Protein-protein binding affinity prediction method and device, equipment and medium

Abstract

The invention discloses a protein-protein binding affinity prediction method and device, equipment and a medium, and can be applied to the technical field of drug target affinity prediction. According to the method, after the first affinity prediction model comprising the structure model and the first monotonic neural network model is constructed, the first affinity prediction model is trained based on the self-attention mechanism and the mutual attention mechanism, so that the interaction mechanism of the receptor protein and the ligand protein can be simulated, and the accuracy of the model is improved; then, after a second affinity prediction model comprising a sequence model and a second monotonic neural network model is constructed, training the second affinity prediction model based on a structural feature vector output by the trained structural model and by adopting an alignment mechanism; therefore, the sequence model has the accuracy of the structure model and keeps the original high-throughput screening characteristic, and the affinity prediction accuracy of the algorithm model can be improved while the algorithm model training cost can be effectively reduced.

Description

Technical Field

[0001] The present application relates to the technical field of drug target affinity prediction, and in particular to a method and apparatus, device and medium for predicting protein-protein binding affinity. Background Art

[0002] In the related art, a core issue of protein drug discovery engineering is to evaluate the strength of binding between drug proteins and target proteins, i.e. affinity. High-throughput protein affinity screening can accelerate the development of protein drugs. Although the method of determining high-throughput protein affinity through experiments is accurate, it requires cumbersome experimental operations, strict experimental conditions, and expensive instruments and consumables. Therefore, algorithm-based affinity prediction is a paradigm with greater potential for high-throughput screening. However, the affinity prediction algorithm is limited in the following aspects: First, due to the lack of affinity experimental data in the strict sense, and the large amount of affinity experimental data in the broad sense, the strict affinity data and the broad affinity data are heterogeneous, making the algorithm model obtained by training based on the current data relatively generalized; second, due to the high time and economic cost of obtaining accurate protein structure in reality, and the most direct determinant of affinity is the structure of receptor protein and ligand protein, especially the structure near the binding interface, the current algorithm model for affinity prediction through protein structure is relatively expensive.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method and apparatus, device and medium for predicting the affinity of protein-protein binding, which can effectively reduce the training cost of the algorithm model and improve the affinity prediction accuracy of the algorithm model.

[0005] To achieve the above object, one aspect of the embodiments of the present application provides a method for predicting the affinity of protein-protein binding, the method comprising the following steps:

[0006] Constructing a first affinity prediction model, wherein the first affinity prediction model includes a structural model and a first monotone neural network model;

[0007] Training the first affinity prediction model based on the self-attention mechanism and the mutual attention mechanism;

[0008] constructing a second affinity prediction model, wherein the second affinity prediction model includes a sequence model and a second monotone neural network model;

[0009] Based on the structural feature vector output by the trained structural model, the second affinity prediction model is trained using an alignment mechanism;

[0010] Affinity prediction is performed using the trained second affinity prediction model.

[0011] In some embodiments, the training of the first affinity prediction model based on the self-attention mechanism and the mutual attention mechanism includes:

[0012] Obtaining ligand protein structural feature information, receptor protein structural feature information and first heterogeneous affinity;

[0013] Inputting the structural feature information of the ligand protein and the structural feature information of the receptor protein into the structural model, training the structural model based on the self-attention mechanism and the mutual attention mechanism, and outputting a first feature vector;

[0014] Inputting the first feature vector into the first monotone neural network model to obtain a first predicted affinity;

[0015] The training process of the first affinity prediction model is adjusted based on the first heterogeneous affinity and the first predicted affinity feedback.

[0016] In some embodiments, the training of the structural model based on the self-attention mechanism and the mutual attention mechanism includes:

[0017] Controlling the receptor protein to update its own characteristics according to the structural characteristic information of the receptor protein, and the ligand protein to update its own characteristics according to the structural characteristic information of the ligand protein;

[0018] The receptor protein is controlled to update its own characteristics according to the structural characteristic information of the ligand protein, and the ligand protein updates its own characteristics according to the structural characteristic information of the receptor protein.

