Intelligent screening and identification method and system for umami peptides of microbial origin

Abstract

The present application belongs to the field of functional microorganism product screening and food biotechnology, and particularly relates to an intelligent screening and identification method and system for umami peptides of microbial origin. The method comprises the following steps: preparing a training set, virtual screening, constructing a receptor model, batch molecular docking, and intelligent screening and identification. That is, first, the taste activity is predicted by machine learning tools, then the appropriate receptor subunit is selected according to the length of the peptide chain, and finally, the umami peptides are screened quickly and effectively through batch molecular docking energy screening and core binding site matching. Compared with the traditional sensory-oriented chemical synthesis identification method, the present application is more green, safe and efficient, and lays a solid foundation for further establishing a deep exploration platform and a safety evaluation system for new functional components of microbial origin.

Description

Technical Field

[0001] This invention belongs to the field of functional microbial product screening and food biotechnology, specifically relating to an intelligent screening and identification method and system for microbial-derived umami peptides. Background Technology

[0002] Umami peptides are a class of small-molecule polypeptide compounds possessing umami flavor properties. They can be extracted from natural sources such as yeast, seafood, and mushrooms, or synthesized artificially. As a green and safe novel food additive, umami peptides primarily enhance the umami flavor of food and improve its overall taste. In recent years, they have begun to be used in food processing, such as in soy sauce, fish sauce, and soup bases. They can also be used to prepare low-salt or salt-free seasonings to help people reduce sodium intake, thus greatly improving the quality and competitiveness of food products. Umami peptides may also possess various biological activities, such as antioxidant, antibacterial, and immune-regulating effects, and can be applied in health foods and pharmaceuticals. In short, the research and application of umami peptides has become a hot topic in the food industry and scientific community, with broad application prospects in deepening basic research on food flavor, improving food quality, and developing health foods.

[0003] In recent years, with the availability of massive amounts of data and the reduction of computing costs, machine learning computing tools based on artificial intelligence technology have developed rapidly. They can be used to independently acquire and integrate knowledge, learn from data, and have been applied to the screening and prediction of flavor-active peptides. Molecular docking technology is a more scientific and faster method than traditional umami peptide screening and identification methods.

[0004] Therefore, the identification of umami peptides requires methods with higher throughput, higher sensitivity, and higher accuracy. Peptidomics can efficiently obtain peptide sequences with potential biological activity, and applying this method to the screening of umami peptides is a future development trend. Rapid screening and evaluation methods for umami peptides also deserve further exploration, and virtual screening and molecular docking are widely considered high-throughput and rapid techniques that can be applied to screen umami peptides from identified peptide libraries. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and system for intelligent screening and identification of microbial umami peptides.

[0006] This invention is implemented by providing an intelligent screening and identification method for microbial-derived umami peptides, comprising the following steps: 1) Prepare training sets: Summarize known umami peptides and non-umami peptides of various sources, lengths and with clear thresholds to form positive and negative training sets respectively; 2) Virtual screening: Compare the positive and negative training sets and select a virtual screening tool. Use the selected virtual screening tool to perform batch screening on the positive training set to screen out representative umami peptides of different peptide chain lengths in the positive training set. 3) Receptor model construction: Obtain the structure of T1R1 / T1R3 receptor proteins, construct and select the optimal model; 4) Batch molecular docking: Construct three-dimensional structures for representative umami peptides screened in step 2), realize molecular docking of the three-dimensional structure of umami peptides with the homology model of T1R1 / T1R3 receptor protein, and output ligand-receptor protein complexes to obtain the relationship between docking energy and umami threshold and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins. 5) Based on the relationship between docking energy and umami threshold obtained in step 4) and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins, intelligent screening and identification of microbial umami peptides are performed.

[0007] Preferably, in step 1), the positive training set includes 50 umami peptides with a chain length of 2-10 and a defined threshold, and the negative training set includes 45 non-umami peptides with a prominent bitter taste and 5 sweet peptides.

[0008] Further optimization, in step 2), the selected virtual screening tools are: UMPED-FRL, a machine learning classifier based on feature representation learning, and Umami-YYDS, a prediction tool based on data augmentation, comparison algorithms, and model optimization.

