Method and device for training all atom structure prediction model, and electronic device

Abstract

To provide a method for training all atom structure prediction model; an all atom structure prediction method; a training device; an all atom structure prediction device; an electronic device; a storage medium; and a computer program.SOLUTION: A method comprises steps of: acquiring information on the structure of a biomolecule and the first dynamic locus of the biomolecule including information on positions of atoms in the biomolecule at different times; adding noise to the first dynamic locus to acquire the second dynamic locus; encoding the structure information to acquire encoded characteristics; decoding the encoded characteristics and the second dynamic locus to acquire a target dynamic locus; and training an initial all atom structure prediction model on the basis of a difference between the target dynamic locus and the first dynamic locus to acquire all atom structure prediction model.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to the field of artificial intelligence such as deep learning and bio-computing. Specifically, it relates to a method, apparatus, and electronic device for training an all-atom structure prediction model.

Background Art

[0002] With the development of artificial intelligence, using artificial intelligence technology to predict the all-atom structure of biomolecules has received more attention than experimental means. How to use deep learning methods to promote the accurate prediction of the all-atom structure of biomolecules has become increasingly important.

Summary of the Invention

[0003] This application provides a method, apparatus, and electronic device for training an all-atom structure prediction model.

[0004] According to one aspect of this application, obtaining the structural information of a biomolecule and the first dynamic trajectory of the biomolecule, wherein the first dynamic trajectory includes the position information of atoms in the biomolecule at different time points; adding noise to the first dynamic trajectory to obtain a second dynamic trajectory; encoding the structural information to obtain encoded features; decoding the encoded features and the second dynamic trajectory to obtain a target dynamic trajectory; and training an initial all-atom structure prediction model based on the difference between the target dynamic trajectory and the first dynamic trajectory to obtain an all-atom structure prediction model. A method for training an all-atom structure prediction model is provided.

[0005] According to another aspect of this application, obtaining the structural information of a biomolecule; Encoding the structural information by a full-atom structure prediction model to obtain encoded features, wherein the full-atom structure prediction model is obtained by training using the training method of the full-atom structure prediction model described above; Decoding by the full-atom structure prediction model based on the encoded features and dynamic trajectory noise to obtain the dynamic trajectory of the biomolecule, wherein the dynamic trajectory includes position information of atoms in the biomolecule at different time points. A full-atom structure prediction method is provided.

[0006] According to another aspect of the present application, An acquisition module configured to acquire the structural information of a biomolecule and the first dynamic trajectory of the biomolecule, wherein the first dynamic trajectory includes position information of atoms in the biomolecule at different time points; A noise addition module configured to add noise to the first dynamic trajectory to obtain a second dynamic trajectory; An encoding module configured to encode the structural information to obtain encoded features; A decoding module configured to decode the encoded features and the second dynamic trajectory to obtain a target dynamic trajectory; A training module configured to train an initial full-atom structure prediction model based on the difference between the target dynamic trajectory and the first dynamic trajectory to obtain a full-atom structure prediction model. A training apparatus for a full-atom structure prediction model is provided.

[0007] According to another aspect of the present application, An acquisition module configured to acquire the structural information of a biomolecule; An encoding module configured to input the structural information into a full-atom structure prediction model, encode the structural information, and obtain encoded features, wherein the full-atom structure prediction model is obtained by training using the training device of the full-atom structure prediction model described above, and A decoding module configured to decode based on the encoded features and dynamic trajectory noise by the full-atom structure prediction model to obtain the dynamic trajectory of the biomolecule, wherein the dynamic trajectory includes position information of atoms in the biomolecule at different time points. A full-atom structure prediction apparatus is provided that includes the decoding module.

[0008] According to another aspect of the present application, At least one processor, and A memory communicably connected to the at least one processor, Instructions executable by the at least one processor are stored in the memory, and when the instructions are executed by the at least one processor, the at least one processor can execute the method described in the above embodiment. An electronic device is provided.

[0009] According to another aspect of the present application, a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions cause the computer to execute the method described in the above embodiment. A non-transitory computer-readable storage medium is provided.

[0010] According to another aspect of the present application, a computer program, wherein when the computer program is executed by a processor, the steps of the method described in the above embodiment are realized. A computer program is provided.

[0011] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will be more easily understood from the following description.

Brief Description of the Drawings

[0012] The drawings are for better understanding of this application and do not limit this application.

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Modes for Carrying Out the Invention

[0013] Hereinafter, exemplary embodiments of the present application will be described with reference to the drawings. For ease of understanding, various details of the embodiments of the present application are included, and they should be regarded as merely illustrative. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Also, for clarity and conciseness, descriptions of well-known functions and configurations are omitted in the following description.

[0014] Hereinafter, a training method, apparatus, electronic device, and storage medium for an all-atom structure prediction model according to an embodiment of the present application will be described with reference to the drawings.

[0015] FIG. 1 is a schematic flowchart of a training method for an all-atom structure prediction model provided by an embodiment of the present application.

[0016] The training method for the all-atom structure prediction model according to the embodiment of the present application can be executed by the training apparatus for the all-atom structure prediction model according to the embodiment of the present application, and the training apparatus for the all-atom structure prediction model can be configured within an electronic device.

