CN116646028A - Training method and device of molecular docking model, electronic equipment and storage medium - Google Patents

Abstract

The disclosure provides a training method and device for a molecular docking model, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, in particular to the technical field of biological calculation and deep learning. The specific implementation scheme is as follows: acquiring a protein and drug molecule combination for training, and simulating the protein and drug molecule combination to obtain labeling data of the protein and drug molecule combination; inputting the protein and drug molecule combination into an initial molecule docking model, and performing multitasking training on the initial molecule docking model to obtain prediction data of the protein and drug molecule combination; and correcting the initial molecular docking model based on the difference between the labeling data and the prediction data, and returning to continue training until a trained molecular docking model is obtained. Therefore, the model performs multitask training based on a large amount of protein and drug molecule combination data, so that the generalization capability of the model can be improved, and the accuracy of model prediction data can be improved.

Description

Molecular docking model training method, device, electronic equipment and storage medium

technical field

The present disclosure relates to the technical field of artificial intelligence, specifically to the technical fields of biological computing and deep learning, and in particular to a training method, device, electronic equipment and storage medium for a molecular docking model.

Background technique

In the process of drug development, usually thousands of compounds need to be tested in order to find safe and effective drugs, so the prediction of the affinity between the drug and the target becomes a crucial step in the drug screening process. However, the existing affinity prediction methods have limited ability to explore proteins and drug molecules, and the cost of exploration time is high, the success rate is low, and they are not suitable for large-scale virtual screening. Therefore, it is urgent to find a better drug target affinity prediction method .

Contents of the invention

The disclosure provides a training method for a molecular docking model, a drug target information acquisition method, a device, an electronic device, and a storage medium.

According to one aspect of the present disclosure, a training method for a molecular docking model is provided, including: obtaining a combination of a protein and a drug molecule for training, and simulating the combination of the protein and a drug molecule to obtain the combination of the protein and a drug molecule Combined labeling data; input the protein and drug molecule combination into the initial molecular docking model, perform multi-task training on the initial molecular docking model, and obtain the prediction data of the protein and drug molecule combination; based on the labeling data and For the difference between the predicted data, the initial molecular docking model is corrected and returned to continue training until a trained molecular docking model is obtained.

According to another aspect of the present disclosure, there is provided a drug target information acquisition method, including: an acquisition module for acquiring a combination of a target protein and a candidate drug molecule, and inputting the combination of the target protein and a candidate drug molecule In the pre-trained target molecule docking model, the binding conformation data of the combination of the target protein and the candidate drug molecule and the affinity data between the target protein and the candidate drug molecule are obtained; wherein, the target molecule docking model adopts the present disclosure The model trained by the training method.

According to another aspect of the present disclosure, a training device for a molecular docking model is provided, including: an acquisition module for acquiring a combination of a protein and a drug molecule for training, and simulating the combination of the protein and a drug molecule to obtain Labeling data of the combination of the protein and the drug molecule; a training module for inputting the combination of the protein and the drug molecule into the initial molecular docking model, performing multi-task training on the initial molecular docking model, and obtaining the protein and the drug molecule Combined prediction data; a correction module, configured to correct the initial molecular docking model based on the difference between the labeled data and the prediction data and return to continue training until a trained molecular docking model is obtained.

According to another aspect of the present disclosure, an information acquisition device for a drug target is provided, including: an acquisition module for acquiring a combination of a target protein and a candidate drug molecule, and inputting the combination of the target protein and a candidate drug molecule In the pre-trained target molecule docking model, the binding conformation data of the combination of the target protein and the candidate drug molecule and the affinity data between the target protein and the candidate drug molecule are obtained. Wherein, the target molecule docking model adopts the model trained by the training method of the present disclosure.

According to another aspect of the present disclosure, there is provided an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; Executable instructions, the instructions are executed by the at least one processor, so that the at least one processor can execute the molecular docking model training method described in the above-mentioned one embodiment.

According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, on which computer programs/instructions are stored, and the computer instructions are used to cause the computer to execute the above-mentioned one aspect embodiment. The training method of the molecular docking model described above.

According to another aspect of the present disclosure, a computer program product is provided, including computer programs/instructions, and when the computer programs/instructions are executed by a processor, the method for training molecular docking models described in the above-mentioned one embodiment is implemented.

It should be understood that what is described in this section is not intended to identify key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will be readily understood through the following description.

Description of drawings

The accompanying drawings are used to better understand the present solution, and do not constitute a limitation to the present disclosure. in:

FIG. 1 is a schematic flowchart of a training method for a molecular docking model provided by an embodiment of the present disclosure;

FIG. 2 is a schematic flowchart of another training method for a molecular docking model provided by an embodiment of the present disclosure;

FIG. 3 is a schematic flowchart of another training method for a molecular docking model provided by an embodiment of the present disclosure;

FIG. 4 is a schematic flowchart of another training method for a molecular docking model provided by an embodiment of the present disclosure;

Fig. 5 is a schematic diagram of the pre-training process of the molecular docking model provided by the embodiment of the present disclosure;

FIG. 6 is a schematic flowchart of another training method for a molecular docking model provided by an embodiment of the present disclosure;

FIG. 7 is a schematic flowchart of a method for obtaining information on a drug target provided by an embodiment of the present disclosure;

Fig. 8 is a schematic diagram of predicted protein-drug molecule affinity data and binding conformation provided by the embodiments of the present disclosure;

9 is a schematic structural diagram of a training device for a molecular docking model provided by an embodiment of the present disclosure;

FIG. 10 is a schematic structural diagram of an information acquisition device for a drug target provided by an embodiment of the present disclosure;

FIG. 11 is a block diagram of an electronic device used to implement the method for training a molecular docking model according to an embodiment of the present disclosure.

Detailed ways

Exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to facilitate understanding, and they should be regarded as exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

The following describes the training method, device and electronic equipment of the molecular docking model of the embodiments of the present disclosure with reference to the accompanying drawings.

Artificial Intelligence (AI for short) is a subject that studies certain thinking processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.) technology. Artificial intelligence hardware technology generally includes computer vision technology, speech recognition technology, natural language processing technology and its learning/deep learning, big data processing technology, knowledge map technology and other major aspects.

Natural Language Processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between humans and computers using natural language. Natural language processing is a science that combines linguistics, computer science, and mathematics. Natural language processing is mainly used in machine translation, public opinion monitoring, automatic summarization, opinion extraction, text classification, question answering, text semantic comparison, speech recognition, etc.

Deep Learning (DL for short) is a new research direction in the field of Machine Learning (ML for short). It is introduced into machine learning to make it closer to the original goal-artificial intelligence. Deep learning is to learn the internal law and representation level of sample data. The information obtained in the learning process is of great help to the interpretation of data such as text, images and sounds. Its ultimate goal is to enable machines to have the ability to analyze and learn like humans, and to be able to recognize data such as text, images, and sounds. Deep learning is a complex machine learning algorithm that has achieved results in speech and image recognition that far exceed previous related technologies.

