Antibody generation method, device, electronic device, and storage medium

Abstract

This invention provides an antibody generation method, apparatus, electronic device, and storage medium. The method includes: S1: obtaining an antigen-antibody complex to be processed; S2: obtaining a representation map based on the antigen-antibody complex; S3: performing intra-chain updates on nodes based on neighbor node representations and coordinates of neighbor nodes, obtaining intra-chain updated node representations and coordinates; S4: performing inter-chain updates on nodes based on intra-chain updated node representations and coordinates, obtaining inter-chain updated node representations, coordinates, and an inter-chain updated representation map; S6: using the inter-chain updated representation map as the representation map, performing a preset number of rounds S3 to S5 to obtain a first updated representation map; S7: using the first updated representation map as the representation map, performing a preset number of rounds S3 to S6 to obtain a target updated representation map, and obtaining an antibody with a complete complementarity-determining region based on the target updated representation map. This invention enables the generated antibody to meet antigen specificity requirements.

Description

Technical Field

[0001] This invention relates to the field of antibody generation technology, and in particular to an antibody generation method, apparatus, electronic device and storage medium. Background Technology

[0002] Antibodies are protective proteins produced by the body in response to antigen stimulation. Antibodies recognize a unique characteristic of a specific foreign substance; this foreign target is the antigen.

[0003] According to relevant technologies, in order to accelerate the research and development process of antibody development and reduce research and development costs, artificial intelligence-assisted design is gradually being used in the antibody development process.

[0004] However, current AI-based antibody design methods have certain flaws, making it difficult to effectively design antibodies that meet the requirements. Summary of the Invention

[0005] This invention provides an antibody generation method, apparatus, electronic device, and storage medium, which enables the generation of antibodies with intact structure and meeting antigen specificity requirements.

[0006] This invention provides an antibody generation method, comprising: S1: obtaining an antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex is an antibody lacking a complementarity-determining region; S2: obtaining a representation map corresponding to the antigen-antibody complex based on the antigen-antibody complex to be processed, wherein the representation map includes multiple nodes, each node including a node representation and node coordinates; S3: performing an intra-chain update on the node based on the neighbor node representations and neighbor node coordinates of the node's neighbor nodes, obtaining an intra-chain updated node representation and intra-chain updated node coordinates of the node; S4: performing an intra-chain update on the node representation and intra-chain updated node coordinates of the node. New node coordinates, and inter-chain updates of the nodes are performed through an attention mechanism to obtain the inter-chain updated node representation and inter-chain updated node coordinates of the nodes; S5: Based on the inter-chain updated node representation and inter-chain updated node coordinates of each node, an inter-chain updated representation map corresponding to the representation map is obtained; S6: The inter-chain updated representation map is used as the representation map and S3 to S5 are repeated for a preset number of rounds to obtain a first updated representation map; S7: The first updated representation map is used as the representation map and S3 to S6 are repeated for a preset number of rounds to obtain a target updated representation map, and an antibody corresponding to the complementarity-determining region of the antigen-antibody complex to be processed is obtained based on the target updated representation map.

[0007] According to an antibody generation method provided by the present invention, the step of updating the node within the chain based on the neighbor node representations and coordinates of the neighbor nodes of the node to obtain the updated node representation within the chain specifically includes: obtaining relative coordinates within the chain based on the node coordinates of the node and the neighbor node coordinates of the neighbor nodes; performing a transpose multiplication on the relative coordinates within the chain to obtain the spatial information within the chain passed from the neighbor nodes to the node; concatenating the node representation of the node, the neighbor node representations of the neighbor nodes, the spatial information within the chain, and the edge type to obtain a first concatenation vector, and inputting the first concatenation vector into a first pre-trained fully connected neural network to obtain the information passed from the neighbor nodes to the node within the chain, wherein the edge type is the type of edge between the node and the neighbor nodes; summing the information passed from each neighbor node to the node within the chain to obtain summed information within the chain; concatenating the summed information within the chain with the node representation of the node to obtain a second concatenation vector, and inputting the second concatenation vector into a second pre-trained fully connected neural network to obtain the updated node representation within the chain.

[0008] According to an antibody generation method provided by the present invention, the step of updating the node within the chain based on the neighbor node representations and coordinates of the neighbor nodes of the node to obtain the updated node coordinates within the chain specifically includes: projecting the intra-chain transmission information through a third pre-trained fully connected neural network to obtain projected intra-chain transmission information with the same dimension as the relative coordinates within the chain; performing element-wise multiplication of the projected intra-chain transmission information with the relative coordinates within the chain to obtain a first node coordinate update amount corresponding to the neighbor nodes; averaging the first node coordinate update amounts corresponding to each of the neighbor nodes to obtain an intra-chain target node coordinate update amount; and performing residual concatenation processing on the intra-chain target node coordinate update amount and the node coordinates of the node to obtain the updated node coordinates within the chain.

[0009] According to an antibody generation method provided by the present invention, based on the intra-chain updated node representation, an attention mechanism is used to perform inter-chain updates on the node to obtain the inter-chain updated node representation of the node. Specifically, the method includes: determining the value vector of the neighbor node to the node, and the attention weight of the node to the neighbor node; multiplying the attention weight of the node to the neighbor node with the value vector of the neighbor node to obtain inter-chain transmission information passed from the neighbor node to the node; summing the inter-chain transmission information passed from each neighbor node to the node to obtain summed inter-chain transmission information; and summing the summed inter-chain transmission information with the intra-chain updated node representation to obtain the inter-chain updated node representation of the node.

[0010] According to an antibody generation method provided by the present invention, the inter-chain updated node coordinates of the node are obtained by performing inter-chain updated node coordinates based on the intra-chain updated node coordinates through an attention mechanism. Specifically, the method includes: projecting the value vector of the neighbor node to the node through a fourth pre-trained fully connected neural network to obtain a projected value vector with the same dimension as the inter-chain relative coordinates, wherein the inter-chain relative coordinates are determined based on the intra-chain updated node coordinates of the node and the intra-chain updated neighbor node coordinates of the neighbor node; multiplying the projected value vector element-wise with the inter-chain relative coordinates, and then multiplying it with the attention weight of the node to the neighbor node to obtain a second node coordinate update amount corresponding to the neighbor node; summing the second node coordinate update amounts corresponding to each neighbor node to obtain an inter-chain target node coordinate update amount; and performing a residual connection process on the inter-chain target node coordinate update amount and the intra-chain updated node coordinates of the node to obtain the inter-chain updated node coordinates of the node.

[0011] According to an antibody generation method provided by the present invention, the value vector of the neighbor node to the node is determined in the following manner: the inter-chain space information passed by the neighbor node to the node and the intra-chain updated neighbor node representation of the neighbor node are concatenated and then input into a fifth pre-trained fully connected neural network to obtain the value vector of the neighbor node to the node.

[0012] According to an antibody generation method provided by the present invention, the attention weights of the node to the neighboring nodes are determined in the following manner: the intra-chain updated node representation of the node is input into a sixth pre-trained fully connected neural network to obtain the query vector of the node; the inter-chain spatial information passed from the neighboring nodes to the node and the intra-chain updated neighboring node representation of the neighboring nodes are concatenated and input into a seventh pre-trained fully connected neural network to obtain the key vector of the neighboring nodes to the node; the query vector of the node is multiplied by the key vector of each neighboring node to the node to obtain the multiplication result, and the multiplication result is processed through a softmax activation layer to obtain the attention weights of the node to the neighboring nodes.

