Prediction method, device and equipment for intermolecular interaction force

Abstract

The invention discloses an intermolecular interaction force prediction method, device and equipment. The method comprises the following steps: calculating the similarity between atoms contained in each molecule through a truncation attention network; truncating each molecule according to the similarity by adopting a pre-configured truncating coefficient to obtain a substructure corresponding to each molecule; performing interaction force calculation according to the substructures corresponding to different molecules to obtain acting force between the substructures; and obtaining an interaction probability of the to-be-predicted molecule pair according to the acting force between the substructures, and determining an interaction force condition between the molecules according to the interaction force probability. According to the method, a substructure corresponding to each molecule is obtained by adopting a truncation coefficient for truncation according to the similarity, so that interference of irrelevant atoms in the substructures is avoided, the purity and the integrity of the substructures are kept, the situation of the intermolecular interaction force is obtained according to the acting force between the substructures determined by the accurately extracted substructures, and the accuracy of the intermolecular interaction force is improved. And the predicted condition of the intermolecular interaction force is more accurate.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and chemical science and technology, and in particular to a method, device and equipment for predicting intermolecular interaction forces. Background Art

[0002] Molecular combinations may lead to unexpected intermolecular interactions, which in turn may trigger adverse molecular events. Especially in chemical pharmaceuticals, adverse molecular events can significantly affect the therapeutic effects of drugs. Therefore, identifying intermolecular interactions plays a very important role in chemical engineering, medicine, and scientific research.

[0003] Traditionally, intermolecular interaction force identification mainly uses a graph-based approach to represent molecules as graph structures for substructure extraction and analysis. However, when extracting substructures, the importance of nodes is usually determined through a specific network structure after the molecular information is transmitted, thereby demarcating the boundaries of the substructure. However, at this time, each node has already integrated information from other nodes, resulting in blurred substructure boundaries. Schemes based on prior knowledge or predefined valid structures (Category II) ensure the purity of the substructure by preventing interference from other atomic information. However, these methods face major challenges. First, different substructures may be very similar. For example, -COOH and -C=O both contain carbonyl (tang) groups, which may cause confusion in the model. Second, inadequately predefined substructures may miss other important atomic nodes. When substructure extraction is inaccurate, the accuracy of intermolecular interaction force prediction will be significantly affected. Summary of the Invention

[0004] The present invention provides a method for predicting intermolecular interaction forces, so as to achieve accurate prediction of intermolecular interaction forces.

[0005] According to a first aspect of the present invention, a method for predicting intermolecular interaction forces is provided, comprising:

[0006] Obtain the molecules contained in the molecular pair to be predicted, and calculate the similarity between the atoms contained in each molecule through the truncated attention network;

[0007] For each molecule, a pre-configured cutoff coefficient is used to perform truncation according to the similarity to obtain the substructure corresponding to each molecule;

[0008] Calculating the interaction forces between the substructures according to the substructures corresponding to the different molecules to obtain the interaction forces between the substructures, wherein the atom pairs include atoms from different molecules;

[0009] The interaction probability of the molecular pair to be predicted is obtained according to the inter-substructure interaction force, and the intermolecular interaction force situation is determined according to the interaction force probability.

[0010] According to another aspect of the present invention, there is provided a device for predicting intermolecular interaction forces, comprising:

[0011] The similarity calculation module is used to obtain the molecules contained in the molecular pair to be predicted and calculate the similarity between the atoms contained in each molecule through the truncated attention network;

[0012] A substructure extraction module, configured to obtain the substructure corresponding to each molecule by truncating each molecule using a pre-configured truncation coefficient according to the similarity;

[0013] an inter-substructure force calculation module, configured to calculate the interaction forces between the substructures corresponding to different molecules to obtain the inter-substructure forces, wherein the atomic pairs include atoms from different molecules;

[0014] The intermolecular interaction force determination module is used to obtain the interaction probability of the molecular pair to be predicted based on the inter-substructure interaction force, and determine the intermolecular interaction force situation based on the interaction force probability.

[0015] According to another aspect of the present invention, an electronic device is provided, comprising:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to perform the method according to any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method described in any embodiment of the present invention when executed.

[0020] The technical solution of the embodiment of the present invention calculates the similarity between atoms in a molecule by a truncated attention network, and obtains the substructure corresponding to each molecule by truncation based on the similarity using a truncation coefficient, thereby avoiding the interference of irrelevant atoms in the substructure, maintaining the purity and integrity of the substructure, and obtaining the intermolecular interaction force based on the inter-substructure force determined by the accurately extracted substructure, thereby making the predicted intermolecular interaction force more accurate.

