Method and system for re-scoring after docking based on ligand similarity and DTI
By combining ligand similarity with DTI, the problem of limited prediction effect of virtual screening in the prior art is solved, and the enrichment and hit rate of virtual screening is significantly improved, especially in kinase and GPCR targets. Excellent performance on the market, suitable for different target types and data volumes.
Patent Information
- Application Number
- CN202411524639.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-30
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, a single strategy based on ligand similarity or drug-target interaction (DTI) has limited predictive effects in virtual screening, especially when the data volume is small or the target structure is complex, it is difficult to effectively improve. The enrichment ability and prediction accuracy of docking virtual screening.
A docking and rescoring method that binds ligand similarity to DTI is adopted. By obtaining the template protein of the target target protein, selecting the co-crystal ligand for similarity score, and molecularly docking the target ligand with the target protein. The figure is used. The attention network performs DTI classification reasoning, and ultimately adds the similarity score and classification result scores, and sorts the target ligands to improve the enrichment ability of virtual screening.
It significantly improves the enrichment and hit rate of virtual screening, especially in kinase and GPCR targets, and can significantly improve the docking ranking ability when the data volume is sufficient, and overcomes the disadvantages of the ligand similarity search method. , suitable for different target types and data volumes.
Smart Images

Figure CN119046700A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer-aided drug design, and in particular to a post-docking re-scoring method and system based on ligand similarity and DTI. Background Art
[0002] Molecular docking technology is widely used in virtual screening for drug discovery, but its success rate is still limited by the robustness and broad applicability of the docking scoring function. In most docking screens, only the top few dozen molecules are tested for activity. However, as the number of molecules in the virtual library grows exponentially, the difficulty of enriching active molecules increases accordingly. Therefore, as the scale of virtual screening libraries increases, new virtual screening methods need to have higher enrichment capabilities to obtain more active molecules without increasing the number of test molecules.
[0003] Currently, a variety of post-docking processing methods have been proposed to further enhance the enrichment capabilities of docking-based virtual screening. For example, consensus scoring methods improve hit rates by combining the results of different docking programs. Researchers often rely on empirical methods to examine the results of different docking programs to confirm whether the docked poses cover consistent protein-ligand interactions, including hydrogen bonds, halogen bonds, π-π stacking, etc. This judgment relies on the researcher's professional experience and is difficult to standardize.
[0004] To improve the accuracy of the docking pose, more precise energy- and free-energy-based methods can be used to re-score the docking pose, such as solvated interaction energy (SIE) [Journal of Chemical Information and Modeling, 2007, 47(1): 122-133.], molecular mechanics / Poisson-Boltzmann surface area (MM / PBSA) [Physical Chemistry Chemical Physics, 2014, 16(31): 16719-16729.], molecular mechanics / generalized Born surface area (MM / GBSA) [Journal of Chemical Information and Modeling, 2021, 61(6): 2844-2856.], and free energy perturbation (FEP) [Proceedings of the National Academy of Sciences, 2020, 117(44): However, these methods require a lot of computing resources and are relatively time-consuming, thus limiting their application in ultra-large-scale virtual screening scenarios.
[0005] Another common approach is a post-docking rescoring strategy based on 2D or 3D ligand similarity alignment. This approach compares the conformations predicted by the docking program with the crystal-resolved conformations of the template active ligand and re-ranks the predicted ligand conformations based on similarity (such as the Tanimoto coefficient). Due to the presence of many rotatable bonds in the compounds, 3D-based methods can be used to search for maximum shape overlap, which generally improves the true positive rate. However, a common limitation of ligand similarity alignment-based strategies is how to select the reference molecules for similarity calculation, especially when there is little information on active compounds. Although advances in structural biology have made a large number of protein-ligand complex structures available, providing rich information on interaction patterns (such as the PDBind and Biolip databases), such methods still rely on high-quality reference data.