Smart search is a new generation search engine combined with artificial intelligence technology. In addition to providing traditional functions such as fast retrieval and relevance ranking, it can also provide functions such as user role registration, automatic identification of user interests, semantic understanding of content, intelligent information filtering and push, etc.

Computer vision, computer vision is a science that studies how to make machines "see". More specifically, it refers to using cameras and computers instead of human eyes to identify, track and measure targets, and further graphics processing. Make the computer process into an image that is more suitable for human observation or sent to the instrument for detection. As a scientific discipline, computer vision studies related theories and technologies, trying to build an artificial intelligence system that can obtain "information" from images or multi-dimensional data, which can be used to help make a "decision". Because perception can be thought of as extracting information from sensory signals, computer vision can also be thought of as the science of how to make artificial systems "perceive" from images or multidimensional data.

Image processing (image processing) technology, using a computer to analyze images to achieve the desired results. Also known as image processing. Image processing generally refers to digital image processing. A digital image refers to a large two-dimensional array obtained by shooting with industrial cameras, video cameras, scanners and other equipment. The elements of this array are called pixels, and their values are called grayscale values. Image processing technology generally includes three parts: image compression, enhancement and restoration, matching, description and recognition.

Machine translation, also known as automatic translation, is the process of using a computer to convert a natural language (source language) into another natural language (target language). It is a branch of computational linguistics and one of the goals of artificial intelligence.

Biocomputing is a field that borrows principles and mechanisms from biological systems to solve computational problems. It applies some properties and processes of biology to computing systems to improve computing efficiency and performance. The goal of biocomputing is to take inspiration from biological systems and translate it into new computational methods and techniques to solve complex problems. It has a wide range of applications in optimization, pattern recognition, data analysis, and simulation, and is constantly being developed and expanded.

FIG. 1 is a schematic flowchart of a training method for a molecular docking model provided by an embodiment of the present disclosure.

As shown in Figure 1, the training method of the molecular docking model may include:

S101, acquiring a combination of protein and drug molecule for training, and simulating the combination of protein and drug molecule to obtain labeled data of the combination of protein and drug molecule.

It should be noted that the execution subject of the training method of the molecular docking model in the embodiment of the present disclosure may be a hardware device with data information processing capability and/or the necessary software required to drive the hardware device to work. Optionally, the execution subject may include servers, computers, user terminals and other intelligent devices. Optionally, the user terminal includes, but is not limited to, a mobile phone, a computer, an intelligent voice interaction device, and the like. Optionally, the server includes but is not limited to a web server, an application server, and may also be a server of a distributed system, or a server combined with a block chain, etc.

In some implementations, pre-built protein structure database (Protein DataBank bind, PDBbind) and drug molecule database can be downloaded from the source database. In the present disclosure, the protein file can be downloaded from the source database in advance, and the protein structure database can be constructed based on the protein file. The protein structure database and the drug molecule database contain large-scale and diverse data, which enriches the training data used to train the molecular docking model and facilitates the improvement of the generalization ability of the molecular docking model.

Further, the protein structure database and the drug molecule database are randomly sampled, and the sampled protein and the sampled drug molecule are combined to obtain a combination of protein and drug molecule for training. Through random sampling, the deviation of sampling results due to the tendency of selection can be avoided, which helps to improve the reliability of sampling results. For large-scale databases, random sampling can simplify the sampling process, thereby reducing computational costs.

In some implementations, the molecular docking algorithm can be used to simulate the combination of protein and drug molecule to obtain the labeled data of the combination of protein and drug molecule. Wherein, the annotation data may include but not limited to: binding conformation data of protein and drug molecule combination, affinity data between protein and drug molecule, and internal conformation data of drug molecule. Molecular docking algorithms can predict the binding conformation and affinity between proteins and drug molecules. By simulating the interaction of different proteins with drug molecules, drug candidates with potential therapeutic effects can be screened. This helps speed up the drug discovery and development process and increases the efficiency of screening.

It can be understood that molecular docking is a technique in computational biology and drug design, which is used to predict and simulate the interaction between drug molecules (ligands) and proteins (targets), and can determine the interaction between drug molecules and proteins. Binding conformation and affinity.

A drug molecule is a compound used to interact with a target protein, which can be a known drug, a natural product, or a computer-generated molecule. The chemical structure and properties of the ligand will determine its interaction mode and affinity with the target.

A protein is a protein located in a target within a certain range near the drug molecule after the drug is combined with the target.

The binding conformation is the image of the arrangement in space when the drug molecule binds to the protein atom. Affinity refers to the binding strength between drug molecules and protein atoms, which characterizes the degree of interaction between proteins and ligands, and can be the degree of interaction between drug molecules and proteins. The greater the affinity, the greater the binding strength, and the more likely the drug will be active when it acts on the target. The internal conformation of drug molecules is the image of the arrangement of drug molecules in space.

In the embodiments of the present disclosure, proteins and drug molecules need to be continuously randomly sampled from the protein structure database and the drug molecule database, and combined to obtain a protein-drug molecule combination. Then, the combination of protein and drug molecule is continuously simulated to obtain a large amount of labeled data of protein and drug molecule combination, which is used to train the molecular docking model.

S102, input the combination of protein and drug molecule into the initial molecular docking model, perform multi-task training on the initial molecular docking model, and obtain the prediction data of the combination of protein and drug molecule.

In some implementations, features can be extracted from the combination of protein and drug molecules, the extracted feature information can be input into the initial molecular docking model, and the deep learning model can be used to perform multi-task training on the initial molecular docking model. Among them, multi-task training assists the molecular docking model to output better prediction data. The multitasking may include but not limited to the first task of predicting the internal conformation of the drug molecule, the second task of predicting the binding conformation of the combination of the protein and the drug molecule, and the third task of predicting the affinity between the protein and the drug molecule. This is not limited in the embodiments of the present disclosure.

Optionally, the feature information of the combination of protein and drug molecule can be input into the first task for training, so as to obtain the predicted internal conformation data of the drug molecule. Optionally, the characteristic information of the combination of protein and drug molecule can be input into the second task for training, so as to obtain the predicted binding conformation data of the combination of protein and drug molecule. Optionally, the characteristic information of the combination of protein and drug molecule can be input into the third task for training, so as to obtain the predicted affinity data between protein and drug molecule.

In some implementations, the feature information of protein and drug molecule combinations can be grouped, and different training tasks can be selected for the feature information in different groups to train the initial molecular docking model.

S103, based on the difference between the labeled data and the predicted data, correct the initial molecular docking model and return to continue training until a trained molecular docking model is obtained.

Wherein, the initial molecular docking model may be a network model with any structure, which is not limited in this disclosure.

In some implementations, the prediction data may include, but not limited to, the prediction data of the internal conformation of the drug molecule, the prediction data of the binding conformation of the combination of the protein and the drug molecule, and the prediction data of the affinity between the protein and the drug molecule. During the training process, the difference between the labeled data and the predicted data can be obtained, the correction gradient can be determined, and then based on the correction gradient, the initial molecular docking model can be reversely corrected until the final trained docking model is obtained.