[0019] In some embodiments, the training of the second affinity prediction model using an alignment mechanism based on the structural feature vector output by the trained structural model includes:

[0020] Obtaining receptor protein sequence information, ligand protein sequence information and second heterogeneous affinity;

[0021] Inputting the receptor protein sequence information and the ligand protein sequence information into the sequence model, and outputting a second feature vector;

[0022] Inputting the second feature vector into the second monotone neural network model to obtain a second predicted affinity;

[0023] An alignment mechanism is used to feedback-regulate the training process of the second affinity prediction model according to the second heterogeneous affinity, the second predicted affinity, the first feature vector, and the second feature vector.

[0024] In some embodiments, the alignment mechanism is as follows:

[0025]

[0026] In the formula, χ structure represents the first eigenvector, χ sequence represents the second eigenvector, || ||2 represents the L2 norm, and ∈ is a very small number.

[0027] In some embodiments, the data processing during the sequence model training process includes the following steps:

[0028] The receptor protein is controlled to update its own characteristics according to the characteristic information of the ligand protein sequence, and the ligand protein updates its own characteristics according to the characteristic information of the receptor protein sequence.

[0029] In some embodiments, the feature update of the structural model during training is as follows:

[0030]

[0031] In the formula, l represents the lth round of feature update. The updated feature vector of the i-th amino acid residue, is the total attention weight of residue i to residue j in the lth round of feature update; The feature vector of the ith amino acid residue; R j and T j Respectively represent the rotation matrix and translation vector from the local coordinate system to the global coordinate system of the j-th amino acid residue; represents the updated feature vector of the amino acid residue pair formed by the i-th amino acid residue and the j-th amino acid residue; in the self-attention mechanism, i and j represent residues within the same protein; in the mutual attention mechanism, i and j represent residues of the receptor protein and residues of the ligand protein, respectively;

[0032] The feature update formula of the sequence model during training is as follows:

[0033]

[0034] h' li =α li,rj Linear(h li );

[0035] In the formula, h' li represents the updated sequence feature of the ith residue in the ligand; h li represents the sequence characteristics of the ith residue in the ligand before updating; h rj represents the sequence characteristics of the jth residue in the receptor before updating; hl represents the sequence characteristics of the ligand; h r Represents the sequence characteristics of the receptor; α li,rj represents the attention of the i-th residue in the ligand to the j-th residue in the receptor; d represents a constant; in the self-attention mechanism, i and j represent residues in the same protein; in the mutual attention mechanism, i and j represent residues in the receptor protein and residues in the ligand protein, respectively.

[0036] To achieve the above object, another aspect of the embodiments of the present application provides a protein-protein binding affinity prediction device, the device comprising:

[0037] A first module is used to construct a first affinity prediction model, wherein the first affinity prediction model includes a structural model and a first monotone neural network model;

[0038] A second module is used to train the first affinity prediction model based on a self-attention mechanism and a mutual attention mechanism;

[0039] A third module is used to construct a second affinity prediction model, wherein the second affinity prediction model includes a sequence model and a second monotone neural network model;

[0040] A fourth module is used to train the second affinity prediction model using an alignment mechanism based on the structural feature vector output by the trained structural model;

[0041] The fifth module is used to perform affinity prediction using the trained second affinity prediction model.

[0042] To achieve the above object, another aspect of an embodiment of the present application provides an electronic device, including:

[0043] at least one processor;

[0044] at least one memory for storing at least one program;

[0045] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0046] To achieve the above objective, another aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0047] The embodiments of the present application include at least the following beneficial effects: The present application provides a protein-protein binding affinity prediction method and apparatus, equipment and medium, which constructs a first affinity prediction model including a structural model and a first monotonic neural network model, and then trains the first affinity prediction model based on a self-attention mechanism and a mutual attention mechanism, thereby simulating the mechanism of interaction between receptor protein and ligand protein, thereby improving the model accuracy; then, after constructing a second affinity prediction model including a sequence model and a second monotonic neural network model, the second affinity prediction model is trained based on the structural feature vector output by the trained structural model and an alignment mechanism, thereby enabling the sequence model to have the accuracy of the structural model and maintain its original high-throughput screening characteristics, thereby effectively reducing the algorithm model training cost while improving the affinity prediction accuracy of the algorithm model. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flow chart of a method for predicting affinity of protein-protein binding provided in an embodiment of the present application;