[0009] Further preferred, in step 3), the receptor model construction includes the following steps: 301) The amino acid sequences of the T1R1 / T1R3 receptor protein were retrieved from UniProtKB. Homology modeling was performed using the ligand-binding region of mGluR1 as a template. The homology model was constructed using modeler 10.1 and then optimized and validated. 302) The structure of the T1R1 / T1R3 receptor protein was predicted using alphafold v2.2.4, the pTMscore was calculated, and the 3D model of the T1R1 / T1R3 receptor protein was rendered and visualized based on the residue confidence score. 303) Evaluate and compare the accuracy of the T1R1 / T1R3 receptor protein models constructed by the two methods in steps 301) and 302) and select the optimal receptor model.

[0010] Further optimization, in step 302), the parameters used for alphafold v2.2.4 are all default parameters, and the database versions used are pdb_mmcif and uniport as 2022-08-03, and the rest of the databases are default databases.

[0011] Further optimization, in step 4), the batch molecular docking includes the following steps: 401) The three-dimensional structures of representative umami peptides were constructed and optimized in UCSF Chimera; 402) The SailVina script software is used to call AutodockVina and openbabel tools in batches to achieve efficient molecular docking of the three-dimensional structure of representative umami peptides with the optimal receptor model of T1R1 / T1R3 receptor protein, and output the ligand-receptor protein complex and docking energy. 403) The binding pocket and central site of the T1R1 / T1R3 receptor protein were predicted using the neural network-based prediction tool Deepsite serve, with the docking box size X, Y, and Z all set to 29. 404) Discovery Studio visualize was used to visualize and analyze the optimal docking posture of ligand-receptor protein complexes of umami peptides of different peptide lengths, and the relationship between docking energy and umami threshold and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins were obtained.

[0012] This invention also provides an intelligent screening and identification system for microbial umami peptides, based on the above-mentioned intelligent screening and identification method for microbial umami peptides, comprising the following modules: The training set acquisition module is used to summarize umami peptides and non-umami peptides from various known sources, of various lengths and with clear thresholds to form positive and negative training sets respectively. The virtual screening module is used to compare and select a virtual screening tool by comparing the positive and negative training sets. The selected virtual screening tool is then used to perform batch screening on the positive training set to select representative umami peptides of different peptide chain lengths from the positive training set. The receptor model construction module is used to obtain the structure of the T1R1 / T1R3 receptor protein, construct and select the optimal model; The batch molecular docking module is used to construct three-dimensional structures of representative umami peptides screened in the virtual screening module, realize molecular docking of the three-dimensional structure of umami peptides with the homology model of T1R1 / T1R3 receptor proteins, and output ligand-receptor protein complexes, obtain the relationship between docking energy and umami threshold and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins. The intelligent screening and identification module is used to intelligently screen and identify microbial umami peptides based on the relationship between docking energy and umami threshold obtained from the batch molecular docking module and the core binding sites of different umami peptides on the T1R1 / T1R3 receptor proteins.

[0013] Compared with the prior art, the advantages of the present invention are as follows: This invention establishes a rapid screening system for umami peptides based on artificial intelligence. First, machine learning tools are used to predict taste activity. Then, appropriate receptor subunits are selected based on peptide chain length, followed by batch molecular docking energy screening and core binding site matching to quickly and effectively screen for umami peptides. Compared to traditional sensory-guided chemical synthesis identification methods, this method is greener, safer, and more efficient, laying a solid foundation for further establishing a platform for in-depth exploration of novel functional components from microbial sources and a safety evaluation system. Attached Figure Description

[0014] Figure 1 This is a technical roadmap of the present invention; Figure 2 This is a rendering of the T1R1 / T1R3 receptor protein model in Example 2 of the present invention.

[0015] Figure 3 This is a comparison of the T1R1 / T1R3 receptor protein model in Example 2 of the present invention. Detailed Implementation

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] As mentioned earlier, current peptide identification primarily employs mass spectrometry following stepwise chromatographic separation to determine the primary structure of the peptide. Traditional screening methods for umami peptides mainly involve extraction, ultrafiltration, chromatographic purification, synthetic characterization, and sensory evaluation. This process has the following drawbacks: 1. Traditional mass spectrometry requires multiple iterations of purification of samples before identification, and a certain number of peptides with the same molecular weight or charge number still exist in the system. This indicates that traditional sensory-guided mass spectrometry identification methods are not only time-consuming and labor-intensive, but also cause the loss of many target peptides due to multiple iterations of separation and chromatographic purification, thus hindering the direct characterization of flavor-active peptides.

