A data processing method based on a bioisostere search platform
By utilizing the data processing methods of the bioelectronic isosteric search platform, the problem of insufficient query results in existing tools has been solved, achieving efficient optimization and cost reduction in drug development. The SMILES coding and machine learning clustering algorithms have been used to optimize the query of bioelectronic isosteric substances, thereby improving the efficiency of information acquisition.
Patent Information
- Application Number
- CN202310237865.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-10
- Publication Date
- 2025-11-04
- Estimated Expiration
- 2043-03-10
AI Technical Summary
Existing bioelectronic isosteric search tools yield few results when searching for chemical groups, leading to increased drug development costs and low efficiency.
This paper provides a data processing method based on a bioelectronic isosteric search platform. By obtaining the SMILES codes of target functional groups, the method queries and optimizes bioelectronic isosteres, and combines machine learning unsupervised clustering algorithms to perform data clustering and visualization output.
It improves the efficiency of bioelectronic isosteric query, reduces drug development time and costs, provides more bioelectronic isosteric information and clustering methods, and expands users' thinking space.
Smart Images
Figure CN116092586B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of drug design, and particularly relates to a data processing method based on a bioisostere search platform. BACKGROUND
[0002] According to reading of relevant literatures, it can be known that in order to help researchers to select bioisostere, currently, a plurality of programs, databases and web servers and other computing tools for querying bioisostere have been developed at home and abroad. However, although the technologies and databases used by the plurality of bioisostere search tools currently available are different, when a user queries bioisostere of certain chemical groups by using these tools, only a very small number or almost no bioisostere can be queried.
[0003] In addition, with the increase of drug research and development cost, the demand for bioisostere is also increasing, however, the information about bioisostere is still insufficient at present. In the past, it is not easy to determine a suitable bioisostere group, and a large number of experiments are required. Since the computing tools only provide possible bioisostere, the more effective bioisostere can be provided, the more effective the tool is. SUMMARY
[0004] The purpose of the present application is to provide a data processing method based on a bioisostere search platform, so as to promote the optimization process of lead compounds in drug research and development, and reduce the time and cost of drug research and development.
[0005] The technical solution of the present application to solve the above technical problems is as follows:
[0006] The present application provides a data processing method based on a bioisostere search platform, which comprises:
[0007] S1: obtaining a target functional group input by a user;
[0008] S2: generating SMILES code of the target functional group;
[0009] S3: querying whether the bioisostere of the target functional group is contained in the database storing bioisostere according to the SMILES code, if yes, entering step S8, otherwise, entering step S4;
[0010] S4: processing the SMILES code to obtain a ligand of the target functional group;
[0011] S5: generating bioisostere of the target functional group according to the ligand of the target functional group;
[0012] S6: optimizing the bioisostere to obtain an optimal bioisostere;
[0013] S7: taking the bioisostere of the target functional group as the optimal bioisostere;
[0014] S8: taking the optimal bioisostere as a data processing result to output visually.
[0015] Optionally, in the step S1, the target functional group comprises a chemical group that needs to query a bioisostere.
[0016] Optionally, the step S4 comprises:
[0017] S41: replacing the wildcard * in the SMILES code with R to obtain a replaced SMILES code;
[0018] S42: converting the replaced SMILES code into a mol format and performing hydrogen addition processing to obtain a new SMILES code;
[0019] S43: querying ligands of the new SMILES code from all ligands;
[0020] S44: taking the ligands of the new SMILES code as ligands of the target functional group.
[0021] Optionally, the step S5 comprises:
[0022] S51: determining a protein in which the ligand of the target functional group exists as a reference protein;
[0023] S52: preprocessing the reference protein to obtain a preprocessed reference protein subchain;
[0024] S53: inputting a FASTA sequence of the preprocessed reference protein subchain into a Blastp tool and comparing with a protein sequence database to obtain a similar protein subchain similar to the preprocessed reference protein subchain;
[0025] S54: overlapping the reference protein subchain and the similar protein subchain to obtain an overlapped structure;
[0026] S55: taking each atom of the overlapped structure as a center, obtaining atoms of similar proteins within a specified radius, which are bioisosteres.
