Complex structure prediction method, model training method and related device

Abstract

A method and apparatus for predicting the structure of a complex are provided. [Solution] A method obtains a biomolecular sequence combination. The biomolecular sequence combination is obtained by performing sequence combination based on a specified plurality of biomolecular sequences. The method also predicts a first probability distribution for the biomolecular sequence combination. The first probability distribution is used to indicate a first probability of a plurality of candidate structural unit groups in the biomolecular sequence combination, where the candidate structural unit groups include structural units in at least two biomolecular sequences, and the first probability is used to indicate the likelihood of an interaction between each structural unit in the candidate structural unit groups. The method further determines at least one first structural unit group from the plurality of candidate structural unit groups based on the first probability distribution, and predicts a target structure of the biomolecular complex based on the interacting structural units in each first structural unit group.

Description

[Technical Field]

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to technical fields such as deep learning and biocomputing, and in particular to a method for predicting the structure of a complex, a model training method, and related devices. [Background technology]

[0002] With the rapid development of computer technology and bioinformatics, biomolecular complexes play an irreplaceable role as fundamental units that perform essential biological functions. In biology, multiple biomolecular sequences (e.g., proteins and nucleic acids) form complexes through precise and dynamic interactions to realize core life activities such as signal transduction, gene regulation, and metabolic catalysis. However, due to the high diversity of biomolecular sequences and the complexity of their interactions, accurately predicting and designing the three-dimensional structures of biomolecular complexes remains one of the current challenges in the fields of computational biology and structural biology. Summary of the Invention

[0003] The present disclosure provides a method for predicting the structure of a complex, a model training method, and related devices.

[0004] According to one aspect of the present disclosure, there is provided a method for predicting a structure of a complex, the method comprising: a step of acquiring a combination of biomolecular sequences, the combination of biomolecular sequences being acquired by combining sequences based on a plurality of designated biomolecular sequences; predicting a first probability distribution for the biomolecular sequence combination, wherein the first probability distribution is used to indicate a first probability of a plurality of candidate structural units in the biomolecular sequence combination, the candidate structural units including at least two structural units in the biomolecular sequences, and the first probability is used to indicate a likelihood of interaction between each structural unit in the candidate structural units; determining at least one first group of structural units from among the plurality of candidate groups of structural units based on the first probability distribution; and predicting a target structure of the biomolecular complex based on the interacting structural units in each of the first structural unit groups.

[0005] According to another aspect of the present disclosure, there is provided a method for training an all-atom structure prediction model, the method comprising: obtaining a training sample, the training sample including a combination of sample biomolecular sequences, the combination of sample biomolecular sequences being obtained based on a combination of a plurality of sample biomolecular sequences; a step of predicting a second probability distribution of the combination of the sample biomolecular sequences using a structure prediction model, and generating a predicted structure of a biomolecular complex based on the second probability distribution, wherein the second probability distribution is used to indicate second probabilities of a plurality of groups of structural units to be selected in the combination of the sample biomolecular sequences, the groups of structural units to be selected including structural units in at least two of the sample biomolecular sequences, and the second probability is used to indicate the possibility of interaction between each structural unit in the groups of structural units to be selected; training the structure prediction model based on differences between the predicted structures and annotation structures corresponding to the plurality of sample biomolecular sequences.

[0006] According to another aspect of the present disclosure, there is provided an apparatus for predicting a structure of a complex, the apparatus comprising: an acquisition module for acquiring a combination of biomolecular sequences, the combination of biomolecular sequences being acquired by combining sequences based on a plurality of designated biomolecular sequences; a prediction module that predicts a first probability distribution of the combination of biomolecular sequences, the probability distribution being used to indicate a first probability of a group of multiple candidate structural units in the combination of biomolecular sequences, the group of candidate structural units including structural units in at least two of the biomolecular sequences, and the probability being used to indicate a likelihood of interaction between each structural unit in the group of candidate structural units; a determination module for determining at least one first group of structural units from among the plurality of candidate groups of structural units based on the first probability distribution; and a generation module for predicting a target structure of a biomolecular complex based on the interacting structural units in each of the first structural unit groups.

[0007] According to another aspect of the present disclosure, there is provided an apparatus for training a structural prediction model, the apparatus comprising: an acquisition module for acquiring training samples, the training samples including a combination of sample biomolecular sequences, the combination of sample biomolecular sequences being acquired based on a combination of a plurality of sample biomolecular sequences; a prediction module for predicting a second probability distribution of the combination of the sample biomolecular sequences using an all-atom structure prediction model, and generating a predicted structure of a biomolecular complex based on the second probability distribution, wherein the second probability distribution is used to indicate second probabilities of a plurality of groups of structural units to be selected in the combination of the sample biomolecular sequences, the groups of structural units to be selected including structural units in at least two of the sample biomolecular sequences, and the second probability is used to indicate the possibility of interaction between each structural unit in the groups of structural units to be selected; a training module for training the structure prediction model based on differences between the predicted structures and annotation structures corresponding to the plurality of sample biomolecular sequences.

[0008] According to another aspect of the present disclosure, there is provided an electronic device, comprising: at least one processor; a memory communicatively coupled to the at least one processor; The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform a method provided by one aspect of the present disclosure above or a method provided by another aspect of the present disclosure above.

[0009] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions causing the computer to perform a method as set forth in one aspect of the present disclosure above or a method as set forth in another aspect of the present disclosure above.

[0010] According to another aspect of the present disclosure, there is provided a computer program, which, when executed by a processor, implements a method as set forth in one of the above aspects of the present disclosure or a method as set forth in another of the above aspects of the present disclosure.

[0011] It should be understood that the contents described in this section are not intended to identify key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will be readily apparent from the following description. [Brief explanation of the drawings]

[0012] The drawings are used for better understanding of the present technical solution and are not intended to limit the present disclosure. [Figure 1] 1 is a schematic flowchart of a method for predicting the structure of a complex provided by Example 1 of the present disclosure. [Figure 2] 1 is a schematic flowchart of a method for predicting the structure of a complex provided by Example 2 of the present disclosure. [Figure 3] 1 is a schematic flowchart of a method for predicting the structure of a complex provided by Example 3 of the present disclosure. [Figure 4] 1 is a schematic flowchart of a method for predicting the structure of a complex provided by Example 4 of the present disclosure. [Figure 5] 10 is a schematic flowchart of a method for training a structure prediction model provided by Example 5 of the present disclosure. [Figure 6] FIG. 1 is a principle schematic diagram of a training method for a structural prediction model provided by an embodiment of the present disclosure. [Figure 7] FIG. 10 is a schematic diagram of a complex structure prediction device provided by Example 6 of the present disclosure. [Figure 8] FIG. 10 is a schematic configuration diagram of a training device for a structural prediction model provided by Example 7 of the present disclosure. [Figure 9] FIG. 1 is a schematic block diagram of an exemplary electronic device for implementing embodiments of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0013]

[0033] The following describes exemplary embodiments of the present disclosure in conjunction with the drawings. For ease of understanding, various details of the embodiments of the present disclosure are included therein and should be considered merely exemplary. 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 disclosure. Similarly, for the sake of clarity and conciseness, the following description will omit descriptions of well-known functions and structures.

