Methods, systems, equipment, and media for determining drug-target match
By using BERT networks and dual-channel GNN networks to process drug molecules and target protein sequences, the problem of low speed and accuracy in determining drug-target protein matching was solved, thereby improving the efficiency of new drug development and reducing costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2023-04-24
- Publication Date
- 2026-05-26
AI Technical Summary
Existing methods for determining whether a drug has a targeted therapeutic effect on a target protein are slow and inaccurate, resulting in a long and costly process for new drug development.
Bert network and dual-channel GNN network were used to perform multimodal processing and feature extraction on drug molecule sequences and target protein sequences. The total confidence score was obtained by weighted summation to determine the match between the drug and the target.
It enables the rapid and accurate determination of the targeting relationship between new drugs and target proteins, improving drug screening efficiency, reducing computational resource consumption, and lowering the cost of new drug development.
Smart Images
Figure CN116486899B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of natural language processing and digital healthcare, and particularly to a method, system, device, and medium for determining drug-target matching. Background Technology
[0002] The process of developing a new drug from laboratory research to commercialization involves four stages: drug discovery, preclinical research, clinical research, and market approval. Determining the match between a drug and its target protein is a crucial aspect of drug discovery. It involves using appropriate methods to assess the biological activity, pharmacological effects, and medicinal value of compounds that may be used as drugs, as well as the target proteins that the drugs may treat.
[0003] Drug development is an extremely challenging process, often referred to in the industry as the "Rule of Thirty," which states that on average, it takes ten years and costs one billion US dollars for a drug to go from laboratory development to clinical approval, with a success rate of only 10%. Statistics show that this rule is currently increasing in both time and cost, with some drugs costing well over one billion US dollars. Traditionally, determining whether a drug can target a specific protein involves cellular screening. This involves treating target cells with candidate drugs and then using biochemical experiments to determine if the drug has a therapeutic effect on the target cells.
[0004] The inventors realized that the above-mentioned method of determining whether a drug has targeted therapy for a target protein requires multiple experiments to determine the targeting relationship between a new drug and the target protein to be determined, and the experimental results are random. Therefore, it is impossible to quickly and accurately determine the targeting relationship between a new drug and the target protein to be determined. Summary of the Invention
[0005] The purpose of this invention is to provide a method, system, device, and medium for determining drug-target matching. This addresses the problem of low speed and accuracy in existing methods for determining whether a drug is effective against a target protein, which suffer from low accuracy in determining the drug-target protein targeting relationship.
[0006] Firstly, a method for determining drug-target matching is provided, including:
[0007] Obtain drug molecule sequences and target protein sequences;
[0008] The drug molecule sequence and the target protein sequence are respectively input into two BERT networks cascaded in the first model, and multimodal processing is performed on the drug molecule sequence and the target protein sequence to obtain a first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence.
[0009] The drug molecule sequence and the target protein sequence are respectively input into the dual-channel GNN network of the second model, and feature extraction and feature splicing are performed on the drug molecule sequence and the target protein sequence respectively to obtain a second confidence score; wherein, the second model is used to obtain the association relationship between the drug molecule sequence and the target protein sequence;
[0010] The first confidence score and the second confidence score are weighted and summed to obtain the total confidence score. When the total confidence score is greater than or equal to a preset confidence threshold, the drug is determined to be a match for the target.
[0011] Secondly, a drug-target matching determination system is provided, including:
[0012] The data acquisition module is used to acquire drug molecule sequences and target protein sequences;
[0013] The first confidence score acquisition module is used to input the drug molecule sequence and the target protein sequence into two BERT networks cascaded in the first model, respectively, and perform multimodal processing on the drug molecule sequence and the target protein sequence to obtain the first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence.
[0014] The second confidence score acquisition module is used to input the drug molecule sequence and the target protein sequence into the dual-channel GNN network of the second model, respectively, and to perform feature extraction and feature splicing on the drug molecule sequence and the target protein sequence to obtain the second confidence score; wherein, the second model is used to obtain the correlation between the drug molecule sequence and the target protein sequence.
[0015] The matching judgment module is used to perform a weighted summation of the first confidence score and the second confidence score to obtain a total confidence score, and to determine that the drug matches the target when the total confidence score is greater than or equal to a preset confidence threshold.
[0016] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described drug and target matching determination method.
[0017] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described method for determining drug-target matching.