[0006] Whether based on 2D or 3D molecular similarity searches, pure ligand-based virtual screening (LBVS) strategies have poor generalization capabilities and can result in a high number of false-positive results. This is because drugs have specific binding modes with their target proteins, and molecules with the same binding mode may have similar biological functions. Therefore, drug-target interaction (DTI) has become a potential method to improve the enrichment of docking scores.
[0007] In recent years, a variety of methods have emerged for DTI prediction tasks. For example, the WGNN-DTA model developed by Mingjian Jiang et al. constructs a molecular graph using amino acid sequences and SMILES and uses graph neural networks to extract features to predict binding affinity. The DTI prediction model developed by Jaechang Lim et al. extracts graphical features of intermolecular interactions from the 3D structural information of protein-ligand binding poses and uses a distance-aware graph attention algorithm to distinguish various types of intermolecular interactions. GanDTI developed by Shuyu Wang et al. uses composite graphs and protein sequence data for interaction classification and affinity prediction tasks.
[0008] Although these DTI-based prediction methods can be closer to actual drug activity and have strong interpretability, their results are highly dependent on accurate target structure information. In addition, since these methods are usually performed under static conditions, kinetics and solvent effects are ignored, these limitations also affect their effectiveness in practical applications. Summary of the Invention
[0009] The present invention aims to address the limited predictive effectiveness of either ligand similarity or DTI alone. To this end, the present invention provides a post-docking rescoring method and system based on ligand similarity and DTI. This method combines the advantages of both ligand similarity and drug-target interaction to improve the enrichment capability and predictive accuracy of docking virtual screening.
[0010] The present invention provides a post-docking re-scoring method based on ligand similarity and DTI, and the technical solution adopted is as follows: Obtain a target protein, retrieve a template pocket from a database that has similar residues and topological structures to the pocket of the target protein through pocket similarity comparison, and obtain a template protein corresponding to the template pocket; Select the N co-crystal ligands of the template protein with the highest pocket similarity comparison scores, and use their co-crystal conformations to score the target ligand to obtain a similarity score of the target ligand; the target ligand is a potential active ligand of the target protein; The target protein is molecularly docked with the target ligand, and the molecular graph of the protein-ligand complex is obtained through the docking conformation. The molecular graph is input into the graph attention network to calculate the classification result of whether the binding posture of the protein-ligand complex is active, and the classification result score of the target ligand is obtained; The similarity score and the classification result score are added together to obtain the final score of the target ligand. The target ligands are sorted from high to low according to the final score to obtain the re-scoring result.
[0011] Furthermore, the process of obtaining the template protein corresponding to the template pocket is as follows: extracting the protein corresponding to the template pocket and its co-crystal ligand from the database by the pocket number, and performing dehydration, hydrogenation and charge treatment on the protein to obtain the template protein.
[0012] Furthermore, the calculation formula of the target ligand similarity score LS_score is: Among them, Max[ ] means to find the maximum value, N probe represents the number of heavy atoms of the target ligand, represents the number of atomic pairs aligned between the cocrystal conformation of the target ligand and the cocrystal ligand, represents the distance between the i-th pair of aligned atoms, represents the difference in relative masses between the ith pair of aligned atoms, express The normalized scale of express The normalized scale of is a constant whose value is 10. represents the weighted Jaccard score, which measures the similarity of the chemical bonds associated with the ith pair of aligned atoms, represents the weight factor of the distance, A weighting factor representing relative quality, A weighting factor representing the similarity of chemical bonds.
[0013] Further, is 0.45, is 0.1, is 0.45.
[0014] Furthermore, N is 50.
[0015] Furthermore, the LS_score ranges from 0 to 1, with a score of 1 indicating a perfect structural match between the two ligands; The classification result is scored as 1 or 0, where 1 indicates that the binding pose of the protein-ligand complex is active and 0 indicates that the binding pose of the protein-ligand complex is inactive.
