A training method and a prediction method of a deep learning model for predicting small molecule products of enzyme-catalyzed reactions
By combining a deep learning model with a pre-trained model and a weighted cross-entropy loss function, the problem of rapid, low-cost, and high-throughput prediction of small molecule products in enzyme-catalyzed reactions was solved, achieving efficient prediction of various enzyme-catalyzed reaction types.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-09
- Publication Date
- 2026-06-09
AI Technical Summary
Current technologies lack rapid, low-cost, and high-throughput methods for predicting small molecule products of enzyme-catalyzed reactions, especially through enzyme sequence and substrate information.
A deep learning model is employed, utilizing pre-trained protein language models and chemical molecule language models to obtain numerical embedding representations of enzyme sequences and substrates. Clustering and model training are performed using training data, and a weighted cross-entropy loss function and optimizer are used for training to achieve the prediction of small molecule products of enzyme-catalyzed reactions.
It enables low-cost, high-throughput prediction of small molecule products of enzyme-catalyzed reactions, improves the generalization ability of the prediction model, is applicable to a variety of enzyme-catalyzed reaction types, and does not require structural information or catalytic mechanism of the enzyme.
Smart Images
Figure CN122177242A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of biodetection technology, specifically to a training method and a prediction method for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions. Background Technology
[0002] For most enzymes, their catalytic reaction information is unknown. Biological experiments can be used to determine which products an enzyme catalyzes from its substrate molecules, but these experiments are time-consuming and expensive. Deep learning models offer a novel, rapid, low-cost, and high-throughput alternative for predicting enzyme-catalyzed reactions, demonstrating broader applicability and higher accuracy. Existing research on enzyme reaction prediction focuses on predicting protein function, substrate, or certain parameters, lacking a method to predict small molecule products solely based on enzyme sequence and substrate. Summary of the Invention
[0003] In view of this, the present invention provides a training method and a prediction method for a deep learning model for predicting small molecule products of enzyme-catalyzed reactions.
[0004] In a first aspect, the present invention provides a training method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions, comprising the following steps: acquiring training data, wherein the training data includes multiple training samples, wherein each training sample includes an enzyme sequence feature, multiple substrate features, and multiple product features; clustering the enzyme sequence features in the training data according to sequence similarity to obtain multiple enzyme sequence feature groups; splitting the training data into a training set, a validation set, and a test set according to the multiple enzyme sequence feature groups; defining a deep learning model to predict small molecule products of enzyme-catalyzed reactions; wherein in the deep learning model, a pre-trained protein language model and a chemical molecule language model are used to obtain numerical embedding representations of the enzyme sequence and substrate, respectively; training the deep learning model using the training set, and validating the deep learning model using the validation set during the training process; and testing the deep learning model using the test set after training is completed.
[0005] The present invention provides a training method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions. This method trains the deep learning model using training data to obtain a predictive model for small molecule products of enzyme-catalyzed reactions. When using the deep learning model to predict small molecule products of enzyme-catalyzed reactions, only enzyme sequence and substrate information are needed; enzyme structural information or EC numbers and other features are not required, nor is an understanding of the enzyme catalytic mechanism necessary. This allows for low-cost, high-throughput prediction of enzyme-catalyzed reaction products. Furthermore, the product prediction method is applicable to various types of enzyme-catalyzed reactions, rather than being limited to a single reaction type. Moreover, during the training of the deep learning model, enzyme sequence features in the training data are clustered according to sequence similarity. This prevents overly similar enzyme sequences from appearing in the training and test sets, thereby improving the generalization ability of the product prediction model.
[0006] In one optional implementation, before clustering the enzyme sequence features in the training data according to sequence similarity to obtain multiple enzyme sequence feature groups, the method further includes: sorting multiple substrate features in the training data from largest to smallest molecular weight; and sorting multiple product features in the training data from largest to smallest molecular weight.
[0007] Therefore, a well-trained deep learning model can learn the molecular weight relationship of products, and the predicted products obtained by using the deep learning model are more likely to be sorted in descending order of molecular weight.
