An electron density-based method and system for molecular modeling, property prediction

Abstract

The application discloses a kind of molecular modeling method and modeling system based on electron density, wherein the method comprises: obtaining atomic coordinate information in target molecule;Based on atomic coordinate information, electron density grid point information of target molecule is obtained at preset resolution;From electron density grid point information, target electron density value greater than preset electron density value is extracted, and 3-dimensional matrix formed by target electron density value is constructed as the molecular model of target molecule.The technical scheme provided by the application solves the problem that the features contained in the traditional molecular representation model are insufficient.

Description

Technical Field

[0001] This invention relates to the field of structural biology analysis technology, specifically to a molecular modeling and property prediction method and system based on electron density. Background Technology

[0002] A molecule is a whole composed of atoms bonded together in a specific order and spatial arrangement; this order and spatial arrangement is called molecular structure. Due to the interactions between atoms within a molecule, its physical and chemical properties depend not only on the types and numbers of constituent atoms but also, and perhaps more importantly, on its structure. In fields such as drug development, designing drugs and other chemicals based on molecular structure is crucial. Currently, building deep learning models based on molecular structure is a hot research topic. By modeling molecules and then training them with deep learning, the efficiency of drug development can be improved. However, the most common method for 3D molecular modeling in current technology is molecular conformation, which uses different elements in the molecule to form a 3D matrix. This modeling method results in overly sparse information in the molecular space, leading to poor feature extraction from training samples for deep learning models and consequently, poor model prediction performance. Therefore, how to build molecular models that can represent more complete molecular features is a pressing problem that needs to be solved. Summary of the Invention

[0003] In view of this, the present invention provides a molecular modeling method based on electron density, thereby obtaining a molecular model with more complete features.

[0004] According to the first aspect, a molecular modeling method based on electron density, the method comprising:

[0005] Obtain atomic coordinate information in the target molecule;

[0006] Based on the atomic coordinate information, the electron density grid information of the target molecule is obtained at a preset resolution;

[0007] The target electron density value greater than the preset electron density value is extracted from the electron density grid information, and the 3D matrix composed of the target electron density values ​​is used to construct the molecular model of the target molecule.

[0008] Optionally, obtaining the electron density grid information of the target molecule at a preset resolution based on the atomic coordinate information includes:

[0009] The atomic coordinate information is transformed from real space to frequency domain space to generate structure factors;

[0010] The structure factor is adjusted to return to real space at a preset resolution to generate the electron density grid information. The preset resolution is a preset range of summation of frequency domain spatial vector components.

[0011] Optionally, the atomic coordinate information is transformed from real space to frequency domain space based on Fourier transform to generate the structure factor. The calculation formula for the transformation process is as follows:

[0012]

[0013] Where r represents atomic coordinate information, x, y, and z represent the three components of atomic coordinate information, and f n Let f(s) represent the diffraction factor of n different elements, s represent the frequency domain space vector, h, k, and l represent the three components of the frequency domain space vector respectively, F(s) represent the structure factor of the s vector, and 2πi is an imaginary number.

[0014] Optionally, the structure factor is adjusted to a preset resolution based on the inverse Fourier transform and returned to the real space to generate the electron density grid information. The calculation formula for the return process is as follows:

[0015]

[0016] Where r represents atomic coordinate information, s represents frequency domain space vector, h, k, and l represent the three components of frequency domain space vector, F(s) is the structure factor of s vector, ρ(r) represents the electron density lattice information, Vcell represents the unit cell volume, 2πi is the imaginary number, and the preset resolution is obtained by summing the preset range of h, k, and l.

[0017] Optionally, the preset resolution includes multiple different sub-resolutions, and the method further includes:

[0018] The process sequentially uses each of the sub-resolutions to perform the following steps: calculating the electron density grid information of the target molecule based on the atomic coordinate information at a preset resolution; extracting the target electron density value greater than the preset electron density value from the electron density grid information; and constructing a molecular model of the target molecule from a 3D matrix composed of the target electron density values, thereby generating multiple 3D matrices.

[0019] The multiple 3D matrices are superimposed to form a 4D matrix, generating a multi-channel model.

