Cryoelectron microscope structure database building method and system and storage medium
By adjusting protein chains and sequences based on density maps from EMDB and combining this with structure prediction optimization, the problem of insufficient protein complex structure prediction in AlphaFoldDB was solved, resulting in a high-quality cryo-electron microscopy structure database.
Patent Information
- Application Number
- CN202511544144.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-28
- Publication Date
- 2026-02-24
AI Technical Summary
The existing AlphaFoldDB database lacks experimental structural information for protein structure prediction, especially in protein complex structures, and the random coil state of amino acid regions lacks intuitiveness, resulting in an overall structural quality lower than that of the PDB database.
By aligning protein chains with the density map based on EMDB, template structures are obtained. Combined with protein sequences and unmodeled annotations, structure prediction and optimization are performed. The model-density map correlation coefficient is calculated, candidate structures are adjusted, and finally imported into the database.
It improves the structural accuracy of the database, overcomes the limitations of protein complex structure prediction, and enables the database structure quality to reach or exceed the level of the PDB database.
Smart Images
Figure CN121565253A_ABST
Abstract
Description
Technical Field This application belongs to the field of structural biology technology, and in particular relates to a method, system and storage medium for building a cryo-electron microscopy structural database. Background Technology
[0001] In bioinformatics and structural biology, the construction of structural databases effectively bridges the gap between sequences and structures, providing structural biologists with valuable data resources and playing a crucial role in biological research. With the improvement in the performance of protein structure prediction algorithms, structural prediction has moved from exploratory research to a practical stage, predicting structures of high quality and providing new research insights into biology. However, the corresponding operational resources required—high-performance GPUs, downloading massive sequence databases, and configuring complex software environments on Linux systems—place a significant burden on researchers with a biological background.
[0002] AlphaFoldDB is a database that integrates the structures predicted from all existing protein sequences (more than 200 million sequences in the UniProt database) using AlphaFold. Using AlphaFoldDB, researchers can easily obtain the structures of proteins of interest without having to run complex structure prediction processes themselves. Furthermore, this database is interoperable with mainstream biological databases such as UniProt, greatly facilitating researchers' structural understanding when exploring protein function.
[0003] However, AlphaFoldDB still has limitations. First, AlphaFoldDB's structure predictions are based solely on protein sequences, lacking structural information from experimental methods, leading to discrepancies between some structures in the database and actual protein structures. Second, the structure prediction algorithm used in AlphaFoldDB remains significantly inadequate for protein complexes, thus it does not yet provide data on protein complex structures, resulting in a lack of systematic structural information in this field. Furthermore, some amino acid regions in AlphaFoldDB structures may exhibit random coil states. Although pLDDT can indicate the reliability of each amino acid, this method lacks the intuitiveness of explicitly removing invalid amino acids in the PDB database, making the overall structure quality of AlphaFoldDB lower than that of the PDB database. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this application provides a method, system, and storage medium for building a cryo-electron microscopy structure database. This application achieves this through the following technical solution: Based on the density map of EMDB, obtain the input structure of each density map; Based on the density map, the protein chains in the input structure are adjusted to align with the density map to obtain the template structure; Based on the density map, the protein sequences and unmodeled annotations in the input structure are adjusted to obtain protein sequences and unmodeled annotations that match the template structure. Based on the density map, protein sequence, and template structure, structure prediction and optimization are performed to obtain candidate structures; Calculate the model-density map correlation coefficient between the candidate structure and the density map, and adjust the candidate structure according to the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure; The final structure is evaluated, and structural annotations are generated by combining the protein sequence and unmodeled annotations. The final structure and structural annotations are then imported into a database.
[0005] Optional, The step of adjusting the protein chains in the input structure based on the density map to align them with the density map and obtaining the template structure includes: The input structure is split into single protein chains; Check if each protein chain exists in the corresponding density map; if not, delete it. The chain that appears more frequently in the density map than the input structure is copied and rotated and translated to align it with the density map until the density map is completely filled and there are no extra protein chains. All protein chains are saved as template structures.
[0006] Optional, The adjustment of protein sequences and unmodeled annotations in the input structure based on density maps includes: The input structure is parsed to obtain the protein sequence and unmodeled annotations; The protein sequence and unmodeled annotation are deleted based on the density map: if the chain corresponding to the protein sequence does not exist in the corresponding density map, the protein sequence and unmodeled annotation are deleted. The protein sequence and unmodeled annotation are copied based on the density map: if the chain corresponding to the protein sequence appears more frequently in the density map than the input structure, then the protein sequence and unmodeled annotation are copied.