[0021] 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 the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 This is a flow chart of a method for predicting intermolecular interaction forces provided in Example 1 of the present invention;

[0024] Figure 2 2. It is a schematic diagram of an application framework of intermolecular interaction forces provided in accordance with the first embodiment of the present invention;

[0025] Figure 3 This is a flow chart of a method for predicting intermolecular interaction forces provided in accordance with the second embodiment of the present invention;

[0026] Figure 4 This is a schematic structural diagram of a device for predicting intermolecular interaction forces provided according to the third embodiment of the present invention;

[0027] Figure 5 It is a structural diagram of an electronic device provided by the fourth embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Example 1

[0031] Figure 1 A flowchart of a method for predicting intermolecular interaction forces is provided for the first embodiment of the present invention. This embodiment is applicable to the case of predicting intermolecular interaction forces. The method can be executed by an intermolecular interaction force prediction device, which can be implemented in the form of hardware and / or software. Figure 1 As shown, the method includes:

[0032] Step S101: obtain the molecules contained in the molecule pair to be predicted, and calculate the similarity between the atoms contained in each molecule through the truncated attention network.

[0033] Specifically, this embodiment aims to mine representative substructures from complex graph data and achieve information complementarity and synergy between different substructures by involving specialized interaction modules, thereby better predicting the impact on intermolecular interactions and providing accurate prediction support for the next task that requires consideration of molecular interactions. Figure 2The figure shows a schematic diagram of the application framework of the intermolecular interaction force in this embodiment, which mainly involves three parts. The first is substructure extraction, which extracts the core substructure formed between the atoms in the molecular graph to ensure that the obtained substructure has high recognition and information purity; the second is substructure interaction. The main task of this part is to establish interactive connections between different substructures, and through cross-attention or fusion strategies, the extracted substructures can transmit and supplement information to each other, thereby further enhancing the expressive power of the entire graph representation. Finally, the intermolecular interaction force prediction is used to complete the final prediction of the target task. The above-mentioned substructure extraction and interaction network based on the truncated attention mechanism not only provides a novel graph representation method in theory, but also shows high robustness and generalization performance in practical applications. Through the fine division and interactive fusion of substructures, the above framework can fully capture the local and global features contained in the graph, providing an efficient and intuitive solution for the understanding and processing of complex graph data.

[0034] Optionally, the similarity between atoms contained in each molecule is calculated through a truncated attention network, including: determining the levels contained in the truncated attention network, wherein the truncated attention network includes at least two levels; determining the interval distance based on the current level of the molecule passing through the truncated attention network; determining the target atom for the current atom based on the interval distance, and calculating the similarity between the current atom and each target atom at the current level.

[0035] In this embodiment, the molecules contained in the molecular pair to be predicted are obtained, and each molecule can be identified as a graph structure, in which the relationship between nodes and edges describes the internal combination and chemical bond relationship of the molecule. For example, the node set V = {v1, v2..vn} in the molecular graph represents the atoms in the molecule, each node vi corresponds to an atom, and the edge set X∈V×V is used to represent the chemical bonding relationship between atoms, so the edges connect those atoms that interact through chemical bonds. In the task of predicting intermolecular interaction forces, each data in the dataset consists of a pair of molecular graphs, such as Figure 2 The starting point of the framework is shown in Figure 1, while the pair of molecular graphs above correspond to the specific graphical representation of the molecules to be predicted. Therefore, graph-based molecular representation methods provide a powerful framework for using machine learning algorithms to analyze and understand molecular structures and their interactions in molecular and chemical research. This provides comprehensive and detailed structural information for the task of predicting intermolecular interaction forces.