[0017] The electronic device may be any device with computing capabilities such as a personal computer, mobile terminal, server, etc. The mobile terminal may be, for example, a hardware device equipped with various operating systems, touchscreens, and / or displays, such as an in-vehicle device, mobile phone, tablet, personal digital assistant, wearable device, etc.

[0018] As shown in FIG. 1, the training method for the all-atom structure prediction model includes the following steps 101 to 105.

[0019] In step 101, the structural information of the biomolecule and the first dynamic trajectory of the biomolecule are acquired.

[0020] In this application, the biomolecule may be any one of proteins, small molecules, RNA (Ribonucleic Acid), DNA (Deoxyribonucleic Acid), and ions, and the biomolecule may be, but is not limited to, a complex of at least two of proteins, small molecules, RNA, DNA, and ions.

[0021] The structural information of a biomolecule can refer to sequence information, molecular formula, etc. For example, when the biomolecule is one of proteins, RNA, DNA, etc., the structural information can refer to the sequence information. Also, for example, when the biomolecule is a complex of a protein and a molecule, the structural information may include the sequence information of the protein and the sequence information of the DNA. Also, for example, when the biomolecule is a complex of a protein and a small molecule, the structural information may include the sequence information of the protein and the molecular formula of the small molecule.

[0022] The first dynamic trajectory of a biomolecule can include the position information of atoms in the biomolecule at different time points. The position information here can refer to three-dimensional coordinate information. Since the three-dimensional structure of a biomolecule is determined by the positions of the atoms in the biomolecule, the first dynamic trajectory can be used to represent the three-dimensional structure of the biomolecule at different time points.

[0023] For example, the shape of the first dynamic trajectory may be F1*S1*3, where F1 represents the number of frames, S1 represents the number of atoms in the biomolecule, 3 represents the three dimensions in the three-dimensional coordinate information, and one frame is considered to correspond to one time point. The first dynamic trajectory includes the position information of the atoms in the biomolecule at F1 time points, that is, it includes the three-dimensional structure of the biomolecule at F1 time points.

[0024] The structural information of the biomolecule corresponds to the first dynamic trajectory. Taking the sequence information of a protein as an example, the sequence information of a protein refers to the order of amino acids in the protein molecule. The atoms in each frame of the first dynamic trajectory are arranged according to the order of amino acids. That is, according to the order of amino acids, the atoms in each amino acid are sorted sequentially. Note that although the order of atoms at different times does not change, the position information of the atoms may change.

[0025] Exemplarily, the first dynamic trajectory may be based on the three-dimensional structures of biomolecules at different time points obtained in the experimental stage, or may be obtained by other methods, and this is not limited.

[0026] In step 102, noise is added to the first dynamic trajectory to obtain a second dynamic trajectory.

[0027] In this application, noise can be added to the first dynamic trajectory to obtain a second dynamic trajectory, which is the dynamic trajectory with noise added. Exemplarily, noise may be added to the position information of the atoms in the biomolecule at some of the multiple time points, or noise may be added to the position information of some atoms at different time points in the first dynamic trajectory.

[0028] The noise used may be Gaussian noise, random noise, or other types of noise.

[0029] In step 103, the structural information is encoded to obtain the encoded features.

[0030] In this application, the initial all-atom structure prediction model can include an encoder and a decoder. By inputting the structural information of the biomolecule into the encoder for encoding, the encoded features can be obtained.

[0031] In step 104, the encoded features and the second dynamic trajectory are decoded to obtain a target dynamic trajectory.

[0032] In this application, the encoded features and the second dynamic trajectory are input into the decoder of the initial all-atom structure prediction model, and the decoder is used to decode the encoded features and the second dynamic trajectory to obtain the target dynamic trajectory.

[0033] Here, the target dynamic trajectory can include the position information of atoms in the biomolecule at different predicted time points.

[0034] In step 105, based on the difference between the target dynamic trajectory and the first dynamic trajectory, the initial all-atom structure prediction model is trained to obtain the all-atom structure prediction model.

[0035] In this application, based on the difference between the position information of atoms in the biomolecule at each time point in the target dynamic trajectory and the position information of atoms in the biomolecule at the same time point in the first dynamic trajectory, the parameters of the initial all-atom structure prediction model are adjusted to obtain the all-atom structure prediction model with adjusted parameters. If the training end condition is not satisfied, the all-atom structure prediction model with adjusted parameters is continuously trained until the training end condition is satisfied to obtain the all-atom structure prediction model. The all-atom structure prediction model is used to predict the dynamic trajectory of the biomolecule based on the structural information of the biomolecule.

[0036] In addition, by training using the position information of atoms in the biomolecule at one time point, an all-atom structure prediction model capable of predicting the position information of atoms in the biomolecule at a specific time point can be obtained, and an all-atom structure prediction model for predicting the static conformation of the biomolecule can also be obtained.

[0037] In an embodiment of the present application, the structural information of a biomolecule is encoded to obtain the encoded features, and the encoded features and a second dynamic trajectory obtained by adding noise to the first dynamic trajectory are decoded to obtain a target dynamic trajectory. Next, an all-atom structure prediction model is obtained by training based on the difference between the target dynamic trajectory and the first dynamic trajectory. In this way, by training based on the structural information of the biomolecule and the first dynamic trajectory of the biomolecule, an all-atom structure prediction model capable of predicting the dynamic trajectory of the biomolecule is obtained, the prediction of the dynamic trajectory of the biomolecule is realized, and the generalization ability of the model can be improved.

