A protein sequence property prediction method, device, equipment and storage medium

Abstract

The application relates to a protein sequence property prediction method, device, equipment and storage medium, wherein the method comprises the following steps: constructing a multi-task classification model based on an ESM protein language model and a multi-task classifier; inputting a protein sequence data set into the multi-task classification model for training, wherein the protein sequence data set comprises multiple specified protein sequences and corresponding labels, the ESM protein language model is used for feature extraction of each specified protein sequence to obtain a first feature vector matrix; the multi-task classifier is used for training the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model; and a target protein sequence is input into the protein property prediction model for prediction to obtain the enzyme class, substrate type and nucleic acid binding property of the target protein, thereby improving the prediction efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a protein sequence property prediction method, device, equipment and storage medium. BACKGROUND

[0002] With the rapid development of genome sequencing technology, protein sequence data is growing exponentially, but the coverage of functional annotation verified by experiments is less than 5%. Traditional protein function prediction technology relies on the following methods:

[0003] Homology alignment-based method: function is inferred by sequence similarity, but the prediction accuracy of distantly related homologous proteins is significantly reduced. Manual feature engineering method: extract shallow features such as amino acid composition and physicochemical properties, and combine machine learning models (such as SVM, random forest) for classification, but cannot capture deep semantic associations of sequences, and has problems such as time-consuming feature engineering and low prediction accuracy. And the existing single-task model has the problems of low computational efficiency and insufficient feature reuse ability in the multi-label prediction scene. SUMMARY

[0004] Therefore, it is necessary to provide a protein sequence property prediction method, device, equipment and storage medium in view of the above technical problems.

[0005] In a first aspect, the embodiments of the present application provide a protein sequence property prediction method, which comprises:

[0006] building a multi-task classification model based on an ESM protein language model and a multi-task classifier;

[0007] inputting a protein sequence dataset into the multi-task classification model for training, wherein the protein sequence dataset comprises a plurality of specified protein sequences and corresponding labels, the ESM protein language model is used to extract features of each specified protein sequence to obtain a first feature vector matrix, and the multi-task classifier is used to train the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model;

[0008] inputting a target protein sequence into the protein property prediction model for prediction to obtain the enzyme class, substrate type and nucleic acid binding property of the target protein.

[0009] In one embodiment, the multi-task classifier is used to train the multi-task classification model based on the first feature vector matrix and the corresponding labels, which comprises:

[0010] inputting the first feature vector matrix and the corresponding labels into the multi-task classifier to perform initial prediction, obtaining initial prediction results of enzyme categories, substrate types and nucleic acid binding properties of each specified protein, and constructing a first vector matrix based on the initial prediction results;

[0011] randomly replacing part of elements in the first vector matrix with real labels to obtain a second vector matrix;

[0012] obtaining an error matrix based on the second vector matrix and the first vector matrix;

[0013] performing feature fusion on the error matrix and the first feature vector matrix to obtain a second feature vector matrix;

[0014] inputting the second feature vector matrix into the multi-task classifier to perform prediction, obtaining enzyme categories, substrate types and nucleic acid binding properties of each specified protein.

[0015] In one of the embodiments, the obtaining of the error matrix based on the second vector matrix and the first vector matrix comprises:

[0016] obtaining the error matrix based on the difference between the second vector matrix and the first vector matrix.

[0017] In one of the embodiments, the performing of the feature fusion on the error matrix and the first feature vector matrix to obtain the second feature vector matrix comprises:

[0018] inputting the error matrix into a self-attention mechanism module to perform feature enhancement to obtain an enhanced error matrix;

[0019] performing feature fusion on the enhanced error matrix and the first feature vector matrix to obtain the second feature vector matrix.

[0020] In one of the embodiments, the multi-task classifier comprises a first classifier, a second classifier and a third classifier, the ESM protein language model is connected in parallel with the first classifier, the second classifier and the third classifier; the first classifier is used to predict the enzyme category of the protein, the second classifier is used to predict the substrate category of the protein, and the third classifier is used to predict the binding property of the protein.