[0020] Optionally, the method further includes:

[0021] The electron density values ​​in the multi-channel model are normalized to obtain a normalized model.

[0022] Optionally, the method further includes:

[0023] Using the center of the normalized model as the rotation center, the normalized matrix is ​​rotated in random directions and at random angles to obtain the rotated molecular model.

[0024] According to the second aspect, a method for establishing a molecular property prediction model includes:

[0025] Obtain molecular samples and generate molecular models of the molecular samples based on the molecular samples using the method described in the first aspect;

[0026] The molecular properties corresponding to the molecular samples are obtained respectively, and the molecular properties and their corresponding molecular models are paired and stored in the training sample set;

[0027] The machine learning model is trained using the training sample set to obtain a property prediction model, which is used to predict molecular properties.

[0028] According to a third aspect, a method for predicting molecular properties, the method comprising:

[0029] Obtain the molecule to be predicted, and generate a molecular model of the molecule to be predicted using the method described in the first aspect;

[0030] The molecular model is input into a property prediction model obtained using the second aspect of the molecular property prediction model establishment method, and the molecular properties of the molecule to be predicted are predicted by the property prediction model.

[0031] According to the fourth aspect, a molecular modeling system based on electron density, the system comprising:

[0032] The information acquisition module obtains the atomic coordinate information of the target molecule;

[0033] The electron density generation module acquires the electron density grid information of the target molecule at a preset resolution based on the atomic coordinate information.

[0034] The molecular model building module extracts target electron density values ​​that are greater than a preset electron density value from the electron density grid information, and constructs a molecular model of the target molecule from a 3D matrix composed of the target electron density values.

[0035] According to the fifth aspect, an electronic device includes:

[0036] A memory and a processor are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the methods described in the first aspect, the second aspect, and the third aspect, or any optional embodiment of the first aspect.

[0037] According to a sixth aspect, embodiments of the present invention provide a computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions for causing the computer to perform the methods described in the first aspect, the second aspect, and the third aspect, or any optional embodiment of the first aspect.

[0038] The technical solution of this invention has the following advantages:

[0039] This invention provides a molecular modeling method and system based on electron density. The modeling method specifically includes: first, under a preset resolution of electron density, calculating the electron density values ​​of each atom's coordinate position and its neighboring coordinate positions based on the atomic coordinates in the target molecule; then, utilizing the characteristics of atomic structure (higher electron density intensity near the nucleus and lower electron density intensity farther from the nucleus), extracting the target electron density values ​​above the preset density value to form a 3D matrix, and using this 3D matrix as a molecular model to represent the target molecule. Since electron density itself comprehensively includes information on element type and chemical properties, the molecular model provides a more complete description of the target molecule's characteristics. Furthermore, compared to molecular models composed solely of different element types, it saves significant computational power in subsequent calculations such as machine learning, improving model efficiency. Attached Figure Description

[0040] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a schematic diagram illustrating the steps of a molecular modeling method based on electron density according to an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of a molecular model structure based on molecular conformation, according to an embodiment of the present invention.

[0043] Figure 3 This is a schematic diagram of a molecular model structure based on electron density in a molecular modeling method based on electron density according to an embodiment of the present invention;

[0044] Figure 4 This is a schematic diagram of the molecular model structure based on electron density at different resolutions in a molecular modeling method based on electron density according to an embodiment of the present invention.

[0045] Figure 5This is a schematic diagram illustrating the steps of a method for establishing a molecular property prediction model according to an embodiment of the present invention;

[0046] Figure 6 This is a schematic diagram illustrating the steps of a molecular property prediction method according to an embodiment of the present invention;

[0047] Figure 7 This is a schematic diagram of the structure of a molecular modeling system based on electron density according to an embodiment of the present invention;

[0048] Figure 8 This is a schematic diagram of the structure of a molecular property prediction system according to an embodiment of the present invention;

[0049] Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention; Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] The technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0052] Please see Figure 1 The present invention provides a molecular modeling method based on electron density, which specifically includes the following steps:

[0053] Step S1: Obtain the atomic coordinate information in the target molecule.

