (AIaaS: AI as a Service) to users. The AIaaS platform divides some common AI services and provides independent or packaged services on the cloud. This service mode is similar to AI themoll, and all users can access and use one or more artificial intelligence services provided by the AIaaS platform through an application programming interface.

[0037] For example, the artificial intelligence service here may be a molecular processing service, that is, a program for molecular processing according to the embodiment of the present application is encapsulated in a server of the cloud. A user calls the molecular processing service in the cloud service through a terminal (where a client such as a compound screening client is running), and the server arranged on the cloud calls the encapsulated program for molecular processing to obtain the three-dimensional structure of the target molecule, calls a neural network model, performs an energy error prediction process on the three-dimensional structure of the target molecule, obtains the energy error of the target molecule, performs a first energy calculation process on the three-dimensional structure of the target molecule, obtains the first energy of the target molecule, and performs an error correction process on the first energy of the target molecule based on the energy error of the target molecule to obtain the second energy of the target molecule. Thus, a researcher can perform subsequent analysis research based on the second energy of the target molecule, for example, determine the binding ability of the target molecule to a protein pocket based on the second energy of the target molecule, and perform screening for candidate pharmaceutical compounds through the binding ability of the target molecule to the protein pocket.

[0038] Referring to FIG. 2, FIG. 2 is a schematic diagram of the architecture of an artificial intelligence based molecular processing system according to an embodiment of the present application, in which a terminal 400 connects to a server 200 via a network 300, which may be a wide area network or a local area network, or a combination of both.

[0039] The terminal 400 (running a client such as a compound screening client) can be used to obtain an energy evaluation request for a target molecule. For example, when a researcher inputs a target molecule through an input interface of the terminal 400, an energy evaluation request for the target molecule is automatically generated. The terminal 400 transmits an energy prediction request for the target molecule to the server 200, and the server 200 obtains a three-dimensional structure of the target molecule, invokes a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule, obtains an energy error of the target molecule, performs a first energy calculation process on the three-dimensional structure of the target molecule, obtains a first energy of the target molecule, performs an error correction process on the first energy of the target molecule based on the energy error of the target molecule, obtains a second energy of the target molecule, and the server 200 returns the second energy of the target molecule to the terminal 400. Thereby, the researcher can perform subsequent analysis research based on the second energy of the target molecule, for example, determine the binding ability of the target molecule to a protein pocket based on the second energy of the target molecule, and perform screening for a candidate pharmaceutical compound through the binding ability of the target molecule to the protein pocket.

[0040] In some embodiments, the molecular processing plug-in can be embedded in a client running on the terminal to locally implement the artificial intelligence-based molecular processing method on the client. For example, after the terminal 400 obtains an energy evaluation request for a target molecule, it calls the molecular processing plug-in to implement the artificial intelligence-based molecular processing method, obtains the three-dimensional structure of the target molecule, calls a neural network model, performs an energy error prediction process on the three-dimensional structure of the target molecule, obtains the energy error of the target molecule, performs a first energy calculation process on the three-dimensional structure of the target molecule, obtains the first energy of the target molecule, performs an error correction process on the first energy of the target molecule based on the energy error of the target molecule, and obtains the second energy of the target molecule. Thus, the researcher can perform subsequent analysis research based on the second energy of the target molecule, for example, determine the binding ability of the target molecule to a protein pocket based on the second energy of the target molecule, and perform screening for candidate pharmaceutical compounds through the binding ability of the target molecule to the protein pocket.

[0041] In some embodiments, after the terminal 400 obtains the energy evaluation request for the target molecule, it calls the molecular processing interface of the server 200 (which can be provided in the form of a cloud service, i.e., as a molecular processing service), and the server 200 obtains the three-dimensional structure of the target molecule, calls the neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule, obtains the energy error of the target molecule, performs a first energy calculation process on the three-dimensional structure of the target molecule, obtains the first energy of the target molecule, and performs an error correction process on the first energy of the target molecule based on the energy error of the target molecule to obtain the second energy of the target molecule. Thus, the researcher can perform subsequent analysis research based on the second energy of the target molecule, for example, determine the binding ability of the target molecule to a protein pocket based on the second energy of the target molecule, and perform screening for candidate pharmaceutical compounds through the binding ability of the target molecule to the protein pocket.

[0042] Referring to FIG. 3, FIG. 3 is an exemplary structural diagram of an electronic device according to an embodiment of the present application, and the electronic device shown in FIG. 3, a terminal 400, includes at least one processor 410, a memory 450, at least one network interface 420, and a user interface 430. Each component in the terminal 400 is coupled by a bus system 440. It can be understood that the bus system 440 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 440 further includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, various buses are labeled as the bus system 440 in FIG. 3.

[0043] The processor 410 may be an integrated circuit chip having signal processing capabilities, such as a general purpose processor, a digital signal processor (DSP), or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., where the general purpose processor may be a microprocessor or any conventional processor, etc.

[0044] The user interface 430 includes one or more output devices 431 that enable media content to be displayed, said output devices 431 including one or more speakers and / or one or more visual display screens, and further includes one or more input devices 432 that include user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons or controls, etc.

[0045] Memory 450 may be removable, non-removable, or a combination thereof, and example hardware devices thereof include solid-state memory, hard disk drives, optical disk drives, etc. Memory 450 optionally includes one or more storage devices that are remote in physical location from processor 410.

[0046] The memory 450 may include volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 450 described in the present embodiments is intended to include any suitable type of memory.

[0047] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as illustratively described below.

[0048] The operating system 451 includes system programs for processing various basic system services and executing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., which are used to realize various basic services and process hardware-based tasks.

[0049] The network communications module 452 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 420; example network interfaces 420 include Bluetooth® technology, Wireless Fidelity (WiFi), and Universal Serial Bus (USB), among others.

[0050] The display module 453 is used to enable information to be displayed via one or more output devices 431 (e.g., a display screen, speakers, etc.) associated with the user interface 430 (e.g., the user interface for operating peripherals and displaying content and information).

[0051] The input processing module 454 is used to detect one or more user inputs or interactions from one of the one or more input devices 432 and to convert the detected inputs or interactions.

[0052] In some embodiments, the artificial intelligence based molecular processing device according to the embodiments of the present application can be realized in the form of software, and FIG. 3 shows an artificial intelligence based molecular processing device 455 stored in memory 450. The artificial intelligence based molecular processing device 455 can be software in the form of a program, plug-in, etc., and includes the following software modules: an acquisition module 4551, a neural network module 4552, a calculation module 4553, a correction module 4554, and a training module 4555. These modules are logical, and therefore can be arbitrarily combined or further divided based on the functions to be realized. The functions of each module will be described later.

[0053] As mentioned above, the artificial intelligence based molecular processing methods according to embodiments of the present application can be implemented by various types of electronic devices.

[0054] In the following, a molecular processing method based on artificial intelligence according to an embodiment of the present application will be described, and similarly, the electronic device that realizes the molecular processing method based on artificial intelligence according to the embodiment of the present application may be a terminal. Therefore, in the following, the execution subject of each step will not be described repeatedly.