[0027] Optionally, the S52 comprises:
[0028] S521: judging a resolution of the reference protein, and retaining a protein subchain with a resolution higher than 1 angstrom as a preprocessed reference protein subchain; reference protein subchains;
[0029] S522: using a sequence comparison algorithm to globally compare all the reference protein subchains with resolution higher than to delete the duplicate reference protein subchains in the reference protein subchains with resolution higher than , obtaining available reference protein subchains;
[0030] S523: comparing the FASTA sequences of all available reference protein subchains pairwise, if the same, keeping the reference protein subchain with smaller resolution value and deleting the reference protein subchain with larger resolution value; if different, keeping all, obtaining pre-processed reference protein subchains.
[0031] Optionally, the S6 comprises: using ShaEP to calculate the similarity index of the superposition of the target functional group and the bioisostere, and setting a threshold for screening to obtain the optimal bioisostere.
[0032] Optionally, before the S8, the data processing method based on the bioisostere search platform further comprises:
[0033] post-processing the optimal bioisostere.
[0034] Optionally, the post-processing comprises:
[0035] A1: performing format conversion operation on the optimal bioisostere to obtain converted data;
[0036] A2: judging whether the converted data is complete, if yes, entering A4, otherwise, entering A3;
[0037] A3: performing data cleaning on the converted data and entering A4;
[0038] A4: calculating the molecular fingerprint of the converted data, and calculating the similarity between molecules according to the molecular fingerprint to obtain a similarity matrix;
[0039] A5: using a machine learning unsupervised clustering algorithm to perform data clustering on the converted data according to the similarity matrix, obtaining a clustering result;
[0040] A6: outputting the clustering result as the data processing result in a visualized manner.
[0041] Optionally, the A4 comprises:
[0042] using RDKit to calculate the molecular fingerprint and the similarity between molecules.
[0043] Optionally, the A5 comprises:
[0044] According to the similarity matrix, the machine learning unsupervised clustering algorithm is parameter adjusted to obtain optimal hyperparameters;
[0045] The number of clusters is determined according to the optimal hyperparameters;
[0046] According to the number of clusters, the converted data is data clustered to obtain a clustering result.
[0047] The present application has the following beneficial effects:
[0048] (1) The present application can make the medicinal chemist no longer need to obtain a small amount of bioisostere through repeated tests, and only through searching can a large amount of bioisostere information be obtained, so as to promote the optimization process of the lead compound in drug research and development to speed up, and reduce the drug research and development time and cost;
[0049] (2) The present application can query more bioisostere results, obtain more corresponding information, and provide a variety of unsupervised clustering methods for clustering the searched bioisostere, thereby providing users with a more broad thinking space and angle. BRIEF DESCRIPTION OF DRAWINGS
[0050] Figure 1 The flowchart of the data processing method based on the bioisostere search platform of the present application;
[0051] Figure 2 The unsupervised clustering algorithm parameter optimization result diagram of bioisostere;
[0052] Figure 3 The visualization result diagram of the unsupervised clustering of bioisostere. DETAILED DESCRIPTION
[0053] The principles and characteristics of the present application are described below in conjunction with the drawings, and the examples are only used to explain the present application and are not used to limit the scope of the present application.
[0054] The present application provides a data processing method based on a bioisostere search platform, as shown in Figure 1 The data processing method based on the bioisostere search platform comprises:
[0055] S1: obtaining a target functional group input by a user;
[0056] The target functional group of the present application is a chemical group of the bioisostere to be queried.
[0057] S2: generating a SMILES code of the target functional group;
[0058] As an embodiment, the skilled in the art can draw the chemical group needed to query the bioisostere in the chemical molecule editor JSME embedded in the front end of Django (open source web application framework), and Django generates the SMILES code of the target functional group by itself.
[0059] S3: According to the SMILES code, query whether the bioisostere containing the target functional group is contained in the database storing the bioisostere, if yes, go to step S8, otherwise, go to step S4;
[0060] S4: Process the SMILES code to obtain the ligand of the target functional group;
[0061] Alternatively, the S4 comprises:
[0062] S41: Replace the wildcard * in the SMILES code with R to obtain a replaced SMILES code;
[0063] S42: Convert the replaced SMILES code into mol format and do hydrogenation treatment to obtain a new SMILES code;
[0064] S43: Query the ligand of the new SMILES code from all ligands;
[0065] S44: Take the ligand of the new SMILES code as the ligand of the target functional group.
