Method and system for crystal analysis

Abstract

Disclosed in the present application are a method and system for crystal analysis. A method for crystal analysis, comprising: acquiring an electron density map of a crystal, the electron density map being generated by performing phase analysis on diffraction data of the crystal; on the basis of electron density peaks of the electron density map, determining coordinates of each non-hydrogen atom of the crystal; on the basis of the coordinates of each non-hydrogen atom of the crystal, building a geometric relationship map of the non-hydrogen atoms; and using a graph neural network-based non-hydrogen atom identification module to determine the element type of each non-hydrogen atom of the crystal on the basis of the geometric relationship map of the non-hydrogen atoms.

Description

Method and system for crystal resolution TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and more particularly, to a method and system for crystal resolution. BACKGROUND

[0002] Observation of the microscopic molecular structure of a substance is a key factor in driving scientific discovery. Crystal resolution is one of the important technologies that can resolve the structure of a crystal in detail, which cannot be achieved by any other spectroscopic analysis method. Crystal resolution plays a crucial role in chemical research, and has been closely related to the award of Nobel Prize in Chemistry for more than a hundred years. Nowadays, research on new materials and new drugs has attracted widespread attention, involving a large number of new substance synthesis, and crystal resolution will be widely used in the structure resolution of new synthesized substances, and has almost become a routine research in the related field. However, crystal resolution needs crystallography experts to manually complete the resolution task through repeated trial and error with quite professional crystallography knowledge. This is very time-consuming and labor-intensive for crystal resolution, which has a large demand in daily research. Therefore, there is an urgent need for a fast and accurate crystal resolution method and system.

[0003] SUMMARY

[0004] It should be understood that the above general description and the following detailed description of the application are exemplary and illustrative, and are intended to provide further explanation of the application as claimed.

[0005] According to an aspect of the present application, a method for crystal resolution is provided, comprising: obtaining an electron density map of a crystal, the electron density map being generated by phase resolution on diffraction data of the crystal; determining coordinates of each non-hydrogen atom of the crystal based on electron density peaks of the electron density map; constructing a geometric relationship graph of non-hydrogen atoms based on the coordinates of each non-hydrogen atom of the crystal; and determining an element type of each non-hydrogen atom of the crystal based on the geometric relationship graph of the non-hydrogen atoms using a non-hydrogen atom recognition module based on a graph neural network.

[0006] According to another aspect of the present application, there is provided a method for crystal resolution, comprising: obtaining coordinates and element types of each non-hydrogen atom of a crystal; constructing a geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein the equivalent atoms comprise non-hydrogen atoms and equivalent atoms; determining a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal based on the geometric relationship graph of the equivalent atoms using a graph neural network-based hydrogenation module; and determining coordinates of each hydrogen atom of the crystal based on the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal.

[0007] According to yet another aspect of the present application, there is provided a system for crystal resolution, comprising means for performing the method according to any one of the above methods.

[0008] The method and system for crystal resolution according to embodiments of the present application can achieve fast and accurate crystal resolution. BRIEF DESCRIPTION OF DRAWINGS

[0009] The accompanying drawings, which are included to provide a further understanding of the present application and are incorporated in and constitute a part of this application, illustrate embodiments of the present application and together with the description serve to explain the principles of the present application. In the drawings:

[0010] FIG. 1 is a flowchart of a method for crystal resolution according to an embodiment of the present application;

[0011] FIG. 2 is a flowchart of a method for constructing a geometric relationship graph of non-hydrogen atoms according to an embodiment of the present application;

[0012] FIG. 3 is a flowchart of a method for training a graph neural network-based non-hydrogen atom recognition module according to an embodiment of the present application;

[0013] FIG. 4 is a flowchart of a method for crystal resolution according to another embodiment of the present application;

[0014] FIG. 5 is a flowchart of a method for constructing a geometric relationship graph of equivalent atoms according to an embodiment of the present application;

[0015] FIG. 6 is a flowchart of a method for training a graph neural network-based hydrogenation module according to an embodiment of the present application;

[0016] FIG. 7 is a flowchart of a method for crystal resolution according to yet another embodiment of the present application;

[0017] FIG. 8 is a flowchart of a method for determining a hydrogenation operation type according to an embodiment of the present application;

[0018] FIG. 9A is a schematic diagram of an electron density map and an electron density peak according to an embodiment of the present application;

[0019] FIG. 9B is a diagram of coordinates of non-hydrogen atoms and element types of a crystal according to one embodiment of the present application;

[0020] FIG. 10 is a block diagram of a system for crystal analysis according to one embodiment of the present application;

[0021] FIG. 11 is a block diagram of a system for crystal analysis according to another embodiment of the present application;

[0022] FIG. 12 is a block diagram of a system for crystal analysis according to still another embodiment of the present application;

[0023] FIG. 13 is a result graph of performance evaluation of crystal analysis; and

[0024] FIG. 14 is a comparison graph of crystal analysis applications. DETAILED DESCRIPTION

[0025] Reference will now be made in detail embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to like elements throughout. The embodiments of the present application are not limited to the specific illustrative

[0026] Although the terms used in the present application are selected from publicly-known and used terms from among the terms having the same or similar meanings, some of the terms mentioned in the specification of the present application can be selected by the applicant from among the terms after his or her consideration, or may

[0027] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In other instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

[0028] Currently, crystal analysis needs to be completed manually by crystallography experts with professional crystallography knowledge, and there is no fully automated tool. Crystal analysis generally uses X-ray to irradiate the crystal to generate diffraction data, and after phase analysis of the diffraction data, the diffraction data is restored to the electron density in the unit cell through Fourier transform. At this time, the crystallographer needs to manually identify the non-hydrogen atom of the electron density, that is, according to the shape of the electron density, a position coordinate (x, y, z) in the 3D space is identified to correspond to a type of non-hydrogen atom (C, N, O, etc.). This identification process requires professional crystallography knowledge and repeated attempts to verify correctness, and after identification, the atomic coordinates need to be refined to better fit the theoretical diffraction data of the current structure and the real diffraction data measured by experiment. If the fitting effect of the theoretical diffraction data and the real diffraction data is poor after refinement, it is usually due to the error of the atomic identification in the previous step, so the incorrectly identified atoms need to be re-identified. Therefore, the crystallographer needs to use professional crystallography tools and rely on professional crystallography knowledge to repeatedly identify non-hydrogen atoms and refine to obtain the correct structure, which consumes a lot of time and human resources. Since crystal analysis has a large demand in daily research in related fields, identification by crystallographers is very time-consuming and laborious, and there are many identification errors.