[0013] According to an antibody generation method provided by the present invention, the inter-chain spatial information passed from the neighboring node to the node is determined by performing a transpose multiplication on the relative coordinates between the chains to obtain the inter-chain spatial information passed from the neighboring node to the node.

[0014] According to an antibody generation method provided by the present invention, the neighboring nodes are determined in the following manner: nodes in the representation graph that are at a preset distance from the node are taken as the neighboring nodes of the node.

[0015] The present invention also provides an antibody generation device, comprising: a first module for acquiring an antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex is an antibody lacking a complementarity-determining region; a second module for obtaining a representation graph corresponding to the antigen-antibody complex to be processed, wherein the representation graph includes multiple nodes, each node including a node representation and node coordinates; a third module for performing an intra-chain update on the node based on the neighbor node representations and neighbor node coordinates of the node's neighbor nodes, obtaining an intra-chain updated node representation and intra-chain updated node coordinates of the node; and a fourth module for performing an attention mechanism based on the intra-chain updated node representation and intra-chain updated node coordinates. The nodes are updated inter-chain to obtain the inter-chain updated node representation and inter-chain updated node coordinates; the fifth module is used to obtain an inter-chain updated representation map corresponding to the representation map based on the inter-chain updated node representation and inter-chain updated node coordinates of each node; the sixth module is used to repeatedly execute the steps of the third to fifth modules for a preset number of rounds using the inter-chain updated representation map as the representation map to obtain a first updated representation map; the seventh module is used to repeatedly execute the steps of the third to sixth modules for a preset number of rounds using the first updated representation map as the representation map to obtain a target updated representation map, and obtain an antibody corresponding to the complementarity-determining region of the antigen-antibody complex to be processed based on the target updated representation map.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the antibody generation method as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the antibody generation method as described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the antibody generation method as described above.

[0019] The antibody generation method, apparatus, electronic device, and storage medium provided by this invention, by processing the antigen-antibody complex to be processed, can utilize complete antigen-antibody complex information, thereby more effectively generating antibodies that complete the complementarity-determining region. Furthermore, by alternately performing intra-chain and inter-chain coding processing on the representation map corresponding to the antigen-antibody complex to be processed, the intra-chain and inter-chain information in the antigen-antibody complex to be processed can be fully utilized, resulting in a structurally complete antibody that meets the antigen's specificity requirements. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts illustrating the antibody generation method provided by the present invention;

[0022] Figure 2 This is a schematic diagram of an application scenario provided by the present invention, which obtains a representation diagram corresponding to the antigen-antibody complex to be treated.

[0023] Figure 3 This is a schematic diagram illustrating an application scenario of alternating intra-chain and inter-chain updates of a representation graph, as provided by the present invention.

[0024] Figure 4 This is a schematic diagram illustrating an application scenario of the antibody generation method provided by the present invention;

[0025] Figure 5 This is a schematic diagram of the process of updating the node representation of the node based on the neighbor node representation and neighbor node coordinates of the node provided by the present invention, and updating the node representation of the node within the chain.

[0026] Figure 6 This is a schematic diagram of the process of updating the coordinates of a node by performing an in-chain update on a node based on the neighbor node representation and neighbor node coordinates of a node provided by the present invention.

[0027] Figure 7 This is a schematic diagram of the process provided by the present invention, which uses an attention mechanism to update nodes between chains based on intra-chain node updates to obtain the inter-chain node updates of nodes.

[0028] Figure 8This is a schematic diagram of the process provided by the present invention to update the coordinates of nodes between chains by using an attention mechanism based on the intra-chain updated node coordinates to obtain the inter-chain updated node coordinates of the nodes.

[0029] Figure 9 This is a flowchart illustrating the process of determining the attention weight of a node to its neighboring nodes, provided by the present invention.

[0030] Figure 10 This is a schematic diagram of the antibody generation device provided by the present invention;

[0031] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0033] The antibody generation method provided by this invention incorporates information about both the antigen and the light chain by extending the input from the heavy chain of the antibody to the entire antigen-antibody complex to be treated. Since the complementarity-determining regions (CDRs) on the antibody have the greatest impact on antigen-antibody binding, in application, the CDR regions of the input antigen-antibody complex to be treated are masked (i.e., CDR regions are removed). Based on the antibody generation method of this application, the generated antibody has a more optimized CDR region to meet the specificity requirements of the antigen.

[0034] Furthermore, in order to extract information related to the CDR region from complexes with a large amount of information (corresponding to antigen-antibody complexes to be treated), the antibody generation method alternately performs intra-chain coding and inter-chain coding processing on the representation map corresponding to the antigen-antibody complex to be treated, so that the intra-chain and inter-chain information in the antigen-antibody complex to be treated can be fully utilized, thereby ensuring that the generated antibody structure is intact and meets the antigen specificity requirements.

[0035] To further illustrate the antibody generation method provided by this invention, the following will be combined with... Figure 1 Please provide an explanation.

[0036] Figure 1 This is one of the flowcharts of the antibody generation method provided by the present invention.

[0037] In an exemplary embodiment of the present invention, combined with Figure 1 As can be seen, the antibody generation method may include steps 110 to 170, and each step will be described below.

[0038] In step 110, an antigen-antibody complex to be treated is obtained, wherein the antibody in the antigen-antibody complex to be treated is an antibody lacking the complementarity-determining region.

[0039] In one embodiment, the antigen-antibody complex to be treated may include the structure and sequence of the antigen-antibody complex. In the antigen-antibody complex to be treated, the antibody is an antibody lacking a complementarity-determining region (CDR). Since the missing CDR is the region that has the greatest impact on antigen-antibody binding, by completing the missing CDR in the antigen-antibody complex to be treated, the generated antibody will effectively meet the antigen specificity requirements.

[0040] In step 120, a representation diagram corresponding to the antigen-antibody complex to be treated is obtained based on the antigen-antibody complex to be treated. The representation diagram includes multiple nodes, and each node includes a node representation and node coordinates.

[0041] In one embodiment, the antigen-antibody complex to be treated can be processed into a representation graph corresponding to the antigen-antibody complex based on the atomic coordinates of the structure in the complex. In one example, the representation graph is a representation graph in tensor form. The representation graph may include multiple nodes, and each node may include a node representation (also known as a vector representation) and node coordinates.

[0042] Figure 2 This is a schematic diagram illustrating an application scenario of obtaining a representation diagram corresponding to the antigen-antibody complex based on the antigen-antibody complex to be treated, as provided by the present invention.

[0043] Combination Figure 2 To illustrate, Figure A shows the molecular formula of the antigen-antibody complex to be treated, Figure B is a simplified structural diagram of Figure A, and Figure C is a representation of the antigen-antibody complex shown in Figure A. (Combined with...) Figure 2 As can be seen, each node in the diagram represents an amino acid residue in the complex. The coordinates of each node contain the coordinates of the backbone atoms within the residue. Figure 2 It is known that the skeleton atom consists of 4 atoms. The skeleton atom can be understood as the most important atom related to the residue.