[0066] S5: According to the ligand of the target functional group, generate the bioisostere of the target functional group;
[0067] Alternatively, the S5 comprises:
[0068] S51: Determine the protein in which the ligand of the target functional group exists as a reference protein;
[0069] That is, all protein PDB files containing the reference ligand are obtained from the protein database, and the protein containing the reference ligand is defined as the reference protein.
[0070] S52: Preprocess the reference protein to obtain a preprocessed reference protein sub-chain;
[0071] Here, the preprocessing comprises:
[0072] S521: Determine the resolution of the reference protein, and retain the reference protein sub-chain with resolution higher than 2.5A;
[0073] In structural resolution, the resolution corresponds to the distance of the smallest observable feature, in X-ray crystallography, the resolution corresponds to the smallest distance between lattice planes resolved in the diffraction pattern. A larger resolution value indicates a lower resolution, and vice versa. In summary, resolution is a measure of the quality of protein, nucleic acid crystal data.
[0074] S522: using a sequence comparison algorithm, all the resolution is higher than The reference protein sub-chain is compared globally to delete the repeated reference protein sub-chain in the reference protein sub-chain with the resolution higher than
[0075] Specifically, the Needleman-Wunsch sequence alignment algorithm is used for comparison, and the comparison command is: needle -asequence seq1.fasta -bsequence seq2.fasta -outfile out.needle -gapopen 10 -gapextend 0.5 -sprotein. Unless specified, needle may guess whether the alignment is nucleic acid or protein according to IUPAC letters, IUPAC is a method of systematically naming chemicals, so the command uses the “-sprotein” flag to indicate that the protein sequence is being aligned, preventing errors from being thrown. After the command is executed, the sequence alignment result file is obtained and analyzed, and the data is selectively retained. After global alignment of the two protein sub-chains, the file obtained is required to check whether the value of Identity in the file is the same. If not, both sub-chains are retained, and if the same, the data of one chain can be retained.
[0076] S523: compare the consistency of the FASTA sequences of all available reference protein sub-chains two by two. If the same, retain the reference protein sub-chain with the smaller resolution value and delete the reference protein sub-chain with the larger resolution value; if different, retain all, to obtain the pre-processed reference protein sub-chain.
[0077] Similarly, the present application uses the Needleman-Wunsch sequence alignment algorithm to compare the consistency of the FASTA sequences of all available reference protein sub-chains two by two.
[0078] S53: input the FASTA sequence of the pre-processed reference protein sub-chain into the Blastp tool and compare it with the protein sequence database to obtain similar protein sub-chains similar to the pre-processed reference protein sub-chain;
[0079] S54: overlap the reference protein sub-chain and the similar protein sub-chain to obtain the overlapped structure;
[0080] S55: Using each atom of the superimposed structure as the center, obtain atoms of similar proteins within a specified radius, which are bioisosteres.
[0081] The processed single-stranded FASTA sequence of the protein is input into the Blastp tool and compared with a protein sequence database to detect protein chains with high similarity. Similar protein chains before and after inputting into Blastp are then overlap-aligned using the US-align tool. In the overlapped structure, all atoms within a specified radius are obtained, centered on each atom of the target functional group. All atoms within this specified radius represent the bioisosteric structure.
[0082] S6: Optimize the bioelectron isosteres to obtain the optimal bioelectron isosteres;
[0083] That is, ShaEP is used to calculate the similarity index between the target functional group and the bioelectron isosteres, and a threshold is set for screening to obtain the optimal bioelectron isosteres.
[0084] S7: The bioelectron isosteric variant of the target functional group is selected as the optimal bioelectron isosteric variant;
[0085] S8: The optimal bioelectroisoster is used as the data processing result and output it in a visual format.
[0086] Optionally, prior to step S8, the data processing method based on the bioelectronic isosteric search platform further includes:
[0087] The optimal bioelectron isosteres are then post-processed.