[0049] Figure 2 It is a training schematic diagram of the protein-protein binding affinity prediction method provided in the examples of the present application;

[0050] Figure 3 It is a schematic diagram of the structure of the protein-protein binding affinity prediction device provided in the embodiments of the present application;

[0051] Figure 4 is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application;

[0052] Figure 5 is a correlation scatter plot of the trained structural model provided in the embodiment of the present application on the test set;

[0053] Figure 6 It is a correlation scatter plot of the trained sequence model provided in the embodiment of the present application on the test set. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application, they are only examples of devices and methods consistent with some aspects of the embodiments of the present application.

[0055] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0056] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0058] In the related technology, a core issue of protein drug discovery engineering is to evaluate the strength of binding between protein drugs and target proteins, that is, affinity. The mechanism of action of protein drugs usually depends on their ability to bind to specific target proteins. Protein drugs with high affinity can bind to target proteins more effectively, thereby exerting a therapeutic effect; on the other hand, high-affinity protein drugs can bind to target proteins more specifically and reduce the effects on non-target proteins, which helps to reduce the side effects of drugs and improve the safety of treatment; in addition, high-affinity protein drugs may have a longer half-life in the body because they are more likely to bind to target proteins and remain stable, thereby extending the effective time of the drug. In summary, in the process of protein drug discovery and development, early evaluation of affinity can help screen out the most promising candidate drugs, reduce the risk of failure that may be encountered in the subsequent development process, and thus accelerate the drug development process.

[0059] Biologists quantify protein-protein binding affinity by the change in Gibbs free energy (ΔG) or the dissociation constant (KD) value during binding, and the measurement methods can be summarized as follows: separation methods (Methods based on separation of comp lex and unbound partners) such as gel filtration, electrophoretic separation, and ultracentrifugation. Direct measurement methods such as isothermal titration calorimetry (ITC) and surface plasmon resonance (SPR).

[0060] In addition, deep mutational scanning (DMS) can be used to measure affinity indices other than ΔG / KD to evaluate protein-protein binding affinity from the side, and obtain enrichment ratios for evaluating affinity changes after amino acid mutations. Another example is high-throughput titration assays, which can compare K values ​​before and after mutation. d,app To quantify the effect of amino acid mutations on affinity. Changes in affinity can also be evaluated by measuring changes in IC50. These non-ΔG / KD affinity indicators are usually monotonically positively or negatively correlated with ΔG / KD, but existing AI algorithms only use ΔG or KD data, and data in other units are not used.

[0061] Although the experimental method is accurate, it requires tedious experimental operations, strict experimental conditions, and expensive instruments and consumables. Therefore, algorithm-based affinity prediction is a paradigm with greater potential for high-throughput screening. However, the affinity prediction algorithm is limited in the following aspects: First, due to the lack of strict affinity experimental data, and the large amount of broad affinity experimental data, the strict affinity data and the broad affinity data are heterogeneous, making the algorithm model trained based on the current data relatively generalized; second, in reality, the time and economic cost of obtaining accurate protein structure are very high, and the most direct determinant of affinity is the structure of receptor protein and ligand protein, especially the structure near the binding interface, resulting in a high cost for the current algorithm model for affinity prediction based on protein structure.

[0062] In view of this, a method and apparatus, device and medium for predicting the affinity of protein-protein binding are provided in the embodiments of the present application. After constructing a first affinity prediction model including a structural model and a first monotonic neural network model, the first affinity prediction model is trained based on a self-attention mechanism and a mutual attention mechanism, so as to simulate the mechanism of interaction between receptor protein and ligand protein, thereby improving the accuracy of the model; then, after constructing a second affinity prediction model including a sequence model and a second monotonic neural network model, the second affinity prediction model is trained based on the structural feature vector output by the trained structural model and an alignment mechanism, so that the sequence model can have the accuracy of the structural model and maintain its original high-throughput screening characteristics, thereby effectively reducing the training cost of the algorithm model while improving the affinity prediction accuracy of the algorithm model.