In some implementations, a loss function for the molecular docking model is calculated from the difference. Correct the model parameters of the molecular docking model based on the loss function, and continue to train the adjusted molecular docking model based on the subsequent combination of protein and drug molecules until the training end condition is met, and a trained molecular docking model is obtained.

It can be understood that the training end condition may be that the training time of the molecular docking model reaches a set value. The training end condition can also be setting the accuracy of the prediction data output by the molecular docking model.

According to the training method of the molecular docking model in the embodiment of the present disclosure, based on the data in the protein structure database and the drug molecule database, randomly sample the protein and the drug molecule, obtain the combination of the protein and the drug molecule, and continuously simulate the combination of the protein and the drug molecule , to get the labeled data. Input the combination of protein and drug molecules into the initial molecular docking model for multi-task training to obtain a molecular docking model with strong generalization ability. The model is corrected based on the difference between the labeled data and the predicted data. By correcting the defects and deviations in the molecular docking model, the prediction of the binding conformation and affinity can be improved, and the reliability and accuracy of the prediction results can be improved. Further, it can improve the model's ability to explore proteins and drug molecules, reduce the exploration time, and improve the accuracy of molecular docking model prediction data.

FIG. 2 is a schematic flowchart of a training method for a molecular docking model provided by an embodiment of the present disclosure.

As shown in Figure 2, the training method of the molecular docking model may include:

S201, acquiring a combination of protein and drug molecule for training, and simulating the combination of protein and drug molecule to obtain labeled data of the combination of protein and drug molecule.

Relevant content of step S201 may refer to the foregoing embodiments, and details are not repeated here.

S202, performing feature extraction on the combination of protein and drug molecule.

In some implementations, molecular descriptors can be used to extract features of the combination of protein and drug molecules to obtain feature information of the combination of protein and drug molecules. It can be understood that characteristic information includes, but is not limited to: molecular weight, number of hydrogen bond acceptors, number of hydrogen bond donors, number of rotatable bonds, lipid-water partition coefficient, number of specific functional groups, atomic structure, atomic electrical properties, atomic bonds Information such as the type and quantity of protein and drug molecules, covalent bonds, non-supply bonds, and distances between proteins and drug molecules. Among them, the covalent bond is that two or more atoms use their outer electrons together, and ideally reach the state of electron saturation, thus forming a relatively stable and strong chemical structure, that is, when protein atoms and When there is a covalent bond between the drug molecules, it can be determined that the binding strength of the two is relatively high. In this disclosure, covalent bond information between protein atoms and drug molecules can be obtained in the dataset.

S203, encoding the extracted feature information to obtain a feature vector representation corresponding to the combination of the protein and the drug molecule.

In some implementations, the characteristic information of the combination of the protein and the drug molecule can be encoded by a pre-trained encoder to obtain the corresponding encoded information. For example, the encoder may be a recurrent neural network (Recurrent neural network, RNN), a variational autoencoder (Variational Auto Encoder, VAE), and a self-attention encoder.

For example, a self-attention encoder can be used to encode the feature information of the combination of protein and drug molecules, and the input feature information can be associated and output through the self-attention mechanism in the encoder, so that the output feature vector representation can contain multiple input The association relationship between the feature information on the dimension. This is not limited in the embodiments of the present disclosure.

S204, performing multi-prediction head training on the feature vector representation to obtain prediction data of combinations of proteins and drug molecules.

It should be noted that multi-prediction head training is a method of using multiple prediction heads in a neural network for parallel prediction. Each prediction head corresponds to a prediction task and can output prediction data correspondingly. In the present disclosure, multi-prediction head training is performed on the feature vector representation corresponding to the combination of protein and drug molecule, so that the molecular docking model can obtain better generalization ability. Wherein, each prediction head outputs different prediction data results.

In some implementations, the feature vector representation can be input into the molecular docking model for multi-prediction head training to obtain the output prediction data. Optionally, the feature vector representation is input into the first prediction head, and the internal conformation data of the drug molecule is obtained by the first prediction head based on the feature vector representation. Among them, the internal conformation data of drug molecules can include information such as interatomic bonds, chemical bond angles, and interactions between electron layers. This information allows the model to learn the fine spatial structure and atomic interactions. For the predicted geometric conformation The legitimacy of the law has important guiding significance.

Optionally, the feature vector representation is input into the second prediction head, and the binding conformation data of the combination of the protein and the drug molecule is obtained by the second prediction head based on the feature vector representation. Among them, the binding conformation between the protein and the drug molecule can include information such as the distance between the drug molecule and the protein atom, the change of the torsion angle after the drug molecule is combined, and the size of the translational rotation. These information allow the model to learn the interaction between the drug molecule and the protein. The function can assist the model to more accurately predict the binding conformation between the bound protein and the drug molecule, and further, the precise binding conformation can help predict the binding force between the protein and the drug molecule.

Optionally, the feature vector representation is input into the third prediction head for training, and the affinity data between the protein and the drug molecule output by the third prediction head is obtained.

In the embodiment of the present disclosure, the prediction data may be the internal conformation data of the drug molecule, the binding conformation data of the combination of the protein and the drug molecule, and the affinity data between the protein and the drug molecule. Among them, the affinity data between proteins and drug molecules can reflect the interaction between proteins and drug molecules. For example, information such as protein binding or repulsion with drug molecules, binding affinity, etc. can be included. Affinity data can help screen out drug molecules that interact well with selected proteins.

S205. Based on the difference between the labeled data and the predicted data, correct the initial molecular docking model and return to continue training until a trained molecular docking model is obtained.

Relevant content of step S205 may refer to the foregoing embodiments, and details are not repeated here.

According to the training method of the molecular docking model of the embodiment of the present disclosure, based on the data in the protein structure database and the drug molecule database, randomly sample the protein and the drug molecule, obtain the combination of the protein and the drug molecule, and simulate the combination of the protein and the drug molecule , to get the labeled data. At the same time, feature extraction is performed on the combination of protein and drug molecules, and the extracted feature information is input into the initial molecular docking model for multi-task training to obtain a molecular docking model with strong generalization ability. The model is corrected based on the difference between the labeled data and the predicted data. By correcting the defects and deviations in the molecular docking model, the prediction of the binding conformation and affinity can be improved, and the reliability and accuracy of the prediction results can be improved. Further, it can improve the model's ability to explore proteins and drug molecules, reduce the exploration time, and improve the accuracy of molecular docking model prediction data.

FIG. 3 is a schematic flowchart of a training method for a molecular docking model provided by an embodiment of the present disclosure.

As shown in Figure 3, the training method of the molecular docking model may include:

S301, acquiring a combination of protein and drug molecule for training, and simulating the combination of protein and drug molecule to obtain labeled data of the combination of protein and drug molecule.

