Amino acid and RMSD features improve glycosidase mutation prediction method and its application

Abstract

The present invention relates to the technical field of enzyme activity prediction, and specifically to a method for improving the prediction of glycosidase mutations using amino acid and RMSD features and its application. The present invention comprises the following steps: selecting amino acid features and constructing a QSAR dictionary, determining feature values and weights and calculating amino acid contribution scores, calculating a mutant-wild-type offset metric and reflecting local influences, introducing protein RMSD as a global feature, setting an enzyme activity binary classification target and constructing a MutShift prediction model, solving the data imbalance problem and optimizing model performance, introducing a QSAR dictionary and protein RMSD global features and training a MutShift model. The present invention achieves accurate prediction of glycosidase mutations and improves the accuracy of the enzyme activity prediction model by constructing a QSAR dictionary and optimizing an amino acid residue feature set, combining local and global structural impact assessments of mutants and wild types.

Description

Technical Field

[0001] The present invention relates to the technical field of enzyme activity prediction, and in particular to a method for predicting glycosidase mutations by improving amino acid and RMSD features and an application thereof. Background Art

[0002] Glycosyltransferases catalyze the transfer of sugar moieties from donors (such as UDP-sugar) to acceptor molecules, generating a diverse array of glycosylated products. In recent years, optimizing their function through mutation has become a hot topic in enzyme engineering. However, the activity of glycosyltransferases is influenced by multiple factors, making predicting the effects of mutations complex. While effective, experimental screening methods are costly and time-consuming, making them unsuitable for large-scale application. To address this issue, researchers have proposed a robust QSAR local model construction method, as described in Chinese patent CN115527619A. This method utilizes support vector machine regression and particle swarm optimization algorithms to establish robust and reliable local models, which are then combined into a consistent model to improve predictive performance. However, existing machine learning-based enzyme mutation prediction methods still face challenges: feature construction is incomplete, focusing primarily on a single dimension; and insufficient analysis of differences between mutants and wild-type results fails to fully reveal the specific effects of mutations on enzyme activity. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method for predicting glycosidase mutations by improving amino acid and RMSD features and its application.

[0004] The technical solution adopted in the present invention is as follows:

[0005] A method for predicting glycosidase mutations based on amino acid and RMSD features, comprising the following steps:

[0006] S1. Select amino acid features and construct QSAR dictionary:

[0007] S11. Collect and organize data on the physicochemical properties of amino acids, including polarity, hydrophobicity, acidity and alkalinity, molecular weight, volume, aromaticity, and donor and acceptor characteristics;

[0008] S12, assigning characteristic values to each amino acid based on the data and constructing a QSAR dictionary;

[0009] S2. Determine the eigenvalues and weights and calculate the amino acid contribution score:

[0010] S21. Determine the number of amino acids in the enzyme and the number of features of each amino acid;

[0011] S22, obtain characteristic values from the QSAR dictionary;

[0012] S23. Determine feature weights through experimental data or model training;

[0013] S24, calculating amino acid contribution scores;

[0014] S3. Calculate the deviation metric between mutant and wild type and reflect the local impact:

[0015] S31. Calculate the amino acid contribution scores of the mutant and wild-type enzymes respectively;

[0016] S32, calculating the deviation metric for each amino acid residue;

[0017] S4. Introducing protein RMSD as a global feature:

[0018] S41. Determine the α-carbon atom coordinates of the wild-type and mutant enzymes;

[0019] S42. Compare the differences between the mutant and wild-type structures and calculate the RMSD;

[0020] S5. Set the enzyme activity binary classification target and build the MutShift prediction model:

[0021] S51. Collect the enzyme activity data of glycosyltransferase, including the enzyme activity values of mutants and wild type;

[0022] S52. Setting a binary classification target according to the change direction of the enzyme activity value;

[0023] S53, using learning algorithms and building MutShift prediction models;

[0024] S6. Solve the data imbalance problem and optimize model performance:

[0025] S61, analyzing enzyme activity data to determine the number of positive and negative samples;

[0026] S62. Use oversampling technology to increase the number of positive samples and balance the two types of samples;

[0027] S63. Perform model training on the balanced data and evaluate model performance;

[0028] S7. Introduce QSAR dictionary and protein RMSD global features and train MutShift model:

[0029] S71, QSAR dictionary features are used as input features and introduced into the MutShift prediction model;

[0030] S72. Use the training dataset to train the model and use the validation dataset to evaluate the model performance;

[0031] S73. Adjust model parameters according to the evaluation results to optimize model performance;

[0032] S74. Use the optimized model to predict enzyme activity and verify the accuracy of the prediction results.