[0008] In one optional implementation, in the deep learning model, obtaining numerical embedding representations of enzyme sequences and substrates using pre-trained protein language models and chemical molecular language models respectively includes: using a pre-trained protein language model to capture the structure, function, and / or biological characteristics of the enzyme sequence and outputting a numerical embedding representation of the enzyme sequence features; and using a pre-trained chemical molecular language model to capture the structure and / or properties of the substrate and outputting a numerical embedding representation of the substrate features.
[0009] In other words, protein language models, through self-supervised learning on a large number of protein sequences, can capture patterns and features in protein sequences, and the generated numerical embeddings can effectively represent the structure, function, and biological characteristics of enzymes; chemical molecule language models, through self-supervised training on a large number of chemical molecule SMILES characters, can capture the patterns and molecular structure information in these SMILES characters, and the resulting numerical embeddings of chemical molecule SMILES characters capture the structure, properties, and other physicochemical characteristics of chemical molecules.
[0010] In one alternative implementation, the protein language model is at least one of ProtBert and ESM2; and / or, the chemical molecule language model is at least one of ChemBERTa-2-MTR and ChemBERTa-2-MLM.
[0011] This allows for a higher accuracy rate in the product prediction model.
[0012] In one alternative implementation, training a deep learning model for predicting small molecule products of enzyme-catalyzed reactions using a training set includes: using a weighted cross-entropy loss function during the training process; and using an optimizer such as Adam, Nadam, SGD, RMSprop, Adagrad, Adadelta, or AdamW during the training process.
[0013] This improves the training efficiency and convergence speed of the model.
[0014] In one optional implementation, the method for calculating the cross-entropy loss function value includes: obtaining multiple prediction products obtained by the deep learning model based on the training set; calculating the cross-entropy loss function value corresponding to each prediction product among the multiple prediction products; for each prediction product among the multiple prediction products, using the ratio of the non-empty number of the corresponding real product to the total number of samples in the training set as the weight of each prediction product among the multiple prediction products; and determining the cross-entropy loss function value based on the cross-entropy loss function value corresponding to each prediction product among the multiple prediction products and the corresponding weight.
[0015] This avoids the impact of empty products on the training process.
[0016] Secondly, the present invention also provides a prediction method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions, comprising: obtaining an enzyme sequence and substrate from input data to be predicted; inputting the enzyme sequence and substrate into a deep learning model trained using the method of the first aspect to obtain the predicted product.
[0017] This invention provides a prediction method for a deep learning model of small molecule products of enzyme-catalyzed reactions. It can predict multiple products of enzyme-catalyzed reactions simultaneously, and only requires enzyme sequence features and substrate features. It does not require enzyme structural information or EC number or other features, nor does it require knowledge of the enzyme catalysis mechanism. Therefore, it can predict the products of enzyme-catalyzed reactions at low cost and high throughput, and can be applied to a variety of enzyme-catalyzed reaction types, rather than being limited to a single reaction type.
[0018] Thirdly, the present invention also provides a training device for a deep learning model for predicting small molecule products of enzyme-catalyzed reactions, comprising a data processing module, a model definition module, a training module, and a testing module. The data processing module is used to acquire training data, which includes multiple training samples. Each training sample includes an enzyme sequence feature, multiple substrate features, and multiple product features. The enzyme sequence features in the training data are clustered according to sequence similarity to obtain multiple enzyme sequence feature groups. The training data is split into a training set, a validation set, and a test set according to the multiple enzyme sequence feature groups. The model definition module is used to define a deep learning model for predicting small molecule products of enzyme-catalyzed reactions. In the deep learning model, pre-trained protein language models and chemical molecule language models are used to obtain numerical embedding representations of enzyme sequences and substrates, respectively. The training module is used to train the deep learning model using the training set and to validate the deep learning model using the validation set during the training process. The testing module is used to test the trained deep learning model using the test set after training is completed.
[0019] Fourthly, the present invention also provides a prediction device for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions, comprising an acquisition module and a product prediction module; the acquisition module is used to acquire the enzyme sequence and substrate from the input data to be predicted; the product prediction module is used to input the enzyme sequence and substrate into the deep learning model trained using the method of the first aspect to obtain the predicted product.