[0038] FIG. 2 is a schematic flowchart of a method for training an all-atom structure prediction model provided by another embodiment of the present application.

[0039] As shown in FIG. 2, the method for training the all-atom structure prediction model includes the following steps 201 to 206.

[0040] In step 201, the structural information of the biomolecule and the first dynamic trajectory of the biomolecule are obtained.

[0041] In the present application, step 201 can adopt any one of the implementation forms in each embodiment of the present application, so the description will be omitted here again.

[0042] In step 202, noise is added to the first dynamic trajectory to obtain a second dynamic trajectory.

[0043] In the present application, step 202 can adopt any one of the implementation forms in each embodiment of the present application, so the description will be omitted here again.

[0044] In step 203, the structural information is encoded to obtain the encoded features.

[0045] In this application, since step 203 can adopt any one of the implementation forms in each embodiment of this application, the description will be omitted here again.

[0046] In step 204, the second dynamic trajectory is subjected to dimensionality reduction processing to obtain a third dynamic trajectory.

[0047] In this application, by fusing the position information of a plurality of adjacent atoms in the second dynamic trajectory, the second dynamic trajectory can be dimensionally reduced to obtain a third dynamic trajectory.

[0048] Exemplarily, the second dynamic trajectory can be block-divided to obtain a third dynamic trajectory. The third dynamic trajectory can include a plurality of first sub-block trajectories, and each first sub-block trajectory can be regarded as one element of the third dynamic trajectory. Each first sub-block trajectory can be obtained by fusing based on the position information of a plurality of adjacent atoms in the second dynamic trajectory. The fusion here may be calculating an average value, or calculating a weighted average value, or randomly selecting the position information of any one atom from the position information of a plurality of adjacent atoms, etc.

[0049] For example, the shape of the dynamic trajectory can be represented as F*S*3, where F represents the number of frames, S represents the number of atoms in the biomolecule, and 3 represents the three dimensions in the three-dimensional coordinate information. When the number of frames of the first dynamic trajectory is 4 and the number of atoms is 6, it can be regarded as a 4*6 matrix, and the matrix can be divided into small 2*2 blocks to obtain a 2*3 matrix. That is, the number of frames of the second dynamic trajectory is 2, the number of atoms is 3, and there are a total of 6 first sub-block trajectories. Thus, the shape of the first dynamic trajectory is dimensionally reduced from 4*6*3 to 2*3*3.

[0050] In step 205, the encoded feature and the third dynamic trajectory are decoded to obtain a target dynamic trajectory.

[0051] In this application, the encoded features and the third dynamic trajectory can be decoded to obtain an intermediate dynamic trajectory, and the intermediate dynamic trajectory can be interpolated to obtain a target dynamic trajectory.

[0052] Exemplarily, when the decoder of the model processes one-dimensional data, after converting the third dynamic trajectory into a one-dimensional vector, it is input into the decoder together with the encoded features for decoding to obtain a new one-dimensional vector. Then, the new one-dimensional vectors are rearranged to obtain an intermediate dynamic trajectory, and then the intermediate dynamic trajectory is interpolated to obtain a target dynamic trajectory.

[0053] Exemplarily, when the first dynamic trajectory includes a plurality of first sub-block trajectories, the plurality of first sub-block trajectories are sorted in a preset order to obtain a first trajectory sequence. Then, the encoded features and the first trajectory sequence are decoded to obtain a target dynamic trajectory. In this way, by sorting the plurality of first sub-block trajectories in a preset order and expanding the third dynamic trajectory into a one-dimensional first trajectory sequence before decoding, the processing needs of the model can be satisfied.

[0054] The preset order may be an order from left to right and an order from top to bottom, and the length of the first trajectory sequence is the same as the number of first sub-block trajectories.

[0055] Exemplarily, when decoding the encoded features and the first trajectory sequence, the encoded features and the first trajectory sequence are decoded to obtain a second trajectory sequence. The plurality of second sub-block trajectories in the second trajectory sequence are rearranged to obtain a fourth dynamic trajectory, and then the fourth dynamic trajectory is interpolated to obtain a target dynamic trajectory.

[0056] The second trajectory sequence is one-dimensional data, and the length of the second trajectory sequence is the same as the number of second sub-block trajectories.

[0057] In this way, by decoding the encoded features and the first trajectory sequence, a one-dimensional second trajectory sequence is obtained, and then, by sorting, a fourth dynamic trajectory is obtained. Subsequently, by performing interpolation processing, a target dynamic trajectory is obtained, and the computing amount of the model can be reduced while improving the accuracy of dynamic trajectory prediction.

[0058] In step 206, based on the difference between the target dynamic trajectory and the first dynamic trajectory, an initial all-atom structure prediction model is trained to obtain an all-atom structure prediction model.

[0059] Exemplarily, based on the difference between the position information of an atom at any one time point in the target dynamic trajectory and the position information of the same atom at any one time point in the first dynamic trajectory, a first loss corresponding to any one time point is determined. Based on the sum of the first losses corresponding to different time points, a second loss is obtained. Based on the second loss, the initial all-atom structure prediction model is trained to obtain an all-atom structure prediction model.