[0033] This technical solution achieves accurate prediction of glycosidase mutations by constructing a QSAR dictionary, calculating amino acid contribution scores, evaluating local and global structural effects, and building and optimizing prediction models. Specifically, amino acid features were selected and a QSAR dictionary was constructed to collect and organize the physicochemical property data of amino acids; these feature values and their weights were determined, and the contribution scores of amino acids in enzymes were calculated based on them; the impact of mutations on the local structure of enzymes was reflected by calculating the offset measure of the amino acid contribution scores between mutants and wild types; the RMSD was calculated to quantify the impact of mutations on the global structure of proteins, thereby more comprehensively evaluating the effects of mutations; a binary classification target for enzyme activity was set, and a MutShift prediction model was constructed using a learning algorithm to predict changes in enzyme activity of glycosyltransferases based on the characteristics and mutation status of amino acid residues; to address the data imbalance problem, the number of positive samples was increased through oversampling technology, so that the model could better learn and predict the increase in enzyme activity; QSAR dictionary features were introduced as input features into the MutShift prediction model, and the training dataset was used for training, while the validation dataset was used to evaluate the model performance; the model parameters were adjusted according to the evaluation results to optimize the model performance, and finally the optimized model was used to predict enzyme activity, and the accuracy of the prediction results was verified.

[0034] In addition, the method for improving glycosidase mutation prediction based on amino acid and RMSD features according to the present invention also has the following additional technical features:

[0035] According to one embodiment of the present invention, in the step S11, the physicochemical property data of amino acids are collected and sorted. The interaction between the active site of the enzyme and the substrate is affected by polarity, hydrophobicity and acidity and alkalinity, which are used to predict the change in substrate binding ability after mutation.

[0036] Molecular weight and volume affect the position and force balance of amino acids in the three-dimensional structure and are used to predict the effect of mutations on enzyme structural stability;

[0037] Hydrogen bond donors and acceptors directly affect the chemical reactivity of the catalytic center and are used to determine whether the enzyme retains catalytic ability after mutation;

[0038] Aromaticity and hydrophobicity reflect the effect of amino acid mutations on the global conformation and are used to predict whether the enzyme is inactivated due to mutation.

[0039] In this technical solution, polarity describes the uniformity of molecular charge distribution; polar amino acids form hydrogen bonds and interact with water molecules or polar groups, including Ser, Thr, Asn, Gln, Tyr, His, Lys, Arg, Asp, Glu; non-polar amino acids include Ala, Val, Ile, Leu, Phe, Met, Trp, Cys, Pro, Gly; polar amino acids often appear in enzyme active sites, affecting binding and catalytic efficiency; non-polar amino acids maintain enzyme stability; hydrophobicity describes the behavior of molecules in water; hydrophobic amino acids aggregate and stay away from polar solvents, including Ala, Val, Ile, Leu, Phe, Met, Trp, Pro; hydrophilic amino acids are others; hydrophobicity affects the characteristics of substrate binding sites and is important for substrate selectivity; in protein folding, hydrophobic amino acids form a hydrophobic core to maintain enzyme structural stability; acidity and alkalinity describe the properties of amino acid side chains; acidic amino acids include Asp, Glu; basic amino acids include Lys, Arg, His; neutral amino acids are others; acidic or basic amino acids participate in acid-base catalytic reactions at the active site and affect charge interactions; molecular weight reflects the volume and structural complexity of amino acids, and is standardized to Trp (204); large amino acids cause steric hindrance and affect binding; small amino acids allow structural changes; volume affects the spatial conformation of the enzyme and the accessibility of the active site, and is standardized to Trp (163); large amino acids restrict substrate entry; small amino acids enhance enzyme flexibility; aromaticity describes whether the amino acid has an aromatic side chain, including Phe, Tyr, and Trp; non-aromatic amino acids are others; aromatic groups bind to substrates through π-π stacking or van der Waals interactions and participate in substrate recognition; whether the amino acid side chain contains hydrogen bond donors or acceptors, Ser, Thr, Asn, etc. are typical donors and acceptors, and Phe, Ala, etc. do not have this property; hydrogen bond donors and acceptors form a network at the active site, which is an important factor in the specific binding of enzymes to substrates, and the number affects the binding energy and catalytic efficiency.