[0014] With the rapid development of artificial intelligence (AI) technology in recent years, AI-based structure prediction methods for biomolecular complexes (including various types of biomolecular complexes, such as proteins, small molecules, RNA, and DNA) have attracted widespread attention from both academia and industry. Improving the accuracy and efficiency of structure prediction for biomolecular complexes through deep learning technology has become a core research topic. Its application scenarios are widespread, including protein 3D structure prediction, protein-small molecule interaction conformation prediction, and RNA structure prediction. Specifically, given a biomolecular sequence (e.g., protein sequence, RNA sequence, or DNA sequence), the molecular formula of the small molecule, covalent bond information, and protein modification information, it is necessary to predict the atomic-level 3D spatial structure of these biomolecules. Obtaining accurate structures, including their structure and binding poses, plays an important role in biologists' analysis of biological processes and downstream applications such as affinity prediction, drug screening, and drug design.

[0015] Related technologies use all-atom structure prediction models to predict the structure of biomolecular complexes. All-atom structure prediction models (e.g., AlphaFold3) typically include two networks: a structure generation network and a structure scoring network. The structure generation network is used to generate corresponding conformations based on input biomolecular sequence information, and the structure scoring network is used to evaluate the generated conformations and assess the quality of the generated conformations. During the inference process of generating the complex, the structure generation network is called multiple times to generate different conformations. These conformations are then scored using the structure scoring network, and the conformation with the highest score is selected as the final predicted conformation. To differentiate the generated different conformations, the following methods are usually used: (1) Perturb the input data, for example, perturb multiple sequence comparison (MSA) information in the input. (2) Perturb the network structure, e.g., enable the dropout module of the network. (3) The structure generation module is based on a diffusion framework, and its output itself is random, so different diffusion random seeds can generate different conformations.

[0016] Related sampling techniques can provide different conformations, but the sampling is highly random and can be inefficient due to, for example, repeatedly sampling very similar conformations or difficulty in sampling the correct conformation.

[0017] To address at least one of the above-mentioned problems, the present disclosure provides a method for predicting the structure of a complex, a model training method, and related devices.

[0018] Hereinafter, a complex structure prediction method, a model training method, and related devices according to embodiments of the present disclosure will be described with reference to the accompanying drawings.

[0019] FIG. 1 is a schematic flowchart of the method for predicting the structure of a complex provided by Example 1 of the present disclosure.

[0020] The embodiments of the present disclosure are described as an example in which the complex structure prediction method is disposed in a complex structure prediction device, and this complex structure prediction device can be applied to any electronic device so that the electronic device can perform the complex structure prediction function.

[0021] The electronic device may be any device with computing capabilities, such as a computer, a mobile terminal, a server, etc., and the mobile terminal may be a hardware device with various operating systems, touchscreens and / or displays, such as an in-vehicle device, a mobile phone, a tablet computer, a personal digital assistant, or a wearable device.

[0022] As shown in FIG. 1, this complex structure prediction method can include the following steps 101 to 104.

[0023] In step 101, a combination of biomolecular sequences is obtained.

[0024] The biomolecular sequence combinations are obtained by combining sequences based on a plurality of designated biomolecular sequences.

[0025] In order to effectively simulate potential interaction patterns between different biomolecular sequences, the disclosed embodiments allow multiple designated biomolecular sequences to be paired or combined in different ways to facilitate further analysis of possible interactions between them, where the multiple designated biomolecular sequences may include, for example, protein sequences and nucleic acid sequences, and may also include, for example, two or more different protein sequences.

[0026] In step 102, a first probability distribution of biomolecular sequence combinations is predicted.

[0027] A first probability distribution is used to indicate a first probability of a group of multiple candidate structural units in a combination of biomolecular sequences, the group of candidate structural units including structural units in at least two biomolecular sequences, and the first probability is used to indicate the possibility of interaction between each structural unit in the group of candidate structural units.

[0028] Each biomolecular sequence can contain multiple structural units (also called tokens), for example, for protein sequences, DNA sequences, and RNA sequences, each residue is treated as one token. In the disclosed examples, for each biomolecular sequence combination, a further probability distribution (also called a first probability distribution) is predicted to evaluate which structural units in the biomolecular sequence combination are most likely to interact with each other.

[0029] In step 103, at least one first group of structural units is determined from among the plurality of candidate groups of structural units based on the first probability distribution.

[0030] In order to accurately determine the structural unit group that is most likely to interact from among a plurality of candidate structural unit groups from a combination of biomolecular sequences, one possible implementation method is to determine at least one first structural unit group from among the plurality of candidate structural unit groups based on a first probability of the plurality of candidate structural unit groups in the combination of biomolecular sequences indicated by a first probability distribution.

[0031] In step 104, a target structure of the biomolecular complex is predicted based on the interacting structural units in each first structural unit group.

[0032] Furthermore, the target structure of the biomolecular complex is predicted based on the association information of the first structural unit group, and the association information of the first structural unit group includes, but is not limited to, detailed information such as the position, type, and interaction pattern of the structural unit.

[0033] For example, if the first structural unit group contains specific amino acid residues (such as lysine and aspartic acid) and it is known that there may be non-covalent interactions (such as salt bridges) between them, such interactions are specifically taken into consideration when predicting the complex structure.

[0034] As described above, by obtaining a biomolecular sequence combination composed of a plurality of specified biomolecular sequence combinations, potential interaction patterns between different biomolecules are simulated, providing a rich candidate space for subsequent structure prediction. Furthermore, based on the first probability distribution of the predicted biomolecular sequence combinations, the possibility of interactions occurring between the structural units in each candidate structural unit group in the biomolecular sequence combination is quantitatively evaluated, thereby filtering out the first structural unit groups with the highest probability of interactions and avoiding unnecessary testing of the vast conformational space. Finally, based on the interacting structural units in the first structural unit group, the overall three-dimensional structure of the target biomolecular complex is further predicted, using the interacting structural units in the first structural unit group as guidance, ensuring that the structure of the generated biomolecular complex meets its interaction requirements, thereby improving the reliability and biological rationality of the structure prediction.

[0035] In any embodiment of the present disclosure, to clearly explain how to predict the target structure of a biomolecular complex based on the interacting structural units in each first structural unit group, the present disclosure presents another method for predicting the structure of a complex.

[0036] FIG. 2 is a schematic flowchart of the method for predicting the structure of a complex provided by Example 2 of the present disclosure.

[0037] As shown in FIG. 2, this complex structure prediction method can include the following steps 201 to 206.

[0038] In step 201, a combination of biomolecular sequences is obtained.

