Molecular modification method for glycosyltransferase based on artificial intelligence

Abstract

The invention provides a molecular modification method for glycosyltransferase based on artificial intelligence. The method comprises the following steps: acquiring description information of enzyme mutants to form a basic data set; performing physicochemical property analysis on the basic data set to obtain an analysis result set; calculating the quantized value of the conservative degree of the mutation site and the quantized value of the influence degree of different mutations on the sequence so as to construct a position specificity scoring matrix; extracting a position specificity scoring value and a frequency from the scoring matrix, determining a functional section, and performing weighting processing; performing three-dimensional structure prediction and energy calculation based on the three-dimensional structure reconstruction model of the mutant to obtain an energy value; calculating an energy change value between the mutant and the wild type; performing molecular docking and dynamic simulation analysis on the mutant protein to obtain quantitative description of an interaction mode and strength and dynamic simulation characteristic data; and constructing a model training set to train a machine learning model for predicting the change of the enzyme activity of the mutant.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for molecular modification of glycosyltransferase based on artificial intelligence. Background Art

[0002] With the rapid development of biotechnology, research in the field of enzyme engineering has become increasingly in-depth. As a class of enzymes that play a key role in biosynthesis and metabolism, the improvement of the functions and properties of glycosyltransferases is of great significance for improving the efficiency and output of biotechnology. However, glycosyltransferases in nature often cannot meet the requirements of high efficiency and stability for industrial applications. Molecular modification and design of existing glycosyltransferases is an important and efficient research direction for optimizing their catalytic performance.

[0003] At present, the molecular modification and design of glycosyltransferases mainly rely on directed evolution technology, which simulates the natural selection process, constructs a mutant library of glycosyltransferase genes through random mutation or recombination techniques, and then uses high-throughput screening methods to find mutants with optimized performance.

[0004] Although this method is simple to operate and can produce rich mutant diversity, the mutant library is generally very large, which leads to cumbersome and inefficient screening process. It also has a certain degree of blindness and it is difficult to accurately control the mutation direction and effect. Summary of the invention

[0005] In order to solve the above technical problems, the present application provides a method for molecular modification of glycosyltransferase based on artificial intelligence, so as to at least solve or alleviate the problems existing in the above-mentioned prior art.

[0006] A method for molecular modification of glycosyltransferase based on artificial intelligence, comprising: Obtaining description information of a glycosyltransferase mutant, wherein the description information includes a mutation site and a description of related experimental results; Preprocessing the description information of the glycosyltransferase mutant to obtain a mutant basic data set; Based on the constructed automated script, the physicochemical properties of the mutation sites are analyzed on the mutant basic data set to obtain a physicochemical properties analysis result set; Based on the original mutation site sequence data corresponding to the mutant basic data set, a similar sequence search is performed in a protein database to calculate a quantitative value of the conservation degree of each mutation site in the protein database and a quantitative value of the degree of influence of different mutations on the sequence; Based on the quantitative value of the conservation degree of each mutation site in the protein database and the quantitative value of the influence degree of different mutations on the sequence, constructing a position-specific scoring matrix; Based on the constructed automated script, extracting the position-specific score value and frequency corresponding to each mutation site from the position-specific scoring matrix; Determining the functional segment of the glycosyltransferase protein sequence in the glycosyltransferase mutant according to the position-specific score and frequency corresponding to each mutation site; Performing weighted processing on key functional points in the functional section to obtain a weighted information set of the functional section; Based on the three-dimensional structure reconstruction model of the glycosyltransferase mutant, the three-dimensional structure of the mutant glycosyltransferase selected from the mutant basic data set is predicted to obtain the predicted three-dimensional structure of the mutant glycosyltransferase; Performing energy calculation on the predicted three-dimensional structure of the mutant glycosyltransferase to obtain an energy value of the three-dimensional structure of the mutant glycosyltransferase; Calculating the difference between the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase to obtain an energy change value; Performing molecular docking on the mutant protein in the variant basic data set to obtain a molecular docking result, wherein the molecular docking result at least includes: a quantitative description of the interaction mode and strength between the mutant protein and the ligand; Performing kinetic simulation analysis on mutants in the variant basic data set to obtain kinetic simulation characteristic data; Based on the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation characteristic data, a model training set is constructed to train the selected machine learning model to obtain an enzyme activity prediction model; Based on the enzyme activity prediction model, predict the change in enzyme activity of the target glycosyltransferase relative to the wild-type mutant to obtain a predicted value of the mutant enzyme activity; Based on the predicted value of the mutant enzyme activity, the target glycosyltransferase is molecularly modified.

[0007] In the scheme of the present application, artificial intelligence-related means are used to first obtain the description information of the glycosyltransferase mutant and pre-process the mutant basic data set, and then perform multi-dimensional analysis and calculation based on this data set, such as building an automated script to analyze the physicochemical properties of the mutation site, searching the protein database to calculate the quantitative value of the degree of conservation and the quantitative value of the degree of influence, and a series of processes. These steps are not blindly screening a large number of mutants one by one as in the traditional way, but to mine and analyze the key features of the mutants from the data level, build a position-specific scoring matrix, etc., and then comprehensively build a model training set of multi-faceted data to train the enzyme activity prediction model. Through this model, the change in the enzyme activity of the target glycosyltransferase relative to the wild-type mutant can be directly predicted, and molecular modification can be performed based on the predicted value. In this way, the high-throughput screening work without clues in the extremely large mutation library is avoided, the efficiency of screening and modification is greatly improved, and the entire molecular modification process is more organized and efficient.

[0008] In addition, in this scheme, multiple links are set up to solve the blindness problem and realize accurate control of mutation direction and effect. On the one hand, by analyzing the physicochemical properties of the mutation site, the physicochemical properties of different mutation sites themselves can be known, which provides a basis for subsequent judgment of the possible impact on the enzyme molecule. For example, the quantitative value of the degree of conservation of each mutation site in the protein database and the quantitative value of the degree of influence of different mutations on the sequence are calculated, and a position-specific scoring matrix is ​​constructed. Based on this, it can be clear which sites are relatively conservative, which sites are more likely to bring functional changes, etc., and the key mutation sites and their importance are accurately located, providing a basis for subsequent determination of functional segments, etc., to avoid blindly considering all mutation sites. The functional segments of the glycosyltransferase protein sequence in the glycosyltransferase mutant are further determined, and the key functional sites in the functional segments are weighted, focusing on the key functional sites that have an important impact on the enzyme activity. Combined with the multi-dimensional data collection of mutant glycosyltransferase three-dimensional structure prediction, energy calculation, molecular docking, and kinetic simulation analysis, these detailed and targeted data are combined to build a model training set to train the enzyme activity prediction model, and finally the molecular modification is carried out based on the value of the mutant enzyme activity predicted by the model. The whole process is based on the scientific analysis of a large amount of data. It is no longer a blind generation of mutants, but can control the effect of mutations in a directional and well-founded way to develop in the expected direction of improving enzyme activity, achieving a more precise control of the direction and effect of mutations. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The drawings constituting a part of this application are used to provide a further understanding of this application, so that other features, purposes and advantages of this application become more obvious. The schematic embodiment drawings and their descriptions of this application are used to explain this application and do not constitute an improper limitation on this application. In the drawings: Figure 1 This is a flow chart of a method for molecular modification of glycosyltransferase based on artificial intelligence according to an embodiment of the present application.

[0010] Figure 2 This application provides a flow chart of another method for molecular modification of glycosyltransferase based on artificial intelligence.

[0011] Figure 3 This application also provides a flow chart of a method for molecular modification of glycosyltransferase based on artificial intelligence. DETAILED DESCRIPTION

[0012] In order to enable those skilled in the art to better understand the solution of the present application, the following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0013] Figure 1 This is a flow chart of a method for molecular modification of glycosyltransferase based on artificial intelligence according to an embodiment of the present application. Figure 1 As shown, a method for molecular modification of glycosyltransferase based on artificial intelligence includes: Obtaining description information of a glycosyltransferase mutant, wherein the description information includes a mutation site and a description of related experimental results; Preprocessing the description information of the glycosyltransferase mutant to obtain a mutant basic data set; Based on the constructed automated script, the physicochemical properties of the mutation sites are analyzed on the mutant basic data set to obtain a physicochemical properties analysis result set; Based on the original mutation site sequence data corresponding to the mutant basic data set, a similar sequence search is performed in a protein database to calculate a quantitative value of the conservation degree of each mutation site in the protein database and a quantitative value of the degree of influence of different mutations on the sequence; Based on the quantitative value of the conservation degree of each mutation site in the protein database and the quantitative value of the influence degree of different mutations on the sequence, constructing a position-specific scoring matrix; Based on the constructed automated script, extracting the position-specific score value and frequency corresponding to each mutation site from the position-specific scoring matrix; Determining the functional segment of the glycosyltransferase protein sequence in the glycosyltransferase mutant according to the position-specific score and frequency corresponding to each mutation site; Performing weighted processing on key functional points in the functional section to obtain a weighted information set of the functional section; Based on the three-dimensional structure reconstruction model of the glycosyltransferase mutant, the three-dimensional structure of the mutant glycosyltransferase selected from the mutant basic data set is predicted to obtain the predicted three-dimensional structure of the mutant glycosyltransferase; Performing energy calculation on the predicted three-dimensional structure of the mutant glycosyltransferase to obtain an energy value of the three-dimensional structure of the mutant glycosyltransferase; Calculating the difference between the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase to obtain an energy change value; Performing molecular docking on the mutant protein in the variant basic data set to obtain a molecular docking result, wherein the molecular docking result at least includes: a quantitative description of the interaction mode and strength between the mutant protein and the ligand; Performing kinetic simulation analysis on mutants in the variant basic data set to obtain kinetic simulation characteristic data; Based on the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation characteristic data, a model training set is constructed to train the selected machine learning model to obtain an enzyme activity prediction model; Based on the enzyme activity prediction model, predict the change in enzyme activity of the target glycosyltransferase relative to the wild-type mutant to obtain a predicted value of the mutant enzyme activity; Based on the predicted value of the mutant enzyme activity, the target glycosyltransferase is molecularly modified.