[0018] This invention discloses a method, system, device, and medium for determining drug-target matching. The method involves inputting the acquired drug molecule sequence and target protein sequence into two cascaded BERT networks in a first model. Multimodal processing is performed on both sequences to obtain a first confidence score. Simultaneously, the drug molecule sequence and target protein sequence are input into a dual-channel GNN network in a second model. Feature extraction and feature concatenation are performed on both sequences to obtain a second confidence score. The first and second confidence scores are then weighted and summed to obtain a total confidence score. A drug-target match is determined when the total confidence score is greater than or equal to a preset confidence threshold. This method enables rapid and accurate determination of the targeting relationship between a new drug and the target protein, effectively improving the efficiency of drug screening. By using BERT networks to perform position encoding and feature embedding on the drug molecule sequence and target protein sequence, the orderliness of the drug molecule sequence and target protein can be improved. To avoid errors in drug molecule or target protein sequences due to random scrambling during BERT model initialization, which can negatively impact subsequent analysis, graph neural networks, despite having fewer layers, can fully utilize the bond connections between atoms in the drug molecule or target protein sequences. This significantly improves accuracy and reduces analysis time. Effective cascading of graph neural networks and BERT networks, due to the smaller model size, greatly increases the scale and speed of drug-target matching assessments, reduces computational resources, and significantly improves the efficiency of new drug development, potentially leading to cost reduction and efficiency improvement. Attached Figure Description
[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort:
[0020] Figure 1 The diagram shows an application environment of a drug and target matching determination method according to an embodiment of the present invention.
[0021] Figure 2 The diagram shows a flowchart of a method for determining drug-target matching in one embodiment of the present invention.
[0022] Figure 3 The diagram shows a flowchart of obtaining a drug molecule matrix in one embodiment of the present invention.
[0023] Figure 4The diagram shows a flowchart of obtaining the second confidence score in one embodiment of the present invention.
[0024] Figure 5 The diagram shows a flowchart of obtaining drug feature vectors in one embodiment of the present invention.
[0025] Figure 6 The diagram shows a flowchart of the first GNN network processing procedure in one embodiment of the present invention.
[0026] Figure 7 The diagram shown is a structural block diagram of a drug and target matching determination system in one embodiment of the present invention.
[0027] Figure 8 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention;
[0028] Figure 9 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation
[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0030] The drug-target matching determination method provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network, and the server can send the drug and target determination results to the client. The acquired drug molecule sequence and target protein sequence are input into two cascaded BERT networks in the first model. Multimodal processing is performed on the drug molecule sequence and target protein sequence to obtain a first confidence score. Simultaneously, the drug molecule sequence and target protein sequence are input into a dual-channel GNN network in the second model, where feature extraction and feature concatenation are performed to obtain a second confidence score. The first and second confidence scores are then weighted and summed to obtain a total confidence score. When the total confidence score is greater than or equal to a preset confidence threshold, the drug and target are determined to be a match. This achieves rapid and accurate determination of the targeting relationship between a new drug and the target protein, effectively improving the efficiency of drug screening. By using BERT networks to perform position encoding and feature embedding on the drug molecule sequence and target protein sequence respectively, the orderliness of the drug molecule sequence and target protein can be improved. To avoid errors in drug molecule or target protein sequences due to random scrambling during BERT model initialization, which can negatively impact subsequent judgments, a graph neural network is used. Although with fewer layers, it fully utilizes the bond connections between atoms in the drug molecule or target protein sequences, significantly improving accuracy and reducing time. Effective cascading of graph neural networks and BERT networks, due to the smaller model size, greatly increases the scale and speed of drug-target matching determination, reduces computational resources, and significantly improves the efficiency of new drug development, offering the potential for cost reduction and efficiency improvement. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a dedicated server or a server cluster. The invention is described in detail below through specific embodiments.
[0031] Please see Figure 2 As shown, Figure 2 A flowchart illustrating a drug-target matching determination method provided in an embodiment of the present invention includes the following steps:
[0032] S10. Obtain the drug molecule sequence and target protein sequence.
[0033] Because it's necessary to determine whether a drug is suitable for acting on a target, and drugs exist in the form of molecular sequences, with different drugs having different molecular sequences. When a drug treats a target, it acts on the target protein to determine if the drug has a targeted effect. The drug molecular sequence refers to the drug molecular structure represented by a string, such as the expression SMILES, similar to a language sequence. The target protein sequence is data representing the protein, composed of 20 different letters (amino acids), and the length of the protein sequence is generally hundreds or thousands. The protein sequence data is the sequence of letters corresponding to all amino acids; for example, glycine-g, alanine-b, valine-j, etc., represent the sequence b1-j2-g3-b4.... Therefore, in this embodiment, the molecular sequence of the drug to be identified and the protein sequence of the target are obtained for subsequent analysis and judgment.
[0034] S20. The drug molecule sequence and the target protein sequence are respectively input into two BERT networks cascaded in the first model, and the drug molecule sequence and the target protein sequence are subjected to multimodal processing to obtain a first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence.