[0054] Step S2: Using atomic coordinate information, obtain the electronic density grid information of the target molecule at a preset resolution;

[0055] Step S3: Extract the target electron density value that is greater than the preset electron density value from the electron density grid information, and construct the 3D matrix composed of the target electron density values ​​as the molecular model of the target molecule.

[0056] Specifically, the most intuitive and traditional way to represent molecules in 3D using existing technologies is through molecular conformation. However, in applications such as deep learning, conformation leads to excessively sparse information about molecules in space. Therefore, in terms of shape, the 3D representation of molecules mainly takes the following forms: expansion using kernel functions such as Gaussian expansion, and expansion based on van der Waals force fields. Furthermore, to distinguish different atoms and chemical properties, for example, in the input of deep learning models, multiple channels are used to represent different element types, or multiple channels are used to represent chemical properties (hydrogen bond acceptors, hydrogen bond donors, aromatic systems, etc.). However, using multiple channels to represent the same molecule increases the sparsity of the 3D representation, making it difficult for deep learning models to learn the correct features and consuming computational resources exponentially. Multi-channel representation of elements in the molecule is used to distinguish features of different dimensions, employing multiple matrices to represent various features. In this example, when representing molecules, to distinguish different element types, each element is stored separately in a channel (matrix), and then all channels are superimposed to obtain a new matrix with an added dimension. For example, all carbon atoms are represented by an x×y×z spatial matrix, all nitrogen atoms are represented by an x×y×z spatial matrix, and so on. All our molecules are composed of 10 elements. Therefore, each molecule will ultimately be represented by a 10×x×y×z matrix.

[0057] like Figure 2 The diagram shown is a schematic representation of a single-channel Gaussian logic cloud that does not distinguish element types (one perspective of a 3D graph). It's evident that the molecular representation occupies less than a quarter of the entire graph. If the graph were further divided into 10 channels, the average effective information per channel would not exceed 0.025% of the total graph. Even with a large number of multi-channel samples on top of the 3D representation, such sparse information is difficult for mathematical tools like neural networks to capture and learn during application.

[0058] The electron density-based representation method proposed in this invention represents the probability of an electron appearing at a specific position within a molecule. The calculation of electron density not only comprehensively considers the element types and chemical properties of the molecule, but also allows for control over its sparsity within a small range. This results in more accurate molecule identification, reduces the need for an excessively large number of channels, and saves on hardware computing costs.

[0059] Step S1 involves obtaining a predetermined number of atomic coordinates in the target molecule. Specifically, to obtain the electron density of the molecule, it is first necessary to know the atomic coordinates of all the atoms in the molecule. In the field of chemistry, obtaining the atomic coordinates of a molecule through actual experiments or methods such as FindIt software is easy, and the specific steps are existing technology, which will not be elaborated here.

[0060] Step S2: Using atomic coordinate information, obtain the electronic density grid information of the target molecule at a preset resolution.

[0061] Specifically, in molecular representation, the electron density is a continuous function. To store the electron density of this molecule in a computer matrix, it needs to be discretized. This allows for the acquisition of electron density lattice information based on the number of atomic coordinates, which is then stored in a 3D matrix. Methods for obtaining electron density lattice information include, but are not limited to: experimental electron density methods based on X-ray crystallography, experimental electron density methods based on electron microscopy, and calculated electron density methods based on quantum chemistry. Traditional X-ray crystallography can analyze the fine three-dimensional structure of macromolecules at the atomic or near-atomic level, thereby obtaining information such as the macromolecular structure experimentally and then calculating the electron density lattice information. Experimental electron densities obtained through electron microscopy are directly saved as spatial lattice files and can be directly read. In the absence of experimental conditions, electron density can be calculated based on quantum chemistry using readily available atomic spatial coordinates in the target molecule structure. This method can divide the target molecule structure into multiple smaller parts, i.e., multiple small molecule systems, and calculate the electron density value of each lattice point within each small molecule system.

[0062] Specifically, in one embodiment, step S2 includes the following steps:

[0063] Step S201: Transform the atomic coordinate information from real space to frequency domain space to generate structure factors.