[0044] In another example, each node in the graph is represented as a node with a one-hot vector form. In application, an embedding layer can be used to transform the node representation with one-hot vector form into a node representation with a continuous vector form. That is, the embedding layer can assign a corresponding continuous vector as the node representation based on the type of amino acid represented by the node.

[0045] In step 130, based on the neighbor node representations and coordinates of the neighbor nodes of the node, the node is updated within the chain to obtain the updated node representation and coordinates within the chain.

[0046] In one embodiment, the neighboring nodes of a node can be determined based on the node, and the node can be updated within the chain based on the neighboring node representation and the neighboring node coordinates to obtain the updated node representation and the updated node coordinates within the chain.

[0047] In another embodiment, neighboring nodes can be determined in the following way:

[0048] The nodes in the graph that are at a preset distance from the node are considered as the node's neighbor nodes.

[0049] The preset distance can be adjusted according to the actual situation, and is not specifically limited in this embodiment.

[0050] It should be noted that the neighbor node representation is a node representation about the neighbor node. The neighbor node coordinates are node coordinates about the neighbor node. The node representation updated within the chain is the node representation obtained after the node has undergone an in-chain update. The node coordinates updated within the chain are the node coordinates obtained after the node has undergone an in-chain update.

[0051] In step 140, based on the intra-chain updated node representation and intra-chain updated node coordinates, the node is updated between chains through an attention mechanism to obtain the inter-chain updated node representation and inter-chain updated node coordinates.

[0052] In step 150, based on the inter-chain update node representation and inter-chain update node coordinates of each node, an inter-chain update representation graph corresponding to the representation graph is obtained.

[0053] In one embodiment, intra-chain updates can be performed on each node in the representation graph based on step 130, thereby obtaining an intra-chain updated representation graph corresponding to the representation graph. Further, based on the intra-chain updated node representations and coordinates of each node in the intra-chain updated representation graph, inter-chain updates can be performed on each node using an attention mechanism, thereby obtaining the inter-chain updated node representations and coordinates of each node.

[0054] In another embodiment, based on the inter-chain update node representation and coordinates of each node, an inter-chain update representation graph corresponding to the representation graph can be obtained. Based on this, each node on the representation graph undergoes a round of intra-chain encoding and inter-chain encoding processing sequentially. In application, by alternately performing intra-chain and inter-chain encoding processing on the representation graph corresponding to the antigen-antibody complex to be processed, the intra-chain and inter-chain information in the antigen-antibody complex to be processed can be fully utilized, thus laying the foundation for generating an antibody with a complete structure that meets the antigen's specificity requirements.

[0055] In step 160, steps 130 to 150 are repeated for a preset number of rounds using the inter-chain update representation graph as the representation graph to obtain the first update representation graph.

[0056] In one embodiment, the inter-chain update representation graph obtained by sequentially performing one round of intra-chain encoding and inter-chain encoding can be used as the representation graph, and steps 130 to 150 are repeated for a preset number of rounds to obtain a first update representation graph. It is understood that the first update representation graph is the representation graph obtained by alternating intra-chain encoding and inter-chain encoding processes for a preset number of rounds. The preset number of rounds can be adjusted according to actual circumstances; in this embodiment, the preset number of rounds is not specifically limited.

[0057] Figure 3 This is a schematic diagram illustrating an application scenario where the graph is alternately updated within and between chains, as provided by the present invention.

[0058] Combination Figure 3 To explain, during the alternating intra-chain and inter-chain updates of the representation graph, intra-chain encoding and inter-chain encoding processes can be performed sequentially. After one round of intra-chain and inter-chain encoding processes, the sequence and structure of the complex can be predicted simultaneously, resulting in an inter-chain updated representation graph. Furthermore, this inter-chain updated representation graph is then used as the new representation graph for another round of intra-chain and inter-chain encoding processes. The predicted sequence and structure are used as new inputs for the intra-chain encoding process. After a predetermined number of rounds, the first updated representation graph is obtained.

[0059] It should be noted that since the antigen-antibody complex to be treated includes antibodies, and antibodies are large protein molecules, the antigen-antibody complex to be treated is also a multi-chain structure. Intra-chain coding processing refers to coding within the same chain of the antigen-antibody complex to be treated, while inter-chain coding processing refers to coding between different chains of the antigen-antibody complex to be treated.

[0060] In step 170, steps 130 to 160 are repeated for a preset number of rounds using the first updated representation map as the representation map to obtain the target updated representation map, and the antibody corresponding to the complementarity determination region of the antigen-antibody complex to be processed is obtained based on the target updated representation map.

[0061] In one embodiment, the first updated representation map is used again as a representation map for sequential intra-chain and inter-chain encoding processing to obtain an inter-chain updated representation map. Further, the inter-chain updated representation map is used again as a representation map for sequential intra-chain and inter-chain encoding processing. After a preset number of rounds of processing, the first updated representation map is obtained. Further, the first updated representation map is used as a representation map to repeat the aforementioned steps for a preset number of rounds to obtain the target updated representation map. The CDR region of the antibody in the target updated representation map is determined based on a large amount of information in the complex, thereby ensuring that the generated antibody structure is complete and meets the antigen specificity requirements.

[0062] It is understandable that, after processing for a preset number of rounds to obtain the first updated representation graph, if the preset number of rounds is n, and if the preset number of rounds is also n when repeatedly executing steps 130 to 160 using the first updated representation graph as the representation graph, then, in the process of obtaining the final target updated representation graph, the intra-chain encoding processing and inter-chain encoding processing are performed n times respectively. 2 Second-rate.

[0063] In this embodiment, by using a non-autoregressive progressive decoding method, the generated antibody can be iteratively brought closer to the target result in fewer steps, thereby making the generation process more efficient and accumulating fewer errors.

[0064] The antibody generation method provided by this invention can be represented by a virtual device. That is, the corresponding virtual device may include an input module, an intra-chain coding module, an inter-chain coding module, and an output module. Specifically, steps 110 to 120 correspond to the input module, step 130 corresponds to the intra-chain coding module, steps 140 and 150 correspond to the inter-chain coding module, and steps 160 and 170 correspond to the output module.

[0065] Figure 4 This is a schematic diagram illustrating an application scenario of the antibody generation method provided by the present invention.

[0066] In one embodiment, combined with Figure 4As can be seen, the input module converts the antigen-antibody complex to be processed into a representation map. Further, it is processed alternately through the intra-chain module (corresponding to the intra-chain encoding module) and the inter-chain module (corresponding to the inter-chain encoding module) for a preset number of rounds. The first output module outputs a first updated representation map, which is then used again as the representation map to repeat the preset number of rounds. This process is repeated multiple times, alternating between the intra-chain and inter-chain modules, to obtain the target updated representation map. In this embodiment, by encoding the entire map using an intra-chain graph encoder and an inter-chain interacting graph encoder, the information within and between chains can be fully transmitted.