[0035] Considering that deep learning models randomly initialize data during processing, this initialization process shuffles ordered sequences, turning them into disordered data. Since drug molecule sequences and target protein sequences possess a certain degree of sequence characteristics, random shuffling might result in a different drug or target protein sequence. Clearly, this processing method is unsuitable for processing drug molecule sequences and target protein sequences. For example, for a drug molecule sequence {ABCD}, where A, B, C, and D are atoms in the drug's molecular formula, random shuffling might result in the sequence {ACDB}, which could correspond to a different drug. In this embodiment, to ensure the orderliness of the drug molecule sequence and target protein, a first network is constructed, consisting of two cascaded BERT networks. The drug molecule sequence and target protein sequence are input into the first model, and through multimodal processing, the outputs of the drug molecule sequence and the target protein sequence in the first model are fused to obtain a first confidence score. It should be noted that the two BERT networks in this embodiment have the same network architecture, but different parameters. The parameters of the first BERT network trained for drug molecule sequences and the second BERT network trained for target protein sequences are different in order to better extract drug features and target features.
[0036] Specifically, in one embodiment of the present invention, in step S20, the step of inputting the drug molecule sequence and the target protein sequence into two BERT networks in a first model cascade, respectively, and performing multimodal processing on the drug molecule sequence and the target protein sequence to obtain a first confidence score includes:
[0037] S21. Input the drug molecule sequence into the first Bert network of the first model to encode the drug molecule sequence and obtain the drug molecule matrix.
[0038] S22. Input the target protein sequence into the second Bert network of the first model to encode the target protein sequence and obtain the target protein matrix.
[0039] S23. Calculate the L1 norm of the drug molecule matrix and the target protein matrix, or the L2 norm of the drug molecule matrix and the target protein matrix, and use the L1 norm or the L2 norm as the first confidence score.
[0040] The drug molecule sequence is input into a first BERT network, where it is encoded using positional encoding and feature embedding to obtain a drug molecule matrix. Similarly, the target protein sequence is input into a second BERT network, where feature embedding and positional encoding are used to obtain a target protein matrix. The drug molecule matrix contains molecular feature information of the drug, and the target protein matrix contains feature information of the target protein. The L1 or L2 norm of these two matrices is calculated, allowing feature fusion to obtain the first confidence score. Specifically, the L1 norm is the sum of the absolute values of all elements in the vector, obtained by summing the absolute values of each element in the drug molecule matrix and the corresponding elements in the target protein matrix. The L2 norm is the square root of the sum of the squares of all elements in the vector, obtained by summing the squares of each element in the drug molecule matrix and the corresponding elements in the target protein matrix. It is understood that in this embodiment, either the L1 norm or the L2 norm can be selected for calculation, and the result can be used as the first confidence score. Preferably, to prevent overfitting and improve the generalization ability of the model, the L2 norm can be selected. It should be noted that when the L1 norm is used as the first confidence score, the total confidence score obtained by weighted summing with the second confidence score is recorded as the first total confidence score. When the L2 norm is used as the first confidence score, the total confidence score obtained by weighted summing with the second confidence score is recorded as the second total confidence score. Since the results of the L1 norm and L2 norm calculations are different, the first total confidence score and the second total confidence score are also different. Therefore, those skilled in the art will find that the confidence threshold set will differ depending on whether the L1 norm or the L2 norm is used as the first confidence score.
[0041] Further, in one embodiment of the present invention, in step S21, the step of inputting the drug molecule sequence into the first BERT network of the first model to encode the drug molecule sequence and obtain a drug molecule matrix includes:
[0042] S210. Input the drug molecule sequence into the first Bert network, perform feature embedding on the drug molecule sequence, and obtain a drug embedding matrix;
[0043] S211. Based on the positional relationship of each atom in the drug molecule sequence, position encoding is performed on the drug embedding matrix to obtain the drug encoding matrix;
[0044] S212. Perform feature fusion on the drug embedding matrix and the drug encoding matrix to output the drug molecule matrix.
[0045] After the drug molecule sequence is input into the first BERT network, it is segmented into words to obtain multiple atoms that make up the drug molecule sequence. For example, for the drug molecule sequence {ABC}, it can be segmented according to the chemical bonds connecting the atoms, resulting in three atoms A, B, and C. The TokenID corresponding to each atom is looked up in a pre-stored lookup table to complete feature embedding, thus obtaining a drug embedding matrix. The drug embedding matrix is composed of the TokenIDs corresponding to each atom in sequence. Then, the drug molecule sequence is positionally encoded according to static or dynamic positional encoding to obtain a drug encoding matrix. It is understood that static or dynamic positional encoding methods are existing knowledge and will not be detailed here. By summing the drug embedding matrix and the drug encoding matrix, feature fusion is achieved, resulting in the drug molecule matrix output by the first BERT network.