[0020] Fifthly, the present invention also provides a computer device, including a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the training method of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions described in the first aspect or any corresponding embodiment above, and / or the prediction method of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions described in the second aspect.
[0021] In a sixth aspect, the present invention also provides a computer-readable storage medium storing computer instructions for causing a computer to execute the product prediction method for enzyme-catalyzed reactions of the first aspect or any corresponding embodiment thereof and / or the prediction method of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions of the second aspect.
[0022] In a seventh aspect, the present invention also provides a computer program product, including computer instructions for causing a computer to execute the product prediction method for enzyme-catalyzed reactions of the first aspect or any corresponding embodiment described above and / or the prediction method of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions of the second aspect. Attached Figure Description
[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0024] Figure 1 This is a flowchart of a training method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention.
[0025] Figure 2 This is a flowchart of a training method for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions according to another embodiment of the present invention.
[0026] Figure 3 This is a schematic diagram of the network architecture of the EPP model according to an embodiment of the present invention;
[0027] Figure 4 This is a flowchart of a prediction method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention.
[0028] Figure 5 This is an algorithm logic diagram of a deep learning model for training and predicting small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention.
[0029] Figure 6 These are schematic diagrams illustrating the prediction effects of Example 1 to Example 4 of the enzyme-catalyzed reaction product prediction method according to embodiments of the present invention;
[0030] Figure 7 This is a structural block diagram of a training device for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention.
[0031] Figure 8 This is a structural block diagram of a prediction device for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention.
[0032] Figure 9 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation
[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0034] According to embodiments of the present invention, a training method and a prediction method embodiment for a deep learning model for predicting small molecule products of enzyme-catalyzed reactions are provided. It should be noted that the steps shown in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowcharts, in some cases, the steps shown or described may be executed in a different order than that shown here.
[0035] This embodiment provides a training method for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions, which can be used in computer devices. Figure 1 This is a flowchart of a training method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:
[0036] Step S101: Obtain training data, which includes multiple training samples, each of which includes an enzyme sequence feature, multiple substrate features, and multiple product features.
[0037] Step S102: Cluster the enzyme sequence features in the training data according to sequence similarity to obtain multiple enzyme sequence feature groups.
[0038] In one alternative implementation, the CD-HIT algorithm can be used to cluster enzyme sequences according to sequence similarity, and the clustered dataset can be split into training set, validation set and test set to prevent overly similar enzyme sequences from appearing in the training set and test set samples, thereby improving the generalization ability of the product prediction model.
[0039] Step S103: Divide the training data into training set, validation set and test set according to multiple enzyme sequence features.
[0040] Step S104: Define a deep learning model for predicting small molecule products of enzyme-catalyzed reactions; wherein the deep learning model uses pre-trained protein language models and chemical molecule language models to obtain numerical embedding representations of enzyme sequences and substrates, respectively.
[0041] In one optional implementation, in the deep learning model, obtaining numerical embedding representations of enzyme sequences and substrates using pre-trained protein language models and chemical molecular language models respectively includes: using a pre-trained protein language model to capture the structure, function, and / or biological characteristics of the enzyme sequence and outputting a numerical embedding representation of the enzyme sequence features; and using a pre-trained chemical molecular language model to capture the structure and / or properties of the substrate and outputting a numerical embedding representation of the substrate features.
[0042] In other words, protein language models, through self-supervised learning on a large number of protein sequences, can capture patterns and features in protein sequences, and the generated numerical embeddings can effectively represent the structure, function, and biological characteristics of enzymes; chemical molecule language models, through self-supervised training on a large number of chemical molecule SMILES characters, can capture the patterns and molecular structure information in these SMILES characters, and the resulting numerical embeddings of chemical molecule SMILES characters capture the structure, properties, and other physicochemical characteristics of chemical molecules.
[0043] Step S105: Train the deep learning model using the training set, and validate the deep learning model using the validation set during the training process.
[0044] Step S106: After training is complete, test the deep learning model using the test set.