In some implementations, protein files can be downloaded from the source database, and a protein structure database can be constructed based on the downloaded protein files; drug molecule files can be downloaded from the source database, and a drug molecule database can be constructed based on the downloaded drug molecule files. The source database usually contains a large amount of protein structure data and drug molecule data. By downloading protein files and drug molecule files and building a database, a wealth of protein and drug information can be provided, including protein structure, protein sequence, drug name, structure, and physical and chemical properties. , biological activity and other information, it is helpful to obtain comprehensive protein and drug data and knowledge, and promote the research on protein and drug molecules.

It is understandable that there may be missing protein-related information from the downloaded protein files in the source database, resulting in failure to construct a protein structure database. Optionally, identify the abnormality of the downloaded protein file, obtain the abnormal protein file, and perform abnormal repair on the abnormal protein file. Further, the abnormal type of the abnormal protein file is obtained, and a repair program is obtained based on the abnormal type, and the protein file is modified based on the repair program. By repairing the missing information of the protein file, the integrity and correctness of the protein sequence can be ensured to obtain a more stable and reasonable protein structure, so that the complete and accurate protein structure can be used for subsequent molecular docking and drug screening .

In some implementations, a database of protein structures is constructed based on the repaired protein files. Proteins are sampled from the Protein Structure Database and drug molecules are obtained from the Drug Molecule Database. Then, the sampled protein and the sampled drug molecule are combined to obtain a combination of protein and drug molecule.

In some implementations, the sampled protein and drug molecule can be combined using the binding pocket of the protein. Sampled proteins and drug molecules can also be combined using protein-based structural information. It can be understood that when a protein and a drug molecule are combined, there is a high-probability binding region, and the binding pocket finds this region on the protein, guiding the drug molecule to combine with the protein in this region.

Optionally, the binding pockets of the sampled proteins can be mined to obtain the binding pockets of the sampled proteins. Based on the binding pocket of the protein, the sampled protein and drug molecule are combined to obtain a combination of protein and drug molecule. The binding pocket of a protein is the key area for the interaction between a drug molecule and a protein. Combining proteins and drug molecules using the binding pocket can screen out drug molecules with high binding activity and selectivity, which helps to guide drug optimization and develop.

Optionally, the structural information of the sampled protein is determined, and based on the structural information of the protein, the sampled protein and the drug molecule are combined to obtain a combination of the protein and the drug molecule. The structural information of proteins can reveal the function of proteins, analyze the interaction mechanism between drugs and proteins, combine proteins and drug molecules according to the structural information of proteins, and screen drug molecules, thus providing goals and directions for drug design.

Furthermore, based on the molecular docking algorithm, the molecular docking simulation is carried out for the combination of protein and drug molecule, and the binding conformation data of the combination of protein and drug molecule, the affinity data between protein and drug molecule, and the internal conformation data of the drug molecule are obtained. The binding conformation data, affinity data and internal conformation data of drug molecules obtained from the simulation are used as labeling data.

S302, input the combination of protein and drug molecule into the initial molecular docking model, perform multi-task training on the initial molecular docking model, and obtain prediction data of the combination of protein and drug molecule.

Relevant content of steps S301-S302 may refer to the foregoing embodiments, and details are not repeated here.

S303. Determine label data corresponding to each prediction head.

In some implementations, the annotation data of the combination of protein and drug molecule can be matched with each prediction head, and the annotation data corresponding to each prediction head can be determined. Optionally, the combined conformational data of the protein and the drug molecule can be used as the annotation data of the first prediction head; the affinity data between the protein and the drug molecule can be used as the annotation data of the second prediction head; the internal conformation of the drug molecule can be The data is used as the annotation data of the third prediction head.

S304. Obtain the difference between the prediction data of the prediction head and the corresponding label data, and obtain the first loss function of the initial molecular docking model.

In some implementations, the loss function of each prediction head can be calculated according to the difference between the prediction data and the corresponding label data. Optionally, a corresponding second loss function can be defined for each prediction head. For example, mean square error may be used as the second loss function, and cross entropy may also be used as the second loss function.

In some implementations, a second loss function for the prediction heads is determined based on the difference of each prediction head. Optionally, the mean square error is used as the second loss function, and the second loss function corresponding to each prediction head is calculated according to the mean square error between the prediction data output by each prediction head and the label data corresponding to each prediction head.

Further, the second loss function of each prediction head is weighted to obtain the first loss function. Optionally, the weighting method may use uniform weighting, that is, each second loss function has the same weight value.

S305. Correct the initial molecular docking model based on the first loss function.

In some implementations, model parameters of the initial molecular docking model can be modified based on the first loss function. Optionally, the correction gradient of the correction model parameters can be determined according to the first loss function, and then based on the correction gradient, the initial molecular docking model is reversely corrected, so as to improve the performance and stability of the molecular docking model.

S306, continue training until a trained molecular docking model is obtained.

Relevant content of step S306 may refer to the foregoing embodiments, and details are not repeated here.

According to the training method of the molecular docking model in the embodiment of the present disclosure, based on the data in the protein structure database and the drug molecule database, randomly sample the protein and the drug molecule, obtain the combination of the protein and the drug molecule, and continuously simulate the combination of the protein and the drug molecule , to get the labeled data. Input the combination of protein and drug molecules into the initial molecular docking model for multi-task training to obtain a molecular docking model with strong generalization ability. The model is corrected based on the difference between the labeled data and the predicted data. By correcting the defects and deviations in the molecular docking model, the prediction of the binding conformation and affinity can be improved, and the reliability and accuracy of the prediction results can be improved. Further, it can improve the model's ability to explore proteins and drug molecules, reduce the exploration time, and improve the accuracy of molecular docking model prediction data.

FIG. 4 is a schematic flowchart of a training method for a molecular docking model provided by an embodiment of the present disclosure.

As shown in Figure 4, the training method of the molecular docking model may include:

S401. Obtain a combination of protein and drug molecule for training, and simulate the combination of protein and drug molecule to obtain labeled data of the combination of protein and drug molecule.

S402, performing feature extraction on the combination of protein and drug molecule.

S403, encoding the extracted feature information to obtain a feature vector representation corresponding to the combination of the protein and the drug molecule.

Relevant content of steps S401-S403 may refer to the foregoing embodiments, and details are not repeated here.

S404. Input the feature vector representations to each prediction head respectively, and the prediction heads learn the feature vector representations to obtain prediction data of the prediction heads.

It is understandable that different prediction heads output different prediction data. The prediction data can be the internal conformation data of the drug molecule, the binding conformation data of the combination of the protein and the drug molecule, and the affinity data between the protein and the drug molecule.

Optionally, the feature vector representation is input into the first prediction head, and the internal conformation data of the drug molecule is obtained by the first prediction head based on the feature vector representation.

Optionally, the eigenvector representation is input into the second prediction head, and the second prediction head obtains the binding conformation data of the combination of the protein and the drug molecule based on the eigenvector representation.

Optionally, the feature vector representation is input into the third prediction head for training, and the affinity data between the protein and the drug molecule output by the third prediction head is obtained.