[0064] Step S202: Adjust the structure factor with a preset resolution to return to the real space to generate electron density grid information. The preset resolution is the summation range of the preset frequency domain spatial vector components.

[0065] Specifically, in the methods for obtaining electron density described above, on the one hand, due to limitations, not all systems can have their electron density measured experimentally; on the other hand, quantum chemical calculations are limited by the scale of the target system, often requiring extensive simplification of the system before calculating the electron density of macromolecular-small molecule complex systems or macromolecular systems. This simplification often leads to deviations in the calculation results. Based on this, this invention proposes a preferred solution: obtaining the electron density lattice information of the target molecular structure based on crystallographic theory. First, the atomic coordinate information is transformed from real space (solid space) to frequency domain space. The diffraction of the lattice with electromagnetic waves yields the pattern in the frequency domain. The frequency domain not only preserves the symmetry of the lattice and information such as molecular structure and physicochemical properties from real space, but also decomposes the "detail level" of the real-space molecule according to different "frequency" values. That is, high-frequency information in the frequency domain reflects the details of the molecule in real space, while low-frequency information reflects the coarse outline of the molecule. After filtering the information in the frequency domain according to frequency and transforming it back to real space, molecular representations with "different levels of detail" are obtained. Specifically, the molecular coordinate information in real space is subjected to Fourier transform to obtain the result after frequency domain space transformation (referred to as structure factor in this embodiment of the invention). Then, the frequency domain space vector is subjected to inverse Fourier transform within a preset frequency range to return to real space. The value obtained by transforming the molecular coordinate information in real space to frequency domain space and then back to real space is the electron density grid information.

[0066] Specifically, in one embodiment, regarding step S101 above, the atomic coordinate information of the target molecular structure is first obtained. This atomic coordinate information includes the atomic coordinates of the atoms in the target molecular structure. The structure factor (including diffraction amplitude and phase) is calculated based on the atomic coordinate information. The formula for calculating the structure factor by converting the atomic coordinate information from real space to frequency domain space based on Fourier transform is as follows:

[0067]

[0068] Where r represents atomic coordinate information (very easy to obtain), x, y, and z represent the three components of atomic coordinate information, and f n Let represent the diffraction factors of n different elements (which can be found in the International Crystallographic Tables), s represent the frequency domain space vector, h, k, and l represent the three components of the frequency domain space vector, F(s) represent the structure factor of the s vector, and 2πi is an imaginary number. Furthermore, by introducing descriptions of the solvent and atomic vibrations over time, the above formula can be improved, resulting in a more accurate conversion. The improved formula is as follows:

[0069]

[0070] Where f(bulk solvent) is the solvent shell description function, bulk solvent is called the solvent shell, f(b) is the temperature factor description function, b is called the temperature factor, the temperature factor is used to measure the degree of atomic thermal motion in the crystal, and the solvent shell describes the contribution of the solvent to diffraction.

[0071] Specifically, in one embodiment, for step S102, the structure factor is adjusted to return to the real space with a preset resolution to generate electron density grid information. The inverse transformation from frequency domain space to real space is performed based on the inverse Fourier transform, and the calculation formula is as follows:

[0072]

[0073] Where r represents atomic coordinates, s(h,k,l) ​​represents the reciprocal lattice vector, F(s) is the diffraction amplitude of the s vector, ρ(r) represents the electron density lattice information, and Vcell is the unit cell volume. Electron density lattice information at a preset resolution is generated using structure factors. The preset resolution is obtained by summing the preset reciprocal lattice vectors; that is, by changing the summing range of h, k, and l, electron densities at different resolutions can be obtained. Higher resolution yields clearer electron density maps with more accurate representation of interatomic bonds; lower resolution results in blurrier electron density maps, but a more accurate representation of the overall molecular framework. The method of calculating electron density based on crystallographic theory not only considers the macroscopic connections between atoms in macromolecular systems but also obtains electron density maps with varying clarity by adjusting different preset resolutions, thus enabling the generation of electron density maps under multiple resolution conditions. After obtaining the electron density map at a preset resolution, the target molecular structure can be divided into a preset number of grid point coordinates. Correspondingly, the electron density map is also discretely represented from continuous variables according to the number of grid points, thus obtaining the electron density grid point information.