[0036] It should be noted that in order to accurately extract the substructure at the atomic level, an L-layer truncated graph attention network (TGAT) is used in this embodiment. The network aggregates information and updates node features. TGAG is a message passing method based on the attention mechanism, which can collect useful information from its neighboring nodes by calculating the correlation scores between nodes. In this process, we set the low correlation coefficient to zero to ensure that information is only obtained from neighboring nodes with high correlation with the target node. This can effectively avoid the interference of redundant information and ensure the accuracy and efficiency of the information transmission process. In this way, TGAT can perform accurate information aggregation in the molecular graph to ensure that the obtained substructure information is closely related to the chemical properties at the atomic level. And in this embodiment, when calculating the similarity between the atoms contained in each molecule through the TGAT network, it is necessary to first determine the levels contained in the network. In this embodiment, L=3 is used as an example for explanation. Each molecule will pass through each level in turn, and the interval distance is determined according to the current level of the molecule through the network. For example, when it is determined that the molecule passes through the first layer, the interval distance is determined to be 1, when it is determined that the molecule passes through the second layer, the interval distance is determined to be 2, and when it is determined that the molecule passes through the third layer, the interval distance is determined to be 3. For the current atom, the target atom is determined based on the interval distance, and the similarity between the current atom and each target atom at the current level is calculated. For example, a molecule is extracted from the molecule to be predicted, and an atom is arbitrarily selected from the molecule as the current atom. When passing through the first level, since the interval distance is 1, the atom that is one step away, that is, the adjacent atom, is required as the target atom, and the number of adjacent target atoms can be multiple. In this embodiment, after determining the target atom associated with the current atom in the first level, the similarity between the current atom and each target atom is calculated, and the following formula (1) can be used specifically when calculating the similarity:

[0037]

[0038] Among them, e ij is the similarity between node i and node j, (·) T Is the transpose operator, used to transpose a matrix or vector. The | symbol represents the concatenation operation of a vector or matrix, which is usually used to merge multiple vectors or matrices into a long vector or matrix. W is a learnable weight vector, which is continuously adjusted during network training through optimization methods such as gradient descent to minimize the loss function. i and h jRepresent the embedding vectors of node i and node j respectively, these embedding vectors are calculated by the hierarchical structure of the network, and they represent the characteristics and relationships of the nodes in high-dimensional space. In this calculation process, it depends on the characteristics of each node and the characteristics of the neighboring nodes in the molecular graph, and the weight of information transmission is determined by calculating the similarity between them. Such a design enables the network to selectively focus on those most relevant nodes in multiple neighboring nodes, thereby avoiding the interference of irrelevant information during message transmission. In this way, TGAT not only enhances the interaction between nodes, but also effectively reduces the noise caused by the information transmitted by irrelevant nodes, thereby improving the performance and robustness when extracting substructures and processing complex chemical information. And when passing through the second level, due to the interval distance of 2, it is necessary to use the adjacent atoms after two steps, that is, one atom apart, as the target atom, and the number of target atoms on the second level can also be multiple, and the mode of calculating the current atom and the target atom on the second level is roughly the same as on the first level, and no further description is given in this embodiment.

[0039] Step S102 : truncating each molecule using a pre-configured truncation coefficient according to the similarity to obtain the substructure corresponding to each molecule.

[0040] Optionally, a pre-configured truncation coefficient is used to perform truncation on each molecule according to the similarity to obtain the substructure corresponding to each molecule, including: determining the non-associated atoms of each molecule at each level whose similarity is less than the truncation coefficient; cutting the chemical bonds between the non-associated atoms on the molecule to obtain the substructure corresponding to each molecule, wherein the substructures from different levels contain different numbers of atoms, and the substructures from the same level contain the same number of atoms.

[0041] Specifically, in this embodiment, after calculating the similarity between atoms at different levels of the molecule, a cutoff coefficient is pre-set to effectively filter out low-similarity nodes that are not related to important relationships between nodes. Only atoms with similarities higher than the cutoff coefficient are retained, thereby removing redundant or irrelevant information that does not contribute to information transfer between nodes. Specifically, atoms with similarities lower than the cutoff coefficient are screened out as non-associated atoms, and the chemical bonds between non-associated atoms are cut off on the molecule to obtain the substructure corresponding to each molecule. The following formula (2) can be used for truncation:

[0042]

[0043] Among them, s is the coefficient value corresponding to index i,j = top-α(e,α), and α is the truncation coefficient. The top-α function selects α indices from the given input vector. The above truncation process is very important in graph learning, because in practical applications, there are often a large number of low-correlation edges between the nodes of the molecular graph. The existence of these edges may increase the complexity of the calculation and even introduce noise. Therefore, by effectively truncating the low-correlation coefficient, unnecessary computational burden can be reduced, and it is ensured that only the most meaningful information for the task is transmitted during the message passing process, thereby improving efficiency and accuracy. Figure 2 The upper and lower arrows in the molecular structure extraction section show examples of substructures extracted from different molecules at different levels of the network. It can be seen that the number of nodes, i.e., atoms, contained in the substructures obtained at different levels is different.