[0014] Optionally, the preprocessing of the description information of the glycosyltransferase mutant to obtain a mutant basic data set includes: Performing data filtering on the description information of the glycosyltransferase mutant to select data directly related to the mutation site and the change of enzyme activity; Based on the data directly related to the mutation site and the enzyme activity change screened out, multiple data sub-tables are constructed, each data sub-table corresponds to a protein mutant or a group of mutants with similar enzyme activity change characteristics; Cleaning the multiple sub-tables, removing the repeated data directly related to the mutation sites and enzyme activity changes between the multiple sub-tables, to obtain a cleaned data sub-table; The data in the cleaned data subtable is standardized to obtain a mutant basic data set.

[0015] Optionally, the constructed automated script performs a physicochemical property analysis of the mutation site on the mutant basic data set to obtain a physicochemical property analysis result set, including: Constructing a regular expression, wherein the regular expression is used to extract a mutation site representation from the mutant basic data set; Creating a first mapping dictionary, a second mapping dictionary, and the amino acid property dictionary, wherein the first mapping dictionary is used to map the three-letter abbreviations of the amino acids in the mutation site representation to single-letter representations, the second mapping dictionary is used to reversely map the single-letter representations of the amino acids to the three-letter representations of the amino acids in the mutation site table, and the amino acid property dictionary is used to define the description of the physicochemical properties of the amino acids; An amino acid mutation impact function is defined, wherein the amino acid mutation impact function is used to query the amino acid property dictionary to obtain the physicochemical property description of the amino acid in the mutation site table, and based on this, the physicochemical property change value of the mutation site is calculated to perform mutation site physicochemical property analysis on the mutant basic data set to obtain a physicochemical property analysis result set; Defining an amino acid representation conversion function to convert the representation of the amino acid in the mutation site table based on the first mapping dictionary and the second mapping dictionary, wherein the conversion includes: converting the three-letter representation of the amino acid into a single-letter representation and / or converting the single-letter representation of the amino acid into a three-letter representation; defining a mutation reading function, the mutation reading function being used to read the mutant basic data set to extract a mutation site representation from the mutant basic data set using the regular expression; The file storage path is set to set the file access path of the mutant basic data set, so as to call the defined mutation reading function to read the mutant basic data set.

[0016] Optionally, the constructed automated script performs a physicochemical property analysis of the mutation site on the mutant basic data set to obtain a physicochemical property analysis result set, including: Calling the defined mutation reading function to read the mutant basic data set from the file storage path; Calling the constructed regular expression to extract the mutation site representation from the mutant basic data set; Calling a defined amino acid representation conversion function to map the three-letter abbreviation of the amino acid in the mutation site representation to a single-letter representation based on the created first mapping dictionary; The defined amino acid mutation impact function is called to query the physicochemical property description of the single-letter representation of the amino acid in the mutation site representation in the constructed amino acid property dictionary, and the physicochemical property change value of the mutation site is calculated accordingly to obtain a physicochemical property analysis result set, wherein the physicochemical property change value includes at least one of the following: hydrophilicity (Hydrophobicity_Change), isoelectric point (pI_Change), molecular weight (MolecularWeight_Change), carboxyl dissociation constant (Ka_Carboxyl_Change), and amino dissociation constant (Ka_Amino_Change) of the amino acid.

[0017] Specifically, in one application scenario, an automated processing program is constructed in the following manner to implement physical and chemical property analysis of the mutation site on the mutant basic data set based on the constructed automated script to obtain a physical and chemical property analysis result set: (1) Import libraries: Import Python’s pandas library and re module for data processing and regular expression matching, respectively.

[0018] (2) Create a mapping dictionary: three_to_one_letter_mapping: maps the three-letter abbreviations of amino acids to their one-letter representations. one_to_three_letter_mapping: maps the one-letter representations back to the three-letter abbreviations.

[0019] (3) Define the amino acid property dictionary: amino_acids_properties, which includes the hydrophilicity, isoelectric point, molecular weight, and dissociation constants of the carboxyl and amino groups of amino acids.

[0020] (4) Calculate the impact of amino acid mutation: Function calculate_amino_acid_mutation: Input the original and mutated amino acids, calculate and return the changes in physical and chemical properties.

[0021] (5) Convert amino acid representation: single_letter_to_three_letter: Converts single-letter amino acids to three-letter abbreviations. convert_amino_acid_representation: Converts between three-letter abbreviations and single-letter representations as needed.

[0022] (6) Read mutation information: Function read_mutations_from_file: reads and parses amino acid mutation information from a file, using regular expression matching.

[0023] (7) File path setting: Specify the text file path containing mutation information.

[0024] (8) Read and store mutation information: call read_mutations_from_file and store the results into the list mutations.

[0025] (9) Initialize the result list: Create an empty list all_results to store the calculation results.

[0026] (10) Calculate and store mutation results: traverse mutations, calculate changes in physical and chemical properties, and store the results in all_results.

[0027] (11) Exception handling: If the amino acid is not in the dictionary, print a warning and skip.

[0028] (12) Result output: If the program is run as the main program, convert the results into a DataFrame and save it as a CSV file.

[0029] The automated process described above has the following technical benefits: (1) Modular design: The clear division of functions makes the code easy to understand and maintain.

[0030] (2) Data-driven: Amino acid properties are stored in a dictionary for easy expansion and updating.

[0031] (3) Flexibility: Supports reading mutation information from files, making it easier to process large amounts of data.

[0032] (4) Error handling: Warning for amino acids that are not in the dictionary to avoid program crashes.

[0033] Optionally, the searching for similar sequences in a protein database based on the original mutation site sequence data corresponding to the mutant basic data set to calculate the quantitative value of the conservation degree of each mutation site in the protein database and the quantitative value of the influence degree of different mutations on the sequence includes: The statistics of the amino acid occurrence frequency and score of each mutation site in the original mutation site sequence data corresponding to the mutant basic data set and the protein database are calculated, and based on this, the quantitative value of the conservation degree of each mutation site in the original mutation site sequence data and the quantitative value of the degree of influence of different mutations on the sequence are calculated.

[0034] Optionally, constructing a position-specific scoring matrix based on the quantified value of the conservation degree of each mutation site in the protein database and the quantified value of the influence degree of different mutations on the sequence comprises: A two-dimensional matrix is ​​created based on the amino acid class and the position corresponding to each mutation site; Based on the quantified value of the conservation degree of each mutation site in the protein database and the quantified value of the influence degree of different mutations on the sequence, the element values ​​in the two-dimensional matrix are calculated to construct a position-specific scoring matrix.

[0035] Optionally, determining the functional segment of the glycosyltransferase protein sequence in the glycosyltransferase mutant according to the position-specific score value and frequency corresponding to each mutation site includes: Based on the GTB family characteristics of glycosyltransferases and the fact that they have two domains, a UGT89C1 sequence with a heat value greater than a set heat threshold is screened from the protein database as a reference sequence; Screening the protein database for the top multiple protein sequences related to the glycosylation process; Comparing the reference sequence with the plurality of protein sequences to determine regions of similarity and difference between the sequences; Based on the similarities and differences between the sequences, protein sequences whose sequence lengths differ from other sequences by more than a set sequence difference are determined as abnormal sequences and removed to obtain valid sequences; Based on the two domains, the position-specific scoring values ​​and frequencies corresponding to each mutation site, the functional segments of the glycosyltransferase protein sequence in the glycosyltransferase mutant are determined.

[0036] Optionally, determining the functional segment of the glycosyltransferase protein sequence in the glycosyltransferase mutant based on the two domains, the position-specific score value and the frequency corresponding to each mutation site, comprises: Determining the segmentation ratio of the two domains based on biochemical experimental data; Matching the position-specific score corresponding to each mutation site with the segment division ratio to evaluate the importance and conservation of each mutation site in the sequence family; Based on the frequency corresponding to each mutation site, weights are assigned to the importance and conservation of each mutation site in the sequence family; Based on the weight allocation of the importance and conservation degree of each mutation site in the sequence family, typical mutation regions are determined to divide the glycosyltransferase protein sequence in the glycosyltransferase mutant into functional segments.

[0037] Optionally, the weighted processing of the key functional points in the functional section to obtain a functional section weighted information set includes: Counting the position-specific scores and frequencies of mutation sites within the functional segment to screen out the key functional sites; Assigning weighting factors based on the location-specific scores and frequencies of the key functional sites; The positions of the key functional points are weighted and settled based on the weighting factors to obtain a functional section weighted information set.