[0046] S30. The drug molecule sequence and the target protein sequence are respectively input into the dual-channel GNN network of the second model, and feature extraction and feature splicing are performed on the drug molecule sequence and the target protein sequence respectively to obtain the second confidence score; wherein, the second model is used to obtain the association relationship between the drug molecule sequence and the target protein sequence.
[0047] To further improve matching accuracy, this invention also uses a dual-channel graph neural network (GNN) as a second model. This model uses a knowledge graph approach to obtain the association between drug molecule sequences and target protein sequences, thereby generating a second confidence score. The dual-channel graph neural network includes a first GNN network for drug molecule sequences and a second GNN network for target protein sequences.
[0048] Specifically, in one embodiment of the present invention, in step S30, the step of inputting the drug molecule sequence and the target protein sequence into the dual-channel GNN network of the second model, respectively, and performing feature extraction and feature splicing on the drug molecule sequence and the target protein sequence to obtain a second confidence score includes:
[0049] S31. Input the drug molecule sequence into the first GNN network of the second model to obtain the drug feature vector;
[0050] S32. Input the target protein sequence into the second GNN network of the second model to obtain the target protein feature vector;
[0051] S33. The drug feature vector and the target protein feature vector are concatenated and input into the fully connected layer of the second model to obtain the matching degree value between the drug feature vector and the target protein feature vector, and the matching degree value is used as the second confidence score.
[0052] The drug molecule sequence is input into a first GNN network. Features learned by each layer of the first GNN are passed to the next layer, and so on. Through the stacking of multiple convolutional layers, the drug molecule information in the first GNN network is aggregated and updated, ultimately yielding a drug feature vector. Similarly, the target protein feature vector output by the second GNN network can be obtained. These two feature vectors are then concatenated. A fully connected layer is used to add each feature value in the concatenated vector according to a predetermined weight, resulting in a matching score. This matching score is then used as the second confidence score.
[0053] Further, in one embodiment of the present invention, in step S31, the step of inputting the drug molecule sequence into the first GNN network of the second model to obtain the drug feature vector includes:
[0054] S310. Based on the connection relationships of atoms and bonds in the drug molecule sequence, construct the drug molecule feature matrix and the drug molecule adjacency matrix;
[0055] S311. Input the drug molecule feature matrix and the drug molecule adjacency matrix into the first GNN network to obtain the drug feature vector.
[0056] Since drug molecule sequences are composed of atoms and bonds, meaning adjacent atoms are connected by bonds, a drug molecule feature matrix and a drug molecule adjacency matrix can be constructed based on the connections between atoms and bonds. Specifically, the drug molecule adjacency matrix refers to whether there are bonds connecting atoms, and it is composed of atoms. If there are 5 atoms, the drug molecule adjacency matrix is a 5*5 matrix. If there is a bond connecting atoms i and j, then atoms i and j are 1; otherwise, they are 0. Alternatively, it can be expressed as: if there is a bond connecting atoms i and j, then atoms i and j are 0; otherwise, they are 1. The drug molecule feature matrix represents the characteristics of each atom. For example, in a 5*5 drug molecule feature matrix, the data at position (1,1) can represent the normalized mass, size, etc., of the first atom. Then, the drug molecule feature matrix and the drug molecule adjacency matrix are input into the first GNN network, and drug feature vectors are obtained by extracting drug features.
[0057] Further, in one embodiment of the present invention, in step S311, the step of inputting the drug molecule feature matrix and the drug molecule adjacency matrix into the first GNN network to obtain the drug feature vector includes:
[0058] S3111. Perform feature perturbation processing on the drug molecule adjacency matrix to obtain the drug molecule perturbation matrix;
[0059] S3112. Input the drug molecule feature matrix and the drug molecule perturbation matrix into the first GNN network to extract the drug molecule feature vector and the drug molecule perturbation vector;
[0060] S3113. Summing the drug molecule feature vector and the drug molecule perturbation vector to obtain the drug feature vector.
[0061] Feature perturbation refers to swapping any two rows or columns of data in the adjacency matrix of a drug molecule. By perturbing, if two features have never co-occurred before, the perturbation causes them to co-occur, effectively improving the subsequent matching degree. Specifically, for example, the adjacency matrix of a drug molecule is a 4x4 matrix, with each row representing one atom. That is, the first row represents the bond connections between the first atom and other atoms. The row numbers of the drug molecule adjacency matrix are 1, 2, 3, and 4. After feature perturbation, the positions of the second and third atoms are swapped. Although the network topology remains unchanged, the features of the atoms at each position have changed, resulting in the drug molecule perturbation matrix with row numbers 1, 3, 2, and 4. Then, the drug molecule feature matrix and the drug molecule perturbation matrix are input together into a first GNN network. The first GNN network reads the information between each atom and bond, extracting the drug molecule feature vector and the drug molecule perturbation vector. The sum of the drug molecule feature vector and the drug molecule perturbation vector is calculated and used as the drug feature vector.