[0039] The biomolecular sequence combinations are obtained by combining sequences based on a plurality of designated biomolecular sequences.

[0040] In step 202, a first probability distribution of biomolecular sequence combinations is predicted.

[0041] A first probability distribution is used to indicate a first probability of a group of multiple candidate structural units in a combination of biomolecular sequences, the group of candidate structural units including structural units in at least two biomolecular sequences, and the first probability is used to indicate the possibility of interaction between each structural unit in the group of candidate structural units.

[0042] In step 203, at least one first group of structural units is determined from among a plurality of candidate groups of structural units based on a first probability distribution.

[0043] In order to accurately filter the group of structural units that are most likely to interact, one possible implementation is to determine a first group of structural units that are most likely to interact from each group of candidate structural units based on a first probability of each group of candidate structural units in a first probability distribution.

[0044] As an example, at least one first structural unit group is determined from each candidate structural unit group based on a first probability of each candidate structural unit group in a first probability distribution, and the first probability of the first structural unit group is greater than a set threshold.

[0045] That is, a threshold is set as a filtering criterion, and only candidate structural units whose first probability exceeds this threshold are selected as the first structural unit group, thereby ensuring that the selected first structural unit group has high reliability and stability, and effectively eliminating structural units with low probability or instability, thereby improving the accuracy and reliability of subsequent structural prediction of biomolecular complexes.

[0046] In step 204, the similarity between any two first structural unit groups is obtained.

[0047] Similarity is used to indicate the degree of similarity between biomolecular complexes generated based on interacting structural units in any two first structural unit groups.

[0048] In order to avoid the possibility of repeatedly sampling highly similar conformations, one possible implementation is to determine the similarity between any two groups of first structural units, and filter the groups of first structural units based on the similarity between any two groups of first structural units, thus first obtaining the similarity between any two groups of first structural units.

[0049] As an example, a structure prediction is performed based on any two groups of first structural units to obtain a predicted structure of the any two groups of first structural units, the predicted structure is used to indicate the structure of a biomolecular complex generated by the first groups of structural units, the degree of similarity between the structures of the biomolecular complex generated by the any two groups of first structural units is evaluated based on the predicted structure of the any two groups of first structural units, and the degree of similarity between the any two groups of first structural units is determined based on the degree of similarity.

[0050] That is, for each pair of selected first structural units (i.e., a combination including at least two structural units), a computational biology tool or algorithm is used to predict the overall three-dimensional structure (predicted structure) of the biomolecular complex that they may form, and the degree of spatial overlap, differences in atomic positions, and other geometric and topological features of the two predicted structures are compared to determine a similarity score for the two predicted structures. The similarity score is used to indicate the degree of similarity between the structures of the biomolecular complex generated by any two first structural unit groups, and the similarity between any two first structural unit groups is determined based on the similarity score. Note that the higher the degree of similarity between the structures of the biomolecular complex generated by any two first structural unit groups, the higher the similarity between any two first structural unit groups.

[0051] In step 205, each of the first structural unit groups is filtered based on the similarity between each of the first structural unit groups to obtain a retained first structural unit group.

[0052] In order to remove highly similar structural unit groups, reduce unnecessary repeated calculations, and save time and computational resources, one possible implementation is to filter the first structural unit groups based on a set similarity threshold. If certain first structural unit groups are found to be highly similar (i.e., exceeding the set threshold), one representative first structural unit group can be selected and retained, and other similar first structural unit groups can be removed, thereby obtaining a retained first structural unit group.

[0053] In step 206, a target structure of the biomolecular complex is predicted based on the interacting structural units in the retained first group of structural units.

[0054] Furthermore, since the retained first structural unit group represents different interaction patterns and structural characteristics, in order to improve the diversity of predicted structures of biomolecular complexes, the target structure of the biomolecular complex is predicted based on the related information of interacting structural units in the retained first structural unit group.

[0055] The process of executing steps 201 and 202 can be realized using any of the methods in the embodiments of the present disclosure, but the embodiments of the present disclosure are not limited thereto, and a description thereof will be omitted.

[0056] As described above, by calculating the similarity between any two groups of first structural units and filtering all groups of first structural units based on this similarity, highly similar or overlapping groups of first structural units can be effectively removed, thereby retaining groups of first structural units with greater representativeness and diversity. This process not only reduces redundant calculations in subsequent structure prediction and improves computational efficiency, but also avoids problems of falling into local optima and repeatedly sampling similar conformations. Furthermore, predicting the target structure of a biomolecular complex based on the interacting structural units in the retained groups of first structural units helps to focus on key interaction patterns and improve the accuracy and biological rationality of the prediction results.

[0057] In any embodiment of the present disclosure, to clearly illustrate a method for predicting a target structure of a biomolecular complex based on interacting structural units in a retained first group of structural units, the present disclosure presents another method for predicting the structure of a complex.

[0058] FIG. 3 is a schematic flowchart of the method for predicting the structure of a complex provided by Example 3 of the present disclosure.

[0059] As shown in FIG. 3, this complex structure prediction method can include the following steps 301 to 308.

[0060] In step 301, a combination of biomolecular sequences is obtained.

[0061] The biomolecular sequence combinations are obtained by combining sequences based on a plurality of designated biomolecular sequences.

[0062] In step 302, a first probability distribution of biomolecular sequence combinations is predicted.

[0063] The first probability distribution is used to indicate a first probability of a group of multiple candidate structural units in a combination of biomolecular sequences, the group of candidate structural units including structural units in at least two biomolecular sequences, and the first probability is used to indicate the possibility of an interaction between each structural unit in the group of candidate structural units.

[0064] In step 303, at least one first group of structural units is determined from among a plurality of candidate groups of structural units based on a first probability distribution.

[0065] In step 304, the similarity between any two first structural unit groups is obtained.

[0066] Similarity is used to indicate the degree of similarity between biomolecular complexes generated based on interacting structural units in any two first structural unit groups.

[0067] In step 305, each of the first structural unit groups is filtered based on the similarity between each of the first structural unit groups to obtain a retained first structural unit group.

[0068] In step 306, a candidate structure of a biomolecular complex corresponding to each retained group of first structural units is predicted based on the interacting structural units in each retained group of first structural units.

[0069] In order to improve computational efficiency and diversity of results and to avoid repeated sampling, one possible implementation is to perform multiple sampling processes on the retained first group of structural units, and each time, select and label one group of unlabeled structural units, and generate the structure of the corresponding biomolecular complex based on the interacting structural units in the sampled group of structural units.

[0070] As an example, a sampling process is performed multiple times on the retained first structural unit group, and the sampling process includes a step of sampling one unlabeled first structural unit group from the retained first structural unit group and labeling the one first structural unit group, where the labeling is used to indicate that the one first structural unit group has already been sampled, and a step of generating one candidate structure based on interacting structural units in the one first structural unit group.

