Method for predicting enzyme catalytic reaction activity, enzyme modification and substrate screening
Patent Information
- Application Number
- CN202611197993.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-07
- Publication Date
- 2026-11-03
AI Technical Summary
In existing technologies, enzyme modification and substrate screening rely on directed evolution or rational design, which is time-consuming, labor-intensive, and costly. Furthermore, existing computational methods are unable to accurately predict the three-dimensional spatial interaction between enzymes and substrates.
A point cloud neural network model based on the Transformer architecture is used to directly process the three-dimensional atomic coordinate information of enzymes and substrates. The three-dimensional structure of enzymes and the three-dimensional conformation of substrates are predicted by AlphaFold3 and RDKit. The reactivity is predicted by the point cloud neural network model, and the model is optimized by training set to improve the prediction accuracy.
It achieves high-precision prediction of enzyme catalytic reaction activity, significantly improves the efficiency of enzyme modification and substrate screening, reduces the number of experimental samples, shortens the research and development cycle, and reduces costs.
Smart Images
Figure CN123033579A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of biocatalyst technology, and in particular to a method for predicting enzyme catalytic reaction activity, modifying enzymes, and screening substrates. Background Technology
[0002] Enzymes, as highly efficient biocatalysts, have wide applications in medicine, chemical engineering, energy, and other fields. One of the core goals of enzyme engineering is to improve the catalytic performance of enzymes or broaden their applications by modifying enzymes or screening new substrates. However, traditional enzyme modification and substrate screening mainly rely on directed evolution or rational design combined with extensive wet experimental verification. This approach requires constructing a large mutant library or screening a large number of compound libraries, which is time-consuming, labor-intensive, costly, and inefficient.
[0003] With the development of structural biology and artificial intelligence, it has become possible to predict enzyme properties using computational methods. However, most existing computational methods are based on the enzyme's amino acid sequence or simplified molecular descriptors, which makes it difficult to fully capture the fine interactions between the enzyme and its substrate in three-dimensional space, resulting in inaccurate predictions of enzyme properties. Summary of the Invention
[0004] This invention provides a method for predicting enzyme catalytic activity, modifying enzymes, and screening substrates, which addresses the shortcomings of inaccurate prediction of enzyme properties in existing technologies. It enables precise prediction of the catalytic activity (Kcat) of enzyme-substrate complexes directly from their three-dimensional structures, thereby guiding efficient enzyme modification and substrate screening.
[0005] This invention provides a method for predicting enzyme catalytic reaction activity, comprising:
[0006] The point cloud of the enzyme to be predicted is obtained by encoding its three-dimensional structure, and the point cloud of the substrate catalyzed by the enzyme to be predicted is obtained by encoding its three-dimensional structure.
[0007] The point cloud of the enzyme to be predicted and the point cloud of the substrate are used as inputs to a point cloud neural network model to obtain the predicted reactivity value of the enzyme to be predicted catalyzing the substrate, which is output by the point cloud neural network model.
[0008] The point cloud neural network model is obtained by training a training set, which includes the point cloud of the sample enzyme, the point cloud of the substrate catalyzed by the sample enzyme, and the true value of the reaction activity of the substrate catalyzed by the sample enzyme.
[0009] According to the enzyme catalytic reaction activity prediction method provided by the present invention, before encoding the three-dimensional structure of the enzyme to be predicted to obtain a point cloud of the enzyme to be predicted, and before encoding the three-dimensional structure of the substrate catalyzed by the enzyme to be predicted to obtain a point cloud of the substrate, the method further includes:
[0010] The three-dimensional structure of the enzyme to be predicted was predicted using AlphaFold3 based on the amino acid sequence of the enzyme to be predicted.
[0011] The RDKit is used to construct a three-dimensional conformation of the substrate based on the chemical structure of the enzyme to be predicted, which is then used as the three-dimensional structure of the substrate.