[0040] According to one embodiment of the present invention, in the calculation of the amino acid contribution score in step S24, the amino acid contribution score The calculation formula is as follows:

[0041]

[0042] in: is the number of amino acids in the enzyme;

[0043] The number of characteristics for each amino acid, including polarity and hydrophobicity;

[0044] For the The j-th feature value of the amino acid is taken from the amino acid residue dictionary;

[0045] For the The weight of a feature indicates the degree of influence of the feature on the change of enzyme activity; the weight is obtained through model training or experimental data;

[0046] is a bias term used to adjust the predicted baseline value.

[0047] This technical solution quantifies the degree to which the characteristics of each amino acid in the enzyme (including polarity, hydrophobicity, etc.) affect the change in enzyme activity by calculating the amino acid contribution score. The score calculation formula involves the number of amino acids, the number of features, the feature value, the feature weight, and the bias term.

[0048] According to one embodiment of the present invention, in the step S32 of calculating the deviation metric of each amino acid residue, the deviation metric between the mutant and the wild type is The calculation formula is as follows:

[0049]

[0050] in: The mutant and wild type The difference in amino acid contribution scores on each amino acid residue directly reflects the local impact of the mutation on the enzyme molecule;

[0051] For mutants The amino acid contribution score of each amino acid residue;

[0052] The wild type The amino acid contribution score of each amino acid residue.

[0053] This technical solution calculates the difference in amino acid contribution scores between the mutant and the wild type at each amino acid residue to intuitively reflect the local impact of the mutation on the enzyme molecule, thereby deriving a deviation metric.

[0054] According to one embodiment of the present invention, in the comparison of the differences between the mutant and the wild type in step S42, the root mean square deviation between the mutant and the wild type is The calculation formula is as follows:

[0055]

[0056] in: is the α-carbon number, calculated by comparing the wild type with the mutant , which can quantify the effect of mutations on the global structure of proteins;

[0057] and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate;

[0058] and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate;

[0059] and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate.

[0060] This technical solution quantifies the effect of mutations on the global structure of proteins by calculating the root mean square deviation of α-carbon atoms in the mutant and wild-type protein structures.

[0061] According to one embodiment of the present invention, in the change direction of the enzyme activity value in step S52, the enzyme activity value is derived from the data of the public PDB website and laboratory measurement, and the binary classification target is:

[0062] Target value 1, representing an increase in the mutant enzyme activity, is used to characterize enhanced activity;

[0063] The target value of 0 indicates decreased enzyme activity in the mutant and is used to characterize decreased activity.

[0064] This technical solution determines the enzyme activity value through the public PDB website and laboratory measurement data, and uses a binary classification target (target value 1 indicates increased enzyme activity, and target value 0 indicates decreased enzyme activity) to determine the direction of change in the mutant enzyme activity, that is, whether the activity is enhanced or decreased.

[0065] According to one embodiment of the present invention, in step S62, when the oversampling technique is used to increase the number of positive samples, the enzyme activity data is imbalanced, that is, the number of positive samples is much smaller than the number of negative samples, wherein:

[0066] The number of positive samples is ;

[0067] The number of negative samples is ;

[0068] The goal of oversampling is to increase the number of positive samples so that the number of samples in the two categories is balanced:

[0069] Number of new samples = - = (Number of positive samples, number of newly added samples).

[0070] This technical solution increases the number of positive samples through oversampling technology to balance the difference between the number of positive samples and the number of negative samples in enzyme activity data, making the number of samples in the two categories equal.

[0071] According to one embodiment of the present invention, the QSAR dictionary features of step S71 are used as input features, and an enzyme activity prediction model is introduced through local mutant and wild-type offset metrics and global RMSD to quantify the impact of mutants on enzyme structure and establish a connection with enzyme activity changes for predicting glycoside transferase activity.

[0072] This technical solution introduces QSAR dictionary features as input, combines local mutant and wild-type offset metrics with global RMSD, and establishes an enzyme activity prediction model to quantify the impact of mutants on enzyme structure and establish a connection with enzyme activity changes, thereby predicting glycosyltransferase activity.

[0073] To achieve the above objectives, the present invention also provides an application of amino acid and RMSD features to enhance glycosidase mutation prediction.

[0074] An application for improving glycosidase mutation prediction by using amino acid and RMSD features, which uses the amino acid and RMSD features to improve the glycosidase mutation prediction method.