[0044] Optionally, after truncating each molecule according to the similarity using a pre-configured truncation coefficient to obtain the substructure corresponding to each molecule, the method also includes: normalizing the similarity between the atoms contained in the substructure; traversing the substructure to obtain the characteristics of the current atom traversed and the similarity between the atoms associated with the current atom; and updating the characteristics of the current atom according to the similarity between the atoms associated with the current atom.

[0045] Specifically, in this embodiment, after obtaining the substructure corresponding to each molecule, the similarities between the atoms contained in the substructure are normalized. Specifically, the normalization process can be performed using the following formula (3):

[0046]

[0047] Among them, e ij is the similarity between node i and node j, c ij It is the normalized similarity. LeakyReLU is an activation function. In this embodiment, normalization is used to unify the similarities between different nodes, which vary greatly in magnitude, into a single unit for calculation, thereby facilitating subsequent calculations. In addition, the features of each atom contained in the substructure are updated. Specifically, the substructure is traversed to obtain the features of the current atom traversed and the similarities between the atoms associated with the current atom. The features of the current atom are updated based on the similarities between the atoms associated with the current atom.

[0048] Step S103 , performing interaction force calculation based on the substructures corresponding to different molecules to obtain the interaction force between the substructures.

[0049] Optionally, the interaction forces are calculated based on the substructures corresponding to different molecules to obtain the interaction forces between substructures, including: importance scoring the substructures corresponding to each molecule; deleting the substructures whose importance score values ​​are less than a preset threshold to obtain the target substructures corresponding to each molecule; and obtaining the interaction forces between substructures based on the target substructures corresponding to different molecules.

[0050] Optionally, importance scoring is performed on the substructure corresponding to each molecule, including: obtaining a characteristic matrix of the substructure corresponding to each molecule, and projecting the characteristic matrix to obtain a one-dimensional vector; and performing importance scoring on the one-dimensional vector to obtain an importance score value.

[0051] Optionally, the inter-substructure force is obtained by performing interaction force calculation based on the target substructures corresponding to different molecules, including: determining the first molecule and the second molecule included in the molecular pair to be predicted; obtaining the first vector of the first target substructure corresponding to the first molecule, and the first weight corresponding to the first vector; obtaining the second vector of the second target substructure corresponding to the second molecule, and the second weight corresponding to the second vector; and calculating the inter-substructure force between the first target substructure and the second target substructure based on the first vector, the first weight, the second vector and the second weight.

[0052] Specifically, in this embodiment, after obtaining the substructure corresponding to each molecule, as shown in FIG. Figure 2 As shown, since the number of substructures corresponding to each molecule is very large, it is obviously impractical to select each substructure separately and perform subsequent processing, especially when facing large-scale graph data. Directly selecting each substructure is not only inefficient, but also affects the feasibility and performance of the calculation. Therefore, in this embodiment, the TopKPool network is used to screen the substructures to significantly reduce the computational overhead caused by redundant substructures. Among them, when screening, the importance score of each substructure is obtained, and the substructures whose importance score meets the requirements are retained as target substructures, and the target substructures from different molecules are subjected to interaction force calculation to obtain the interaction force between substructures.

[0053] Step S104 , obtaining the interaction probability of the molecular pair to be predicted based on the interaction force between the substructures, and determining the intermolecular interaction force situation based on the interaction force probability.

[0054] Optionally, the interaction probability of the molecular pair to be predicted is obtained based on the inter-substructure forces, and the intermolecular interaction force situation is determined based on the interaction force probability, including: obtaining a learning matrix for the molecular pair to be predicted based on the inter-substructure forces; calculating based on the inter-substructure forces and the learning matrix corresponding to different molecules to obtain the interaction probability of the molecular pair to be predicted; judging whether the interaction probability is 0, if so, determining that there is no intermolecular force between the molecules to be predicted, otherwise, determining that there is an intermolecular interaction force between the molecules to be predicted.