[0062] S40. The first confidence score and the second confidence score are weighted and summed to obtain the total confidence score. When the total confidence score is greater than or equal to a preset confidence threshold, it is determined that the drug matches the target.
[0063] The total confidence score L = αL1 + βL2, where L1 is the first confidence score, L2 is the second confidence score, α is the weight of the first confidence score, and β is the weight of the second confidence score. The values of α and β are both in the range [0,1]. Those skilled in the art can choose the appropriate values according to actual needs, and no specific limitation is made here. By comparing the total confidence score with a preset confidence threshold, if the total confidence score is greater than or equal to the confidence threshold, it indicates a high degree of matching between the drug and the target, and the drug can be used to treat the target, thus determining that the drug and target are matched. Conversely, if the total confidence score is less than the confidence threshold, it indicates a low degree of matching between the drug and the target, and the drug is not suitable for treating the target, thus determining that the drug and target are mismatched, and other targets need to be found to match the drug.
[0064] It should be noted that, for ease of explanation, this application only explains the processing of drug molecule sequences in the first and second models. Since the processing of target protein sequences in the first and second models is the same as that of drug molecule sequences, it will not be described in detail here.
[0065] Further, in one embodiment of the present invention, in step S40, after determining that the drug matches the target when the total confidence score is greater than or equal to a preset confidence threshold, the method further includes: verifying whether the drug meets a preset drugability standard, wherein verifying whether the drug meets the preset drugability standard includes:
[0066] S41. Based on the molecular sequence of the drug, search for the clinical trial data of the drug, and determine whether the clinical trial data is within the preset drug-likeness threshold range;
[0067] S42. If the clinical trial data are within the drugability threshold range, then the drug meets the drugability standard;
[0068] S43. If the clinical trial data is not within the range of the drugability threshold, then the drug does not meet the drugability standard.
[0069] The above matching process verifies the drug's compatibility with its target, but this is only a preliminary assessment of the drug's potential for production. For actual use, to ensure patient health, after verifying the drug's compatibility with the target, its drug-likeness must be further considered. Drug-likeness refers to the early evaluation of preliminary pharmacodynamic studies, pharmacokinetic characteristics, and safety, demonstrating the potential to be developed into a drug. Early evaluation of the drug-likeness of selected candidate drugs to determine the best candidate can improve the success rate of new drug development and reduce its risks and costs. Specifically, this can be achieved by querying clinical trial data and determining whether the data falls within a pre-defined drug-likeness threshold range. If it does, the drug meets the drug-likeness standard and has good drug-likeness, suitable for further development. If not, the drug does not meet the drug-likeness standard and its properties still have some shortcomings, making it unsuitable for market. Common reasons for a drug not meeting the drug-likeness standard include poor chemical structure, non-specific activity, unclear mechanism of action, unreasonable pharmacokinetic properties, and significant toxic side effects. Because the market launch of such drugs may cause adverse effects such as damage to the body's functions or failure to effectively treat the patient's disease, further drugability screening after determining drug-target compatibility is a crucial process. Only when a drug meets drugability standards can subsequent drug development and clinical application proceed. This drugability screening significantly improves the efficiency of new drug discovery and creation, minimizing or avoiding the frustration and regret of a candidate drug being halted in clinical trials due to unforeseen circumstances. It is understood that this application, in addition to drugability screening, also includes drug activity screening, drug toxicity screening, and drug physicochemical property screening, thereby obtaining more practical drugs. It is understood that the threshold ranges required for the above-mentioned drug activity screening, drug toxicity screening, and drugability screening are defined by information such as the disease and target of the drug, and will not be detailed here. It should be noted that the above-mentioned deep learning model can be applied to various fields, such as medical diagnosis and application testing.
[0070] It should be understood that the sequence number of each step in the above embodiments does not imply the order of steps. The order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0071] In one embodiment, a drug-target matching determination system is provided. The drug-target matching determination system 10 corresponds one-to-one with the drug-target matching determination methods described in the above embodiments. For example... Figure 6 As shown, the drug-target matching system 10 includes a data acquisition module 11, a first confidence score acquisition module 12, a second confidence score acquisition module 13, and a matching determination module 14. Detailed descriptions of each functional module are as follows:
[0072] Data acquisition module 11 is used to acquire drug molecule sequences and target protein sequences;
[0073] The first confidence score acquisition module 12 is used to input the drug molecule sequence and the target protein sequence into two BERT networks cascaded in the first model, respectively, and perform multimodal processing on the drug molecule sequence and the target protein sequence to obtain the first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence.