[0063] The protein-protein binding affinity prediction method provided in the embodiment of the present application relates to the technical field of drug target affinity prediction. The protein-protein binding affinity prediction method provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server, and can also be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and a car terminal, etc., but is not limited to this; the server side can be configured as an independent physical server, or a server cluster or distributed system composed of multiple physical servers, and can also be configured to provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The cloud server of basic cloud computing services, the server can also be a node server in a blockchain network; the software can be an application of the affinity prediction method for implementing protein-protein binding, etc., but is not limited to the above form.

[0064] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0065] Figure 1 is an optional flow chart of the protein-protein binding affinity prediction method provided in the embodiments of the present application, Figure 1 The method may include but is not limited to steps S110 to S150:

[0066] Step S110: constructing a first affinity prediction model, wherein the first affinity prediction model includes a structural model and a first monotone neural network model;

[0067] Step S120: training the first affinity prediction model based on the self-attention mechanism and the mutual attention mechanism;

[0068] Step S130: constructing a second affinity prediction model, wherein the second affinity prediction model includes a sequence model and a second monotone neural network model;

[0069] Step S140: Based on the structural feature vector output by the trained structural model, the second affinity prediction model is trained using an alignment mechanism;

[0070] Step S150: performing affinity prediction using the trained second affinity prediction model.

[0071] In the embodiment of the present application, the present application performs the training process of the structure model and the sequence model as follows: Figure 2 It can be understood that the training process of this embodiment uses the real protein complex three-dimensional structure and the first heterogeneous affinity data to train the structural model and the first monotone neural network, and obtains a pre-trained AI model for predicting protein-protein binding affinity based on complex structure information as the first affinity prediction model. Among them, the real three-dimensional structure refers to the crystal structure of the protein complex obtained by physical means; the first heterogeneous affinity data refers to the affinity data with different unit dimensions measured by different experimental methods, for example, the change value of Gibbs free energy during protein-protein binding ΔG (unit is kcal / mol), the dissociation constant KD (unit is mol / L), the enrichment ratio (no unit) and other data; the structural model is a geometric graph neural network with interactive feature updates, which represents the three-dimensional structure of the complex as a fully connected graph consisting of points and edges, where a point represents an amino acid residue and an edge represents an amino acid residue pair formed by any amino acid residue and any amino acid residue; the first monotonic neural network is an AI algorithm that controls the numerical monotonicity between its two inputs. It is used as a priori constraint to control the monotonicity between data of different unit dimensions, so that the model integrates heterogeneous affinity data for training to enhance the generalization performance of the model.

[0072] In the embodiment of the present application, the process of training the first affinity prediction model based on the self-attention mechanism and the mutual attention mechanism includes but is not limited to the following steps:

[0073] Obtaining ligand protein structural feature information, receptor protein structural feature information and first heterogeneous affinity;

[0074] Inputting the structural feature information of the ligand protein and the structural feature information of the receptor protein into the structural model, training the structural model based on the self-attention mechanism and the mutual attention mechanism, and outputting the first feature vector;

[0075] Inputting the first feature vector into a first monotone neural network model to obtain a first predicted affinity;

[0076] The training process of the first affinity prediction model is adjusted based on the first heterogeneous affinity and the first predicted affinity.

[0077] It can be understood that this embodiment includes two modes in a complete feature update process of the structural model: first perform self-attention feature update, and then perform mutual attention feature update. In the self-attention feature update process, the receptor protein is controlled to update its own features according to the structural feature information of the receptor protein, and the ligand protein updates its own features according to the structural feature information of the ligand protein. In the mutual attention feature update process, the receptor protein is controlled to update its own features according to the structural feature information of the ligand protein, and the ligand protein updates its own features according to the structural feature information of the receptor protein, thereby simulating the process of interaction between the ligand protein and the receptor protein.