[0055] Specifically, after obtaining all the inter-substructural forces between two molecules in the molecule to be predicted, this embodiment constructs a learnable matrix M for the pair of molecules to be predicted based on these inter-substructural forces, where each element in M ​​represents an inter-substructural force. This embodiment introduces a matrix M to encode each sub-structural force, facilitating the assignment of appropriate weights to each possible interaction type during model training. This learnable matrix not only helps the model learn the underlying relationships between different interaction types but also allows for adaptive adjustment based on training data, gradually refining its judgment of interaction types. This approach simplifies the previously complex multi-classification problem into a binary classification problem, significantly reducing computational complexity and improving model training efficiency. Once the learning matrix M is assigned to all intermolecular interaction types, the predicted probability of intermolecular interaction can be calculated based on this matrix. This calculation process is based on the input molecular graph and interaction type, combined with the learned weights in the matrix M, to determine the probability of interaction for each pair of molecules. This approach enables the automatic identification of potential interactions between molecules through a learning process, providing an efficient and reliable solution for the task of predicting molecular interactions. After obtaining M, the interaction probability of the molecular pair to be predicted is calculated according to the following formula (4):

[0056]

[0057] in, and It is a vector of substructures extracted from different molecules, and M is a learning matrix. In this embodiment, the intermolecular interaction force can be predicted based on the value of p. In this embodiment, the intermolecular interaction force of two molecules in the molecule to be predicted can be determined based on the numerical value of p. When the value of p is 0, it means that there is no intermolecular interaction force between the two molecules, and when the value of p is non-0, it means that there is an intermolecular interaction force between the two molecules. Therefore, in this embodiment, potential substructures can be generated comprehensively, and the TopKPool pooling layer is used for effective screening to select the most important target substructures. These substructures play a key role in the subsequent interaction process, thereby promoting the prediction of more efficient intermolecular interaction forces, and not only improves the accuracy of the prediction, but also effectively reduces the computational overhead, showing lower time complexity and computational complexity. In addition, by extracting and analyzing substructures in a fine-grained manner, the potential interaction mechanism between molecules can be better revealed, providing more powerful theoretical support. This not only helps to improve the efficiency of the molecular discovery process, but also lays a solid foundation for the in-depth study of intermolecular interaction forces.

[0058] In the embodiment of the present application, the similarity between atoms in the molecule is calculated by truncating the attention network, and the substructure corresponding to each molecule is obtained by truncation based on the similarity using a truncation coefficient, thereby avoiding the interference of irrelevant atoms in the substructure, maintaining the purity and integrity of the substructure, and obtaining the intermolecular interaction force based on the inter-substructure force determined by the accurately extracted substructure, thereby making the predicted intermolecular interaction force more accurate.

[0059] Example 2

[0060] Figure 3 A flowchart of a method for predicting intermolecular interaction forces is provided in Example 2 of the present invention. This embodiment, based on the above embodiment, specifically describes how to calculate the interaction forces between substructures corresponding to different molecules to obtain the interaction forces between substructures. Figure 3 As shown, the method includes:

[0061] Step S201: obtain the molecules contained in the molecule pair to be predicted, and calculate the similarity between the atoms contained in each molecule through the truncated attention network.

[0062] Optionally, the similarity between atoms contained in each molecule is calculated through a truncated attention network, including: determining the levels contained in the truncated attention network, wherein the truncated attention network includes at least two levels; determining the interval distance based on the current level of the molecule passing through the truncated attention network; determining the target atom for the current atom based on the interval distance, and calculating the similarity between the current atom and each target atom at the current level.

[0063] Step S202 : truncate each molecule using a pre-configured truncation coefficient according to the similarity to obtain the substructure corresponding to each molecule.

[0064] Optionally, a pre-configured truncation coefficient is used to perform truncation on each molecule according to the similarity to obtain the substructure corresponding to each molecule, including: determining the non-associated atoms of each molecule at each level whose similarity is less than the truncation coefficient; cutting the chemical bonds between the non-associated atoms on the molecule to obtain the substructure corresponding to each molecule, wherein the substructures from different levels contain different numbers of atoms, and the substructures from the same level contain the same number of atoms.

[0065] Optionally, after truncating each molecule according to the similarity using a pre-configured truncation coefficient to obtain the substructure corresponding to each molecule, the method also includes: normalizing the similarity between the atoms contained in the substructure; traversing the substructure to obtain the characteristics of the current atom traversed and the similarity between the atoms associated with the current atom; and updating the characteristics of the current atom according to the similarity between the atoms associated with the current atom.

[0066] Step S203: performing importance scoring on the substructures corresponding to each molecule.

[0067] Optionally, importance scoring is performed on the substructure corresponding to each molecule, including: obtaining a characteristic matrix of the substructure corresponding to each molecule, and projecting the characteristic matrix to obtain a one-dimensional vector; and performing importance scoring on the one-dimensional vector to obtain an importance score value.