[0038] To this end, by performing relevant statistics on the original mutation site sequence data corresponding to the mutant basic data set in the protein database, the quantitative value of the degree of conservation of each mutation site and the quantitative value of the degree of influence of different mutations on the sequence are calculated, so that the importance and influence of each mutation site in the entire protein sequence system can be understood from a macro level. This helps to accurately grasp which sites are relatively stable and which changes are more likely to have a greater impact on the overall sequence, provide basic data support for targeted molecular modification and other work, avoid blindly treating all mutation sites "equally", and improve the scientificity and accuracy of the modification. In addition, a two-dimensional matrix is ​​created based on the amino acid category and mutation site position, and the matrix element value is determined based on the quantitative value of conservation and the quantitative value of influence calculated previously, so that the complex site-related information is presented in an intuitive matrix form. This facilitates the subsequent rapid review and comparison of relevant features under different sites and different amino acids, making the integration and utilization of data more efficient, and providing a clear data architecture for further in-depth analysis of the functional characteristics of glycosyltransferases. Furthermore, the matrix can clearly show the differences of each mutation site in different dimensions, which helps to dig out sites with special properties or important roles, and then focus on key links such as the subsequent determination of functional segments, and improve the accuracy of key site judgment in the entire molecular modification process. Finally, combined with the GTB family characteristics of glycosyltransferases, the UGT89C1 sequence with a heat value greater than the set threshold was screened as a reference sequence, and multiple protein sequences related to the glycosylation process were comprehensively compared and analyzed, taking into full account the family commonality and glycosylation, a key function-related factor. In this way, the similarity and difference regions between sequences can be more accurately found, and valid sequences can be obtained after removing abnormal sequences, so that the subsequent determination of functional segments based on these valid sequences is more reliable and more in line with the actual function of glycosyltransferases. Moreover, the functional segments are further determined based on the position-specific scoring values ​​and frequencies corresponding to the two domains and each mutation site, and the structural characteristics and site-related quantitative data are combined to analyze the protein sequence in multiple dimensions. It can accurately locate the areas that play a key role in the function of glycosyltransferases, which helps to focus on these functional segments during molecular modification, optimize them in a targeted manner, and improve the performance of the modified enzymes.

[0039] In addition, by first screening out key functional sites by statistically analyzing the position-specific scores and frequencies of mutation sites within the functional segments, it is possible to identify the sites that have a more prominent impact on enzyme function from many sites, avoid paying too much attention to insignificant sites, and make subsequent transformation work more targeted, concentrating resources on optimizing key positions. Based on the relevant numerical values ​​of key functional sites, weighted factors are assigned and weighted settlement is performed to obtain a weighted information set for the functional segment, which can highlight the relative importance differences of different key sites. When performing molecular transformation and other operations, a more reasonable transformation strategy can be formulated based on the weights, such as focusing on optimizing key sites with high weights, thereby more effectively improving the overall function and performance of glycosyltransferases and enhancing the predictability and controllability of the transformation effect. To this end, an automated program is constructed as follows to perform the above-mentioned technical processing of obtaining the functional segment weighted information set: (1) Import library: Import the pandas library, abbreviated as pd.

[0040] (2) Define the processing function: Define the process_pssm function, which receives the file path file_path and the threshold threshold.

[0041] (3) Define amino acid representation: Initialize the string amino_acids containing the single-letter representations of all amino acids.

[0042] (4) Initialize the result list: Create an empty list results to store the mutation results after screening.

[0043] (5) Read file data: Use the with statement to open the file, read all row data and store them into the pssm_data list, skipping the header row.

[0044] (6) Process PSSM data: Traverse each row in pssm_data (skip the header row). Use the split() method to split the row data and extract the position number, original amino acid, PSSM score, and mutant amino acid frequency.

[0045] (7) Screening mutations: Traverse each amino acid and its index in amino_acids. Obtain the PSSM score and frequency of the corresponding amino acid. If the score is higher than the threshold and the amino acid is different from the original amino acid, then consider the mutation.

[0046] (8) Build result dictionary: For each selected mutation, create a dictionary containing the key combination (original amino acid, position, mutant amino acid), PSSM score and frequency. Add the dictionary to the results list.

[0047] (9)Convert to DataFrame: Use pandas’ DataFrame to convert the results list into a DataFrame object.

[0048] (10) Save the results: Save the DataFrame as a CSV file named 'PSSM_Results.csv', excluding the index column.

[0049] (11) Set the threshold and call the function: In the last part of the code, set the threshold to 5 to ensure that the frequency is not 0, and call the process_pssm function, passing in the path of the PSSM file 'PSSM.TXT' and the threshold.

[0050] (12) Result output: After the function is executed, the results will be saved in CSV format, which contains all mutations above the threshold and their related information.

[0051] The automated procedure for implementing the technical process of the functional segment weighted information set described above has the following technical benefits: (1) Automated processing: Automatically read and process PSSM data from files to reduce manual operations.

[0052] (2) Flexible threshold setting: Users can set thresholds according to their needs and flexibly control the criteria for mutation screening.

[0053] Optionally, the three-dimensional structure reconstruction model based on the glycosyltransferase mutant is used to predict the three-dimensional structure of the mutant glycosyltransferase selected from the mutant basic data set to obtain the predicted three-dimensional structure of the mutant glycosyltransferase, including: The mutant basic data set is analyzed to screen out the mutation site position, the amino acid types before and after the mutation, and the corresponding glycosyltransferase sequence; The collected three-dimensional structure data of the wild-type glycosyltransferase is used as the initial prediction framework, and the mutation site position, the amino acid types before and after the mutation, and the corresponding glycosyltransferase sequence are input into the three-dimensional structure reconstruction model of the glycosyltransferase mutant to perform three-dimensional structure prediction to obtain the predicted mutant glycosyltransferase three-dimensional structure.

[0054] Optionally, performing energy calculation on the predicted three-dimensional structure of the mutant glycosyltransferase to obtain the energy value of the three-dimensional structure of the mutant glycosyltransferase comprises: An energy calculation task is created to input the predicted mutant glycosyltransferase three-dimensional structure into the constructed energy calculation protocol and analyze and calculate the atomic interactions and chemical bond energies in the mutant glycosyltransferase three-dimensional structure accordingly to obtain the energy value of the mutant glycosyltransferase three-dimensional structure.

[0055] Optionally, the calculating the difference between the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase to obtain the energy change value comprises: Based on the constructed energy change value calculation function, the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase are subjected to difference processing to obtain the energy change value.

[0056] To this end, by first carefully analyzing the mutant basic data set, we screened out key information such as the location of the mutation site, the type of amino acids before and after the mutation, and the corresponding glycosyltransferase sequence, and then used the three-dimensional structure data of the wild-type glycosyltransferase as the initial framework for prediction. This makes full use of the known wild-type structure and combines the specific mutation-related information to make the prediction of the three-dimensional structure of the mutant glycosyltransferase more in line with the actual situation, and can more accurately present the changes in the structure after the mutation, providing a reliable structural basis for subsequent in-depth research on the impact of mutations on enzyme function. In addition, instead of blindly predicting the structure, we focus on the elements closely related to the mutation extracted from the mutant basic data set, and input them into the three-dimensional structure reconstruction model in a targeted manner, avoiding the interference of irrelevant information, helping to improve the prediction efficiency and the availability of the prediction results for subsequent analysis, and better serve the overall process of glycosyltransferase molecular modification.

[0057] In terms of energy calculation, by creating an energy calculation task and inputting the predicted three-dimensional structure of the mutant glycosyltransferase into a specially constructed energy calculation protocol, it is possible to perform detailed analysis and calculation of its interatomic interactions and chemical bond energy to obtain energy values. This allows us to deeply understand the stability, activity and other characteristics of the mutant three-dimensional structure from an energy perspective, because the energy state is closely related to the function of the enzyme. Through energy calculation, we can dig out the manifestation of structural changes at the energy level and provide a quantitative basis for evaluating the impact of mutations. After obtaining the energy value of the three-dimensional structure of the mutant glycosyltransferase, we can further calculate the difference between its energy value and that of the wild-type glycosyltransferase (i.e., the energy change value), and intuitively see the energy change caused by the mutation. This comparison is critical for determining whether the mutation enhances or weakens the performance of the enzyme. For example, energy changes may affect the binding ability of the enzyme to the substrate, catalytic efficiency, etc., thereby helping to more scientifically evaluate the advantages and disadvantages of different mutation schemes and provide strong support for screening out better molecular transformation directions.

[0058] In the specific technical implementation, the initial prediction framework is constructed with the collected three-dimensional structural data of the wild-type glycosyltransferase. The source of this wild-type data should be high-quality data obtained through experimental measurement or reliable database acquisition. Then, the previously extracted mutation-related data is accurately input into the model according to the input format and specifications required by the three-dimensional structure reconstruction model. The model will simulate, construct and predict the three-dimensional structure of the mutant according to its preset algorithms and rules, such as molecular mechanics, molecular dynamics and other principles, and finally output the predicted three-dimensional structure of the mutant glycosyltransferase.

[0059] When calculating energy, we first create an energy calculation task, which involves determining the calculation target, scope, and calculation method used. Then we build an energy calculation protocol, which needs to comprehensively consider the various types of interactions between atoms in biological macromolecules such as glycosyltransferases (such as van der Waals forces, electrostatic interactions, etc.) and the different energy contributions of chemical bonds. By integrating physical and chemical principles and related mathematical models, we form a set of rules that can accurately calculate the energy of its three-dimensional structure.