[0016] Furthermore, the conformation with the highest final score is retained among multiple conformations of the target ligand, and its final score is used as the final score of the target ligand.
[0017] The present invention also provides a post-docking re-scoring system based on ligand similarity and DTI, which adopts the following technical solution: comprising: a data preprocessing module, a ligand similarity search module, a DTI reasoning module and a sorting module, wherein the data preprocessing module is connected to the ligand similarity search module and the DTI reasoning module respectively, and the sorting module is connected to the ligand similarity search module and the DTI reasoning module respectively; The data preprocessing module is used to obtain a target protein, retrieve a template pocket having residues and a topological structure similar to those of the pocket of the target protein from a database through pocket similarity comparison, and obtain a template protein corresponding to the template pocket; The ligand similarity search module is used to select the co-crystal ligands of the N template proteins with the highest pocket similarity comparison scores, and use their co-crystal conformations to score the target ligands to obtain the similarity scores of the target ligands; the target ligands are potential active ligands of the target protein; The DTI inference module is used to molecularly dock the target protein with the target ligand, obtain a molecular graph of the protein-ligand complex through the docking conformation, input the molecular graph into the graph attention network, calculate the classification result of whether the binding posture of the protein-ligand complex is active, and obtain the classification result score of the target ligand; The ranking module is used to add the similarity score and the classification result score to obtain the final score of the target ligand, and to sort the target ligands from high to low according to the final score to obtain a re-scoring result.
[0018] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: The present invention not only takes ligand similarity into account, but also introduces drug-target interaction prediction, thereby improving the enrichment and hit rate of virtual screening. The present invention retains all docking conformations and performs DTI classification reasoning on these conformations, overcoming the drawback that the ligand similarity-based search method can only identify conformations of known active compounds. Experiments have shown that the present invention performs particularly well on kinase and GPCR targets, especially when the amount of data is sufficient, and can significantly improve the docking ranking ability. In the case of less data or complex target structure, its performance is slightly affected, but it still shows good optimization capabilities. Overall, the present invention has strong adaptability to different target types and data amounts, and has great application potential in specific target categories.
[0019] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0021] Figure 1 It is a flow chart of the method provided by the present invention.
[0022] Figure 2 The present invention provides experimental result diagrams of the hit rates using only ligand similarity and using both ligand similarity and DTI.
[0023] Figure 3 This is a comparison chart of the ranking capabilities of the method provided by the present invention and Smina.
[0024] Figure 4 It is a structural block diagram of the system provided by the present invention.
[0025] Reference numerals: 1. Data preprocessing module; 2. Ligand similarity search module; 3. DTI inference module; 4. Sorting module. DETAILED DESCRIPTION
[0026] To make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Obviously, the embodiments described are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.
[0027] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.
[0028] The following combination Figures 1 to 4 The present invention is further described in detail, and a post-docking re-scoring method and system based on ligand similarity and DTI is described: In this embodiment, Figure 1 As shown, a post-docking re-scoring method based on ligand similarity and DTI is provided, comprising the following steps: Step 1: Obtain the target protein, and retrieve a template pocket from the BioLip database that is similar to the residues and topological structure of the pocket of the target protein through pocket similarity comparison. Then, extract the protein corresponding to the template pocket and its co-crystal ligand from the BioLip database by pocket number, and perform dehydration, hydrogenation, and charge addition treatment on the protein to obtain the template protein.
[0029] Step 2: Select the co-crystal ligands of the N template proteins with the highest pocket similarity comparison scores, and use their co-crystal conformations to score the target ligand to obtain a similarity score of the target ligand; the target ligand is a potential active ligand of the target protein.