[0074] The second confidence score acquisition module 13 is used to input the drug molecule sequence and the target protein sequence into the dual-channel GNN network of the second model, respectively, and to perform feature extraction and feature splicing on the drug molecule sequence and the target protein sequence to obtain the second confidence score; wherein, the second model is used to obtain the correlation between the drug molecule sequence and the target protein sequence.
[0075] The matching determination module 14 is used to perform a weighted summation of the first confidence score and the second confidence score to obtain a total confidence score, and to determine that the drug matches the target when the total confidence score is greater than or equal to a preset confidence threshold.
[0076] In one embodiment, the first confidence score acquisition module 12 is specifically used for:
[0077] The drug molecule sequence is input into the first BERT network of the first model to encode the drug molecule sequence and obtain the drug molecule matrix;
[0078] The target protein sequence is input into the second Bert network of the first model to encode the target protein sequence, thereby obtaining the target protein matrix.
[0079] Calculate the L1 norm of the drug molecule matrix and the target protein matrix, or the L2 norm of the drug molecule matrix and the target protein matrix, and use the L1 norm or the L2 norm as the first confidence score.
[0080] In one embodiment, the first confidence score acquisition module 12 is further configured to:
[0081] The drug molecule sequence is input into the first Bert network, and feature embedding is performed on the drug molecule sequence to obtain a drug embedding matrix;
[0082] Based on the positional relationship of each atom in the drug molecule sequence, the drug embedding matrix is positionally encoded to obtain the drug encoding matrix;
[0083] The drug embedding matrix and the drug encoding matrix are fused to output a drug molecule matrix.
[0084] In one embodiment, the second confidence score acquisition module 13 is specifically used for:
[0085] The drug molecule sequence is input into the first GNN network of the second model to obtain the drug feature vector;
[0086] The target protein sequence is input into the second GNN network of the second model to obtain the target protein feature vector;
[0087] The drug feature vector and the target protein feature vector are concatenated and input into the fully connected layer of the second model to obtain the matching degree value between the drug feature vector and the target protein feature vector, and the matching degree value is used as the second confidence score.
[0088] In one embodiment, the second confidence score acquisition module 13 is further configured to:
[0089] Based on the connection relationships of atoms and bonds in the drug molecule sequence, a drug molecule feature matrix and a drug molecule adjacency matrix are constructed.
[0090] The drug molecule feature matrix and the drug molecule adjacency matrix are input into the first GNN network to obtain the drug feature vector.
[0091] In one embodiment, the second confidence score acquisition module 13 is further configured to:
[0092] The adjacency matrix of the drug molecules is subjected to feature perturbation processing to obtain the perturbation matrix of the drug molecules;
[0093] The drug molecule feature matrix and the drug molecule perturbation matrix are input into the first GNN network to extract the drug molecule feature vector and the drug molecule perturbation vector;
[0094] The drug molecule feature vector is obtained by summing the drug molecule feature vector and the drug molecule perturbation vector.
[0095] In one embodiment, the matching determination module 14 is further configured to:
[0096] Based on the molecular sequence of the drug, search for the clinical trial data of the drug and determine whether the clinical trial data is within the preset drug-likeness threshold range;
[0097] If the clinical trial data are within the drugability threshold range, then the drug meets the drugability standard;
[0098] If the clinical trial data is not within the drugability threshold range, then the drug meets the drugability standard.
[0099] This invention provides a drug-target matching determination device, addressing the problems of current laboratory drug screening methods, which require extensive biological equipment, are extremely slow, and have a limited number of compounds screened. The device inputs the obtained drug molecule sequence and target protein sequence into two cascaded BERT networks in a first model. Multimodal processing of the drug molecule sequence and target protein sequence yields a first confidence score. Simultaneously, the drug molecule sequence and target protein sequence are input into a dual-channel GNN network in a second model for feature extraction and feature concatenation, yielding a second confidence score. The first and second confidence scores are then weighted and summed to obtain a total confidence score. When the total confidence score is greater than or equal to a preset confidence threshold, the drug is determined to match the target. This achieves rapid and accurate determination of the targeting relationship between a new drug and the target protein, effectively improving the efficiency of drug screening. This allows pharmaceutical companies and drug discovery researchers without coding skills to screen drugs simply by clicking a mouse. They only need to synthesize and test a small number of drugs and targets, discarding many compounds that are theoretically impossible to become drugs, thus saving significant manpower and financial costs. By using BERT networks, positional encoding and feature embedding are performed on drug molecule sequences and target protein sequences respectively, improving the orderliness of these sequences. This avoids the random scrambling of sequences during initialization when inputting into the BERT model, which can lead to errors in drug molecule or target protein sequences and affect subsequent judgment results. Using graph neural networks, although the number of layers is small, it can fully utilize the bond connections between atoms in the drug molecule or target protein sequences, greatly improving the accuracy of judgments and shortening the judgment time. The effective cascading of graph neural networks and BERT networks, due to the small model size, significantly increases the scale and speed of drug-target matching judgments while reducing computational resources. By effectively connecting data, computing power, algorithms (machine learning, deep learning, natural language processing, knowledge graphs, etc.) and platforms, the efficiency of new drug development is greatly improved, enabling processes such as drug target discovery and drug performance screening, and providing the possibility for cost reduction and efficiency improvement.
