A method for finding proteins with specific functions in organisms
Through the Alphafold prediction and structural similarity alignment method, the problem of low sequence homology in protein function annotation is solved, and the accuracy and scope of functional prediction is improved, especially in proteins with large sequence differences but similar structural functions are displayed. A unique advantage.
Patent Information
- Application Number
- CN202410346157.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-25
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-03-25
AI Technical Summary
The prior art relies on sequence homology comparison in protein function annotation, and cannot fully utilize protein structure information. Especially for proteins with large sequence differences, their prediction accuracy is reduced and their function cannot be accurately annotated.
Through the whole genomic protein structure predicted by Alphafold, combined with uniprot database sequence alignment and local AlphaFold software structure prediction, the structural similarity comparison was performed using US-align software to screen out proteins with similar structural functions to the target.
Improve the accuracy of protein function prediction, expand the range of annotable proteins, and overcome the limitations of functional annotation accuracy of traditional methods when proteins with large sequence differences but similar structures and functions.
Smart Images

Figure CN118230822B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of bioinformatics, and specifically relates to a method for mining enzymes with target functions in unknown functional proteins and proteins that have huge sequence differences with known target proteins but have consistent structures and functions through whole-genome protein structures predicted by Alphafold. Background Art
[0002] With the rapid development of high-throughput sequencing technology, biological research has entered an unprecedented era of genomic big data. This technological advancement enables us to obtain genome sequences of whole species at an unprecedented speed and cost efficiency. Especially in the field of microorganisms, tens of thousands of strains have been sequenced, generating a large amount of protein sequence data. These data are crucial for understanding biodiversity, disease mechanisms, and the development of new drugs and enzymes.
[0003] However, although it is becoming easier to obtain sequence data, accurate annotation of protein function remains a huge challenge. Most current functional annotations are based on comparative analysis of sequence homology, that is, predicting function by comparing the sequence similarity of known functional proteins with that of unknown proteins. This method is effective to a certain extent, but it relies on a sequence database of known proteins, and its prediction accuracy is greatly reduced for proteins with low sequence homology.
[0004] To complicate matters further, even proteins with very different sequences may have very similar three-dimensional (3D) structures and biological functions. This suggests that sequence-based function prediction methods may not fully utilize protein structure information to identify functionally related proteins. Therefore, it is necessary to develop a new method that can directly use protein 3D structure information for functional annotation.
[0005] Recently, thanks to breakthroughs in protein structure prediction technologies such as AlphaFold, we now have the ability to predict the 3D structure of proteins with high accuracy. This provides unprecedented opportunities to use structural information to improve protein function prediction. With these advanced structure prediction tools, we can perform large-scale predictions of protein structure at the genome-wide level, thus overcoming the limitations of traditional functional annotation methods based on sequence comparison.
[0006] In summary, although the advancement of genome sequencing technology has provided a large amount of protein sequence data for biological research, how to accurately annotate the functions of these proteins remains an unresolved problem. Current annotation methods based on sequence homology have obvious limitations, especially when dealing with proteins with large sequence differences but similar structures and functions. The accuracy of these methods is greatly reduced and they cannot accurately annotate the functions of proteins. Although the 3D structure of a protein is crucial to its function, existing methods fail to fully utilize structural information for function prediction. Therefore, it is particularly urgent and important to develop a new method that can use the 3D structural information of a protein for function prediction. Summary of the invention
[0007] The purpose of the present invention is to provide a method for searching for proteins with specific functions in organisms, and to mine enzymes with target functions in unknown functional proteins and proteins that have huge sequence differences with known target proteins but maintain consistent structure and function through the whole genome protein structure predicted by Alphafold.
[0008] In order to achieve the above object, the technical solution of the present invention is summarized as follows:
[0009] First, the whole genome protein to be searched is compared with the uniprot database sequence, and the corresponding public AlphaFold structure is downloaded for the 100% similar sequence according to the matched uniprot database protein ID. For proteins that do not have a public predicted structure and proteins that are not 100% matched to the uniprot database, the locally installed AlphaFold software is used for structural prediction. After collecting the structural data set of the whole protein, the US-align software is used to align the whole protein with the structure of the known functional protein or the local structure of its active center, and the proteins with a structure similar to the target functional protein are screened according to the alignment score.