[0029] In view of the above problems, the present application provides a method and system for crystal analysis.

[0030] FIG. 1 is a flowchart of a method 100 for crystal analysis according to an embodiment of the present application. The method can be used for single crystals, or can be used for polycrystals.

[0031] At step 102, an electron density map of the crystal can be obtained. One example of the electron density map is shown on the left side of FIG. 9A. Although the electron density map shown in FIG. 9A is in two-dimensional form, it should be understood that the electron density map actually reflects the three-dimensional spatial information of the crystal structure. The electron density map can be generated by phase analysis of the diffraction data of the crystal. The diffraction data of the crystal can include diffraction angle, diffraction intensity, structure factor, space group and symmetry, lattice parameter, etc. For polycrystals, diffraction data can be obtained from polycrystals by serial femtosecond crystallography, 3D electron diffraction, etc. For single crystals, in addition to the aforementioned methods, diffraction data can also be obtained from single crystals by X-ray, etc. The diffraction data of the crystal can also be obtained from a database. In order to convert the diffraction data of the crystal into an electron density map, phase analysis can be achieved by shelxs, shelxt, charge flipping, etc. crystallography tools.

[0032] At step 104, the coordinates of each non-hydrogen atom of the crystal can be determined based on the electron density peaks of the electron density map. FIG. 9A shows the electron density peaks determined from the electron density map, each corresponding to one non-hydrogen atom. Since there are multiple electron density peaks in the electron density map, each corresponding to one non-hydrogen atom, the coordinates of each non-hydrogen atom can be determined by finding the coordinates of each electron density peak. Specifically, the coordinates of each electron density peak can be determined by finding the local maxima of the electron density map and then determining the coordinates corresponding to the location of each maximum. Although there are also electron density peaks corresponding to hydrogen atoms in the electron density map, since the peaks corresponding to hydrogen atoms are lower, the electron density peaks corresponding to hydrogen atoms can be excluded based on the peak value compared to the electron density peaks corresponding to non-hydrogen atoms.

[0033] At step 106, a geometric relationship graph of non-hydrogen atoms can be constructed based on the coordinates of each non-hydrogen atom of the crystal. The geometric relationship graph of non-hydrogen atoms can include the coordinates of each non-hydrogen atom and the connection relationship between non-hydrogen atoms.

[0034] In one embodiment, step 106 can include steps 202-204, as shown in FIG. 2.

[0035] At step 202, the distance between each atom in all non-hydrogen atoms and every other atom can be calculated. For example, the distance between each non-hydrogen atom and each other non-hydrogen atom can be calculated based on the coordinates of each non-hydrogen atom.

[0036] At step 204, each pair of atoms in all non-hydrogen atoms with a distance less than a distance threshold can be connected as nodes to construct the geometric relationship graph of non-hydrogen atoms. The distance threshold can be set as needed.

[0037] Returning to FIG. 1, at step 108, the element type of each non-hydrogen atom of the crystal can be determined based on the geometric relationship graph of non-hydrogen atoms using a graph neural network-based non-hydrogen atom identification module. In some applications, for crystals without hydrogen atoms, the coordinates and element types of each non-hydrogen atom determined according to the above method constitute a complete crystal structure. In other applications, the coordinates and element types of each non-hydrogen atom determined according to the above method constitute the skeleton of the crystal and can be used for preliminary crystal structure analysis.

[0038] In one embodiment, the bond between non-hydrogen atoms can be calculated based on the covalent radius corresponding to the element type of the non-hydrogen atom.

[0039] Instead of requiring a crystallography expert to perform crystal analysis by repeatedly trying manually with crystallography knowledge, by using the method for crystal analysis, the application can achieve automatic crystal analysis without manual identification by human experts. In addition, the application can achieve thousands of times the speed of human experts and higher accuracy, which can greatly free up time and labor from tedious manual crystal analysis.

[0040] FIG. 3 is a flowchart of a method 300 of training a graph neural network-based non-hydrogen atom identification module according to an embodiment of the application. The graph neural network-based non-hydrogen atom identification module can be trained by the method 300.

[0041] At step 302, a dataset can be received. The dataset can include a geometric relationship graph of non-hydrogen atoms of each crystal of a plurality of crystals, and a true value of an element type of the non-hydrogen atoms. The plurality of crystals can include single crystals, or can include polycrystals. The geometric relationship graph of non-hydrogen atoms of each crystal can be constructed based on diffraction data of the crystal by a method similar to steps 102-106.

[0042] At step 304, an element type prediction result of each non-hydrogen atom of each crystal of the plurality of crystals can be generated based on the dataset using the graph neural network-based non-hydrogen atom identification module. The graph neural network-based non-hydrogen atom identification module can be initialized based on a graph neural network model, and classify nodes (non-hydrogen atoms) in the geometric relationship graph of non-hydrogen atoms of each crystal to predict an element type corresponding to each non-hydrogen atom. Since the graph neural network is particularly suitable for processing graph structure data, the graph neural network-based non-hydrogen atom identification module can achieve fast and accurate classification of nodes in the geometric relationship graph of non-hydrogen atoms.