[0060] Here, the first loss corresponding to any one time point may be the sum of the sub-losses corresponding to each atom, and the sub-loss corresponding to each atom may be determined based on the difference between the position information of each atom in the target dynamic trajectory and the position information of the same atom in the first dynamic trajectory.

[0061] In this way, by training the initial all-atom structure prediction model based on the difference between the position information of each atom at each time point in the target dynamic trajectory and the position information of the same atom at each time point in the first dynamic trajectory to obtain an all-atom structure prediction model, the accuracy of the model is improved.

[0062] In the embodiments of the present application, the second dynamic trajectory is subjected to dimensionality reduction processing to obtain a third dynamic trajectory, and the encoded features and the third dynamic trajectory are decoded to obtain a target dynamic trajectory. Thereby, since the second dynamic trajectory is dimensionally reduced and then decoded, the computing amount can be reduced, and the processing speed of the model can be improved.

[0063] FIG. 3 is a schematic flowchart of a method for training an all-atom structure prediction model provided by another embodiment of the present application.

[0064] As shown in FIG. 3, the method for training the all-atom structure prediction model includes the following steps 301 to 307.

[0065] In step 301, the structural information of the biomolecule is obtained.

[0066] In the present application, since step 301 can adopt any one of the implementation forms in each embodiment of the present application, the description will not be repeated here.

[0067] In step 302, based on the structural information of the biomolecule, the interaction energy between biomolecules is calculated to generate a plurality of static conformations of the biomolecule.

[0068] In the present application, physical-based docking software is used to calculate the interaction energy between molecules based on the structural information of the biomolecule, thereby generating a plurality of static conformations of the biomolecule. Since these static conformations cover different binding modes and orientations, the data diversity is very rich.

[0069] For example, when the biomolecule is a complex of a protein and a small molecule, docking software is used to calculate the interaction energy between molecules based on the sequence information of the protein and the small molecule information, thereby predicting and generating a plurality of static conformations of the complex of the protein and the small molecule.

[0070] In step 303, a plurality of static conformations are simulated to capture the first dynamic trajectory of the biomolecule.

[0071] In the present application, molecular dynamics simulation software can be used to simulate a plurality of static conformations for a long time to capture the first dynamic trajectory of the biomolecule. Here, the first dynamic trajectory not only shows the flexibility of the biomolecule, but also provides rich dynamic interaction information, which helps the model learn the more realistic movement of the biomolecule.

[0072] For example, a plurality of static conformations of a complex of a plurality of biomolecules can be simulated for a long time to capture the dynamic movement of the complex of the biomolecules in the solvent environment, such as structural changes and dynamic trajectories.

[0073] In step 304, noise is added to the first dynamic trajectory to obtain the second dynamic trajectory.

[0074] In the present application, step 304 can adopt any one of the implementation forms in each embodiment of the present application, so the description is omitted here again.

[0075] In step 305, the structural information is encoded to obtain the encoded features.

[0076] In the present application, step 305 can adopt any one of the implementation forms in each embodiment of the present application, so the description is omitted here again.

[0077] In step 306, the encoded features and the second dynamic trajectory are decoded to obtain the target dynamic trajectory.

[0078] In the present application, step 306 can adopt any one of the implementation forms in each embodiment of the present application, so the description is omitted here again.

[0079] Exemplarily, the position information of each atom in the second dynamic trajectory is sorted in a preset order to obtain a third trajectory sequence, the encoded features and the third trajectory sequence are decoded to obtain a fourth trajectory sequence, and the position information of the atoms in the fourth trajectory sequence is rearranged to obtain a target dynamic trajectory. Thereby, since the second dynamic trajectory is converted into a one-dimensional third trajectory sequence and then decoded, the processing needs of the model can be satisfied.

[0080] In step 307, based on the difference between the target dynamic trajectory and the first dynamic trajectory, an initial all-atom structure prediction model is trained to obtain an all-atom structure prediction model.

[0081] In this application, since step 307 can adopt any one of the implementation forms in each embodiment of this application, the description is omitted here again.

[0082] In the embodiments of this application, by calculating the interaction energy between biomolecules, a plurality of static conformations of the biomolecules are generated. Since the static conformations cover different binding modes and orientations, the data diversity becomes very rich. The first dynamic trajectory obtained based on the plurality of static conformations not only shows the flexibility of the biomolecules, but also provides rich dynamic interaction information, which helps the model to learn more realistic movements of the biomolecules. Thus, by training the all-atom structure prediction model based on the first dynamic trajectory training, the model can capture the time-dependence and complex interactions from the first dynamic trajectory, improve the accuracy of structure prediction, and achieve high-precision all-atom structure prediction.

[0083] For ease of understanding, the following will be described with reference to FIG. 4. FIG. 4 is a schematic diagram of the process by which the all-atom structure prediction model provided by the embodiment of this application predicts a dynamic trajectory.

[0084] As shown in FIG. 4, the all-atom structure prediction model includes an encoder and a decoder. Taking the complex of biomolecules such as proteins, DNA, and RNA as an example, the encoder is used to encode sequence information such as proteins, DNA, and RNA to obtain encoded features, and the encoded features are input into the decoder.