[0071] That is, in each sampling, one group of structural units is selected randomly or based on a specific strategy from among the unlabeled (i.e., unprocessed) first group of structural units, and once a specific group of first structural units is selected, it is labeled. The labeling serves to indicate that this group of structural units has already been sampled and processed, thereby avoiding repeated processing of the same group of structural units in the same or multiple sampling processes, and for each sampled and labeled group of first structural units, a structure of one biomolecular complex, i.e., one candidate structure, is generated based on the interaction information between the structural units.

[0072] In step 307, a structure score for each candidate structure is predicted.

[0073] The structural score is used to indicate the degree of match between the candidate structure and the corresponding structural unit of the first group of structural units.

[0074] To effectively evaluate the rationality of each candidate structure and filter the optimal conformation, one possible implementation is to obtain a structural score for each candidate structure by evaluating the degree of match between the candidate structure of the biomolecular complex and its corresponding structural unit in the first structural unit group. The structural score can be calculated using various methods, such as energy function, geometric match, or comparison with known similar structures. A higher structural score indicates that the candidate structure is more compatible with the expected combination of structural units and their interaction pattern.

[0075] In step 308, a target structure is determined from among the candidate structures based on the structure score of each candidate structure.

[0076] To improve the quality of structural predictions of biomolecular complexes, one possible implementation is to sort all candidate structures based on their scores and directly select the candidate structure with the highest score as the sole target structure.

[0077] The execution process of steps 301 to 305 can be realized using any of the methods of the embodiments of the present disclosure, but the embodiments of the present disclosure are not limited thereto, and the description thereof will be omitted.

[0078] As described above, by predicting candidate structures of biomolecular complexes based on the interacting structural units in the retained first structural unit group, a comprehensive search of potential complex conformations is realized, ensuring that all possible structural patterns are evaluated and avoiding the leakage of potentially important structural information. Next, a structural score is calculated for each generated candidate structure, and this score reflects the degree of match and stability between the candidate structure and its corresponding first structural unit group. Finally, a target structure is determined from each candidate structure based on the calculated structural score, thereby improving the quality and accuracy of structure prediction of biomolecular complexes.

[0079] To clearly explain how to predict the probability distribution of biomolecular sequence combinations in any embodiment of the present disclosure, the present disclosure presents another method for predicting the structure of a complex.

[0080] FIG. 4 is a schematic flowchart of the method for predicting the structure of a complex provided by Example 4 of the present disclosure.

[0081] As shown in FIG. 4, this complex structure prediction method can include the following steps 401 to 407.

[0082] In step 401, a combination of biomolecular sequences is obtained.

[0083] The biomolecular sequence combinations are obtained by combining sequences based on a plurality of designated biomolecular sequences.

[0084] In step 402, a group of multiple candidate structural units for a combination of biomolecular sequences is obtained.

[0085] In the examples disclosed herein, a group of candidate structural units refers to a set of child molecular units that may form a stable complex in a combination of biomolecular sequences, recognized based on specific criteria (physicochemical properties, known interaction databases, etc.). Each group of candidate structural units contains at least two structural units (e.g., amino acid residues or nucleotides), and there may be interactions between these units.

[0086] In step 403, the spatial distance between a plurality of structural units in any candidate structural unit group is calculated.

[0087] In order to determine whether multiple structural units in any group of candidate structural units are physically "close" and therefore have the potential for interaction to occur, one possible implementation is to calculate the spatial distance between multiple structural units in any group of candidate structural units using a method such as the Euclidean distance formula.

[0088] In step 404, a first probability of any group of candidate structural units is determined based on the spatial distances between the structural units in the group of candidate structural units.

[0089] The first probability is negatively correlated with spatial distance.

[0090] In one possible implementation, the greater the spatial distance between multiple structural units in any group of candidate structural units, the smaller the first probability of the group of candidate structural units, i.e., the smaller the possibility of interaction between each structural unit in the group of candidate structural units; and the smaller the spatial distance between multiple structural units in any group of candidate structural units, the greater the first probability of the group of candidate structural units, i.e., the greater the possibility of interaction between each structural unit in the group of candidate structural units.

[0091] In step 405, a first probability distribution of biomolecular sequence combinations is generated based on the first probability of each candidate structural unit group.

[0092] Furthermore, the first probabilities of all candidate structural unit groups are integrated into a comprehensive probability distribution map, i.e., a first probability distribution of biomolecular sequence combinations, which not only reflects the interaction possibility (i.e., the first probability) of each individual structural unit group, but also provides information about which structural unit groups are more likely to form more stable complexes in the biomolecular sequence combinations.

[0093] In step 406, at least one first group of structural units is determined from among the plurality of candidate groups of structural units based on the first probability distribution.

[0094] In step 407, a target structure of the biomolecular complex is predicted based on the interacting structural units in each first structural unit group.

[0095] The execution process of step 401 and steps 406 to 407 can be realized using any of the methods in the embodiments of the present disclosure, but the embodiments of the present disclosure are not limited thereto, and a description thereof will be omitted.

[0096] As described above, the spatial distances between the structural units in any candidate structural unit group are calculated, and a first probability for any candidate structural unit group is determined based on the spatial distance. Because the first probability has a negative correlation with the spatial distance, a small spatial distance corresponds to a high first probability, thereby effectively quantifying the likelihood of interactions between the structural units in each candidate structural unit group. Furthermore, a first probability distribution of biomolecular sequence combinations is generated based on the first probability for each candidate structural unit group. This first probability distribution provides a basis for filtering out first structural unit groups with high potential, thereby improving the reliability and accuracy of biomolecular complex structure prediction.

[0097] The above is an embodiment corresponding to the application method of the structure prediction model (i.e., the structure prediction of a complex), and the present invention further provides a method for training the structure prediction model.

[0098] FIG. 5 is a schematic flowchart of a method for training a structure prediction model provided by Example 5 of the present disclosure.

[0099] As shown in FIG. 5, the training method for the structure prediction model can include the following steps 501 to 503.

[0100] In step 501, training samples are obtained.

[0101] The training samples include a combination of sample biomolecular sequences, which are obtained based on a combination of a plurality of sample biomolecular sequences.

[0102] In one possible implementation, a plurality of sample biomolecular sequences are collected from a database, sequence combination is performed on the plurality of sample biomolecular sequences to obtain combinations of sample biomolecular sequences, and training samples are generated based on the combinations of sample biomolecular sequences, where the training samples further include annotation structures of complexes corresponding to each combination of sample biomolecular sequences.

[0103] In step 502, a second probability distribution of sample biomolecular sequence combinations is predicted using the structure prediction model, and a predicted structure of a biomolecular complex is generated based on the second probability distribution.

[0104] The second probability distribution is used to indicate a second probability of a group of multiple structural units to be selected in a combination of sample biomolecular sequences, where the group of structural units to be selected includes structural units in at least two sample biomolecular sequences, and the second probability is used to indicate the possibility of interaction between each structural unit in the group of structural units to be selected.