[0068] Specifically, in this embodiment, since the number of substructures corresponding to each molecule is relatively large, which affects the feasibility and performance of the calculation, the TopKPool network is used in this embodiment to screen the substructures. In order to ensure the connectivity of the substructures during screening and avoid losing the key information of the graph due to oversimplification, the feature matrix of the substructure is projected into a one-dimensional space in this embodiment, which can more effectively select the key substructures in the graph and reduce the interference of redundant information in this process. After obtaining the one-dimensional vector of the substructure, the importance score can be obtained by performing an importance score on the one-dimensional vector. When calculating the importance score value based on the one-dimensional vector, the elements in the vector can be added and summed, and the sum result can be used as the importance score value, or corresponding weights can be assigned to each element in each vector, and the product of each element and the corresponding weight can be used as the importance score value. Of course, this embodiment is only an example, and does not limit the specific calculation process of the importance score value.

[0069] Step S204 : Delete the substructures whose importance score values ​​are less than a preset threshold value to obtain target substructures corresponding to each molecule.

[0070] Specifically, this embodiment sets a preset threshold, and after obtaining the importance score of each substructure, the importance score of each substructure is compared with the preset threshold. If the importance score of a substructure is lower than the preset threshold, it indicates that the substructure does not contain key information of the molecular graph and has little impact on the intermolecular forces. Therefore, this embodiment deletes the substructure, and the remaining substructures are used as the target substructures corresponding to the molecule. The number of target substructures obtained is related to the preset threshold set by the user, which can be determined by the user based on multiple factors such as the accuracy of the intermolecular interaction forces and the current computing resources.

[0071] Step S205 , performing interaction force calculation based on the target substructures corresponding to different molecules to obtain the interaction force between the substructures.

[0072] Optionally, the inter-substructure force is obtained by performing interaction force calculation based on the target substructures corresponding to different molecules, including: determining the first molecule and the second molecule included in the molecular pair to be predicted; obtaining the first vector of the first target substructure corresponding to the first molecule, and the first weight corresponding to the first vector; obtaining the second vector of the second target substructure corresponding to the second molecule, and the second weight corresponding to the second vector; and calculating the inter-substructure force between the first target substructure and the second target substructure based on the first vector, the first weight, the second vector and the second weight.

[0073] Specifically, in this embodiment, after the target substructure corresponding to each molecule is obtained by screening through the importance score value, the interaction force between the target substructure in each molecule and the target substructure in other molecules is calculated, such as Figure 2 As shown in the Attention part of the molecule, for example, when the first molecule contains two target substructures A and B, and the second molecule contains two target substructures C and D, the inter-substructure forces between A and C, A and D, B and C, and B and D will be calculated respectively. Subsequently, the intermolecular interaction force will be calculated based on all the determined inter-substructure forces. When calculating the inter-substructure forces, the following formula (5) is used to obtain:

[0074]

[0075] in, is the vector of the i-th target substructure in the first molecule, is the vector of the j-th target substructure in the second molecule, W1 is the weight corresponding to the i-th target substructure, W2 is the weight corresponding to the j-th target substructure, and b is a learnable weight vector.

[0076] Step S206 , obtaining the interaction probability of the molecular pair to be predicted based on the interaction force between the substructures, and determining the intermolecular interaction force situation based on the interaction force probability.

[0077] Optionally, the interaction probability of the molecular pair to be predicted is obtained based on the inter-substructure forces, and the intermolecular interaction force situation is determined based on the interaction force probability, including: obtaining a learning matrix for the molecular pair to be predicted based on the inter-substructure forces; calculating based on the inter-substructure forces and the learning matrix corresponding to different molecules to obtain the interaction probability of the molecular pair to be predicted; judging whether the interaction probability is 0, if so, determining that there is no intermolecular force between the molecules to be predicted, otherwise, determining that there is an intermolecular interaction force between the molecules to be predicted.

[0078] In the embodiment of the present application, the similarity between atoms in the molecule is calculated by truncating the attention network, and the substructure corresponding to each molecule is obtained by truncation based on the similarity using a truncation coefficient, thereby avoiding the interference of irrelevant atoms in the substructure, maintaining the purity and integrity of the substructure, and obtaining the intermolecular interaction force based on the inter-substructure force determined by the accurately extracted substructure, thereby making the predicted intermolecular interaction force more accurate.