[0055] Please refer to FIG. 4A, which is an exemplary flow chart of an artificial intelligence based molecular processing method according to an embodiment of the present application, which will be described with reference to steps 101 to 104 shown in FIG. 4A.

[0056] In step 101, the three-dimensional structure of the target molecule is obtained.

[0057] As an example, a target molecule has multiple molecular conformations, each of which has a corresponding three-dimensional structure, the target molecule comprising at least one atom, and for a target molecule having multiple atoms, the three-dimensional structure is a conformation consisting of multiple atoms and at least one chemical bond.

[0058] In step 102, the neural network model is invoked to perform an energy error prediction process on the three-dimensional structure of the target molecule to obtain the energy error of the target molecule.

[0059] As an example, a neural network model is used to perform feature extraction processing on the three-dimensional structure of a target molecule, energy error features are obtained, and the energy error features are mapped to the difference in the calculation results obtained when performing energy calculation on the target molecule using two given energy calculation mechanisms. The neural network model performs fitting based on the feature mapping to obtain the energy error, and in fact, it is not necessary to perform energy calculation on the target molecule using two given energy calculation mechanisms. This is because the function that the neural network model can realize depends on the training method. The neural network model is trained by fitting the energy error of the sample molecule, and the energy error of the sample molecule refers to the difference in the calculation results obtained when calculating the energy of the sample molecule according to two energy calculation mechanisms, respectively, and the energy calculation mechanism represents the method of performing energy calculation on a given molecule, and the method of energy calculation may be an energy calculation process based on computational chemistry, and the energy calculation process based on computational chemistry includes a quantum mechanics method and a molecular mechanics method. The quantum mechanics method includes a first-principles method (e.g., calculation using the Psi4 tool) and a semi-empirical quantum mechanics method (e.g., calculation using the GFN2-xTB program and the GFN-xTB program), and the two kinds of energy calculation mechanisms respectively correspond to a first energy calculation process and a second energy calculation process, the accuracy of the first energy calculation process is smaller than the accuracy of the second energy calculation process, and the speed of the first energy calculation process is faster than the speed of the second energy calculation process.

[0060] In step 103, a first energy calculation process is performed on the three-dimensional structure of the target molecule to obtain a first energy of the target molecule.

[0061] As an example, the first energy calculation process and the second energy calculation process may be energy calculation processes of computational chemistry, and the energy calculation process of computational chemistry includes a quantum mechanics method and a molecular mechanics method. The quantum mechanics method includes a first-principles method (e.g., calculation using the Psi4 tool) and a semi-empirical quantum mechanics method (e.g., using the GFN2-xTB program and the GFN-xTB program), and the first energy calculation process may be a semi-empirical quantum mechanics method, and the second energy calculation process may be a first-principles method, and the first-principles method is a method that solves the Schrödinger equation using rigorous approximation based on the first principles of quantum chemistry. The first-principles method includes a method based on a wave function and a method based on a density functional, and a representative method based on a wave function is a method based on the Hartree-Fock equation, and a representative method based on a density functional is a method based on density functional theory.

[0062] In step 104, an error correction process is performed on the first energy of the target molecule based on the energy error of the target molecule to obtain a second energy of the target molecule.

[0063] As an example, the error correction process refers to performing an addition or subtraction process on the first energy and the energy error. When the energy error represents a difference in the calculation result between the first energy calculation process and the second energy calculation process, the error correction process is to perform a subtraction process on the first energy and the energy error to obtain the second energy of the target molecule. When the energy error represents a difference in the calculation result between the second energy calculation process and the first energy calculation process, the error correction process is to perform an addition process on the first energy and the energy error to obtain the second energy of the target molecule.

[0064] In some embodiments, refer to Fig. 4B, which is an exemplary flowchart of a molecular processing method based on artificial intelligence according to an embodiment of the present application. In step 102, the step of calling a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule and obtaining the energy error of the target molecule can be realized through steps 1021 to 1022 shown in Fig. 4B.

[0065] In step 1021, a feature extraction process is performed on the three-dimensional structure using the neural network model to obtain the energy error features of the target molecule.

[0066] In step 1022, a neural network model is used to perform a full-connection process on the energy error features to obtain the energy error of the target molecule.

[0067] The process of extracting energy error features of the three-dimensional structure using an artificial intelligence method and performing the feature extraction process in step 1021 will be described in detail in the following content. Based on the energy error features, the energy error of the target molecule can be predicted, and the difference between the calculation results of the molecular energy of the target molecule calculated using two types of energy calculation processes can be intelligently determined.

[0068] In some embodiments, the neural network model includes N cascaded feature networks, and the step of performing feature extraction processing on the three-dimensional structure using the neural network model in step 1021 to obtain energy error features of the target molecule can be realized through the following steps A to C.

[0069] In step A, an initial feature extraction process is performed on each atom in the three-dimensional structure to obtain the initial features of each atom.

[0070] As an example, when the target molecule E includes three atoms (atom A, atom B and atom C) and one chemical bond (atom A and atom B are connected by a chemical bond), an initial feature extraction process is performed for each atom to obtain an initial attribute feature of each atom in the three-dimensional structure, the initial attribute feature represents the attribute information of the atom (the attribute information includes at least one of the atom type, the atom characteristics, etc.), the initial coordinate feature of each atom in the three-dimensional structure is obtained, and the initial coordinate feature represents the position information of the atom (the position information refers to the three-dimensional coordinate of each atom, and the three-dimensional coordinate of the other two atoms can be determined by establishing a three-dimensional coordinate system with any one atom of the three atoms as the origin), and the following process is performed for each atom in the three-dimensional structure, for example, the following process is performed for atom A. At least one other atom (i.e., atom B and atom C) other than atom A in the three-dimensional structure is obtained, and the initial relationship feature between the atom and each other atom is obtained. The initial relationship feature represents the bond relationship between the atom and other atoms. The initial relationship feature represents the bond relationship between atom A and atom B and the bond relationship between atom A and atom C. There are two types of bond relationships: bonded relationship and non-bonded relationship. If two atoms are connected by a chemical bond, the two atoms are in a bonded relationship. If two atoms are not connected by a chemical bond, the two atoms are in a non-bonded relationship. The initial attribute feature of each atom, the initial coordinate feature of each atom, and the initial relationship feature of each atom are the initial features of each atom.

[0071] By using the initial attribute features, initial relationship features and initial coordinate features as initial features, it is possible to not only represent the type attributes of atoms, but also represent the relative positions of atoms and their connection status through chemical bonds, thereby effectively improving the expressive ability of the initial features.

[0072] In step B, if the value of n is 1≦n≦N-1, the nth feature network among the N cascaded feature networks is used to perform the nth feature extraction process on the input of the nth feature network to obtain the nth feature of each atom, and then the nth feature is transmitted to the n+1th feature network.

[0073] As an example, the value range of N satisfies 2≦N, the value of n is an integer increasing from 1, and the value range of n satisfies 1≦n≦N. When the value of n is 1, the input of the nth feature network is the initial feature of each atom, and when the value of n is 2≦n≦N, the input of the nth feature network is the n-1th feature of each atom output by the n-1th feature network.