[0105] In one possible implementation, predicted sample biomolecular sequence combinations in a training sample are input into a structure prediction model, a second probability distribution of the sample biomolecular sequence combinations is predicted by the structure prediction model, and a predicted structure of a biomolecular complex is generated based on the second probability distribution, wherein the second probability distribution is used to indicate a second probability of a group of multiple structural units to be selected in the combination of sample biomolecular sequences, the group of structural units to be selected includes structural units in at least two sample biomolecular sequences, and the second probability is used to indicate the possibility of interaction between each structural unit in the group of structural units to be selected.

[0106] In step 503, a structure prediction model is trained based on the differences between the annotation structures and the predicted structures corresponding to the plurality of sample biomolecular sequences.

[0107] In an embodiment of the present disclosure, a value of a loss function (referred to as a loss value in this disclosure) is determined based on the difference between the labeled structure and the predicted structure, and a structure prediction model can be trained based on the loss value to minimize the loss value.

[0108] Note that the above only exemplifies the minimization of the loss value as the termination condition for model training. In actual applications, other termination conditions can be set. For example, the termination conditions can further include the training time reaching a set length, the number of training times reaching a set number, etc., and this embodiment is not limited to these.

[0109] In order to improve the accuracy and efficiency of structural prediction of biomolecular complexes, any one embodiment of the present disclosure exemplifies that the structural prediction model includes a prediction network and a generation network, in which the prediction network is used to predict a second probability distribution of combinations of sample biomolecular sequences, the generation network is used to determine at least one second structural unit group from among multiple structural unit groups to be selected based on the second probability distribution, and the generation network is used to generate a predicted structure of the biomolecular complex based on the structural units interacting in the at least one second structural unit group.

[0110] In order to improve the prediction accuracy of the structural prediction model, in any one embodiment of the present disclosure, a generative network is used to perform at least one prediction process on interacting structural units in at least one group of second structural units, and the prediction process includes a step of generating a predicted structure of a biomolecular complex based on the interacting structural units in the retained group of second structural units, wherein the retained group of second structural units is obtained by filtering at least one group of second structural units based on the similarity between any two groups of second structural units.

[0111] In any one embodiment of the present disclosure, for example, the structure prediction model further includes a structure scoring network, and the step of generating a predicted structure of a biomolecular complex based on interacting structural units in the retained second structural unit group mainly includes the following steps 1 to 3: 1. Using the generative network, predict candidate structures of biomolecular complexes corresponding to each retained group of second structural units based on the interacting structural units in each retained group of second structural units. As an example, a sampling process is performed on the retained second structural unit group multiple times, the sampling process including sampling one unlabeled second structural unit group from the retained second structural unit group, labeling the one second structural unit group, the labeling being used to indicate that the one second structural unit group has already been sampled, and generating one candidate structure based on interacting structural units in the one second structural unit group. 2. Using a structure scoring network, predicting a structure score for a candidate structure of a biomolecular complex corresponding to each retained group of second structural units, the structure score being used to indicate the degree of match between the candidate structure and the structural unit of the corresponding group of second structural units. 3. Based on the structural scores of the candidate structures of the biomolecular complex corresponding to each of the retained groups of second structural units, a target structure is determined from among the candidate structures of the biomolecular complex corresponding to each of the retained groups of second structural units.

[0112] In an embodiment of the present disclosure, a method for training a structure prediction model includes obtaining training samples containing combinations of sample biomolecular sequences, where the combinations of sample biomolecular sequences are generated by a plurality of sample biomolecular sequences to provide abundant input data for the structure prediction model, predicting a second probability distribution for the combinations of these sample biomolecular sequences using the model, and generating a predicted structure of a biomolecular complex based on the second probability distribution. The second probability distribution quantizes the probability of interactions between structural units in each group of structural units to be selected, filtering out groups of structural units with the highest probability of interactions based on the second probability distribution, and predicting the 3D structure of the biomolecular complex, thereby avoiding aimless testing of a vast conformational space and using interacting structural units in the group of structural units as guidance to ensure that the generated structure of the biomolecular complex meets the interaction requirements. Finally, optimizing and training the structure prediction model based on the differences between known annotation structures corresponding to the plurality of sample biomolecular sequences and the predicted structure generated by the model, thereby improving the accuracy and reliability of the model's prediction of biomolecular complex structures.

[0113] Based on any of the above embodiments, as shown in FIG. 6 , the structure prediction model may be an all-atom structure prediction model, which includes a prediction network (also referred to as a contact prediction network), a generation network (also referred to as a structure generation network with contact guidance), and a structure scoring network (also referred to as a structure scoring network). The method for predicting the structure of a complex according to the embodiments of the present disclosure can also be realized based on the following steps (1) to (3). (1) A combination of biomolecular sequences is obtained, and this combination of biomolecular sequences includes a biomolecular sequence A (A1-A2-A3-A4) and a biomolecular sequence B (B1-B2-B3). (2) A prediction network is used to predict the probability distribution of biomolecular sequence combinations (also called contact probability distribution), and based on this probability distribution, a set of contact sets (first structural unit group) with high probability is recognized. (3) Based on the retained high-probability contact set (the retained first group of structural units), sampling is performed to generate conformations that satisfy the contacts. Since these contacts may correspond to similar bonding patterns, contact sets with similar bonding patterns are grouped together, i.e., the high-probability contact set is filtered to obtain the retained high-probability contact set. This method includes steps 1) to 4). 1) Sample one contact from the retained set of high-probability contacts (i.e., sample one of the retained first structural units). 2) Structure prediction is performed using a generative network, and this model generates a conformation that fits the input contacts. 3) Resample any unsampled contacts and repeat the previous steps. 4) If all high-probability contact sets have been sampled and the sampling process is still ongoing, sample all high-probability contact sets again. Finally, all predicted structures are sorted based on the scores given by the structure scoring network, and the structure with the highest score is selected as the final predicted result.

[0114] Corresponding to the complex structure prediction method provided by the examples of Figures 1 to 4 above, the present disclosure further provides a complex structure prediction device, and since the complex structure prediction device provided by the examples of the present disclosure corresponds to the complex structure prediction method provided by the examples of Figures 1 to 4 above, embodiments of the complex structure prediction method are also applicable to the complex structure prediction device provided by the examples of the present disclosure, and will not be described in detail in the examples of the present disclosure.

[0115] FIG. 7 is a schematic diagram of the complex structure prediction device provided in Example 6 of the present disclosure.

[0116] As shown in FIG. 7, the complex structure prediction device 700 includes an acquisition module 710, a prediction module 720, a determination module 730, and a generation module 740.