S405. Based on the prediction data of the prediction head, the prediction data of the combination of the protein and the drug molecule is obtained.

It can be understood that, based on the prediction data of the prediction head, the prediction data of the combination of protein and drug molecule can be predicted.

In some implementations, the internal conformation of drug molecules contains information such as interatomic bonds, chemical bond angles, and interactions between electron shells. These information can enable molecular docking models to learn fine spatial structures and atomic interactions. Therefore, based on the internal conformation data of the drug molecule, the geometric conformation of the combination of the protein and the drug molecule can be predicted.

In some implementations, the binding conformation of the combination of protein and drug molecule includes information such as the distance between the drug molecule atom and the protein atom, the change of the torsion angle after the drug molecule is combined, and the size of translation and rotation. This information can allow the molecular docking model to learn the drug Interaction between molecules and proteins. Therefore, based on the binding conformation data of the protein-drug molecule combination, the binding force of the protein-drug molecule combination can be predicted.

In some implementations, the affinity data between proteins and drug molecules includes information such as binding or repulsion between proteins and drug molecules, and the magnitude of binding affinity. These information can help molecular docking models to screen out more likely to interact with selected proteins. drug molecule. Therefore, based on the affinity data between the protein and the drug molecule, the binding or repelling affinity of the combination of the protein and the drug molecule can be predicted.

S406. Based on the difference between the labeled data and the predicted data, correct the initial molecular docking model and return to continue training until a trained molecular docking model is obtained.

Relevant content of step S406 may refer to the foregoing embodiments, and details are not repeated here.

In the pre-training process of the molecular docking model shown in Figure 5, the protein and drug molecules collected from the protein structure database and drug molecule database are combined to obtain the combination of protein and drug molecule. The combination of protein and drug molecule is simulated to obtain the labeled data of the combination of protein and drug molecule. Extract features from the combination of protein and drug molecules, input them into the neural network of the molecular docking model, perform multi-prediction head training, and obtain the output prediction data. Calculate the prediction data of the prediction head and the corresponding label data loss function, based on the loss function, correct the initial molecular docking model, and continue to train the revised molecular docking model based on the subsequent combination of protein and drug molecules until the end of training is satisfied condition, the trained molecular docking model is obtained.

According to the training method of the molecular docking model of the embodiment of the present disclosure, based on the data in the protein structure database and the drug molecule database, randomly sample the protein and the drug molecule, obtain the combination of the protein and the drug molecule, and simulate the combination of the protein and the drug molecule , to get the labeled data. At the same time, feature extraction is performed on the combination of protein and drug molecules, and the extracted feature information is input into the initial molecular docking model for multi-task training to obtain a molecular docking model with strong generalization ability. The model is corrected based on the difference between the labeled data and the predicted data. By correcting the defects and deviations in the molecular docking model, the prediction of the binding conformation and affinity can be improved, and the reliability and accuracy of the prediction results can be improved. Further, it can improve the model's ability to explore proteins and drug molecules, reduce the exploration time, and improve the accuracy of molecular docking model prediction data.

On the basis of the above embodiments, in the embodiments of the present disclosure, the process of obtaining the target molecular docking model after starting the training method of the molecular docking model can be explained.

FIG. 6 is a schematic flowchart of obtaining a target molecular docking model in a molecular docking model training method provided by an embodiment of the present disclosure.

As shown in Figure 6, the method for obtaining the docking model of the target molecule may include:

S601, performing multi-task training on the initial molecular docking model based on the sampled protein and drug molecule combinations, to obtain a trained molecular docking model.

Relevant content of step S601 may refer to the foregoing embodiments, and details are not repeated here.

S602. Based on the prediction demand information, obtain the target molecular docking model from the target prediction heads that need to be retained among the multi-prediction heads in the trained molecular docking model, and decouple the remaining prediction heads.

In some implementations, the target prediction head that needs to be retained among the multiple prediction heads in the trained molecular docking model can be determined according to the prediction requirement information, and the target prediction head is related to the result of the prediction data. Optionally, one or more prediction heads may be determined as target prediction heads. Optionally, the first prediction head may be retained, and the prediction data of the internal conformation of the drug molecule may be retained. Optionally, the second prediction head can be retained to retain the prediction data of the binding conformation of the protein combined with the drug molecule. Optionally, the third prediction head can be retained, retaining the prediction data of the affinity between the protein and the drug molecule.

Further, the remaining prediction heads are separated from the target prediction heads, and the remaining prediction heads are decoupled from the trained molecular docking model. Optionally, the prediction head can be stored, and when a specified prediction task needs to be performed, the target prediction head corresponding to the prediction task can be called, that is, the target prediction head is connected behind the backbone network of the molecular docking model, and through the backbone network and the target prediction head to obtain the prediction data corresponding to the prediction task.

According to the training method of the molecular docking model of the embodiment of the present disclosure, based on the data in the protein structure database and the drug molecule database, randomly sample the protein and the drug molecule, obtain the combination of the protein and the drug molecule, and simulate the combination of the protein and the drug molecule , to get the labeled data. Input the combination of protein and drug molecules into the initial molecular docking model for multi-task training to obtain a molecular docking model with strong generalization ability. The model is corrected based on the difference between the labeled data and the predicted data. By correcting the defects and deviations in the molecular docking model, the prediction of the binding conformation and affinity can be improved, and the reliability and accuracy of the prediction results can be improved. Further, it can improve the model's ability to explore proteins and drug molecules, reduce the exploration time, and improve the accuracy of molecular docking model prediction data. Furthermore, the target prediction head is reserved based on the demand, and the remaining prediction heads are decoupled, so that the molecular docking model can not only realize the parallel processing of multi-tasks, but also enable the molecular docking model to perform individual tasks. In the single-task execution mode It can reduce the complexity of the molecular docking model and reduce the occupation of computing resources.

FIG. 7 is a schematic flowchart of a method for obtaining information on a drug target provided by an embodiment of the present disclosure.

As shown in Figure 7, the information acquisition method of the drug target may include:

S701, obtaining the combination of target protein and candidate drug molecules.

It can be understood that the target protein refers to the protein located in the target within a certain range near the drug molecule after the drug binds to the target. Common target proteins can be enzymes, receptors, and other regulatory proteins. A candidate drug molecule can be a compound.

S702, inputting the combination of the target protein and the candidate drug molecule into the pre-trained target molecule docking model, obtaining binding conformation data of the combination of the target protein and the candidate drug molecule and affinity data between the target protein and the candidate drug molecule.

In some implementations, the protein structure database and the drug molecule database can be downloaded from the source database, and according to the goals and needs of drug research, the proteins related to the research in the protein structure database are designated as target proteins. Alternatively, target proteins can be assigned based on protein structure information. Screen the drug molecules that may interact with the target protein from the drug molecule database as candidate drug molecules. Combining the target protein and candidate drug molecules to obtain a combination of target protein and candidate drug molecules.