[0074] As an example, assuming that the value of N is 3, the following process is still performed by taking atom A as an example: a first feature extraction process is performed on the initial feature of atom A using a first feature network to obtain a first feature of atom A, the first feature is transmitted to a second feature network to continue with a second feature extraction process, and a second feature extraction process is performed on the first feature of atom A using the second feature network to obtain a second feature of atom A.

[0075] In step C, when the value of n is N, perform attribute feature extraction processing on the (n-1)th feature of each atom using the (n)th feature network to obtain the (n)th attribute feature of each atom, perform coordinate feature extraction processing on the (n-1)th feature of each atom using the (n)th feature network to obtain the (n)th coordinate feature of each atom, and form an energy error feature using the (n)th attribute feature and the (n)th coordinate feature of each atom.

[0076] As an example, assuming that the value of N is 3, the following process is still performed by taking atom A as an example: a third attribute feature extraction process is performed on the second feature of atom A using the third feature network to obtain a third attribute feature of atom A, a third coordinate feature extraction process is performed on the second feature of atom A using the third feature network to obtain a third coordinate feature of atom A, and the third attribute feature of atom A and the third coordinate feature of atom A are taken as energy error features.

[0077] In the embodiment of the present application, the coordinate features and attribute features of each atom are obtained using a method of successive substitution, and an energy error feature is formed using the coordinate features of multiple atoms and the attribute features of multiple atoms, thereby effectively improving the feature representation ability of the energy error feature, and thereby improving the accuracy of subsequent predicted energy error.

[0078] In some embodiments, the step of performing an n-th feature extraction process on the input of the n-th feature network using an n-th feature network among the N cascaded feature networks in step B to obtain the n-th feature of each atom can be realized by executing the following steps B1 to B3 for each atom (atom A in target molecule E will be described as an example) using the n-th feature network.

[0079] In step B1, other atoms excluding the atoms in the three-dimensional structure are obtained, still taking the target molecule E as an example, the other atoms are other atoms B and other atoms C excluding atom A.

[0080] In step B2, a first mapping process is performed on the n-1th feature of the atom and the n-1th feature of each other atom to obtain an n-th associated feature of the atom that corresponds to each other atom.

[0081] The n-1th coordinate feature of the atom is extracted from the n-1th feature of the atom, the n-1th coordinate feature of each of the other atoms is extracted from the n-1th feature of each of the other atoms, the n-1th attribute feature of the atom is extracted from the n-1th feature of the atom, the n-1th attribute feature of each of the other atoms is extracted from the n-1th feature of each of the other atoms, the n-1th relationship feature of the atom is extracted from the n-1th feature of the atom, and the following processing is performed for each of the other atoms (atom B is used as an example in the following description): the n-1th relationship feature of the atom with respect to the other atoms is extracted from the n-1th relationship feature of the atom, a first feature distance is obtained between the n-1th coordinate feature of the atom and the n-1th coordinate feature of the other atoms, and a first fusion processing is performed on the square of the first feature distance, the n-1th attribute feature of the atom, the n-1th attribute feature of the other atoms, and the n-1th relationship feature of the atom with respect to the other atoms to obtain the nth association feature of the atom corresponding to the other atoms.

[0082] According to the embodiment of the present application, the feature distance between any two atoms and the bonding relationship between any two atoms are taken into consideration in the association feature, so that the association feature can learn global information of the three-dimensional structure, and the global information learning ability of the neural network model can be improved.

[0083] As an example, suppose the value of N is 3, still take atom A and another atom B as an example, extract the second coordinate feature of atom A from the second feature of atom A, extract the second coordinate feature of atom B from the second feature of atom B, extract the second attribute feature of atom from the second feature of atom A, and extract the second attribute feature of atom B from the second feature of atom B. The n-1th relationship feature is the initial relationship feature, that is, the relationship feature does not change through feature iterative substitution, and extract the second relationship feature of atom A (the initial relationship feature of atom A) from the second feature of atom A, and extract the second relationship feature of atom B (the initial relationship feature of atom B) from the second feature of atom B.

[0084] As an example, referring to formula (1), subsequent processing is performed on another atom B.

[0085]

number

number

number

[0086] In step B3, a second mapping process is performed on the (n-1)th feature of the atom and the nth associated features of the atom corresponding to each other atom to obtain the nth feature of the atom.

[0087] performing an addition process on the n-th association feature of the atom corresponding to the multiple other atoms to obtain the n-th association feature of the atom; performing a second fusion process on the n-1th attribute feature of the atom and the n-th association feature of the atom to obtain the n-th attribute feature of the atom; obtaining a first feature difference between the n-1th coordinate feature of the atom and the n-1th coordinate feature of each other atom; performing a linear mapping process on the n-th association feature of the atom corresponding to each other atom to obtain a weight of each other atom; performing a weighted averaging process on the first feature differences of the multiple other atoms based on the weight of each other atom to obtain a weighted average result corresponding to the atom; performing an addition process on the weighted average result of the atom and the n-1th coordinate feature of the atom to obtain the n-th coordinate feature of the atom; taking the initial relationship feature of the atom as the n-th relationship feature of the atom; and forming the n-th feature of the atom using the n-th relationship feature of the atom, the n-th attribute feature of the atom and the n-th coordinate feature of the atom.

[0088] According to the embodiment of the present application, the feature distance between any two atoms and the bonding relationship between any two atoms are taken into consideration in the n-th feature, so that the n-th feature can learn global information of the three-dimensional structure, and the global information learning ability of the neural network model can be improved.

[0089] As an example, referring to formula (2), an addition process is performed on the n-th associated feature of atom A corresponding to a plurality of other atoms (other atom B and other atom C) to obtain the n-th associated feature of atom A.

[0090]

number

[0091] As an example, referring to formula (3), a second fusion process is performed on the (n-1)th attribute feature of atom A and the nth association feature of atom A to obtain the nth attribute feature of atom A.

[0092]

number

number

[0093] As an example, a first feature difference between the n-1th coordinate feature of atom A and the n-1th coordinate feature of other atoms (other atom B and other atom C) is obtained, and a weighted average process is performed on the first feature differences of multiple other atoms (other atom B and other atom C) using the linear mapping result of the nth association feature of atom A corresponding to each other atom (other atom B and other atom C) as a weight to obtain a weighted average result corresponding to atom A, and an addition process is performed on the weighted average result of atom A and the n-1th coordinate feature of atom A to obtain the nth coordinate feature of atom A. The above processes for atom A and other atoms can refer to formula (4).

[0094]

number

number

[0095] As an example, the initial relation feature of atom A is the n-th relation feature of atom A, and the n-th relation feature a of atom A corresponding to each other atom is ij , n-th attribute feature h of atom A i n and the nth coordinate feature of atom A is used to form the nth feature of the atom.