[0045] The present invention provides a training method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions. This method trains the deep learning model using training data to obtain a predictive model for small molecule products of enzyme-catalyzed reactions. When using the deep learning model to predict small molecule products of enzyme-catalyzed reactions, only enzyme sequence and substrate information are needed; enzyme structural information or EC numbers and other features are not required, nor is an understanding of the enzyme catalytic mechanism necessary. This allows for low-cost, high-throughput prediction of enzyme-catalyzed reaction products. Furthermore, the product prediction method is applicable to various types of enzyme-catalyzed reactions, rather than being limited to a single reaction type. Moreover, during the training of the deep learning model, enzyme sequence features in the training data are clustered according to sequence similarity. This prevents overly similar enzyme sequences from appearing in the training and test sets, thereby improving the generalization ability of the product prediction model.
[0046] This embodiment provides a training method for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions, which can be used in computer devices. Figure 2 This is a flowchart of another method for training a deep learning model to predict small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:
[0047] Step S201: Obtain training data, which includes multiple training samples, each of which includes an enzyme sequence feature, multiple substrate features, and multiple product features.
[0048] Step S202: Sort the multiple substrate features in the training data in descending order of molecular weight.
[0049] Step S203: Sort the multiple product features in the training data from largest to smallest molecular weight.
[0050] Through the above steps S202 to S203, the trained deep learning model can learn the molecular weight relationship of the products, and the predicted products obtained by the deep learning model are more likely to be sorted in descending order of molecular weight.
[0051] Step S204: Cluster the enzyme sequence features in the training data according to sequence similarity to obtain multiple enzyme sequence feature groups.
[0052] Step S205: The training data is split into training set, validation set and test set according to multiple enzyme sequence features.
[0053] Step S206: Define a deep learning model for predicting small molecule products of enzyme-catalyzed reactions; wherein the deep learning model uses pre-trained protein language models and chemical molecule language models to obtain numerical embedding representations of enzyme sequences and substrates, respectively.
[0054] Step S207: Train the deep learning model using the training set, and validate the deep learning model using the validation set during the training process.
[0055] As shown above, the training process uses a weighted cross-entropy loss function; the optimizer used during training is Adam, Nadam, SGD, RMSprop, Adagrad, Adadelta, or AdamW.
[0056] Specifically, the method for calculating the cross-entropy loss function value includes: obtaining multiple prediction products obtained by the deep learning model based on the training set; calculating the cross-entropy loss function value corresponding to each prediction product among the multiple prediction products; for each prediction product among the multiple prediction products, using the ratio of the number of non-empty real products to the total number of samples in the training set as the weight of each prediction product among the multiple prediction products; and determining the cross-entropy loss function value based on the cross-entropy loss function value and the corresponding weight of each prediction product among the multiple prediction products.
[0057] In other words, a weighted cross-entropy loss function is used in the training process of a deep learning model to learn and predict non-empty products. The weighting is based on the number of non-empty products to avoid the influence of empty products on the training process. For example, if the number of three non-empty products is 100, 90, and 80, and the total number of samples is 100, then the weighted cross-entropy loss = (100 / 100*loss1 + 90 / 100*loss2 + 80 / 100*loss3) / 3.
[0058] Step S208: After training is complete, test the deep learning model using the test set.
[0059] In one alternative implementation, such as Figure 3 As shown, the deep learning model (referred to as the EPP model) includes a protein language model, a chemical molecule language model, an encoder module, and three decoder modules (decoder1, decoder2, and decoder3). A pre-trained protein language model captures the structure, function, and / or biological characteristics of the enzyme sequence, outputting a numerical embedding representation of the enzyme sequence. A pre-trained chemical molecule language model captures the structure and / or properties of the substrate, outputting a numerical embedding representation of the substrate. The encoder module uses a multilayer perceptron (MLP) architecture, containing at least two linear layers, to receive the numerical embedding representations of the enzyme sequence and small molecule substrates, generating intermediate features. All three decoder modules are also multilayer perceptron (MLP) architectures, each containing at least two linear layers, outputting the codes for three predicted products. Through a deep neural network based on a multilayer perceptron (MLP) architecture, it is possible to learn the data patterns between enzymes, substrates, and products from a large amount of enzyme catalytic reaction data.