[0007] Optional, The process of predicting and optimizing the structure based on the density map, protein sequence, and template structure to obtain candidate structures includes: Density maps, protein sequences, and template structures are used as input data and fed into a structure prediction network to predict the structure and obtain a preliminary structure. The geometric features and density map fit of the preliminary structure are optimized to obtain candidate structures.
[0008] Optional, The method further includes: Check the degree of fit between the preliminary structure, candidate structure, and density map, and perform iterative optimization until the degree of fit meets the preset threshold.
[0009] Optional, The calculation of the model-density map correlation coefficient between the candidate structure and the density map, and the adjustment of the candidate structure based on the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure, includes: Calculate the model-density map correlation coefficient between candidate structures and density maps, and obtain the correlation coefficient for each amino acid; For each protein chain, multiple fragments are extracted based on unmodeled annotations, and the average correlation coefficient of each fragment is calculated based on the correlation coefficient of each amino acid. If the average correlation coefficient of the fragment is lower than the average correlation coefficient of the protein chain, then all amino acids contained in the fragment are deleted from the candidate structure. If there are still amino acids in the candidate structures after deletion that are not present in the density map, then modify the unmodeled annotation to mark the extra amino acids as unmodeled as well, and recalculate the average correlation coefficient of each fragment. If the candidate structure after deletion does not contain any amino acids that appear in the density map, the final structure will be output.
[0010] Optional, The structural evaluation of the final structure includes: Calculate the evaluation metrics, which include: model geometric metrics and model-density plot correlation coefficients.
[0011] This application also provides a cryo-electron microscopy structure database construction system for implementing the aforementioned method, the system comprising: The input structure module is used to obtain the input structure of each density map based on the EMDB density map; The template structure module is used to adjust the protein chains in the input structure based on the density map to align them with the density map, thereby obtaining the template structure. The protein sequence module is used to adjust the protein sequences and unmodeled annotations in the input structure based on the density map to obtain protein sequences and unmodeled annotations that match the template structure. The candidate structure module is used to predict and optimize the structure based on the density map, protein sequence and template structure to obtain candidate structures. The final structure module is used to calculate the model-density map correlation coefficient between the candidate structure and the density map, and adjust the candidate structure according to the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure; The data import module is used to perform structural evaluation on the final structure, generate structural annotations by combining protein sequences and unmodeled annotations, and import the final structure and structural annotations into the database.
[0012] This application also provides a computer-readable storage medium storing one or more programs, which, when executed, can implement the aforementioned method for building a cryo-electron microscopy structure database.
[0013] Compared with the prior art, this application has the following advantages: 1. The cryo-electron microscopy (CELM) structure database construction method proposed in this application obtains the input structure of each density map based on the density map data in the EMDB. By adjusting the input structure, a template structure and its corresponding protein sequence and unmodeled annotation are obtained. Candidate structures are obtained through structure prediction and optimization. The candidate structures are adjusted according to the model-density map correlation coefficient between the unmodeled annotation, the candidate structure, and the density map to obtain the final structure. The final structure is evaluated, and a structure annotation is generated by combining the protein sequence and the unmodeled annotation. The final structure and the structure annotation are imported into the database. This method enables the structures in the database to achieve the accuracy of experimental resolution and overcomes the limitations of protein complex structure prediction.
[0014] 2. Through thorough manual inspection and revision, the structural quality of the database can be made to reach the same or better level as the PDB database, thereby improving the structural quality of the database.
[0015] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 A flowchart illustrating the method for building a cryo-electron microscopy structure database is shown. Figure 2 A schematic block diagram of the cryo-electron microscopy structure database construction system is shown. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] The cryo-electron microscopy structure database construction method of this application embodiment can be used to establish a novel cryo-electron microscopy structure database construction process, so that the structures in the database can achieve the accuracy of experimental resolution and overcome the limitations of protein complex structure prediction. At the same time, through detailed manual inspection and revision, the quality of the structures in the database can reach the same or better level as the PDB database.
[0020] See appendix Figure 1 The method of this application includes: S1. Based on the density map of EMDB, obtain the input structure of each density map.
[0021] Specifically, based on the density map data of EMDB, the input structure of each density map is obtained from the PDB database. The input structure is the standard structure or homologous structure of the density map.
[0022] In this embodiment, based on the density maps in the EMDB, the standard structure or homologous structure of each density map is obtained from the PDB database through database search, literature review, and other methods, and used as the "input structure".