[0067] In this embodiment, the intra-chain module and inter-chain module alternately update the representation vectors and coordinates of the nodes. After obtaining the representation vector and coordinates of each node, the final design result is obtained through the output module. The output module works in an iterative manner, with each iteration using the generation result of the previous iteration as input to update the results of all masked nodes and update the representation vectors and coordinates.

[0068] The antibody generation method provided by this invention, by treating the antigen-antibody complex, can utilize complete antigen-antibody complex information to more effectively generate antibodies that complete the complementarity-determining region. Furthermore, by alternately performing intra-chain and inter-chain coding processing on the representation map corresponding to the antigen-antibody complex, the intra-chain and inter-chain information in the antigen-antibody complex can be fully utilized, resulting in a structurally complete antibody that meets the antigen's specificity requirements.

[0069] To further illustrate the antibody generation method provided by this invention, the following will be combined with... Figure 5 Please provide an explanation.

[0070] Figure 5 This is a schematic diagram of the process of updating the node representation within the chain by using the neighbor node representation and neighbor node coordinates of the node based on the node's neighbor node, as provided by the present invention.

[0071] In an exemplary embodiment of the present invention, combined with Figure 5 As can be seen, updating a node within the chain based on the neighbor node representation and neighbor node coordinates can include steps 510 and 550, which will be described in detail below.

[0072] In step 510, relative coordinates within the chain are obtained based on the node coordinates of the node and the neighbor node coordinates of the neighbor node.

[0073] In one embodiment, for the current node i and one of its neighboring nodes j, the relative coordinates of the two nodes can be obtained by subtracting the node coordinates from the neighboring node coordinates (corresponding to relative coordinates within the chain). Where l represents the number of rounds, This represents the relative coordinates within the chain obtained during the l-th round of encoding. It can be understood that completing one round of encoding and one round of inter-chain encoding can be considered as completing one round of encoding.

[0074] In step 520, the relative coordinates within the chain are transposed and multiplied to obtain the spatial information within the chain passed from the neighboring node to the node.

[0075] In one embodiment, the relative coordinates within the chain can be transposed and multiplied to obtain the spatial information within the chain that neighbor node j passes to node i. This embodiment enables the spatial information within the chain to be invariant to rotation and translation.

[0076] In step 530, the node representation of the node, the neighbor node representation of the neighbor node, the intra-chain spatial information, and the edge type are concatenated to obtain a first concatenated vector. This first concatenated vector is then input into a first pre-trained fully connected neural network to obtain the intra-chain transmission information passed from the neighbor node to the node. Here, the edge type refers to the type of edge connection between the node and its neighbor nodes.

[0077] In one embodiment, node i can be represented as a node. The neighbor node representation of neighbor node j In-chain space information passed from neighbor node j to node i And the types of edges between node i and its neighbor node j The first concatenated vector is formed by concatenating the vectors. This first concatenated vector is then used as the first pre-trained fully connected neural network φ. m By taking the input, we can obtain the intra-chain information passed from neighbor node j to node i.

[0078] In one example, neighbor node j passes the in-chain information to node i. This can be achieved using the following formula (1):

[0079]

[0080] It should be noted that the first pre-trained fully connected neural network φ m It can be achieved through prior training.

[0081] The edge type can be preset. Since neighbor node j and node i can be connected directly within a chain or indirectly between chains, the edge type can be determined based on the connection method between neighbor node j and node i.

[0082] In step 540, the intra-chain transmission information passed to the node by each neighbor node is summed to obtain the summed intra-chain transmission information.

[0083] In one embodiment, for node i, the intra-chain transmission information passed from all its neighboring nodes can be calculated and summed to obtain the summed intra-chain transmission information.

[0084] In step 550, the summed intra-chain information is concatenated with the node representation to obtain a second concatenated vector, which is then input into a second pre-trained fully connected neural network to obtain the intra-chain updated node representation of the node.

[0085] In one embodiment, the summed information passed within the chain can be compared with the node representation of node i. Row concatenation is performed to obtain a second concatenated vector. This second concatenated vector is then input into a second pre-trained fully connected neural network φ. h Thus, the in-chain update node representation h of node i can be obtained. i (l+0.5) That is, the updated node representation of node i is obtained.

[0086] Among them, h i (l+0.5) This represents the updated intra-chain node representation of node i obtained during the (l+0.5)th round of encoding. It can be understood that completing one intra-chain encoding process and one inter-chain encoding process sequentially constitutes completing one round of encoding.

[0087] Among them, the second pre-trained fully connected neural network φ h It can be achieved through prior training.

[0088] In one embodiment, the in-chain update of a node representation can be achieved using the following formula (2):

[0089]

[0090] It should be noted that during the in-chain encoding process, all neighboring nodes, except for the global node, are in-chain neighbors. For example, the set of neighboring nodes of node i is represented as N(i|ε). in ).

[0091] Figure 6This is a schematic diagram of the process of updating the coordinates of a node by performing an in-chain update on a node based on the neighbor node representation and neighbor node coordinates of a node, as provided by the present invention.

[0092] The following will combine Figure 6 This paper explains the process of updating the coordinates of a node within its chain based on the neighbor node representation and coordinates of its neighbor nodes.

[0093] In an exemplary embodiment of the present invention, combined with Figure 6 As can be seen, updating a node within its chain based on the neighbor node representation and neighbor node coordinates can include steps 610 to 640, which will be described in detail below.

[0094] In step 610, the intra-chain transmitted information is projected through a third pre-trained fully connected neural network to obtain the projected intra-chain transmitted information with the same dimension as the relative coordinates within the chain.

[0095] In one embodiment, information can be transferred within the chain. After the third pre-trained fully connected neural network φ Z Perform projection processing to obtain the relative coordinates with the chain. Information is passed within the chain after projection processing with the same dimension.

[0096] Among them, the third pre-trained fully connected neural network φ Z It can be obtained through pre-training.

[0097] In step 620, the information transmitted within the chain after projection processing is multiplied element-wise with the relative coordinates within the chain to obtain the first node coordinate update amount corresponding to the neighboring node.

[0098] In step 630, the coordinate update amounts of the first node corresponding to each neighbor node are averaged to obtain the coordinate update amounts of the target node within the chain.

[0099] In one embodiment, the information transmitted within the chain after projection processing can be compared with the relative coordinates within the chain. Perform element-wise multiplication to obtain the coordinate update of the first node corresponding to neighbor node j. Further, average the results of the above element-wise multiplication on all neighbor nodes of node i to obtain the coordinate update of the target node within the chain. Update the coordinates of the target node within the chain. The amount of time used to update the coordinates of node i.

[0100] In step 640, the target node coordinate update amount and the node coordinates within the chain are processed by residual connection to obtain the updated node coordinates within the chain.

[0101] In one embodiment, the coordinates of the target node within the chain can be updated using a residual join approach. and node coordinates Process to obtain the updated node coordinates within the chain.

[0102] In one embodiment, updating the node coordinates within the chain can be achieved using formula (3):

[0103]

[0104] Among them, Z i (l+0.5) This represents the updated node coordinates within the chain of node i, which are the updated node coordinates within the chain of node i obtained during the encoding process in round (l+0.5).

[0105] Figure 7 This is a schematic diagram of the process provided by the present invention, which uses an attention mechanism to update nodes between chains to obtain the node representation of the node with an inter-chain update.