[0060] During energy calculation and analysis, after the predicted three-dimensional structure of the mutant glycosyltransferase is input according to the requirements of the energy calculation protocol, the calculation protocol will analyze the atomic interactions and chemical bond energy in the structure in turn according to the set steps. It may apply relevant theories and algorithms such as quantum mechanics and classical mechanics for quantitative calculations, and finally summarize the energy value of the three-dimensional structure of the mutant glycosyltransferase.

[0061] When calculating the energy change value, an energy change value calculation function is constructed based on mathematical operation logic. The input parameters of this function are the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase. A simple subtraction operation is performed inside the function to perform difference processing on the two energy values, and finally an accurate energy change value is output, thereby achieving a quantitative presentation of the energy change caused by the mutation. The above process of calculating the energy change value can be implemented by, but not limited to, the rosetta tool.

[0062] Optionally, the molecular docking is performed on the mutant protein in the variant basic data set to obtain a molecular docking result, and the molecular docking result at least includes: a quantitative description of the interaction mode and strength between the mutant protein and the ligand, including: Screening out mutant proteins for molecular docking from the variant basic data set to form a protein list, wherein the protein list at least includes a file path and a name corresponding to each mutant protein; Opening the mutant protein structures and corresponding ligand structures in the protein list one by one to perform dehydration and hydrogenation operations on them to obtain processed mutant protein structure data; Performing structural transformation on the mutant protein structure data to add atomic charges and atomic types therein to obtain the mutant protein structure data to be docked; Obtaining set docking parameters, wherein the docking parameters define a spatial range for searching for a binding conformation of a ligand and a protein; The docking parameters were set to the central cavity of the protein; Based on the central cavity of the protein, simulating the orientation and position changes of the ligand around the mutant protein structure data to be docked, so as to calculate the interaction energy between the ligand and the mutant protein structure data to be docked and generate a plurality of different binding energy scores between the ligand and the mutant protein structure data to be docked; Determine the ligand corresponding to the highest binding energy score and the structure data of the mutant protein to be docked as the docking conformation; The binding energy, hydrogen bond and hydrophobic interaction force numbers of the docked conformation are calculated to obtain the molecular docking result.

[0063] To this end, mutant proteins for molecular docking are screened out from the variant basic data set to form a protein list, and key information such as the file path and name corresponding to each mutant protein is clearly listed. In this way, the objects to be analyzed can be selected in a targeted manner, avoiding indiscriminate processing of all data, so that subsequent molecular docking work can focus on the target mutant protein, improving the accuracy of the operation and the overall analysis efficiency. Dehydration and hydrogenation operations are performed on the mutant protein structure and the corresponding ligand structure to remove water molecules that may interfere with the docking analysis, and hydrogen atoms are added to make the structure more consistent with the actual chemical environment, which helps to more accurately simulate the interaction in the subsequent molecular docking process. Then, structural transformation is performed to add key information such as atomic charge and atom type, and the structural data is further improved so that it can better participate in the docking calculation based on physical and chemical principles and improve the reliability of the docking results.

[0064] Get the set docking parameters and set the docking parameters to the central cavity of the protein. This approach of focusing the spatial range of the search for ligand-protein binding conformations on the central cavity of the protein is based on the consideration of protein structure and functional characteristics. Usually, the key binding sites of ligands and proteins are often concentrated in the central cavity area. By reasonably limiting the range in this way, unnecessary search calculations can be reduced, and the interaction between ligands and proteins can be simulated in a space that is more in line with biological reality, which not only improves the simulation efficiency, but also makes the simulation results more in line with the actual binding situation, and enhances the practicality of the results. Based on the set central cavity of the protein, the orientation and position changes of the ligand around the structure data of the mutant protein to be docked are simulated, and multiple different binding energy scores are generated by calculating the interaction energy, which can comprehensively explore the possible binding modes of ligands and proteins and the corresponding energy states. This helps to dig out the most stable and most likely binding conformation, and provides a rich data basis for accurately judging the interaction mode and strength between ligands and mutant proteins, so as to have a deeper and more detailed understanding of the binding situation between the two.

[0065] By determining the ligand with the highest binding energy score and the structure data of the mutant protein to be docked as the docking conformation, the conformation with the best energy and the most stable one is selected from many possible binding situations. This is in line with the principle that substances tend to form low-energy and stable states in chemical reactions, so that the most likely binding mode can be focused on, providing a key reference for subsequent in-depth research or practical applications (such as optimizing molecular modification strategies based on docking results, etc.). The binding energy, hydrogen bond and hydrophobic interaction force of the docking conformation are further calculated to achieve a comprehensive quantitative description of the molecular docking results. These quantitative data can intuitively reflect the important characteristics of the intensity and mode of interaction between the mutant protein and the ligand. Whether from the perspective of energy (binding energy) or from the perspective of specific intermolecular forces (hydrogen bonds, hydrophobic interaction forces), researchers can clearly grasp the binding characteristics between the two, which is convenient for comparing the binding differences between different mutant proteins and ligands, and then better evaluate the effect of mutations on the interaction between proteins and ligands, and provide strong quantitative support for related work such as molecular modification of glycosyltransferases. In a specific application scenario, the automated program for realizing the above-mentioned molecular docking can be constructed as follows: Use AutoDock Vina for molecular docking, prepare the docking protein list file, call pymol to dehydrate and hydrogenate each mutant protein and ligand, then use mgltools to convert to PDBQT format, set the docking box to the central cavity of the protein, and then perform command docking. Take the binding conformation with the highest score in the docking results as the analysis object, and organize and collect the binding energy of the docking conformation, as well as the number of hydrogen bonds and hydrophobic interaction forces obtained by pymol analysis, and further analyze it.

[0066] The following is an algorithmic idea for automated molecular docking and saving the binding energy of the docked conformation: (1) Import library: Import the os and subprocess libraries, abbreviated as os and subprocess respectively.

[0067] (2) Define the processing function: Define the perform_docking function, which receives the working directory work_dir and the ligand file path ligand_pdb.

[0068] (3) Define file paths: Initialize variables containing all input and output file paths.

[0069] (4) Initialize the result list: create an empty list results to store the molecular docking results.

[0070] (5) Convert ligand file: Use the PyMOL script to convert the ligand PDB file into a PDB file with hydrogen and check whether the file exists.

[0071] (6) Convert protein files: traverse the protein list, add hydrogen atoms to each protein using a PyMOL script, and convert it to a PDBQT file.

[0072] (7) Molecular docking: Use AutoDock Vina to perform molecular docking on each protein and extract affinity scores.

[0073] (8) Build result dictionary: For each protein docking result, create a dictionary containing protein name and affinity score and add it to the results list.

[0074] (9) Save results: Convert the results list into a DataFrame object and save it as a CSV file.

[0075] (10) Set parameters and call function: In the last part of the code, set the working directory and ligand file path, and call the perform_docking function.

[0076] (11) Result output: After the function is executed, the results will be saved in CSV format, and the file contains the molecular docking scores of all proteins.

[0077] Use pymol to analyze the structure of molecular docking and count the number of hydrogen bonds and hydrophobic interaction forces in the docking conformation. The algorithm is as follows: (1) Import necessary libraries: import PyMOL library, cmd, and operating system library os.

[0078] (2) Set the working path: specify the working directory variable work_dir.

[0079] (3) Specify the protein list file: Use the os.path.join method to merge the working directory with the proteins_list.txt file name to determine the path of the protein list file proteins_list_file.

[0080] (4) Read protein data: Open the file using the with statement and store the read row data in the protein list proteins.

[0081] (5) Determine the output file path: Use the os.path.join method again, combining the working directory and the interactions_summary.txt file name to determine the output file path output_file.

[0082] (6) Prepare the output file: Use the with statement to open the file, prepare for writing, and add a header line "Protein\tHydrogen Bonds\tHydrophobic Interactions" to the top of the file.

[0083] (7) Process proteins one by one: For each protein in the proteins list, perform the following operations: Set the file path: Use the os.path.join method to determine the paths of the protein PDB file protein_pdb, the hydrogenated PDBQT file protein_pdbqt, and the ligand PDBQT file ligand_pdbqt based on the working directory and protein name.

[0084] Restart PyMOL: Call cmd.reinitialize() to restart the PyMOL session.

[0085] Load files: Use cmd.load method to load protein and ligand files.

[0086] Add hydrogen atoms: Call cmd.h_add method to add hydrogen atoms to the loaded file.

[0087] Define Selection Sets: Set donor and acceptor selection sets for proteins and ligands.

[0088] Search for hydrogen bonds: Call cmd.find_pairs method to search for hydrogen bond pairs between protein and ligand.

[0089] Create a hydrogen bond selection set: Use the cmd.select method to create an hbonds selection set containing hydrogen bonds.

[0090] Count the number of hydrogen bonds: Call the cmd.count_atoms method to count the number of atoms in the hbonds selection set, that is, the number of hydrogen bonds.

[0091] Define hydrophobic residue selection sets: Set hydrophobic residue selection sets for proteins and ligands.

[0092] Search for hydrophobic interaction: Call cmd.select method to search for hydrophobic interaction pairs.

