Enzyme modification method based on artificial intelligence

Abstract

The invention provides an enzyme modification method based on artificial intelligence. The enzyme modification method comprises the following steps: training an AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model; performing multi-sequence comparison on the protein and nucleic acid sequence database and the non-redundant protein feature sequence database based on the active point location recognition model to obtain a comparison result; initializing a matrix of which the length is equal to that of a comparison result, so as to construct a PSSM matrix and screen out potential mutation sites; according to the screened potential mutation sites, constructing an initial mutation library to form a virtual screening mutation library; performing structure reconstruction on the sequence in the virtual screening mutation library based on a glycosyl transferase tertiary structure prediction model so as to predict the glycosyl transferase tertiary structure and perform molecular docking and dynamic simulation on the glycosyl transferase tertiary structure; the method is used for analyzing binding pockets of proteins and ligands and determining dynamic interaction of amino acid residues and ligands in the binding pockets, and the contact number of each amino acid residue and the ligands is used for modifying the tertiary structure of the glycosyltransferase.

Description

Technical Field

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

[0002] As a biocatalyst, enzymes play an important role in industrial production, environmental governance, medical health and other fields. The catalytic efficiency and specificity of enzymes are directly related to the performance and application range of related products. With the development of biotechnology, enzyme engineering has gradually become a key technology to improve enzyme performance. Enzyme engineering includes methods such as directed evolution of enzymes, rational design and computational simulation, aiming to improve the stability, catalytic efficiency and substrate adaptability of enzymes by modifying their structure and function.

[0003] However, traditional enzyme modification methods mainly rely on experimental trial and error or rational design based on known structures. For example, site-directed mutagenesis usually requires a large number of experiments to verify the effects of different mutation sites, which is time-consuming and costly. In addition, although rational design based on known structures can reduce the number of experiments to a certain extent, it is often difficult to achieve the expected results due to the limited accuracy of protein structure prediction. Summary of the invention

[0004] In order to solve the above technical problems, the present application provides an enzyme modification method based on artificial intelligence to at least solve or alleviate the problems existing in the above-mentioned prior art.

[0005] An enzyme modification method based on artificial intelligence, comprising: Obtain the sequence of glycosyltransferase to train the AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model; Based on the active site recognition model, multiple sequence alignment is performed on the protein and nucleic acid sequence database and the non-redundant protein feature sequence database to obtain alignment results, wherein the alignment results include conserved sites and functional segments of sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database; Initialize a matrix with the same length as the alignment result. The rows of the matrix correspond to the positions of the amino acid residues in the sequence, and the columns correspond to different amino acid types. Traversing the alignment results, counting the frequency of occurrence of each amino acid type, initializing the matrix after the initialization, and constructing a PSSM matrix; According to the PSSM matrix, PSSM scoring is performed to screen out potential mutation sites; Constructing an initial mutation library based on the screened potential mutation sites; Performing spatial restriction screening on the initial mutation library to form a virtual screening mutation library; Based on the glycosyltransferase tertiary structure prediction model, the sequences in the virtual screening mutation library are structurally reconstructed to predict the tertiary structure of the glycosyltransferase; Molecular docking and dynamic simulation were performed on the predicted tertiary structure of glycosyltransferase to analyze the binding pocket between the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand, as well as the number of contacts between each amino acid residue and the ligand; According to the binding pocket between the protein and the ligand, and the dynamic interaction between the amino acid residues and the ligand, and the number of contacts between each amino acid residue and the ligand, the optimal mutation site is screened from the virtual screening mutation library to transform the tertiary structure of the glycosyltransferase.

[0006] In the scheme of the present application, the tertiary structure prediction of glycosyltransferase using the AlphaFold2 model can provide accurate structural information and provide a solid foundation for subsequent enzyme modification. AlphaFold2 has proven its excellent ability in the field of protein structure prediction, which enables this method to guide the design and modification of enzymes more accurately. By constructing PSSM and scoring PSSM, potential mutation sites can be efficiently screened. This method not only takes into account the conservatism of amino acid residues, but also integrates the frequency of occurrence of different amino acid types at specific positions, ensuring that the selected mutation sites have a high potential modification value. Then, the initial mutation library is subjected to spatial restriction screening, further excluding those mutation sites that may have an adverse effect on the three-dimensional structure, and improving the practical application value of the mutation sites finally screened. Finally, by performing molecular docking and kinetic simulation on the predicted tertiary structure of glycosyltransferase, the binding pocket between protein and ligand and its dynamic interaction can be deeply analyzed, providing a scientific basis for selecting the optimal mutation site, from sequence acquisition to the final enzyme modification, forming a set of artificial intelligence-based processes, ensuring the scientificity and rationality of each step, and improving the success rate of enzyme modification. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] 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 an enzyme modification method based on artificial intelligence according to an embodiment of the present application. DETAILED DESCRIPTION

[0008] 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.

[0009] An enzyme modification method based on artificial intelligence, characterized by comprising: Obtain the sequence of glycosyltransferase to train the AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model; Based on the active site recognition model, multiple sequence alignment is performed on the protein and nucleic acid sequence database and the non-redundant protein feature sequence database to obtain alignment results, wherein the alignment results include conserved sites and functional segments of sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database; Initialize a matrix with the same length as the alignment result. The rows of the matrix correspond to the positions of the amino acid residues in the sequence, and the columns correspond to different amino acid types. Traversing the alignment results, counting the frequency of occurrence of each amino acid type, initializing the matrix after the initialization, and constructing a PSSM matrix; According to the PSSM matrix, PSSM scoring is performed to screen out potential mutation sites; Constructing an initial mutation library based on the screened potential mutation sites; Performing spatial restriction screening on the initial mutation library to form a virtual screening mutation library; Based on the glycosyltransferase tertiary structure prediction model, the sequences in the virtual screening mutation library are structurally reconstructed to predict the tertiary structure of the glycosyltransferase; Molecular docking and dynamic simulation were performed on the predicted tertiary structure of glycosyltransferase to analyze the binding pocket between the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand, as well as the number of contacts between each amino acid residue and the ligand; According to the binding pocket between the protein and the ligand, and the dynamic interaction between the amino acid residues and the ligand, and the number of contacts between each amino acid residue and the ligand, the optimal mutation site is screened from the virtual screening mutation library to transform the tertiary structure of the glycosyltransferase.

[0010] In the scheme of the present application, the tertiary structure prediction of glycosyltransferase using the AlphaFold2 model can provide accurate structural information and provide a solid foundation for subsequent enzyme modification. AlphaFold2 has proven its excellent ability in the field of protein structure prediction, which enables this method to guide the design and modification of enzymes more accurately. By constructing PSSM and scoring PSSM, potential mutation sites can be efficiently screened. This method not only takes into account the conservatism of amino acid residues, but also integrates the frequency of occurrence of different amino acid types at specific positions, ensuring that the selected mutation sites have a high potential modification value. Then, the initial mutation library is subjected to spatial restriction screening, further excluding those mutation sites that may have an adverse effect on the three-dimensional structure, and improving the practical application value of the mutation sites finally screened. Finally, by performing molecular docking and kinetic simulation on the predicted tertiary structure of glycosyltransferase, the binding pocket between protein and ligand and its dynamic interaction can be deeply analyzed, providing a scientific basis for selecting the optimal mutation site, from sequence acquisition to the final enzyme modification, forming a set of artificial intelligence-based processes, ensuring the scientificity and rationality of each step, and improving the success rate of enzyme modification.