[0096] In some embodiments, the step C of performing attribute feature extraction processing on the n-1th feature of each atom using the nth feature network to obtain the nth attribute feature of each atom is realized through the following technical solutions: Extract the n-1th coordinate feature of the atom, the n-1th attribute feature of the atom and the n-1th relationship feature of the atom from the n-1th feature of each atom, and perform the following processing for each atom: Obtain other atoms excluding the atom in the three-dimensional structure, and perform the following processing for each other atom. An n-1th relationship feature of the atom with respect to other atoms is extracted from the n-1th relationship feature of the atom, a second feature distance between the n-1th coordinate feature of the atom and the n-1th coordinate feature of the other atoms is obtained, a first fusion process is performed on the square of the second feature distance, the n-1th attribute feature of the atom, the n-1th attribute feature of the other atoms, and the n-1th relationship feature of the atom with respect to other atoms to obtain an n-th association feature of the atom corresponding to the other atoms, an addition process is performed on the n-th association features of the atom corresponding to the multiple other atoms to obtain the n-th association feature of the atom, and a second fusion process is performed on the n-1th attribute feature of the atom and the n-th association feature of the atom to obtain the n-th attribute feature of the atom.

[0097] As an example, extract the n-1th coordinate feature of the atom, the n-1th attribute feature of the atom and the n-1th relationship feature of the atom from the n-1th feature of each of the atoms, specifically, assuming that the value of N is 3, still take atoms A, B and C as an example to perform the following processing: extract the second coordinate feature of atom A from the second feature of atom A, extract the second coordinate feature of atom B from the second feature of atom B, extract the second coordinate feature of atom C from the second feature of atom C, extract the second attribute feature of atom A from the second feature of atom A, extract the second attribute feature of atom B from the second feature of atom B, and extract the second attribute feature of atom C from the second feature of atom C. The (n-1)th relationship feature is the initial relationship feature, i.e., the relationship feature does not change due to the iterative substitution of features, and the second relationship feature of atom A (the initial relationship feature of atom A) is extracted from the second feature of atom A, the second relationship feature of atom B (the initial relationship feature of atom B) is extracted from the second feature of atom B, and the second relationship feature of atom C (the initial relationship feature of atom C) is extracted from the second feature of atom C.

[0098] As an example, referring to formula (5), for each atom (atom A will be described as an example), other atoms B and C excluding the atoms in the three-dimensional structure are obtained, and subsequent processing is performed on each other atom (other atom B will be described as an example).

[0099]

number

number

number

[0100] As an example, referring to formula (6), an addition process is performed on the n-th associated feature of atom A corresponding to a plurality of other atoms (other atom B and other atom C) to obtain the n-th associated feature of atom A.

[0101]

number

[0102] As an example, referring to formula (7), a second fusion process is performed on the (n-1)th attribute feature of atom A and the nth association feature of atom A to obtain the nth attribute feature of atom A.

[0103]

number

number

[0104] In step C, a coordinate feature extraction process is performed on the (n-1)th feature of each of the atoms using the n feature network. For a specific embodiment of obtaining the n-th coordinate feature of each of the atoms, reference may be made to the specific embodiment of obtaining the (n-1)th coordinate feature.

[0105] In some embodiments, referring to FIG. 4C , before invoking a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule and obtaining the energy error of the target molecule, steps 105 to 109 shown in FIG. 4C can be further performed.

[0106] In step 105, a sample molecule is obtained, and a three-dimensional structure generation process is performed on the sample molecule to obtain a plurality of three-dimensional structures of the sample molecule.

[0107] As an example, the sample molecule may be from any molecular dataset, such as the QMugs dataset, and a structure generation process may be performed on the sample molecule to obtain a plurality of sample molecular structures, the structure generation process including twisting chemical bonds, and the obtained plurality of sample molecular structures may include the molecular structure of the sample molecule itself and other molecular structures obtained by twisting the chemical bonds, and the structure generation process may not change the type of each atom in the molecule, but may change the coordinates of each atom and the distance between each atom in the molecule.

[0108] In step 106, the label energy error of each sample molecular conformation is obtained.

[0109] In some embodiments, obtaining the label energy error of each sample molecular conformation in step 106 can be realized through the following technical solution: For each sample molecular conformation, perform the following process: perform a first energy calculation process on the sample molecular conformation to obtain a first energy of the sample molecular conformation, perform a second energy calculation process on the sample molecular conformation to obtain a second energy of the sample molecular conformation, and obtain a first difference between the second energy of the sample molecule and the first energy of the sample molecule as the label energy error of the sample molecular conformation.

[0110] As an example, the molecular energy of a molecule is calculated using a highly accurate second energy processing method, for example, using the DFT calculation theory level and basis set "WB97X-D3 / def2-TZVP" to calculate the quantum mechanical energy (single-point energy) of the molecule to be used as the molecular energy, and the molecular energy obtained by the calculation is denoted as E_dft, and at the same time, the molecular energy of the molecule is calculated using a high-speed first energy processing method, for example, using a semi-empirical quantum mechanical method to calculate the semi-empirical quantum mechanical energy of the molecule to be used as the molecular energy, and the molecular energy obtained by the calculation is denoted as E_xtb, and the difference between the two, E_delta = E_dft - E_xtb, is used as the label energy error (Label) of the molecular three-dimensional structure.

[0111] In step 107, each sample molecular conformation is forward propagated through the initialized neural network model to obtain a predicted energy error for each sample molecular conformation.

[0112] As an example, the initialized neural network model is used to perform feature extraction processing on the three-dimensional structure to obtain the energy error feature of the target molecule, and the initialized neural network model is used to perform a full-connection processing on the energy error feature to obtain the energy error of the target molecule. The processing process of the initialized neural network model can refer to the above embodiment, with the only difference being that the parameters used in the processing process are initialized parameters, not parameters obtained by training.

[0113] In step 108, a total loss corresponding to the neural network model is determined based on the label energy error of each sample molecular conformation and the predicted energy error of each sample molecular conformation.

[0114] In some embodiments, the step of determining the overall loss corresponding to the neural network model based on the label energy error of each sample molecular conformation and the predicted energy error of each sample molecular conformation in step 108 can be realized through the following technical solutions. For each sample molecular conformation, perform a process of determining a first root mean square error of the sample molecular conformation according to the label energy error of the sample molecular conformation and the predicted energy error of the sample molecular conformation; obtain other sample molecular conformations other than the sample molecular conformation of the sample molecule; for each other sample molecular conformation, determine a second difference between the label energy error of the sample molecular conformation and the label energy error of the other sample molecular conformation; perform a process of determining a third difference between the predicted energy error of the sample molecular conformation and the predicted energy error of the other sample molecular conformation; perform a root mean square process on the second difference and the third difference to obtain a second root mean square error of the sample molecular conformation corresponding to the other sample molecular conformation; perform an addition process on the second root mean square errors of the sample molecular conformations corresponding to the multiple other sample molecular conformations to obtain a third root mean square error of the sample molecular conformation; perform a third fusion process on the first root mean square errors of the multiple sample molecular conformations and the third root mean square errors of the multiple sample molecules to obtain a comprehensive loss corresponding to the neural network model.

[0115] As an example, the total loss of a neural network model is given by equation (8).

[0116]

number

number

number

number

number

number

[0117] In step 109, backpropagation processing is performed on the total loss in the neural network model to obtain parameter change values ​​of the neural network model when the total loss converges, and the parameters of the neural network model are updated based on the parameter change values.