[0018] 2. With the rapid development and widespread application of technologies such as peptidomics and computer-simulated digestion, a large number of peptide sequences have been identified, which may contain a large amount of umami peptide resources. However, it is difficult to achieve high-throughput screening by simply using traditional screening methods such as structure-activity relationship or evaluation of the frequency of bioactive fragments.

[0019] In summary, traditional methods for identifying peptides and screening umami peptides are inefficient, hindering the development and utilization of umami peptides. Therefore, in a typical embodiment of this invention, an intelligent screening and identification method for microbial-derived umami peptides is provided, which can greatly improve the screening efficiency of umami peptides. The technical route is as follows: Figure 1 As shown, specifically, the method includes: Prepare the training set: Summarize the known umami and non-umami peptides of various sources, lengths and with clear thresholds to form positive and negative training sets; Virtual screening: The peptide dataset is screened in batches using the UMPRD-FRL machine learning classifier based on feature representation learning and another prediction tool, Umami-YYDS, developed based on data augmentation, comparison algorithms, and model optimization.

[0020] Receptor model construction: The amino acid sequence of the protein receptor T1R1 / T1R3 (heterodimer) was retrieved from UniProtKB. Homology modeling was performed using the ligand-binding region (closed-open state) of mGluR1 as a template (PDB ID: 1EWK). The homology model was constructed using modeler 10.1 and optimized for validation. Simultaneously, the structure of the T1R1 / T1R3 receptor protein was predicted again using alphafold v2.2.4, with all parameters set to default. The database versions used were pdb_mmcif and uniport (2022-08-03), and the remaining databases were default. The optimal results were selected, pTMscores were calculated, and the 3D receptor model was visualized using UCSF Chimera software (v 1.16) based on residue confidence scores (pLDDT). The accuracy of the initial models constructed using the two methods was evaluated and compared using the SAVES server, and the optimal receptor model and peptide substrate were selected.

[0021] Batch molecular docking: The 3D structures of umami peptides were constructed and optimized in UCSF Chimera (v 1.15). SailVina scripting software was used to batch call AutodockVina and OpenBabel tools to achieve efficient molecular docking of peptides with receptor models, outputting the ligand-receptor complex and docking energy. The neural network-based prediction tool Deepsite serve was used to predict the binding pocket and central site of the umami receptors T1R1 / T1R3. The docking box size (X, Y, Z) was set to 29 to fully account for the interaction between the peptide and receptor. The docking process was considered semi-flexible (the peptide conformation is flexible, while the protein structure is rigid). Finally, Discovery Studio visualize was used to visualize and analyze the optimal docking posture to explore the relationship between docking energy and the umami threshold, and to summarize the core binding sites.

[0022] Based on the relationship between docking energy and umami threshold, as well as the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins, microbial umami peptides are intelligently screened and identified.

[0023] The following examples further illustrate the present invention, but do not constitute a limitation thereof. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0024] Example 1: Preparing the training set and virtual screening Methods: We used the UMPRD-FRL machine learning classifier based on feature representation learning and another prediction tool, Umami-YYDS, developed based on data augmentation, comparison algorithms, and model optimization, to perform batch screening of peptide datasets.

[0025] (2) Results Determination: The study first evaluated the performance of two different machine learning tools (UMPred-FRL and Umami-YYDS). Studies have shown that machine learning may not be accurate enough for peptides with chain lengths greater than 10. Therefore, this study randomly collected 50 peptides (2-10 peptides) with clearly defined thresholds from known flavor-active peptides. Simultaneously, 50 non-umami peptides were used as negative controls for the performance testing of the prediction tool. Furthermore, since the same peptide may exhibit multiple basic tastes, when selecting positive controls for umami peptides, peptides with a more prominent umami flavor were chosen based on relevant literature descriptions to minimize errors. Because bitter peptides and umami peptides have significantly different amino acid compositions, peptides with a prominent bitter flavor were selected as the main non-umami controls (45 peptides), and 5 sweet peptides were selected together as negative controls for umami flavor.