[0043] At step 306, a loss function between the element type prediction result of each non-hydrogen atom of each crystal of the plurality of crystals and the true value can be calculated. The loss function can be cross-entropy loss, mean square error, etc.

[0044] At step 308, the parameters of the graph neural network-based non-hydrogen atom identification module can be updated based on the loss function using a parameter updating method to obtain a trained graph neural network-based non-hydrogen atom identification module. The parameter updating method can be gradient backpropagation, gradient descent, momentum optimization, etc. By updating the parameters through the loss function and the parameter updating method, the performance of the module can be optimized.

[0045] By performing the method 300, a trained graph neural network-based non-hydrogen atom identification module can be obtained. By using the graph neural network-based non-hydrogen atom identification module, the application can achieve fast and accurate identification of non-hydrogen atoms.

[0046] In crystal analysis, in addition to the identification of non-hydrogen atoms, another problem that needs to be solved is how to identify the hydrogen atoms attached to the non-hydrogen atoms. Since the scattering ability of hydrogen atoms to X-rays is weak, hydrogen atoms cannot be directly observed from the electron density map of the crystal diffraction data in most cases. At present, crystallographers also need to judge which non-hydrogen atoms to add hydrogen atoms according to their professional crystallographic knowledge. Since hydrogen atoms are difficult to observe from the electron density map, it is very easy to miss or add errors, and crystallographers need to try repeatedly to correctly add hydrogen. In view of the above problems, another method and system for crystal analysis are proposed.

[0047] FIG. 4 is a flowchart of a method 400 for crystal analysis according to another embodiment of the present application. The method can be used for single crystals, or can be used for polycrystals.

[0048] At step 402, the coordinates and element types of each non-hydrogen atom of the crystal can be obtained. FIG. 9B shows a schematic diagram representing the coordinates and element types of each non-hydrogen atom of the crystal, wherein the connection relationship between the non-hydrogen atoms indicates the bond that can be calculated according to the covalent radius corresponding to the element type of the non-hydrogen atom, but the information obtained in step 402 does not need to include this bond information. In one embodiment, the coordinates and element types of each non-hydrogen atom of the crystal can be obtained from the diffraction data of the crystal. In one embodiment, the coordinates and element types of each non-hydrogen atom of the crystal can be determined based on the diffraction data of the crystal using a non-hydrogen atom identification module based on a graph neural network. In another embodiment, the coordinates and element types of each non-hydrogen atom of the crystal can be determined based on the diffraction data of the crystal by automated software. In yet another embodiment, the coordinates and element types of each non-hydrogen atom of the crystal can be determined based on the diffraction data of the crystal by manual identification. The diffraction data of the crystal can include diffraction angle, diffraction intensity, structure factor, space group and symmetry, lattice parameter, etc. For polycrystals, diffraction data can be obtained from polycrystals by serial femtosecond crystallography, 3D electron diffraction, etc. For single crystals, in addition to the aforementioned methods, diffraction data can also be obtained from single crystals by X-ray, etc. The diffraction data of the crystal can also be obtained from a database.

[0049] At step 404, an equivalent atom geometric relationship graph can be constructed based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein the equivalent atom contains the non-hydrogen atom and the equivalent atom.

[0050] In one embodiment, step 404 can include steps 502-506, as shown in FIG. 5.

[0051] At step 502, the coordinates and element types of equivalent atoms within a predetermined distance of each non-hydrogen atom of the crystal can be determined using the symmetry and periodicity of the crystal. In chemical research, based on the symmetry and periodicity of the crystal, the position and element type of the equivalent atoms can be determined given the position and element type of the non-hydrogen atoms. Since the internal structure of the crystal is infinitely repeated, only the equivalent atoms within a predetermined distance of the non-hydrogen atoms need to be determined. The predetermined distance can be set as needed.

[0052] At step 504, the distance between each of the equivalent atoms to every other equivalent atom can be calculated, where the equivalent atoms include the non-hydrogen atoms and the determined equivalent atoms. Specifically, the distance between each equivalent atom and every other equivalent atom can be calculated based on the coordinates of each equivalent atom.

[0053] At step 506, each pair of equivalent atoms with a distance less than a distance threshold in the equivalent atoms can be connected as a node to construct a geometric relationship graph of the equivalent atoms. The distance threshold can be set as needed. The distance threshold can be the same as or different from the distance threshold used in step 204 to construct the geometric relationship graph of the non-hydrogen atoms.

[0054] Returning to FIG. 4, at step 406, a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal can be determined using the graph neural network-based hydrogenation module based on the geometric relationship graph of the equivalent atoms. The hydrogenation operation type can include, for example, no hydrogenation atom, 1 hydrogen atom to carbon atom, 2 hydrogen atoms to carbon atom, 3 hydrogen atoms to carbon atom, 1 hydrogen atom to oxygen atom, etc.

[0055] At step 408, the coordinates of each hydrogen atom of the crystal can be determined based on the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal. When the hydrogenation operation type is determined, the coordinates of the hydrogen atom corresponding to the non-hydrogen atom can be uniquely determined based on the bond formation calculated according to the covalent radius corresponding to the element type of the non-hydrogen atom.

[0056] After determining the coordinates of the hydrogen atom corresponding to each non-hydrogen atom of the crystal, in combination with the coordinates and element types of each non-hydrogen atom of the crystal obtained, the coordinates and element types of all atoms of the molecule in the real environment can be obtained. In one embodiment, visualization of the crystal structure can be implemented based on the coordinates and element types of all atoms. In another embodiment, a Crystallographic Information File (CIF) can be generated based on the coordinates and element types of all atoms. The CIF file contains crystal structure information and can be read and / or processed by computer software.