[0030] The present invention uses ligand similarity to compare the atomic structure of ligand molecules. The calculation parameters include interatomic distances, atomic masses, and chemical bond connections. The present invention uses a rigid ligand structure comparison method to calculate the similarity of 3D structure and chemical properties between the cocrystal conformation of the target ligand and the cocrystal ligand. The similarity score LS_score of the target ligand is calculated as follows: Among them, Max[ ] means to find the maximum value, N probe represents the number of heavy atoms of the target ligand, represents the number of atomic pairs aligned between the cocrystal conformation of the target ligand and the cocrystal ligand, represents the distance between the i-th pair of aligned atoms, represents the difference in relative masses between the ith pair of aligned atoms, express The normalized scale of express The normalized scale of is a constant whose value is 10. represents the weighted Jaccard score, which measures the similarity of the chemical bonds associated with the ith pair of aligned atoms, represents the weight factor of the distance, A weighting factor representing relative quality, Represents the weight factor of chemical bond similarity. In this embodiment, is 0.45, is 0.1, is 0.45, and N is 50. The LS_score ranges from 0 to 1, with a score of 1 indicating a perfect structural match between the two ligands.
[0031] Step 3: Molecular docking is performed between the target protein and the target ligand. The molecular graph of the protein-ligand complex is obtained through the docking conformation. The molecular graph includes the node feature set x, the adjacency matrix A0 and the adjacency matrix A1. Each node in the node feature set x represents an atom. The adjacency matrix A0 represents the pure covalent interaction between atoms, and the adjacency matrix A1 represents the covalent interaction between atoms and the non-covalent atomic interaction.
[0032] The molecular graph is input into the graph attention network (GAT), and the molecular graph is enhanced by the gate of the graph attention network to calculate two new node feature sets x0 and x1, where x0 represents the node feature set of pure covalent interactions, and x1 represents the node feature set of covalent interactions and non-covalent atomic interactions; x0 is subtracted from x1 to obtain the output node feature. By subtracting x0 from x1, the graph attention network understands the difference between the structure of the bound pose and the separated structure. After updating the feature vector through several gate-enhanced graph attention layers, the feature vector of the ligand atom is summarized into an output node feature vector representing the protein-ligand complex graph. Applying the Multilayer Perceptron of Graph Attention Network to , to classify whether the binding posture of the protein-ligand complex is active. ReLU activation functions are used between layers, and a sigmoid function is used after the final layer to output classification probabilities. Finally, the classification result of whether the binding posture of the protein-ligand complex is active is calculated to obtain the classification result score of the target ligand. The classification result score is 1 or 0, where 1 indicates an active binding posture of the protein-ligand complex and 0 indicates an inactive binding posture of the protein-ligand complex.
[0033] The graph attention network used in step 3 is pre-trained. By comparing pocket similarities, a template pocket with residues and topological structures similar to those of the pocket of the target protein is retrieved from the BioLip database. The protein corresponding to the template pocket and its co-crystal ligand are then extracted from the BioLip database by pocket number. The protein is dehydrated, hydrogenated, and charged to obtain the template protein. The template protein is re-docked with the co-crystal ligand of the target protein to obtain a new docking conformation. The new docking conformation is evaluated. The present invention calculates the RMSD value of the co-crystal conformation of the template protein and the target protein, and labels the new docking conformation. The new docking conformation with an RMSD less than 2Å is labeled "1", and the new docking conformation with an RMSD greater than 4Å is labeled "0". The training set and test set for DTI inference are generated. New docking conformations between 2Å and 4Å are discarded to avoid difficulty in distinguishing between negative and positive data. The graph attention network is trained using the training set and test set, and the trained graph attention network is used for DTI classification prediction of molecular graphs of docking conformations of protein-target ligands. In order to improve the computing speed and reduce the consumption of computing resources, only protein atoms with a minimum distance to ligand atoms less than 7Å are selected for graphical representation during training.
[0034] Step 4: Add the similarity score and the classification result score to obtain the final score of the target ligand, which ranges from (1, 2). Since each target ligand corresponds to multiple conformations, the conformation with the highest final score is retained among the multiple conformations of the target ligand, and its final score is used as the final score of the target ligand. Sort the target ligands from high to low according to the final score to obtain the re-scoring result.