[0074] Step S3: Extract target electron density values ​​that are greater than the preset electron density value from the electron density grid information, and construct a molecular model of the target molecule from the 3D matrix composed of the target electron density values.

[0075] Specifically, such as Figure 3 As shown, since the electron density value is relatively high at the atomic nucleus and gradually decreases away from the nucleus, it is similar to the contour lines on a map (the highest point is at the top of the mountain, and the lowest point is at the foot of the mountain). By setting a preset electron density value as the boundary of the contour line, and extracting the target electron density value that is higher than the contour line, the target electron density value is stored in the 3D matrix. This makes the outline and details of the molecular model represented by the electron density map clearer and includes most of the features of the molecule. Finally, the extracted 3D matrix is ​​used as the molecular model of the target molecule.

[0076] Specifically, in one embodiment, the molecular modeling method provided by the present invention further includes the following steps:

[0077] Step S4: The preset resolution includes multiple different sub-resolutions. Steps S1 to S3 are performed sequentially using each sub-resolution to obtain multiple 3D matrices. These multiple 3D matrices are then superimposed to form a 4D matrix, generating a multi-channel model.

[0078] Specifically, to further enhance the feature set of the molecular model, a multi-channel representation method is employed. Unlike traditional molecular conformation techniques, which use multiple channels based on electron density, this method uses different resolutions and has a significantly fewer channel count than conformation-based representations. Combining the advantage of low sparsity in electron density-based molecular models, a smaller number of channels can encompass the vast majority of molecular features. For example... Figure 4 As shown, the high-resolution molecular models have clearer key bond information, while the low-resolution molecular models have clearer molecular skeletons. Therefore, molecular models obtained under multiple sub-resolution conditions are stacked in space to obtain a 4D matrix containing more molecular features, resulting in more accurate modeling.

[0079] Specifically, in one embodiment, the molecular modeling method provided by the present invention further includes the following steps:

[0080] Step S5: Normalize the electron density values ​​in the multi-channel model to obtain a normalized model. Specifically, the default electron density values ​​(i.e., electron density values) are distributed within the range of 1 to 100, which is a large range. This molecular model is not conducive to mathematical calculations. To facilitate the application of mathematical tools such as neural networks, the values ​​in this 4-dimensional multi-channel model are adjusted to the 0-1 range to obtain the normalized model.

[0081] Specifically, in one embodiment, the molecular modeling method provided by the present invention further includes the following steps:

[0082] Step S6: Using the center of the normalized model as the rotation center, rotate the normalized matrix in random directions and at random angles to obtain the rotated molecular model. Specifically, in space, the molecular model has rotation and translation invariance, meaning that rotation in any direction and angle, and changes in position, do not alter the original molecular properties. Random rotation provides a variety of molecules, which can improve the generalization ability of deep learning models and alleviate overfitting in the field of deep learning, thus further improving the applicability of the molecular model provided in this embodiment of the invention.

[0083] By executing the above steps, the present invention provides a molecular modeling method based on electron density. First, under a preset resolution of electron density, the electron density values ​​of each atom's coordinate position and its neighboring coordinate positions are calculated based on the atomic coordinates in the target molecule. Then, utilizing the characteristics of atomic structure (higher electron density intensity at positions closer to the nucleus and lower electron density intensity at positions farther from the nucleus), the target electron density values ​​above the preset density value are extracted to form a 3D matrix, which is then used as a molecular model to represent the target molecule. To improve the applicability of this molecular model, it is then upgraded to a 4D matrix based on different resolutions, and normalized and rotated, thereby increasing the diversity of the molecular model and further enhancing its applicability. Generalization ability is improved because electron density itself comprehensively includes information on element type and chemical properties, making the molecular model's description of the target molecule more complete. Compared to molecular models composed solely of different element types, it also saves significant computational power in subsequent calculations such as machine learning, improving model efficiency.