[0106] The following will combine Figure 7 This paper explains the process of updating the node representation between chains by using an attention mechanism to update the node representation between chains based on the intra-chain updated node representation.

[0107] In an exemplary embodiment of the present invention, combined with Figure 7 As can be seen, based on the intra-chain updated node representation, the inter-chain updated node representation can be obtained by using the attention mechanism to update the node. This can include steps 710 to 740, which will be described in detail below.

[0108] In step 710, the value vector of a neighbor node to a node and the attention weight of a node to its neighbor node are determined.

[0109] In application, the intra-chain updated node representations and coordinates of each node obtained after intra-chain encoding can be used to obtain an intra-chain updated representation graph corresponding to the representation graph. This intra-chain updated representation graph is then used as input for the inter-chain encoding process.

[0110] In the inter-chain encoding process, the neighbor node information can be expanded and aggregated through the attention mechanism. The query vector, key vector, and value vector required to calculate the attention weights are also introduced with spatial features that are invariant to rotation and translation by multiplying the relative coordinates element by element, so as to ensure that the result of the attention weights is not affected by the rotation and translation of the input.

[0111] In one embodiment, the value vector of a neighboring node can be determined in the following way:

[0112] The inter-chain spatial information passed from neighboring nodes to the node, and the neighboring node's intra-chain update representation are concatenated and then input into the fifth pre-trained fully connected neural network to obtain the value vector of the neighboring node to the node.

[0113] In one embodiment, the inter-chain space information passed from a neighboring node to a node can be determined in the following way:

[0114] relative coordinates between chains Perform transpose multiplication to obtain the inter-chain space information passed from neighboring nodes to the node.

[0115] In one example, for node i, the inter-chain space information passed to node i by its neighbor node j can be used. And the in-chain update of neighbor node j's neighbor node representation h j (l+0.5) After being concatenated, the result is used as input and processed by the fifth pre-trained fully connected neural network φ. v This yields the value vector v of neighbor node j for node i. ij .

[0116] In one embodiment, the value vector v of neighbor node j relative to node i is... ij This can be achieved using the following formula (4):

[0117]

[0118] Figure 9 This is a flowchart illustrating the process of determining the attention weight of a node to its neighboring nodes, as provided by the present invention.

[0119] In one embodiment, combined with Figure 9 As can be seen, the attention weight of a node to its neighboring nodes can include steps 910 to 930, and each step will be described below.

[0120] In step 910, the node's in-chain update node representation is input into the sixth pre-trained fully connected neural network to obtain the node's query vector.

[0121] In one embodiment, for node i, the in-chain update node representation h of node i can be... i (l+0.5) Input to the sixth pre-trained fully connected neural network φ q This yields the query vector q of the node. i Among them, the sixth pre-trained fully connected neural network φ q It can be obtained through pre-training.

[0122] In one embodiment, the node's query vector q t This can be achieved using the following formula (5):

[0123] q i =φ q (h i (l+0.5) (5)

[0124] In step 920, the inter-chain spatial information passed from the neighboring node to the node and the neighboring node's intra-chain updated neighboring node representation are concatenated and then input into the seventh pre-trained fully connected neural network to obtain the key vector of the neighboring node to the node.

[0125] In one example, for node i, the inter-chain space information passed to node i by its neighbor node j can be used. And the in-chain update of neighbor node j's neighbor node representation h j (l+0.5) After being concatenated, the result is used as input and processed through the seventh pre-trained fully connected neural network φ. k This yields the key vector k of neighbor node j to node i. ij Among them, the seventh pre-trained fully connected neural network φ k It can be obtained through pre-training.

[0126] In one embodiment, the key vector k between neighboring nodes is... ij This can be achieved using the following formula (6):

[0127]

[0128] In step 930, the query vector of a node is multiplied by the key vector of each neighboring node to obtain the multiplication result. The multiplication result is then processed through a softmax activation layer to obtain the attention weights of the node to its neighboring nodes.

[0129] In one embodiment, the query vector q of the node can be... i Each node is compared with the k of its neighboring nodes. ijPerform a dot product operation to obtain the result. Then, process the result through a softmax activation layer to obtain the attention weight α of node i to its neighbor node j. ij .

[0130] In one embodiment, the attention weight α of node i to its neighbor node j is... ij This can be achieved using formula (7):

[0131]

[0132] Here, exp(·) represents the exponential form of the natural constant e.

[0133] It should be noted that during inter-chain processing, all neighbors are inter-chain neighbors, meaning the neighboring nodes of a node are obtained based on the intra-chain update of the representation graph corresponding to the representation graph. For example, the set of neighboring nodes of node i is represented as N(i|ε ex ).

[0134] In step 720, the attention weight of a node to its neighboring node is multiplied by the value vector of the neighboring node to the node to obtain the inter-chain information passed from the neighboring node to the node.

[0135] In step 730, the inter-chain transmission information passed from each neighbor node to the node is summed to obtain the summed inter-chain transmission information.

[0136] In one embodiment, for any neighbor node j, the attention weight α of the corresponding node i assigned to neighbor node j can be determined. ij The value vector v of its corresponding neighbor node j to node i ij Perform multiplication to obtain the inter-chain information passed from neighboring nodes to the node. Furthermore, the inter-chain information passed from each neighbor node to the node can be summed to obtain the summed inter-chain information.

[0137] In step 740, the summed inter-chain transmission information is summed with the intra-chain updated node representation to obtain the inter-chain updated node representation of the node.

[0138] In one embodiment, the inter-chain transmission information passed from all neighboring nodes to node i is summed (corresponding to the summed inter-chain transmission information), and then compared with the intra-chain update node representation h of node i itself. i (l+0.5) Add them together to obtain the inter-chain update node representation h of node i. i (l+1) .

[0139] In one embodiment, the inter-chain update node representation h of node i i (l+1) This can be achieved using formula (8):

[0140]

[0141] It should be noted that h i (l+1) This represents the in-chain updated node representation of node i obtained during the (l+1)th round of encoding processing.

[0142] Figure 8 This is a schematic diagram of the process provided by the present invention, which uses an attention mechanism to update nodes between chains based on intra-chain updated node coordinates to obtain the inter-chain updated node coordinates.

[0143] The following will combine Figure 8 This paper explains the process of updating the coordinates of nodes between chains by using an attention mechanism based on the intra-chain updated node coordinates to obtain the inter-chain updated node coordinates.

[0144] In an exemplary embodiment of the present invention, combined with Figure 8 As can be seen, updating the node between chains based on the intra-chain updated node coordinates and using the attention mechanism to obtain the inter-chain updated node coordinates can include steps 810 to 840, and each step will be described below.

[0145] In step 810, the value vector of the neighbor node to the node is projected through the fourth pre-trained fully connected neural network to obtain the projected value vector with the same dimension as the relative coordinates between the chains. The relative coordinates between the chains are determined based on the updated coordinates of the node within the chain and the updated coordinates of the neighbor node within the chain.

[0146] In step 820, the value vector after projection processing is multiplied element-wise with the relative coordinates between chains, and then multiplied with the attention weight of the node to its neighboring nodes to obtain the second node coordinate update amount corresponding to the neighboring node.