[0010] Specifically, a method for searching for a protein with a specific function in an organism comprises the following steps:
[0011] (1) collecting and extracting DNA of the organism, sequencing and assembling the genome, and annotating the assembly results with proteins;
[0012] (2) using the uniprot database to perform sequence alignment on the obtained protein sequences to identify sequences that match known protein sequences; downloading the known protein structures corresponding to the matching sequences from the A1phaFold Protein Structure Database;
[0013] (3) For proteins with no known structure, the locally installed AlphaFold program was used to predict the structure of the protein sequence;
[0014] (4) Organize and rename the obtained protein structure files and build a new protein structure database;
[0015] (5) using US-align software to perform a full protein structure alignment on the new protein structure database constructed in step (4) to identify proteins with similar structures to the specific target protein;
[0016] (6) based on the structural comparison results, selecting a protein having a structure similar to the target protein;
[0017] (7) Express, purify and functionally verify the selected protein.
[0018] The specific operation of step (2) is as follows: download the uniref100.fasta protein sequence set from the uniprot database, then establish an index library of known protein sequences, compare the known protein sequences with the whole genome protein sequence allp.fasta that needs to be compared, and according to the results, select the sequence IDs with an identical matching percentage of 100.0, and then download the corresponding protein structure from the AlphaFoldProtein Structure Database.
[0019] When using protein structure prediction tools to perform structure prediction in step (3), the number of predictions generated by each model is 5.
[0020] The specific operation of step (4) is as follows: the obtained protein structure files are renamed to the corresponding protein names, and collected and organized into a data set as the structural data set of the whole genome protein; and the plddts value in the ranking_debug.json file of each protein is used as the evaluation indicator of its prediction accuracy, and the protein structure with a plddts value lower than 50 is marked.
[0021] When using US-align software to align protein structures, the alignment results are stored in the result file, where each line in the output file is the alignment result of a pair of proteins, including #PDBchain1 (the name of the first protein in the protein pair), PDBchain2 (the name of the second protein in the protein pair), TM1 (TM-score normalized according to the length of the first protein), TM2 (TM-score normalized according to the length of the second protein), RMSD (root mean square deviation of the two protein structures), etc., and the formula Calculate the characteristic value VD of the similarity between the two proteins in each row. The closer the VD value is to zero, the closer the structures of the two proteins are. Generally, when the VD value is less than 0.85, the two protein structures are considered to be relatively close. The TM-score value is in the (0, 1] interval, where 1 indicates a perfect match of the structures. When TM-score ≥ 0.5, it means that the protein structures share the same global topology.
[0022] According to the TM1, TM2 and RMSD values in the output result file of US-align calculation and the VD values calculated based on these three values, the matched protein structures are sorted respectively, and the selection is made based on this. The selection criteria are as follows: when there is a protein pair with a VD value less than 0.85, the protein obtained from the protein pair corresponding to the smallest VD value is used as the main screening result, and the protein with a difference between the VD value and the smallest VD value less than 0.1 is used as the candidate protein; when the VD values of all protein pairs are greater than 0.85, check whether there is a protein pair with TM1 or TM2 greater than 0.6 and RMSD value less than 5. If so, select up to 5 candidate proteins from small to large according to the RMSD value. If not, it is considered that there is no protein with similar structure in the species.
[0023] RMSD stands for root mean square deviation, which is a metric used to quantify the difference between two protein structures. By calculating the average distance between atoms (usually backbone atoms) between two superimposed proteins, RMSD can reflect the similarity of the structures. The lower the RMSD value, the higher the similarity between the two structures, while the higher the RMSD value means that there are significant differences between the two.
[0024] TM-score is a metric used to measure the similarity between two protein structures. It is used in bioinformatics to compare the global similarity of full-length protein structures. It aims to provide a more accurate measure of similarity than the commonly used RMSD measurement. The value of TM-score ranges from 0 to 1, where 1 indicates that the two structures match exactly. In general, structures with a TM-score below 0.20 are considered to be randomly selected unrelated proteins, while those above 0.5 are assumed to have roughly the same folding structure.