[0057] Instead of requiring a crystallography expert to perform crystal analysis by repeatedly trying manually with crystallography knowledge, by using the method for crystal analysis, the application can achieve automatic crystal analysis without manual identification by human experts. In addition, the application can achieve thousands of times the speed of human experts and higher accuracy, which can greatly liberate time and labor from tedious manual crystal analysis.

[0058] FIG. 6 is a flowchart of a method 600 of training a graph neural network-based hydrogenation module according to one embodiment of the application. The graph neural network-based hydrogenation module can be trained by the method 600.

[0059] At step 602, a dataset can be received. The dataset can include an equivalent atom-geometric relationship graph of each crystal in a plurality of crystals, and a true value of a hydrogenation operation type of a non-hydrogen atom. The plurality of crystals can include single crystals, or can include polycrystals. The equivalent atom-geometric relationship graph of each crystal can be constructed based on the coordinates and element types of the non-hydrogen atoms of the crystal, by a method similar to step 404.

[0060] At step 604, a hydrogenation operation type prediction result of each non-hydrogen atom in each crystal in the plurality of crystals can be generated based on the dataset using the graph neural network-based hydrogenation module. The graph neural network-based hydrogenation module can be initialized based on a graph neural network model, and classify the non-hydrogen atoms in the equivalent atom-geometric relationship graph of each crystal to predict the hydrogenation operation type corresponding to each non-hydrogen atom. Since the graph neural network is particularly suitable for processing graph structure data, the graph neural network-based hydrogenation module can achieve fast and accurate classification of non-hydrogen atoms in the equivalent atom-geometric relationship graph.

[0061] At step 606, a loss function between the hydrogenation operation type prediction result of each non-hydrogen atom in each crystal in the plurality of crystals and the true value can be calculated. The loss function can be cross-entropy loss, mean square error, etc.

[0062] At step 608, the parameters of the graph neural network-based hydrogenation module can be updated to obtain a trained graph neural network-based hydrogenation module based on the loss function using a parameter updating method. The parameter updating method can be gradient backpropagation, gradient descent, momentum optimization, etc. By updating the parameters through the loss function and the parameter updating method, the performance of the module can be optimized.

[0063] By performing the method 600, a trained graph neural network-based hydrogenation module can be obtained. By using the graph neural network-based hydrogenation module, the application can achieve fast and accurate determination of the hydrogenation operation type of the non-hydrogen atom.

[0064] FIG. 7 is a flowchart of a method 700 for crystal resolution according to yet another embodiment of the present application. The method can be used for a single crystal, or can be used for a polycrystal.

[0065] Steps 702-708 are the same as steps 102-108 in the method 100 described with reference to FIG. 1. For the sake of brevity, the specific details of steps 702-708 are not repeated here.

[0066] Based on the coordinates and element types of the non-hydrogen atoms that have been determined, in order to determine the complete crystal structure, the hydrogenation operation type corresponding to each non-hydrogen atom can be further determined to determine the complete coordinates of all atoms including the non-hydrogen atoms and hydrogen atoms.

[0067] At step 710, the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal can be determined. In one embodiment, the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal can be determined based on the coordinates and element types of each non-hydrogen atom of the crystal determined in steps 704 and 708, by manually identifying the hydrogenation operation type. In another embodiment, the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal can be determined based on the coordinates and element types of each non-hydrogen atom of the crystal determined in steps 704 and 708, using a graph neural network-based hydrogenation operation module.

[0068] In one embodiment, step 710 can include steps 802-804, as shown in FIG. 8.

[0069] At step 802, a geometric relationship graph of equivalent atoms can be constructed based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein the equivalent atoms contain the non-hydrogen atom and equivalent atoms. The geometric relationship graph of equivalent atoms can include the coordinates of each equivalent atom, and the connection relationship between the equivalent atoms.

[0070] In one embodiment, constructing the geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal can include: determining the coordinates and element types of the equivalent atoms of each non-hydrogen atom of the crystal within a predetermined distance using the symmetry and periodicity of the crystal, calculating the distance between each of the equivalent atoms to every other equivalent atom, wherein the equivalent atoms contain the non-hydrogen atom and the determined equivalent atoms, and connecting the edges between each pair of equivalent atoms in the equivalent atoms whose distance is less than a distance threshold, to construct the geometric relationship graph of equivalent atoms.

[0071] At step 804, a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal can be determined using the graph neural network-based hydrogenation module based on the geometry graph of equivalent atoms. The hydrogenation operation type can include, for example, no hydrogenation atom, add 1 hydrogen atom to a carbon atom, add 2 hydrogen atoms to a carbon atom, add 3 hydrogen atoms to a carbon atom, add 1 hydrogen atom to an oxygen atom, and the like.

[0072] In one embodiment, the graph neural network-based hydrogenation module can be trained by receiving a dataset including a geometry graph of equivalent atoms of each crystal of a plurality of crystals and a true value of a hydrogenation operation type of a non-hydrogen atom, generating a hydrogenation operation type prediction result of each non-hydrogen atom of each crystal of the plurality of crystals based on the dataset using the graph neural network-based hydrogenation module, calculating a loss function between the hydrogenation operation type prediction result of each non-hydrogen atom of each crystal of the plurality of crystals and the true value, and updating parameters of the graph neural network-based hydrogenation module using a parameter update method based on the loss function to obtain a trained graph neural network-based hydrogenation module.

[0073] Returning to FIG. 7, at step 712, coordinates of each hydrogen atom of the crystal can be determined based on the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal. When the hydrogenation operation type is determined, the coordinates of the hydrogen atom corresponding to the non-hydrogen atom can be uniquely determined based on the bond formation calculated according to the covalent radius corresponding to the element type of the non-hydrogen atom.