[0011] Optionally, the step of obtaining the sequence of a glycosyltransferase to train the AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model includes: Capturing the dependency between amino acids based on the row-level and column-level attention modules in the AlphaFold2 model to obtain a predicted pairing table, which reflects the possible interactions and pairing relationships between amino acids; The predicted pairing table is expanded based on the structure generation module in the AlphaFold2 model to obtain the co-evolution relationship characteristics between the residues and predict the tertiary structure of the glycosyltransferase based on this.

[0012] To this end, the above-mentioned application of AlphaFold2 to the structure prediction of glycosyltransferases has the following significant technical benefits: (1) AlphaFold2 can capture the long-distance dependencies between amino acids in detail through row-level and column-level attention modules, which is crucial for understanding the complex interaction network in glycosyltransferases. This precise capture of dependencies makes the predicted tertiary structure closer to the real structure, improving the accuracy of the prediction.

[0013] (2) Compared with traditional experimental methods, AlphaFold2 can provide high-quality protein structure predictions in a shorter time, greatly shortening the research cycle. This is of great significance for quickly responding to scientific research needs and accelerating the development of new drugs.

[0014] (3) Using AlphaFold2 for structure prediction does not require expensive experimental equipment and materials, only computing resources are needed, which greatly reduces research costs and enables more research institutions and individuals to participate in the structural and functional research of glycosyltransferases.

[0015] (4) Accurate three-dimensional structure prediction provides a solid foundation for the functional study of glycosyltransferases. Based on the predicted structure, the active site and substrate binding mode of the enzyme can be further analyzed, providing theoretical guidance for enzyme functional annotation, modification and inhibitor design.

[0016] The following is an example code showing how to use AlphaFold2 (assuming a pre-trained model and necessary data processing tools) to predict the tertiary structure of a glycosyltransferase: import tensorflow as tf from alphafold import AlphaFold from alphafold.data import pipeline from alphafold.model import model from alphafold.common import protein # Amino acid sequence of glycosyltransferase amino_acid_sequence = "MSTN..." # Amino acid sequence # Data preprocessing, converting amino acid sequences into model input format feature_dict = pipeline.make_atomic_features( sequence=amino_acid_sequence, pdb_code=None, use_msa=True# Use multiple sequence alignment information ) # Load AlphaFold2 pre-trained model model_params = { 'data_dir': '. / data / ',# data directory 'model_dir': '. / models / alphafold' # Model directory } af_model = model.AlphaFoldModel(**model_params) # Prediction structure prediction_result = af_model.predict(feature_dict) # Extract the predicted tertiary structure predicted_structure = prediction_result['structure'] print("Predicted structure:", predicted_structure) Optionally, the active site recognition model includes a protein language model, a multimodal cross-attention model, and a multiple sequence alignment model. The active site recognition model performs multiple sequence alignment on a protein and nucleic acid sequence database and a non-redundant protein feature sequence database to obtain an alignment result, and the alignment result includes conserved sites and functional segments of sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database, including: Based on the protein language model, the sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database are converted into a protein language description of each amino acid residue; Based on the multimodal cross attention frame model, the protein language descriptions corresponding to the protein and nucleic acid sequence databases and the non-redundant protein feature sequence databases are aligned with the three-dimensional structure sample features of glycosyltransferases and the sample feature representations of enzymatic reactions, and then fused to obtain enzyme representation fusion features; Based on the multiple sequence alignment model, the protein and nucleic acid sequence databases and the non-redundant protein feature sequence databases are aligned to obtain sequence similarity or homology metrics, and the sites, structural motifs or functional domains that remain unchanged or vary within a set limited range in different sequences are determined based on the sequence similarity or homology metrics, so as to determine the conserved sites of the sequences in the protein and nucleic acid sequence databases and the non-redundant protein feature sequence database based on the sites that remain unchanged or vary within a set limited range, and to determine the functional segments of the sequences in the protein and nucleic acid sequence databases and the non-redundant protein feature sequence database based on the structural motifs or functional domains.

[0017] To this end, based on the active site recognition model, the protein language model, multimodal cross-attention model and multiple sequence alignment model are integrated to efficiently and accurately identify the conserved sites and functional segments related to glycosyltransferase activity from massive protein and nucleic acid sequence data, which has the following technical benefits: (1) Through the protein language model, complex protein and nucleic acid sequences are converted into easy-to-process protein language descriptions, which helps to capture key information in the sequence and reduce noise interference. The multimodal cross-attention model further aligns and fuses the sequence information with the three-dimensional structural sample features of glycosyltransferases and the sample feature representations of enzymatic reactions, thereby more accurately identifying activity-related sites.

[0018] (2) The multiple sequence alignment model can process a large amount of sequence data and find out the sites or structural motifs that remain unchanged in different sequences by comparing the similarities and homologies between different sequences. These are usually the key parts of protein function. This method is not only applicable to known glycosyltransferases, but can also effectively predict unknown or newly discovered glycosyltransferases.

[0019] (3) Accurate identification of active sites is the basis of drug design and enzyme engineering. This technical solution can quickly and accurately locate the active site of glycosyltransferases, providing strong support for designing drug molecules targeting specific glycosyltransferases or modifying enzymes to improve their catalytic efficiency.

[0020] The following is a sample code that shows how to use the model in this technical solution to identify active sites on protein and nucleic acid sequence databases: # Protein and nucleic acid sequence databases and non-redundant protein feature sequence databases protein_db = [...] nucleic_acid_db = [...] non_redundant_protein_db = [...] # Load pre-trained protein language model, multimodal cross-attention model and multiple sequence alignment model protein_language_model = ... multimodal_cross_attention_model = ... multiple_sequence_alignment_model = ... # Convert sequence to protein language description protein_lang_desc = protein_language_model.transform(protein_db +nucleic_acid_db + non_redundant_protein_db) # Prepare the sample features of the three-dimensional structure of glycosyltransferase and the sample features of the enzymatic reaction gt_structure_features = ... enzyme_reaction_features = ... # Align and fuse using multimodal cross-attention model enzyme_repr_fusion = multimodal_cross_attention_model.align_and_fuse( protein_lang_desc, gt_structure_features, enzyme_reaction_features ) # Perform multiple sequence alignment to obtain sequence similarity or homology metrics alignment_results = multiple_sequence_alignment_model.align( protein_db + nucleic_acid_db + non_redundant_protein_db, enzyme_repr_fusion ) # Determine conserved sites and functional segments based on the alignment results conserved_sites = [] functional_segments = [] for result in alignment_results: # Extract sites that remain unchanged or change within a limited range as conservative sites conserved_sites.append(result['conserved_sites']) # Extract structural motifs or functional domains as functional segments functional_segments.append(result['functional_segments']) # Output results print("Conserved Sites:", conserved_sites) print("Functional Segments:", functional_segments) Optionally, the initialization of a matrix having the same length as the alignment result, wherein the rows of the matrix correspond to the positions of the amino acid residues in the sequence and the columns correspond to different amino acid types, includes: Based on the protein language model, the sequences in the protein and nucleic acid sequence databases and the non-redundant protein feature sequence database are Initialize the number of rows of the matrix by the number of amino acid residues corresponding to the sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database according to the protein language model; Initialize the number of columns of the matrix according to the number of amino acid types corresponding to the sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database according to the protein language model; Based on the occurrence frequencies of different amino acid types in the comparison results, the participation degrees of different amino acid residues in different amino acid types are determined to initialize the elements in the matrix targeted by the initialization.

[0021] To this end, a method based on initializing a matrix, in which the rows correspond to the positions of the amino acid residues in the sequence and the columns correspond to the different amino acid types, has the following technical benefits: (1) By initializing a matrix with the same length as the alignment result, the distribution and frequency of amino acid residues in different sequences can be easily counted and compared. This helps to quickly identify conserved sites, functional segments, and possible mutation sites in the sequence, thereby improving the efficiency of data analysis.