[0023] S2. Based on the density map, adjust the protein chains in the input structure to align them with the density map to obtain the template structure; The step of adjusting the protein chains in the input structure based on the density map to align them with the density map and obtaining the template structure includes: The input structure is split into single protein chains; Check if each protein chain exists in the corresponding density map; if not, delete it. Copy the protein chains that appear more frequently in the density map than the input structure, and rotate and translate them to align them with the density map until the density map is completely filled and there are no extra protein chains. All protein chains are saved as template structures.
[0024] A protein chain is a molecule composed of multiple amino acids linearly linked by peptide bonds in a protein structure. A protein structure typically contains multiple protein chains.
[0025] In this embodiment, the density map and input structure are opened in ChimeraX software, and the following operations are performed on the input structure: the input structure is split into individual protein chains using the split command; protein chains present in the input structure but not in the density map are deleted using the close command; protein chains that appear more frequently in the density map than in the input structure are copied using the combine command, rotated, and translated, and then matched with the density map using the fitmap command until the entire density map is completely filled and there are no extra protein chains; all protein chains are saved as a "template structure" using the save command. Through the above operations, the input structure can be fitted to the density map to obtain the template structure.
[0026] S3. Based on the density map, adjust the protein sequences and unmodeled annotations in the input structure to obtain protein sequences and unmodeled annotations that match the template structure.
[0027] The protein sequences and unmodeled annotations in the input structure are adjusted (deleted and / or copied) based on the density map, including: The input structure is parsed to obtain the protein sequence and unmodeled annotations; The protein sequence and unmodeled annotation are deleted based on the density map: if the chain corresponding to the protein sequence does not exist in the corresponding density map, the protein sequence and unmodeled annotation are deleted. The protein sequence and unmodeled annotation are copied based on the density map: if the chain corresponding to the protein sequence appears more frequently in the density map than the input structure, then the protein sequence and unmodeled annotation are copied.
[0028] By integrating the above adjustments, protein sequences and unmodeled annotations that match the template structure can be obtained.
[0029] In this embodiment, a Python script is used to extract protein sequences and unmodeled annotations from the .cif format file of the input structure. The density map is checked in ChimeraX software to see if it matches the protein chain corresponding to the protein sequence. The extracted protein sequences and unmodeled annotations are then deleted and copied to obtain protein sequences and unmodeled annotations that match the template structure.
[0030] In this context, unmodeled annotation refers to whether each amino acid in a protein sequence has three-dimensional coordinate information in its corresponding structure. If there is no three-dimensional coordinate information, it is called "unmodeled".
[0031] S4. Based on the density map, protein sequence and template structure, perform structure prediction and optimization to obtain candidate structures.
[0032] Specifically, based on the density map, protein sequence, and template structure, structure prediction and optimization are performed to obtain candidate structures, including: Density maps, protein sequences, and template structures are used as input data and fed into a structure prediction network to predict the structure and obtain a preliminary structure. The geometric features and density map fit of the preliminary structure are optimized to obtain candidate structures.
[0033] In this embodiment, a deep neural network containing the Cryoformer cryofold model can be used for structure prediction, including: Density maps, protein sequences, and template structures are used as input data and fed into a deep neural network. The deep neural network includes multiple cryoformer modules, each of which includes an encoder and a decoder. The decoder is used to learn the matching of protein sequence correlation characterization and density map three-dimensional spatial information. It cross-fuses the protein sequence correlation characterization and the three-dimensional spatial information output by the encoder to output the protein's full atomic structure model as a preliminary structure.
[0034] In this embodiment, protein optimization software can be used to optimize the geometric features and density map fit of the preliminary structure. For example, the phenix.real_space_refine command of PHENIX software can be used to optimize the geometric features and density map fit of the preliminary structure to obtain a "candidate structure".
[0035] In this embodiment, the degree of agreement between the preliminary structure, candidate structure, and density map is checked, and iterative optimization is performed until the degree of agreement meets a preset threshold.
[0036] This includes checking the degree of fit between the preliminary structure, candidate structure, and density map, and performing iterative optimization, including: If the degree of fit between the preliminary structure and the density map does not meet the preset threshold, the structure prediction is repeated. If necessary, the protein chains in the preliminary structure that do not match the density map can be visualized and checked in ChimeraX. Then, step S2 is repeated to generate the template structure and step S3 is repeated to generate the protein sequence. If the degree of fit between the candidate structure and the density map does not meet the preset threshold, the structure optimization will be performed again. If necessary, the weights of the density map can be modified during the structure optimization.