[0025] The above step (2) is more specifically as follows: download the uniref100.fasta protein sequence set from uniprot, use the diamond makedb--in uniref100.fasta-d uniref100 command to establish a known protein sequence index library, and for the whole genome protein sequence allp.fasta that needs to be compared, use the diamondblastp--db uniref100--query allp.fasta--out allp.tab--outfmt 6--max-target-seqs 1--evalue 1e-5--id30--block-size 20.0--index-chunks 1 command to compare the known protein sequences. According to the results, select the id of the sequence with an identical matching percentage (pident) of 100.0, and use the command "wget https: / / alphafold.ebi.ac.uk / files / AF-″+id+″-F1-model_v4.pdb" to download the corresponding protein structure in the AlphaFold Protein Structure Database.
[0026] A more specific step of the above step (3) is as follows: For other proteins that are not 100 matched and are not in the AlphaFold Protein Structure Database, the locally installed AlphaFold software is used for local protein structure prediction. The specific command is "run_2.2.sh-d${datadir}-o${output_dir}-f${FASTAFILE}-t 2021-11-01-e 5", where data_dir is the installed support data directory, output_dir is the output data directory, FASTAFILE is the protein sequence FASTA format file, the template date format is 2021-11-01, and the number of predictions generated by each model is 5.
[0027] Advantages of the present invention:
[0028] The present invention innovatively combines protein sequence alignment, the use of the existing AlphaFold structural database, and the locally installed AlphaFold software structure prediction and structural similarity search, providing an efficient and accurate new approach for protein function annotation and identification. Specifically,
[0029] (1) Combining sequence alignment and structure prediction: By first aligning the whole genome protein sequence with the Uniprot database, this method uses known information to accelerate the process of collecting protein structures. For proteins for which no existing structure is found, the locally installed AlphaFold software is used for structure prediction, ensuring that all target proteins have structural information available for subsequent analysis.
[0030] (2) Construction of a universally applicable structural dataset: By combining the publicly available AlphaFold structure and local predictions, this method creates a comprehensive protein structure dataset, which provides a basis for in-depth structure-function analysis.
[0031] (3) Structural similarity search: Using the USalign tool to align the entire protein based on the structure of a known functional protein or the local structure of its active center is an innovative strategy. This method not only takes into account global structural similarity, but also allows for detailed searches of functionally critical local domains, thereby improving the accuracy of discovering proteins with similar functions.
[0032] (4) Efficient screening mechanism: By screening based on the alignment score, this method can effectively identify proteins that are structurally similar to the target functional protein from a large number of proteins, especially those with large sequence differences, thereby solving the limitations of traditional methods in the accuracy of functional annotation.
[0033] Therefore, this method overcomes the limitation of existing technologies when sequence homology is low by utilizing the 3D structural information of proteins, rather than just sequence information. Through structural similarity search, functionally related proteins can be accurately identified even between proteins with different sequences but similar structures and functions. This method not only improves the accuracy of functional prediction, but also expands the range of annotated proteins, especially for those proteins that are difficult to accurately annotate using traditional sequence alignment-based methods.
[0034] In summary, this method effectively solves the limitations of existing technologies in protein function annotation through an innovative strategy that combines sequence alignment, structure prediction and structural similarity search, especially showing its unique advantages when dealing with proteins with large sequence differences but potentially similar structures and functions. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 It is a flow chart of the main body of the method of the present invention;
[0036] Figure 2 It is a diagram showing the sequence similarity intervals between the target protein and the four proteins obtained by screening;
[0037] Figure 3It is a diagram showing the structural similarity between the target protein and the four proteins obtained by screening;
[0038] Figure 4 It is an expression vector that verifies the activity of the protein screened;
[0039] Figure 5 It is the hydrolysate of the four expressed proteins on agar powder. DETAILED DESCRIPTION
[0040] The present invention will be further described below in conjunction with specific examples, and the advantages and features of the present invention will become clearer as the description proceeds. However, the specific experimental methods involved in the following examples are all conventional methods or implemented under the conditions recommended by the manufacturer's instructions unless otherwise specified.