[0021] In one of the embodiments, the multi-task classifier is used to train the multi-task classification model based on the first feature vector matrix and the corresponding labels, comprising:

[0022] inputting the first feature vector matrix into the first classifier, the second classifier and the third classifier respectively for prediction to obtain a predicted probability of the enzyme class, a predicted probability of the substrate class and a predicted probability of the nucleic acid binding property;

[0023] calculating a first cross-entropy loss based on the predicted probability of the enzyme class and a first label, calculating a second cross-entropy loss based on the predicted probability of the substrate class and a second label, and calculating a third cross-entropy loss based on the predicted probability of the nucleic acid binding property and a third label;

[0024] fine-tuning parameters of the multi-task classification model based on an average of the first cross-entropy loss, the second cross-entropy loss and the third cross-entropy loss.

[0025] In one of the embodiments, before the protein sequence dataset is input into the multi-task classification model for training, the method further comprises:

[0026] performing data preprocessing on each of the specified protein sequences and padding to a fixed length.

[0027] In a second aspect, the embodiments of the present application further provide a protein sequence property prediction device, the device comprising:

[0028] a model construction module configured to construct a multi-task classification model based on an ESM protein language model and a multi-task classifier;

[0029] a model training module configured to input a protein sequence dataset into the multi-task classification model for training, wherein the protein sequence dataset comprises a plurality of specified protein sequences and corresponding labels, the ESM protein language model is configured to perform feature extraction on each of the specified protein sequences to obtain a first feature vector matrix, and the multi-task classifier is configured to train the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model;

[0030] a model prediction module configured to input a target protein sequence into the protein property prediction model for prediction to obtain an enzyme class, a substrate type and a nucleic acid binding property of the target protein.

[0031] In a third aspect, the embodiments of the present application further provide a computer device comprising a memory and a processor, the memory storing a computer program, and the processor is configured to run the computer program to execute the method of the first aspect described above.

[0032] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, wherein the storage medium stores a computer program, and the computer program is executed by a processor to implement the method in the first aspect.

[0033] The protein sequence property prediction method, device, equipment and readable storage medium described above, by constructing a multi-task classification model based on an ESM protein language model and a multi-task classifier, inputting a protein sequence dataset into the multi-task classification model for training, wherein the protein sequence dataset includes a plurality of specified protein sequences and corresponding labels, the ESM protein language model is used to extract features of each specified protein sequence to obtain a first feature vector matrix, and the multi-task classifier is used to train the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model, and inputting a target protein sequence into the protein property prediction model for prediction to obtain the enzyme class, substrate type and nucleic acid binding property of the target protein, which realizes the prediction of multiple properties of the protein and improves the prediction efficiency and accuracy.

[0034] The details of one or more embodiments of the present application are presented in the following drawings and description to make other features, objects and advantages of the present application more apparent. BRIEF DESCRIPTION OF DRAWINGS

[0035] The accompanying drawings illustrated herein are used to provide further understanding of the present application, constitute a part of the present application, the schematic embodiments of the present application and the description thereof are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0036] Figure 1 It is a hardware structure block diagram of a terminal device of the protein sequence property prediction method in an embodiment;

[0037] Figure 2 It is a flowchart of the protein sequence property prediction method in an embodiment;

[0038] Figure 3 It is a flowchart of the multi-task classification model training in an embodiment;

[0039] Figure 4 It is a structure block diagram of the protein sequence property prediction device in an embodiment;

[0040] Figure 5 It is a structure schematic diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and explained below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0042] It is obvious that the drawings in the following description are only some examples or embodiments of the present application, and for those of ordinary skill in the art, the present application can also be applied to other similar scenarios without creative labor on the basis of these drawings. In addition, it can be understood that although the efforts made in this development process can be complex and lengthy, for those of ordinary skill in the art related to the content disclosed in the present application, some design, manufacture or production changes based on the technology disclosed in the present application are only routine technical means and should not be understood as insufficient disclosure of the present application.

