Method for predicting activity of enzyme mutants

By selecting characteristics such as amino acid hydrophobicity, volume change and substrate volume, combined with the random forest algorithm, the problem of insufficient accuracy in predicting enzyme mutant activity is solved, and efficient prediction of enzyme mutant activity is achieved, reducing experimental costs.

CN119028441APending Publication Date: 2024-11-26青岛奔月生物技术有限公司 +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411471318.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-22
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prior art is inaccurate and computationally expensive in predicting changes in enzyme mutant activity, especially for complex enzyme mutants, where machine learning methods have challenges in feature selection and model training.

Method used

Specific feature selections such as amino acid hydrophobicity changes, volume changes and substrate volume characteristics, combined with random forest algorithms, training sets and test set divisions, model training and optimization, calculate substrate volume through RDKit tool, and use 50% cross-validation. and grid search optimization model hyperparameters.

Benefits of technology

It significantly improves the accuracy of the activity prediction of enzyme mutants, reduces the workload and cost of experimental screening, and can effectively screen enzyme mutants with potential application value.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028441A_ABST
    Figure CN119028441A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of enzyme mutant activity prediction, in particular to a method for predicting enzyme mutant activity. The method comprises the following steps: inputting glycosidase transferase data; key features influencing the activity of the glycosidase transferase are extracted; predefining an amino acid characteristic dictionary; dividing a training set and a test set; training and optimizing the model; and evaluating and verifying the model. According to the method, the accuracy of predicting the activity of the mutant can be remarkably improved by using specific feature selection, such as hydrophobic change and volume change of amino acid and volume features of a substrate, and combining a random forest algorithm aiming at the glycosyl transferase enzyme mutant.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the activity of enzyme mutants, and in particular to a method for predicting the activity of enzyme mutants. Background Art

[0002] In enzyme engineering and protein design, predicting the activity changes of enzyme mutants is a key challenge. Existing methods usually rely on molecular dynamics simulations or quantitative structure-activity relationship models, but these methods have problems of insufficient accuracy and high computational cost when dealing with complex enzyme mutants. Machine learning methods have been introduced into this field in recent years. However, how to select appropriate features and effectively apply them to model training is still an open problem. For example, a Chinese patent with publication number CN116884472A discloses a method for predicting the antimicrobial activity of antimicrobial peptides. It takes into account the intrinsic connection of antimicrobial peptide genes and solves the problem of overfitting of the model and the problem of missing some labels in the prediction of antimicrobial peptide activity by improving the prediction model. However, the above patent seems to be unable to predict the activity changes of mutants. 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 the activity of enzyme mutants, which can significantly improve the prediction accuracy of the change in the activity of the mutants.

[0004] The technical solution of the present invention is: A method for predicting the activity of an enzyme mutant, the method comprising: Glycosidase transferase data entry; Extract key features that influence glycosidase transferase activity; Dictionary of predefined amino acid properties; Training set and test set division; Model training and optimization; Model evaluation and validation.

[0005] As a further embodiment of the present invention, the glycosidase transferase data input includes the following features: changes in the amino acid site of the active pocket and three-dimensional volume features of the substrate.

[0006] As a further embodiment of the present invention, the amino acid feature extraction includes the hydrophobicity and volume changes of the amino acids in the active pocket and the changes in the substrate volume.

[0007] As a further embodiment of the present invention, the amino acid property dictionary stores various physical and chemical properties of amino acids, including hydrophobicity, volume, and polarity.

[0008] As a further embodiment of the present invention, the substrate volume is calculated using the RDKit tool, and the calculation formula for the volume V(M) of the substrate molecule M is as follows:

[0009] Where, represents the contribution of the i-th atom in the molecule, and N is the total number of atoms in the molecule.

[0010] As a further solution of the present invention, the training set and test set division randomly divides the data set into a training set and a test set through a train_test_split function.

[0011] As a further solution of the present invention, the model selection is a random forest model.