It should be noted that the target molecular docking model can be obtained by using the training method of the molecular docking model shown in Fig. 1 to Fig. 6 , which will not be repeated here.

In the embodiment of the present disclosure, feature extraction is performed on the combination of target protein and candidate drug molecule, the extracted feature information is input into the target molecule docking model, and the combination of target protein and candidate drug molecule is predicted to obtain the target protein Binding conformational data combined with candidate drug molecules and affinity data between the target protein and the candidate drug molecule.

According to the information acquisition method of a drug target in an embodiment of the present disclosure, the conformational data and affinity data of a combination of a target protein and a candidate drug molecule can be predicted by using a pre-trained target molecule docking model. The target molecular docking model is obtained by using the training method of the disclosed molecular docking model. The molecular docking model has strong generalization ability, and can predict the conformation data and affinity data of the combination of the target protein and the candidate drug molecule more quickly. High accuracy rate.

A schematic diagram of the predicted protein-drug molecule affinity data and binding conformation as shown in FIG. 8 . Obtain the target protein and candidate drug molecules from the database, combine the target protein and candidate drug molecules, and obtain the combination of target protein and candidate drug molecules. Furthermore, the combination of target protein and candidate drug molecule is input into the pre-trained molecular docking model, which can predict and output the binding conformation data of the combination of target protein and candidate drug molecule and the interaction between target protein and candidate drug molecule. affinity data.

Corresponding to the molecular docking model training methods provided in the above several embodiments, an embodiment of the present disclosure also provides a molecular docking model training device, since the molecular docking model training device provided by the embodiment of the present disclosure is the same as the above The training methods of the molecular docking model provided in several embodiments correspond to each other, so the implementation of the above-mentioned training method of the molecular docking model is also applicable to the training device of the molecular docking model provided in the embodiments of the present disclosure, and will not be described in detail in the following examples describe.

FIG. 9 is a schematic structural diagram of a training device for a molecular docking model provided by an embodiment of the present disclosure.

As shown in FIG. 9 , a training device 900 for a molecular docking model according to an embodiment of the present disclosure includes an acquisition module 901 , a training module 902 and a correction module 903 .

The obtaining module 901 is used to obtain the combination of protein and drug molecule used for training, and simulate the combination of protein and drug molecule to obtain the labeled data of the combination of protein and drug molecule;

A training module 902, configured to input the combination of the protein and the drug molecule into the initial molecular docking model, perform multi-task training on the initial molecular docking model, and obtain the prediction data of the combination of the protein and the drug molecule;

The correction module 903 is configured to correct the initial molecular docking model based on the difference between the labeled data and the predicted data, and return to continue training until a trained molecular docking model is obtained.

In an embodiment of the present disclosure, the training module 902 is also used to: extract features from the combination of the protein and the drug molecule; encode the extracted feature information to obtain the features corresponding to the combination of the protein and the drug molecule Vector representation; multi-prediction head training is performed on the feature vector representation to obtain prediction data of the combination of the protein and the drug molecule.

In an embodiment of the present disclosure, the correction module 903 is further configured to: determine the label data corresponding to each prediction head; obtain the difference between the prediction data of the prediction head and the corresponding label data, and obtain the initial molecule A first loss function of the docking model; modifying the initial molecular docking model based on the first loss function.

In an embodiment of the present disclosure, the correction module 903 is further configured to: determine the second loss function of the prediction head according to the difference of each prediction head; The second loss function performs weighting to obtain the first loss function.

In an embodiment of the present disclosure, the training module 902 is further configured to: respectively input the feature vector representations to each prediction head, and let the prediction heads learn the feature vector representations to obtain the prediction The predicted data of the head; based on the predicted data of the predicted head, the predicted data of the combination of the protein and the drug molecule is obtained.

In an embodiment of the present disclosure, the training module 902 is further configured to: input the feature vector representation into the first prediction head, and the first prediction head obtains the drug molecule based on the feature vector representation The internal conformation data of the internal conformation data; the feature vector representation is input into the second prediction head, and the second prediction head is based on the feature vector representation to obtain the binding conformation data of the combination of the protein and the drug molecule; the feature vector Indicates that it is input into the third prediction head for training, and the affinity data between the protein and the drug molecule output by the third prediction head is obtained.

In an embodiment of the present disclosure, the training module 902 is further configured to: based on the prediction demand information, predict the target heads that need to be reserved from the multi-prediction heads in the trained molecular docking model, and perform the remaining prediction heads Decoupling is performed to obtain the docking model of the target molecule.

In an embodiment of the present disclosure, the acquisition module 901 is further configured to: perform molecular docking simulation on the combination of the protein and the drug molecule based on a molecular docking algorithm, to obtain binding conformation data of the combination of the protein and the drug molecule, The affinity data between the protein and the drug molecule and the internal conformation data of the drug molecule; the binding conformation data, the affinity data and the internal conformation data of the drug molecule obtained by simulation are used as the labeling data.

In an embodiment of the present disclosure, the acquisition module 901 is further configured to: sample proteins from the protein structure database, and obtain drug molecules from the drug molecule database; combine the sampled proteins and sampled drug molecules to obtain The protein is combined with a drug molecule.

In an embodiment of the present disclosure, the acquisition module 901 is further configured to: mine the binding pocket of the sampled protein to obtain the binding pocket of the sampled protein; based on the binding pocket of the protein, The sampled protein and drug molecule are combined to obtain a combination of the protein and drug molecule.

In an embodiment of the present disclosure, the acquisition module 901 is further configured to: determine the structural information of the sampled protein, and combine the sampled protein and drug molecules based on the structural information of the protein , to obtain a combination of the protein and the drug molecule.

In an embodiment of the present disclosure, the obtaining module 901 is further configured to: download protein files from a source database, and construct the protein structure database based on the downloaded protein files.

In an embodiment of the present disclosure, the obtaining module 901 is further configured to: identify abnormality of the downloaded protein file, obtain the abnormal protein file, and repair the abnormal protein file.

In an embodiment of the present disclosure, the obtaining module 901 is further configured to: obtain the abnormal type of the abnormal protein file, and obtain a repair program based on the abnormal type, and repair the protein file based on the repair program to modify.

According to the training method of the molecular docking model of the embodiment of the present disclosure, based on the data in the protein structure database and the drug molecule database, randomly sample the protein and the drug molecule, obtain the combination of the protein and the drug molecule, and simulate the combination of the protein and the drug molecule , to get the labeled data. At the same time, feature extraction is performed on the combination of protein and drug molecules, and the extracted feature information is input into the initial molecular docking model for multi-task training to obtain a molecular docking model with strong generalization ability. The model is corrected based on the difference between the labeled data and the predicted data. By correcting the defects and deviations in the molecular docking model, the prediction of the binding conformation and affinity can be improved, and the reliability and accuracy of the prediction results can be improved. Further, it can improve the model's ability to explore proteins and drug molecules, reduce the exploration time, and improve the accuracy of molecular docking model prediction data.