[0012] According to the enzyme catalytic reaction activity prediction method provided by the present invention, the point cloud of the enzyme to be predicted is obtained by encoding the three-dimensional structure of the enzyme to be predicted, including:
[0013] The atoms within a predetermined range surrounding the catalytic active center in the three-dimensional structure of the enzyme to be predicted are encoded to obtain the point cloud of the enzyme to be predicted.
[0014] According to the enzyme catalytic reaction activity prediction method provided by the present invention, the method encodes the three-dimensional structure of the enzyme to be predicted to obtain a point cloud of the enzyme, and encodes the three-dimensional structure of the substrate catalyzed by the enzyme to be predicted to obtain a point cloud of the substrate, including:
[0015] Each atom in the three-dimensional structure of the enzyme to be predicted and its corresponding substrate is regarded as a point in three-dimensional space;
[0016] Based on the three-dimensional coordinates, atomic radius, and charge of each atom, a feature vector is constructed for the points corresponding to each atom.
[0017] Based on the points and feature vectors corresponding to each atom of the enzyme to be predicted, the three-dimensional structure of the enzyme to be predicted is transformed into a multi-dimensional point cloud.
[0018] Based on the points and feature vectors corresponding to each atom of the substrate of the enzyme to be predicted, the three-dimensional structure of the substrate of the enzyme to be predicted is transformed into a multi-dimensional point cloud.
[0019] According to the enzyme catalytic reaction activity prediction method provided by the present invention, the point cloud neural network model is based on the Transformer architecture, and the reaction activity prediction value is the Kcat prediction value.
[0020] According to the enzyme catalytic reaction activity prediction method provided by the present invention, after the point cloud neural network model is trained with a training set, it further includes:
[0021] The test set is input into the point cloud neural network model for testing to obtain the predicted value of the reaction activity corresponding to the test set. The test set includes the point cloud of the sample enzyme, the point cloud of the substrate catalyzed by the sample enzyme, and the true value of the reaction activity of the substrate catalyzed by the sample enzyme.
[0022] The Spearman correlation coefficient between the predicted and actual reactivity values corresponding to the test set is calculated and used as the prediction performance index of the point cloud neural network model.
[0023] This invention also provides an enzyme modification method, comprising:
[0024] The target enzyme sequence was modified in different ways to obtain the modified target enzyme;
[0025] Based on any of the enzyme catalytic reaction activity prediction methods described above, obtain the predicted reaction activity of the target enzyme before and after modification of the same substrate.
[0026] The final modification result of the target enzyme is determined based on the predicted reactivity of the target enzyme before and after modification of the same substrate.
[0027] According to the enzyme modification method provided by the present invention, the final modification result of the target enzyme is determined based on the predicted reactivity of the target enzyme before and after modification of the same substrate, including:
[0028] If the predicted reactivity value of the target enzyme after modification is greater than the predicted reactivity value of the target enzyme before modification, and the difference between the predicted reactivity values of the target enzyme before and after modification is greater than a preset threshold, then the target enzyme after modification is determined to be the final modification result of the target enzyme.
[0029] The present invention also provides a substrate screening method, comprising:
[0030] Based on any of the enzyme catalytic reaction activity prediction methods described above, the predicted reaction activity values of each substrate in the target enzyme catalytic substrate molecule library are obtained.
[0031] Based on the predicted reactivity values of the target enzymes catalyzing each substrate, the substrates were selected as the screening results.
[0032] According to the substrate screening method provided by the present invention, the substrate library includes known substrates, and substrates are selected as screening results based on the predicted reactivity values of each substrate catalyzed by the target enzyme, including:
[0033] If the predicted reactivity of the target enzyme to catalyze each unknown substrate is greater than the predicted reactivity of the target enzyme to catalyze a known substrate, then the unknown substrate is used as the screening result.
[0034] The method for predicting enzyme catalytic reaction activity, modifying enzymes, and screening substrates provided by this invention has the following beneficial effects:
[0035] High prediction accuracy: This invention employs a point cloud neural network, which can directly process the three-dimensional atomic coordinate information of enzymes and substrates, effectively learning their spatial structural features and interaction patterns. Compared with methods based on sequences or two-dimensional structures, it can more accurately predict the Kcat value of enzyme-catalyzed reactions. This technical solution ultimately achieved a Spearman correlation coefficient of approximately 0.7 with a significance p-value < 0.5, indicating a very high correlation between the predicted and actual values.