[0012] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a method for predicting the activity of enzyme mutants, which is particularly targeted at glycosyltransferase enzyme mutants. By using specific feature selection, such as changes in the hydrophobicity of amino acids, changes in volume, and features such as the volume of substrates, combined with a random forest algorithm, the accuracy of mutant activity prediction can be significantly improved. This method can be used to screen enzyme mutants with potential application value to reduce the workload and cost of experimental screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0014] Figure 1 It is a schematic diagram of the process of the present invention.

[0015] Figure 2 2 is an importance analysis diagram of the features of the present invention. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0017] Example like Figure 1As shown, this embodiment provides a method for predicting the activity of an enzyme mutant, the method comprising: Glycosidase transferase data entry; Extract key features that influence glycosidase transferase activity; Dictionary of predefined amino acid properties; Training set and test set division; Model training and optimization; Model evaluation and validation.

[0018] As a further embodiment of the present invention, the glycosidase transferase data input includes the following features: Changes in amino acid sites in the active pocket: Compared with the wild type, the replacement of key amino acids in the active pocket of the mutant has a direct effect on the catalytic activity of the glycosidase transferase; Three-dimensional volume characteristics of the substrate: Calculated from the SMILES string of the substrate, the volume of the substrate affects its binding mode with the active pocket of the glycosidase transferase, thereby affecting the activity of the enzyme.

[0019] Part of the data is from the public documents on the Protein Data Bank website. The present invention selects data related to glycosidic transferase, which includes: Structural data: 3D structural information of glycosyltransferase extracted from PDB was used to further analyze the effects of mutations on the conformation of the enzyme active pocket; Activity data: Experimentally measured enzyme activities of glycosyltransferases and their mutants are provided in the public literature and used as label data (i.e., target variables) for model training to help the model learn the relationship between amino acid sequence changes and enzyme activity.

[0020] Another part of the data comes from internal laboratory research results. These data mainly include mutant experimental analysis of specific glycosidic transferases through technical means such as site-directed mutagenesis and saturation mutagenesis, and the enzyme activity characterization indicators before and after mutation were obtained.

[0021] As a further embodiment of the present invention, the amino acid feature extraction includes the hydrophobicity and volume changes of the amino acids in the active pocket and the changes in the substrate volume, and these features are closely related to the catalytic activity of the enzyme.

[0022] As a further embodiment of the present invention, the amino acid property dictionary is a predefined data structure for storing multiple physical and chemical properties of each amino acid, including hydrophobicity, volume, and polarity, which can have a significant impact on the structure and function of enzymes (proteins).

[0023] The specific values ​​of amino acid properties can be obtained from literature, databases (such as AAindex database), or existing bioinformatics tools. AAindex is a widely used amino acid index database that provides physical and chemical property data of various amino acids, which are organized into tables and stored in a dictionary structure.

[0024] The amino acid property dictionary is designed as a Python dictionary with key-value pairs in the form of {'A': {'hydrophobicity': 1.8, 'volume': 88.6}, 'R': {...}, ...}, where the key is the single-letter code of the amino acid and the value is a sub-dictionary containing multiple properties. When processing mutant data, the model queries the corresponding property values ​​of the amino acids before and after the mutation through the dictionary. For example, to calculate the change in hydrophobicity before and after the amino acid mutation: Indicates the change in hydrophobicity before and after mutation, represents the hydrophobicity after mutation, represents the hydrophobicity of the wild type, then = -

[0025] By comparing the changes in the characteristics of amino acids before and after mutation, the potential impact of the mutation on the active pocket environment and enzyme activity can be inferred. These characteristic changes are then used as input features of the model for training and prediction.

[0026] Stores various physicochemical properties of amino acids, including hydrophobicity, volume, and polarity.

[0027] As a further embodiment of the present invention, the substrate volume is calculated using the RDKit tool, and the calculation formula for the volume V(M) of the substrate molecule M is as follows:

[0028] in, represents the contribution of the i-th atom in the molecule, and N is the total number of atoms in the molecule.

[0029] Glycosidase activity is directly affected by mutations in the active pocket amino acid sites, which may change the physicochemical properties of the active pocket, such as hydrophobicity and spatial volume. In addition, the fit of the substrate volume in the active pocket is also a key factor affecting enzyme activity. Therefore, the input amino acid site changes and substrate volume characteristics, after extraction and calculation, as the core input of the model, can effectively predict the activity of glycosidase transferase.