[0085] In addition, noise is added to the dynamic trajectory sample of the complex to obtain a noise-added dynamic trajectory. By block-splitting the noise-added dynamic trajectory of the complex, dimensionality reduction of the noise-added dynamic trajectory is realized. Then, it is expanded into a one-dimensional vector and input into the decoder. The decoder decodes the encoded features and the one-dimensional vector to obtain a new one-dimensional vector. After rearranging the new one-dimensional vectors and performing interpolation, a predicted dynamic trajectory is obtained.

[0086] Here, the shape of the noise-added dynamic trajectory can be represented as the number of frames * the number of atoms * 3, where 3 represents the three dimensions in three-dimensional coordinate information. In FIG. 4, each block of the noise-added dynamic trajectory can represent the position information of one atom, and the sequence information corresponds to the dynamic trajectory. For example, the sequence information of a protein refers to the order of amino acids in the protein molecule, and the atoms in each frame of the dynamic trajectory are arranged according to the order of amino acids. In the case of a complex, after sorting the atoms of one biomolecule is completed, the atoms of another biomolecule can be sorted, and the order of biomolecules is not limited.

[0087] In the training stage of the model, based on the difference between the predicted dynamic trajectory and the dynamic trajectory sample, the initial all-atom structure prediction model can be trained to obtain the all-atom structure prediction model.

[0088] To implement the above embodiments, the embodiments of the present application further provide an all-atom structure prediction method. FIG. 5 is a schematic flowchart of the all-atom structure prediction method provided by an embodiment of the present application.

[0089] As shown in FIG. 5, the all-atom structure prediction method includes the following steps 501 to 503.

[0090] In step 501, the structural information of the biomolecule is obtained.

[0091] In this application, since the description of the structural information of the biomolecule can refer to the above embodiments, the repeated description is omitted here.

[0092] In step 502, the structural information is encoded by the all-atom structure prediction model to obtain the encoded features.

[0093] The all-atom structure prediction model may be obtained by training using the training method described in the above embodiments.

[0094] The all-atom structure prediction model can include an encoder and a decoder. By inputting the structural information into the encoder, the encoder can encode the structural information of the biomolecule to obtain the encoded features.

[0095] In step 503, the all-atom structure prediction model decodes based on the encoded features and the dynamic trajectory noise to obtain the dynamic trajectory of the biomolecule.

[0096] The dynamic trajectory noise can include noises at different time points, for example, noises of a plurality of frames.

[0097] In this application, the encoded features and the dynamic trajectory noise are input into the decoder for decoding, and the noise is sequentially removed by decoding to obtain the dynamic trajectory of the biomolecule. The dynamic trajectory can include the position information of the atoms in the biomolecule at different time points.

[0098] In the embodiments of the present application, by using the all-atom structure prediction model obtained by training based on the training method described in the above embodiments, the dynamic trajectory of a biomolecule can be predicted, and the prediction accuracy can be improved.

[0099] To implement the above embodiments, the embodiments of the present application further provide a training apparatus for an all-atom structure prediction model. FIG. 6 is a schematic configuration diagram of a training apparatus for an all-atom structure prediction model provided by an embodiment of the present application.

[0100] As shown in FIG. 6, the training apparatus 600 for the all-atom structure prediction model includes an acquisition module 610 configured to acquire the structure information of a biomolecule and the first dynamic trajectory of the biomolecule, where the first dynamic trajectory includes the position information of atoms in the biomolecule at different time points, a noise addition module 620 configured to add noise to the first dynamic trajectory to obtain a second dynamic trajectory, an encoding module 630 configured to encode the structure information to obtain encoded features, a decoding module 640 configured to decode the encoded features and the second dynamic trajectory to obtain a target dynamic trajectory, and a training module 650 configured to train an initial all-atom structure prediction model based on the difference between the target dynamic trajectory and the first dynamic trajectory to obtain an all-atom structure prediction model.

[0101] Optionally, the decoding module 640 performs dimensionality reduction processing on the second dynamic trajectory to obtain a third dynamic trajectory, and is configured to decode the encoded features and the third dynamic trajectory to obtain the target dynamic trajectory.

[0102] Optionally, the decoding module 640 Sort the plurality of first sub-block trajectories in a preset order to obtain a first trajectory sequence, and configured to decode the encoded feature and the first trajectory sequence to obtain the target dynamic trajectory.

[0103] Optionally, the decoding module 640 decodes the encoded feature and the first trajectory sequence to obtain a second trajectory sequence, rearranges a plurality of second sub-block trajectories in the second trajectory sequence to obtain a fourth dynamic trajectory, and is configured to interpolate the fourth dynamic trajectory to obtain the target dynamic trajectory.

[0104] Optionally, the acquisition module 610 generates a plurality of static conformations of the biomolecule by calculating the interaction energy between the biomolecules based on the structural information of the biomolecule, and is configured to simulate the plurality of static conformations to capture a first dynamic trajectory of the biomolecule.