[0036] Revolutionary improvement in screening efficiency: This invention enables virtual screening of an exponential number of enzyme modification schemes and a vast array of potential substrates, rapidly predicting their catalytic activity on a computer, thus focusing on the few most promising candidates before experiments. This significantly reduces the number of samples required for actual construction and testing, substantially shortens the R&D cycle, and lowers costs.
[0037] Wide range of applications: This invention provides a powerful computational tool for enzyme engineering and synthetic biology. Whether it's improving catalytic efficiency through enzyme modification or discovering new catalytic reactions through substrate screening, this method provides reliable theoretical guidance and has extremely high scientific research and industrial value.
[0038] The physical significance is clear: this method makes predictions directly based on the three-dimensional structure at the atomic level, and the features learned by the model are closely related to the real physicochemical interactions, providing valuable clues for understanding the enzyme catalysis mechanism and making subsequent rational designs. Attached Figure Description
[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0040] Figure 1 This is a schematic flowchart of the enzyme catalytic reaction activity prediction method provided by the present invention;
[0041] Figure 2 This is a schematic diagram of the point cloud neural network model based on the Transformer architecture in the enzyme catalytic reaction activity prediction method provided by the present invention;
[0042] Figure 3 This is a schematic flowchart of the enzyme modification method provided by the present invention;
[0043] Figure 4 This is a schematic flowchart of the substrate screening method provided by the present invention. Detailed Implementation
[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0045] The following is combined with Figure 1 A method for predicting enzyme catalytic reaction activity according to the present invention includes:
[0046] Step 101: Encode the three-dimensional structure of the enzyme to be predicted to obtain the point cloud of the enzyme to be predicted, and encode the three-dimensional structure of the substrate catalyzed by the enzyme to be predicted to obtain the point cloud of the substrate.
[0047] Step 102: Use the point cloud of the enzyme to be predicted and the point cloud of the substrate as input to the point cloud neural network model to obtain the predicted reactivity value of the enzyme to be predicted catalyzing the substrate output by the point cloud neural network model.
[0048] The point cloud neural network model is obtained by training a training set, which includes the point cloud of the sample enzyme, the point cloud of the substrate catalyzed by the sample enzyme, and the true value of the reaction activity of the substrate catalyzed by the sample enzyme.
[0049] By encoding the three-dimensional structures of the enzyme and substrate to be predicted, the three-dimensional structures of the enzyme and substrate to be predicted are transformed into numerical representations that can be processed by point cloud neural networks.
[0050] Point cloud neural networks are a type of deep learning model that can directly process unordered point sets in three-dimensional space. In this embodiment, they are used to learn features from the three-dimensional atomic coordinates of enzymes and substrates.
[0051] A point cloud neural network can simultaneously receive the point clouds of the enzyme to be predicted and the substrate (two point clouds) as input. After internal multi-layer feature extraction and calculation, the model finally outputs a predicted reactivity value.
[0052] Before using the point cloud neural network model for reactivity prediction, the point cloud neural network model is trained.
[0053] A dataset can be constructed by collecting enzymes, their catalytic substrates, and actual enzyme reactivity values from publicly available literature. The units for all actual reactivity values are then standardized.
[0054] The three-dimensional structures of enzymes and substrates in the dataset are encoded to obtain point clouds. The point clouds of enzymes, point clouds of substrates, and true values of enzyme reactivity in the dataset are divided to obtain training sets and test sets.
[0055] During the training of a point cloud neural network model using a training set, the predicted reactivity values and the true reactivity labels output by the model are input into a loss function (such as the mean squared error loss function) to calculate the prediction error. The gradient is calculated using the backpropagation algorithm, and an optimizer (such as Adam) is used to update the weight parameters of the neural network. This process is repeated for multiple rounds of training until the model's performance on the test set no longer improves or reaches the preset number of training rounds.