[0088] Alternatively, the post-processing includes:
[0089] A1: Perform a format conversion operation on the optimal bioelectronic isosteres to obtain the converted data;
[0090] This will convert the SMILES encoding format to SDF format.
[0091] A2: Determine if the converted data is complete. If yes, proceed to A4; otherwise, proceed to A3.
[0092] SDF files have a standard format. By writing judgment code, we can check whether they conform to the standard format and thus determine whether the data is complete.
[0093] A3: Perform data cleaning on the transformed data and proceed to A4;
[0094] Here, the data cleaning includes deletion of invalid data and deletion of duplicate data. As an embodiment, after obtaining the corresponding 2228 molecular data by the calculation process, the present application needs to clean the 2228 data, first delete the invalid data by writing code to obtain 1236 data points, and then delete the duplicate data to obtain 1157 available data.
[0095] A4: Calculate the molecular fingerprints of the converted data, and calculate the similarity between molecules according to the molecular fingerprints to obtain a similarity matrix;
[0096] The present application calculates the molecular fingerprints and the similarity between molecules by using RDKit.
[0097] The 2048-bit length Morgan molecular fingerprints (MFF) of these data points are calculated by using the RDKit method, and the "Tanimoto similarity coefficient" of each molecule is calculated according to the MFF to obtain the similarity matrix between molecules
[0098] A5: According to the similarity matrix, the converted data is clustered by using a machine learning unsupervised clustering algorithm to obtain a clustering result;
[0099] That is, the similarity coefficient of 1157 data is used to form a molecular similarity matrix as a model input, and the data is automatically classified by using a machine learning unsupervised clustering algorithm.
[0100] It should be noted that the machine learning unsupervised clustering algorithm includes types that need to adjust the number of clusters and types that do not need to adjust the number of clusters. The type of algorithm that needs to adjust the number of clusters is illustrated by taking K-means as an example: K-means algorithm needs to specify the super parameter "number of clusters" (n_components), so it is necessary to search for the best value of this super parameter in a certain range, and the index used to evaluate the clustering effect of the algorithm is called Silhouette Coefficient. In the parameter space 2-150, the super parameter is searched, the relationship between the super parameter and the Silhouette Coefficient is obtained, and the super parameter range corresponding to the region where the slope changes sharply in the figure is selected according to the elbow rule 5-20 as Figure 2 (a), and then further select parameters such as Figure 2 (b), it is calculated that the super parameter is 6, that is, when the number of clusters is 6, the classification effect of unsupervised clustering is the best.
[0101] Therefore, if a person skilled in the art selects an algorithm type that needs to adjust the super parameter, then:
[0102] According to the similarity matrix, the machine learning unsupervised clustering algorithm is adjusted in parameters to obtain the optimal super parameter;
[0103] determining the number of clusters according to the optimal hyperparameters;
[0104] data clustering the converted data according to the number of clusters to obtain a clustering result.
[0105] The type of algorithm that needs to adjust the hyperparameters is taken as an example of DBSCAN: DBSCAN clusters according to data density. When performing DBSCAN unsupervised clustering, only the initial parameters (eps = 1.5, min_samples = 3) need to be set, and the hyperparameter cluster number does not need to be set. Then, unsupervised clustering can be directly performed. The clustering result shows that when the number of clusters is 142, the best clustering effect is achieved, and the silhouette coefficient is 0.216.
[0106] After clustering by the unsupervised clustering algorithm, the data also needs to be visualized to intuitively present the clustering result to the user. TSNE and PCA are used to reduce the dimensionality of the 2048-bit molecular fingerprint to 2D and 3D, respectively. Then, matplotlib is used to draw two-dimensional and three-dimensional graphs of the reduced data. Different colors are used to represent different categories in the graphs, such as Figure 3 .
[0107] A6: outputting the clustering result as the data processing result for visualization.