[0035] In order to verify the virtual screening performance of the present invention, an ablation experiment was conducted to compare the hit rate (Hit_Rate1%) of using only ligand similarity and using both ligand similarity and DTI. Figure 2 As shown in the figure, for 11 of the 15 targets, the pipeline performance remained unchanged or improved. In particular, for the VDR target, the HR1% was 0 when using only ligand similarity, but increased to 1.98 after the introduction of DTI. This indicates that DTI has a significant positive impact on the overall performance of the pipeline.
[0036] To further evaluate the effectiveness of the present invention (DockRefine), its ranking ability was compared with that of the molecular docking tool Smina. Figure 3As shown, DockRefine performed exceptionally well on 11 of the 15 targets, demonstrating that DockRefine significantly improves the ranking capabilities of docking algorithms in most cases, thereby enhancing the accuracy and efficiency of screening. Kinase targets, such as MTORC1, IDH1, PKM2, and TP53, performed exceptionally well in DockRefine. These targets typically possess well-defined active sites and good druggability, making them suitable for docking screening. DockRefine was particularly effective in leveraging large datasets, such as TP53, which had 7,160 positive and 17,901 negative data points in its training set, significantly improving its ranking capabilities. This demonstrates that DockRefine is highly effective in optimizing the screening pipeline for kinase targets, especially when sufficient target data is available. GPCR targets, such as ADRB2 and OPRK1, also performed well in the DockRefine pipeline. GPCR targets have highly conserved domains, making them generally amenable to docking screening. Among these targets, DockRefine leveraged its rich training and test data sets, such as ADRB2, which had 1,335 positive data points and 3,701 negative data points in its training set, effectively improving docking ranking performance. This demonstrates that DockRefine also possesses strong optimization capabilities when working with GPCR targets.
[0037] Data size plays a key role in DockRefine's performance. For targets with larger datasets, such as ESR1, ago, and TP53, DockRefine fully leverages its strengths, leveraging the rich data to optimize docking rankings. However, for targets with smaller datasets, such as MTORC1, where the training set had 166 positive data points and 414 negative data points, DockRefine still saw some improvement, but its performance was not as significant as for targets with more abundant data.
[0038] DockRefine also demonstrated robust docking performance for targets with balanced data, such as PPARG, where the training set had 1,162 positive data points and 2,903 negative data points. This suggests that while data size has a certain impact on DockRefine's performance, it still demonstrates good optimization capabilities when handling targets with balanced data.
[0039] DockRefine underperformed basic docking methods for targets such as KAT2A, ESR1_ant, FEN1, and VDR. This may be due to the following factors: KAT2A and FEN1 are enzyme targets with complex active site structures, which may make it difficult for DockRefine to accurately capture key interactions. ESR1_ant, as an estrogen receptor antagonist, may require high ligand specificity, and DockRefine's recognition ability is limited for such targets. VDR, as a nuclear receptor, has low druggability, which may also affect DockRefine's performance.
[0040] DockRefine performs particularly well on kinase and GPCR targets, significantly improving docking ranking capabilities when sufficient data is available. While performance is slightly affected when data is limited or target structures are complex, it still demonstrates good optimization capabilities. Overall, DockRefine is highly adaptable to different target types and data volumes, and has great potential for application in specific target categories.
[0041] To further evaluate the effectiveness of this method, we selected the common docking software Glide-SP (Glide Docking in Precision Mode) and Vina (Vina Molecular Docking Software) as benchmark comparison methods and compared their virtual screening capabilities with DockRefine. The results, shown in Table 1, demonstrate that among the 15 targets in the benchmark LIT-PCBA dataset, DockRefine exhibited high enrichment factors (EF1%) of 11.33, 2.00, 15.25, 2.00, and 2.55 for targets such as ADRB2, ALDH1, ESR1_ago, PKM2, and TP53, respectively. Compared to docking-based methods such as Vina and Glide-SP, DockRefine significantly outperformed these methods in enrichment for ADRB2 and ESR1_ago, demonstrating the advantages of this pipeline in identifying highly active compounds.