[0056] This embodiment employs a point cloud neural network, which can directly process the three-dimensional atomic coordinate information of enzymes and substrates, effectively learning their spatial structural features and interaction patterns. Compared to methods based on sequences or two-dimensional structures, it can more accurately predict the activity of enzyme-catalyzed reactions.
[0057] Based on the above embodiments, this embodiment further includes, before encoding the three-dimensional structure of the enzyme to be predicted to obtain the point cloud of the enzyme to be predicted, and before encoding the three-dimensional structure of the substrate catalyzed by the enzyme to be predicted to obtain the point cloud of the substrate:
[0058] The three-dimensional structure of the enzyme to be predicted was predicted using AlphaFold3 based on the amino acid sequence of the enzyme to be predicted.
[0059] The RDKit is used to construct a three-dimensional conformation of the substrate based on the chemical structure of the enzyme to be predicted, which is then used as the three-dimensional structure of the substrate.
[0060] AlphaFold3 is an artificial intelligence system that can predict the three-dimensional structure of proteins with high accuracy based on their amino acid sequences.
[0061] RDKit is an open-source cheminformatics and machine learning toolkit, which in this example is used to process and construct the three-dimensional structure of substrate molecules.
[0062] Based on the above embodiments, this embodiment encodes the three-dimensional structure of the enzyme to be predicted to obtain a point cloud of the enzyme, including:
[0063] The atoms within a predetermined range surrounding the catalytic active center in the three-dimensional structure of the enzyme to be predicted are encoded to obtain the point cloud of the enzyme to be predicted.
[0064] The three-dimensional structure of the enzyme is processed by using code to retain only all atoms within a preset range around the catalytic active site, which will then be used as the enzyme structure input for subsequent models.
[0065] The preset range can be within 20 Å around the catalytic active center.
[0066] Based on the above embodiments, this embodiment encodes the three-dimensional structure of the enzyme to be predicted to obtain a point cloud of the enzyme, and encodes the three-dimensional structure of the substrate catalyzed by the enzyme to be predicted to obtain a point cloud of the substrate, including:
[0067] Point cloud generation: Each atom in the three-dimensional structure of the enzyme to be predicted and its corresponding substrate is regarded as a point in three-dimensional space;
[0068] Feature encoding: Based on the three-dimensional coordinates (x, y, z), atomic radius (r), and charge magnitude (e) of each atom, a feature vector is constructed for the points corresponding to each atom;
[0069] Based on the points and feature vectors corresponding to each atom of the enzyme to be predicted, the three-dimensional structure of the enzyme to be predicted is transformed into a multi-dimensional point cloud.
[0070] Based on the points and feature vectors corresponding to each atom of the substrate of the enzyme to be predicted, the three-dimensional structure of the substrate of the enzyme to be predicted is transformed into a multi-dimensional point cloud.
[0071] After the above processing, a three-dimensional enzyme structure containing N atoms is transformed into a five-dimensional point cloud composed of N points, and a three-dimensional substrate structure containing M atoms is transformed into a five-dimensional point cloud composed of M points.
[0072] Based on the above embodiments, the point cloud neural network model in this embodiment is based on the Transformer architecture, and the reactivity prediction value is the Kcat prediction value.
[0073] A schematic diagram of the structure of a point cloud neural network model based on the Transformer architecture is shown below. Figure 2 As shown, the predicted reactivity value output by the model is the Kcat predicted value. A Kcat predicted value ≥ 0 indicates a higher predicted reactivity, meaning the enzyme has a stronger ability to catalyze the substrate.
[0074] Kcat is the enzyme turnover number, a core kinetic parameter for measuring enzyme catalytic efficiency. It represents the maximum number of substrate molecules that a single enzyme molecule can convert into product per unit time.
[0075] Based on the above embodiments, this embodiment further includes, after the point cloud neural network model is trained using a training set:
[0076] The test set is input into the point cloud neural network model for testing to obtain the predicted value of the reaction activity corresponding to the test set. The test set includes the point cloud of the sample enzyme, the point cloud of the substrate catalyzed by the sample enzyme, and the true value of the reaction activity of the substrate catalyzed by the sample enzyme.