[0105] Optionally, the decoding module 640 sorts the position information of each atom in the second dynamic trajectory in a preset order to obtain a third trajectory sequence, decodes the encoded feature and the third trajectory sequence to obtain a fourth trajectory sequence, rearranges the position information of the atoms in the fourth trajectory sequence to obtain the target dynamic trajectory.

[0106] Optionally, the training module 650 Based on the difference between the position information of any one atom at a certain time point in the target dynamic trajectory and the position information of the same atom at the same time point in the first dynamic trajectory, a first loss corresponding to the any one time point is determined. Based on the first losses corresponding to different time points, a second loss is obtained. Based on the second loss, the initial all-atom structure prediction model is trained to obtain the all-atom structure prediction model.

[0107] It should be noted that the interpretation and description regarding the embodiments of the training method of the all-atom structure prediction model described above are also applicable to the training device of the all-atom structure prediction model according to the embodiments, so the repeated description is omitted here.

[0108] In the embodiments of the present application, the structural information of a biomolecule is encoded to obtain the encoded features, and based on the encoded features and the second dynamic trajectory obtained by adding noise to the first dynamic trajectory, decoding is performed to obtain the target dynamic trajectory. Based on the difference between the target dynamic trajectory and the first dynamic trajectory, an all-atom structure prediction model is obtained by training. In this way, by training based on the structural information of the biomolecule and the first dynamic trajectory of the biomolecule, an all-atom structure prediction model capable of predicting the dynamic trajectory of the biomolecule is obtained, so that the prediction of the dynamic trajectory of the biomolecule is realized and the generalization ability of the model is improved.

[0109] To implement the above embodiments, the embodiments of the present application further provide an all-atom structure prediction device. FIG. 7 is a schematic configuration diagram of an all-atom structure prediction device provided by an embodiment of the present application.

[0110] As shown in FIG. 7, the all-atom structure prediction device 700 An acquisition module 710 configured to acquire the structural information of a biomolecule An encoding module 720 configured to encode the structural information by means of a full-atom structure prediction model to obtain encoded features, wherein the full-atom structure prediction model is the encoding module 720 obtained by training using the device described in the above embodiments, and A decoding module 730 configured to decode based on the encoded features and dynamic trajectory noise by means of the full-atom structure prediction model to obtain the dynamic trajectory of the biomolecule, wherein the dynamic trajectory includes position information of atoms in the biomolecule at different time points.

[0111] It should be noted that the interpretation and explanation regarding the embodiments of the above-described full-atom structure prediction method are also applicable to the full-atom structure prediction device according to this embodiment, and thus the repeated explanation is omitted here.

[0112] In the embodiments of the present application, by using the full-atom structure prediction model obtained by training based on the training device described in the above embodiments, the dynamic trajectory of a biomolecule can be predicted, and the prediction accuracy can be improved.

[0113] According to the embodiments of the present application, the present application further provides an electronic device, a readable storage medium, and a computer program.

[0114] FIG. 8 shows a schematic block diagram of an exemplary electronic device 800 according to an embodiment of the present application. The electronic device is intended to represent various forms of digital computers such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices such as personal digital assistants, mobile phones, smart phones, wearable devices, and other similar computing devices. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0115] As shown in FIG. 8, the device 800 includes a computing unit 801, and the computing unit 801 can execute various suitable operations and processes based on a computer program stored in a ROM (Read-Only Memory) 802 or a computer program loaded from a storage unit 808 into a RAM (Random Access Memory) 803. The RAM 803 can also store various programs and data necessary for the operation of the device 800. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An I / O (Input / Output) interface 805 is also connected to the bus 804.

[0116] A plurality of components in device 800 are connected to I / O interface 805, including input unit 806 such as a keyboard and a mouse, output unit 807 such as various displays and speakers, storage unit 808 such as a magnetic disk and an optical disk, and communication unit 809 such as a network card, a modem, and a wireless communication transceiver. Communication unit 809 enables device 800 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0117] The computing unit 801 may be various general-purpose and / or dedicated processing components having processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units that execute machine learning model algorithms, a DSP (Digital Signal Processor), and any suitable processor, controller, microcontroller, etc. The computing unit 801 executes each of the above-described methods and processes, such as the training method of the all-atom structure prediction model. For example, in some embodiments, the training method of the all-atom structure prediction model can be realized as a computer software program tangibly included in a machine-readable medium such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed into the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the above-described training method of the all-atom structure prediction model can be executed. Optionally, in other embodiments, the computing unit 801 may be configured to execute the training method of the all-atom structure prediction model in any other suitable manner (e.g., by firmware).

[0118] Each embodiment of the systems and techniques described in this specification can be implemented in digital electronic circuitry, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can be implemented in one or more computer programs, which can be executed and / or interpreted in a programmable system including at least one programmable processor, where the programmable processor can be a dedicated or general-purpose programmable processor, and which receives data and instructions from a storage system, at least one input device, and at least one output device, and can transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0119] The program code for implementing the method of the present application can be described using any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a dedicated computer, or other programmable data processing devices, and when the program code is executed by the processor or controller, the functions / operations defined in the flowchart and / or block diagram will be executed. The program code can be fully executed on a machine, partially executed on a machine, partially executed on a machine as a stand-alone software package and partially executed on a remote machine, or fully executed on a remote machine or server.