[0075] According to one embodiment of the present invention, the physicochemical property data of a single amino acid are selected in the initial stage, and then the number of features is gradually increased to form a more complex feature set; the feature set is input into the MutShift prediction model for predicting enzyme activity.

[0076] This technical solution improves the glycosidase mutation prediction method by adopting amino acid and RMSD features. In the initial stage, the physicochemical property data of a single amino acid is selected, and the number of features is gradually increased to form a complex feature set. These feature sets are then input into the MutShift prediction model to predict the enzymatic activity of glycosidases.

[0077] Compared with the prior art, the present invention has the following beneficial effects:

[0078] (1) By constructing a QSAR dictionary and optimizing the amino acid residue feature set, combined with the local and global structural impact assessment of mutants and wild-type, we can achieve accurate prediction of glycosidase mutations and improve the accuracy of enzyme activity prediction models;

[0079] (2) By introducing oversampling technology to solve the data imbalance problem and optimizing the MutShift prediction model parameters, this technical solution can more effectively learn and predict the increase in enzyme activity, providing support for the mutation research and application of glycosidases. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a flowchart of the method of the present invention.

[0081] Figure 2 is a diagram showing the effects of local and global characteristics of the mutant on enzyme activity.

[0082] Figure 3 This is a diagram of the construction and verification process of the Mutshift prediction model.

[0083] Figure 4 It is a training and screening diagram of the QSAR amino acid residue dictionary.

[0084] Figure 5 This is the evaluation and screening diagram of the Mutshift prediction model. DETAILED DESCRIPTION

[0085] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0086] like Figure 1 As shown, this embodiment provides a method for predicting glycosidase mutations using amino acid and RMSD features, comprising the following steps:

[0087] S1. Select amino acid features and construct QSAR dictionary:

[0088] S11. Collect and organize data on the physicochemical properties of amino acids, including polarity, hydrophobicity, acidity and alkalinity, molecular weight, volume, aromaticity, and donor and acceptor characteristics;

[0089] S12, assigning characteristic values to each amino acid based on the data and constructing a QSAR dictionary;

[0090] S2. Determine the eigenvalues and weights and calculate the amino acid contribution score:

[0091] S21. Determine the number of amino acids in the enzyme and the number of features of each amino acid;

[0092] S22, obtain characteristic values from the QSAR dictionary;

[0093] S23. Determine feature weights through experimental data or model training;

[0094] S24, calculating amino acid contribution scores;

[0095] S3. Calculate the deviation metric between mutant and wild type and reflect the local impact:

[0096] S31. Calculate the amino acid contribution scores of the mutant and wild-type enzymes respectively;

[0097] S32, calculating the deviation metric for each amino acid residue;

[0098] S4. Introducing protein RMSD as a global feature:

[0099] S41. Determine the α-carbon atom coordinates of the wild-type and mutant enzymes;

[0100] S42. Compare the differences between the mutant and wild-type structures and calculate the RMSD;

[0101] S5. Set the enzyme activity binary classification target and build the MutShift prediction model:

[0102] S51. Collect the enzyme activity data of glycosyltransferase, including the enzyme activity values of mutants and wild type;

[0103] S52. Setting a binary classification target according to the change direction of the enzyme activity value;

[0104] S53, using learning algorithms and building MutShift prediction models;

[0105] S6. Solve the data imbalance problem and optimize model performance:

[0106] S61, analyzing enzyme activity data to determine the number of positive and negative samples;

[0107] S62. Use oversampling technology to increase the number of positive samples and balance the two types of samples;

[0108] S63. Perform model training on the balanced data and evaluate model performance;

[0109] S7. Introduce QSAR dictionary and protein RMSD global features and train MutShift model:

[0110] S71, QSAR dictionary features are used as input features and introduced into the MutShift prediction model;

[0111] S72. Use the training dataset to train the model and use the validation dataset to evaluate the model performance;

[0112] S73. Adjust model parameters according to the evaluation results to optimize model performance;

[0113] S74. Use the optimized model to predict enzyme activity and verify the accuracy of the prediction results.