[0117] The acquisition module 710 acquires a combination of biomolecular sequences, the combination of biomolecular sequences being obtained by performing sequence combination based on a specified plurality of biomolecular sequences; the prediction module 720 predicts a first probability distribution of the combination of biomolecular sequences, the probability distribution being used to indicate a first probability of a plurality of candidate structural unit groups in the combination of biomolecular sequences, the candidate structural unit group including structural units in at least two biomolecular sequences, and the probability being used to indicate the possibility of interaction between each structural unit in the candidate structural unit group; the determination module 730 determines at least one first structural unit group from among the plurality of candidate structural unit groups based on the first probability distribution; and the generation module 740 predicts the target structure of the biomolecular complex based on the interacting structural units in each first structural unit group.

[0118] In one possible implementation, the generation module 740 obtains a similarity between any two first structural unit groups, the similarity is used to indicate the degree of similarity between biomolecular complexes generated by the interacting structural units in any two first structural unit groups, filters each first structural unit group based on the degree of similarity between each first structural unit group to obtain a retained target structural unit group, and predicts the target structure of the biomolecular complex based on the interacting structural units in the retained first structural unit group.

[0119] In one possible implementation, the generation module 740 predicts candidate structures of biomolecular complexes corresponding to each retained first structural unit group based on the interacting structural units in each retained first structural unit group, predicts a structural score for each candidate structure, the structural score is used to indicate the degree of match between the candidate structure and the structural unit of the corresponding first structural unit group, and determines a target structure from among the candidate structures based on the structural score of each candidate structure.

[0120] In one possible implementation, the generation module 740 performs multiple sampling processes on the retained first structural unit groups, the sampling processes including the steps of sampling one unlabeled first structural unit group from the retained first structural unit groups and labeling the one first structural unit group, where the labeling is used to indicate that the one first structural unit group has already been sampled, and generating one candidate structure based on interacting structural units in the one first structural unit group.

[0121] In one possible implementation, the generation module 740 performs structure prediction based on any two first structural unit groups to obtain predicted structures of the any two first structural unit groups, the predicted structures are used to indicate the structure of a biomolecular complex generated by the first structural unit groups, evaluates the degree of similarity between the structures of the biomolecular complex generated by the any two first structural unit groups based on the predicted structures of the any two first structural unit groups, and determines the similarity between the any two first structural unit groups based on the degree of similarity.

[0122] In one possible implementation, the determination module 730 determines at least one first structural unit group from among each candidate structural unit group based on a first probability of each candidate structural unit group in the first probability distribution, and the first probability of the first structural unit group is greater than a set threshold.

[0123] In one possible implementation, the prediction module 720 obtains a plurality of candidate structural unit groups in a combination of biomolecular sequences, calculates the spatial distance between the plurality of structural units in any of the candidate structural unit groups, determines a first probability of any of the candidate structural unit groups based on the spatial distance between the plurality of structural units in any of the candidate structural unit groups, wherein the probability has a negative correlation with the spatial distance, and generates a first probability distribution of the combination of biomolecular sequences based on the first probability of each candidate structural unit group.

[0124] The complex structure prediction device of the embodiment of the present disclosure can obtain biomolecular sequence combinations composed of multiple specified biomolecular sequence combinations, thereby simulating potential interaction patterns between different biomolecules and providing a rich candidate space for subsequent structure prediction. Furthermore, based on a first probability distribution of the predicted biomolecular sequence combinations, it quantitatively evaluates the possibility of interactions occurring between structural units in each candidate structural unit group in the biomolecular sequence combination, thereby filtering out first structural unit groups with the highest probability of interactions and avoiding aimless testing of a vast conformational space. Finally, based on the information on the interacting structural units in the first structural unit group, it further predicts the overall three-dimensional structure of the target biomolecular complex, thereby realizing the use of interacting structural units in the first structural unit group as guidance and ensuring that the structure of the generated biomolecular complex meets its interaction requirements, thereby improving the reliability and biological rationality of structure prediction.

[0125] The present disclosure further provides a training apparatus for a structural prediction model, which corresponds to the training method for a structural prediction model provided by the examples of Figures 5 to 6 above, and the training apparatus for a structural prediction model provided by the examples of the present disclosure corresponds to the training method for a structural prediction model provided by the examples of Figures 5 to 6 above. Therefore, the embodiments of the training method for a structural prediction model are also applicable to the training apparatus for a structural prediction model provided by the examples of the present disclosure, and will not be described in detail in the examples of the present disclosure.

[0126] FIG. 8 is a schematic diagram of a training device for a structural prediction model provided by the seventh embodiment of the present disclosure.

[0127] As shown in FIG. 8, the structural prediction model training apparatus 800 may include an acquisition module 810, a prediction module 820, and a training module 830.

[0128] The acquisition module 810 acquires a training sample, the training sample including a combination of sample biomolecular sequences, the combination of sample biomolecular sequences being obtained based on a combination of multiple sample biomolecular sequences; the prediction module 820 predicts a second probability distribution of the combination of sample biomolecular sequences using a structure prediction model, and generates a predicted structure of a biomolecular complex based on the second probability distribution, the second probability distribution being used to indicate a second probability of a group of multiple structural units to be selected in the combination of sample biomolecular sequences, the group of structural units to be selected including structural units in at least two sample biomolecular sequences, and the second probability being used to indicate the possibility of interaction between each structural unit in the group of structural units to be selected; and the training module 830 trains the structure prediction model based on the difference between the annotation structure corresponding to the multiple sample biomolecular sequences and the predicted structure.

[0129] In one possible implementation, the structure prediction model includes a prediction network and a generation network, and the prediction module 820 uses the prediction network to predict a second probability distribution of the combination of the sample biomolecular sequence, uses the generation network to determine at least one second structural unit group from among multiple structural unit groups to be selected based on the second probability distribution, and uses the generation network to generate a predicted structure of the biomolecular complex based on the interacting structural units in the at least one second structural unit group.

[0130] In one possible implementation form, the prediction module uses the generation network to perform at least one prediction process on interacting structural units in at least one group of second structural units, and the prediction process includes a step of generating a predicted structure of a biomolecular complex based on the interacting structural units in the retained group of second structural units, wherein the retained group of second structural units is obtained by filtering at least one group of second structural units based on the similarity between any two groups of second structural units.

[0131] In an embodiment of the structure prediction model training device of the present disclosure, training samples containing combinations of sample biomolecular sequences are obtained, and the combinations of sample biomolecular sequences are generated by a plurality of sample biomolecular sequences to provide abundant input data for the structure prediction model. The model is used to predict a second probability distribution of these combinations of sample biomolecular sequences, and a predicted structure of a biomolecular complex is generated based on the second probability distribution. The second probability distribution quantizes the probability of interactions between structural units in each group of structural units to be selected, and the group of structural units most likely to interact can be filtered based on the second probability distribution. The three-dimensional structure of the biomolecular complex is then predicted, thereby avoiding aimless testing of a vast conformational space and realizing guidance based on interacting structural units in the group of structural units, ensuring that the generated structure of the biomolecular complex meets the interaction requirements. Finally, the structure prediction model is optimized and trained based on the differences between known annotation structures corresponding to the plurality of sample biomolecular sequences and the predicted structure generated by the model, thereby improving the accuracy and reliability of the model's prediction of biomolecular complex structures.