[0074] After determining the coordinates and element types of each non-hydrogen atom of the crystal and the coordinates of the hydrogen atom corresponding to each non-hydrogen atom, all atomic coordinates and element types of the molecule in a real environment can be obtained. In one embodiment, visualization of the crystal structure can be implemented based on the atomic coordinates and element types of all atoms. In another embodiment, a CIF file can be generated based on the atomic coordinates and element types of all atoms.

[0075] Instead of requiring a crystallography expert to perform crystal analysis by repeatedly manually trying with crystallography knowledge, by using the above-described method for crystal analysis, the present application can achieve automatic crystal analysis and can achieve faster speed and higher accuracy.

[0076] FIG. 10 is a block diagram of a system 1000 for crystal resolution, according to one embodiment of the present application. The system 1000 can include an electron density map obtaining module 1002, a non-hydrogen atom coordinate determining module 1004, a non-hydrogen atom geometric relationship graph constructing module 1006, and a graph neural network based non-hydrogen atom identifying module 1008. The electron density map obtaining module 1002 can be configured to obtain an electron density map of a crystal, the electron density map being generated by phase resolution on diffraction data of the crystal. The non-hydrogen atom coordinate determining module 1004 can be configured to determine coordinates of each non-hydrogen atom of the crystal based on electron density peaks of the electron density map. The non-hydrogen atom geometric relationship graph constructing module 1006 can be configured to construct a geometric relationship graph of non-hydrogen atoms based on the coordinates of each non-hydrogen atom of the crystal. The graph neural network based non-hydrogen atom identifying module 1008 can be configured to determine an element type of each non-hydrogen atom of the crystal based on the geometric relationship graph of non-hydrogen atoms.

[0077] In one embodiment, the non-hydrogen atom geometric relationship graph constructing module 1006 can be configured to construct the geometric relationship graph of non-hydrogen atoms based on the coordinates of each non-hydrogen atom of the crystal by: calculating distances between each atom in all non-hydrogen atoms and every other atom, and connecting edges between each pair of atoms in all non-hydrogen atoms with distances less than a distance threshold to construct the geometric relationship graph of non-hydrogen atoms.

[0078] In one embodiment, the graph neural network based non-hydrogen atom identifying module 1008 can be trained by: receiving a dataset, the dataset including the geometric relationship graph of non-hydrogen atoms of each crystal in a plurality of crystals, and ground truth values of element types of non-hydrogen atoms; generating, using the graph neural network based non-hydrogen atom identifying module, element type prediction results of each non-hydrogen atom of each crystal in the plurality of crystals based on the dataset; calculating a loss function between the element type prediction results of each non-hydrogen atom of each crystal in the plurality of crystals and the ground truth values; and updating, based on the loss function, parameters of the graph neural network based non-hydrogen atom identifying module using a parameter updating method to obtain a trained graph neural network based non-hydrogen atom identifying module.

[0079] FIG. 11 is a block diagram of a system 1100 for crystal resolution according to another embodiment of the present application. The system 1100 can include a non-hydrogen atom obtaining module 1102, a geometry graph of equivalent atoms constructing module 1104, a graph neural network based hydrogenation module 1106, and a hydrogen atom coordinate determining module 1108. The non-hydrogen atom obtaining module 1102 can be configured to obtain coordinates and element types of each non-hydrogen atom of a crystal. The geometry graph of equivalent atoms constructing module 1104 can be configured to construct a geometry graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein an equivalent atom contains a non-hydrogen atom and an equivalent atom. The graph neural network based hydrogenation module 1106 can be configured to determine a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal based on the geometry graph of equivalent atoms. The hydrogen atom coordinate determining module 1108 can be configured to determine coordinates of each hydrogen atom of the crystal based on the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal.

[0080] In one embodiment, the geometry graph of equivalent atoms constructing module 1104 can be configured to construct the geometry graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal by: determining coordinates and element types of equivalent atoms of each non-hydrogen atom of the crystal within a predetermined distance using symmetry and periodicity of the crystal; calculating distances between each of the equivalent atoms to every other equivalent atom, wherein an equivalent atom contains a non-hydrogen atom and a determined equivalent atom; and connecting edges between each pair of equivalent atoms in the equivalent atoms with distances less than a distance threshold as nodes to construct the geometry graph of equivalent atoms.

[0081] In one embodiment, the graph neural network based hydrogenation module 1106 can be trained by: receiving a dataset, the dataset including a geometry graph of equivalent atoms of each of a plurality of crystals, and ground truth values of hydrogenation operation types of non-hydrogen atoms; generating, using the graph neural network based hydrogenation module, prediction results of the hydrogenation operation types of each non-hydrogen atom of each of the plurality of crystals based on the dataset; calculating a loss function between the prediction results and the ground truth values of the hydrogenation operation types of each non-hydrogen atom of each of the plurality of crystals; and updating, based on the loss function, parameters of the graph neural network based hydrogenation module using a parameter updating method to obtain a trained graph neural network based hydrogenation module.

[0082] FIG. 12 is a block diagram of a system 1200 for crystal resolution according to yet another embodiment of the present application. The system 1200 can include an electron density map obtaining module 1202, a non-hydrogen atom coordinate determining module 1204, a geometric relationship map of equivalent atoms constructing module 1206, a graph neural network based non-hydrogen atom identifying module 1208, a hydrogenation operation type determining module 1210, and a hydrogen atom coordinate determining module 1212. The electron density map obtaining module 1202 ~ the graph neural network based non-hydrogen atom identifying module 1208 are the same as the electron density map obtaining module 1002 ~ the graph neural network based non-hydrogen atom identifying module 1008 in the system 1000 described with reference to FIG. 10. To avoid redundancy, the specific details of the electron density map obtaining module 1202 ~ the graph neural network based non-hydrogen atom identifying module 1208 are not described herein. The hydrogenation operation type determining module 1210 can be configured to determine a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal. The hydrogen atom coordinate determining module 1212 can be configured to determine a coordinate of each hydrogen atom of the crystal based on the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal.