[0147] In one embodiment, neighboring node j and node i are used as examples for illustration. The value vector v of neighboring node j relative to node i can be... ij After the fourth pre-trained fully connected neural network φ n Perform projection processing to obtain the relative coordinates between the chains. After the value vector is processed by projection with the same dimension, it is then compared with the relative coordinates between the chains. Perform element-wise multiplication, then multiply with the attention weight α of node i to its neighbor node j. ijPerform multiplication to obtain the updated coordinates of the second node corresponding to the neighbor node j.

[0148] In step 830, the coordinate update amounts of the second node corresponding to each neighbor node are summed to obtain the coordinate update amount of the target node between the chains.

[0149] In step 840, the inter-chain target node coordinate update amount and the node's intra-chain updated node coordinate are processed by residual connection to obtain the node's inter-chain updated node coordinate.

[0150] In one embodiment, all neighboring nodes of node i are processed according to steps 810 to 820, and the resulting second node coordinate update amounts corresponding to each neighboring node are summed to obtain the inter-chain target node coordinate update amount.

[0151] Furthermore, update the coordinates of the target nodes between chains. Update node coordinates within the node's chain. Perform residual join processing to obtain the inter-link updated node coordinates.

[0152] In one embodiment, the inter-chain update node coordinates to a node can be determined using the following formula (9):

[0153]

[0154] Among them, Z i (l+1) This represents the inter-chain updated node coordinates of node i, which is the intra-chain updated node coordinates of node i obtained during the encoding process in round (l+1).

[0155] The advantages of the antibody generation method provided by this invention are as follows:

[0156] (1) It utilizes complete antigen-antibody complex information, which can more effectively generate antibodies that bind to antigens;

[0157] (2) Using a graph neural network model with equivariance in three-dimensional space, it can automatically handle the rotation and translation of the input;

[0158] (3) Non-autoregressive iterative decoding is efficient and has low cumulative error;

[0159] (4) Compared with the traditional experimental method of obtaining antibodies, the antibody generation process in this application is low-cost and effective.

[0160] In one example, for the antigen-antibody complex numbered 2vis in the PDB, the antibody generation method provided by this invention can generate an effective CDR region. Based on existing affinity calculation models, the CDR-H3 generated by the antibody generation method provided by this invention (i.e., the antibody generated from the antigen-antibody complex numbered 2vis in the PDB according to this invention) shows an affinity improvement of -5.13 compared to the original CDR-H3, where a more negative affinity is better.

[0161] As described above, the antibody generation method provided by this invention, by processing the antigen-antibody complex to be treated, can utilize complete antigen-antibody complex information, thereby more effectively generating antibodies that complete the complementarity-determining region. Furthermore, by alternately performing intra-chain and inter-chain coding processing on the representation map corresponding to the antigen-antibody complex to be treated, the intra-chain and inter-chain information in the antigen-antibody complex to be treated can be fully utilized, ensuring that the generated antibody structure is complete and meets the antigen specificity requirements.

[0162] Based on the same concept, the present invention also provides an antibody generation device.

[0163] The antibody generation apparatus provided by the present invention is described below. The antibody generation apparatus described below and the antibody generation method described above can be referred to in correspondence.

[0164] Figure 10 This is a schematic diagram of the antibody generation device provided by the present invention.

[0165] In an exemplary embodiment of the present invention, combined with Figure 10 As can be seen, the antibody generation device may include the first module 1010 to the seventh module 1070, and each module will be described below.

[0166] The first module 1010 can be configured to obtain an antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex to be processed is an antibody lacking a complementarity-determining region.

[0167] The second module 1020 can be configured to obtain a representation graph corresponding to the antigen-antibody complex to be processed, based on the antigen-antibody complex to be processed, wherein the representation graph includes multiple nodes, and each node includes a node representation and node coordinates.

[0168] The third module 1030 can be configured to perform intra-chain updates on a node based on the neighbor node representation and neighbor node coordinates of the node's neighbor node, thereby obtaining the intra-chain updated node representation and intra-chain updated node coordinates.

[0169] The fourth module 1040 can be configured to perform inter-chain updates on nodes based on the intra-chain updated node representation and intra-chain updated node coordinates, using an attention mechanism to obtain the inter-chain updated node representation and inter-chain updated node coordinates.

[0170] The fifth module 1050 can be configured to update the node representation and node coordinates between each node to obtain an inter-chain updated representation graph corresponding to the representation graph.

[0171] The sixth module 1060 can be configured to repeatedly execute the steps performed by the third to fifth modules for a preset number of rounds using the inter-chain update representation graph as a representation graph to obtain the first update representation graph.

[0172] The seventh module 1070 can be configured to repeatedly execute the steps performed by the third to sixth modules for a preset number of rounds using the first updated representation map as the representation map, to obtain a target updated representation map, and to obtain an antibody corresponding to the complementarity-determining region of the antigen-antibody complex to be processed based on the target updated representation map.

[0173] In an exemplary embodiment of the present invention, the third module 1030 may perform an in-chain update on the node based on the neighbor node representations and coordinates of the node's neighbor nodes to obtain the updated node representation of the node:

[0174] Based on the node coordinates of a node and the neighbor node coordinates of its neighbor nodes, the relative coordinates within the chain are obtained. The relative coordinates within the chain are then transposed and multiplied to obtain the spatial information within the chain passed from the neighbor node to the node. The node representation of the node, the neighbor node representations of its neighbor nodes, the spatial information within the chain, and the edge types are concatenated to obtain a first concatenated vector. This first concatenated vector is then input into a first pre-trained fully connected neural network to obtain the information transmitted within the chain from the neighbor node to the node, where the edge type is the type of edge between the node and its neighbor node. The information transmitted within the chain from each neighbor node to the node is then summed to obtain summed information within the chain. This summed information within the chain is then concatenated with the node representation of the node to obtain a second concatenated vector. This second concatenated vector is then input into a second pre-trained fully connected neural network to obtain the updated node representation within the chain.

[0175] In an exemplary embodiment of the present invention, the third module 1030 may perform an in-chain update on the node based on the neighbor node representations and coordinates of the node's neighbor nodes to obtain the updated node coordinates:

[0176] The intra-chain transmitted information is projected through a third pre-trained fully connected neural network to obtain the projected intra-chain transmitted information with the same dimension as the relative coordinates within the chain. The projected intra-chain transmitted information is then multiplied element-wise with the relative coordinates within the chain to obtain the first node coordinate update corresponding to the neighboring nodes. The first node coordinate update corresponding to each neighboring node is averaged to obtain the target node coordinate update within the chain. The target node coordinate update and the node coordinates are then joined using a residual connection to obtain the updated node coordinates within the chain.

[0177] In an exemplary embodiment of the present invention, the fourth module 1040 may perform inter-chain updates on nodes based on the intra-chain updated node representation and through an attention mechanism to obtain the inter-chain updated node representation of the node:

[0178] Determine the value vector of a neighbor node to a node, and the attention weight of a node to its neighbor node; multiply the attention weight of a node to its neighbor node with the value vector of a neighbor node to obtain the inter-chain transmission information passed from the neighbor node to the node; sum the inter-chain transmission information passed from each neighbor node to the node to obtain the summed inter-chain transmission information; sum the summed inter-chain transmission information with the intra-chain updated node representation to obtain the inter-chain updated node representation of the node.