[0093] Count the number of hydrophobic interactions: Call the cmd.count_atoms method to count the number of atoms in the hydrophobic interaction selection set ligand_hydrophobic_interactions.

[0094] Record interaction data: Record protein name, number of hydrogen bonds, and number of hydrophobic interactions to an output file.

[0095] (8) Output result: Print information "Interaction data saved to:" and output file path output_file.

[0096] Then use the script to save the calculation of affinity change and interaction force number change in the document for subsequent analysis.

[0097] Optionally, the performing kinetic simulation analysis on the mutants in the variant basic data set to obtain kinetic simulation feature data includes: Determine the mutation sites in the variant basic data set that are located in the known key functional regions of the protein and the mutation sites whose impact factors on the overall structural stability of the protein are greater than a set factor threshold as mutants that need to be subjected to kinetic simulation analysis; Performing simulation calculations on the mutant that needs to be subjected to dynamic simulation analysis to obtain a protein topology file, wherein the protein topology file includes atom types, charges, and chemical bond connection relationships; Obtaining an existing ligand topology file obtained by wild-type docking, wherein the ligand topology file contains topological structure information of the ligand molecule under a corresponding simulation environment; Putting the protein topology file and the ligand topology file into a virtual container to construct a simulated virtual system; Performing energy minimization and temperature-pressure balance digital operations on the simulated virtual system to generate a dynamic simulation example; Based on the set virtual force field and simulation configuration parameters, the dynamics simulation instance is run to calculate the atomic motion and interaction within the dynamics simulation instance time step by time, simulate the dynamic change process of the system over time and generate dynamics simulation characteristic data accordingly, the dynamics simulation characteristic data including temperature change (Temperature), energy change (Interaction_Energy), RMSD (Root Mean Square Deviation, used to measure the deviation of the protein structure relative to the initial structure during the simulation process) and RMSF (Root Mean Square Fluctuation, reflecting the fluctuation amplitude of each atom of the protein during the simulation process) value changes.

[0098] To this end, the above-mentioned technical processing of the kinetic simulation analysis of mutants in the variant basic data set has the following technical advantages: (1) Focus on key mutants to improve analysis targeting By identifying the mutation sites in the variant basic data set located in the known key functional regions of the protein and the mutation sites whose impact factors on the overall structural stability of the protein are greater than the set factor threshold as mutants that need to be subjected to dynamic simulation analysis, we can focus on those key mutations that may have important effects on protein function and structural stability. Avoiding indiscriminate simulation analysis of all mutants can not only reduce unnecessary consumption of computing resources, but also more specifically mine mutation-related information that plays a key role in the function of glycosyltransferases, providing a key focus for subsequent in-depth understanding and optimization of molecular modification.

[0099] (2) Build a complete simulation system to ensure the scientific nature of the simulation For the selected mutants that need to be subjected to dynamic simulation analysis, simulation calculations are performed to obtain the protein topology file, which covers key information such as atom type, charge, and chemical bond connection relationship. At the same time, the ligand topology file obtained by the existing wild-type docking is obtained, which contains the topological structure information of the ligand molecule in the corresponding simulation environment. The preparation of these two topology files improves the basic data required for simulation from two important aspects, protein and ligand, so that the virtual system constructed subsequently can accurately reflect the actual molecular composition and structural characteristics, laying a solid foundation for scientific and reasonable dynamic simulation.

[0100] The protein topology file and the ligand topology file are placed in a virtual container to construct a simulated virtual system, simulating a scenario close to the interaction between proteins and ligands in a real physiological environment. In this way, the molecular system can be controlled and observed in a computer environment, avoiding the interference of many complex factors and experimental conditions that may be encountered in actual experiments. It is more convenient to study the dynamic changes of mutants in a specific system, which helps to deeply analyze the impact mechanism of mutations on the entire system.

[0101] (3) Multi-step simulation operation to gain in-depth insights into the dynamic properties of molecules The energy minimization and temperature and pressure balance digital operations are performed on the simulated virtual system. This step can make the virtual system reach a relatively stable initial state that conforms to the laws of physics and chemistry, eliminate problems such as unreasonable energy distribution and structural tension, ensure that subsequent dynamic simulations are carried out on a reasonable and stable basis, improve the accuracy and reliability of the simulation results, and more realistically reflect the behavior of the system in the actual environment.

[0102] Based on the set virtual force field and simulation configuration parameters, the dynamic simulation instance is run, and the atomic motion and interaction in the dynamic simulation instance are calculated step by step. The dynamic change process of the system over time is simulated and the dynamic simulation characteristic data is generated, such as temperature change, energy change, RMSD and RMSF value change, etc. These rich characteristic data show the dynamic characteristics of the system in the simulation process from different angles. For example, RMSD can intuitively understand the deviation of the protein structure relative to the initial structure during the simulation and judge the change in the stability of the structure; RMSF can reflect the fluctuation amplitude of each atom of the protein, which is helpful for analyzing the flexibility differences in different regions. Through these quantitative data, we can have a deep insight into the dynamic behavior of mutants at the molecular level and the change of interaction with ligands over time, which provides a detailed basis for accurately evaluating the impact of mutations on the function and structural stability of glycosyltransferases, and then guide the formulation of more effective molecular transformation strategies.

[0103] In a specific application scenario, the automated program for implementing the above dynamics simulation analysis can be constructed as follows: Each mutant was analyzed by the dynamics simulation software Gromacs. However, it should be noted that when preparing the protein and ligand topology files, the ligand topology file of the mutant was still analyzed using the ligand topology file obtained by docking with the wild type. After the entire system was prepared, energy minimization and temperature-pressure equilibrium were performed. The simulation was run for 50 ns. After the run, the analysis tool provided by Gromacs was used to obtain the temperature change (Temperature), energy change (Interaction_Energy), RMSD and RMSF value changes during the running period. Re-analysis was performed to compare the temperature change (mean and maximum fluctuation), energy change (mean and maximum fluctuation) and RMSD (mean and maximum fluctuation) of the mutant with that of the wild type. The RMSF value is used for segmented mean comparison. The segmentation is based on the sequence alignment of the glycosyltransferase family. Considering that glycosyltransferase exists in the GTB family and has two domains, the UGT89C1 sequence, which has been studied more, is used in the Uniport database for alignment. The top 100 sequences are selected for sequence alignment using MEGA11, and the data with sequence lengths significantly longer than other sequences are removed. GeneDoc software is used to observe the conservation of the sequence, and the domain range of UGT89C1 is combined for division, and the ratio of the region to the full length is calculated. The domain division ratio of UGT89C1 is 4%, 48%, 34%, and 14%, but considering the application of the entire glycosyltransferase, the segmented region is generalized to adapt to the functional domain division of different glycosyltransferases. Finally, according to the alignment results, the segmentation is designed according to 2%, 2%-50%, 50%-90%, and 90%-100% of the full length of the protein and the mean comparison calculation is performed. Temperature, energy change, RMSD and RMSF were selected as analysis indicators because they can comprehensively reflect the thermal stability, interaction, overall structural stability and local flexibility of proteins. These indicators are commonly used parameters to evaluate the effects of protein mutations and can provide important information about protein function and stability.

[0104] The code algorithm for analysis is as follows: (1) Initialize variables: mutants: a list containing mutants, properties: a list containing the physicochemical properties to be analyzed, column_defaults: defines the column name of the CSV file for each property.

[0105] (2) Define the function process_non_rmsf_data: used to read non-RMSF (Root Mean Square Fluctuation) data files, calculate the difference between the mean and the maximum and minimum values, and round the result to 8 decimal places.

[0106] (3) Preprocess all non-RMSF data: traverse each property and each mutant, call the process_non_rmsf_data function to process the data, and store the results in the all_data dictionary.

[0107] (4) Calculate the difference of non-RMSF attributes and save to CSV: Traverse the all_data dictionary, calculate the mean and the difference between the mutant and the wild type for each property, and store these data in the list non_rmsf_results. Convert the non_rmsf_results list to a Pandas DataFrame and save it to a CSV file.

[0108] (5) Define the function process_rmsf_data: used to read the RMSF data file, calculate the average RMSF value of each amino acid residue segment, and round the result to 8 decimal places.

[0109] (6) Read wild-type data: Call the process_rmsf_data function to process the wild-type RMSF data.

[0110] (7) Initialize the RMSF data structure and create a dictionary rmsf_diff_data to store the RMSF difference of each amino acid residue segment.

[0111] (8) Process the RMSF data of each mutant: traverse each mutant and call the process_rmsf_data function to process the RMSF data. Calculate the RMSF difference between the mutant and the wild type for each amino acid residue segment and store the result in the rmsf_diff_data dictionary.

[0112] (9) Create RMSF difference DataFrame: Convert the rmsf_diff_data dictionary to a Pandas DataFrame. Transpose the DataFrame so that the RMSF difference for each amino acid residue segment is a column. Save the transposed DataFrame to CSV. Save the transposed DataFrame to a CSV file.

[0113] Temperature change: Temperature is a measure of thermal motion in a simulated system, which reflects the energy distribution of the system. During the simulation, the change in temperature can indicate the thermal stability and dynamic behavior of the system.

[0114] Energy change (Interaction_Energy): Interaction energy describes the strength of intermolecular interactions, including van der Waals forces, hydrogen bonds, electrostatic interactions, etc. Energy changes can reflect the effects of mutations on protein stability and function.