[0118] As an example, backpropagation can be realized by using the backpropagation algorithm, which mainly cycles through two segments (excitation propagation, weight update) in an iterative manner until the network's response to the input reaches a predetermined target range. The learning process of the backpropagation algorithm consists of a forward propagation process and a backpropagation process. In the forward propagation process, the input information is processed layer by layer from the input layer through the implicit layer and sent to the output layer. If the desired output value is not obtained in the output layer, the sum of squares of the output and the desired error is used as the target function and passed to backpropagation, and the partial derivative of the target function with respect to each neuron weight is obtained along each line, and the gradient of the target function with respect to the weight orientation is constructed as the basis for correcting the weights, and the learning of the network is completed in the weight correction process. When the overall loss converges to the desired value, the learning of the neural network model is completed.

[0119] The following describes an exemplary application of the present embodiment in one practical application scenario.

[0120] A terminal (running a client such as a compound screening client) can be used to obtain an energy evaluation request for a target molecule. For example, when a researcher inputs a target molecule through an input interface of the terminal, an energy evaluation request for the target molecule is automatically generated. The terminal transmits an energy prediction request for the target molecule to a server, and the server obtains a three-dimensional structure of the target molecule, invokes a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule, obtains an energy error of the target molecule, performs a first energy calculation process on the three-dimensional structure of the target molecule, obtains a first energy of the target molecule, performs an error correction process on the first energy of the target molecule based on the energy error of the target molecule, obtains a second energy of the target molecule, and the server returns the second energy of the target molecule to the terminal. Thus, the researcher can perform subsequent analysis studies based on the second energy of the target molecule, for example, determine the binding ability of the target molecule to a protein pocket based on the second energy of the target molecule, and perform screening for candidate pharmaceutical compounds through the binding ability of the target molecule to the protein pocket.

[0121] The molecular processing method based on artificial intelligence according to the embodiment of the present application can be applied to the new drug research and development process and the material research and development process. Taking the new drug research and development process as an example, as shown in FIG. 1, in the new drug research and development process, after completing target identification and validation, screening needs to be performed on the candidate drug compounds. In the screening process, it is necessary to calculate the energy of the molecule in scenarios such as calculating molecular properties and calculating the binding ability of the molecule to the protein pocket. The calculation results can help drug researchers analyze molecular properties, the binding ability of the molecule to the protein pocket, etc., and help researchers design more effective drug molecules, greatly improve the efficiency of research and development, and reduce the cost of drug research and development.

[0122] First, a training dataset for the Deep Quantum Chemistry (DeepQC) model needs to be constructed. An enhanced dataset is constructed based on an arbitrary dataset (e.g., the QMugs dataset). The enhanced dataset contains 660,000 types of molecules, and each type of molecule has three molecular conformations, resulting in a total of nearly 2 million data.

[0123] For each molecular conformation in the enhanced dataset, the molecular energy of the molecule is calculated using the highly accurate second energy processing method. For example, the quantum mechanical energy (single-point energy) of the molecule is calculated using the DFT calculation theory level and basis set "WB97X-D3 / def2-TZVP" to obtain the molecular energy, and the molecular energy obtained by the calculation is denoted as E_dft. At the same time, the molecular energy of the molecule is calculated using the fast first energy processing method. For example, the semi-empirical quantum mechanical method is used to calculate the semi-empirical quantum mechanical energy of the molecule to obtain the molecular energy, and the molecular energy obtained by the calculation is denoted as E_xtb. The difference between the two, E_delta=E_dft-E_xtb, is used as the label energy error (Label) of the molecular conformation, and the quantum mechanical calculation related to the second energy processing method is implemented using the Psi4 tool, and the semi-empirical quantum mechanical calculation related to the first energy processing method is implemented using the xTB tool. The enriched dataset is split in a ratio of 8:1:1 and used as the training set, validation set, and test set for DeepQC, respectively.

[0124] The architecture of the DeepQC model is as shown in FIG. 5, and the DeepQC model includes a neural network model related to deep learning and a first energy processing method related to computational chemistry. The neural network model may be an equivariant graph neural network, and the input of the neural network model is the three-dimensional structure coordinates of the molecule, and the output of the neural network model is the E_delta predicted value. In the training process, the overall loss between the E_delta predicted value and the E_delta label value obtained by forward propagation is calculated, and backpropagation is performed on the overall loss to obtain the gradient of each network layer, and the parameters of the neural network model are updated using a self-adaptive moment estimation algorithm. The overall loss of the neural network model can be seen in Equation (9).

[0125]

number

number

number

number

number

number

[0126] The training hyperparameters for the DeepQC model are shown in Table 1. [Table 1] The energy of a molecule can be predicted using a trained DeepQC model. The three-dimensional coordinates of the molecule are input into the DeepQC model, and a first energy E_xtb can be obtained by a first energy calculation process (e.g., a process using the GFN2-xTB program). An energy error E_delta can be predicted using a neural network model (e.g., an equivariant graph neural network), and the final predicted energy value E_dft is obtained by accumulating these two.

[0127] In the research and development process of new drugs and new materials, it is necessary to calculate the energy of molecules. In the calculation methods of related technologies, the quantum mechanical method requires a large amount of calculation and a long calculation time, making it impossible to perform large-scale calculations. In addition, the deep learning method has poor prediction accuracy in long-range actions, and there is a need to improve the accuracy.

[0128] The DeepQC model proposed by the embodiment of the present application aims to maintain high accuracy while maintaining a small amount of calculation. In order to better verify the effect of the DeepQC model, the DeepQC model is verified with a test set, and the calculation accuracy is tested on two data sets, Conformer Benchmark and TorsionNet500, and compared with the computational chemistry method, the semi-empirical quantum mechanics method and the deep learning method. The vertical axis of FIG. 6 shows the correlation between the molecular energy calculated by various methods in Conformer Benchmark and the theoretical value, and the vertical axis of FIG. 7 shows the correlation between the molecular energy calculated by various methods in TorsionNet500 and the theoretical value, where the theoretical value is the molecular energy calculated by the high-precision quantum mechanics method. In terms of algorithm speed, when the DeepQC model proposed by the embodiment of the present application is compared with the high-precision quantum mechanics method, the DeepQC model has a speed increase of several hundred times.

[0129] The following will now describe an exemplary structure of the artificial intelligence-based molecular processing device 455 implemented as a software module according to an embodiment of the present application. In some embodiments, as shown in FIG. 3, the software modules in the artificial intelligence-based molecular processing device 455 stored in the memory 450 include an acquisition module 4551 configured to acquire a three-dimensional structure of a target molecule, and a neural network module 4552 configured to call a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule to obtain an energy error of the target molecule, where the neural network model is trained by fitting the energy error of the sample molecule, and the energy error of the sample molecule is calculated according to two kinds of energy calculation mechanisms. The energy calculation mechanism includes a neural network module 4552, a calculation module 4553 configured to perform the first energy calculation process on the three-dimensional structure of the target molecule to obtain a first energy of the target molecule, and a correction module 4554 configured to perform an error correction process on the first energy of the target molecule based on the energy error of the target molecule, and obtain a second energy of the target molecule. The energy calculation mechanism includes a first energy calculation process and a second energy calculation process, the accuracy of the first energy calculation process being smaller than that of the second energy calculation process, and the speed of the first energy calculation process being faster than that of the second energy calculation process.