[0042] Compared to other empirical virtual screening methods, such as AA-score (amino acid scoring function) and X-score (binding affinity scoring function), DockRefine not only considers ligand similarity but also incorporates drug-target interaction prediction, resulting in superior performance across multiple targets, as shown in Table 1. For example, compared to the EF1% values of X-score and AA-score, the DockRefine pipeline achieved average EF1% values that were 1.83-fold and 1.96-fold higher, respectively, further demonstrating the superior performance of this method in virtual screening.
[0043] Compared to the EVIS (Enhanced Virtual Screening System) method, which primarily relies on pocket and ligand similarity, DockRefine outperformed EVIS on multiple targets, particularly ESR1_ago, where both methods performed equally well. However, on four other targets, DockRefine's EF1% values were almost twice that of EVIS. This is primarily due to the fact that DockRefine retains all docked conformations and performs DTI classification inference on these conformations, overcoming the drawback of ligand similarity-based search methods that can only identify conformations similar to known active compounds.
[0044] Table 1 Comparison of EF1% indicators of various methods on 15 targets of LIT-PCBA The above experiments show that DockRefine not only significantly improves the enrichment and hit rate of virtual screening, but also overcomes the limitations of traditional virtual screening methods based on ligand similarity, providing an efficient and accurate tool for drug discovery.
[0045] In this embodiment, Figure 4 As shown, a post-docking re-scoring system based on ligand similarity and DTI is also provided, and the technical solution adopted is as follows: comprising: a data preprocessing module 1, a ligand similarity search module 2, a DTI reasoning module 3 and a sorting module 4, wherein the data preprocessing module 1 is connected to the ligand similarity search module 2 and the DTI reasoning module 3 respectively, and the sorting module 4 is connected to the ligand similarity search module 2 and the DTI reasoning module 3 respectively. The data preprocessing module 1 is used to obtain a target protein, retrieve a template pocket with similar residues and topological structure to the pocket of the target protein from a database through pocket similarity comparison, and obtain a template protein corresponding to the template pocket; The ligand similarity search module 2 is used to select the co-crystal ligands of the N template proteins with the highest pocket similarity comparison scores, and score the target ligands using their co-crystal conformations to obtain similarity scores of the target ligands; the target ligands are potential active ligands of the target protein; The DTI reasoning module 3 is used to perform molecular docking between the target protein and the target ligand, obtain a molecular graph of the protein-ligand complex through the docking conformation, input the molecular graph into the graph attention network, calculate the classification result of whether the binding posture of the protein-ligand complex is active, and obtain the classification result score of the target ligand; The ranking module 4 is used to add the similarity score and the classification result score to obtain the final score of the target ligand, and to sort the target ligands from high to low according to the final score to obtain a re-scoring result.
[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.
Claims
1. A post-docking re-scoring method based on ligand similarity and DTI, It is characterized in that include: Obtain a target protein, retrieve a template pocket having residues and a topological structure similar to those of the pocket of the target protein from a database through pocket similarity comparison, and obtain a template protein corresponding to the template pocket; Select the co-crystal ligands of the N template proteins with the highest pocket similarity comparison scores, and use their co-crystal conformations to score the target ligand to obtain the similarity score of the target ligand; The target ligand is a potential active ligand of the target protein; The target protein is molecularly docked with the target ligand, and the molecular graph of the protein-ligand complex is obtained through the docking conformation. The molecular graph is input into the graph attention network to calculate the classification result of whether the binding posture of the protein-ligand complex is active, and the classification result score of the target ligand is obtained; The similarity score and the classification result score are added together to obtain the final score of the target ligand. The target ligands are sorted from high to low according to the final score to obtain the re-scoring result.