[0084] like Figure 5 As shown in the figure, this embodiment of the invention also provides a method for establishing a molecular property prediction model, which specifically includes the following steps:

[0085] Step S7: Obtain molecular samples and generate molecular models of the molecular samples using the methods in steps S1 to S6. Specifically, a large number of molecular models are obtained through steps S1 to S6. These molecular models are then bundled with corresponding molecular properties to obtain training samples, which can be used to train deep learning models and achieve the function of property analysis and prediction.

[0086] Step S8: Obtain the molecular properties corresponding to each molecular sample, and store the molecular properties and their corresponding molecular models in the training sample set. Specifically, in this embodiment of the invention, the molecular properties used include, but are not limited to: the average kinetic energy, mass, volume, elements contained in the molecule, and state (excited or unexcited). Storing these molecular properties as labels and their corresponding molecular models in the training sample set completes the construction of the training samples.

[0087] Step S9: Train the machine learning model using the training sample set to obtain a property prediction model, which is used to predict molecular properties. Specifically, train the deep learning model using the training samples obtained in step S8 to train a prediction model capable of predicting molecular properties. This invention uses a convolutional neural network to establish the property prediction model; its specific training process is existing technology and will not be described in detail here. Because the molecular model established in steps S1 to S6 contains very complete features, the accuracy of the prediction results of this property prediction model is greatly improved, enabling its widespread application in the field of medicinal chemistry.

[0088] like Figure 6 As shown in the figure, this invention also provides a method for predicting molecular properties, which specifically includes the following steps:

[0089] Step S10: Obtain the molecule to be predicted and generate a molecular model of the molecule to be predicted using the methods in steps S1 to S6;

[0090] Step S11: Input the molecular model into the property prediction model obtained in step S9, and generate the molecular properties corresponding to the molecular model through the property prediction model.

[0091] Specifically, after the molecule to be predicted is converted into a molecular model, the property prediction model obtained in step S9 can be input to predict the molecular properties, thereby obtaining properties such as the average kinetic energy, mass, volume, contained elements, and state of the predicted molecule. Experimental researchers can then select the required properties according to the actual situation. Using the property prediction model provided in this embodiment of the invention further reduces the workload of manual experiments and calculations, and improves the efficiency of work in fields such as drug development.

[0092] like Figure 7 As shown, this embodiment also provides a molecular modeling system based on electron density, which includes:

[0093] The information acquisition module 101 acquires the coordinate information of a preset number of atoms in the target molecule. For details, please refer to the relevant description of step S1 in the above method embodiment, which will not be repeated here.

[0094] The electron density generation module 102 calculates the electron density grid information of the target molecule using atomic coordinate information at a preset resolution. For details, please refer to the relevant description of step S2 in the above method embodiments, which will not be repeated here.

[0095] The molecular model building module 103 extracts target electron density values ​​that are greater than a preset electron density value from the electron density grid information, and uses the 3D matrix composed of the target electron density values ​​as the molecular model. For details, please refer to the relevant description of step S3 in the above method embodiment, which will not be repeated here.

[0096] The electron density-based molecular modeling system provided in this embodiment of the invention is used to execute the electron density-based molecular modeling method provided in the above embodiment. Its implementation method and principle are the same. For details, please refer to the relevant descriptions in the above method embodiments, which will not be repeated here.

[0097] like Figure 8 As shown, this embodiment also provides a molecular property prediction system, which includes:

[0098] The information acquisition module 104 acquires the molecule to be predicted and generates a molecular model of the molecule to be predicted using the methods in steps S1 to S6. For details, please refer to the relevant description of step S10 in the above method embodiments, which will not be repeated here.

[0099] The property prediction module 105 inputs the molecular model into the property prediction model obtained in step S9, and generates the molecular properties corresponding to the molecular model through the property prediction model. For details, please refer to the relevant description of step S11 in the above method embodiments, which will not be repeated here.

[0100] The present invention provides a molecular property prediction system for executing the molecular property prediction method provided in the above embodiments. Its implementation method and principle are the same. For details, please refer to the relevant descriptions of the above method embodiments, which will not be repeated here.

[0101] Figure 9 An electronic device according to an embodiment of the present invention is shown. The device includes a processor 901 and a memory 902, which can be connected via a bus or other means. Figure 9 Taking the example of a connection between China and Israel via a bus.