[0026] The prediction results are shown in Table 1. UMPred-FRL predicted 38 out of 50 known umami peptides with umami taste characteristics, while Umami-YYDS accurately predicted 42, indicating that both tools have high accuracy in positive result determination. Regarding the determination of non-umami peptide controls (Table 2), UMPred-FRL accurately predicted 43 known non-umami peptides, while Umami-YYDS accurately predicted 37. Particularly in the determination of sweet peptides, UMPred-FRL showed higher accuracy, which is related to the difference in their core algorithms. Thus, it can be seen that the two tools each have their own characteristics and advantages. Combining the results of the two machine learning assessments, the evaluation results for the 37 peptides were consistent, with 34 peptides being identified as umami peptides, achieving an accuracy rate of 91.9%. Although the combined prediction results may lose some true umami peptides, it is more rigorous and accurate than individual determinations. In conclusion, combining UMPred-FRL (based on feature representation learning) and Umami-YYDS (primarily based on a random forest model) can improve the reliability of machine learning screening.

[0027] Table 1. Positive validation of known umami peptides by UMPred-FRL and Umami-YYDS.

[0028] Table 2 shows the negative validation of UMPred-FRL and Umami-YYDS for known non-umami peptides.

[0029] Example 2: Receptor Model Construction (1) Methods: The amino acid sequences of the protein receptors T1R1 / T1R3 (heterodimer) were retrieved from UniProtKB. Homology modeling was performed using the ligand-binding region (closed-open state) of mGluR1 as a template (PDB ID: 1EWK). The homology model was constructed using modeler 10.1 and optimized for validation. Simultaneously, the structure of the T1R1 / T1R3 receptor protein was predicted again using alphafold v2.2.4. All parameters were default. The database versions used were pdb_mmcif and uniport (2022-08-03), and the rest were default databases. The optimal results were selected to calculate the pTMscore, and the 3D model of the receptor was rendered and visualized using UCSF Chimera software (v 1.16) based on the residue confidence score (pLDDT). The accuracy of the initial models constructed by the two methods was evaluated and compared using the SAVES server, and the optimal receptor model and peptide substrate were selected.

[0030] (2) Result determination: The final active conformation of the T1R1 / T1R3 receptor constructed based on homology modeling is basically consistent with the modeling results of existing studies, that is, the crystal structure of T1R1 is a closed conformation and T1R3 is an open conformation. At the same time, the structure of the umami peptide receptor T1R1 / T1R3 was predicted by AlphaFold2 based on artificial intelligence, and the receptor model was successfully constructed. From the confidence level of the structure predicted by AlphaFold2 ( Figure 2 The majority of amino acid residues were rendered in blue and light blue (pLDDT>70), indicating that the predicted structure was reasonable.

[0031] Figure 3 Figure a shows the structural comparison results of the two models, clearly demonstrating the differences between them. The subunit on the left is the T1R1 portion of the umami receptor, while the subunit on the right is the T1R3 portion. The model clearly shows the receptor-binding region of the T1R1 / T1R3 heterodimer, which is a heterodimer linked by a CRD. The two leaves are connected by a hinge, forming a cavity. This cavity is the main region where the ligand interacts with the taste receptor, namely the Venus flytrap domain (VFTD). Meanwhile, the Ramachandran diagram (…) Figure 3 b and Figure 3c) The analysis results show that all residues in the AI-modeled receptor model are located within reasonable regions, with 91.9% of the residues located in the most favorable region and 8.1% in the allowed region. Compared with the receptor structure constructed based on homology modeling, the number of amino acid residues located in unreasonable regions (red dots in the figure) is significantly reduced. Therefore, the research group will select the receptor model constructed based on artificial intelligence for subsequent molecular docking studies to further accurately screen umami peptides and explore binding sites and structure-activity relationships.

[0032] Example 3: Batch Molecular Docking (1) Methods: The three-dimensional structures of umami peptides were constructed and optimized in UCSF Chimera (v 1.15). SailVina scripting software was used to batch call AutodockVina and OpenBabel tools to achieve efficient molecular docking between peptides and receptor models, outputting the ligand-receptor complex and docking energy. The binding pocket and central site of the umami receptors T1R1 / T1R3 were predicted using the neural network-based prediction tool Deepsite serve. The docking box sizes X, Y, and Z were all set to 29 to fully account for the interaction between the peptide and the receptor. The docking process was considered semi-flexible (the peptide conformation is flexible, while the protein structure is rigid). Finally, Discovery Studio visualize was used to visualize and analyze the optimal docking posture to explore the relationship between docking energy and the umami threshold, and to summarize the core binding sites.

[0033] (2) Result determination: Studies have shown that T1R1 / T1R3 is the only known heterodimer that reacts with glutamate and umami peptides. However, due to the different structures of umami peptides, each taste pattern may affect receptor cells through different mechanisms, and the binding sites of umami peptides are not yet clear. Therefore, it is necessary to study the docking energy relationship and core binding sites between umami peptides and umami receptors.