[0079] Example 3

[0080] Figure 4 A schematic diagram of the structure of a device for predicting intermolecular interaction forces provided in Example 3 of the present invention. Figure 4 As shown, the device includes:

[0081] A similarity calculation module 310 is used to obtain molecules included in the molecular pair to be predicted and calculate the similarity between the atoms included in each molecule through a truncated attention network;

[0082] The substructure extraction module 320 is used to obtain the substructure corresponding to each molecule by truncating each molecule according to its similarity using a pre-configured truncation coefficient;

[0083] The inter-substructure force calculation module 330 is used to calculate the interaction force based on the substructures corresponding to different molecules to obtain the inter-substructure force;

[0084] The intermolecular interaction force determination module 340 is used to obtain the interaction probability of the molecular pair to be predicted based on the inter-substructure interaction force, and determine the intermolecular interaction force situation based on the interaction force probability.

[0085] Optionally, a similarity calculation module is used to determine the layers included in the truncated attention network, wherein the truncated attention network includes at least two layers;

[0086] Determine the separation distance based on the molecule by truncating the current level of the attention network;

[0087] The target atom is determined based on the interval distance for the current atom, and the similarity between the current atom and each target atom at the current level is calculated.

[0088] Optionally, a substructure extraction module is used to determine the non-correlated atoms of each molecule at each level whose similarity is less than a cutoff coefficient;

[0089] The chemical bonds between non-associated atoms in the molecule are cut off to obtain the substructure corresponding to each molecule, where substructures from different levels contain different numbers of atoms, and substructures from the same level contain the same number of atoms.

[0090] Optionally, the apparatus further comprises a feature updating module for normalizing the similarities between atoms contained in the substructure;

[0091] Traverse the substructure to obtain the characteristics of the current atom and the similarities between the atoms associated with the current atom;

[0092] The feature of the current atom is updated according to the similarity between the atoms associated with the current atom.

[0093] Optionally, the inter-substructure force calculation module includes: an importance scoring unit for scoring the importance of the substructures corresponding to each molecule;

[0094] A target substructure acquisition unit is used to delete substructures whose importance scores are less than a preset threshold value to obtain target substructures corresponding to each molecule;

[0095] The inter-substructure force acquisition unit is used to calculate the interaction force based on the target substructures corresponding to different molecules to obtain the inter-substructure force.

[0096] Optionally, an important scoring unit is used to obtain a feature matrix of the substructure corresponding to each molecule, and project the feature matrix to obtain a one-dimensional vector;

[0097] Perform importance scoring on a one-dimensional vector to obtain an importance score value.

[0098] Optionally, an inter-substructure force acquisition unit is used to determine the first molecule and the second molecule included in the molecular pair to be predicted;

[0099] Obtaining a first vector of a first target substructure corresponding to the first molecule and a first weight corresponding to the first vector;

[0100] Obtaining a second vector of a second target substructure corresponding to the second molecule and a second weight corresponding to the second vector;

[0101] An inter-substructure force between the first target substructure and the second target substructure is calculated according to the first vector, the first weight, the second vector and the second weight.

[0102] Optionally, an intermolecular interaction force determination module is used to obtain a learning matrix for the molecular pair to be predicted based on the inter-substructure forces;

[0103] Calculate the interaction probability of the molecular pairs to be predicted based on the interaction forces between the substructures corresponding to different molecules and the learning matrix;

[0104] It is determined whether the interaction probability is 0. If so, it is determined that there is no intermolecular interaction force between the molecules to be predicted. Otherwise, it is determined that there is an intermolecular interaction force between the molecules to be predicted.

[0105] The device for predicting intermolecular interaction forces provided by the embodiment of the present invention can execute the method for predicting intermolecular interaction forces provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0106] Example 4

[0107] Figure 5 The present invention is a block diagram of an electronic device 10 that can be used to implement an embodiment of the present invention. 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 processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0108] like Figure 5As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0109] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0110] The processor 11 can be a variety of general-purpose and / or specialized processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors that run machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the method for predicting intermolecular interaction forces.

[0111] In some embodiments, the method for predicting intermolecular interaction forces may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for predicting intermolecular interaction forces described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the method for predicting intermolecular interaction forces in any other appropriate manner (e.g., by means of firmware).

[0112] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0113] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0114] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0115] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory 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).

[0116] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can 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), a blockchain network, and the Internet.