[0108] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A data processing method based on a bioelectronic isosteric search platform, characterized in that, The data processing method based on the bioelectronic isosteric search platform includes: S1: Obtain the target functional groups input by the user; S2: Generate the SMILES encoding of the target functional group; S3: Query the database storing bioelectron isosteres according to the SMILES code to see if it contains bioelectron isosteres of the target functional group. If yes, proceed to step S8; otherwise, proceed to step S4. S4: Process the SMILES encoding to obtain the ligand of the target functional group; S5: Generate a bioisostere of the target functional group based on the ligand of the target functional group; S6: Optimize the bioelectron isosteres to obtain the optimal bioelectron isosteres; S7: The bioelectron isosteres of the target functional group are taken as the optimal bioelectron isosteres; S8: The optimal bioelectroisoster is used as the data processing result and output it in a visual format. Step S4 includes: S41: Replace the wildcard * in the SMILES encoding with R to obtain the SMILES encoding after replacement; S42: Convert the replaced SMILES code into mol format and perform hydrogenation to obtain a new SMILES code; S43: Search for the ligands encoded by the new SMILES from all ligands; S44: Use the new SMILES-encoded ligand as the ligand for the target functional group; Step S5 includes: S51: The protein containing the ligand of the target functional group is identified as the reference protein; S52: Pre-treat the reference protein to obtain a pre-treated reference protein sub-chain; S53: Input the FASTA sequence of the preprocessed reference protein sub-chain into the Blastop tool and compare it with the protein sequence database to obtain a similar protein sub-chain that is similar to the preprocessed reference protein sub-chain. S54: Overlap the reference protein sub-chain and the similar protein sub-chain to obtain the overlapped structure; S55: Using each atom of the superimposed structure as the center, obtain atoms of similar proteins within a specified radius, which are bioisosteres; S52 includes: S521: Determine the resolution of the reference protein and retain reference protein sub-chains with a resolution higher than 2.7 Å; S522: Use a sequence comparison algorithm to perform pairwise global comparisons on all the reference protein sub-chains with a resolution higher than 2.7 Å, so as to delete the duplicate reference protein sub-chains in the reference protein sub-chains with a resolution higher than 2.7 Å, and obtain usable reference protein sub-chains; S523: Compare the FASTA sequences of all available reference protein sub-chains pairwise. If they are the same, retain the reference protein sub-chain with the smaller resolution value and delete the reference protein sub-chain with the larger resolution value. If they are different, retain all of them to obtain the preprocessed reference protein sub-chain.
2. The data processing method based on the bioelectronic isosteric search platform according to claim 1, characterized in that, In step S1, the target functional group includes chemical groups for which bioisosteres need to be queried.
3. The data processing method based on the bioelectronic isosteric search platform according to claim 1, characterized in that, S6 includes: using ShaEP to calculate the similarity index of the target functional group and the bioelectron isosteres, and setting a threshold for screening to obtain the optimal bioelectron isosteres.
4. The data processing method based on a bioelectronic isosteric search platform according to any one of claims 1-3, characterized in that, Prior to step S8, the data processing method based on the bioelectronic isosteric search platform further includes: The optimal bioelectron isosteres are then post-processed.
5. The data processing method based on the bioelectronic isosteric search platform according to claim 4, characterized in that, The post-processing includes: A1: Perform a format conversion operation on the optimal bioelectronic isosteres to obtain the converted data; A2: Determine if the converted data is complete. If yes, proceed to A4; otherwise, proceed to A3. A3: Perform data cleaning on the transformed data and proceed to A4; A4: Calculate the molecular fingerprint of the converted data, and calculate the similarity between molecules based on the molecular fingerprint to obtain a similarity matrix; A5: Based on the similarity matrix, the transformed data is clustered using an unsupervised clustering algorithm based on machine learning to obtain the clustering results; A6: The clustering results are used as the data processing results and output as a visualization.
6. The data processing method based on the bioelectronic isosteric search platform according to claim 5, characterized in that, The A4 includes: The molecular fingerprint and the similarity between molecules were calculated using RDKit.
7. The data processing method based on the bioelectronic isosteric search platform according to claim 5, characterized in that, The A5 includes: Based on the similarity matrix, the parameters of the machine learning unsupervised clustering algorithm are adjusted to obtain the optimal hyperparameters; The number of clusters is determined based on the optimal hyperparameters; The transformed data is clustered based on the cluster number to obtain the clustering result.