[0034] ① Correlation analysis between threshold and docking energy: First, representative umami peptides were screened from a known dataset of flavor-active peptides as research subjects for batch molecular docking. The screening requirements included: 1. having a clear umami threshold; 2. to avoid differences caused by different sensory evaluators and batches, peptides evaluated by the same sensory evaluation group should be selected as much as possible; 3. covering various peptide lengths and divided into four categories according to peptide chain length: short peptides (2-4 peptides), medium-length peptides (5-7 peptides), long peptides (8-10 peptides), and others (more than 10 peptides). Finally, 5 short peptides, 9 medium-length peptides, 7 long peptides, and 7 other peptides were selected, which were derived from synthetic peptides, clams, and anchovy sauce, respectively. The correlation analysis results are shown in Table 3. It was found that most thresholds were not significantly correlated with the docking energy of the two receptor monomers, but for medium-length peptides (5-7 peptides), the change in the threshold was related to the docking energy of the T1R3 subunit. P <0.05).

[0035] Meanwhile, analysis of docking results for peptides of different chain lengths revealed that the docking energy for the T1R1 receptor subunit was lower when docking oligopeptides (2-4 amino acids). As the peptide chain length increased, the peptide may have a greater tendency to bind to the T1R3 subunit. This is likely because the T1R3 subunit possesses an open functional domain (VFTD), providing a sufficiently large cavity to bind longer peptide chains. Furthermore, some longer peptides, such as AGAGPTP, TETKTFTLK, and NALKSVECYDAR, exhibited relatively lower docking energies. This may be due to the increased number of side chain groups, leading to a higher probability of binding to the receptor's active site. Additionally, longer peptides may be more flexible, resulting in shorter docking distances with the receptor. In conclusion, when using docking energy to assess and predict whether a peptide possesses umami flavor, it is necessary to classify the peptide chain length and then analyze the docking results for the corresponding appropriate receptor subunit.

[0036] Table 3 Correlation analysis between docking energy and threshold

[0037] ② Screening of core binding sites: Based on the results of batch molecular docking, the core binding sites of umami peptides were further summarized. The docking frequency was the sum of the number of effective forces (hydrogen bonds, electrostatic interactions, hydrophobic interactions, etc.) at each binding site. The core binding sites of the T1R1 receptor subunit are shown in Table 4. The table summarizes the top three core binding sites of each peptide chain length in terms of docking frequency. Among them, the core binding sites of short peptide T1R1 receptor subunits are mainly concentrated in ARG 249 and TYR 192, etc.; the core binding sites of medium-length peptides are mainly concentrated in LEU 23, GLU 273, ASP 80 and PHE 219, etc.; the core binding sites of long peptides are mainly concentrated in ARG 123, ARG 279, HIS 280 and ASP 80, etc.; and the core binding site of other longer peptide segments is LEU 23.

[0038] Table 4. Core binding sites of T1R1 receptor subunits

[0039] Table 5 shows the core binding sites of the T1R3 receptor subunits. For short peptides, the core binding sites are mainly concentrated at TYR 198, ALA 282, and SER 150; for medium-length peptides, they are mainly concentrated at TYR 198, HIS258, HIS 125, and ALA 282; for long peptides, they are mainly concentrated at HIS 258, HIS 125, and ALA 282; and for other even longer peptides, the core binding site is HIS 258. These core binding sites can serve as one of the criteria for our peptide set screening, establishing and improving the entire high-throughput safety screening system for microbial-derived umami peptides.

[0040] Table 5. Core binding sites of T1R3 receptor subunits

[0041] In summary, combining peptidomics and virtual screening techniques can effectively obtain active peptides. Therefore, applying this method to the screening of umami peptides will be a future development direction.

[0042] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of them. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention. Although the specific embodiments of the present invention have been described above, they are not intended to limit the protection scope of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A method for intelligent screening and identification of microbial umami peptides, characterized in that, The steps include the following: 1) Prepare training sets: Summarize known umami peptides and non-umami peptides of various sources, lengths and with clear thresholds to form positive and negative training sets respectively; 2) Virtual screening: Compare the positive and negative training sets and select a virtual screening tool. Use the selected virtual screening tool to perform batch screening on the positive training set to screen out representative umami peptides of different peptide chain lengths in the positive training set. 3) Receptor model construction: Obtain the structure of T1R1 / T1R3 receptor proteins, construct and select the optimal model; 4) Batch molecular docking: Construct three-dimensional structures for representative umami peptides screened in step 2), realize molecular docking of the three-dimensional structure of umami peptides with the homology model of T1R1 / T1R3 receptor protein, and output ligand-receptor protein complexes to obtain the relationship between docking energy and umami threshold and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins. 5) Based on the relationship between docking energy and umami threshold obtained in step 4) and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins, intelligent screening and identification of microbial umami peptides are performed.