[0115] RMSD (root mean square deviation) is a measure of the difference between two three-dimensional structures. In molecular dynamics simulations, RMSD is often used to evaluate the stability of protein or other molecular structures relative to the initial structure or reference structure during the simulation. The smaller the RMSD value, the smaller the structural change and the higher the stability. It is crucial to understand the flexibility, binding properties and function of proteins.

[0116] RMSF (root mean square fluctuation) is used to describe the average position fluctuation of a single atom or residue in a molecular dynamics simulation over the simulation time. Regions with high RMSF values ​​usually represent flexible regions in the protein that may be critical to the function of the protein, such as binding sites or hinge regions between domains. RMSF analysis helps identify flexible regions in proteins that may be related to the biological function and dynamic behavior of the molecule.

[0117] In the analysis of dynamic simulation results, these two RMSD and RMSF are often drawn into a curve graph to facilitate a more intuitive analysis and understanding of the dynamic behavior of the molecule. For example, when analyzing the RMSD value, if the RMSD value remains low and stable for most of the simulation time, this indicates that the binding of the drug molecule to the target protein is stable. If the RMSD value increases over time, this may indicate that the drug molecule dissociates from its binding site or that the target protein undergoes a significant conformational change; when analyzing the RMSF value, the RMSF curve will reveal that certain specific regions of the protein (such as the loop region or the unstructured tail) show higher RMSF values ​​during the simulation, indicating that these regions are more flexible than other parts.

[0118] After obtaining the RMSD and RMSF graphs, the graphical analysis was converted into numerical analysis. For the RMSD value, the mean and maximum fluctuation were calculated to replace the analysis of the continuous fluctuations of the RMSD graph; for the RMSF value, the mean of the change in each area was calculated to replace the analysis of the fluctuations in each area of ​​the RMSF graph, and the changes in each section were analyzed. In combination with the actual situation, the functional domains of the glycosyltransferase were divided, and the fluctuation areas that the data could reflect were focused on the areas that had a greater impact on the enzyme activity.

[0119] Moreover, by comparing the mutants with the wild type, the effects of mutations on protein properties can be intuitively demonstrated. In addition, a segmented comparison approach was used, which enables a more detailed analysis of local changes in proteins, especially considering the sequence conservation and domain structure of the glycosyltransferase family.

[0120] Converting RMSD and RMSF analysis in molecular dynamics simulations to numerical analysis has the following benefits: (1) Accurate quantification: Numerical analysis provides an accurate method to quantify the dynamic behavior of molecules, making the comparison of different simulation or experimental data more direct and objective.

[0121] (2) Simplified interpretation: By simplifying complex curve data into a few key values ​​(such as RMSD mean and maximum fluctuation, RMSF area mean), it is easier to interpret and understand the stability and flexibility areas of the molecule, that is, the changes in various regions of the protein during the reaction.

[0122] (3) Rapid identification of key regions: Numerical analysis helps to quickly identify key flexible regions in protein structures, which may be closely related to biological functions, such as binding sites or hinge regions between domains.

[0123] (4) Easy to compare and help assist decision-making: Numerical results can be easily used to compare different simulation conditions, different proteins, or different drug molecules to evaluate their effects on the dynamic behavior of the protein.

[0124] (5) Improved efficiency: Automated numerical analysis can save a lot of time, especially when dealing with large-scale or long-term simulation data. Compared with manual analysis of curve graphs, numerical methods can generate results quickly.

[0125] (6) Statistical analysis: Numerical data facilitate statistical analysis, such as calculation of standard deviation, confidence interval, etc., which helps in assessing the reliability and variability of the results.

[0126] (7) Data integration: Numerical analysis can be integrated with other bioinformatics data, such as protein sequence information, domain information, etc., to provide a more comprehensive perspective for systems biology research.

[0127] (8) Support for further research: The flexible regions identified by numerical analysis can serve as a starting point for further experimental studies, such as mutation analysis to study the effects of these regions on protein function.

[0128] The above dynamic simulation process has the following technical features: (1) Domain considerations: By combining sequence alignment and domain information of the glycosyltransferase family, your method can more accurately identify and analyze the key functional regions of proteins.

[0129] (2) Segment comparison: By segmenting based on sequence conservation and domain extent, your analysis can more specifically relate protein dynamics to functional domains.

[0130] (3) Generalized segmented region: Not only applicable to specific glycosyltransferases, but also generalizable to other glycosyltransferases, increasing the universality and application scope of the analysis.

[0131] Optionally, the model training set is constructed based on the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation feature data to train the selected machine learning model to obtain an enzyme activity prediction model, including: Preprocessing the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation characteristic data to form a preprocessed data set; Performing feature selection on the preprocessed data set to generate model feature data and corresponding target feature data; Constructing a model training set based on the model feature data and the corresponding target feature data, and dividing the model training set into a training set and a test set according to a set division rule; Dividing the training set into multiple batches of sample data, so as to train the selected machine learning model multiple times according to a set number of iterations to obtain a model to be verified; The model to be verified is verified based on the test set. If the verification condition is met, the training is terminated and the model to be verified that passes the verification is used as the enzyme activity prediction model.

[0132] To this end, the above-mentioned technical processing process for constructing an enzyme activity prediction model has the following technical benefits: (1) Data preprocessing Preprocessing of data from diverse and complex sources, such as the result set of physicochemical property analysis, weighted information set of functional segments, energy change values, quantitative descriptions, and kinetic simulation feature data, can remove possible noise, outliers, and irregular data formats. For example, unifying the dimensions of different data and filling in missing values ​​can make the data used for subsequent model training more accurate and standardized, laying a good data foundation for training high-quality enzyme activity prediction models and avoiding overfitting or inaccurate predictions due to poor data quality.

[0133] Through the preprocessing integration operation, the data originally scattered in various data sets can be better integrated together, and the potential correlation between them can be excavated, so that data of different dimensions can work together to more comprehensively reflect the various characteristics of glycosyltransferases and their relationship with enzyme activity, thereby providing strong support for the model to learn more accurate rules.

[0134] (2) Feature selection Feature selection is performed to screen out model feature data and corresponding target feature data that have an important impact on enzyme activity prediction from numerous original data features, and to eliminate those redundant or weakly correlated features. This can not only reduce the computational complexity of model training and speed up training, but also avoid the interference of irrelevant features in the model learning process, allowing the model to focus more on key factors closely related to enzyme activity, thereby improving the model's prediction accuracy and generalization ability, so that it can better capture the essential characteristics and laws that affect enzyme activity changes.

[0135] (3) Construction and division of training sets The model training set is constructed based on the selected model feature data and the corresponding target feature data, and is divided into a training set and a test set according to the set division rules. This scientific division method can ensure that the model is trained on a part of the data, and the performance is tested on another part of the data that is not involved in the training. In this way, the generalization ability of the model, that is, the model's ability to predict new data, can be truly and objectively evaluated, avoiding the situation where the model simply remembers the training data (overfitting), so that the final enzyme activity prediction model has stronger reliability and practicality, and can accurately predict enzyme activity changes for different glycosyltransferase situations in practical applications.

[0136] (4) Batch training and verification Dividing the training set into multiple batches of sample data and training the selected machine learning model multiple times according to the set number of iterations will help the model gradually adjust parameters, learn the rules in the data more stably, and make the model convergence process more stable and accurate. Then the trained model to be verified is verified based on the test set. Only when the verification conditions are met will the training be terminated and the final enzyme activity prediction model be determined. Through such a strict verification mechanism, it can be ensured that the obtained model meets the requirements in terms of accuracy, stability, etc. The output prediction results have a high degree of credibility, which can provide an accurate reference for the subsequent molecular modification of glycosyltransferases based on enzyme activity prediction.

[0137] In a specific application scenario, the above model can be constructed and trained as follows: The information of amino acid rational property change, energy change, affinity, hydrogen bond number, temperature change mean, energy change mean, RMSD mean and RMSF segment mean were collected and sorted as training data sets; the sorted training data sets were used to build models using different machine learning models, and the models were trained and debugged. The prediction performance of the models was verified by using 20% ​​of the training data sets as independent test sets, using MSE, RMSE and R square as evaluation indicators, and selecting the model with the smallest MSE and RMSE values ​​and the R square closest to 1 as the initial prediction model. Finally, it was found that the model was constructed using a neural network model, with one input layer, five hidden layers and one output layer as the basic model, and the tanh function as the activation function, and the model with the smallest MSE and RMSE values ​​and the R square closest to 1 was used as the prediction model.