[0083] In one embodiment, the hydrogenation operation type determining module 1210 can include a geometric relationship map of equivalent atoms constructing module and a graph neural network based hydrogenation module. The geometric relationship map of equivalent atoms constructing module can be configured to construct a geometric relationship map of equivalent atoms based on the coordinate and the element type of each non-hydrogen atom of the crystal, wherein the equivalent atom contains the non-hydrogen atom and the equivalent atom. The graph neural network based hydrogenation module can be configured to determine the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal based on the geometric relationship map of equivalent atoms.

[0084] In one embodiment, the geometric relationship map of equivalent atoms constructing module can be configured to construct the geometric relationship map of equivalent atoms based on the coordinate and the element type of each non-hydrogen atom of the crystal by: determining the coordinate and the element type of the equivalent atom of each non-hydrogen atom of the crystal within a predetermined distance by utilizing the symmetry and the periodicity of the crystal; calculating the distance between each of the equivalent atoms to every other equivalent atom, wherein the equivalent atom contains the non-hydrogen atom and the determined equivalent atom; and connecting edges between each pair of equivalent atoms in the equivalent atoms with a distance less than a distance threshold to construct the geometric relationship map of equivalent atoms.

[0085] In one embodiment, the graph neural network based hydrogenation module is trained by the following steps: receiving a dataset, the dataset comprising an equivalent atom graph of each crystal in a plurality of crystals and a ground truth of hydrogenation operation type of non-hydrogen atoms; generating, using the graph neural network based hydrogenation module, a prediction of hydrogenation operation type of each non-hydrogen atom of each crystal in the plurality of crystals based on the dataset; calculating a loss function between the prediction of hydrogenation operation type of each non-hydrogen atom of each crystal in the plurality of crystals and the ground truth; and updating, based on the loss function, a parameter of the graph neural network based hydrogenation module using a parameter update method to obtain a trained graph neural network based hydrogenation module.

[0086] Figure 13 is a result plot of performance evaluation of crystal analysis. The present application performed performance evaluation on 10,000 real measured single crystal diffraction data. (a) part shows the performance of the non-hydrogen atom identification module in non-hydrogen atom identification. The non-hydrogen atom identification module generally achieved very high performance on a large number of element types covering almost the entire periodic table. (b) part shows the performance of the hydrogenation module in hydrogenation operation type determination. The hydrogenation module achieved very high performance on a variety of common hydrogenation operation types. (c) part shows the crystal prediction accuracy, i.e. the proportion of crystals in which all non-hydrogen atoms are correctly predicted. The non-hydrogen atom identification module of the present application achieved a very significant performance improvement compared to the traditional crystallography tools that currently have some non-hydrogen atom identification capability. Top-1 refers to the accuracy of directly taking the element type with the highest model prediction probability, i.e. the accuracy of the model directly predicting correctly. Top-2 refers to the accuracy of directly taking the element type with the second highest model prediction probability, which contains at most one incorrect atom, and the correct element type of the incorrect atom can be corrected by taking the second highest model prediction probability. (d), (e) part shows that the analysis results of the non-hydrogen atom identification module combined with the hydrogenation module and the analysis results of human experts have very high consistency in terms of indicators reflecting the quality of analysis (residual (R1), goodness of fit (S)), while the analysis of the present application is fully automatic and the analysis speed is thousands of times faster than that of human experts.

[0087] When the method of the present application is applied to the crystal diffraction data provided in multiple published papers, the method of the present application finds multiple errors in the crystal structure. Figure 14 is a comparison plot of crystal analysis application, the left side is the crystal structure diagram shown in the published paper, and the right side is the non-hydrogen atom and hydrogen atom identified according to the method of the present application. As indicated by the dashed line box, there are three typical errors in the crystal structure in the published paper, which are non-hydrogen atom identification error (part a), hydrogen atom addition error (part b), and hydrogen atom addition omission (part c). In addition, the residual (R1) of the crystal structure identified by the present application compared with the real experimental data is also reduced compared with the published paper.

[0088] Compared with the crystallographer using crystallographic knowledge to perform crystal analysis by repeatedly trying manually, the automatic crystal analysis method realized by using artificial intelligence technology can complete the crystal analysis task without any professional knowledge background, greatly improve the analysis speed, and ensure the accuracy of the analysis.

[0089] Further examples:

[0090] Example 1. A method for crystal analysis, comprising:

[0091] obtaining an electron density map of a crystal, the electron density map being generated by phase analysis on diffraction data of the crystal;

[0092] determining coordinates of each non-hydrogen atom of the crystal based on electron density peaks of the electron density map;

[0093] constructing a geometric relationship graph of non-hydrogen atoms based on the coordinates of each non-hydrogen atom of the crystal; and

[0094] determining an element type of each non-hydrogen atom of the crystal using a graph neural network-based non-hydrogen atom recognition module based on the geometric relationship graph of the non-hydrogen atoms.

[0095] Example 2. The method of example 1, wherein constructing a geometric relationship graph of non-hydrogen atoms based on the coordinates of each non-hydrogen atom of the crystal comprises:

[0096] calculating a distance between each atom in all non-hydrogen atoms and every other atom; and

[0097] connecting edges between each pair of atoms in all non-hydrogen atoms with a distance less than a distance threshold to construct the geometric relationship graph of the non-hydrogen atoms.