2. The intelligent screening and identification method for microbial-derived umami peptides according to claim 1, characterized in that, In step 1), the positive training set includes 50 umami peptides with a chain length of 2-10 and a defined threshold, and the negative training set includes 45 non-umami peptides with a prominent bitter taste and 5 sweet peptides.

3. The intelligent screening and identification method for microbial umami peptides according to claim 1, characterized in that, In step 2), the selected virtual screening tools are: UMPred-FRL, a machine learning classifier based on feature representation learning, and Umami-YYDS, a prediction tool based on data augmentation, comparison algorithms, and model optimization.

4. The intelligent screening and identification method for microbial-derived umami peptides according to claim 1, characterized in that, Step 3) involves the following steps in constructing the receptor model: 301) The amino acid sequences of the T1R1 / T1R3 receptor protein were retrieved from UniProtKB. Homology modeling was performed using the ligand-binding region of mGluR1 as a template. The homology model was constructed using modeler 10.1 and then optimized and validated. 302) The structure of the T1R1 / T1R3 receptor protein was predicted using alphafold v2.2.4, the pTMscore was calculated, and the 3D model of the T1R1 / T1R3 receptor protein was rendered and visualized based on the residue confidence score. 303) Evaluate and compare the accuracy of the T1R1 / T1R3 receptor protein models constructed by the two methods in steps 301) and 302), and select the optimal receptor model.

5. The intelligent screening and identification method for microbial-derived umami peptides according to claim 1, characterized in that, In step 302), all parameters used in alphafold v2.2.4 are default parameters, and the database versions used are pdb_mmcif and uniport as 2022-08-03, while the other databases are default databases.

6. The intelligent screening and identification method for microbial-derived umami peptides according to claim 1, characterized in that, Step 4) involves the following steps for batch molecular docking: 401) The three-dimensional structures of representative umami peptides were constructed and optimized in UCSF Chimera; 402) The SailVina script software is used to call the Autodock Vina and openbabel tools in batches to achieve efficient molecular docking of the three-dimensional structure of representative umami peptides with the optimal receptor model of T1R1 / T1R3 receptor protein, and output the ligand-receptor protein complex and docking energy. 403) The binding pocket and central site of the T1R1 / T1R3 receptor protein were predicted using the neural network-based prediction tool Deepsite serve, with the docking box size X, Y, and Z all set to 29. 404) Discovery Studio visualize was used to visualize and analyze the optimal docking posture of ligand-receptor protein complexes of umami peptides of different peptide lengths, and the relationship between docking energy and umami threshold and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins were obtained.

7. A smart screening and identification system for microbial-derived umami peptides, characterized in that, The intelligent screening and identification method for microbial umami peptides according to claim 1 includes the following modules: The training set acquisition module is used to summarize umami peptides and non-umami peptides from various known sources, of various lengths and with clear thresholds to form positive and negative training sets respectively. The virtual screening module is used to compare and select a virtual screening tool by comparing the positive and negative training sets. The selected virtual screening tool is then used to perform batch screening on the positive training set to select representative umami peptides of different peptide chain lengths from the positive training set. The receptor model construction module is used to obtain the structure of the T1R1 / T1R3 receptor protein, construct and select the optimal model; The batch molecular docking module is used to construct three-dimensional structures of representative umami peptides screened in the virtual screening module, realize molecular docking of the three-dimensional structure of umami peptides with the homology model of T1R1 / T1R3 receptor proteins, and output ligand-receptor protein complexes, obtain the relationship between docking energy and umami threshold and the core binding sites of different umami peptides on T1R1 / T1R3 receptor proteins. The intelligent screening and identification module is used to intelligently screen and identify microbial umami peptides based on the relationship between docking energy and umami threshold obtained from the batch molecular docking module and the core binding sites of different umami peptides on the T1R1 / T1R3 receptor proteins.

Images