[0078] Specifically, the feature update process of the first round of the structural model is shown in Formula 1, Formula 2, Formula 3 and Formula 4:

[0079]

[0080]

[0081]

[0082]

[0083] In the formula, Represents the feature vector of the i-th amino acid residue; Represents the attention weight of residue i on residue j from the amino acid residue category information; represents the attention weight of residue i to residue j extracted from the category information of amino acid residue pairs in the lth round of feature update; represents the attention weight of residue i to residue j extracted from the three-dimensional structure information in the lth round of feature update; is the total attention weight of residue i to residue j in the first round of feature update; i and j represent the i-th and j-th amino acid residues. In the self-attention mechanism, i and j both represent residues in the same protein (receptor protein or ligand protein); when i and j represent residues on different proteins, that is, i represents the residue of receptor protein and j represents the residue of ligand protein or i represents the residue of ligand protein and j represents the residue of receptor protein, it is a mutual attention mechanism; R i and T i Respectively represent the rotation matrix and translation vector from the local coordinate system to the global coordinate system of the ith amino acid residue; R j and T j They represent the rotation matrix and translation vector from the local coordinate system to the global coordinate system of the j-th amino acid residue respectively; d is a predefined constant.

[0084] Based on the above formula, the overall feature update process of the first round of the structural model is as shown in Formula 5:

[0085]

[0086] In the formula, The updated feature vector of the i-th amino acid residue, is the total attention weight of residue i to residue j in the lth round of feature update; The feature vector of the ith amino acid residue; R j and T j Respectively represent the rotation matrix and translation vector from the local coordinate system to the global coordinate system of the j-th amino acid residue; Represents the updated feature vector of the amino acid residue pair formed by the ith amino acid residue and the jth amino acid residue.

[0087] In the embodiment of the present application, the feature update process of the structural model includes the above-mentioned several rounds of feature update process, so that the deep features of the interaction between the receptor protein and the ligand protein can be extracted. Specifically, after the training of the structural model is completed, the trained structural model is tested on the test set, and the correlation point diagram between the predicted value and the true value is as follows: Figure 5 shown.

[0088] It can be understood that after completing the updating process of the structural model, this embodiment uses an alignment mechanism to train the second affinity prediction model based on the structural feature vector output by the trained structural model, and the process includes but is not limited to the following steps:

[0089] Obtaining receptor protein sequence information, ligand protein sequence information and second heterogeneous affinity;

[0090] Inputting the receptor protein sequence information and the ligand protein sequence information into the sequence model, and outputting a second feature vector;

[0091] Inputting the second feature vector into a second monotone neural network model to obtain a second predicted affinity;

[0092] An alignment mechanism is used to feedback-regulate the training process of the second affinity prediction model according to the second heterogeneous affinity, the second predicted affinity, the first eigenvector and the second eigenvector.

[0093] Specifically, this embodiment uses sequence information and heterogeneous affinity data, and after training the sequence model and the second monotone neural network with an alignment mechanism, obtains a pre-trained AI model for predicting protein-protein binding affinity based on protein sequence information as the second affinity prediction model. Among them, the sequence model is a neural network with interactive feature updates.

[0094] In order to allow the sequence model to learn structural information, this embodiment trains the sequence model through an alignment mechanism, thereby simplifying the model training process and making the features that the sequence model can extract from the sequence close to the features extracted by the structural model. Specifically, the alignment mechanism can be regarded as a mechanism to minimize the template, and its formula is shown in Formula 6:

[0095]

[0096] In the formula, χ structure represents the first eigenvector, χ sequence represents the second eigenvector, || ||2 represents the L2 norm, and ∈ is a very small number.

[0097] In the feature update process of the sequence model, the feature update process of the interaction between the receptor protein and the ligand protein is described by Formula 7 and Formula 8:

[0098]

[0099] h′ li =α li，rj Linear(h li ) Formula 8;

[0100] In the formula, h′ li represents the updated sequence feature of the ith residue in the ligand; h li represents the sequence characteristics of the ith residue in the ligand before updating; h rj represents the sequence characteristics of the jth residue in the receptor before updating; h l represents the sequence characteristics of the ligand; h r Represents the sequence characteristics of the receptor; α li，rjrepresents the attention of the i-th residue in the ligand to the j-th residue in the receptor; d represents a constant; in the self-attention mechanism, i and j both represent residues within the same protein (receptor protein or ligand protein); when i and j represent residues on different proteins, that is, i represents the residue of the receptor protein and j represents the residue of the ligand protein, or i represents the residue of the ligand protein and j represents the residue of the receptor protein, it is a mutual attention mechanism.