[0179] In an exemplary embodiment of the present invention, the fourth module 1040 can perform inter-chain updates on nodes based on intra-chain updated node coordinates using an attention mechanism to obtain the inter-chain updated node coordinates:

[0180] The value vectors of neighbor nodes are projected onto the node through a fourth pre-trained fully connected neural network to obtain a projected value vector with the same dimension as the inter-chain relative coordinates. The inter-chain relative coordinates are determined based on the node's intra-chain updated node coordinates and the neighbor nodes' intra-chain updated neighbor node coordinates. The projected value vector is then multiplied element-wise with the inter-chain relative coordinates, and then multiplied with the attention weights of the node to its neighbor nodes to obtain the second node coordinate update amount corresponding to the neighbor node. The second node coordinate update amounts corresponding to each neighbor node are summed to obtain the inter-chain target node coordinate update amount. The inter-chain target node coordinate update amount and the node's intra-chain updated node coordinates are then residually joined to obtain the node's inter-chain updated node coordinates.

[0181] In an exemplary embodiment of the present invention, the fourth module 1040 may determine the value vector of a neighbor node in the following manner:

[0182] The inter-chain space information passed from the neighbor node to the node, and the neighbor node's intra-chain updated neighbor node representation are concatenated and then input into the fifth pre-trained fully connected neural network to obtain the value vector of the neighbor node for the node.

[0183] In an exemplary embodiment of the present invention, the fourth module 1040 may determine the attention weight of a node to its neighboring nodes in the following manner:

[0184] The node's intra-chain updated node representation is input into the sixth pre-trained fully connected neural network to obtain the node's query vector. The inter-chain spatial information passed from neighboring nodes to the node, and the neighboring node's intra-chain updated neighboring node representation are concatenated and input into the seventh pre-trained fully connected neural network to obtain the neighboring node's key vector. The node's query vector is then multiplied by the key vectors of each neighboring node to obtain the multiplication result. The multiplication result is then processed through a softmax activation layer to obtain the node's attention weights to its neighboring nodes.

[0185] In an exemplary embodiment of the present invention, the fourth module 1040 may determine the inter-chain space information passed to the node by the neighbor node in the following manner:

[0186] By performing a transpose multiplication on the relative coordinates between chains, we obtain the inter-chain spatial information passed from the neighboring node to the node.

[0187] In an exemplary embodiment of the present invention, the third module 1030 may determine neighboring nodes in the following manner:

[0188] The nodes in the graph that are at a preset distance from the node are considered as the node's neighbor nodes.

[0189] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11As shown, the electronic device may include: a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute an antibody generation method, which includes: S1: obtaining an antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex to be processed is an antibody lacking a complementarity-determining region; S2: obtaining a representation graph corresponding to the antigen-antibody complex to be processed based on the antigen-antibody complex to be processed, wherein the representation graph includes multiple nodes, and each node includes a node representation and node coordinates; S3: performing an intra-chain update on the node based on the neighbor node representations and neighbor node coordinates of the node's neighbor nodes, obtaining the intra-chain updated node representation and intra-chain updated node coordinates of the node; S4: based on the intra-chain... S5: Based on the inter-chain updated node representations and coordinates of each node, an inter-chain updated representation graph corresponding to the representation graph is obtained; S6: The inter-chain updated representation graph is used as the representation graph and S3 to S5 are repeated for a preset number of rounds to obtain the first updated representation graph; S7: The first updated representation graph is used as the representation graph and S3 to S6 are repeated for a preset number of rounds to obtain the target updated representation graph, and the antibody corresponding to the complementarity-determining region of the antigen-antibody complex to be processed is obtained based on the target updated representation graph.

[0190] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0191] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the antibody generation method provided by the above methods. The method includes: S1: obtaining an antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex to be processed is an antibody lacking a complementarity-determining region; S2: obtaining a representation graph corresponding to the antigen-antibody complex to be processed based on the antigen-antibody complex to be processed, wherein the representation graph includes multiple nodes, and each node includes a node representation and node coordinates; S3: performing intra-chain updates on the nodes based on the neighbor node representations and neighbor node coordinates of the neighbor nodes of the nodes. S4: Based on the intra-chain updated node representation and coordinates, perform inter-chain updates on the nodes using an attention mechanism to obtain the inter-chain updated node representation and coordinates; S5: Based on the inter-chain updated node representation and coordinates of each node, obtain the inter-chain updated representation graph corresponding to the representation graph; S6: Repeat steps S3 to S5 for a preset number of rounds using the inter-chain updated representation graph as the representation graph to obtain the first updated representation graph; S7: Repeat steps S3 to S6 for a preset number of rounds using the first updated representation graph as the representation graph to obtain the target updated representation graph, and obtain the antibody corresponding to the complementarity-determining region of the antigen-antibody complex to be processed based on the target updated representation graph.

[0192] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the antibody generation method provided by the methods described above. The method includes: S1: obtaining an antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex is an antibody lacking a complementarity-determining region; S2: obtaining a representation graph corresponding to the antigen-antibody complex based on the antigen-antibody complex to be processed, wherein the representation graph includes multiple nodes, and each node includes a node representation and node coordinates; S3: performing an in-chain update on the node based on the neighbor node representations and neighbor node coordinates of the node's neighbor nodes, to obtain an in-chain updated node representation of the node. S4: Based on the intra-chain updated node representation and intra-chain updated node coordinates, perform inter-chain updates on the nodes through an attention mechanism to obtain the inter-chain updated node representation and inter-chain updated node coordinates of the nodes; S5: Based on the inter-chain updated node representation and inter-chain updated node coordinates of each node, obtain the inter-chain updated representation graph corresponding to the representation graph; S6: Repeat S3 to S5 for a preset number of rounds using the inter-chain updated representation graph as the representation graph to obtain the first updated representation graph; S7: Repeat S3 to S6 for a preset number of rounds using the first updated representation graph as the representation graph to obtain the target updated representation graph, and obtain the antibody corresponding to the complementarity-determining region of the antigen-antibody complex to be processed based on the target updated representation graph.