[0100] Specific limitations regarding the drug-target matching determination system can be found in the limitations of the drug-target matching determination method described above, and will not be repeated here. Each module in the aforementioned drug-target matching determination system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware format or independent of it, or stored in the memory of a computer device in software format, so that the processor can call the corresponding operations of each module.
[0101] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communication with external suppliers via a network connection. When the computer program is processed, it implements the functions or steps of a drug-target matching determination method on the server side.
[0102] In one embodiment, a computer device is provided, which may be from a supplier, and its internal structure diagram may be as follows: Figure 9 As shown. The computer device includes a processor, memory, network interface, display screen, and input system connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is processed, it implements the functions or steps of a drug and target matching determination method on the supplier's side.
[0103] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor computer program performs the following steps:
[0104] Obtain drug molecule sequences and target protein sequences;
[0105] The drug molecule sequence and the target protein sequence are respectively input into two BERT networks cascaded in the first model, and multimodal processing is performed on the drug molecule sequence and the target protein sequence to obtain a first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence.
[0106] The drug molecule sequence and the target protein sequence are respectively input into the dual-channel GNN network of the second model, and feature extraction and feature splicing are performed on the drug molecule sequence and the target protein sequence respectively to obtain a second confidence score; wherein, the second model is used to obtain the association relationship between the drug molecule sequence and the target protein sequence;
[0107] The first confidence score and the second confidence score are weighted and summed to obtain the total confidence score. When the total confidence score is greater than or equal to a preset confidence threshold, the drug is determined to be a match for the target.
[0108] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when processed by a processor:
[0109] Obtain drug molecule sequences and target protein sequences;
[0110] The drug molecule sequence and the target protein sequence are respectively input into two BERT networks cascaded in the first model, and multimodal processing is performed on the drug molecule sequence and the target protein sequence to obtain a first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence.
[0111] The drug molecule sequence and the target protein sequence are respectively input into the dual-channel GNN network of the second model, and feature extraction and feature splicing are performed on the drug molecule sequence and the target protein sequence respectively to obtain a second confidence score; wherein, the second model is used to obtain the association relationship between the drug molecule sequence and the target protein sequence;
[0112] The first confidence score and the second confidence score are weighted and summed to obtain the total confidence score. When the total confidence score is greater than or equal to a preset confidence threshold, the drug is determined to be a match for the target.
[0113] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and the supplier side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.
[0114] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is present, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various formats, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.
[0115] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.
[0116] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A method for determining drug-target matching, characterized in that, include: Obtain drug molecule sequences and target protein sequences; The drug molecule sequence and the target protein sequence are respectively input into two BERT networks cascaded in the first model, and multimodal processing is performed on the drug molecule sequence and the target protein sequence to obtain a first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence. The drug molecule sequence and the target protein sequence are respectively input into the dual-channel GNN network of the second model, and feature extraction and feature splicing are performed on the drug molecule sequence and the target protein sequence respectively to obtain a second confidence score; wherein, the second model is used to obtain the association relationship between the drug molecule sequence and the target protein sequence; The first confidence score and the second confidence score are weighted and summed to obtain the total confidence score. When the total confidence score is greater than or equal to a preset confidence threshold, the drug is determined to match the target. The process involves inputting the drug molecule sequence and the target protein sequence into two BERT networks in a first model cascade, respectively, and performing multimodal processing on the drug molecule sequence and the target protein sequence to obtain a first confidence score, including: The drug molecule sequence is input into the first BERT network of the first model to encode the drug molecule sequence and obtain the drug molecule matrix; The target protein sequence is input into the second Bert network of the first model to encode the target protein sequence and obtain the target protein matrix. Calculate the L1 norm of the drug molecule matrix and the target protein matrix, or the L2 norm of the drug molecule matrix and the target protein matrix, and use the L1 norm or the L2 norm as the first confidence score. The step of inputting the drug molecule sequence into the first BERT network of the first model to encode the drug molecule sequence and obtain the drug molecule matrix includes: The drug molecule sequence is input into the first Bert network, and feature embedding is performed on the drug molecule sequence to obtain a drug embedding matrix; Based on the positional relationship of each atom in the drug molecule sequence, the drug embedding matrix is positionally encoded to obtain the drug encoding matrix; The drug embedding matrix and the drug encoding matrix are fused to output a drug molecule matrix.