[0041] Unless otherwise specified, the technical means used in the examples are conventional means known to those skilled in the art. The test methods in the following examples are conventional methods unless otherwise specified. Unless otherwise specified, the reagents and materials used can all be purchased from the market.
[0042] Unless otherwise defined, all professional and scientific terms used herein have the same meanings as those familiar to those skilled in the art. In addition, any methods and materials similar or equivalent to those described herein may be applied to the present invention. The preferred implementation methods and materials described herein are for demonstration purposes only.
[0043] The method uses the whole genome protein structure predicted by AlphaFold to mine enzymes with target functions in unknown functional proteins and proteins with huge sequence differences from known target proteins but consistent structure and function. The main flow chart of the method is as follows Figure 1 As shown, the specific implementation process is shown below by taking the search for glycoside hydrolase capable of hydrolyzing agarose in Agarivorans ablus strain JK6 isolated from the waters near the Mariana Trench as an example.
[0044] 1. Genome sequencing, assembly and protein annotation of JK6 strain
[0045] The purified cultured JK6 strain was collected, genomic DNA was extracted, and sent to Beijing Novogene for second-generation genome sequencing. The sequencing results were quality filtered using Fastp software, assembled using SPAdes-3.15.2-Linux software, and annotated using Prokka software to obtain the protein sequence file jk6.faa.
[0046] 2. Compare the uniprot database of the whole protein of JK6 strain and download the corresponding predicted structure
[0047] The diamond blastp --db uniref100 --query jk6.faa --out jk6.tab --outfmt6 --max-target-seqs 1 --evalue 1e-5 --id 30 --block-size 20.0 --index-chunks 1 command was used to perform local comparison on the uniprot database, and the comparison result file jk6.tab was obtained. The percentage of identical matches in the third column of the result file (100.0) was used as the screening condition to obtain the corresponding relationship between the protein ID in the JK6 strain and the protein ID with a known similarity of 100. The command "wget https: / / alphafold.ebi.ac.uk / files / AF-"+id+"-F1-model_v4.pdb "Download the corresponding protein structure from the AlphaFold Protein Structure Database.
[0048] 3. Alphafold structure prediction for the remaining proteins
[0049] For other proteins that are not 100 matches and are not in the AlphaFold Protein Structure Database, after removing proteins with a length of less than 50 amino acids (the length of the active domain of glycoside hydrolase protein is much longer than this value), use the AlphaFold software installed on the computer to perform local protein structure prediction. The specific command is "run_2.2.sh-d${data_dir}-o${output_dir}-f${FASTAFILE}-t2021-11-01-e 5", where data_dir is the installed support data directory, output_dir is the output data directory, FASTAFILE is the protein sequence FASTA format file, the template date format is 2021-11-01, and the number of predictions generated by each model is 5.
[0050] 4. Collection and organization of protein structure database of JK6 strain
[0051] For the structure files downloaded directly from the AlphaFold Protein Structure Database, the file names were changed to the corresponding protein IDs in the JK6 strain. The ranked_0.pdb files obtained from the results of each sequence predicted locally by the AlphaFold software were renamed to the corresponding protein IDs, and the data sets were collected and organized as the structural data sets of the whole genome protein. The plddts value in the ranking_debug.json file of each protein was used as the evaluation index of its prediction accuracy, and the protein structures with plddts values lower than 50 were annotated, and such proteins were specially reviewed and processed in the subsequent use.
[0052] 5. Comparison of the whole protein structure database with the reported resolved structure 4BQ2 as the search target
[0053] Download the structure file 4BQ2.pdb of the GH50 family exo-β-agarase Aga50D, which has a crystal structure, from the RCSB PDB database. According to the reported protein function, extract its glycoside hydrolase domain and store it as 4bq2gh.pdb. Use US-align software and the command "USalign 4bq2gh.pdb-dir2 / pdbs / / list.txt-outfmt 2>allp.4bq2ghresult" to perform a structural alignment of the whole genome protein. Among them, pdbs is the folder for the whole protein structure data, list.txt is the file list of the whole protein structure, the output format is 2, and the output file is allp.4bq2ghresult.