[0077] The Spearman correlation coefficient between the predicted and actual reactivity values corresponding to the test set is calculated and used as the prediction performance index of the point cloud neural network model.
[0078] After training the point cloud neural network model, a batch of enzyme-substrate pairing data (test set) that did not appear during training is loaded and input into the trained model for prediction. The predicted reactivity values are compared with the actual reactivity values, and the Spearman correlation coefficient is calculated to evaluate the model's predictive performance.
[0079] The Spearman correlation coefficient is a statistical indicator that measures the strength and direction of the monotonic relationship between two variables. In this embodiment, it is used to assess the correlation between the predicted reactivity value and the actual reactivity value.
[0080] In the experiment, the specific steps for model training and performance evaluation included:
[0081] Data preparation: 11,171 sets of enzyme-substrate-Kcat pairing data were collected from the literature.
[0082] Data preprocessing and encoding: Three-dimensional structures of the enzyme and substrate were constructed using AlphaFold3 and RDKit, and atoms within a 20 Å range of the enzyme active site were extracted. Each atom was encoded as a five-dimensional point cloud of (x, y, z, r, e).
[0083] Model Training: Construct a point cloud neural network based on the Transformer architecture. Divide the dataset into training, validation, and test sets in a 6:2:2 ratio. Set the batch size to 32, the learning rate to 0.001, and train for 100 epochs.
[0084] Performance Results: After training, the model was evaluated on the test set. With a significance level of <0.5, the Spearman correlation coefficient between the model's predictions and the actual values reached approximately 0.7, indicating that the model has excellent predictive ability.
[0085] like Figure 3 As shown, this embodiment also provides an enzyme modification method, including:
[0086] Step 301: Modify the sequence of the target enzyme in different ways to obtain the modified target enzyme;
[0087] Step 302: Based on the enzyme catalytic reaction activity prediction method in any of the above embodiments, obtain the predicted reaction activity of the target enzyme before and after modification of the same substrate.
[0088] Step 303: Determine the final modification result of the target enzyme based on the predicted reactivity of the target enzyme before and after modification of the same substrate.
[0089] The specific steps for enzyme modification applications include:
[0090] 1. Determine a target substrate structure.
[0091] 2. Modify the target enzyme sequence in different ways (such as site-directed mutagenesis) to obtain multiple modified enzyme sequences, predict their three-dimensional structure using AlphaFold3, and then perform encoding processing to obtain their respective enzyme point clouds.
[0092] 3. Pair all enzyme point clouds before and after modification with the same substrate point cloud and input them into the trained model to obtain the predicted reaction activity value.
[0093] 4. Compare the predicted values. If the predicted value of a certain modified enzyme is significantly higher than that of the enzyme before modification, it indicates that the modification scheme may effectively improve the catalytic ability of the enzyme and can be used as a priority for experimental verification.
[0094] Alternatively, the enzyme with the highest predicted reactivity value can be selected as the final modification result from among the enzymes whose predicted reactivity value is higher than that before modification.
[0095] This embodiment enables virtual screening of an exponential number of enzyme modification schemes and rapid prediction of their catalytic activity on a computer. This allows for focusing on the few most promising candidates before experiments, which greatly reduces the number of samples that need to be actually constructed and tested, significantly shortens the R&D cycle, and reduces costs.
[0096] Based on the above embodiments, this embodiment determines the final modification result of the target enzyme according to the predicted reactivity of the target enzyme before and after modification of the same substrate, including:
[0097] If the predicted reactivity value of the target enzyme after modification is greater than the predicted reactivity value of the target enzyme before modification, and the difference between the predicted reactivity values of the target enzyme before and after modification is greater than a preset threshold, then the target enzyme after modification is determined to be the final modification result of the target enzyme.
[0098] In the experiment, the virtual screening steps for enzyme modification included:
[0099] Screening objective: To improve the catalytic activity of a certain enzyme on a specific substrate A.