[0138] The model building algorithm is as follows: (1) Import the required Python libraries, including pandas, modules in sklearn, and components in tensorflow and keras; (2) Read the Excel file 'mutation sites and enzyme activity changes.xlsx' into DataFrame df; (3) Create a new column named 'Functional_Region_Weight' in the DataFrame df and assign weight values ​​to specific locations to indicate the importance of the functional region; (4) Select feature columns for the model, including changes in the physicochemical properties of amino acids, functional region weights, energy change information, affinity change information, hydrogen bond number change information, temperature mean and fluctuation changes, energy mean and fluctuation changes, RMSD mean and fluctuation changes, and RMSF segment mean changes, and store these features in the list features; (5) Select the target column 'Enzyme_Activity', which is the enzyme activity to be predicted; (6) Extract features and target data from DataFrame df, store feature data in X, store target data in y and convert them into NumPy arrays; (7) StandardScaler is used to standardize the feature data to eliminate the influence of different dimensions; (8) Use the train_test_split function to divide the data set into a training set and a test set. The test set accounts for 20%, and the random state is set to 42 to ensure the repeatability of the results; (9) Build a Sequential model with multiple hidden layers, each layer uses the 'tanh' activation function, and the output layer uses a linear activation function; (10) Compile the model, use the Adam optimizer, set the learning rate to 0.001, and use the mean square error as the loss function; (11) Train the model using the training set data for 100 epochs, with the batch size set to 32, and 20% of the training data used for validation; (12) Evaluate model performance, using the test set data, and calculate the mean square error (MSE), root mean square error (RMSE), and coefficient of determination (R2) on the test set; (13) Save the trained model to the file 'enzyme_activity_predictor.keras'; (14) Load the saved model for subsequent prediction; (15) Read the new CSV data file 'mutation_results.csv' and perform the same preprocessing steps on the new data; (16) Use the loaded model to predict new data and obtain the predicted value of enzyme activity; (17) Print the prediction results and save them in the DataFrame new_df, then sort them in descending order. Finally, save the DataFrame containing the prediction results as a CSV file 'predicted_enzyme_activities.csv'.

[0139] 2. Prediction model optimization: Optimize the parameters of the prediction model through methods such as cross-validation to improve its prediction accuracy.

[0140] Experimental verification and dynamic adjustment 1. Experimental verification: Select enzymes with production value but low enzyme activity as experimental objects, prepare the protein sequence and ligand information, collect the obtained characteristic indicators, input them into the enzyme activity prediction model, predict the enzyme activity at the mutation site, select mutants with higher predicted enzyme activity for experimental verification, obtain the enzyme activity change data of mutants and wild types, compare the experimental results with the prediction results of the prediction model to verify the effectiveness of the enzyme activity prediction model.

[0141] 2. Dynamic adjustment: Dynamically adjust the enzyme activity prediction model according to the experimental results, including adjusting weights, changing the prediction model, etc., to improve the accuracy of subsequent screening.

[0142] Optionally, the prediction of the change in the enzyme activity of the target glycosyltransferase relative to the wild-type mutant based on the enzyme activity prediction model to obtain a predicted value of the mutant enzyme activity includes: For the target glycosyltransferase, determining the physicochemical property data, functional segment weighted information, energy change value, molecular docking quantitative description and kinetic simulation characteristic data of the corresponding mutant to generate an input feature vector of the model; The input feature vector is input into the enzyme activity prediction model, and based on the relationship between the parameters and features learned in the training phase and the enzyme activity, the input feature vector is calculated, and the predicted enzyme activity value is output as the predicted value of the mutant enzyme activity.

[0143] Optionally, the molecular modification of the target glycosyltransferase based on the predicted value of the mutant enzyme activity comprises: Determine the difference between the predicted value of the mutant enzyme activity and the value of the wild-type enzyme activity to determine the change trend of the enzyme activity; Based on the location of the mutation site in the target glycosyltransferase and the change trend of the enzyme activity, a molecular modification strategy is generated; Designing primers based on the molecular modification strategy to perform molecular modification on the target glycosyltransferase; After introducing the modified target glycosyltransferase into a host cell, the host cell is allowed to express the mutant glycosyltransferase protein; The enzyme activity of the mutant glycosyltransferase protein expressed by the host cell is measured to determine whether the expected molecular modification index is achieved.

[0144] To this end, the above-mentioned technical processing related to molecular modification based on enzyme activity prediction values ​​has the following technical advantages: (1) The input feature vector is generated by integrating the physicochemical property data, functional segment weighted information, energy change value, molecular docking quantitative description and kinetic simulation feature data of the target glycosyltransferase corresponding mutant, and then the enzyme activity prediction model is used to calculate the predicted value of the mutant enzyme activity. This approach integrates the key data reflecting the characteristics of glycosyltransferases from many aspects and fully explores the intrinsic relationship between various factors and enzyme activity. Compared with estimating enzyme activity changes based on experience or single-dimensional data, its prediction results are more accurate and reliable, and can provide clear, scientifically analyzed and quantitative references for subsequent molecular modification work in advance, avoiding the uncertainty and waste of resources caused by blind attempts at modification.

[0145] (2) Determine the difference between the predicted value of the mutant enzyme activity and the value of the wild-type enzyme activity to judge the trend of enzyme activity change, and generate a molecular modification strategy based on the location of the mutation site, so that the entire modification work has a clear direction. It is possible to analyze from the root why the enzyme activity increases, decreases, or changes insignificantly, and to clarify the impact of the mutation at a specific position on the enzyme activity, so as to formulate a modification strategy "for the right disease", focus on key structural regions and functional sites for targeted adjustments, greatly improve the accuracy and success rate of molecular modification, and avoid aimless modification operations.

[0146] (3) Primers are designed based on the generated molecular modification strategy, and then the target glycosyltransferase is molecularly modified. This process transforms the theoretical modification strategy into a practical and operational gene-level change through the key tool of primers. The design of primers can accurately locate the gene sequence region that needs to be modified, ensure the accurate introduction, replacement or adjustment of the corresponding bases according to the established strategy, and achieve the directional optimization of the molecular structure of the glycosyltransferase, so that the molecular modification work can be carried out in an orderly and controllable manner, which meets the requirements of modern biotechnology for precise gene editing and helps to achieve the expected functional improvement effect.

[0147] (4) The modified target glycosyltransferase is introduced into the host cell to make it express the mutant glycosyltransferase protein. With the help of the host cell expression system, the modified gene information is converted into a protein product with actual function. This is a conventional and effective means to achieve gene function research and protein application in biotechnology. On this basis, the enzyme activity of the mutant glycosyltransferase protein expressed by the host cell is measured, and the data obtained by actual measurement is compared with the expected molecular modification index to intuitively judge whether the modification work has achieved the expected goal. If not, the reasons can be further analyzed based on the measurement results, and the entire modification process can be traced back to adjust the modification strategy, optimize the primer design and other links, forming a closed-loop process of continuous feedback and optimization, so as to ensure that the glycosyltransferase molecules that meet the actual application requirements and whose enzyme activity can be effectively improved can be obtained in the end, which will effectively promote its efficient application in related fields such as industrial production.

[0148] Based on the above embodiments, Figure 2 This application provides a flow chart of another method for molecular modification of glycosyltransferase based on artificial intelligence. Figure 2 As shown, it includes the following steps: Obtain descriptive information of enzyme mutants, including mutation sites and experimental results, and preprocess this information to form a basic data set; Use automated scripts to analyze the physical and chemical properties of the basic data set and obtain an analysis result set; Search similar sequences of the original mutation site sequence in the protein database, calculate the quantitative value of the conservation degree of the mutation site and the quantitative value of the influence of different mutations on the sequence; Based on the quantitative values ​​of conservation and impact, a position-specific scoring matrix is ​​constructed; Extracting location-specific scores and frequencies from the scoring matrix, determining functional segments, and performing weighted processing; Based on the three-dimensional structural reconstruction model of the mutant, three-dimensional structure prediction and energy calculation are performed to obtain the energy value; Calculate the energy change between the mutant and the wild type; Molecular docking and dynamics simulation analysis were performed on the mutant proteins to obtain quantitative descriptions of interaction patterns and strengths as well as dynamics simulation feature data; Based on the above analysis results, a model training set was constructed to train a machine learning model to predict changes in the enzyme activity of the mutants; According to the predicted enzyme activity change value, the target glycosyltransferase is molecularly modified.

[0149] Figure 2 For an exemplary description of each step in Figure 1 Description.

[0150] Further, in the above Figure 2 Based on the embodiment, a method for molecular modification of glycosyltransferase based on artificial intelligence is provided, such as Figure 3 As shown, it includes: Determine the comprehensive characteristic data of the mutation sites in the mutant basic data set; Based on the comprehensive feature data, a model training set is constructed to train the selected machine learning model to obtain an enzyme activity prediction model; Based on the enzyme activity prediction model, predict the change in the enzyme activity of the target glycosyltransferase to obtain a predicted value of the mutant enzyme activity; Based on the predicted value of the mutant enzyme activity, the target glycosyltransferase is molecularly modified.

[0151] Optionally, the comprehensive characteristic data includes at least one of the following data: a set of physicochemical property analysis results, a quantitative value of conservation degree and a quantitative value of the influence degree of different mutations on the sequence, a molecular docking result, a kinetic simulation characteristic data, a functional segment weighted information set, and an energy change value; Optionally, the comprehensive characteristic data includes at least one of the following: the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation characteristic data.

[0152] To implement the above embodiments, the embodiments of the present application further provide an electronic device, which includes a memory and a processor, wherein a computer executable program is stored in the memory, and the computer executable program is executed by the processor to implement the various steps in the above embodiments of the present application.

[0153] The specific form of the electronic device is not limited to uniqueness.

[0154] For exemplary explanations of the various steps performed in the above-mentioned device, please refer to the above-mentioned Figure 1 Description.