[0130] In some embodiments, the neural network module 4552 is further configured to perform a feature extraction process on the three-dimensional structure using the neural network model to obtain an energy error feature of the target molecule, and to perform a full-connection process on the energy error feature using the neural network model to obtain an energy error of the target molecule.

[0131] In some embodiments, the neural network model includes N cascaded feature networks, and the neural network module 4552 further comprises: performing an initial feature extraction process for each atom in the three-dimensional structure to obtain an initial feature for each atom; when the value of n is 1≦n≦N−1, using an n-th feature network of the cascaded N feature networks to perform an n-th feature extraction process on an input of the n-th feature network to obtain an n-th feature for each atom; and transmitting the n-th feature to the n+1-th feature network; when the value of n is N, using the n-th feature network to perform an attribute feature extraction process on the n-1-th feature of each atom. the nth attribute feature of each atom is extracted using a coordinate feature extraction process; the nth attribute feature of each atom is extracted using a coordinate feature extraction process for the n-1th feature of each atom using the nth feature network to obtain an nth coordinate feature of each atom; and the nth attribute feature and the nth coordinate feature of each atom are used to form an energy error feature, where the value range of N satisfies 2≦N, the value of n is an integer that increases from 1, and the value range of n satisfies 1≦n≦N-1, when the value of n is 1, the input of the nth feature network is the initial feature of each atom, and when the value of n is 2≦n≦N, the input of the nth feature network is the n-1th feature of each atom output by the n-1th feature network.

[0132] In some embodiments, the neural network module 4552 is further configured to perform a process of obtaining initial attribute features of each atom in the three-dimensional structure, obtaining initial coordinate features of each atom in the three-dimensional structure, where the initial attribute features represent attribute information of the atom and the initial coordinate features represent position information of the atom, for each atom in the three-dimensional structure, obtaining at least one other atom excluding the atom in the three-dimensional structure, and obtaining initial relationship features between the atom and each other atom, where the initial relationship features represent bonding relationships between the atom and the other atoms, and forming an initial feature of each atom using the initial attribute features of each atom, the initial coordinate features of each atom, and the initial relationship features of each atom.

[0133] In some embodiments, the neural network module 4552 is further configured to use the nth feature network to perform the following processes for each atom, the processes including obtaining other atoms excluding the atom in the three-dimensional structure; performing a first mapping process on the n-1th feature of the atom and the n-1th feature of each other atom to obtain an nth associated feature of the atom corresponding to each other atom; and performing a second mapping process on the n-1th feature of the atom and the nth associated feature of the atom corresponding to each other atom to obtain an nth feature of the atom.

[0134] In some embodiments, the neural network module 4552 is further configured to extract an n-1 coordinate feature of the atom from the n-1 feature of the atom, extract an n-1 coordinate feature of each other atom from the n-1 feature of each other atom, extract an n-1 attribute feature of the atom from the n-1 feature of the atom, extract an n-1 attribute feature of each other atom from the n-1 feature of each other atom, extract an n-1 relationship feature of the atom from the n-1 feature of the atom, and perform the following processes for each other atom, the processes including: extracting the n-1 relationship feature of the atom to the other atom from the n-1 relationship feature of the atom; obtaining a first feature distance between the n-1 coordinate feature of the atom and the n-1 coordinate feature of the other atom; and performing a first fusion process on the square of the first feature distance, the n-1 attribute feature of the atom, the n-1 attribute feature of the other atom, and the n-1 relationship feature of the atom to the other atom to obtain an n association feature of the atom corresponding to the other atom.

[0135] In some embodiments, when the other atoms are multiple, the neural network module 4552 is further configured to: perform an addition operation on the n-th association feature of the atom corresponding to the multiple other atoms to obtain an n-th association feature of the atom; perform a second fusion operation on the n-1th attribute feature and the n-th association feature of the atom to obtain an n-th attribute feature of the atom; obtain a first feature difference between the n-1th coordinate feature of the atom and the n-1th coordinate feature of each other atom; perform a weighted averaging operation on the first feature differences of the multiple other atoms using the n-th association feature of the atom corresponding to each other atom as a weight to obtain a weighted average result corresponding to the atom; perform an addition operation on the weighted average result of the atom and the n-1th coordinate feature of the atom to obtain an n-th coordinate feature of the atom; take the initial relationship feature of the atom as the n-th relationship feature of the atom; and form the n-th feature of the atom using the n-th relationship feature of the atom, the n-th attribute feature of the atom, and the n-th coordinate feature of the atom.

[0136] In some embodiments, the neural network module 4552 is further configured to extract an n-1 coordinate feature of the atom from the n-1 feature of the atom, extract an n-1 coordinate feature of each other atom from the n-1 feature of each other atom, extract an n-1 attribute feature of the atom from the n-1 feature of the atom, extract an n-1 attribute feature of each other atom from the n-1 feature of each other atom, extract an n-1 relationship feature of the atom from the n-1 feature of the atom, and perform the following processing for each other atom, said processing comprising: extracting an n-1 relationship feature of the atom to other atoms from the n-1 relationship feature of the atom; obtaining a second feature distance between the n-1th coordinate feature of the other atom; and performing a first fusion process on the square of the second feature distance, the n-1th attribute feature of the atom, the n-1th attribute feature of the other atoms, and the n-1th relationship feature of the atom to the other atoms to obtain an n-th association feature of the atom corresponding to the other atom, and if there are multiple other atoms, the process includes performing an addition process on the n-th association features of the atoms corresponding to the multiple other atoms to obtain an n-th association feature of the atom; and performing a second fusion process on the n-1th attribute feature and the n-th association feature of the atom to obtain an n-th attribute feature of the atom.

[0137] In some embodiments, before invoking the neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule and obtaining the energy error of the target molecule, the apparatus further comprises a training module 4555 configured to obtain a sample molecule, perform a conformation generation process on the sample molecule to obtain a plurality of sample molecular conformations, obtain a label energy error of each sample molecular conformation, forward propagate each sample molecular conformation in the neural network model to obtain a predicted energy error of each sample molecular conformation, determine a comprehensive loss corresponding to the neural network model according to the label energy error of each sample molecular conformation and the predicted energy error of each sample molecular conformation, perform a backpropagation process on the comprehensive loss in the neural network model, obtain a parameter change value of the neural network model when the comprehensive loss converges, and update the parameters of the neural network model according to the parameter change value.

[0138] In some embodiments, the training module 4555 is further configured to perform the following processes for each sample molecular conformation, the processes including: performing a first energy calculation process for the sample molecular conformation to obtain a first energy of the sample molecular conformation; performing a second energy calculation process for the sample molecular conformation to obtain a second energy of the sample molecular conformation; and obtaining a first difference between the second energy of the sample molecule and the first energy of the sample molecule as the label energy error of the sample molecular conformation.

[0139] In some embodiments, the first energy calculation process is a semi-empirical quantum mechanics based energy calculation process, and the second energy calculation process is a density functional based energy calculation process.