[0100] Screening process: Fifty different site-directed mutagenesis designs were performed on the enzyme to obtain 50 mutant structures. The point clouds of these 50 mutants and the wild-type enzyme were paired with the point cloud of substrate A, and then input into the model trained in Example 1 to predict the reactivity.
[0101] Screening results: Model predictions showed that five mutants had a reactivity value that was more than 20% higher than that of the wild-type enzyme.
[0102] Conclusion: These five mutants were identified as high-priority experimental targets, and it is expected that enzyme variants with significantly enhanced catalytic activity can be obtained.
[0103] like Figure 4 As shown, this embodiment also provides a substrate screening method, including:
[0104] Step 401: Based on the enzyme catalytic reaction activity prediction method described in any of the above embodiments, obtain the predicted reaction activity values of each substrate in the target enzyme catalytic substrate molecule library;
[0105] Step 402: Select substrates as screening results based on the predicted reactivity values of the target enzyme catalyzing each substrate.
[0106] The specific steps involved in substrate screening applications include:
[0107] 1. Determine the structure of a target enzyme.
[0108] 2. Prepare a large library of potential substrate molecules, including known substrates of the enzyme, and construct their three-dimensional structures using RDKit. Then process them according to steps one and two to obtain their respective substrate point clouds.
[0109] 3. Pair the same enzyme point cloud with all substrate point clouds in the library and input them into the trained model to predict the Kcat score.
[0110] 4. Compare the predicted scores. If the predicted reactivity of an unknown substrate is higher than that of a known substrate, it indicates that the molecule may be a new substrate with higher catalytic efficiency, thus broadening the substrate spectrum of the enzyme and making it a priority for experimental verification. The substrate with the highest predicted reactivity can also be used as the screening result.
[0111] This embodiment enables virtual screening of a vast number of potential substrates and rapid prediction of their catalytic activity on a computer. This allows for focusing on the few most promising candidates before experiments, which greatly reduces the number of samples that need to be actually constructed and tested, significantly shortens the R&D cycle, and reduces costs.
[0112] Based on the above embodiments, the substrate library in this embodiment includes known substrates. Substrates are selected as screening results based on the predicted reactivity values of each substrate catalyzed by the target enzyme, including:
[0113] If the predicted reactivity of the target enzyme to catalyze each unknown substrate is greater than the predicted reactivity of the target enzyme to catalyze a known substrate, then the unknown substrate is used as the screening result.
[0114] In the experiment, the specific steps of virtual screening of new substrates included:
[0115] Screening objective: To find new substrates with higher catalytic efficiency for a certain enzyme.
[0116] Screening process: A molecular library containing 1000 structural analogs was constructed, including one known substrate of the enzyme. The point clouds of these 1000 substrates were paired with the point cloud of the enzyme, and then input into the model trained in Example 1 for Kcat prediction.
[0117] Screening results: The model prediction results show that 10 unknown compounds have higher predicted Kcat scores than the known substrates of the enzyme.
[0118] Conclusion: These 10 compounds have been identified as potential new substrates that can be used for subsequent experimental validation to broaden the substrate spectrum of this enzyme.
[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting the activity of an enzyme catalytic reaction, characterized in that, include: The point cloud of the enzyme to be predicted is obtained by encoding its three-dimensional structure, and the point cloud of the substrate catalyzed by the enzyme to be predicted is obtained by encoding its three-dimensional structure. The point cloud of the enzyme to be predicted and the point cloud of the substrate are used as inputs to a point cloud neural network model to obtain the predicted reactivity value of the enzyme to be predicted catalyzing the substrate, which is output by the point cloud neural network model. The point cloud neural network model is obtained by training a training set, which includes the point cloud of the sample enzyme, the point cloud of the substrate catalyzed by the sample enzyme, and the true value of the reaction activity of the substrate catalyzed by the sample enzyme.