[0101] In the feature interaction mechanism of the sequence model, the receptor and the ligand are separated instead of being treated as a whole. Therefore, in the feature updating process, the receptor protein is controlled to update its own features according to the sequence feature information of the ligand protein, and the ligand protein updates its own features according to the sequence feature information of the receptor protein.

[0102] Specifically, the feature update process of the sequence model includes the above-mentioned several feature update processes, so that the deep features of the interaction between the receptor protein and the ligand protein can be extracted. Specifically, after the training of the sequence model is completed, the trained sequence model is tested on the test set, and the correlation dot plot between the predicted value and the true value is as follows: Figure 6 shown.

[0103] After completing the updating process of the sequence model, this embodiment can perform affinity prediction in the actual application process through the second affinity prediction model corresponding to the sequence model, thereby effectively improving the accuracy of the prediction result.

[0104] In addition, the affinity change prediction value ΔΔG of the receptor or ligand before and after the mutation in the protein complex can be used to train the model to calculate the affinity prediction value, thereby reflecting the affinity relationship between the receptor and the ligand.

[0105] In summary, this application simulates this interaction mechanism by adopting the feature update process of self-attention and mutual attention sets, so the accuracy of the affinity prediction model can be improved. In addition, this application combines the advantages of the two models through the alignment mechanism, so that the sequence model has both the accuracy of the structural model and the advantages of the original high-throughput screening, and can integrate affinity data of multiple different unit dimensions, so that the model has better robustness, and further improves the performance of the model through the feature interaction update mechanism.

[0106] Reference Figure 3 , the present application embodiment provides a protein-protein binding affinity prediction device, the device comprising:

[0107] A first module 310 is used to construct a first affinity prediction model, wherein the first affinity prediction model includes a structural model and a first monotone neural network model;

[0108] The second module 320 is used to train the first affinity prediction model based on the self-attention mechanism and the mutual attention mechanism;

[0109] The third module 330 is used to construct a second affinity prediction model, wherein the second affinity prediction model includes a sequence model and a second monotone neural network model;

[0110] A fourth module 340 is used to train a second affinity prediction model using an alignment mechanism based on the structural feature vector output by the trained structural model;

[0111] The fifth module 350 is used to perform affinity prediction using the trained second affinity prediction model.

[0112] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0113] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above protein-protein binding affinity prediction method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0114] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0115] See also Figure 4 , Figure 4 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0116] The processor 410 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0117] The memory 420 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 420 can store an operating system and other application programs. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 420, and the processor 410 calls and executes the protein-protein binding affinity prediction method of the embodiment of the present application;

[0118] Input / output interface 430, used to implement information input and output;

[0119] Communication interface 440, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WI FI, Bluetooth, etc.);

[0120] bus 450 , which transmits information between the various components of the device (e.g., processor 410 , memory 420 , input / output interface 430 , and communication interface 440 );

[0121] The processor 410 , the memory 420 , the input / output interface 430 , and the communication interface 440 are connected to each other in communication within the device via the bus 450 .

[0122] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned protein-protein binding affinity prediction method is implemented.

[0123] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0124] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0125] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0126] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0127] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0128] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0129] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application 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 data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising 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.

[0130] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0131] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0132] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0133] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0134] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0135] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for predicting affinity of protein-protein binding, characterized in that: The method comprises the following steps: Constructing a first affinity prediction model, wherein the first affinity prediction model includes a structural model and a first monotone neural network model; Training the first affinity prediction model based on the self-attention mechanism and the mutual attention mechanism; constructing a second affinity prediction model, wherein the second affinity prediction model includes a sequence model and a second monotone neural network model; Based on the structural feature vector output by the trained structural model, the second affinity prediction model is trained using an alignment mechanism; Affinity prediction is performed using the trained second affinity prediction model.