[0140] In some embodiments, the training module 4555 further performs, for each sample molecular conformation, a process of determining a first root mean square error of the sample molecular conformation based on the label energy error of the sample molecular conformation and the predicted energy error of the sample molecular conformation, obtaining other sample molecular conformations other than the sample molecular conformation of the sample molecule, and for each other sample molecular conformation, determining a second difference between the label energy error of the sample molecular conformation and the label energy error of the other sample molecular conformation, and determining a second difference between the predicted energy error of the sample molecular conformation and the predicted energy error of the other sample molecular conformation. The neural network model is configured to perform a process of determining a third difference between the measured energy error and the measured energy error, perform a root mean square process on the second difference and the third difference to obtain a second root mean square error of the sample molecular conformation corresponding to the other sample molecular conformation, perform an addition process on the second root mean square errors of the sample molecular conformation corresponding to the plurality of other sample molecular conformations to obtain a third root mean square error of the sample molecular conformation, and perform a third fusion process on the first root mean square errors of the plurality of sample molecular conformations and the third root mean square errors of the plurality of sample molecules to obtain a comprehensive loss corresponding to the neural network model.

[0141] An embodiment of the present application provides a computer program product including a computer program or computer executable instructions stored in a computer readable storage medium, wherein a processor of an electronic device reads the computer executable instructions from the computer readable storage medium, and the processor executes the computer executable instructions to cause the electronic device to perform the artificial intelligence based molecular processing method of the embodiment of the present application.

[0142] An embodiment of the present application provides a computer-readable storage medium having stored thereon computer-executable instructions which, when executed by a processor, cause the processor to perform an artificial intelligence based molecular processing method according to an embodiment of the present application, for example, the artificial intelligence based molecular processing method shown in Figures 4A to 4C.

[0143] In some embodiments, the computer readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM, or may be any of a variety of devices including one or any combination of the above memories.

[0144] In some embodiments, the computer executable instructions may take the form of a program, software, software module, script or code, and may be written in any form of programming language (including compiled or interpreted, or declarative or procedural languages), and arranged in any form, such as as a separate program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0145] As an example, the computer-executable instructions may, but do not necessarily, correspond to a file in a file system, and may be stored in a portion of a file that stores other programs or data, such as in one or more scripts within a HyperText Markup Language (HTML) document, in a single file of the discussed program, or in multiple joint files (e.g., files storing one or more modules, subprograms, or code portions).

[0146] As an example, the computer-executable instructions may be arranged to be executed on one electronic device, or on multiple electronic devices at the same location, or on multiple electronic devices distributed at multiple locations and interconnected by a communications network.

[0147] In summary, the first energy of the target molecule is obtained by performing a first energy calculation process on the three-dimensional structure of the target molecule in the embodiment of the present application. Since the speed of the first energy calculation process is faster than the second energy calculation process, the speed of the energy calculation process is increased, and an energy error prediction process is performed on the three-dimensional structure of the target molecule using a neural network model to obtain the energy error of the target molecule, and an error correction process is performed on the first energy calculated based on the energy error to obtain the second energy of the target molecule. Since the energy error can represent the difference in the calculation result between the highly accurate second energy calculation process and the less accurate first energy calculation process, the accuracy of the second energy can be improved by correcting the first energy calculated using the energy error predicted by deep learning.

[0148] The above description is merely an embodiment of the present application, and does not limit the scope of protection of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and scope of the present application shall be included in the scope of protection of the present application.

Claims

1. 1. An electronic device implemented artificial intelligence based molecular processing method comprising: obtaining a three-dimensional structure of a target molecule; calling a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule to obtain an energy error of the target molecule, the neural network model being trained by fitting the energy error of a sample molecule, the energy error of the sample molecule referring to the difference between the calculation results obtained when calculating the energy of the sample molecule according to two kinds of energy calculation mechanisms, the two kinds of energy calculation mechanisms including a first energy calculation process and a second energy calculation process, the accuracy of the first energy calculation process being lower than the accuracy of the second energy calculation process, and the speed of the first energy calculation process being faster than the speed of the second energy calculation process; performing the first energy calculation process on the three-dimensional structure of the target molecule to obtain a first energy of the target molecule; and performing an error correction process on the first energy of the target molecule based on the energy error of the target molecule to obtain a second energy of the target molecule.

2. The step of calling the neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule and obtaining an energy error of the target molecule includes: performing a feature extraction process on the three-dimensional structure using the neural network model to obtain an energy error feature of the target molecule; and performing a fully coupled process on the energy error features using the neural network model to obtain an energy error of the target molecule.

3. The neural network model includes N feature networks connected in cascade, and the step of performing a feature extraction process on the three-dimensional structure using the neural network model to obtain an energy error feature of the target molecule includes: performing an initial feature extraction process for each atom in the three-dimensional structure to obtain an initial feature for each atom; if the value of n is 1≦n≦N−1, using an n-th feature network among the N cascaded feature networks to perform an n-th feature extraction process on the input of the n-th feature network to obtain an n-th feature of each of the atoms, and transmitting the n-th feature to the n+1-th feature network; where n is a value of N, performing an attribute feature extraction process on the (n-1)th feature of each of the atoms using the (n)th feature network to obtain an (n)th attribute feature of each of the atoms, performing a coordinate feature extraction process on the (n-1)th feature of each of the atoms using the (n)th feature network to obtain an (n)th coordinate feature of each of the atoms, and forming the energy error feature using the (n)th attribute feature and the (n)th coordinate feature of each of the atoms; 3. The molecular processing method of claim 2, wherein the value range of N satisfies 2≦N, the value of n is an integer that increases from 1, and the value range of n satisfies 1≦n≦N-1, where when the value of n is 1, the input of the n feature network is the initial feature of each atom, and when the value of n is 2≦n≦N, the input of the n feature network is the n-1th feature of each atom output by the n-1th feature network.

4. The step of performing an initial feature extraction process for each atom in the three-dimensional structure to obtain an initial feature of each atom includes: obtaining initial attribute features for each atom in the three-dimensional structure, and obtaining initial coordinate features for each atom in the three-dimensional structure, the initial attribute features representing attribute information of the atoms, and the initial coordinate features representing position information of the atoms; performing a process for obtaining, for each of the atoms in the three-dimensional structure, at least one other atom other than the atom in the three-dimensional structure, and obtaining an initial relationship feature between the atom and each of the other atoms, the initial relationship feature representing a bonding relationship between the atom and the other atoms; forming an initial feature for each of said atoms using initial attribute features for each of said atoms, initial coordinate features for each of said atoms, and initial relationship features for each of said atoms.

5. performing an n-th feature extraction process on an input of the n-th feature network using an n-th feature network of the cascaded N feature networks to obtain an n-th feature of each of the atoms, performing a process for each of the atoms using the n feature network, the process comprising: obtaining other atoms in the three-dimensional structure excluding the atom; performing a first mapping operation on the n-1 feature of the atom and the n-1 feature of each of the other atoms to obtain an n associated feature of the atom corresponding to each of the other atoms; and performing a second mapping operation on the n-1th feature of the atom and the nth associated features of the atom corresponding to each of the other atoms to obtain an nth feature of the atom.