According to an embodiment of the present disclosure, the present disclosure also provides a drug target information acquisition device, which is used to implement the above-mentioned molecular docking model training method.

Fig. 10 is a schematic structural diagram of an information acquisition device for drug targets according to the first embodiment of the present disclosure.

As shown in FIG. 10 , an information acquisition device 1000 of a drug target according to an embodiment of the present disclosure includes: an acquisition module 1001 .

The obtaining module 1001 is used to obtain the combination of the target protein and the candidate drug molecule, and input the combination of the target protein and the candidate drug molecule into the pre-trained target molecule docking model, and obtain the combination of the target protein and the candidate drug molecule. Binding conformational data and affinity data between the target protein and the candidate drug molecule.

According to the information acquisition method of a drug target in an embodiment of the present disclosure, the conformational data and affinity data of a combination of a target protein and a candidate drug molecule can be predicted by using a pre-trained target molecule docking model. The target molecular docking model is obtained by using the training method of the disclosed molecular docking model. The molecular docking model has strong generalization ability, and can predict the conformation data and affinity data of the combination of the target protein and the candidate drug molecule more quickly. High accuracy rate.

In the technical solution of the present disclosure, the acquisition, storage and application of the user's personal information involved are in compliance with relevant laws and regulations, and do not violate public order and good customs.

According to the embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

FIG. 11 shows a schematic block diagram of an example electronic device 1100 that may be used to implement embodiments of the present disclosure. Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the disclosure described and/or claimed herein.

As shown in FIG. 11 , the device 1100 includes a computing unit 1101 that can be configured according to computer programs/instructions stored in a read-only memory (ROM) 1102 or loaded from a storage unit 1106 into a random access memory (RAM) 1103/instructions. instructions to perform various appropriate actions and processing. In the RAM 1103, various programs and data necessary for the operation of the device 1100 can also be stored. The computing unit 1101 , ROM 1102 , and RAM 1103 are connected to each other through a bus 1104 . An input/output (I/O) interface 1105 is also connected to the bus 1104 .

Multiple components in the device 1100 are connected to the I/O interface 1105, including: an input unit 1106 such as a keyboard, a mouse, etc.; an output unit 1107, such as various types of displays, speakers, etc.; a storage unit 1108, such as a magnetic disk, an optical disk, etc.; And a communication unit 1109, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 1109 allows the device 1100 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.

The computing unit 1101 may be various general-purpose and/or special-purpose processing components having processing and computing capabilities. Some examples of computing units 1101 include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processing processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1101 executes the various methods and processes described above, such as the training method of the molecular docking model. For example, in some embodiments, the training method of the molecular docking model can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit 1106. In some embodiments, part or all of the computer program/instructions may be loaded and/or installed onto device 1100 via ROM 1102 and/or communication unit 1109 . When the computer program/instruction is loaded into the RAM 1103 and executed by the computing unit 1101, one or more steps of the above-described training method of the molecular docking model can be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured in any other appropriate way (for example, by means of firmware) to execute the training method of the molecular docking model.

Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips Implemented in a system of systems (SOC), load programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being embodied in one or more computer programs/instructions executable and/or interpretable on a programmable system including at least one programmable processor, the The programmable processor may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from the storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device , and the at least one output device.

Program codes for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special purpose computer, or other programmable data processing devices, so that the program codes, when executed by the processor or controller, make the functions/functions specified in the flow diagrams and/or block diagrams Action is implemented. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer discs, hard drives, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

To provide for interaction with the user, the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user. ); and a keyboard and pointing device (eg, a mouse or a trackball) through which a user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.

The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

A computer system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by computer programs/instructions running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, each step described in the disclosure may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution disclosed in the present disclosure can be achieved, no limitation is imposed herein.

The specific implementation manners described above do not limit the protection scope of the present disclosure. It should be apparent to those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made depending on design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present disclosure shall be included within the protection scope of the present disclosure.

Claims (33)

1. A method of training a molecular docking model, comprising:

acquiring a protein and drug molecule combination for training, and simulating the protein and drug molecule combination to obtain labeling data of the protein and drug molecule combination;

inputting the protein and drug molecule combination into an initial molecule docking model, and performing multitasking training on the initial molecule docking model to obtain prediction data of the protein and drug molecule combination;

And correcting the initial molecular docking model based on the difference between the labeling data and the prediction data, and returning to continue training until a trained molecular docking model is obtained.

2. The method of claim 1, wherein the inputting the protein and drug molecule combination into an initial molecular docking model, and the multitasking the initial molecular docking model to obtain predicted data for the protein and drug molecule combination, comprises:

extracting the characteristics of the combination of the protein and the drug molecules;

encoding the extracted characteristic information to obtain a characteristic vector representation corresponding to the combination of the protein and the drug molecule;

and performing multi-prediction head training on the feature vector representation to obtain prediction data of the combination of the protein and the drug molecules.

3. The method of claim 1, wherein the modifying the initial molecular docking model based on the differences between the annotation data and the prediction data comprises:

determining the marking data corresponding to each prediction head;

obtaining the difference between the predicted data of the predicted head and the corresponding labeling data, and obtaining a first loss function of the initial molecular docking model;

And correcting the initial molecular docking model based on the first loss function.

4. A method according to claim 3, wherein said obtaining the difference between the predicted data of the predicted head and the corresponding labeling data, obtaining a first loss function of the initial molecular docking model, comprises:

determining a second loss function for each of the pre-measurement heads based on the difference for that pre-measurement head;

and weighting the second loss function of each pre-measurement head to obtain the first loss function.

5. The method of claim 2, wherein said performing multi-predictive head training on said feature vector representation to obtain predictive data for said protein in combination with a drug molecule comprises:

inputting the characteristic vector representations to each prediction head respectively, and learning the characteristic vector representations by the prediction heads to obtain prediction data of the prediction heads;

based on the predicted data of each predicted head, predicted data of the combination of the protein and the drug molecule is obtained.

6. The method of claim 5, wherein the inputting the feature vector representations into each prediction head, the feature vector representations being learned by the prediction heads to obtain the prediction data of the prediction heads, comprises:

Inputting the feature vector representation into a first prediction head, obtaining internal conformational data of the drug molecule based on the feature vector representation by the first prediction head;

inputting the feature vector representation into a second prediction head, and obtaining binding conformation data of the combination of the protein and the drug molecule based on the feature vector representation by the second prediction head;

inputting the characteristic vector representation into a third prediction head for training, and obtaining affinity data between the protein and the drug molecules output by the third prediction head.

7. The method of claim 2, wherein the method further comprises:

and based on the prediction demand information, target pre-measurement heads which need to be reserved in the multiple pre-measurement heads in the trained molecular docking model are used for decoupling the residual pre-measurement heads, so that the target molecular docking model is obtained.