[0022] (2) In multiple sequence alignment and evolutionary analysis, it is usually necessary to calculate the similarities and differences of amino acid residues between different sequences. By initializing the matrix, the information of amino acid residues in different sequences can be easily stored and compared, thus supporting more accurate alignment and evolutionary analysis.

[0023] The following is a sample code showing how to initialize the matrix according to the technical solution and fill the matrix elements based on the comparison results: import numpy as np # Alignment results of protein and nucleic acid sequence databases and non-redundant protein feature sequence databases alignment_result = ...# Alignment result data structure, including sequence information and amino acid residue positions, etc. # The protein language model has been trained and provides information on the type and quantity of amino acids amino_acid_types = ['A', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'K', 'L','M', 'N', 'P', 'Q', 'R', 'S', 'T', 'V', 'W', 'Y'] num_amino_acid_types = len(amino_acid_types) # Initialize the matrix, the number of rows is the number of amino acid residues in the alignment result, and the number of columns is the number of amino acid types num_residues = len(alignment_result['sequences'][0])# Assume that all sequences have the same length, take the length of the first sequence matrix = np.zeros((num_residues, num_amino_acid_types)) # Fill in the matrix elements according to the comparison results for seq_index, sequence in enumerate(alignment_result['sequences']): for pos_index, amino_acid in enumerate(sequence): # Find the index of the amino acid type in the amino acid type list aa_index = amino_acid_types.index(amino_acid) # Based on the frequency of occurrence of amino acid types in the comparison results # Accumulate the elements at corresponding positions in the matrix matrix[pos_index, aa_index] += 1 # Output the initialized matrix print("Initialized Matrix:") print(matrix) Optionally, traversing the alignment results, counting the frequency of occurrence of each amino acid type, and initializing the matrix after the initialization to construct a PSSM matrix, includes: Mapping the alignment result into a two-dimensional array, wherein the rows of the two-dimensional array represent sequences and the columns of the two-dimensional array represent the aligned positions of the amino acid residues in the alignment; Based on the columns of the two-dimensional array, the frequency of occurrence of each amino acid type is counted, and the frequency is used as the score of the amino acid and filled into the matrix element where the corresponding amino acid type and the corresponding amino acid residue are located to construct a PSSM matrix.

[0024] To this end, the Position-Specific Scoring Matrix (PSSM) is used to represent the conservation and preference of amino acid residues at each position in the sequence. The PSSM matrix is ​​initialized by traversing the alignment results and counting the frequency of each amino acid type, which has the following technical benefits: (1) The PSSM matrix takes into account the distribution of amino acids at each position in the sequence, and therefore can more accurately reflect the conservation and characteristics of the sequence. Compared with simple amino acid composition statistics, the PSSM matrix provides more detailed positional information, which helps to identify key functional sites and structural features.

[0025] (2) By aligning multiple sequences and constructing a PSSM matrix, we can reveal the conserved amino acid patterns in different sequences. These patterns are often closely related to the function, structure or evolutionary relationship of the protein, so the PSSM matrix helps to enhance the ability to identify important patterns in biological sequences.

[0026] (3) The PSSM matrix can be used as an input feature for a variety of sequence analysis tasks, such as protein structure prediction, functional annotation, and evolutionary tree construction. Its rich position-specific information provides strong support for these tasks and helps improve the accuracy and reliability of the analysis results.

[0027] (4) The data structure of the PSSM matrix is ​​concise and clear, and can be easily understood and utilized by machine learning algorithms. By inputting the PSSM matrix as a feature into a classifier, regression model, or clustering algorithm, more efficient bioinformatics prediction and classification tools can be developed.

[0028] (5) The construction method of PSSM matrix is ​​universal and can be extended to the analysis of other biological sequences (such as DNA and RNA). This helps promote cross-research and cooperation between different biological fields and promote the development of bioinformatics.

[0029] The following is a sample code showing how to traverse the comparison results and construct the PSSM matrix according to the technical solution: import numpy as np from collections import defaultdict # The comparison results are presented in a two-dimensional list, where each sublist represents a sequence alignment_result = [ ['A', 'C', 'D', 'E', 'F'], ['A', 'C', 'G', 'E', 'F'], ['A', 'H', 'D', 'E', 'F'], # ... ] # Initialize the PSSM matrix, the number of rows is the length of the sequence in the alignment result (i.e. the number of positions of the amino acid residues), # The number of columns is the number of amino acid types (standard 20 amino acids) num_positions = len(alignment_result[0]) amino_acid_types = 'ACDEFGHIKLMNPQRSTVWY' num_amino_acid_types = len(amino_acid_types) pssm_matrix = np.zeros((num_positions, num_amino_acid_types)) # Count the frequency of each amino acid type at each position frequency_dict = defaultdict(lambda: defaultdict(int)) for sequence in alignment_result: for pos_index, amino_acid in enumerate(sequence): frequency_dict[pos_index][amino_acid]+= 1 # Calculate frequencies and fill the PSSM matrix for pos_index in range(num_positions): total_count = sum(frequency_dict[pos_index].values()) for aa_index, amino_acid in enumerate(amino_acid_types): if amino_acid in frequency_dict[pos_index]: frequency = frequency_dict[pos_index][amino_acid] / total_count pssm_matrix[pos_index, aa_index] = frequency# Logarithmic transformation of frequency # Output PSSM matrix print("PSSM Matrix:") print(pssm_matrix) Optionally, the frequency of occurrence of each amino acid type is counted based on the columns of the two-dimensional array, and the frequency is used as the score of the amino acid and filled into the matrix element where the corresponding amino acid type and the corresponding amino acid residue are located to construct a PSSM matrix, including: Based on the columns of the two-dimensional array, counting the frequency of occurrence of each amino acid type to construct a position frequency matrix; The position frequency matrix is ​​subjected to logarithmic ratio mapping to obtain a position weight matrix as the constructed PSSM matrix.

[0030] To this end, in the process of constructing the position-specific score matrix (PSSM), the frequency of occurrence of each amino acid type is counted based on the columns of the two-dimensional array, and these frequencies are further mapped to logarithmic ratios to obtain a position weight matrix, which has technical benefits: (1) Directly counting the amino acid frequency may be affected by the length of the sequence and the total number of amino acids, but through logarithmic ratio mapping, the frequency can be converted into a relative ratio, making the feature comparison between different sequences more fair and accurate. This conversion enhances the discrimination of sequence features and helps to identify truly conservative and specific amino acid sites.

[0031] (2) The position weight matrix reflects the preference of different amino acid types at each position. This preference is often closely related to the function, structure or evolutionary relationship of the protein. By analyzing the position weight matrix, we can gain a deep understanding of the distribution of amino acids in the protein sequence, providing strong support for subsequent protein function prediction, structural modeling and evolutionary analysis.

[0032] (3) When the PSSM matrix is ​​used as the input feature of the machine learning algorithm, its quality directly affects the performance of the algorithm. The position weight matrix constructed by logarithmic ratio mapping provides more accurate and discriminative feature information, which helps to improve the accuracy and generalization ability of the machine learning algorithm in protein sequence analysis tasks.

[0033] (4) Since the log-ratio mapping considers the relative proportion of amino acids at specific positions rather than the absolute frequency, the PSSM matrix is ​​more applicable in comparing protein sequences of different species. Even if the protein sequences of different species differ in length and amino acid composition, the PSSM matrix can still reveal their conservation and specificity at key positions.