2. The method for determining drug-target matching according to claim 1, characterized in that, The process involves inputting the drug molecule sequence and the target protein sequence into a dual-channel GNN network of the second model, respectively, to extract and concatenate features from the drug molecule sequence and the target protein sequence to obtain a second confidence score, including: The drug molecule sequence is input into the first GNN network of the second model to obtain the drug feature vector; The target protein sequence is input into the second GNN network of the second model to obtain the target protein feature vector; The drug feature vector and the target protein feature vector are concatenated and input into the fully connected layer of the second model to obtain the matching degree value between the drug feature vector and the target protein feature vector, and the matching degree value is used as the second confidence score.
3. The method for determining drug-target matching according to claim 2, characterized in that, The step of inputting the drug molecule sequence into the first GNN network of the second model to obtain the drug feature vector includes: Based on the connection relationships of atoms and bonds in the drug molecule sequence, a drug molecule feature matrix and a drug molecule adjacency matrix are constructed. The drug molecule feature matrix and the drug molecule adjacency matrix are input into the first GNN network to obtain the drug feature vector.
4. The method for determining drug-target matching according to claim 3, characterized in that, The step of inputting the drug molecule feature matrix and the drug molecule adjacency matrix into the first GNN network to obtain the drug feature vector includes: The adjacency matrix of the drug molecules is subjected to feature perturbation processing to obtain the perturbation matrix of the drug molecules; The drug molecule feature matrix and the drug molecule perturbation matrix are input into the first GNN network to extract the drug molecule feature vector and the drug molecule perturbation vector; The drug molecule feature vector is obtained by summing the drug molecule feature vector and the drug molecule perturbation vector.
5. The method for determining drug-target matching according to claim 1, characterized in that, After determining that the drug matches the target when the total confidence score is greater than or equal to a preset confidence threshold, the method further includes: verifying whether the drug meets preset drugability standards. Verifying whether the drug meets preset drugability standards includes: Based on the molecular sequence of the drug, search for the clinical trial data of the drug and determine whether the clinical trial data is within the preset drug-likeness threshold range; If the clinical trial data are within the drugability threshold range, then the drug meets the drugability standard; If the clinical trial data is not within the drugability threshold range, then the drug does not meet the drugability standard.
6. A drug-target matching determination system, characterized in that, The system includes: The data acquisition module is used to acquire drug molecule sequences and target protein sequences; The first confidence score acquisition module is used to input the drug molecule sequence and the target protein sequence into two BERT networks cascaded in the first model, respectively, and perform multimodal processing on the drug molecule sequence and the target protein sequence to obtain the first confidence score; wherein, the first model is used to obtain the positional encoding relationship between the drug molecule sequence and the target protein sequence. The second confidence score acquisition module is used to input the drug molecule sequence and the target protein sequence into the dual-channel GNN network of the second model, respectively, and to perform feature extraction and feature splicing on the drug molecule sequence and the target protein sequence to obtain the second confidence score; wherein, the second model is used to obtain the correlation between the drug molecule sequence and the target protein sequence. The matching determination module is used to perform a weighted summation of the first confidence score and the second confidence score to obtain a total confidence score, and to determine that the drug matches the target when the total confidence score is greater than or equal to a preset confidence threshold. The process involves inputting the drug molecule sequence and the target protein sequence into two BERT networks in a first model cascade, respectively, and performing multimodal processing on the drug molecule sequence and the target protein sequence to obtain a first confidence score, including: The drug molecule sequence is input into the first BERT network of the first model to encode the drug molecule sequence and obtain the drug molecule matrix; The target protein sequence is input into the second Bert network of the first model to encode the target protein sequence and obtain the target protein matrix. Calculate the L1 norm of the drug molecule matrix and the target protein matrix, or the L2 norm of the drug molecule matrix and the target protein matrix, and use the L1 norm or the L2 norm as the first confidence score. The step of inputting the drug molecule sequence into the first BERT network of the first model to encode the drug molecule sequence and obtain the drug molecule matrix includes: The drug molecule sequence is input into the first Bert network, and feature embedding is performed on the drug molecule sequence to obtain a drug embedding matrix; Based on the positional relationship of each atom in the drug molecule sequence, the drug embedding matrix is positionally encoded to obtain the drug encoding matrix; The drug embedding matrix and the drug encoding matrix are fused to output a drug molecule matrix.
7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.
8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.