[0155] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for molecular modification of glycosyltransferase based on artificial intelligence, characterized in that: include: Obtaining description information of a glycosyltransferase mutant, wherein the description information includes a mutation site and a description of related experimental results; Preprocessing the description information of the glycosyltransferase mutant to obtain a mutant basic data set; Based on the constructed automated script, the physicochemical properties of the mutation sites are analyzed on the mutant basic data set to obtain a physicochemical properties analysis result set; Based on the original mutation site sequence data corresponding to the mutant basic data set, a similar sequence search is performed in a protein database to calculate a quantitative value of the conservation degree of each mutation site in the protein database and a quantitative value of the degree of influence of different mutations on the sequence; Based on the quantitative value of the conservation degree of each mutation site in the protein database and the quantitative value of the influence degree of different mutations on the sequence, constructing a position-specific scoring matrix; Based on the constructed automated script, extracting the position-specific score value and frequency corresponding to each mutation site from the position-specific scoring matrix; Determining the functional segment of the glycosyltransferase protein sequence in the glycosyltransferase mutant according to the position-specific score and frequency corresponding to each mutation site; Performing weighted processing on key functional points in the functional section to obtain a weighted information set of the functional section; Based on the three-dimensional structure reconstruction model of the glycosyltransferase mutant, the three-dimensional structure of the mutant glycosyltransferase selected from the mutant basic data set is predicted to obtain the predicted three-dimensional structure of the mutant glycosyltransferase; Performing energy calculation on the predicted three-dimensional structure of the mutant glycosyltransferase to obtain an energy value of the three-dimensional structure of the mutant glycosyltransferase; Calculating the difference between the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase to obtain an energy change value; Performing molecular docking on the mutant protein in the variant basic data set to obtain a molecular docking result, wherein the molecular docking result at least includes: a quantitative description of the interaction mode and strength between the mutant protein and the ligand; Performing kinetic simulation analysis on mutants in the variant basic data set to obtain kinetic simulation characteristic data; Based on the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation characteristic data, a model training set is constructed to train the selected machine learning model to obtain an enzyme activity prediction model; Based on the enzyme activity prediction model, predict the change in enzyme activity of the target glycosyltransferase relative to the wild-type mutant to obtain a predicted value of the mutant enzyme activity; Based on the predicted value of the mutant enzyme activity, the target glycosyltransferase is molecularly modified.

2. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The preprocessing of the description information of the glycosyltransferase mutant to obtain a mutant basic data set includes: Performing data filtering on the description information of the glycosyltransferase mutant to select data directly related to the mutation site and the change of enzyme activity; Based on the data directly related to the mutation site and the enzyme activity change screened out, multiple data sub-tables are constructed, each data sub-table corresponds to a protein mutant or a group of mutants with similar enzyme activity change characteristics; Cleaning the multiple sub-tables, removing the repeated data directly related to the mutation sites and enzyme activity changes between the multiple sub-tables, to obtain a cleaned data sub-table; The data in the cleaned data subtable is standardized to obtain a mutant basic data set.

3. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The constructed automated script performs a physicochemical property analysis of the mutation site on the mutant basic data set to obtain a physicochemical property analysis result set, including: Constructing a regular expression, wherein the regular expression is used to extract a mutation site representation from the mutant basic data set; Creating a first mapping dictionary, a second mapping dictionary, and the amino acid property dictionary, wherein the first mapping dictionary is used to map the three-letter abbreviations of the amino acids in the mutation site representation to single-letter representations, the second mapping dictionary is used to reversely map the single-letter representations of the amino acids to the three-letter representations of the amino acids in the mutation site table, and the amino acid property dictionary is used to define the description of the physicochemical properties of the amino acids; An amino acid mutation impact function is defined, wherein the amino acid mutation impact function is used to query the amino acid property dictionary to obtain the physicochemical property description of the amino acid in the mutation site table, and based on this, the physicochemical property change value of the mutation site is calculated to perform mutation site physicochemical property analysis on the mutant basic data set to obtain a physicochemical property analysis result set; Defining an amino acid representation conversion function to convert the representation of the amino acid in the mutation site table based on the first mapping dictionary and the second mapping dictionary, wherein the conversion includes: converting the three-letter representation of the amino acid into a single-letter representation and / or converting the single-letter representation of the amino acid into a three-letter representation; defining a mutation reading function, the mutation reading function being used to read the mutant basic data set to extract a mutation site representation from the mutant basic data set using the regular expression; The file storage path is set to set the file access path of the mutant basic data set, so as to call the defined mutation reading function to read the mutant basic data set.

4. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The step of determining the functional segment of the glycosyltransferase protein sequence in the glycosyltransferase mutant according to the position-specific score and frequency corresponding to each mutation site comprises: Based on the GTB family characteristics of glycosyltransferases and the fact that they have two domains, a UGT89C1 sequence with a heat value greater than a set heat threshold is screened from the protein database as a reference sequence; Screening the protein database for the top multiple protein sequences related to the glycosylation process; Comparing the reference sequence with the plurality of protein sequences to determine regions of similarity and difference between the sequences; Based on the similarities and differences between the sequences, protein sequences whose sequence lengths differ from other sequences by more than a set sequence difference are determined as abnormal sequences and removed to obtain valid sequences; Based on the two domains, the position-specific scoring values ​​and frequencies corresponding to each mutation site, the functional segments of the glycosyltransferase protein sequence in the glycosyltransferase mutant are determined.

5. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The weighted processing of the key functional points in the functional section to obtain a functional section weighted information set includes: Counting the position-specific scores and frequencies of mutation sites within the functional segment to screen out the key functional sites; Assigning weighting factors based on the location-specific scores and frequencies of the key functional sites; The positions of the key functional points are weighted and settled based on the weighting factors to obtain a functional section weighted information set.

6. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The three-dimensional structure reconstruction model based on the glycosyltransferase mutant is used to predict the three-dimensional structure of the mutant glycosyltransferase selected from the mutant basic data set to obtain the predicted three-dimensional structure of the mutant glycosyltransferase, including: The mutant basic data set is analyzed to screen out the mutation site position, the amino acid types before and after the mutation, and the corresponding glycosyltransferase sequence; The collected three-dimensional structure data of the wild-type glycosyltransferase is used as the initial prediction framework, and the mutation site position, the amino acid types before and after the mutation, and the corresponding glycosyltransferase sequence are input into the three-dimensional structure reconstruction model of the glycosyltransferase mutant to perform three-dimensional structure prediction to obtain the predicted mutant glycosyltransferase three-dimensional structure.

7. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The energy calculation of the predicted mutant glycosyltransferase three-dimensional structure to obtain the energy value of the mutant glycosyltransferase three-dimensional structure includes: An energy calculation task is created to input the predicted mutant glycosyltransferase three-dimensional structure into the constructed energy calculation protocol and analyze and calculate the atomic interactions and chemical bond energies in the mutant glycosyltransferase three-dimensional structure accordingly to obtain the energy value of the mutant glycosyltransferase three-dimensional structure.

8. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The step of calculating the difference between the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase to obtain the energy change value comprises: Based on the constructed energy change value calculation function, the energy value of the three-dimensional structure of the mutant glycosyltransferase and the energy value of the wild-type glycosyltransferase are subjected to difference processing to obtain the energy change value.

9. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The molecular docking is performed on the mutant protein in the variant basic data set to obtain a molecular docking result, and the molecular docking result at least includes: a quantitative description of the interaction mode and strength between the mutant protein and the ligand, including: Screening out mutant proteins for molecular docking from the variant basic data set to form a protein list, wherein the protein list at least includes a file path and a name corresponding to each mutant protein; Opening the mutant protein structures and corresponding ligand structures in the protein list one by one to perform dehydration and hydrogenation operations on them to obtain processed mutant protein structure data; Performing structural transformation on the mutant protein structure data to add atomic charges and atomic types therein to obtain the mutant protein structure data to be docked; Obtaining set docking parameters, wherein the docking parameters define a spatial range for searching for a binding conformation of a ligand and a protein; The docking parameters were set to the central cavity of the protein; Based on the central cavity of the protein, simulating the orientation and position changes of the ligand around the mutant protein structure data to be docked, so as to calculate the interaction energy between the ligand and the mutant protein structure data to be docked and generate a plurality of different binding energy scores between the ligand and the mutant protein structure data to be docked; Determine the ligand corresponding to the highest binding energy score and the structure data of the mutant protein to be docked as the docking conformation; The binding energy, hydrogen bond and hydrophobic interaction force numbers of the docked conformation are calculated to obtain the molecular docking result.

10. The method for molecular modification of glycosyltransferase based on artificial intelligence according to claim 1, characterized in that: The method comprises: constructing a model training set based on the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation characteristic data to train the selected machine learning model to obtain an enzyme activity prediction model, including: preprocessing the physicochemical property analysis result set, the functional segment weighted information set, the energy change value, the quantitative description, and the kinetic simulation characteristic data to form a preprocessed data set; Performing feature selection on the preprocessed data set to generate model feature data and corresponding target feature data; Constructing a model training set based on the model feature data and the corresponding target feature data, and dividing the model training set into a training set and a test set according to a set division rule; Dividing the training set into multiple batches of sample data, so as to train the selected machine learning model multiple times according to a set number of iterations to obtain a model to be verified; The model to be verified is verified based on the test set. If the verification condition is met, the training is terminated and the model to be verified that passes the verification is used as the enzyme activity prediction model.

Images