[0034] The following is a sample code showing how to count amino acid frequencies, perform logarithmic ratio mapping, and construct a PSSM matrix according to the technical solution: import numpy as np from collections import defaultdict import math # The comparison results are presented in the form of a two-dimensional list, each sublist represents a sequence alignment_result = [ ['A', 'C', 'D', 'E', 'F'], ['A', 'C', 'G', 'E', 'F'], ['A', 'H', 'D', 'E', 'F'], # ... ] # Initialize variables num_positions = len(alignment_result[0]) amino_acid_types = 'ACDEFGHIKLMNPQRSTVWY' num_amino_acid_types = len(amino_acid_types) position_frequency_matrix = np.zeros((num_positions, num_amino_acid_types)) pssm_matrix = np.zeros((num_positions, num_amino_acid_types)) # Count the frequency of each amino acid type at each position frequency_dict = defaultdict(lambda: defaultdict(int)) for sequence in alignment_result: for pos_index, amino_acid in enumerate(sequence): frequency_dict[pos_index][amino_acid]+= 1 # Build a position frequency matrix for pos_index in range(num_positions): total_count = sum(frequency_dict[pos_index].values()) for aa_index, amino_acid in enumerate(amino_acid_types): if amino_acid in frequency_dict[pos_index]: frequency = frequency_dict[pos_index][amino_acid] / total_count position_frequency_matrix[pos_index, aa_index] = frequency # Perform logarithmic ratio mapping to obtain the position weight matrix (PSSM matrix) # Pseudo-counts are used here to avoid zero logarithms and smooth the frequency distribution pseudocount = 0.01 for pos_index in range(num_positions): for aa_index in range(num_amino_acid_types): # Calculate background frequency background_frequency = 1 / num_amino_acid_types # Apply pseudo counts observed_frequency = position_frequency_matrix[pos_index, aa_index] +pseudocount expected_frequency = background_frequency + pseudocount * (num_amino_acid_types - 1) # Calculate the logarithmic ratio and fill it into the PSSM matrix pssm_matrix[pos_index, aa_index] = math.log(observed_frequency / expected_frequency, 2) # Output PSSM matrix print("PSSM Matrix:") print(pssm_matrix) Optionally, performing PSSM scoring according to the PSSM matrix to screen out potential mutation sites includes: The PSSM matrix is ​​normalized and then smoothed, and input into the constructed scoring model for PSSM scoring to obtain the PSSM score of the amino acid type at each position in the sequence; Based on the PSSM score, determining the conservation estimate and functional annotation of the amino acid type; According to the conservation estimation and functional annotation of the amino acid type, the degree to which the protein structure is affected by the mutation site is determined; According to the degree to which the protein structure is affected by the mutation site, the random probability and the degree of deviation of the site are calculated to screen out potential mutation sites.

[0035] Therefore, in the process of using the PSSM matrix for scoring to screen out potential mutation sites, the PSSM matrix is ​​normalized and smoothed and input into the scoring model for evaluation, which has the following technical benefits: (1) Normalization can eliminate the dimensional differences between the scores of different amino acid types, making the scoring results more fair and comparable. Smoothing can reduce the impact of noise and outliers on the scoring results and improve the robustness of the scoring. These processing steps together ensure the accuracy and reliability of PSSM scoring.

[0036] (2) PSSM scoring can be used to obtain the score of the amino acid type at each position in the sequence, and then determine the conservation estimate of the amino acid type. Highly conservative amino acid types are less likely to change during evolution, so they may be more important to the structure and function of proteins. At the same time, combined with functional annotation information, we can further understand the role of these conserved amino acids in protein function, providing strong support for subsequent mutation site screening.

[0037] (3) Based on the PSSM scoring results, the conservation estimate and functional annotation of amino acid types can be determined, and the degree to which the protein structure is affected by the mutation site can be evaluated. This evaluation can help identify mutation sites that have a significant impact on protein structure and function, providing strong guidance for subsequent experimental verification and drug design.

[0038] (4) By calculating the probability of randomness and the degree of deviation of the site, potential mutation sites can be quickly screened out. These sites may be key points that affect protein structure and function, and are of great significance for understanding the pathogenesis of diseases and developing new treatments. This screening method not only improves efficiency, but also reduces experimental costs and time costs.

[0039] The following is a sample code that shows how to normalize and smooth the PSSM matrix and perform PSSM scoring to screen out potential mutation sites: import numpy as np from scipy.stats import zscore from scipy.ndimage import gaussian_filter1d # Constructed PSSM matrix pssm_matrix = np.array([ [2.0, -1.0, 0.5, 1.5], [1.0, 2.0, -0.5, -1.0], [-0.5, 1.0, 2.0, 0.5], # ... ]) # Normalization (by column) pssm_normalized = (pssm_matrix - pssm_matrix.mean(axis=0)) / pssm_matrix.std(axis=0) # Smoothing (using Gaussian filter) sigma = 1.0 # smoothing parameter pssm_smoothed = gaussian_filter1d(pssm_normalized, sigma, axis=0) # Build a scoring model def scoring_model(pssm_smoothed): return pssm_smoothed.sum(axis=1)# Add the scores of each amino acid type as the final score # Perform PSSM scoring pssm_scores = scoring_model(pssm_smoothed) # Determine the conservation estimate of amino acid types conservation_scores = np.abs(pssm_scores) # Function annotation information functional_annotations = { 0: 'Function A', 1: 'Function B', 2: 'Function C', # ... } # Based on conservation estimation and functional annotation, determine the extent to which the protein structure is affected by the mutation site impact_scores = conservation_scores # Calculate the accidental probability and deviation degree of the site mean_impact = impact_scores.mean() std_impact = impact_scores.std() z_scores = zscore(impact_scores)# Calculate the Z score as the degree of deviation # Screen out potential mutation sites (here the Z score is greater than 2 or less than -2 as the screening criteria) threshold = 2.0 potential_mutations = np.where(np.abs(z_scores)>threshold)[0] # Output potential mutation sites and related information for mutation_index in potential_mutations: print(f"Potential mutation at position {mutation_index}, " f"conservation score: {conservation_scores[mutation_index]}, " f"functional annotation: {functional_annotations.get(mutation_index,'Unknown')}, " f"Z-score: {z_scores[mutation_index]}") Optionally, constructing an initial mutation library based on the screened potential mutation sites comprises: Obtaining virtual specific primers designed according to simulated target gene sequences; Amplifying the screened potential mutation sites based on the virtual specific primers to generate target sites; Performing high-throughput sequencing on all the target sites to detect changes in the DNA sequence near each target site; Based on the changes in the DNA sequence near the target site, the target site where base substitution, insertion or deletion occurs is determined to form an initial mutation site library.

[0040] To this end, by obtaining virtual specific primers designed according to the simulated target gene sequence, the potential mutation sites screened out are amplified to generate target sites, and then high-throughput sequencing and DNA sequence change detection are performed, and finally the target sites where base substitutions, insertions or deletions occur are determined to form the initial mutation library. This process brings significant technical benefits: 1. Improve screening efficiency and accuracy Precisely locate mutation sites: Directly amplify potential mutation sites through virtual specific primers, avoiding the blindness of extensive screening in traditional methods and improving the accuracy of screening.

[0041] High-throughput sequencing accelerates the process: High-throughput sequencing of target sites can detect changes in a large number of DNA sequences at the same time, significantly speeding up the screening process and reducing time costs.

[0042] 2. Optimizing the properties and functions of enzymes Enhance enzyme stability and activity: By modifying specific mutation sites, key properties of the enzyme, such as thermal stability and catalytic activity, can be optimized, making it more suitable for industrial production or experimental research.