[0054] 6. Filter and display the results
[0055] According to the TM1, TM2 and RMSD values in the US-align calculation output result file allp.4bq2ghresult and the VD values calculated based on these three values, the matched protein structures are sorted and selected. The selection criteria are as follows: when there is a protein pair with a VD value less than 0.85, the protein obtained from the protein pair corresponding to the smallest VD value is used as the main screening result, and the protein with a difference between the VD value and the smallest VD value less than 0.1 is used as the candidate protein; when the VD value of all protein pairs is greater than 0.85, check whether there is a protein pair with TM1 or TM2 greater than 0.6 and an RMSD value less than 5. If so, select up to 5 candidate proteins from small to large according to the RMSD value. If not, it is considered that there is no protein with similar structure in the species. According to the difference in the value sorting, 4 proteins with extremely high similarity to 4bq2gh.pdb are obtained. Their score values are shown in Table 1. The intuitive display of the four proteins and 4bq2gh.pdb is shown in Figure 3 .
[0056] Table 1 Sequence similarity values between the target protein and the four proteins screened
[0057]
[0058]
[0059] Table 2 Predicted sequence expression verification primers
[0060]
[0061] Table 3 Similarity characteristic values of the target protein and the four proteins screened
[0062]
[0063]
[0064] VII. Protein expression and enzyme activity verification
[0065] The four protein-related genes from the genomic DNA of the JK6 strain were amplified by PCR. The primer sequences used for cloning are listed in Table 2. Each amplified DNA fragment was inserted into the EcoRI-XhoI site of pPEI-His-Sumo. Each expression plasmid (see Figure 4Escherichia coli BL21 (DE3) cells (Vazyme) were cultured in 2L LB medium containing 50μg / mL kanamycin at 37°C and induced at 16°C for 16 hours with 1mM isopropyl β-D-1-thiogalactoside (A100487, Sangon Biotech) when OD600 reached 0.8. The cells were collected by centrifugation at 4000g for 10 minutes at 4°C. The cell pellet was resuspended in 80mL lysis buffer (20mM Tris-HCl pH 8.0, 500mM NaCl, 5mM imidazole) and supplemented with 1mM phenylmethylsulfonyl fluoride (PMSF, A100754, Sangon Biotech). Then, the cells were perturbed for 3 minutes using a high pressure homogenizer (ATSEngineering Limited, AH1500) at 800bar, and the extracted protein was separated from the cell debris by centrifugation at 12000g for 1 hour at 4°C. The His-tagged protein was purified using a fast protein purification system (AKTA Pure 25) through a HisTrap HP column (5 mL, Cytiva) and eluted with an elution buffer (20 mM HEPES pH 7.5, 300 mM imidazole, 100 mM KCl). The protein was further enriched by a HiTrap Q HP column (5 mL, Cytiva) and separated by gradient elution with buffer A (20 mM HEPES pH 7.5, 50 mM KCl, 1 mM DTT) and buffer B (20 mM HEPES pH 7.5, 1 M KCl, 1 mM DTT). The peak fraction containing the agarase protein was collected and concentrated to about 10 mg / mL for further experiments.
[0066] Take 20ul of the expressed and purified enzyme sample, add 50uL of PBS buffer (pH=7.4) containing excess agar powder and digest at room temperature for 1h. Centrifuge at 12000rpm to take the supernatant, spot 5ul of the sample in batches on a silica gel 60 plate (Merck Millipore), and then air dry and chromatograph. The chromatographic solution is 1-butanol / acetic acid / distilled water (2:1:1, v / v / v), then spray with 10% sulfuric acid / ethanol solution, and heat in a microwave oven for 30s to develop color. See the figure of hydrolyzed agar powder. Figure 5 .