[0132] To realize the above embodiments, the present disclosure further provides an electronic device, the electronic device including at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor such that the at least one processor can perform the method for predicting a complex structure as provided by any of the above embodiments of the present disclosure, or the method for training a structure prediction model as provided by any of the above embodiments of the present disclosure.

[0133] To realize the above embodiments, the present disclosure further provides a non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions causing a computer to execute the method for predicting a complex structure as provided by any of the above embodiments of the present disclosure, or the method for training a structure prediction model as provided by any of the above embodiments of the present disclosure.

[0134] To realize the above embodiments, the present disclosure further provides a computer program, which, when executed by a processor, realizes the method for predicting a complex structure as provided by any of the above embodiments of the present disclosure or the method for training a structure prediction model as provided by any of the above embodiments of the present disclosure.

[0135] According to an embodiment of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium, and a computer program.

[0136] 9 is a schematic block diagram of an exemplary electronic device for implementing embodiments of the present disclosure. 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 may also represent various forms of mobile devices, such as personal digital assistants, mobile phones, smartphones, 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 description herein and / or the practice of the present disclosure as sought.

[0137] 9, the electronic device 900 includes a computing unit 901 that performs various appropriate operations and processes in accordance with a computer program stored in a read-only memory (ROM) 902 or loaded from a storage unit 908 into a random access memory (RAM) 903. The RAM 903 may also store various programs and data necessary for the operation of the electronic device 900. The computing unit 901, the ROM 902, and the RAM 903 are connected to one another via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0138] The components of the electronic device 900 are connected to an I / O interface 905, including an input unit 906 such as a keyboard, a mouse, etc., an output unit 907 such as various types of displays, speakers, etc., a storage unit 908 such as a magnetic disk, an optical disk, etc., and a communication unit 909 such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 enables the electronic device 900 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0139] The computing unit 901 may be various general-purpose and / or dedicated processing components having processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphic processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various machine learning model algorithm computing units, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 executes the methods and processes described above, such as the complex structure prediction method or the molecular structure prediction model training method. For example, in some embodiments, the complex structure prediction method or the molecular structure prediction model training method described above can be implemented as a computer software program tangibly included in a machine-readable medium such as the storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed into the electronic device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into RAM 903 and executed by the computing unit 901, one or more steps of the above-described bicyclic peptide generation method or molecular structure prediction model training method may be executed. Alternatively, in other embodiments, the computing unit 901 may be configured in any other appropriate manner (e.g., via firmware) to execute the above-described complex structure prediction method or structure prediction model training method.

[0140] Various implementations of the systems and techniques described herein above may be realized in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include being embodied in one or more computer programs that can be executed and / or interpreted by a programmable system including at least one programmable processor, which may be a special-purpose or general-purpose programmable processor, and that can receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0141] Program code for carrying out the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, a special purpose computer, or other programmable data processing apparatus such that, when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are performed. The program code may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine, partially on a remote machine, or entirely on a remote machine or server.

[0142] In the context of this disclosure, a machine-readable medium may be a tangible medium that contains or can store a program for use by or in connection 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 may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the above. More specific examples of machine-readable storage media include one or more line-based electrical connections, a portable computer disk, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, a CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0143] 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 a user, and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other types of devices can also provide interaction with a user; for example, the feedback provided to the user can be any form of sensing feedback (e.g., visual feedback, auditory feedback, or haptic feedback) and can receive input from the user in any form (including acoustic, speech, or tactile input).

[0144] The systems and techniques described herein may be implemented on a computing system including a back-end component (e.g., a data server), or a computing system including a middleware component (e.g., an application server), or a computing system including a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with embodiments of the systems and techniques described herein), or on a computing system including any combination of such back-end, middleware, and front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0145] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact via a communication network. The client-server relationship is created by computer programs running on corresponding computers. The server may be a cloud server, also known as a cloud computing server or cloud host. This is a host product in a cloud computing service system that solves the problems of traditional physical hosts and VPS (Virtual Private Server) services, such as difficulty in management and poor business scalability. The server may be a server in a distributed system or a server incorporating blockchain.

[0146] Artificial intelligence is a field that studies how computers can simulate certain human thought processes and intelligent behaviors (learning, reasoning, thinking, planning, etc.), and it includes both hardware-level and software-level technologies. AI hardware technology generally includes sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, etc. AI software technology mainly includes several aspects such as computer vision technology, speech recognition technology, natural language processing technology, machine learning / deep learning, big data processing technology, and knowledge graph technology.

[0147] According to the technical solution of the embodiments of the present disclosure, by obtaining a biomolecular sequence combination composed of a combination of multiple specified biomolecular sequences, it is possible to simulate the potential interaction patterns between different biomolecules and provide a rich candidate space for subsequent structure prediction. Furthermore, based on the first probability distribution of the predicted biomolecular sequence combination, the possibility of interactions occurring between the structural units in each candidate structural unit group in the biomolecular sequence combination is quantitatively evaluated, thereby filtering out the first structural unit group with the highest probability of interactions and avoiding aimless testing of the vast conformation space. Finally, based on the information on the interacting structural units in the first structural unit group, the overall three-dimensional structure of the target biomolecular complex is further predicted, using the interacting structural units in the first structural unit group as guidance, ensuring that the structure of the generated biomolecular complex meets its interaction requirements, thereby improving the reliability and biological rationality of the structure prediction.

[0148] It should be understood that steps can be rearranged, added, or deleted using the various types of flows shown above. For example, the steps described in this disclosure may be performed in parallel, sequentially, or in a different order, but this specification is not limited thereto as long as the technical proposals presented in this disclosure can achieve the desired results.

[0149] The above specific embodiments do not limit the scope of protection of the present disclosure. It should be understood that those skilled in the art can make various modifications, combinations, subcombinations, and substitutions according to design requirements and other factors. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present disclosure should be included within the scope of protection of the present disclosure.