[0030] As a further solution of the present invention, the training set and test set division randomly divides the data set into a training set and a test set through a train_test_split function, wherein the training set is used for model training and the test set is used for model evaluation. Before the data set is divided, the balance of the data is ensured to avoid bias.

[0031] The model was trained and optimized through 5-fold cross validation and used for the final prediction.

[0032] 5-fold cross validation is a technique for evaluating the performance of machine learning models. This paper divides the data set into 5 subsets and performs model training and validation on each fold to reduce the deviation of model evaluation. The specific steps are as follows: 1a. Data partitioning: Randomly divide the entire dataset into 5 subsets of equal size.

[0033] 2b. Training and validation: In the first iteration, the first subset is selected as the validation set, and the remaining four subsets are used as training sets. The model is trained using the training set, and the model performance is evaluated on the validation set.

[0034] In the second iteration, the second subset is selected as the validation set, and the remaining four subsets are used as training sets.

[0035] This process is repeated until each subset has been used as a validation set once.

[0036] 3c. Performance evaluation: Calculate the average performance of 5 validations, and use this average as the final evaluation result of the model.

[0037] This approach allows for a more stable evaluation of the model’s performance because it uses different parts of the data to train and test the model, thus avoiding evaluation bias caused by uneven data distribution.

[0038] As a further solution of the present invention, the model selects a random forest model. Random forest is an integrated learning method that reduces the overfitting risk of a single decision tree by constructing multiple decision trees.

[0039] When a decision tree splits a node, it will The value of selects the best split point. Let the objective function (enzyme activity) be y, then when splitting, find a split point s to minimize the purity of the node:

[0040] in, and are the data volume of the left and right child nodes respectively, and The purity of the left and right child nodes respectively (such as Gini coefficient or entropy) The final output of the random forest model is the weighted average or majority vote of the prediction results of each tree. For classification tasks, it can be expressed as:

[0041] in, It represents the prediction result of the t-th tree for the input sample x, and T is the total number of decision trees.

[0042] Use Grid Search and Cross-Validation to optimize the model's hyperparameters, including n_estimators (the number of trees), max_depth (the maximum depth of the tree), etc.

[0043] During the parameter adjustment process, 5-fold cross validation was used to evaluate the effects of different parameter combinations and select the optimal parameters.

[0044] Specific steps: Construct a grid search object: grid_search = GridSearchCV(estimator=model, param_grid=param_grid,cv=5,n_jobs=-1,scoring='accuracy').

[0045] Perform a grid search: grid_search.fit(X_train, y_train).

[0046] The model is evaluated using the training set and the test set, and the model's accuracy, confusion matrix, classification report and other indicators are calculated. The accuracy reflects the overall correctness of the model's prediction, while the confusion matrix and classification report provide a more detailed analysis of the classification effect.

[0047] Let A be the accuracy, and assume that the model's predicted enzyme activity on the test set is The true enzyme activity is , then the accuracy A is:

[0048] in, is the indicator function, indicating the number of samples predicted correctly; n is the total number of samples.

[0049] The importance of each feature to the prediction is analyzed through the model's feature importance attribute. Use the feature_importances_ attribute to extract the feature importance and use visualization tools (such as matplotlib) to draw a feature importance graph to show the contribution of the feature to the model prediction. Feature importance can be calculated by the average contribution of each feature in all trees. Assume is the jth feature, To express its importance, it can be expressed by the following formula:

[0050] in, is the contribution of the jth feature in the tth tree.