2. A post-docking re-scoring method based on ligand similarity and DTI as claimed in claim 1, It is characterized in that The process of obtaining the template protein corresponding to the template pocket is as follows: extracting the protein corresponding to the template pocket and its co-crystal ligand from the database by the pocket number, performing dehydration, hydrogenation and charge addition treatments on the protein to obtain the template protein.
3. A post-docking re-scoring method based on ligand similarity and DTI as claimed in claim 1, It is characterized in that The calculation formula of the similarity score LS_score of the target ligand is: Among them, Max[ ] means to find the maximum value, N probe represents the number of heavy atoms of the target ligand, represents the number of atomic pairs aligned between the target ligand and the cocrystal conformation of the cocrystal ligand, represents the distance between the i-th pair of aligned atoms, represents the difference in relative masses between the ith pair of aligned atoms, express The normalized scale of express The normalized scale of is a constant whose value is 10. represents the weighted Jaccard score, which measures the similarity of the chemical bonds associated with the ith pair of aligned atoms, represents the weight factor of the distance, A weighting factor representing relative quality, A weighting factor representing the similarity of chemical bonds.
4. A post-docking re-scoring method based on ligand similarity and DTI as claimed in claim 3, It is characterized in that is 0.45, is 0.1, is 0.
45.
5. A post-docking re-scoring method based on ligand similarity and DTI as claimed in claim 3, It is characterized in that N is 50.
6. A post-docking re-scoring method based on ligand similarity and DTI as claimed in claim 3, It is characterized in that The LS_score ranges from 0 to 1, with a score of 1 indicating a perfect structural match between the two ligands; The classification result is scored as 1 or 0, where 1 indicates that the binding pose of the protein-ligand complex is active and 0 indicates that the binding pose of the protein-ligand complex is inactive.
7. A post-docking re-scoring method based on ligand similarity and DTI as claimed in claim 1, It is characterized in that The conformation with the highest final score is retained among multiple conformations of the target ligand, and its final score is used as the final score of the target ligand.
8. A post-docking re-scoring system based on ligand similarity and DTI, It is characterized in that The method for performing the post-docking re-scoring method based on ligand similarity and DTI according to any one of claims 1 to 7 comprises: A data preprocessing module, a ligand similarity search module, a DTI reasoning module and a sorting module, wherein the data preprocessing module is connected to the ligand similarity search module and the DTI reasoning module respectively, and the sorting module is connected to the ligand similarity search module and the DTI reasoning module respectively; The data preprocessing module is used to obtain the target protein, retrieve the template pocket with similar residues and topological structure to the pocket of the target protein from the database through pocket similarity comparison, and obtain the template protein corresponding to the template pocket; The ligand similarity search module is used to select the co-crystal ligands of the N template proteins with the highest pocket similarity comparison scores, and score the target ligands using their co-crystal conformations to obtain the similarity scores of the target ligands; the target ligands are potential active ligands of the target protein; The DTI reasoning module is used to perform molecular docking of the target protein and the target ligand, obtain a molecular graph of the protein-ligand complex through the docking conformation, input the molecular graph into the graph attention network, calculate the classification result of whether the binding posture of the protein-ligand complex is active, and obtain the classification result score of the target ligand; The sorting module is used to add the similarity score and the classification result score to obtain the final score of the target ligand, and sort the target ligands from high to low according to the final score to obtain a re-scoring result.
Citation Information
Patent Citations
Template pocket searching method based on energy function
CN111180006A
Three-dimensional protein-ligand activity prediction method based on attention mechanism
CN115512785A
Cited By
Drug virtual screening method and system based on adaptation during testing
CN120727141A
A method and system for virtual screening of drugs based on on-the-fly adaptation
CN120727141B