[0114] This technical solution achieves accurate prediction of glycosidase mutations by constructing a QSAR dictionary, calculating amino acid contribution scores, evaluating local and global structural effects, and building and optimizing prediction models. Specifically, amino acid features were selected and a QSAR dictionary was constructed to collect and organize the physicochemical property data of amino acids; these feature values and their weights were determined, and the contribution scores of amino acids in enzymes were calculated based on them; the impact of mutations on the local structure of enzymes was reflected by calculating the offset measure of the amino acid contribution scores between mutants and wild types; the RMSD was calculated to quantify the impact of mutations on the global structure of proteins, thereby more comprehensively evaluating the effects of mutations; a binary classification target for enzyme activity was set, and a MutShift prediction model was constructed using a learning algorithm to predict changes in enzyme activity of glycosyltransferases based on the characteristics and mutation status of amino acid residues; to address the data imbalance problem, the number of positive samples was increased through oversampling technology, so that the model could better learn and predict the increase in enzyme activity; QSAR dictionary features were introduced as input features into the MutShift prediction model, and the training dataset was used for training, while the validation dataset was used to evaluate the model performance; the model parameters were adjusted according to the evaluation results to optimize the model performance, and finally the optimized model was used to predict enzyme activity, and the accuracy of the prediction results was verified.

[0115] In addition, the method for improving glycosidase mutation prediction based on amino acid and RMSD features according to the present invention also has the following additional technical features:

[0116] According to one embodiment of the present invention, in the step S11, the physicochemical property data of amino acids are collected and sorted. The interaction between the active site of the enzyme and the substrate is affected by polarity, hydrophobicity and acidity and alkalinity, which are used to predict the change in substrate binding ability after mutation.

[0117] Molecular weight and volume affect the position and force balance of amino acids in the three-dimensional structure and are used to predict the effect of mutations on enzyme structural stability;

[0118] Hydrogen bond donors and acceptors directly affect the chemical reactivity of the catalytic center and are used to determine whether the enzyme retains catalytic ability after mutation;

[0119] Aromaticity and hydrophobicity reflect the effect of amino acid mutations on the global conformation and are used to predict whether the enzyme is inactivated due to mutation.

[0120] In this technical solution, polarity describes the uniformity of molecular charge distribution; polar amino acids form hydrogen bonds and interact with water molecules or polar groups, including Ser, Thr, Asn, Gln, Tyr, His, Lys, Arg, Asp, Glu; non-polar amino acids include Ala, Val, Ile, Leu, Phe, Met, Trp, Cys, Pro, Gly; polar amino acids often appear in enzyme active sites, affecting binding and catalytic efficiency; non-polar amino acids maintain enzyme stability; hydrophobicity describes the behavior of molecules in water; hydrophobic amino acids aggregate and stay away from polar solvents, including Ala, Val, Ile, Leu, Phe, Met, Trp, Pro; hydrophilic amino acids are others; hydrophobicity affects the characteristics of substrate binding sites and is important for substrate selectivity; in protein folding, hydrophobic amino acids form a hydrophobic core to maintain enzyme structural stability; acidity and alkalinity describe the properties of amino acid side chains; acidic amino acids include Asp, Glu; basic amino acids include Lys, Arg, His; neutral amino acids are others; acidic or basic amino acids participate in acid-base catalytic reactions at the active site and affect charge interactions; molecular weight reflects the volume and structural complexity of amino acids, and is standardized to Trp (204); large amino acids cause steric hindrance and affect binding; small amino acids allow structural changes; volume affects the spatial conformation of the enzyme and the accessibility of the active site, and is standardized to Trp (163); large amino acids restrict substrate entry; small amino acids enhance enzyme flexibility; aromaticity describes whether the amino acid has an aromatic side chain, including Phe, Tyr, and Trp; non-aromatic amino acids are others; aromatic groups bind to substrates through π-π stacking or van der Waals interactions and participate in substrate recognition; whether the amino acid side chain contains hydrogen bond donors or acceptors, Ser, Thr, Asn, etc. are typical donors and acceptors, and Phe, Ala, etc. do not have this property; hydrogen bond donors and acceptors form a network at the active site, which is an important factor in the specific binding of enzymes to substrates, and the number affects the binding energy and catalytic efficiency.

[0121] According to one embodiment of the present invention, in the calculation of the amino acid contribution score in step S24, the amino acid contribution score The calculation formula is as follows:

[0122]

[0123] in: is the number of amino acids in the enzyme;

[0124] The number of characteristics for each amino acid, including polarity and hydrophobicity;

[0125] For the The j-th feature value of the amino acid is taken from the amino acid residue dictionary;

[0126] For the The weight of a feature indicates the degree of influence of the feature on the change of enzyme activity; the weight is obtained through model training or experimental data;

[0127] is a bias term used to adjust the predicted baseline value.