[0060] The present invention provides a training method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions. This method trains the deep learning model using training data to obtain the predicted product. When using a deep learning model to predict small molecule products of enzyme-catalyzed reactions, only the enzyme sequence and substrate information are needed; no structural information or EC number or other features are required, nor is an understanding of the enzyme catalysis mechanism necessary. This allows for low-cost, high-throughput prediction of enzyme-catalyzed reaction products. Furthermore, this product prediction method is applicable to various enzyme-catalyzed reaction types, rather than being limited to a single reaction type.
[0061] This embodiment provides a prediction method for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions, which can be used in computer devices. Figure 4 This is a flowchart of a prediction method using a deep learning model to predict small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention. Figure 5This is an algorithm logic diagram of a deep learning model for training and predicting small molecule products of enzyme-catalyzed reactions according to an embodiment of the present invention, as shown in the following figure. Figure 4 and Figure 5 As shown, the process includes the following steps:
[0062] Step S401: Obtain the enzyme sequence and substrate from the input data to be predicted.
[0063] Step S402: Input the enzyme sequence and substrate into the deep learning model trained using the above method to obtain the predicted product.
[0064] To illustrate the predictive performance of a general deep learning model for predicting small molecule products of enzyme-catalyzed reactions according to embodiments of the present invention, the following four examples are provided.
[0065] Example 1: During the training phase of the EPP model, the numerical embedding of the enzyme sequence used the pre-trained protein language model ProtBERT; the numerical embedding of the substrate used the pre-trained ChemBERTa-2-MTR. The final trained model achieved a prediction accuracy of over 91% on the independent test set. Specifically, as follows... Figure 6 As shown, the prediction accuracies for the three products were 91.5%, 92.2%, and 96.5%, respectively.
[0066] Example 2: During the training phase of the EPP model, the numerical embedding of the enzyme sequence used the pre-trained protein language model ProtBERT; the numerical embedding of the substrate used the pre-trained ChemBERTa-2-MLM. The trained model achieved a prediction accuracy of over 89% on the independent test set, such as... Figure 6 As shown, the prediction accuracies for the three products were 90.7%, 89.7%, and 97.1%, respectively.
[0067] Example 3: During the training phase of the EPP model, the numerical embedding of the enzyme sequence used a pre-trained ESM2 protein language model; the numerical embedding of the substrate used a pre-trained ChemBERTa-2-MLR model. The trained model achieved a prediction accuracy of over 90% on the independent test set, such as... Figure 6 As shown, the prediction accuracies for the three products were 90.7%, 91.5%, and 97.0%, respectively.
[0068] Example 4: During the training phase of the EPP model, the numerical embedding of the enzyme sequence used a pre-trained ESM2 protein language model; the numerical embedding of the substrate used a pre-trained ChemBERTa-2-MLM. The trained model achieved a prediction accuracy of over 89% on the independent test set, such as... Figure 6 As shown, the prediction accuracies for the three products were 89.4%, 89.7%, and 97.3%, respectively.
[0069] In summary, this invention provides a general deep learning computational framework for predicting small molecule products of enzyme-catalyzed reactions. High-quality protein data with catalytic information is obtained from a protein database. Pre-trained protein language models and chemical molecule language models are used to represent enzyme sequences and small molecule substrates. A deep neural network is constructed to receive numerical representations of enzyme sequences and small molecule substrates and output three small molecule products. The model is trained using a weighted cross-entropy loss function and an Adam optimizer, with an early stopping strategy employed to prevent overfitting. The final trained model can predict multiple small molecule products of enzyme-catalyzed reactions based on enzyme sequences and small molecule substrates.
[0070] This embodiment also provides a training and prediction device for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.
[0071] This embodiment provides a training device for a deep learning model to predict small molecule products of enzyme-catalyzed reactions, such as... Figure 7 As shown, it includes a data processing module 701, a model definition module 702, a training module 703, and a testing module 704.