[0051] Practical Application: In order to verify the accuracy of the model in predicting glycosidic transferase, a UDP-glucose transferase (protein sequence: MATNLRVLMFPWLAYGHISPFLNIAKQLADRGFLIYLCSTRINLESIIKKIPEKYADSIHLIELQLPELPELPPHYHTTNGLPPHLNPTLHKALKMSKPNFSRILQNLKPDLLIYDVLQPWAEHVANEQNIPAGKLLTSCAAVFSYFFSFRKNPGVEFPFPAIHLPEVEKVKIREILAKEPEEGGRLDEGNKQMMLMCTSRTIEAKYIDYCTELCNWKVVPVGPPFQDL Different mutants of ITNDADNKELIDWLGTKHENSTVFVSFGSEYFLSKEDMEEVAFALELSNVNFIWVARFPKGEERNLEDALPKGFLERIGERGRVLDKFAPQPRILNHPSTGGFISHCGWNSAMESIDFGVPIIAMPIHFDQPINAKLMVELGVAVEIVRDDDGKIHRGEIAETLKSVVTGETGEILRAKVREISKNLKSIRDEEMDAVAEELIQLCRNSNKSK were used for verification. The letters in the table below represent the amino acid and sequence of the mutation site, for example, N338A represents the mutation of asparagine at No. 338 to alanine.

[0052] The results are shown in Table 1. GlycoML accurately predicted the activities of the five mutants of UDP-glucose transferase, which proved the effectiveness of the model.

[0053] Table 1 Mutation site amino acids and sequences

[0054] In the present invention, the contribution of each feature to the prediction of enzyme activity was verified by inputting different features into the model (GlycoML) for comparison. Figure 2 As shown in the table, the experimental results show that the accuracy of the single feature model is between 0.35 and 0.52, as shown in the table: Table 2 Comparison of the accuracy of different feature combination models

[0055] In contrast, the accuracy of the model containing all four features is 0.60, which is significantly better than the single feature model. This shows that the combination of multiple features significantly improves the accuracy of the model, especially in the prediction of enzyme activity, and there is a certain synergy between the features.

[0056] Although the present invention has been described in detail with reference to the accompanying drawings and in combination with the 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 these modifications or substitutions shall be within the scope of the present invention. Any person of ordinary skill in the art may easily think of changes or substitutions within the technical scope disclosed by the present invention, and these shall be within the scope of protection of the present invention.

Claims

1. A method for predicting the activity of an enzyme mutant, characterized in that: The method comprises: Glycosidase transferase data entry; Extract key features that influence glycosidase transferase activity; Dictionary of predefined amino acid properties; Training set and test set division; Model training and optimization; Model evaluation and validation.

2. The method for predicting the activity of an enzyme mutant according to claim 1, wherein The glycosidase transferase data input includes the following features: Changes in the amino acid site of the active pocket and the three-dimensional volume characteristics of the substrate.

3. The method for predicting the activity of an enzyme mutant according to claim 1, wherein The amino acid feature extraction includes the hydrophobicity and volume changes of the amino acids in the active pocket and the changes in the substrate volume.

4. The method for predicting the activity of an enzyme mutant according to claim 1, wherein The amino acid property dictionary stores a variety of physicochemical properties of amino acids, including hydrophobicity, volume, and polarity.

5. The method for predicting the activity of an enzyme mutant according to claim 3, wherein: The substrate volume is calculated using the RDKit tool, and the calculation formula for the volume V(M) of the substrate molecule M is as follows: Among them, V i represents the contribution of the i-th atom in the molecule, and N is the total number of atoms in the molecule.

6. The method for predicting the activity of an enzyme mutant according to claim 1, wherein The training set and test set division step randomly divides the data set into a training set and a test set through the train_test_split function.

7. The method for predicting the activity of an enzyme mutant according to claim 1, wherein The model selected is a random forest model.

8. The method for predicting the activity of an enzyme mutant according to claim 1, wherein The model training and optimization is performed through 5-fold cross validation, which specifically includes the following sub-steps: a. Data partitioning: randomly divide the entire data set into 5 subsets of equal size; b. Training and validation: In the first iteration, the first subset is selected as the validation set, and the remaining four subsets are used as training sets. The model is trained using the training set, and the model performance is evaluated on the validation set. In the second iteration, the second subset is selected as the validation set, and the remaining four subsets are used as training sets; This process is repeated until each subset has been used as a validation set once.

9. The method for predicting the activity of an enzyme mutant according to claim 1, wherein The model evaluation and verification includes the following steps: calculating the average performance of several verifications, and using the average as the final evaluation result of the model.

Citation Information

Cited By

  • Amino acid and RMSD feature improved glycosidase mutation prediction method and application thereof

    CN119943141A

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

    CN119943141B