[0117] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0118] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0119] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for predicting intermolecular interaction forces, characterized in that: include: Obtain the molecules contained in the molecular pair to be predicted, and calculate the similarity between the atoms contained in each molecule through the truncated attention network; For each molecule, a pre-configured cutoff coefficient is used to perform truncation according to the similarity to obtain the substructure corresponding to each molecule; The interaction force between substructures is calculated based on the substructures corresponding to different molecules to obtain the interaction force between substructures; The interaction probability of the molecular pair to be predicted is obtained according to the inter-substructure interaction force, and the intermolecular interaction force situation is determined according to the interaction force probability.

2. The method according to claim 1, characterized in that The calculation of the similarity between atoms contained in each molecule by the truncated attention network includes: Determining the layers included in the truncated attention network, wherein the truncated attention network includes at least two layers; determining a separation distance based on a current level of the molecule through the truncated attention network; Target atoms are determined for the current atom according to the interval distances, and similarities between the current atom and each of the target atoms at the current level are calculated.

3. The method according to claim 2, characterized in that The method of truncating each molecule according to the similarity using a pre-configured truncation coefficient to obtain a substructure corresponding to each molecule includes: Determining non-correlated atoms of each molecule at each level whose similarity is less than the cutoff coefficient; The chemical bonds between the non-associated atoms are cut on the molecules to obtain substructures corresponding to each molecule, wherein substructures from different levels contain different numbers of atoms, and substructures from the same level contain the same number of atoms.

4. The method according to claim 1, wherein After truncating each molecule according to the similarity using a pre-configured truncation coefficient to obtain the substructure corresponding to each molecule, the method further includes: Normalizing the similarities between atoms contained in the substructure; Traversing the substructure to obtain features of a current atom traversed and similarities between atoms associated with the current atom; The feature of the current atom is updated according to the similarity between atoms associated with the current atom.

5. The method according to claim 1, characterized in that The calculation of the interaction forces based on the substructures corresponding to different molecules to obtain the interaction forces between the substructures includes: Score the importance of the substructures corresponding to each molecule; Delete the substructures whose importance scores are less than a preset threshold to obtain the target substructures corresponding to each molecule; The interaction force between substructures is obtained by calculating the interaction force based on the target substructures corresponding to different molecules.

6. The method according to claim 5, characterized in that The importance scoring of the substructures corresponding to each molecule includes: Obtaining a characteristic matrix of a substructure corresponding to each molecule, and projecting the characteristic matrix to obtain a one-dimensional vector; An importance score is performed on the one-dimensional vector to obtain the importance score value.

7. The method according to claim 5, characterized in that The method of calculating the interaction forces between target substructures corresponding to different molecules to obtain the interaction forces between substructures includes: determining a first molecule and a second molecule included in the molecule pair to be predicted; Obtaining a first vector of a first target substructure corresponding to the first molecule and a first weight corresponding to the first vector; Obtaining a second vector of a second target substructure corresponding to the second molecule and a second weight corresponding to the second vector; An inter-substructure force between the first target substructure and the second target substructure is calculated according to the first vector, the first weight, the second vector, and the second weight.

8. The method according to claim 1, characterized in that Obtaining the interaction probability of the molecular pair to be predicted based on the inter-substructure interaction force, and determining the intermolecular interaction force situation based on the interaction force probability, includes: Acquiring a learning matrix for the molecular pair to be predicted according to the inter-substructure forces; Calculating based on the inter-substructure forces corresponding to different molecules and the learning matrix to obtain the interaction probability of the molecular pair to be predicted; It is determined whether the interaction probability is 0. If so, it is determined that there is no intermolecular interaction force between the molecule to be predicted; otherwise, it is determined that there is an intermolecular interaction force between the molecule to be predicted.

9. A device for predicting intermolecular interaction forces, characterized in that: include: The similarity calculation module is used to obtain the molecules contained in the molecular pair to be predicted and calculate the similarity between the atoms contained in each molecule through the truncated attention network; A substructure extraction module, configured to obtain the substructure corresponding to each molecule by truncating each molecule using a pre-configured truncation coefficient according to the similarity; The substructure interaction force calculation module is used to calculate the interaction force between substructures according to the substructures corresponding to different molecules to obtain the interaction force between substructures; The intermolecular interaction force determination module is used to obtain the interaction probability of the molecular pair to be predicted based on the inter-substructure interaction force, and determine the intermolecular interaction force situation based on the interaction force probability.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program to be executed by the at least one processor, where the computer program is executed by the at least one processor so as to enable the at least one processor to perform the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method according to any one of claims 1 to 8 when the processor executes the computer instructions.

Images