2. The method for predicting enzyme catalytic reaction activity according to claim 1, characterized in that, Before encoding the three-dimensional structure of the enzyme to be predicted to obtain a point cloud of the enzyme, and before encoding the three-dimensional structure of the substrate catalyzed by the enzyme to obtain a point cloud of the substrate, the method further includes: The three-dimensional structure of the enzyme to be predicted was predicted using AlphaFold3 based on the amino acid sequence of the enzyme to be predicted. The RDKit is used to construct a three-dimensional conformation of the substrate based on the chemical structure of the enzyme to be predicted, which is then used as the three-dimensional structure of the substrate.
3. The method for predicting enzyme catalytic reaction activity according to claim 1, characterized in that, Encoding the three-dimensional structure of the enzyme to be predicted to obtain the point cloud of the enzyme includes: The atoms within a predetermined range surrounding the catalytic active center in the three-dimensional structure of the enzyme to be predicted are encoded to obtain the point cloud of the enzyme to be predicted.
4. The method for predicting enzyme catalytic reaction activity according to claim 1, characterized in that, Encoding the three-dimensional structure of the enzyme to be predicted to obtain a point cloud of the enzyme, and encoding the three-dimensional structure of the substrate catalyzed by the enzyme to obtain a point cloud of the substrate, includes: Each atom in the three-dimensional structure of the enzyme to be predicted and its corresponding substrate is regarded as a point in three-dimensional space; Based on the three-dimensional coordinates, atomic radius, and charge of each atom, a feature vector is constructed for the points corresponding to each atom. Based on the points and feature vectors corresponding to each atom of the enzyme to be predicted, the three-dimensional structure of the enzyme to be predicted is transformed into a multi-dimensional point cloud. Based on the points and feature vectors corresponding to each atom of the substrate of the enzyme to be predicted, the three-dimensional structure of the substrate of the enzyme to be predicted is transformed into a multi-dimensional point cloud.
5. The method for predicting enzyme catalytic reaction activity according to claim 1, characterized in that, The point cloud neural network model is based on the Transformer architecture, and the reactivity prediction value is the Kcat prediction value.
6. The method for predicting enzyme catalytic reaction activity according to claim 1, characterized in that, After the point cloud neural network model is trained using the training set, it also includes: The test set is input into the point cloud neural network model for testing to obtain the predicted value of the reaction activity corresponding to the test set. The test set includes the point cloud of the sample enzyme, the point cloud of the substrate catalyzed by the sample enzyme, and the true value of the reaction activity of the substrate catalyzed by the sample enzyme. The Spearman correlation coefficient between the predicted and actual reactivity values corresponding to the test set is calculated and used as the prediction performance index of the point cloud neural network model.
7. A method for modifying an enzyme, characterized in that, include: The target enzyme sequence was modified in different ways to obtain the modified target enzyme; Based on the enzyme catalytic reaction activity prediction method according to any one of claims 1-6, obtain the predicted reaction activity of the target enzyme before and after modification of the same substrate; The final modification result of the target enzyme is determined based on the predicted reactivity of the target enzyme before and after modification of the same substrate.
8. The enzyme modification method according to claim 7, characterized in that, Based on the predicted reactivity of the target enzyme before and after modification of the same substrate, the final modification result of the target enzyme is determined, including: If the predicted reactivity value of the target enzyme after modification is greater than the predicted reactivity value of the target enzyme before modification, and the difference between the predicted reactivity values of the target enzyme before and after modification is greater than a preset threshold, then the target enzyme after modification is determined to be the final modification result of the target enzyme.
9. A substrate screening method, characterized in that, include: Based on the enzyme catalytic reaction activity prediction method according to any one of claims 1-6, the predicted reaction activity values of each substrate in the target enzyme catalytic substrate molecule library are obtained. Based on the predicted reactivity values of the target enzymes catalyzing each substrate, the substrates were selected as the screening results.
10. The substrate screening method according to claim 9, characterized in that, The substrate library includes known substrates. Based on the predicted reactivity of each substrate catalyzed by the target enzyme, substrates are selected as screening results, including: If the predicted reactivity of the target enzyme to catalyze each unknown substrate is greater than the predicted reactivity of the target enzyme to catalyze a known substrate, then the unknown substrate is used as the screening result.