[0072] The data processing module 701 is used to acquire training data, which includes multiple training samples. Each training sample includes an enzyme sequence feature, multiple substrate features, and multiple product features. The enzyme sequence features in the training data are clustered according to sequence similarity to obtain multiple enzyme sequence feature groups. The training data is then split into a training set, a validation set, and a test set according to the multiple enzyme sequence feature groups.
[0073] The model definition module 702 is used to define a deep learning model for predicting small molecule products of enzyme-catalyzed reactions; wherein in the deep learning model, pre-trained protein language models and chemical molecule language models are used to obtain numerical embedding representations of enzyme sequences and substrates, respectively.
[0074] The training module 703 is used to train the deep learning model using the training set and to validate the deep learning model using the validation set during the training process.
[0075] The testing module 704 is used to test the trained deep learning model using the test set after training is completed.
[0076] In some optional implementations, before clustering the enzyme sequence features in the training data according to sequence similarity to obtain multiple enzyme sequence feature groups, the data processing module 701 is further configured to: sort multiple substrate features in the training data from largest to smallest molecular weight; and sort multiple product features in the training data from largest to smallest molecular weight.
[0077] In some optional implementations, the model definition module 702 is specifically used to: use a pre-trained protein language model to capture the structure, function and / or biological characteristics of an enzyme sequence and output a numerical embedding representation of the enzyme sequence features; and use a pre-trained chemical molecule language model to capture the structure and / or properties of a substrate and output a numerical embedding representation of the substrate features.
[0078] In some optional implementations, the protein language model is at least one of ProtBert and ESM2; and / or, the chemical molecule language model is at least one of ChemBERTa-2-MTR and ChemBERTa-2-MLM.
[0079] In some alternative implementations, the training module 703 is specifically configured to: use a weighted cross-entropy loss function during the training process; and employ an optimizer such as Adam, Nadam, SGD, RMSprop, Adagrad, Adadelta, or AdamW during the training process.
[0080] In some optional implementations, the training module 703 is specifically used to obtain multiple prediction products obtained by the deep learning model based on the training set; calculate the cross-entropy loss function value corresponding to each prediction product among the multiple prediction products; for each prediction product among the multiple prediction products, use the ratio of the non-empty number of the corresponding real product to the total number of samples in the training set as the weight of each prediction product among the multiple prediction products; and determine the cross-entropy loss function value based on the cross-entropy loss function value corresponding to each prediction product among the multiple prediction products and the corresponding weight.
[0081] This embodiment provides a prediction device for a deep learning model to predict small molecule products of enzyme-catalyzed reactions, such as... Figure 8 The diagram includes an acquisition module 801 and a product prediction module 802.
[0082] The acquisition module 801 is used to acquire the enzyme sequence and substrate from the input data to be predicted;
[0083] The product prediction module 802 is used to input the enzyme sequence and substrate into the deep learning model trained by the above method to obtain the predicted product.
[0084] In this embodiment, the training and prediction devices of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions are presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit), a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.
[0085] This invention also provides a computer device having the above-described features. Figure 7 The training apparatus shown is for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions. Figure 8 The device shown is a prediction device for a deep learning model that predicts small molecule products of enzyme-catalyzed reactions.
[0086] Please see Figure 9 , Figure 9 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 9 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 9 Take a processor 10 as an example.
[0087] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.
[0088] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.
[0089] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0090] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.
[0091] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 9 Taking the example of a connection between China and Israel via a bus.
[0092] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.
[0093] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded over a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.
[0094] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.
[0095] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.
Claims
1. A method for training a deep learning model to predict small molecule products of enzyme-catalyzed reactions, characterized in that, include: Acquire training data, which includes multiple training samples, each of which includes an enzyme sequence feature, multiple substrate features, and multiple product features; The enzyme sequence features in the training data are clustered and grouped according to sequence similarity to obtain multiple enzyme sequence feature groups; The training data is grouped and split into training set, validation set and test set according to multiple enzyme sequence features; A deep learning model for predicting small molecule products of enzyme-catalyzed reactions is defined; wherein, in the deep learning model, pre-trained protein language models and chemical molecule language models are used to obtain numerical embedding representations of enzyme sequences and substrates, respectively; The deep learning model is trained using the training set, and the deep learning model is validated using the validation set during the training process. After training, the deep learning model is tested using the test set.