[0043] Broaden the application scope of enzymes: The modified enzymes may have new or enhanced functions, so they can be applied in a wider range of fields, such as biopharmaceuticals, environmental protection, etc.

[0044] 3. Promoting the intelligent development of enzyme engineering Combining artificial intelligence to improve transformation effects: This method makes full use of the advantages of artificial intelligence technology, predicts potential mutation sites through data-driven models, and combines high-throughput sequencing and other technologies for verification, thus realizing the intelligent and precise enzyme transformation.

[0045] Driving innovation in enzyme engineering: This process not only improves the efficiency and accuracy of enzyme modification, but also provides researchers in the field of enzyme engineering with new ideas and methods, helping to promote continued innovation and development in this field.

[0047] The following is a sample code showing how to construct a partial process of initial mutation library based on the potential mutation sites screened: from Bio.Seq import Seq from Bio.SeqUtils import MeltingTemp as mt from Bio import SeqIO import random import subprocess #Target gene sequence target_gene_sequence = "ATGCGTACGTAGCTAGCTAGCTAGC..." # Design virtual specific primers def design_primers(target_sequence, mutation_sites): primers = [] for site in mutation_sites: # Primer length is 20 primer_length = 20 start_pos = max(0, site - primer_length / / 2) end_pos = min(len(target_sequence), site + primer_length / / 2) primer = target_sequence[start_pos:end_pos] primers.append((primer, site)) return primers #Screened out some potential mutation sites potential_mutation_sites = [50, 100, 150] # Here are the example sites # Design primers primers = design_primers(target_gene_sequence, potential_mutation_sites) #Amplification process def simulate_amplification(primers, target_sequence): amplified_sequences = [] for primer, site in primers: #Only amplify the sequence near the target site start_pos = max(0, site - 50)#Amplify 50 bases before and after the site end_pos = min(len(target_sequence), site + 50) amplified_sequence = target_sequence[start_pos:end_pos] amplified_sequences.append((amplified_sequence, site)) return amplified_sequences # Amplify sequence amplified_sequences = simulate_amplification(primers, target_gene_sequence) # High-throughput sequencing def simulate_sequencing(amplified_sequences): sequencing_results = [] for seq, site in amplified_sequences: mutations = [] for i in range(len(seq)): if random.random()<0.01: # 1% mutation rate mutations.append((i, random.choice(['A', 'T', 'C', 'G']))) # Construct sequencing result object result = { 'sequence': seq, 'site': site, 'mutations': mutations } sequencing_results.append(result) return sequencing_results # Sequencing sequencing_results = simulate_sequencing(amplified_sequences) # Build initial mutation library def build_initial_mutation_library(sequencing_results): mutation_library = [] for result in sequencing_results: for mutation in result['mutations']: mutation_library.append({ 'position': result['site'] + mutation[0]- len(result['sequence']) / / 2+ 50,# Calculate the position in the original sequence based on the amplified fragment and the mutation position 'change': mutation[1], 'original': result['sequence'][mutation[0]]#This position in the original sequence is not mutated }) return mutation_library # Build mutation library initial_mutation_library = build_initial_mutation_library(sequencing_results) # Print mutation library for mutation in initial_mutation_library: print(f"Mutation at position {mutation['position']}: {mutation['original']} ->{mutation['change']}") # For example, primer design needs to consider factors such as melting temperature and secondary structure; the amplification and sequencing processes require laboratory operations and external services.

[0048] This sample code covers the basic process from designing primers, simulated amplification, simulated sequencing to constructing an initial mutation library. However, please note that this is just a very simplified model, and actual biological experiments and data analysis are much more complicated. In actual applications, you may need to use more professional bioinformatics tools and software. Optionally, the initial mutation library is subjected to spatial restriction screening to form a virtual screening mutation library, including: Determine the distance range between the active site and key residues, the volume threshold of the substrate binding pocket, the charge distribution on the protein surface, and the distribution of hydrophobic / hydrophilic regions; Generate spatial restriction screening conditions based on the distance range between the active site and key residues, the volume threshold of the substrate binding pocket, the charge distribution on the protein surface, and the distribution of hydrophobic / hydrophilic regions; Based on the spatial restriction screening conditions, the docking energy of the initial mutation library is calculated to perform spatial restriction screening on the initial mutation library to form a virtual screening mutation library.

[0049] Therefore, after constructing the initial mutation library, it is a crucial step to perform spatial restriction screening to form a virtual screening mutation library. This step can significantly improve the accuracy and efficiency of mutation screening by considering the distance range between the active site and the key residues, the volume threshold of the substrate binding pocket, the charge distribution on the protein surface, and the distribution of hydrophobic / hydrophilic regions. The technical benefits are as follows: (1) By comprehensively considering the spatial structural characteristics of the protein, such as the distance between the active site and key residues, the volume of the substrate binding pocket, etc., mutations that have a significant impact on protein function can be screened more accurately. This screening method based on spatial restriction can exclude sites that have mutated but have little effect on protein function, thereby improving the accuracy of screening.

[0050] (2) Spatially restricted screening conditions can ensure that the selected mutations have a reasonable distribution in the three-dimensional structure of the protein, avoiding situations where the mutations are too concentrated or too dispersed. This helps to optimize the quality of the mutation library and makes subsequent experimental verification and data analysis more reliable.

[0051] (3) Spatial restriction screening not only considers the sequence information of the mutation, but also combines the three-dimensional structure information of the protein. This comprehensive approach can provide useful guidance for subsequent structural biology research and help to deeply understand the function and regulatory mechanism of the protein.

[0052] The following is an example code that shows how to screen an initial mutation library based on spatial restriction screening criteria to form a virtual screening mutation library: # Initial mutation library initial_mutation_library = [...] # Define space restriction filter conditions spatial_constraints = { 'distance_range': (min_distance, max_distance), # The distance range between the active site and the key residues 'pocket_volume_threshold': volume_threshold, # Volume threshold of substrate binding pocket 'charge_distribution': charge_dist,# Charge distribution on the protein surface 'hydrophobicity_distribution': hydrophobicity# Distribution of hydrophobic / hydrophilic regions } # Define a function to calculate docking energy and filter based on spatial restriction filter conditions def calculate_docking_energy_and_filter(mutation, spatial_constraints): # Calculate the docking energy between mutation and protein # docking_energy = calculate_docking_energy(mutation) docking_energy = some_energy_calculation_function(mutation) # Check if the mutation meets the space restriction filter condition if (mutation_satisfies_constraints(mutation, spatial_constraints)): return docking_energy, mutation else: return None # Initialize virtual screening mutation library virtual_screening_library = [] # Screen the initial mutation library for mutation in initial_mutation_library: result = calculate_docking_energy_and_filter(mutation, spatial_constraints) if result is not None: docking_energy, filtered_mutation = result # Based on docking energy and / or other criteria, decide whether to add the mutation to the virtual screening mutation library if docking_energy <some_energy_threshold: virtual_screening_library.append(filtered_mutation) # Output virtual screening mutation library print(virtual_screening_library) Optionally, the structural reconstruction of the sequences in the virtual screening mutation library based on the glycosyltransferase tertiary structure prediction model to predict the tertiary structure of the glycosyltransferase comprises: Comparing the sequences in the virtual screening mutation library with known glycosyltransferase reference sequences to determine the specific location and type of the mutation; Based on the glycosyltransferase tertiary structure prediction model, according to the specific position and type of the mutation, the known glycosyltransferase reference sequence is subjected to amino acid replacement and segment adjustment operations to perform structural reconstruction and obtain a predicted glycosyltransferase tertiary structure.