2. The method according to claim 1, characterized in that The training of the first affinity prediction model based on the self-attention mechanism and the mutual attention mechanism includes: Obtaining ligand protein structural feature information, receptor protein structural feature information and first heterogeneous affinity; Inputting the structural feature information of the ligand protein and the structural feature information of the receptor protein into the structural model, training the structural model based on the self-attention mechanism and the mutual attention mechanism, and outputting a first feature vector; Inputting the first feature vector into the first monotone neural network model to obtain a first predicted affinity; The training process of the first affinity prediction model is adjusted based on the first heterogeneous affinity and the first predicted affinity feedback.

3. The method according to claim 2, characterized in that The training of the structural model based on the self-attention mechanism and the mutual attention mechanism includes: Controlling the receptor protein to update its own characteristics according to the structural characteristic information of the receptor protein, and the ligand protein to update its own characteristics according to the structural characteristic information of the ligand protein; The receptor protein is controlled to update its own characteristics according to the structural characteristic information of the ligand protein, and the ligand protein updates its own characteristics according to the structural characteristic information of the receptor protein.

4. The method according to claim 2, characterized in that: The step of training the second affinity prediction model based on the structure feature vector output by the trained structure model using an alignment mechanism includes: Obtaining receptor protein sequence information, ligand protein sequence information and second heterogeneous affinity; Inputting the receptor protein sequence information and the ligand protein sequence information into the sequence model, and outputting a second feature vector; Inputting the second feature vector into the second monotone neural network model to obtain a second predicted affinity; An alignment mechanism is used to feedback-regulate the training process of the second affinity prediction model according to the second heterogeneous affinity, the second predicted affinity, the first feature vector, and the second feature vector.

5. The method according to claim 4, characterized in that The alignment mechanism is as follows: In the formula, χ structure represents the first eigenvector, χ sequence represents the second eigenvector, || ||2 represents the L2 norm, and ∈ is a very small number.

6. The method according to claim 4, characterized in that The data processing during the sequence model training process includes the following steps: The receptor protein is controlled to update its own characteristics according to the characteristic information of the ligand protein sequence, and the ligand protein updates its own characteristics according to the characteristic information of the receptor protein sequence.

7. The method according to claim 1, characterized in that The characteristic update formula of the structural model during the training process is as follows: In the formula, l represents the lth round of feature update. The updated feature vector of the i-th amino acid residue, is the total attention weight of residue i to residue j in the lth round of feature update; The feature vector of the ith amino acid residue; R j and T j Respectively represent the rotation matrix and translation vector from the local coordinate system to the global coordinate system of the j-th amino acid residue; represents the updated feature vector of the amino acid residue pair formed by the i-th amino acid residue and the j-th amino acid residue; in the self-attention mechanism, i and j represent residues within the same protein; in the mutual attention mechanism, i and j represent residues of the receptor protein and residues of the ligand protein, respectively; The feature update formula of the sequence model during training is as follows: h' li =α li,rj Linear(h li ); In the formula, h' li represents the updated sequence feature of the ith residue in the ligand; h li represents the sequence characteristics of the ith residue in the ligand before updating; h rj represents the sequence characteristics of the jth residue in the receptor before updating; h l represents the sequence features of the ligand; h r Represents the sequence characteristics of the receptor; α li,rj represents the attention of the i-th residue in the ligand to the j-th residue in the receptor; d represents a constant; in the self-attention mechanism, i and j represent residues in the same protein; in the mutual attention mechanism, i and j represent residues in the receptor protein and residues in the ligand protein, respectively.

8. A protein-protein binding affinity prediction device, characterized in that: The device comprises: A first module is used to construct a first affinity prediction model, wherein the first affinity prediction model includes a structural model and a first monotone neural network model; A second module is used to train the first affinity prediction model based on a self-attention mechanism and a mutual attention mechanism; A third module is used to construct a second affinity prediction model, wherein the second affinity prediction model includes a sequence model and a second monotone neural network model; A fourth module is used to train the second affinity prediction model using an alignment mechanism based on the structural feature vector output by the trained structural model; The fifth module is used to perform affinity prediction using the trained second affinity prediction model.

9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images