8. The method of any one of claims 1-7, wherein simulating the combination of the protein and the drug molecule results in labeling data for the combination of the protein and the drug molecule, comprising:

based on a molecular docking algorithm, carrying out molecular docking simulation on the protein and drug molecule combination to obtain binding conformation data of the protein and drug molecule combination, affinity data between the protein and drug molecule and internal conformation data of the drug molecule;

And taking the binding conformation data, the affinity data and the internal conformation data of the drug molecules obtained through simulation as the labeling data.

9. The method of claim 1, wherein the obtaining a protein for training in combination with a drug molecule comprises:

sampling proteins from a protein structure database, and obtaining drug molecules from a drug molecule database;

combining the sampled protein with the sampled drug molecule to obtain the combination of the protein and the drug molecule.

10. The method of claim 9, wherein the combining the sampled protein and sampled drug molecule results in a combination of the protein and drug molecule comprising:

excavating the binding pocket of the sampled protein to obtain the binding pocket of the sampled protein;

combining the sampled protein with a drug molecule based on the binding pocket of the protein, resulting in a combination of the protein and the drug molecule.

11. The method of claim 9, wherein the combining the sampled protein and sampled drug molecule results in a combination of the protein and drug molecule comprising:

and determining the structural information of the sampled protein, and combining the sampled protein with a drug molecule based on the structural information of the protein to obtain the combination of the protein and the drug molecule.

12. The method of claim 8, wherein prior to sampling the protein from the protein structure database, further comprising:

downloading a protein file from a source database, and constructing the protein structure database based on the downloaded protein file.

13. The method of claim 12, wherein the constructing the protein structure database is based on the downloaded protein files;

and carrying out abnormality recognition on the downloaded protein file, obtaining an abnormal protein file, and carrying out abnormality repair on the abnormal protein file.

14. The method of claim 13, wherein the performing an exception repair on the exception protein file comprises:

obtaining the abnormal type of the abnormal protein file, obtaining a repair program based on the abnormal type, and modifying the protein file based on the repair program.

15. An information acquisition method of a drug target, comprising:

obtaining a target protein and candidate drug molecule combination, inputting the target protein and candidate drug molecule combination into a pre-trained target molecule docking model, and obtaining binding conformation data of the target protein and candidate drug molecule combination and affinity data between the target protein and candidate drug molecule;

Wherein the target molecule docking model is a model obtained based on the training method according to any one of claims 1-14.

16. A training device for a molecular docking model, comprising:

the acquisition module is used for acquiring a protein and drug molecule combination for training and simulating the protein and drug molecule combination to obtain labeling data of the protein and drug molecule combination;

the training module is used for inputting the protein and drug molecule combination into an initial molecule docking model, and performing multitasking training on the initial molecule docking model to obtain prediction data of the protein and drug molecule combination;

and the correction module is used for correcting the initial molecular docking model based on the difference between the marking data and the prediction data and returning to continue training until a trained molecular docking model is obtained.

17. The apparatus of claim 16, wherein the training module is further to:

extracting the characteristics of the combination of the protein and the drug molecules;

encoding the extracted characteristic information to obtain a characteristic vector representation corresponding to the combination of the protein and the drug molecule;

and performing multi-prediction head training on the feature vector representation to obtain prediction data of the combination of the protein and the drug molecules.

18. The apparatus of claim 16, wherein the correction module is further configured to:

determining the marking data corresponding to each prediction head;

obtaining the difference between the predicted data of the predicted head and the corresponding labeling data, and obtaining a first loss function of the initial molecular docking model;

and correcting the initial molecular docking model based on the first loss function.

19. The apparatus of claim 18, wherein the correction module is further configured to:

determining a second loss function for each of the pre-measurement heads based on the difference for that pre-measurement head;

and weighting the second loss function of each pre-measurement head to obtain the first loss function.

20. The apparatus of claim 17, wherein the training module is further to:

inputting the characteristic vector representations to each prediction head respectively, and learning the characteristic vector representations by the prediction heads to obtain prediction data of the prediction heads;

and obtaining the prediction data of the combination of the protein and the drug molecule based on the prediction data of the prediction head.

21. The apparatus of claim 20, wherein the training module is further to:

Inputting the feature vector representation into a first prediction head, obtaining internal conformational data of the drug molecule based on the feature vector representation by the first prediction head;

inputting the feature vector representation into a second prediction head, and obtaining binding conformation data of the combination of the protein and the drug molecule based on the feature vector representation by the second prediction head;

inputting the characteristic vector representation into a third prediction head for training, and obtaining affinity data between the protein and the drug molecules output by the third prediction head.

22. The apparatus of claim 17, wherein the training module is further to:

and based on the prediction demand information, target pre-measurement heads which need to be reserved in the multiple pre-measurement heads in the trained molecular docking model are used for decoupling the residual pre-measurement heads, so that the target molecular docking model is obtained.

23. The apparatus of any of claims 16-22, wherein the acquisition module is further to:

based on a molecular docking algorithm, carrying out molecular docking simulation on the protein and drug molecule combination to obtain binding conformation data of the protein and drug molecule combination, affinity data between the protein and drug molecule and internal conformation data of the drug molecule;

And taking the binding conformation data, the affinity data and the internal conformation data of the drug molecules obtained through simulation as the labeling data.

24. The apparatus of claim 16, wherein the acquisition module is further configured to:

sampling proteins from a protein structure database, and obtaining drug molecules from a drug molecule database;

combining the sampled protein with the sampled drug molecule to obtain the combination of the protein and the drug molecule.

25. The apparatus of claim 24, wherein the means for obtaining is further configured to:

excavating the binding pocket of the sampled protein to obtain the binding pocket of the sampled protein;

combining the sampled protein with a drug molecule based on the binding pocket of the protein, resulting in a combination of the protein and the drug molecule.

26. The apparatus of claim 24, wherein the means for obtaining is further configured to:

and determining the structural information of the sampled protein, and combining the sampled protein with a drug molecule based on the structural information of the protein to obtain the combination of the protein and the drug molecule.

27. The apparatus of claim 23, wherein the means for obtaining is further configured to:

Downloading a protein file from a source database, and constructing the protein structure database based on the downloaded protein file.

28. The apparatus of claim 27, wherein the means for obtaining is further configured to:

and carrying out abnormality recognition on the downloaded protein file, obtaining an abnormal protein file, and carrying out abnormality repair on the abnormal protein file.

29. The apparatus of claim 28, wherein the means for obtaining is further configured to:

obtaining the abnormal type of the abnormal protein file, obtaining a repair program based on the abnormal type, and modifying the protein file based on the repair program.

30. An information acquisition device of a drug target, comprising:

the acquisition module is used for acquiring a target protein and candidate drug molecule combination, inputting the target protein and candidate drug molecule combination into a pre-trained target molecule docking model, and acquiring binding conformation data of the target protein and candidate drug molecule combination and affinity data between the target protein and candidate drug molecule;

wherein the target molecule docking model is a model obtained based on the training method according to any one of claims 1-14.

31. An electronic device, comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-15.

32. A non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-15.

33. A computer program product comprising computer programs/instructions which, when executed by a processor, implement the method steps of any of claims 1-15.