[0053] To this end, based on the virtual screening mutation library, the glycosyltransferase tertiary structure prediction model was used to reconstruct the sequence structure to predict the tertiary structure of glycosyltransferase, which has the following technical benefits: (1) By comparing the sequences in the virtual screening mutation library with the known glycosyltransferase reference sequences, the specific location and type of the mutation can be accurately determined. This enables targeted replacement of amino acids or adjustment of chain segments during the structural reconstruction process, thereby accurately predicting the impact of the mutation on the tertiary structure of the glycosyltransferase.

[0054] (2) The predicted tertiary structure information of glycosyltransferases can provide important guidance for the design of enzyme variants. By understanding how mutations change the spatial conformation of enzymes, enzyme variants with specific catalytic performance or stability can be designed to meet the needs of different application scenarios.

[0055] (3) Traditional studies on the function of glycosyltransferases often rely on experimental methods, such as site-directed mutagenesis, protein crystallization, and X-ray diffraction, which are time-consuming and costly. However, using the tertiary structure prediction model, the effect of mutations on enzyme structure can be quickly simulated on a computer, thereby accelerating the process of functional research.

[0056] (4) The predicted tertiary structure information of glycosyltransferases can also provide a theoretical basis for subsequent molecular dynamics simulations, substrate binding site analysis, and studies on the interaction mechanism between enzymes and substrates.

[0057] The following is an example code that shows how to reconstruct the structure of sequences in the virtual screening mutation library based on the glycosyltransferase tertiary structure prediction model: # Virtual screening of a mutant library and a known glycosyltransferase reference sequence virtual_screening_library = [...] # List of sequences in the virtual screening mutation library reference_sequence = "..." # Known glycosyltransferase reference sequence # Align sequences and determine mutation location and type def align_sequences_and_identify_mutations(virtual_sequence,reference_sequence): # Perform sequence alignment and return mutation position and type # mutations = sequence_alignment(virtual_sequence, reference_sequence) mutations = some_sequence_alignment_function(virtual_sequence,reference_sequence) return mutations # Reconstruct the structure according to the mutation position and type def rebuild_structure_based_on_mutations(reference_structure,mutations): # Replace amino acids or adjust chain segments based on mutation position and type # new_structure = modify_structure(reference_structure, mutations) new_structure = some_structure_modification_function(reference_structure, mutations) return new_structure # Glycosyltransferase tertiary structure prediction model # reference_structure = predict_structure(reference_sequence) # Perform structural reconstruction for each sequence in the virtual screening mutation library predicted_structures = [] for virtual_sequence in virtual_screening_library: mutations = align_sequences_and_identify_mutations(virtual_sequence,reference_sequence) # Tertiary structure of the reference sequence reference_structure = some_predicted_structurenew_structure =rebuild_structure_based_on_mutations(reference_structure, mutations) predicted_structures.append(new_structure) # Output the predicted tertiary structure list of glycosyltransferases print(predicted_structures) Optionally, the molecular docking and dynamic simulation of the predicted glycosyltransferase tertiary structure are performed to analyze the binding pocket of the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand and the number of contacts between each amino acid residue and the ligand, including: The predicted glycosyltransferase tertiary structure was placed Based on the constructed molecular search conformational space, the predicted tertiary structure of glycosyltransferase was used as a receptor for molecular docking to analyze the binding pocket between the protein and the ligand; Based on the distance between the center of mass of each residue in the binding pocket and the geometric center of the binding pocket, a dynamic simulation is performed on the docked protein-ligand complex to determine the dynamic interactions between the amino acid residues and the ligand and the number of contacts between each amino acid residue and the ligand.

[0058] To this end, molecular docking and dynamic simulation of the predicted tertiary structure of glycosyltransferases are key steps to gain a deeper understanding of the interaction mechanism between enzymes and ligands. This technical solution can not only reveal the characteristics of the binding pocket between proteins and ligands, but also dynamically analyze the interactions and contact numbers between amino acid residues and ligands, thereby providing important information for enzyme functional research, inhibitor design, and drug development. The following are the main benefits of this technical solution: (1) Molecular docking can accurately locate the key region where glycosyltransferase binds to the ligand, namely the binding pocket. This helps to understand how the ligand enters and binds to the enzyme, providing target information for subsequent inhibitor design and drug development.

[0059] (2) Kinetic simulation can simulate the dynamic interaction process between enzymes and ligands in a real environment, including hydrogen bonds, van der Waals forces, and electrostatic interactions between amino acid residues and ligands. This dynamic analysis can better reflect the actual situation than static structural analysis and helps to deeply understand the catalytic mechanism and substrate specificity of the enzyme.

[0060] (3) Through kinetic simulation, the number of contacts between each amino acid residue and the ligand can be calculated, thereby identifying the amino acid residues that play a key role in ligand binding and enzyme catalysis. This information is of great significance for enzyme function research, site-directed mutagenesis, and enzyme engineering.

[0061] (4) The information on the characteristics of the binding pocket and key amino acid residues can provide important guidance for drug design. By designing ligands that can tightly bind to the binding pocket and form favorable interactions with key amino acid residues, drug molecules with efficient inhibitory activity or catalytic properties can be developed.

[0062] (5) Traditional studies on enzyme-ligand interactions often rely on experimental methods such as crystallography and nuclear magnetic resonance, which are time-consuming and costly. Molecular docking and dynamics simulations can be performed quickly on computers, greatly accelerating the research process.

[0063] The following is an example code to show how to perform molecular docking and dynamics simulation on the predicted tertiary structure of glycosyltransferase: # Predicted tertiary structures of glycosyltransferases (receptors) and ligand molecules receptor_structure = "..."# Predicted glycosyltransferase tertiary structure file path ligand_molecule = "..."# Ligand molecule file path # Molecular docking def perform_molecular_docking(receptor, ligand): # Molecular docking tools, such as AutoDock, Dock6, etc. # docking_result = dock(receptor, ligand) docking_result = some_docking_function(receptor, ligand) return docking_result # Dynamics simulation def perform_dynamics_simulation(docking_result): # Dynamics simulation tools, such as GROMACS, Amber, etc. # simulation_result = simulate(docking_result) simulation_result = some_simulation_function(docking_result) return simulation_result # Perform molecular docking docking_result = perform_molecular_docking(receptor_structure,ligand_molecule) # Perform dynamic simulation simulation_result = perform_dynamics_simulation(docking_result) # Analyze the simulation results and extract information such as the binding pocket characteristics of the protein and the ligand, the dynamic interactions between the amino acid residues and the ligand, and the number of contacts # Process the simulation results and extract the required information #binding_pocket_features = analyze_binding_pocket(simulation_result) # amino_acid_interactions = analyze_amino_acid_interactions(simulation_result) # contact_numbers = analyze_contact_numbers(simulation_result) binding_pocket_features = some_analysis_function_for_binding_pocket(simulation_result) amino_acid_interactions = some_analysis_function_for_amino_acid_interactions(simulation_result) contact_numbers = some_analysis_function_for_contact_numbers(simulation_result) # Output analysis results print("Binding Pocket Features:", binding_pocket_features) print("Amino Acid Interactions:", amino_acid_interactions) print("Contact Numbers:", contact_numbers) Optionally, the method of screening out the optimal mutation site from the virtual screening mutation library according to the binding pocket of the protein and the ligand and determining the dynamic interaction between the amino acid residues and the ligand and the number of contacts between each amino acid residue and the ligand to transform the tertiary structure of the glycosyltransferase comprises: According to the binding pocket of the protein and the ligand, and the dynamic interaction between the amino acid residues and the ligand, and the number of contacts between each amino acid residue and the ligand, each mutation site of the virtual screening mutation library is estimated to obtain an evaluation value of the degree of influence of the mutation on the protein structure, the shape of the binding pocket, and the interaction between the amino acid residues and the ligand; According to the impact evaluation value, the mutation sites that can improve ligand binding affinity, protein stability or enzyme catalytic activity are screened out from the virtual screening mutation library as the optimal mutation sites; A set mutation strategy is obtained to generate simulated mutant genes based on the mutation sites screened out in the virtual screening mutation library to transform the tertiary structure of the glycosyltransferase.