[0128] This technical solution quantifies the degree to which the characteristics of each amino acid in the enzyme (including polarity, hydrophobicity, etc.) affect the change in enzyme activity by calculating the amino acid contribution score. The score calculation formula involves the number of amino acids, the number of features, the feature value, the feature weight, and the bias term.

[0129] According to one embodiment of the present invention, in the step S32 of calculating the deviation metric of each amino acid residue, the deviation metric between the mutant and the wild type is The calculation formula is as follows:

[0130]

[0131] in: The mutant and wild type The difference in amino acid contribution scores on each amino acid residue directly reflects the local impact of the mutation on the enzyme molecule;

[0132] For the mutant The amino acid contribution score of each amino acid residue;

[0133] The wild type The amino acid contribution score of each amino acid residue.

[0134] This technical solution calculates the difference in amino acid contribution scores between the mutant and the wild type at each amino acid residue to intuitively reflect the local impact of the mutation on the enzyme molecule, thereby deriving a deviation metric.

[0135] According to one embodiment of the present invention, in the comparison of the differences between the mutant and the wild type in step S42, the root mean square deviation between the mutant and the wild type is The calculation formula is as follows:

[0136]

[0137] in: is the α-carbon number, calculated by comparing the wild type with the mutant , which can quantify the effect of mutations on the global structure of proteins;

[0138] and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate;

[0139] and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate;

[0140] and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate.

[0141] This technical solution quantifies the effect of mutations on the global structure of proteins by calculating the root mean square deviation of α-carbon atoms in the mutant and wild-type protein structures.

[0142] According to one embodiment of the present invention, in the change direction of the enzyme activity value in step S52, the enzyme activity value is derived from the data of the public PDB website and laboratory measurement, and the binary classification target is:

[0143] Target value 1, representing an increase in the mutant enzyme activity, is used to characterize activity enhancement;

[0144] The target value of 0 indicates decreased enzyme activity in the mutant and is used to characterize decreased activity.

[0145] This technical solution determines the enzyme activity value through the public PDB website and laboratory measurement data, and uses a binary classification target (target value 1 indicates increased enzyme activity, and target value 0 indicates decreased enzyme activity) to determine the direction of change in the mutant enzyme activity, that is, whether the activity is enhanced or decreased.

[0146] According to one embodiment of the present invention, in step S62, when the oversampling technique is used to increase the number of positive samples, the enzyme activity data is imbalanced, that is, the number of positive samples is much smaller than the number of negative samples, wherein:

[0147] The number of positive samples is ;

[0148] The number of negative samples is ;

[0149] The goal of oversampling is to increase the number of positive samples so that the number of samples in the two categories is balanced:

[0150] Number of new samples = - = (Number of positive samples, number of newly added samples).

[0151] This technical solution increases the number of positive samples through oversampling technology to balance the difference between the number of positive samples and the number of negative samples in enzyme activity data, making the number of samples in the two categories equal.

[0152] According to one embodiment of the present invention, the QSAR dictionary features of step S71 are used as input features, and an enzyme activity prediction model is introduced through local mutant and wild-type offset metrics and global RMSD to quantify the impact of mutants on enzyme structure and establish a connection with enzyme activity changes for predicting glycoside transferase activity.

[0153] This technical solution introduces QSAR dictionary features as input, combines local mutant and wild-type offset metrics with global RMSD, and establishes an enzyme activity prediction model to quantify the impact of mutants on enzyme structure and establish a connection with enzyme activity changes, thereby predicting glycosyltransferase activity.

[0154] Example 1

[0155] Assume that the enzyme contains 5 amino acids and the number of features for each amino acid is 3 (polarity, hydrophobicity, molecular weight). After inputting the residue features and weights, calculate the mutant :

[0156] Feature matrix T:

[0157]

[0158]

[0159]

[0160]

[0161] S rating:

[0162] This embodiment provides an application for improving glycosidase mutation prediction using amino acid and RMSD features. In the initial stage, the physicochemical property data of a single amino acid are selected, and then the number of features is gradually increased to form a more complex feature set; the feature set is input into the MutShift prediction model for predicting enzyme activity.

[0163] like Figure 2As shown in the figure, the local characteristics of the mutant, such as polarity, hydrophobicity, acidity and alkalinity, and the global characteristics RMSD, jointly act on the change of enzyme activity, forming a complementary mechanism; the upper part reflects that the local characteristics affect the binding of the enzyme to the substrate and the catalytic efficiency; the lower part quantifies the RMSD, reflecting the potential impact of global structural changes on enzyme activity; local characteristics describe the direct impact of the mutation site, and global characteristics provide the overall change background. Combining the two enables the prediction model to fully capture the complex effects of mutations on enzyme activity.