[0098] Example 3. The method of example 1, wherein the graph neural network-based non-hydrogen atom recognition module is trained by:

[0099] receiving a data set comprising a geometric relationship graph of non-hydrogen atoms and a true value of an element type of non-hydrogen atoms of each crystal in a plurality of crystals;

[0100] generating an element type prediction result of each non-hydrogen atom of each crystal in the plurality of crystals using the graph neural network-based non-hydrogen atom recognition module based on the data set;

[0101] calculating a loss function between the element type prediction result and the true value of each non-hydrogen atom of each crystal in the plurality of crystals; and

[0102] update the parameters of the graph neural network based non-hydrogen atom identification module using a parameter updating method based on the loss function to obtain a trained graph neural network based non-hydrogen atom identification module.

[0103] Example 4. The method of example 1, further comprising:

[0104] determining a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal; and

[0105] determining coordinates of each hydrogen atom of the crystal based on the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal.

[0106] Example 5. The method of example 4, wherein determining a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal comprises:

[0107] constructing a geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein the equivalent atoms comprise the non-hydrogen atom and equivalent atoms; and

[0108] determining the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal based on the geometric relationship graph of equivalent atoms using a graph neural network based hydrogenation module.

[0109] Example 6. The method of example 5, wherein constructing a geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal comprises:

[0110] determining coordinates and element types of equivalent atoms of each non-hydrogen atom of the crystal within a predetermined distance using symmetry and periodicity of the crystal;

[0111] calculating distances between each of the equivalent atoms to every other equivalent atom, wherein the equivalent atoms comprise the non-hydrogen atom and the determined equivalent atoms; and

[0112] connecting edges between each pair of equivalent atoms in the equivalent atoms having a distance less than a distance threshold to construct the geometric relationship graph of equivalent atoms.

[0113] Example 7. The method of example 5, wherein the graph neural network based hydrogenation module is trained by:

[0114] receiving a dataset comprising a geometric relationship graph of equivalent atoms and a ground truth of hydrogenation operation types of non-hydrogen atoms for each of a plurality of crystals;

[0115] using the graph neural network based hydrogenation module, generating a hydrogenation operation type prediction result of each non-hydrogen atom of each crystal in the plurality of crystals based on the dataset;

[0116] calculating a loss function between the hydrogenation operation type prediction result of each non-hydrogen atom of each crystal in the plurality of crystals and a true value; and

[0117] updating parameters of the graph neural network based hydrogenation module based on the loss function using a parameter updating method to obtain a trained graph neural network based hydrogenation module.

[0118] Example 8. A method for crystal resolution, comprising:

[0119] obtaining coordinates and element types of each non-hydrogen atom of a crystal;

[0120] constructing a geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein the equivalent atoms include non-hydrogen atoms and equivalent atoms;

[0121] determining a hydrogenation operation type corresponding to each non-hydrogen atom of the crystal using a graph neural network based hydrogenation module based on the geometric relationship graph of the equivalent atoms; and

[0122] determining coordinates of each hydrogen atom of the crystal based on the hydrogenation operation type corresponding to each non-hydrogen atom of the crystal.

[0123] Example 9. The method of example 8, wherein constructing a geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal comprises:

[0124] determining coordinates and element types of equivalent atoms of each non-hydrogen atom of the crystal within a predetermined distance using symmetry and periodicity of the crystal;

[0125] calculating a distance between each equivalent atom to every other equivalent atom, wherein the equivalent atoms include non-hydrogen atoms and the determined equivalent atoms; and

[0126] connecting edges between each pair of equivalent atoms in the equivalent atoms with a distance less than a distance threshold to construct the geometric relationship graph of the equivalent atoms.

[0127] Example 10. The method of example 8, wherein the graph neural network based hydrogenation module is trained by:

[0128] receiving a dataset, the dataset including a geometric relationship graph of equivalent atoms and a true value of a hydrogenation operation type of a non-hydrogen atom of each crystal in a plurality of crystals;

[0129] generate, using the graph neural network based hydrogenation module, a hydrogenation operation type prediction result for each non-hydrogen atom of each crystal in the plurality of crystals based on the dataset;

[0130] calculate a loss function between the hydrogenation operation type prediction result for each non-hydrogen atom of each crystal in the plurality of crystals and a ground truth value; and

[0131] update, based on the loss function, a parameter of the graph neural network based hydrogenation module using a parameter update method to obtain a trained graph neural network based hydrogenation module.

[0132] Example 11. A system for crystal resolution, comprising means for performing the method of any one of examples 1-10.

[0133] Example 12. A computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the method of any one of examples 1-10.

[0134] Example 13. A computer device comprising a memory and a processor, having stored on the memory a computer program capable of running on the processor, wherein the processor implements the method of any one of examples 1-10 when executing the computer program.

[0135] Example 14. A computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the method of any one of examples 1-10.

[0136] The present application can be a system, a method, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present application.

[0137] Computer readable storage media can be tangible storage media which can retain and store instructions for use by an instruction execution device. Computer readable storage media can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer readable storage media include the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0138] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network can comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0139] Computer readable program instructions for carrying out operations of the present application can be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate array (FPGA), or programmable logic array (PLA) can execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present application.

[0140] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0141] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0142] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0143] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0144] Embodiments of the present application have been described above, and the description is intended to be illustrative, and not restrictive, of the disclosed embodiments. Many modifications and variations of the disclosed embodiments are possible in light of the above teachings. It is therefore to be understood that within the scope of the disclosed embodiments, modifications and variations of the disclosed embodiments can be practiced. It is also to be understood that the specific order or hierarchy of steps in the processes disclosed is an illustration of exemplary processes. Based upon the description and illustrations provided herein, those skilled in the art will understand that changes can be made to the order of steps in the processes and that many of the individual steps can be modified or eliminated. Additionally, the description and illustrations provided herein are not meant to limit the scope of the disclosed embodiments. The scope of the disclosed embodiments is limited only by the claims.