[0064] Therefore, in the research and modification of glycosyltransferases, it is an efficient and accurate strategy to identify and introduce beneficial mutation sites through virtual screening of mutation libraries. This technical solution is based on the in-depth analysis of the binding pocket characteristics of proteins and ligands, the dynamic interactions between amino acid residues and ligands, and the number of contacts. It aims to screen out the optimal mutation sites from the virtual screening mutation library to achieve effective modification of the tertiary structure of glycosyltransferases. The benefits of this technical solution mainly include: (1) Traditional mutation screening methods often rely on experimental methods, which are time-consuming and costly. However, this technical solution can quickly identify potential candidates from a large number of mutation sites through computer simulation and virtual screening, significantly improving the screening efficiency.

[0065] (2) Through detailed analysis of the protein-ligand binding pocket and dynamic simulation of the interaction between amino acid residues and ligands, this technical solution can accurately predict the impact of each mutation site on protein structure, binding pocket shape and interaction, thereby providing a strong basis for the selection of mutation sites.

[0066] (3) By screening out the optimal mutation sites, this technical solution is expected to improve the ligand binding affinity, protein stability or enzyme catalytic activity of glycosyltransferases, thereby optimizing the performance of the protein and meeting specific application requirements.

[0067] (4) The virtual screening results provided by this technical solution can provide clear guidance for subsequent experimental verification, reduce the blindness and uncertainty in the experiment, and improve the success rate of the experiment.

[0068] (5) Through the precise modification of glycosyltransferases, this technical solution is expected to provide enzyme preparations with better performance for drug development, enzyme engineering and other fields, and promote the progress and development of related fields.

[0069] The following is a sample code that shows how to select the optimal mutation site from the virtual screening mutation library according to the technical solution and generate simulated mutant genes: # Information about the protein-ligand binding pocket, dynamic interactions between amino acid residues and ligands, and the number of contacts # And a virtual screening mutation library containing multiple mutation sites # Virtual screening mutation library example mutation_library = [ {"position": 10, "mutation": "A->T", "score": 0.5},# Mutation position and its score {"position": 25, "mutation": "C->G", "score": 0.8}, {"position": 50, "mutation": "T->A", "score": 0.3}, # ... other mutation points ] # Evaluate the effects of mutations on protein structure, binding pockets, and interactions def evaluate_mutation(mutation, binding_pocket, interactions, contact_numbers): return mutation["score"] # Estimate each mutation point in the virtual screening mutation library evaluated_mutations = [] for mutation in mutation_library: score = evaluate_mutation(mutation, binding_pocket, interactions, contact_numbers) evaluated_mutations.append({"mutation": mutation, "evaluated_score":score}) # Filter out the best mutation point based on the evaluation value optimal_mutation = max(evaluated_mutations, key=lambda x: x["evaluated_score"])["mutation"] # Get the set mutation strategy mutation_strategy = "simple_replacement" # Generate mutant genes based on the selected mutation sites and mutation strategies # Generate mutant genes def generate_mutated_gene(mutation, strategy): print(f"Selected mutation: {mutation['position']} {mutation['mutation']} (Strategy: {strategy})") # Generate mutant genes generate_mutated_gene(optimal_mutation, mutation_strategy) The embodiment of the present application further provides an electronic device, which includes a memory and a processor, wherein a computer executable program is stored in the memory, and the processor is used to run the computer executable program to implement the following steps: Obtain the sequence of glycosyltransferase to train the AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model; Based on the active site recognition model, multiple sequence alignment is performed on the protein and nucleic acid sequence database and the non-redundant protein feature sequence database to obtain alignment results, wherein the alignment results include conserved sites and functional segments of sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database; Initialize a matrix with the same length as the alignment result. The rows of the matrix correspond to the positions of the amino acid residues in the sequence, and the columns correspond to different amino acid types. Traversing the alignment results, counting the frequency of occurrence of each amino acid type, initializing the matrix after the initialization, and constructing a PSSM matrix; According to the PSSM matrix, PSSM scoring is performed to screen out potential mutation sites; Constructing an initial mutation library based on the screened potential mutation sites; Performing spatial restriction screening on the initial mutation library to form a virtual screening mutation library; Based on the glycosyltransferase tertiary structure prediction model, the sequences in the virtual screening mutation library are structurally reconstructed to predict the tertiary structure of the glycosyltransferase; Molecular docking and dynamic simulation were performed on the predicted tertiary structure of glycosyltransferase to analyze the binding pocket between the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand, as well as the number of contacts between each amino acid residue and the ligand; According to the binding pocket between the protein and the ligand, and the dynamic interaction between the amino acid residues and the ligand, and the number of contacts between each amino acid residue and the ligand, the optimal mutation site is screened from the virtual screening mutation library to transform the tertiary structure of the glycosyltransferase.

[0070] For exemplary explanations of the various steps executed in the above electronic device, please refer to the above Figure 1 Description.

[0071] The present application also provides an artificial intelligence-based enzyme modification device, which includes: The first program unit is used to obtain the sequence of glycosyltransferase to train the AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model; The second program unit is used to perform multiple sequence alignment on the protein and nucleic acid sequence database and the non-redundant protein characteristic sequence database based on the active site recognition model to obtain an alignment result, wherein the alignment result includes conserved sites and functional segments of sequences in the protein and nucleic acid sequence database and the non-redundant protein characteristic sequence database; The third program unit is used to initialize a matrix with the same length as the comparison result, wherein the rows of the matrix correspond to the positions of the amino acid residues in the sequence, and the columns correspond to different amino acid types; A fourth program unit is used to traverse the comparison results, count the frequency of occurrence of each amino acid type, initialize the matrix element of the initial matrix, and construct a PSSM matrix; A fifth program unit is used to perform PSSM scoring according to the PSSM matrix to screen out potential mutation sites; The sixth program unit is used to construct an initial mutation library based on the screened potential mutation sites; a seventh program unit, for performing spatial restriction screening on the initial mutation library to form a virtual screening mutation library; An eighth program unit is used to perform structural reconstruction on the sequences in the virtual screening mutation library based on the glycosyltransferase tertiary structure prediction model to predict the tertiary structure of glycosyltransferase; The ninth program unit is used to perform molecular docking and dynamic simulation on the predicted tertiary structure of glycosyltransferase to analyze the binding pocket between the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand and the number of contacts between each amino acid residue and the ligand; The tenth program unit is used to screen out the optimal mutation site from the virtual screening mutation library according to the binding pocket of the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand and the number of contacts between each amino acid residue and the ligand, so as to transform the tertiary structure of the glycosyltransferase.

[0072] For exemplary explanations of the various steps performed in the above-mentioned device, please refer to the above-mentioned Figure 1 Description.