[0037] In this embodiment, the degree of agreement between the preliminary structure, candidate structure, and density map is checked in ChimeraX: if the preliminary structure has three or more alpha spirals that are more than 10 angstroms away from the density map, the structure prediction is performed again; if the candidate structure has three or more alpha spirals that are more than 10 angstroms away from the density map, the structure optimization is performed again.
[0038] S5. Calculate the model-density map correlation coefficient between the candidate structure and the density map, and adjust the candidate structure according to the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure.
[0039] The process includes calculating the model-density map correlation coefficient between the candidate structure and the density map, processing the candidate structure based on the unmodeled annotations and the model-density map correlation coefficient, and obtaining the final structure, including: The model-density map correlation coefficient between the candidate structure and the density map is calculated, and the correlation coefficient for each amino acid is obtained. The correlation coefficient ranges from 0 to 1 and is used to measure the degree of matching between the candidate structure and the density map. For each protein chain, multiple segments (i.e. multiple consecutive unmodeled amino acids) are split based on unmodeled annotations. The average correlation coefficient of each segment is calculated based on the correlation coefficient of each amino acid. If the average correlation coefficient of a segment is lower than the average correlation coefficient of the protein chain, all amino acids contained in that segment are deleted from the candidate structure. If there are still amino acids in the candidate structure after deletion that are not present in the density map, then modify the unmodeled annotation to mark these extra amino acids as unmodeled as well, and then recalculate the average correlation coefficient of each fragment; if there are no amino acids in the candidate structure after deletion that are not present in the density map, then the final structure is output.
[0040] In this embodiment, the phenix.model_map_cc command of the PHENIX software is used to calculate the model-density map correlation coefficient between the candidate structure and the density map, that is, each amino acid can be calculated to have a value between 0 and 1; the ChimeraX software is used to check whether there are any extra amino acids in the density map after the candidate structure has been deleted.
[0041] S6. Perform structural evaluation on the final structure, generate structural annotations by combining the protein sequence and unmodeled annotations, and import the final structure and corresponding structural annotations into the database.
[0042] The process includes: performing structural evaluation on the final structure; generating structural annotations by combining the protein sequence and unmodeled annotations; and importing the final structure and its corresponding structural annotations into a database. The final structure is evaluated, and evaluation indicators are calculated, including: model geometric indicators and model-density map correlation coefficient; Protein sequences, unmodeled annotations, and evaluation metrics are integrated into structural annotations and imported into the database along with the final structure.
[0043] In this embodiment, protein structure evaluation software can be used to automatically evaluate the "final structure," such as using the phenix.validation_cryoem command in PHENIX software to automatically evaluate the "final structure," calculating nine model geometric indices (including: angle RMSD, bond RMSD, clash score, C-beta deviations, Ramachhardran favored, Ramachhardran outliers, Ramachhardran Z score, rotamer favored, and rotamer outliers) and six model-density plot correlation coefficients (including: CC...). mask CC box CC peaks CC volume CC mc and CC sc The protein sequence, unmodeled annotations, evaluation metrics, and other data are integrated into structural annotations and imported into the database along with the final structure.
[0044] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0045] See appendix Figure 2The diagram illustrates the structure of a cryo-electron microscopy structure database construction system for implementing the above method, including an input structure module, a template structure module, a protein sequence module, a candidate structure module, a final structure module, and a data import module.
[0046] The input structure module is used to obtain the input structure of each density map based on the EMDB density map; The template structure module is used to adjust the protein chains in the input structure based on the density map to align them with the density map, thereby obtaining the template structure. The protein sequence module is used to adjust the protein sequences and unmodeled annotations in the input structure based on the density map to obtain protein sequences and unmodeled annotations that match the template structure. The candidate structure module is used to predict and optimize the structure based on the density map, protein sequence and template structure to obtain candidate structures. The final structure module is used to calculate the model-density map correlation coefficient between the candidate structure and the density map, and adjust the candidate structure according to the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure; The data import module is used to perform structural evaluation on the final structure, generate structural annotations by combining protein sequences and unmodeled annotations, and import the final structure and structural annotations into the database.
[0047] Based on the same concept, this application also provides a computer-readable storage medium storing one or more programs, which, when executed, can realize the aforementioned method for building a cryo-electron microscopy structure database.
[0048] Although this application 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; and these 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 this application.