[0164] Figure 3 The process of predicting and verifying mutant enzyme activity is demonstrated; 145 sets of PDB data and 45 sets of laboratory data are collected, covering samples with increased, decreased and balanced enzyme activity; local features are extracted to quantify amino acid residue properties and mutation offsets, and global features are used to calculate RMSD to reflect overall conformational changes; features are used to train the Mutshift model, and its effectiveness is verified by calculating accuracy and wet experiments, and 5 mutants are tested using sucrose synthase as an example; the Mutshift model is verified to provide theoretical and experimental support for analyzing the impact of mutations.

[0165] Figure 4 Demonstrates the structured process of QSAR analysis of chemical activity; the core is to characterize electron distribution, including polarity, acidity and basicity, and aromaticity; analyzes the effect of steric hindrance on molecular weight and volume, considers H-bond formation ability and emphasizes acceptor and donor functions; associates 255 mutants with biological offsets, integrates them into an integrated system, and improves the ability to predict the biological activity of chemicals.

[0166] The source and generation process of the feature set are based on the rational design of the protein sequence and structure of glycosidase, and are generated by combining the molecular physical and chemical properties and data-driven analysis. The feature combination generation follows a systematic screening method, combining multiple feature sets according to rules to cover different physical and chemical properties; a single feature is selected in the initial stage, and gradually increased to form a complex set; recursive feature elimination, correlation analysis, etc. are used to sort the importance of features and create new combinations. Model evaluation and screening The feature set is input into the graph convolutional network to predict enzyme activity. The optimal feature combination is selected based on the prediction accuracy and the accuracy is recorded to evaluate its effectiveness, such as Figure 5 shown.

[0167] The best feature combination was feature set 240_6 (accuracy: 0.72). Key findings: Features such as hydrophobicity, molecular weight, and aromaticity significantly contribute to prediction accuracy, while excessive features lead to decreased performance (see Table 1 for details).

[0168] Table 1 Feature set and its corresponding accuracy table

[0169]

[0170] Data source: Multiple glycosidic transferase datasets, including activity change data measured by different experimental methods.

[0171] Practical Application: To validate the model's accuracy in predicting glycosidic transferases, we used different mutants of the SUSY sucrose synthase enzyme. The letters in the table below represent the amino acid sequence at the mutation site. For example, V159T represents a valine-to-threonine mutation at position 159. The results show that MutShift accurately predicts the activity of all five SUSY sucrose synthase mutants, demonstrating the model's effectiveness. See Table 2 for details.

[0172] Table 2 Mutation site amino acid and sequence list

[0173]

[0174] This invention represents significant innovation in the field of enzyme activity prediction. Through node feature dictionary design and differential feature modeling, it not only improves prediction accuracy but also significantly enhances the practical application value of the model, providing an efficient mutant screening tool for glycosidic transferases.

[0175] Although the present invention is described in detail with reference to the accompanying drawings and in combination with preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and such modifications or substitutions shall be within the scope of the present invention. Any person skilled in the art who can easily conceive of changes or substitutions within the technical scope disclosed in the present invention shall be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be based on the scope of protection of the claims.

Claims

1. A method for predicting glycosidase mutations based on amino acid and RMSD features, characterized in that: The steps include: S1. Select amino acid features and construct QSAR dictionary: S11. Collect and organize data on the physicochemical properties of amino acids, including polarity, hydrophobicity, acidity and alkalinity, molecular weight, volume, aromaticity, and donor and acceptor characteristics; S12, assigning characteristic values to each amino acid based on the data and constructing a QSAR dictionary; S2. Determine the eigenvalues and weights and calculate the amino acid contribution score: S21. Determine the number of amino acids in the enzyme and the number of features of each amino acid; S22, obtain characteristic values from the QSAR dictionary; S23. Determine feature weights through experimental data or model training; S24, calculating amino acid contribution scores; S3. Calculate the deviation metric between mutant and wild type and reflect the local impact: S31. Calculate the amino acid contribution scores of the mutant and wild-type enzymes respectively; S32, calculating the deviation metric for each amino acid residue; S4. Introducing protein RMSD as a global feature: S41. Determine the α-carbon atom coordinates of the wild-type and mutant enzymes; S42. Compare the differences between the mutant and wild-type structures and calculate the RMSD; S5. Set the enzyme activity binary classification target and build the MutShift prediction model: S51. Collect the enzyme activity data of glycosyltransferase, including the enzyme activity values of mutants and wild type; S52. Setting a binary classification target according to the change direction of the enzyme activity value; S53, using learning algorithms and building MutShift prediction models; S6. Solve the data imbalance problem and optimize model performance: S61, analyzing enzyme activity data to determine the number of positive and negative samples; S62. Use oversampling technology to increase the number of positive samples and balance the two types of samples; S63. Perform model training on the balanced data and evaluate model performance; S7. Introduce QSAR dictionary and protein RMSD global features and train MutShift model: S71, QSAR dictionary features are used as input features and introduced into the MutShift prediction model; S72. Use the training dataset to train the model and use the validation dataset to evaluate the model performance; S73. Adjust model parameters according to the evaluation results to optimize model performance; S74. Use the optimized model to predict enzyme activity and verify the accuracy of the prediction results.