[0102] Processor 901 can be a Central Processing Unit (CPU). Processor 901 can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0103] The memory 902, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the above method embodiments. The processor 901 executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory 902, thereby implementing the methods in the above method embodiments.

[0104] The memory 902 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor 901, etc. Furthermore, the memory 902 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 902 may optionally include memory remotely located relative to the processor 901, and these remote memories may be connected to the processor 901 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0105] One or more modules are stored in memory 902, and when executed by processor 901, they perform the methods described in the above method embodiments.

[0106] The specific details of the aforementioned electronic device can be understood by referring to the relevant descriptions and effects in the above method embodiments, and will not be repeated here.

[0107] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The implemented program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0108] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for establishing a molecular property prediction model, characterized in that, The method includes: Obtain molecular samples and generate molecular models of the molecular samples based on the molecular samples; The molecular properties corresponding to the molecular samples are obtained respectively, and the molecular properties and their corresponding molecular models are paired and stored in the training sample set; The machine learning model is trained using the training sample set to obtain a property prediction model, which is used to predict molecular properties. The process of generating a molecular model based on the molecular sample includes: Obtain atomic coordinate information in the target molecule; Based on the atomic coordinate information, the electron density grid information of the target molecule is obtained at a preset resolution; the process of obtaining the electron density grid information of the target molecule based on the atomic coordinate information at a preset resolution includes: transforming the atomic coordinate information from real space to frequency domain space based on Fourier transform to generate a structure factor; adjusting the structure factor back to real space based on inverse Fourier transform at a preset resolution to generate the electron density grid information, wherein the preset resolution is a preset range of summation of frequency domain space vector components; The target electron density value greater than the preset electron density value is extracted from the electron density grid information, and the 3D matrix composed of the target electron density values ​​is used to construct the molecular model of the target molecule.

2. The method according to claim 1, characterized in that, The atomic coordinate information is transformed from real space to frequency domain space based on Fourier transform to generate the structure factor. The calculation formula for the transformation process is as follows: in r Represents atomic coordinate information, x, y, z These are three components representing atomic coordinate information. f n represent n Diffraction factors of different elements, s Represents a frequency domain space vector. h, k, l These represent the three components of the frequency domain spatial vector. F(s) The structure factor representing the s vector. It is an imaginary number.

3. The method according to claim 1, characterized in that, The structure factor is adjusted to a preset resolution based on the inverse Fourier transform and returned to the real space to generate the electron density grid information. The calculation formula for the return process is as follows: in, r Represents atomic coordinate information, s Represents a frequency domain space vector. h, k, l These represent the three components of the frequency domain spatial vector. F(s) The structure factor of vector s, ρ(r) Represents the electron density grid information, Vcell Represents the volume of the unit cell. The preset resolution is an imaginary number, determined by a preset value. h, k, l The range of summation is obtained.

4. The method according to claim 1, characterized in that, The preset resolution includes multiple different sub-resolutions, and the method further includes: The process sequentially uses each of the sub-resolutions to perform the following steps: calculating the electron density grid information of the target molecule based on the atomic coordinate information at a preset resolution; extracting the target electron density value greater than the preset electron density value from the electron density grid information; and constructing a molecular model of the target molecule from a 3D matrix composed of the target electron density values, thereby generating multiple 3D matrices. The multiple 3D matrices are superimposed to form a 4D matrix, generating a multi-channel model.

5. The method according to claim 4, characterized in that, The method further includes: The electron density values ​​in the multi-channel model are normalized to obtain a normalized model.

6. The method according to claim 5, characterized in that, The method further includes: Using the center of the normalized model as the rotation center, the normalized matrix is ​​rotated in random directions and at random angles to obtain the rotated molecular model.

7. A method for predicting molecular properties, characterized in that, The method includes: Obtain the molecule to be predicted and generate a molecular model of the molecule to be predicted; The molecular model is input into a property prediction model obtained using the molecular property prediction model establishment method as described in any one of claims 1-6, and the molecular properties of the molecule to be predicted are predicted by the property prediction model.

8. An electronic device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method as described in any one of claims 1-7.

Images