Claims
1. A method for constructing a cryo-electron microscopy structure database, characterized in that, include: Based on the density map of EMDB, obtain the input structure of each density map; Based on the density map, the protein chains in the input structure are adjusted to align with the density map to obtain the template structure; Based on the density map, the protein sequences and unmodeled annotations in the input structure are adjusted to obtain protein sequences and unmodeled annotations that match the template structure. Based on the density map, protein sequence, and template structure, structure prediction and optimization are performed to obtain candidate structures; Calculate the model-density map correlation coefficient between the candidate structure and the density map, and adjust the candidate structure according to the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure; The final structure is evaluated, and structural annotations are generated by combining the protein sequence and unmodeled annotations. The final structure and structural annotations are then imported into a database.
2. The method according to claim 1, characterized in that, The step of adjusting the protein chains in the input structure based on the density map to align them with the density map and obtaining the template structure includes: The input structure is split into single protein chains; Check if each protein chain exists in the corresponding density map; if not, delete it. Copy the protein chains that appear more frequently in the density map than the input structure, and rotate and translate them to align them with the density map until the density map is completely filled and there are no extra protein chains. All protein chains are saved as template structures.
3. The method according to claim 2, characterized in that, The adjustment of protein sequences and unmodeled annotations in the input structure based on density maps includes: The input structure is parsed to obtain the protein sequence and unmodeled annotations; The protein sequence and unmodeled annotation are deleted based on the density map: if the chain corresponding to the protein sequence does not exist in the corresponding density map, the protein sequence and unmodeled annotation are deleted. The protein sequence and unmodeled annotation are copied based on the density map: if the chain corresponding to the protein sequence appears more frequently in the density map than the input structure, then the protein sequence and unmodeled annotation are copied.
4. The method according to claim 1, characterized in that, The process of predicting and optimizing the structure based on the density map, protein sequence, and template structure to obtain candidate structures includes: Density maps, protein sequences, and template structures are used as input data and fed into a structure prediction network to predict the structure and obtain a preliminary structure. The geometric features and density map fit of the preliminary structure are optimized to obtain candidate structures.
5. The method according to claim 4, characterized in that, The method further includes: Check the degree of fit between the preliminary structure, candidate structure, and density map, and perform iterative optimization until the degree of fit meets the preset threshold.
6. The method according to any one of claims 1-5, characterized in that, The calculation of the model-density map correlation coefficient between the candidate structure and the density map, and the adjustment of the candidate structure based on the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure, includes: Calculate the model-density map correlation coefficient between candidate structures and density maps, and obtain the correlation coefficient for each amino acid; For each protein chain, multiple fragments are extracted based on unmodeled annotations, and the average correlation coefficient of each fragment is calculated based on the correlation coefficient of each amino acid. If the average correlation coefficient of the fragment is lower than the average correlation coefficient of the protein chain, then all amino acids contained in the fragment are deleted from the candidate structure. If there are still amino acids in the candidate structures after deletion that are not present in the density map, then modify the unmodeled annotation to mark the extra amino acids as unmodeled as well, and recalculate the average correlation coefficient of each fragment. If the candidate structure after deletion does not contain any amino acids that appear in the density map, the final structure will be output.
7. The method according to claim 1, characterized in that, The structural evaluation of the final structure includes: Calculate the evaluation metrics, which include: model geometric metrics and model-density plot correlation coefficients.
8. A cryo-electron microscopy structure database construction system, characterized in that, The system includes: The input structure module is used to obtain the input structure of each density map based on the EMDB density map; The template structure module is used to adjust the protein chains in the input structure based on the density map to align them with the density map, thereby obtaining the template structure. The protein sequence module is used to adjust the protein sequences and unmodeled annotations in the input structure based on the density map to obtain protein sequences and unmodeled annotations that match the template structure. The candidate structure module is used to predict and optimize the structure based on the density map, protein sequence and template structure to obtain candidate structures. The final structure module is used to calculate the model-density map correlation coefficient between the candidate structure and the density map, and adjust the candidate structure according to the unmodeled annotation and the model-density map correlation coefficient to obtain the final structure; The data import module is used to perform structural evaluation on the final structure, generate structural annotations by combining protein sequences and unmodeled annotations, and import the final structure and structural annotations into the database.
9. The system according to claim 8, characterized in that, The structural evaluation of the final structure includes: Calculate the evaluation metrics, which include: model geometric metrics and model-density plot correlation coefficients.
10. A computer-readable storage medium storing one or more programs, characterized in that, When one or more of these programs are executed, the cryo-electron microscopy structure database construction method described in any one of claims 1-7 can be implemented.