Claims

1. A method for crystal resolution, comprising: obtaining an electron density map of a crystal, the electron density map being generated by phase resolution on diffraction data of the crystal; determining coordinates of each non-hydrogen atom of the crystal based on electron density peaks of the electron density map; constructing a geometric relationship graph of non-hydrogen atoms based on the coordinates of each non-hydrogen atom of the crystal; and determining element types of each non-hydrogen atom of the crystal based on the geometric relationship graph of non-hydrogen atoms using a graph neural network based non-hydrogen atom identification module.

2. The method of claim 1, wherein, constructing a geometric relationship graph of non-hydrogen atoms based on the coordinates of each non-hydrogen atom of the crystal comprises: calculating distances between each atom in all non-hydrogen atoms to every other atom; and connecting edges between each pair of atoms in all non-hydrogen atoms with distances less than a distance threshold as nodes to construct the geometric relationship graph of non-hydrogen atoms.

3. The method of claim 1, wherein, the graph neural network based non-hydrogen atom identification module is trained by: receiving a dataset comprising geometric relationship graphs of non-hydrogen atoms and ground truths of element types of non-hydrogen atoms of each crystal in a plurality of crystals; generating element type predictions of each non-hydrogen atom of each crystal in the plurality of crystals based on the dataset using the graph neural network based non-hydrogen atom identification module; calculating a loss function between the element type predictions and the ground truths of each non-hydrogen atom of each crystal in the plurality of crystals; and updating parameters of the graph neural network based non-hydrogen atom identification module based on the loss function using a parameter updating method to obtain a trained graph neural network based non-hydrogen atom identification module.

4. The method of claim 1, further comprising: determining hydrogenation operation types corresponding to each non-hydrogen atom of the crystal; and determining coordinates of each hydrogen atom of the crystal based on the hydrogenation operation types corresponding to each non-hydrogen atom of the crystal. determining hydrogenation operation types corresponding to each non-hydrogen atom of the crystal comprises: constructing a geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein the equivalent atoms contain non-hydrogen atoms and equivalent atoms; and 5. The method of claim 4, wherein, determining the hydrogenation operation types corresponding to each non-hydrogen atom of the crystal based on the geometric relationship graph of equivalent atoms using a graph neural network based hydrogenation module. constructing a geometric relationship graph of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal comprises: determining coordinates and element types of equivalent atoms of each non-hydrogen atom of the crystal within a predetermined distance using symmetry and periodicity of the crystal; 6. The method of claim 5, wherein, calculating distances between each equivalent atom in the equivalent atoms to every other equivalent atom, wherein the equivalent atoms contain non-hydrogen atoms and the determined equivalent atoms; and connecting edges between each pair of equivalent atoms in the equivalent atoms with distances less than a distance threshold as nodes to construct the geometric relationship graph of equivalent atoms. the graph neural network based hydrogenation module is trained by: ​ 7. The method of claim 5, wherein, ​ receiving a dataset comprising a graph of geometric relationships of equivalent atoms and ground truth values of hydrogenation operation types of non-hydrogen atoms for each crystal in a plurality of crystals; generating, using the graph neural network based hydrogenation module, predicted results of hydrogenation operation types of each non-hydrogen atom for each crystal in the plurality of crystals based on the dataset; computing a loss function between the predicted results of hydrogenation operation types of each non-hydrogen atom for each crystal in the plurality of crystals and the ground truth values; and updating, based on the loss function, parameters of the graph neural network based hydrogenation module using a parameter updating method to obtain a trained graph neural network based hydrogenation module.

8. A method for crystal analysis, comprising: obtaining coordinates and element types of each non-hydrogen atom of a crystal; constructing a graph of geometric relationships of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal, wherein the equivalent atoms comprise non-hydrogen atoms and equivalent atoms; determining, using a graph neural network based hydrogenation module, hydrogenation operation types corresponding to each non-hydrogen atom of the crystal based on the graph of geometric relationships of equivalent atoms; and determining coordinates of each hydrogen atom of the crystal based on the hydrogenation operation types corresponding to each non-hydrogen atom of the crystal. constructing a graph of geometric relationships of equivalent atoms based on the coordinates and element types of each non-hydrogen atom of the crystal comprises:

9. The method of claim 8, wherein, determining coordinates and element types of equivalent atoms within a predetermined distance of each non-hydrogen atom of the crystal using symmetry and periodicity of the crystal; computing distances between each of the equivalent atoms to every other equivalent atom, wherein the equivalent atoms comprise non-hydrogen atoms and the determined equivalent atoms; and connecting edges between each pair of equivalent atoms in the equivalent atoms with distances less than a distance threshold to construct the graph of geometric relationships of equivalent atoms. the graph neural network based hydrogenation module is trained by:

10. The method of claim 8, wherein, receiving a dataset comprising a graph of geometric relationships of equivalent atoms and ground truth values of hydrogenation operation types of non-hydrogen atoms for each crystal in a plurality of crystals; generating, using the graph neural network based hydrogenation module, predicted results of hydrogenation operation types of each non-hydrogen atom for each crystal in the plurality of crystals based on the dataset; computing a loss function between the predicted results of hydrogenation operation types of each non-hydrogen atom for each crystal in the plurality of crystals and the ground truth values; and updating, based on the loss function, parameters of the graph neural network based hydrogenation module using a parameter updating method to obtain a trained graph neural network based hydrogenation module.

11. A system for crystal analysis, comprising means for performing the method of any one of claims 1-10. The computer program is executed by a processor to implement the method of any one of claims 1-10. The processor executes the computer program to implement the method of any one of claims 1-10.

12. A computer readable storage medium having stored thereon a computer program, wherein, The computer program is executed by a processor to implement the method of any one of claims 1-10.

13. A computer device comprising a memory and a processor, having stored on the memory a computer program capable of running on the processor, wherein, The computer program is executed by a processor to implement the method of any one of claims 1-10.

14. A computer program product comprising a computer program, wherein, ​

Images