[0067] From Table 2, Figure 2 , Table 3 and Figure 3From the comparison, it can be seen that for sequences with great sequence differences (similarity less than 60%), the structures show extremely high similarity (RSMD value less than 1.8), and the verification of protein expression activity shows that the four proteins do show agarase activity. It can be seen that the method of the present invention can be used to quickly find proteins with specific functions in organisms, and is reliable. Therefore, it has high application value and is worthy of promotion.
[0068] The embodiments described above are only preferred embodiments of the present invention and are only used to explain the present invention, not to limit the scope of implementation of the present invention. For those skilled in the art, other implementation methods can certainly be easily made by replacement or modification based on the technical contents disclosed in this specification. Therefore, all changes and improvements made on the principles of the present invention should be included in the scope of the patent application of the present invention.
Claims
1. A method for searching for proteins with specific functions in organisms, characterized in that: The method comprises the following steps: (1) collecting and extracting DNA of the organism, sequencing and assembling the genome, and annotating the assembly results with proteins; (2) Using the uniprot database to perform sequence alignment on the obtained protein sequences to identify sequences that match known protein sequences; downloading the known protein structures corresponding to the matching sequences from the AlphaFold Protein Structure Database; (3) For proteins that did not fully match the database, the locally installed AlphaFold software was used to predict the structure of the protein sequence; (4) organizing and renaming the obtained protein structure files to construct a protein structure database corresponding to the biological protein name; (5) using US-align software to perform a full protein structure alignment on the new protein structure database constructed in step (4) to identify proteins with similar structures to the specific target protein; (6) based on the structural comparison results, selecting a protein having a structure similar to the target protein; (7) Expression, purification and functional verification of the selected proteins; When using US-align software to align protein structures, the alignment results are stored in the result file, where each line in the output file is the alignment result of a pair of proteins, including #PDBchain1, PDBchain2, TM1, TM2, RMSD value, and the formula Calculate the characteristic value VD of the similarity between the two proteins in each row, where the closer the VD value is to zero, the closer the structures between the two proteins are; where #PDBchain1 is the name of the first protein in the protein pair, PDBchain2 is the name of the second protein in the protein pair, TM1 is the TM-score normalized according to the length of the first protein, TM2 is the TM-score normalized according to the length of the second protein, and RMSD is the root mean square deviation of the two protein structures; where the TM-score value is in the interval (0, 1], where 1 indicates a perfect match in structure, and when TM-score ≥ 0.5, it means that the protein structures share the same global topology; According to the TM1, TM2 and RMSD values in the output result file calculated by US-align software and the VD values calculated based on these three values, the matched protein structures are sorted respectively, and the selection is made based on this. The selection criteria are as follows: when there is a protein pair with a VD value less than 0.85, the protein obtained from the protein pair corresponding to the smallest VD value is used as the main screening result, and the protein with a difference between the VD value and the smallest VD value less than 0.1 is used as the candidate protein; when the VD values of all protein pairs are greater than 0.85, check whether there is a protein pair with TM1 or TM2 greater than 0.6 and RMSD value less than 5. If so, select up to 5 candidate proteins from small to large according to the RMSD value. If not, it is considered that there is no protein with a structure similar to the specific target protein in the organism.
2. The method according to claim 1, characterized in that The specific operation of step (2) is as follows: download the uniref100.fasta protein sequence set from the uniprot database, then establish an index library of known protein sequences, compare the known protein sequences with the whole genome protein sequence allp.fasta that needs to be compared, and according to the results, select the names of the sequences with an identical matching percentage of 100.0, and then download the corresponding protein structures in the AlphaFold Protein Structure Database.
3. The method according to claim 1, characterized in that When using protein structure prediction tools to perform structure prediction in step (3), the number of predictions generated by each model is 5.
4. The method according to claim 1, characterized in that: The specific operation of step (4) is as follows: the obtained protein structure files are renamed to the corresponding protein names, and collected and organized into a data set as the structural data set of the whole genome protein; and the plddt value in the ranking_debug.json file of each predicted protein is used as the evaluation indicator of its prediction accuracy, and the protein structure with a plddt value lower than 50 is marked.