Claims

1. A method for predicting a complex structure, comprising: a step of acquiring a combination of biomolecular sequences, the combination of biomolecular sequences being acquired by combining sequences based on a plurality of designated biomolecular sequences; predicting a first probability distribution for the biomolecular sequence combination, wherein the first probability distribution is used to indicate a first probability of a plurality of candidate structural units in the biomolecular sequence combination, the candidate structural units including structural units in at least two of the biomolecular sequences, and the first probability is used to indicate a likelihood of interaction between each structural unit in the candidate structural units; determining at least one first group of structural units from among the plurality of candidate groups of structural units based on the first probability distribution; predicting a target structure of a biomolecular complex based on the interacting structural units in each of the first structural unit groups; A method for predicting the structure of a complex, comprising:

2. The step of predicting a target structure of a biomolecular complex based on interacting structural units in each of the first structural unit groups includes: Obtaining a similarity between any two first structural unit groups, the similarity being used to indicate the degree of similarity between biomolecular complexes generated based on interacting structural units in the any two first structural unit groups; filtering each of the first structural unit groups based on a similarity between each of the first structural unit groups to obtain a retained first structural unit group; predicting a target structure of a biomolecular complex based on interacting structural units in the retained first group of structural units; The method for predicting the structure of a complex according to claim 1, comprising:

3. The step of predicting a target structure of a biomolecular complex based on interacting structural units in the retained first structural unit group includes: predicting a candidate structure of a biomolecular complex corresponding to each of the retained first structural unit groups based on interacting structural units in each of the retained first structural unit groups; predicting a structural score for each of said candidate structures, said structural score being used to indicate the degree of match between the candidate structure and a structural unit of the corresponding first set of structural units; determining a target structure from among the candidate structures based on a structure score of each of the candidate structures; The method for predicting the structure of a complex according to claim 2, comprising:

4. The step of predicting a candidate structure of a biomolecular complex corresponding to each of the retained first structural unit groups based on interacting structural units in each of the retained first structural unit groups includes: performing a sampling process on the retained first structural unit group multiple times; The sampling step comprises: a step of sampling one unlabeled first structural unit group from the retained first structural unit groups and labeling the one first structural unit group, the labeling being used to indicate that the one first structural unit group has been sampled; generating a candidate structure based on interacting structural units in the first structural unit group; The method for predicting the structure of a complex according to claim 3, comprising:

5. The step of obtaining the similarity between any two first structural unit groups includes: A step of performing a structure prediction based on any two first structural unit groups to obtain a predicted structure of the any two first structural unit groups, wherein the predicted structure is used to indicate the structure of a biomolecular complex generated by the first structural unit groups; Evaluating the degree of similarity between the structures of biomolecular complexes generated by the any two first structural unit groups based on the predicted structures of the any two first structural unit groups; determining a degree of similarity between any two first structural unit groups based on the degree of similarity; The method for predicting the structure of a complex according to claim 2, comprising:

6. The step of determining at least one first group of structural units from among the plurality of candidate groups of structural units based on the first probability distribution includes: determining at least one first group of structural units from among the candidate group of structural units based on a first probability of each of the candidate group of structural units in the first probability distribution; The method for predicting a structure of a complex according to claim 1 , wherein the first probability of the first group of structural units is greater than a set threshold value.

7. The step of predicting a first probability distribution of biomolecular sequence combinations includes: obtaining a group of candidate structural units for the combination of biomolecular sequences; Calculating spatial distances between a plurality of structural units in any group of candidate structural units; determining a first probability of the arbitrary group of candidate structural units based on a spatial distance between a plurality of structural units in the arbitrary group of candidate structural units, the first probability being negatively correlated with the spatial distance; generating a first probability distribution of the biomolecular sequence combinations based on a first probability of each of the candidate structural units; The method for predicting the structure of a complex according to claim 1, comprising:

8. A method for training a structure prediction model, comprising: obtaining a training sample, the training sample including a combination of sample biomolecular sequences, the combination of sample biomolecular sequences being obtained based on a combination of a plurality of sample biomolecular sequences; a step of predicting a second probability distribution of the combination of the sample biomolecular sequences using a structure prediction model, and generating a predicted structure of a biomolecular complex based on the second probability distribution, wherein the second probability distribution is used to indicate second probabilities of a plurality of groups of structural units to be selected in the combination of the sample biomolecular sequences, the groups of structural units to be selected including structural units in at least two of the sample biomolecular sequences, and the second probability is used to indicate the possibility of interaction between each structural unit in the groups of structural units to be selected; training the structure prediction model based on differences between annotation structures corresponding to the plurality of sample biomolecular sequences and the predicted structures; A method for training a structure prediction model, including:

9. the structural prediction model includes a predictive network and a generative network; The predicted biomolecular complex is determined by the structure prediction model. predicting a second probability distribution of the sample biomolecular sequence combinations using the prediction network; determining at least one second group of structural units from among the plurality of groups of structural units to be selected based on the second probability distribution using the generative network; The method for training a structure prediction model according to claim 8, wherein the structure is generated using the generative network and a step of generating a predicted structure of a biomolecular complex based on the interacting structural units in the at least one second group of structural units.

10. generating a predicted structure of a biomolecular complex based on interacting structural units in the at least one second group of structural units using the generative network, and performing at least one prediction process for interacting structural units in at least one second group of structural units using the generative network, the prediction process comprising:

10. The method for training a structure prediction model according to claim 9, further comprising a step of generating a predicted structure of a biomolecular complex based on interacting structural units in the retained group of second structural units, wherein the retained group of second structural units is obtained by filtering the at least one group of second structural units based on the similarity between any two groups of second structural units.

11. The complex structure prediction device an acquisition module for acquiring a combination of biomolecular sequences, the combination of biomolecular sequences being acquired by combining sequences based on a plurality of designated biomolecular sequences; a prediction module that predicts a first probability distribution of the combination of biomolecular sequences, the probability distribution being used to indicate a first probability of a group of multiple candidate structural units in the combination of biomolecular sequences, the group of candidate structural units including structural units in at least two of the biomolecular sequences, and the probability being used to indicate a possibility of interaction between each structural unit in the group of candidate structural units; a determination module for determining at least one first group of structural units from among the plurality of candidate groups of structural units based on the first probability distribution; a generation module for predicting a target structure of a biomolecular complex based on the interacting structural units in each of the first structural unit groups; A complex structure prediction device comprising:

12. A training device for a structural prediction model, comprising: an acquisition module for acquiring training samples, the training samples including a combination of sample biomolecular sequences, the combination of sample biomolecular sequences being acquired based on a combination of a plurality of sample biomolecular sequences; a prediction module for predicting a second probability distribution of the combination of the sample biomolecular sequences using a structure prediction model and generating a predicted structure of a biomolecular complex based on the second probability distribution, wherein the second probability distribution is used to indicate second probabilities of a plurality of groups of structural units to be selected in the combination of the sample biomolecular sequences, the groups of structural units to be selected including structural units in at least two of the sample biomolecular sequences, and the second probability is used to indicate the possibility of interaction between each structural unit in the groups of structural units to be selected; a training module for training the structure prediction model based on differences between annotation structures corresponding to the plurality of sample biomolecular sequences and the predicted structures; A training device for a structural prediction model, comprising:

13. 1. An electronic device comprising: at least one processor; a memory communicatively coupled to the at least one processor; An electronic device, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 7 or the method of any one of claims 8 to 10.

14. A non-transitory computer-readable storage medium having computer instructions stored thereon, comprising: A non-transitory computer-readable storage medium, the computer instructions causing a computer to perform the method of any one of claims 1 to 7 or the method of any one of claims 8 to 10.

15. A computer program comprising: A computer program which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7 or the steps of the method according to any one of claims 8 to 10.

Images