[0073] 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. An enzyme modification method based on artificial intelligence, characterized in that: include: Obtain the sequence of glycosyltransferase to train the AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model; Based on the active site recognition model, multiple sequence alignment is performed on the protein and nucleic acid sequence database and the non-redundant protein feature sequence database to obtain alignment results, wherein the alignment results include conserved sites and functional segments of sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database; Initialize a matrix with the same length as the alignment result. The rows of the matrix correspond to the positions of the amino acid residues in the sequence, and the columns correspond to different amino acid types. Traversing the alignment results, counting the frequency of occurrence of each amino acid type, initializing the matrix after the initialization, and constructing a PSSM matrix; According to the PSSM matrix, PSSM scoring is performed to screen out potential mutation sites; Constructing an initial mutation library based on the screened potential mutation sites; Performing spatial restriction screening on the initial mutation library to form a virtual screening mutation library; Based on the glycosyltransferase tertiary structure prediction model, the sequences in the virtual screening mutation library are structurally reconstructed to predict the tertiary structure of the glycosyltransferase; Molecular docking and dynamic simulation were performed on the predicted tertiary structure of glycosyltransferase to analyze the binding pocket between the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand, as well as the number of contacts between each amino acid residue and the ligand; According to the binding pocket between the protein and the ligand, and the dynamic interaction between the amino acid residues and the ligand, and the number of contacts between each amino acid residue and the ligand, the optimal mutation site is screened from the virtual screening mutation library to transform the tertiary structure of the glycosyltransferase.

2. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The step of obtaining the sequence of the glycosyltransferase to train the AlphaFold2 model to obtain a glycosyltransferase tertiary structure prediction model includes: Based on the row-level and column-level attention modules in the AlphaFold2 model, the dependencies between amino acids are captured to obtain a predicted pairing table, which reflects the possible interactions and pairing relationships between amino acids; based on the structure generation module in the AlphaFold2 model, the predicted pairing table is expanded to obtain the co-evolutionary relationship characteristics between residues and predict the tertiary structure of glycosyltransferase based on this.

3. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The active site recognition model includes a protein language model, a multimodal cross-attention model, and a multiple sequence alignment model. The active site recognition model performs multiple sequence alignment on the protein and nucleic acid sequence database and the non-redundant protein feature sequence database to obtain an alignment result. The alignment result includes conserved sites and functional segments of sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database, including: Based on the protein language model, the sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database are converted into a protein language description of each amino acid residue; Based on the multimodal cross attention frame model, the protein language descriptions corresponding to the protein and nucleic acid sequence databases and the non-redundant protein feature sequence databases are aligned with the three-dimensional structure sample features of glycosyltransferases and the sample feature representations of enzymatic reactions, and then fused to obtain enzyme representation fusion features; Based on the multiple sequence alignment model, the protein and nucleic acid sequence databases and the non-redundant protein feature sequence databases are aligned to obtain sequence similarity or homology metrics, and the sites, structural motifs or functional domains that remain unchanged or vary within a set limited range in different sequences are determined based on the sequence similarity or homology metrics, so as to determine the conserved sites of the sequences in the protein and nucleic acid sequence databases and the non-redundant protein feature sequence database based on the sites that remain unchanged or vary within a set limited range, and to determine the functional segments of the sequences in the protein and nucleic acid sequence databases and the non-redundant protein feature sequence database based on the structural motifs or functional domains.

4. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The initialization of a matrix of the same length as the alignment result, wherein the rows of the matrix correspond to the positions of the amino acid residues in the sequence, and the columns correspond to different amino acid types, including: Based on the protein language model, the sequences in the protein and nucleic acid sequence databases and the non-redundant protein feature sequence database are Initialize the number of rows of the matrix by the number of amino acid residues corresponding to the sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database according to the protein language model; Initialize the number of columns of the matrix according to the number of amino acid types corresponding to the sequences in the protein and nucleic acid sequence database and the non-redundant protein feature sequence database according to the protein language model; Based on the occurrence frequencies of different amino acid types in the comparison results, the participation degrees of different amino acid residues in different amino acid types are determined to initialize the elements in the matrix targeted by the initialization.

5. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The traversing of the comparison results, counting the frequency of occurrence of each amino acid type, and initializing the matrix after the initialization to construct a PSSM matrix, includes: Mapping the alignment result into a two-dimensional array, wherein the rows of the two-dimensional array represent sequences and the columns of the two-dimensional array represent the aligned positions of the amino acid residues in the alignment; Based on the columns of the two-dimensional array, the frequency of occurrence of each amino acid type is counted, and the frequency is used as the score of the amino acid and filled into the matrix element where the corresponding amino acid type and the corresponding amino acid residue are located to construct a PSSM matrix.

6. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The frequency of occurrence of each amino acid type is counted based on the columns of the two-dimensional array, and the frequency is used as the score of the amino acid and filled into the matrix element where the corresponding amino acid type and the corresponding amino acid residue are located to construct a PSSM matrix, including: Based on the columns of the two-dimensional array, counting the frequency of occurrence of each amino acid type to construct a position frequency matrix; The position frequency matrix is ​​subjected to logarithmic ratio mapping to obtain a position weight matrix as the constructed PSSM matrix.

7. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: According to the PSSM matrix, PSSM scoring is performed to screen out potential mutation sites, including: The PSSM matrix is ​​normalized and then smoothed, and input into the constructed scoring model for PSSM scoring to obtain the PSSM score of the amino acid type at each position in the sequence; Based on the PSSM score, determine the conservation estimate and functional annotation of the amino acid type; According to the conservation estimation and functional annotation of the amino acid type, the degree to which the protein structure is affected by the mutation site is determined; According to the degree to which the protein structure is affected by the mutation site, the random probability and the degree of deviation of the site are calculated to screen out potential mutation sites.

8. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The performing spatial restriction screening on the initial mutation library to form a virtual screening mutation library comprises: Determine the distance range between the active site and key residues, the volume threshold of the substrate binding pocket, the charge distribution on the protein surface, and the distribution of hydrophobic / hydrophilic regions; Generate spatial restriction screening conditions based on the distance range between the active site and key residues, the volume threshold of the substrate binding pocket, the charge distribution on the protein surface, and the distribution of hydrophobic / hydrophilic regions; Based on the spatial restriction screening conditions, the docking energy of the initial mutation library is calculated to perform spatial restriction screening on the initial mutation library to form a virtual screening mutation library.

9. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The structural reconstruction of the sequences in the virtual screening mutation library based on the glycosyltransferase tertiary structure prediction model to predict the tertiary structure of the glycosyltransferase comprises: Comparing the sequences in the virtual screening mutation library with known glycosyltransferase reference sequences to determine the specific location and type of the mutation; Based on the glycosyltransferase tertiary structure prediction model, according to the specific position and type of the mutation, the known glycosyltransferase reference sequence is subjected to amino acid replacement and segment adjustment operations to perform structural reconstruction and obtain a predicted glycosyltransferase tertiary structure.

10. The enzyme modification method based on artificial intelligence according to claim 1, characterized in that: The molecular docking and dynamic simulation of the predicted glycosyltransferase tertiary structure are performed to analyze the binding pocket between the protein and the ligand and determine the dynamic interaction between the amino acid residues and the ligand and the number of contacts between each amino acid residue and the ligand, including: The predicted glycosyltransferase tertiary structure was placed Based on the constructed molecular search conformational space, the predicted tertiary structure of glycosyltransferase was used as a receptor for molecular docking to analyze the binding pocket between the protein and the ligand; Based on the distance between the center of mass of each residue in the binding pocket and the geometric center of the binding pocket, a dynamic simulation is performed on the docked protein-ligand complex to determine the dynamic interactions between the amino acid residues and the ligand and the number of contacts between each amino acid residue and the ligand.

Images