[0120] In the description of the present application, the machine-readable medium may be a tangible medium, including or capable of storing a program used by or in combination with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination of the above. More specific examples of the machine-readable storage medium include electrical connections based on one or more lines, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory), or flash memory, optical fibers, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0121] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user, a keyboard, and a pointing device (e.g., a mouse or trackball), and the user can provide input to the computer via the keyboard and the pointing device. Other types of devices can further provide interaction with the user, for example, the feedback provided to the user can be any form of sensing feedback (e.g., visual feedback, auditory feedback, or tactile feedback), and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0122] The systems and techniques described herein can be implemented in a computing system that includes back-end components (such as, for example, a data server), or a computing system that includes middleware components (such as, for example, an application server), or a computing system that includes front-end components (such as, for example, a user computer having a graphical user interface or a web browser, wherein the user interacts with embodiments of the systems and techniques described herein via the graphical user interface or the web browser), or a computing system that includes any combination of such back-end components, middleware components, and front-end components. The components of the system can be interconnected to each other via digital data communication in any form or medium (such as, for example, a communication network). Examples of communication networks include a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, and a blockchain network.

[0123] A computer system can include a client side and a server. The client side and the server are generally separated from each other and typically interact via a communication network. A client-side / server relationship is created by computer programs that are executed on corresponding computers and have a client-side / server relationship with each other. The server can be a cloud server, also referred to as a cloud computing server or a cloud host, which is one of the host products of a cloud computing service system and solves the drawbacks of difficult management and weak business scalability in conventional physical hosts and VPS (Virtual Private Server) services. The server can also be a server of a distributed system or a server combined with a blockchain.

[0124] Note that the above electronic device can implement the all-atom structure prediction method described in the above embodiments.

[0125] According to the embodiments of the present application, the present application further provides a computer program, and when the computer program is executed by a processor, the training method of the all-atom structure prediction model provided by the above embodiments of the present application, or the all-atom structure prediction method is executed.

[0126] It should be understood that the above various forms of flows can be used to rearrange, add, or delete steps. For example, each step described in the present application may be executed in parallel, sequentially, or in a different order, but as long as the expected results of the technical solutions disclosed in the present application can be achieved, it is not limited in this specification.

[0127] The above-described specific implementation manners do not limit the protection scope of the present application. Those skilled in the art can be aware that various modifications, combinations, sub-combinations, and substitutions can be made based on design requirements and other factors. Any changes, equivalent substitutions, and improvements made within the spirit and principles of the present application should all be included within the protection scope of the present application.

Claims

1. A method for training a full-atom structure prediction model, comprising: obtaining the structural information of a biomolecule and the first dynamic trajectory of the biomolecule, wherein the first dynamic trajectory includes the position information of atoms in the biomolecule at different time points; adding noise to the first dynamic trajectory to obtain a second dynamic trajectory; encoding the structural information to obtain encoded features; decoding the encoded features and the second dynamic trajectory to obtain a target dynamic trajectory; training an initial full-atom structure prediction model based on the difference between the target dynamic trajectory and the first dynamic trajectory to obtain a full-atom structure prediction model. A method for training a full-atom structure prediction model, comprising the above steps.

2. The step of decoding the encoded features and the second dynamic trajectory to obtain a target dynamic trajectory includes: performing dimensionality reduction processing on the second dynamic trajectory to obtain a third dynamic trajectory; decoding the encoded features and the third dynamic trajectory to obtain the target dynamic trajectory. The method for training a full-atom structure prediction model according to Claim 1, comprising the above steps.

3. The third dynamic trajectory includes a plurality of first sub-block trajectories, and the step of decoding the encoded features and the third dynamic trajectory to obtain the target dynamic trajectory includes: sorting the plurality of first sub-block trajectories in a preset order to obtain a first trajectory sequence; decoding the encoded features and the first trajectory sequence to obtain the target dynamic trajectory. The method for training a full-atom structure prediction model according to Claim 2, comprising the above steps.

4. The step of decoding the encoded features and the first trajectory sequence to obtain the target dynamic trajectory includes: decoding the encoded features and the first trajectory sequence to obtain a second trajectory sequence; rearranging a plurality of second sub-block trajectories in the second trajectory sequence to obtain a fourth dynamic trajectory; performing interpolation processing on the fourth dynamic trajectory to obtain the target dynamic trajectory. The method for training a full-atom structure prediction model according to Claim 3, comprising the above steps.

5. The step of obtaining the first dynamic trajectory of the biomolecule is: Based on the structural information of the biomolecule, generating a plurality of static conformations of the biomolecule by calculating the interaction energy between the biomolecules; Simulating the plurality of static conformations to capture the first dynamic trajectory of the biomolecule; The method for training an all-atom structure prediction model according to claim 1, comprising the above steps.

6. The step of decoding the encoded features and the second dynamic trajectory to obtain a target dynamic trajectory is: Sorting the position information of each atom in the second dynamic trajectory in a preset order to obtain a third trajectory sequence; Decoding the encoded features and the third trajectory sequence to obtain a fourth trajectory sequence; Rearranging the position information of the atoms in the fourth trajectory sequence to obtain the target dynamic trajectory; The method for training an all-atom structure prediction model according to claim 1, comprising the above steps.