6. performing a first mapping process on the n-1 feature of the atom and the n-1 feature of each of the other atoms to obtain an n-th associated feature of the atom corresponding to each of the other atoms; extracting from the n-1 feature of the atom a n-1 coordinate feature of the atom, a n-1 attribute feature of the atom, and a n-1 relationship feature of the atom; extracting from the n-1 feature of each of the other atoms a n-1 coordinate feature of each of the other atoms and a n-1 attribute feature of each of the other atoms; and for each said other atom, performing a process comprising: extracting from the n-1-th relation feature of the atom an n-1-th relation feature of the atom with respect to the other atom; obtaining a first feature distance between the n-1 th coordinate feature of the atom and the n-1 th coordinate feature of the other atom; and performing a first fusion process on the square of the first feature distance, the n-1th attribute feature of the atom, the n-1th attribute feature of the other atoms, and the n-1th relationship feature of the atom to the other atoms to obtain an n-th association feature of the atom corresponding to the other atoms.

7. performing a second mapping process on the n-1 feature of the atom and the n-th associated feature of the atom corresponding to each of the other atoms to obtain an n-th feature of the atom; performing an addition process on n-th associated features of the atom corresponding to a plurality of the other atoms to obtain an n-th associated feature of the atom; performing a second fusion process on the (n-1)th attribute feature of the atom and the nth association feature of the atom to obtain an nth attribute feature of the atom; obtaining a first feature difference between the n-1 th coordinate feature of the atom and the n-1 th coordinate feature of each of the other atoms; performing a linear mapping process on the n-th associated feature of the atom corresponding to each of the other atoms to obtain a weight for each of the other atoms; performing a weighted average process on the first characteristic differences of the other atoms based on the weight of each of the other atoms to obtain a weighted average result corresponding to the atom; performing an addition process on the weighted average result of the atom and the (n-1)-th coordinate feature of the atom to obtain the (n)-th coordinate feature of the atom; and forming the nth feature of the atom using the nth relationship feature of the atom, the nth attribute feature of the atom, and the nth coordinate feature of the atom.

8. The step of performing an attribute feature extraction process on the (n-1)th feature of each of the atoms using the nth feature network to obtain an nth attribute feature of each of the atoms includes: extracting from each of the n-1 features of the atom a n-1 coordinate feature of the atom, a n-1 attribute feature of the atom, and a n-1 relationship feature of the atom; For each of the atoms, obtaining other atoms in the three-dimensional structure other than the atom, and performing a process for each of the other atoms, the process comprising: extracting an n-1-th relation feature of the atom with respect to the other atom from the n-1-th relation feature of the atom, and performing a process of obtaining a second feature distance between the n-1-th coordinate feature of the atom and the n-1-th coordinate feature of the other atom; performing a first fusion process on the square of the second feature distance, the n-1 attribute feature of the atom, the n-1 attribute feature of the other atom, and the n-1 relationship feature of the atom to the other atom to obtain an n association feature of the atom corresponding to the other atom; performing an addition process on n-th associated features of the atom corresponding to a plurality of the other atoms to obtain an n-th associated feature of the atom; and performing a second fusion process on the (n-1)th attribute feature of the atom and the nth association feature of the atom to obtain the nth attribute feature of the atom.

9. The molecular processing method includes: calling a neural network model to perform an energy error prediction process on the three-dimensional structure of the target molecule; and before obtaining the energy error of the target molecule, the molecular processing method includes: obtaining a sample molecule, and performing a three-dimensional structure generation process on the sample molecule to obtain a plurality of sample molecule three-dimensional structures; obtaining a label energy error for each of the sample molecular conformations; forward propagating each of the sample molecular conformations through the initialized neural network model to obtain a predicted energy error for each of the sample molecular conformations; determining an overall loss based on the label energy error of each of the sample molecular conformations and the predicted energy error of each of the sample molecular conformations; 9. The molecular processing method according to claim 1, further comprising the steps of: performing a backpropagation process on the overall loss in the initialized neural network model, obtaining a parameter change value of the initialized neural network model when the overall loss converges, and updating parameters of the initialized neural network model based on the parameter change value.

10. The step of obtaining the label energy error of each of the sample molecular conformations includes: For each of the sample molecular conformations, a process is performed, the process comprising: performing a first energy calculation process on the sample molecular three-dimensional structure to obtain a first energy of the sample molecular three-dimensional structure; performing a second energy calculation process on the sample molecular three-dimensional structure to obtain a second energy of the sample molecular three-dimensional structure; The molecular processing method of claim 9, further comprising: obtaining a first difference between the second energy of the sample molecular conformation and the first energy of the sample molecular conformation as a label energy error of the sample molecular conformation.

11. Determining an overall loss based on the label energy error of each of the sample molecular conformations and the predicted energy error of each of the sample molecular conformations, For each of the sample molecular conformations, performing a process of determining a first root mean square error of the sample molecular conformation based on a label energy error of the sample molecular conformation and a predicted energy error of the sample molecular conformation; acquiring another sample molecule conformation other than the sample molecule conformation; performing, for each of the other sample molecular conformations, a process of determining a second difference between the label energy error of the sample molecular conformation and the label energy error of the other sample molecular conformation, and a process of determining a third difference between the predicted energy error of the sample molecular conformation and the predicted energy error of the other sample molecular conformation; performing a root mean square process on the second difference and the third difference to obtain a second root mean square error of the sample molecular conformation corresponding to the other sample molecular conformation; performing an addition process on the second root mean square errors of the sample molecular 3D structures corresponding to the plurality of other sample molecular 3D structures to obtain a third root mean square error of the sample molecular 3D structure; and performing a third fusion process on the first root mean square errors of the plurality of sample molecular conformations and the third root mean square errors of the plurality of sample molecular conformations to obtain a comprehensive loss corresponding to the neural network model.

12. 1. An artificial intelligence based molecular processing system, comprising: an acquisition module configured to acquire a three-dimensional structure of the target molecule; A neural network module configured to call a neural network model to perform an energy error prediction process on a three-dimensional structure of the target molecule to obtain an energy error of the target molecule, a neural network module, the neural network model being trained by fitting an energy error of a sample molecule, the energy error of the sample molecule referring to a difference between calculation results obtained when calculating the energy of the sample molecule according to two kinds of energy calculation mechanisms, the two kinds of energy calculation mechanisms including a first energy calculation process and a second energy calculation process, the accuracy of the first energy calculation process being lower than the accuracy of the second energy calculation process, and the speed of the first energy calculation process being faster than the speed of the second energy calculation process; a calculation module configured to perform the first energy calculation process on a three-dimensional structure of the target molecule to obtain a first energy of the target molecule; and a correction module configured to perform an error correction process on a first energy of the target molecule based on an energy error of the target molecule to obtain a second energy of the target molecule.

13. a memory storing computer executable instructions; and a processor for executing computer executable instructions stored in the memory to implement the artificial intelligence based molecular processing method of any one of claims 1 to 8.

14. A computer program product enabling a computer to carry out the artificial intelligence based molecular processing method according to any one of claims 1 to 8.

Images