2. The method according to claim 1, characterized in that, Before clustering and grouping the enzyme sequence features in the training data according to sequence similarity to obtain multiple enzyme sequence feature groups, the process further includes: The multiple substrate features in the training data are sorted from largest to smallest molecular weight; The multiple product features in the training data are sorted from largest to smallest molecular weight.
3. The method according to claim 1, characterized in that, The step of obtaining numerical embedding representations of enzyme sequences and substrates using pre-trained protein language models and chemical molecule language models in the deep learning model includes: A pre-trained protein language model is used to capture the structural, functional, and / or biological characteristics of an enzyme sequence and output a numerical embedding representation of the enzyme sequence features. A pre-trained chemical molecular language model is used to capture the structure and / or properties of the substrate and output a numerical embedding representation of the substrate features.
4. The method according to claim 3, characterized in that: The protein language model is at least one of ProtBert and ESM2; And / or, The chemical molecular language model is at least one of ChemBERTa-2-MTR and ChemBERTa-2-MLM.
5. The method according to claim 1, characterized in that, The step of training the deep learning model using the training set, and validating the deep learning model using the validation set during the training process, includes: The training process uses a weighted cross-entropy loss function; The optimizers used during training are Adam, Nadam, SGD, RMSprop, Adagrad, Adadelta, or AdamW.
6. The method according to claim 5, characterized in that, The method for calculating the cross-entropy loss function value includes: Obtain multiple prediction products obtained by the deep learning model based on the training set; Calculate the cross-entropy loss function value for each of the multiple predicted products; for each of the multiple predicted products, use the ratio of the number of non-empty real products to the total number of samples in the training set as the weight of each of the multiple predicted products. The cross-entropy loss function value is determined based on the cross-entropy loss function value and the corresponding weight for each of the multiple predicted products.
7. A prediction method for a deep learning model to predict small molecule products of enzyme-catalyzed reactions, characterized in that, include: Obtain the enzyme sequence and substrate from the input data to be predicted; The enzyme sequence and the substrate are input into a deep learning model trained using the method of any one of claims 1 to 6 to obtain a predicted product.
8. A training device for a deep learning model to predict small molecule products of enzyme-catalyzed reactions, characterized in that, include: The data processing module is used to acquire training data, which includes multiple training samples. Each training sample includes one enzyme sequence feature, multiple substrate features, and multiple product features; The enzyme sequence features in the training data are clustered and grouped according to sequence similarity to obtain multiple enzyme sequence feature groups; the training data is then split into a training set, a validation set, and a test set according to the multiple enzyme sequence feature groups. The model definition module is used to define a deep learning model for predicting small molecule products of enzyme-catalyzed reactions; wherein, in the deep learning model, pre-trained protein language models and chemical molecule language models are used to obtain numerical embedding representations of enzyme sequences and substrates, respectively. The training module is used to train the deep learning model using the training set, and to validate the deep learning model using the validation set during the training process. The testing module is used to test the trained deep learning model using the test set after training is completed.
9. A predictive device for a deep learning model to predict small molecule products of enzyme-catalyzed reactions, characterized in that, include: The acquisition module is used to obtain the enzyme sequence and substrate from the input data to be predicted; The product prediction module is used to input the enzyme sequence and the substrate into a deep learning model trained using the method of any one of claims 1 to 6 to obtain the predicted product.
10. A computer device, characterized in that, include: The system includes a memory and a processor, which are interconnected and communicate with each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the training method of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions according to any one of claims 1 to 6 and / or the prediction method of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions according to claim 7.
11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the product prediction method of the enzyme-catalyzed reaction according to any one of claims 1 to 6 and / or the prediction method of the deep learning model for predicting small molecule products of the enzyme-catalyzed reaction according to claim 7.
12. A computer program product, characterized in that, Includes computer instructions for causing a computer to execute the product prediction method for enzyme-catalyzed reactions according to any one of claims 1 to 6 and / or the prediction method of the deep learning model for predicting small molecule products of enzyme-catalyzed reactions according to claim 7.