7. The step of training an initial all-atom structure prediction model based on the difference between the target dynamic trajectory and the first dynamic trajectory to obtain an all-atom structure prediction model is: Determining a first loss corresponding to any one time point based on the difference between the position information of an atom at any one time point in the target dynamic trajectory and the position information of the same atom at the same time point in the first dynamic trajectory; Obtaining a second loss based on the first losses corresponding to different time points; Training the initial all-atom structure prediction model based on the second loss to obtain the all-atom structure prediction model; The method for training an all-atom structure prediction model according to claim 1, comprising the above steps.

8. An all-atom structure prediction method, comprising: Obtaining the structural information of a biomolecule; Encoding the structural information by an all-atom structure prediction model to obtain encoded features, wherein the all-atom structure prediction model is obtained by training using the method for training an all-atom structure prediction model according to claim 1. A step of decoding based on the encoded features and the dynamic trajectory noise by the all-atom structure prediction model to obtain the dynamic trajectory of the biomolecule, wherein the dynamic trajectory includes position information of atoms in the biomolecule at different time points; An all-atom structure prediction method including the above.

9. An apparatus for training an all-atom structure prediction model, comprising: An acquisition module configured to acquire the structure information of a biomolecule and the first dynamic trajectory of the biomolecule, wherein the first dynamic trajectory includes position information of atoms in the biomolecule at different time points; A noise addition module configured to add noise to the first dynamic trajectory to obtain a second dynamic trajectory; An encoding module configured to encode the structure information to obtain encoded features; A decoding module configured to decode the encoded features and the second dynamic trajectory to obtain a target dynamic trajectory; A training module configured to train an initial all-atom structure prediction model based on the difference between the target dynamic trajectory and the first dynamic trajectory to obtain an all-atom structure prediction model; An apparatus for training an all-atom structure prediction model including the above.

10. The decoding module: Performs dimensionality reduction processing on the second dynamic trajectory to obtain a third dynamic trajectory; The apparatus for training an all-atom structure prediction model according to Claim 9, wherein the apparatus is configured to decode the encoded features and the third dynamic trajectory to obtain the target dynamic trajectory.

11. The decoding module: Sorts the plurality of first sub-block trajectories in a preset order to obtain a first trajectory sequence; The apparatus for training an all-atom structure prediction model according to Claim 10, wherein the apparatus is configured to decode the encoded features and the first trajectory sequence to obtain the target dynamic trajectory.

12. The decoding module: Decodes the encoded features and the first trajectory sequence to obtain a second trajectory sequence; Rearranges a plurality of second sub-block trajectories in the second trajectory sequence to obtain a fourth dynamic trajectory. The training device for the all-atom structure prediction model according to claim 11, configured to interpolate the fourth dynamic trajectory to obtain the target dynamic trajectory.

13. The acquisition module generates a plurality of static conformations of the biomolecule by calculating the interaction energy between the biomolecules based on the structure information of the biomolecule, The training device for the all-atom structure prediction model according to claim 9, configured to simulate the plurality of static conformations to capture the first dynamic trajectory of the biomolecule.

14. The decoding module sorts the position information of each atom in the second dynamic trajectory in a preset order to obtain a third trajectory sequence, decodes the encoded feature and the third trajectory sequence to obtain a fourth trajectory sequence, The training device for the all-atom structure prediction model according to claim 9, configured to rearrange the position information of the atoms in the fourth trajectory sequence to obtain the target dynamic trajectory.

15. The training module determines a first loss corresponding to any one time point based on the difference between the position information of any one time point atom in the target dynamic trajectory and the position information of the same atom at the same one time point in the first dynamic trajectory, obtains a second loss based on the first losses corresponding to different time points, The training device for the all-atom structure prediction model according to any one of claims 9 to 14, configured to train the initial all-atom structure prediction model based on the second loss to obtain the all-atom structure prediction model.

16. An all-atom structure prediction device, comprising an acquisition module configured to acquire the structure information of the biomolecule, an encoding module configured to encode the structure information by an all-atom structure prediction model to obtain an encoded feature, wherein the all-atom structure prediction model is obtained by training using the training device for the all-atom structure prediction model according to any one of claims 9 to 14 A decoding module configured to decode based on the encoded features and dynamic trajectory noise by the all-atom structure prediction model to obtain the dynamic trajectory of the biomolecule, wherein the dynamic trajectory includes position information of atoms in the biomolecule at different time points. An all-atom structure prediction device including the above. **Claim 17** An electronic device, including at least one processor, and a memory communicably connected to the at least one processor, wherein instructions executable by the at least one processor are stored in the memory, and when the instructions are executed by the at least one processor, the at least one processor is caused to execute the training method of the all-atom structure prediction model according to any one of claims 1 to 7, or the all-atom structure prediction method according to claim 8. An electronic device. **Claim 18** A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions cause a computer to execute the training method of the all-atom structure prediction model according to any one of claims 1 to 7, or the all-atom structure prediction method according to claim 8. A non-transitory computer-readable storage medium. **Claim 19** A computer program, wherein when the computer program is executed by a processor, the steps of the training method of the all-atom structure prediction model according to any one of claims 1 to 7, or the steps of the all-atom structure prediction method according to claim 8 are realized. A computer program.

Images