2. The method for predicting glycosidase mutations based on amino acid and RMSD features according to claim 1, wherein: In the step S11, the physicochemical property data of amino acids are collected and sorted. The interaction between the active site of the enzyme and the substrate is affected by polarity, hydrophobicity and acidity and alkalinity, which are used to predict the change in substrate binding ability after mutation. Molecular weight and volume affect the position and force balance of amino acids in the three-dimensional structure and are used to predict the effect of mutations on enzyme structural stability; Hydrogen bond donors and acceptors directly affect the chemical reactivity of the catalytic center and are used to determine whether the enzyme retains catalytic ability after mutation; Aromaticity and hydrophobicity reflect the effect of amino acid mutations on the global conformation and are used to predict whether the enzyme is inactivated due to mutation.

3. The method for predicting glycosidase mutations based on amino acid and RMSD features according to claim 1, wherein: In the calculation of the amino acid contribution score in step S24, the amino acid contribution score The calculation formula is as follows: in: is the number of amino acids in the enzyme; The number of characteristics for each amino acid, including polarity and hydrophobicity; For the The j-th feature value of the amino acid is taken from the amino acid residue dictionary; For the The weight of a feature indicates the degree of influence of the feature on the change of enzyme activity; the weight is obtained through model training or experimental data; is a bias term used to adjust the predicted baseline value.

4. The method for predicting glycosidase mutations based on amino acid and RMSD features according to claim 1, wherein: In the step S32 of calculating the deviation metric for each amino acid residue, the deviation metric between the mutant and the wild type is The calculation formula is as follows: in: The mutant and wild type The difference in amino acid contribution scores on each amino acid residue directly reflects the local impact of the mutation on the enzyme molecule; For mutants The amino acid contribution score of each amino acid residue; The wild type The amino acid contribution score of each amino acid residue.

5. The method for predicting glycosidase mutations based on amino acid and RMSD features according to claim 1, wherein: In the step S42, the root mean square deviation between the mutant and the wild type is The calculation formula is as follows: in: is the α-carbon number, calculated by comparing the wild type with the mutant , which can quantify the effect of mutations on the global structure of proteins; and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate; and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate; and , respectively, in the wild-type and mutant protein structures. α-carbon atoms coordinate.

6. The method for predicting glycosidase mutations based on amino acid and RMSD features according to claim 1, wherein: In the change direction of the enzyme activity value in step S52, the enzyme activity value is derived from the public PDB website and laboratory measurement data, and the binary classification target is: Target value 1, representing an increase in the mutant enzyme activity, is used to characterize enhanced activity; The target value of 0 indicates decreased enzyme activity in the mutant and is used to characterize decreased activity.

7. The method for predicting glycosidase mutations based on amino acid and RMSD features according to claim 1, wherein: In step S62, when the oversampling technique is used to increase the number of positive samples, the enzyme activity data is imbalanced, that is, the number of positive samples is much smaller than the number of negative samples, wherein: The number of positive samples is ; The number of negative samples is ; The goal of oversampling is to increase the number of positive samples so that the number of samples in the two categories is balanced: Number of new samples = - = (Number of positive samples, number of newly added samples).

8. The method for predicting glycosidase mutations based on amino acid and RMSD features according to claim 1, wherein: The QSAR dictionary features of step S71 are used as input features, and an enzyme activity prediction model is introduced through local mutant and wild-type offset metrics and global RMSD to quantify the impact of mutants on enzyme structure and establish a connection with enzyme activity changes for predicting glycosyltransferase activity.

Images