[0193] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0194] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0195] It is further understood that although the operations are described in a specific order in the accompanying drawings in the embodiments of the present invention, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all the operations shown to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0196] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating antibodies, characterized in that, The antibody generation method includes: S1: Obtain the antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex to be processed is an antibody lacking the complementarity-determining region; S2: Based on the antigen-antibody complex to be treated, a representation diagram corresponding to the antigen-antibody complex to be treated is obtained, wherein the representation diagram includes multiple nodes, and each node includes a node representation and node coordinates; S3: Based on the neighbor node representations and coordinates of the neighbor nodes of the node, perform an in-chain update on the node to obtain the in-chain updated node representation and coordinates of the node. S4: Based on the intra-chain updated node representation and the intra-chain updated node coordinates, perform inter-chain updates on the node through an attention mechanism to obtain the inter-chain updated node representation and inter-chain updated node coordinates of the node. Specifically, performing inter-chain updates on the node based on the intra-chain updated node representation includes: determining the value vector of the neighbor node to the node, and the attention weight of the node to the neighbor node; multiplying the attention weight of the node to the neighbor node with the value vector of the neighbor node to obtain the inter-chain transmission information passed from the neighbor node to the node; summing the inter-chain transmission information passed from each neighbor node to the node to obtain summed inter-chain transmission information; and summing the summed inter-chain transmission information with the intra-chain updated node representation to obtain the inter-chain updated node representation of the node. S5: Based on the inter-chain update node representation and the inter-chain update node coordinates of each node, obtain the inter-chain update representation graph corresponding to the representation graph; S6: Repeat S3 to S5 for a preset number of rounds using the inter-chain updated representation map as the representation map to obtain a first updated representation map, wherein the first updated representation map is a representation map obtained by alternating intra-chain coding processing and inter-chain coding processing for a preset number of rounds. Intra-chain coding processing refers to coding processing within the same chain of the antigen-antibody complex to be treated, and inter-chain coding processing refers to coding processing between different chains of the antigen-antibody complex to be treated. S7: Repeat S3 to S6 for a preset number of rounds using the first updated representation map as the representation map to obtain a target updated representation map, and obtain an antibody corresponding to the complementarity determination region of the antigen-antibody complex to be processed based on the target updated representation map.

2. The antibody generation method according to claim 1, characterized in that, The process of updating the node within the chain based on the neighbor node representations and coordinates of the node's neighbor nodes to obtain the updated node representation within the chain specifically includes: Based on the node coordinates of the node and the neighbor node coordinates of the neighbor node, the relative coordinates within the chain are obtained; Perform transpose multiplication on the relative coordinates within the chain to obtain the spatial information within the chain that the neighboring node passes to the node; The node representation of the node, the neighbor node representation of the neighbor node, the intra-chain spatial information and the edge type are concatenated to obtain a first concatenation vector, and the first concatenation vector is input into a first pre-trained fully connected neural network to obtain the intra-chain transmission information passed from the neighbor node to the node, wherein the edge type is the edge type between the node and the neighbor node. The intra-chain transmission information passed from each of the neighboring nodes to the node is summed to obtain the summed intra-chain transmission information; The summed intra-chain transmission information is concatenated with the node representation of the node to obtain a second concatenated vector. The second concatenated vector is then input into a second pre-trained fully connected neural network to obtain the intra-chain updated node representation of the node.

3. The antibody generation method according to claim 2, characterized in that, The process of updating the node within the chain based on the neighbor node representations and coordinates of the node's neighbor nodes to obtain the updated node coordinates within the chain specifically includes: The intra-chain transmitted information is projected through a third pre-trained fully connected neural network to obtain the projected intra-chain transmitted information with the same dimension as the relative coordinates within the chain. The information transmitted within the chain after the projection processing is multiplied element-wise with the relative coordinates within the chain to obtain the first node coordinate update amount corresponding to the neighbor node; The coordinate update amounts of the first node corresponding to each of the neighboring nodes are averaged to obtain the coordinate update amounts of the target node within the chain. The updated coordinates of the target node in the chain and the node coordinates of the node are subjected to residual join processing to obtain the updated node coordinates of the node in the chain.

4. The antibody generation method according to claim 1, characterized in that, Based on the intra-chain updated node coordinates, an attention mechanism is used to perform inter-chain updated node coordinates to obtain the inter-chain updated node coordinates of the node, specifically including: The neighbor node's value vector to the node is projected through a fourth pre-trained fully connected neural network to obtain a projected value vector with the same dimension as the inter-chain relative coordinates. The inter-chain relative coordinates are determined based on the node's intra-chain updated node coordinates and the neighbor node's intra-chain updated neighbor node coordinates. After the projection processing, the value vector is multiplied element-wise with the relative coordinates between the chains, and then multiplied with the attention weight of the node to the neighbor node to obtain the second node coordinate update amount corresponding to the neighbor node. The coordinate update amounts of the second node corresponding to each of the neighboring nodes are summed to obtain the inter-chain target node coordinate update amounts. The inter-chain target node coordinate update amount and the intra-chain updated node coordinate of the node are subjected to residual connection processing to obtain the inter-chain updated node coordinate of the node.

5. The antibody generation method according to claim 1, characterized in that, The value vector of the neighboring node is determined in the following way: The inter-chain space information passed from the neighbor node to the node, and the intra-chain updated neighbor node representation of the neighbor node are concatenated and then input into the fifth pre-trained fully connected neural network to obtain the value vector of the neighbor node to the node.

6. The antibody generation method according to claim 1, characterized in that, The attention weight of a node to its neighboring nodes is determined in the following way: The in-chain update node representation of the node is input into the sixth pre-trained fully connected neural network to obtain the query vector of the node; The inter-chain space information passed from the neighbor node to the node and the intra-chain updated neighbor node representation of the neighbor node are concatenated and then input into the seventh pre-trained fully connected neural network to obtain the key vector of the neighbor node to the node. The query vector of the node is multiplied by the key vector of each of the neighboring nodes to obtain the dot product result. The dot product result is then processed through a softmax activation layer to obtain the attention weight of the node to the neighboring nodes.

7. The antibody generation method according to claim 5 or 6, characterized in that, The inter-chain space information passed from the neighbor node to the node is determined in the following way: The relative coordinates between chains are transposed and multiplied to obtain the inter-chain spatial information passed from the neighboring node to the node.

8. The antibody generation method according to claim 1, characterized in that, The neighbor nodes are determined in the following way: The nodes in the representation graph that are at a preset distance from the node are designated as the neighbor nodes of the node.

9. An antibody generation device, characterized in that, The antibody generating apparatus is used to implement the antibody generating method according to any one of claims 1 to 8, the apparatus comprising: The first module is used to obtain an antigen-antibody complex to be processed, wherein the antibody in the antigen-antibody complex is an antibody lacking the complementarity-determining region; The second module is used to obtain a representation diagram corresponding to the antigen-antibody complex to be processed based on the antigen-antibody complex to be processed, wherein the representation diagram includes multiple nodes, and each node includes a node representation and node coordinates; The third module is used to perform an in-chain update on the node based on the neighbor node representation and neighbor node coordinates of the node's neighbor nodes, so as to obtain the in-chain updated node representation and in-chain updated node coordinates of the node. The fourth module is used to perform inter-chain updates on the node based on the intra-chain update node representation and the intra-chain update node coordinates, through an attention mechanism, to obtain the inter-chain update node representation and inter-chain update node coordinates of the node. The fifth module is used to obtain an inter-chain update representation graph corresponding to the representation graph based on the inter-chain update node representation and the inter-chain update node coordinates of each of the nodes. The sixth module is used to repeatedly execute the steps of the third to fifth modules for a preset number of rounds using the inter-chain update representation graph as the representation graph, so as to obtain the first update representation graph; The seventh module is used to repeatedly execute the steps of the third to sixth modules for a preset number of rounds using the first updated representation map as the representation map, so as to obtain a target updated representation map, and obtain an antibody corresponding to the complementarity-determining region of the antigen-antibody complex to be processed based on the target updated representation map.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the antibody generation method as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the antibody generation method as described in any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the antibody generation method as described in any one of claims 1 to 8.

Images