[0043] The method embodiments provided in the present embodiment can be executed in a terminal, a computer or a similar computing device. For example, the method embodiments are run on a terminal, Figure 1 is a hardware structure block diagram of the terminal of the protein sequence property prediction method of the present embodiment. As shown in Figure 1 , the terminal can include one or more (only one is shown in Figure 1 ) processor 102 and memory 104 for storing data, wherein the processor 102 can include but is not limited to processing devices such as microprocessor MCU or programmable logic device FPGA. The above terminal can also include a transmission device 106 for communication function and an input and output device 108. Those of ordinary skill in the art can understand that Figure 1 The structure shown is only schematic, which does not limit the structure of the above terminal. For example, the terminal can also include more or less components than Figure 1 shown, or have a different configuration from Figure 1 shown.

[0044] The memory 104 can be used to store computer programs, such as software programs of application software and modules, such as a computer program corresponding to the protein sequence property prediction method in the embodiment. The processor 102 can execute various functional applications and data processing, that is, implement the method described above, by running the computer program stored in the memory 104. The memory 104 can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 104 can further include a memory remotely arranged with respect to the processor 102, which can be connected to the terminal through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0045] The transmission device 106 is used to receive or send data via a network. The network described above includes a wireless network provided by a communication provider of the terminal. In an example, the transmission device 106 includes a network interface controller (NIC) which can be connected to other network devices through a base station so as to communicate with the Internet. In an example, the transmission device 106 can be a radio frequency (RF) module which is used to communicate with the Internet in a wireless manner.

[0046] The embodiment of the present application provides a protein sequence property prediction method. The method is applied to a terminal in Figure 1 , as shown in Figure 2 , the method comprises the following steps:

[0047] In step 201, a multi-task classification model is constructed based on an ESM protein language model and a multi-task classifier.

[0048] The ESM (Evolutionary Scale Modeling) protein language model is an advanced model based on deep learning technology, which can effectively process long sequence data and capture long-range dependencies in protein sequences.

[0049] The embodiment of the present application constructs a multi-task classification model based on an ESM protein language model and a multi-task classifier, which is used to predict multiple properties of a protein. The ESM protein language model is used to extract features of a protein sequence and capture deep semantic associations in the protein sequence, so as to obtain high-dimensional semantic features shared by the multi-task classifier. The multi-task classifier is used to predict multiple properties of a protein based on the shared high-dimensional semantic features, and output corresponding prediction results.

[0050] For example, the multi-task classification model includes an ESM protein language model and three independent classifiers connected in parallel with the ESM protein language model, and the three independent classifiers are respectively used for predicting the enzyme class, substrate class and nucleic acid binding characteristics of the protein.

[0051] In step 202, a protein sequence dataset is input into the multi-task classification model for training, wherein the protein sequence dataset includes a plurality of specified protein sequences and corresponding labels, the ESM protein language model is used for feature extraction of each specified protein sequence to obtain a first feature vector matrix, and the multi-task classifier is used for training the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model.

[0052] In this step, the multi-task classification model is trained by the protein sequence dataset to obtain a protein property prediction model, which is used to predict the enzyme class, substrate class and nucleic acid binding characteristics of the protein.

[0053] In step 203, a target protein sequence is input into the protein property prediction model for prediction to obtain the enzyme class, substrate type and nucleic acid binding characteristics of the target protein.

[0054] In the above protein sequence property prediction method, a multi-task classification model is constructed based on an ESM protein language model and a multi-task classifier; a protein sequence dataset is input into the multi-task classification model for training, wherein the protein sequence dataset includes a plurality of specified protein sequences and corresponding labels, the ESM protein language model is used for feature extraction of each specified protein sequence to obtain a first feature vector matrix, and the multi-task classifier is used for training the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model; a target protein sequence is input into the protein property prediction model for prediction to obtain the enzyme class, substrate type and nucleic acid binding characteristics of the target protein, thereby realizing prediction of multiple properties of the protein and improving the prediction efficiency and accuracy.

[0055] In one embodiment, the multi-task classifier is used for training the multi-task classification model based on the first feature vector matrix and the corresponding labels, including the following steps:

[0056] In step 301, the first feature vector matrix and the corresponding labels are input into the multi-task classifier for initial prediction to obtain initial prediction results of the enzyme class, substrate type and nucleic acid binding characteristics of each specified protein, and a first vector matrix is constructed based on each initial prediction result.

[0057] Specifically, the first feature vector corresponding to each specified protein sequence and the corresponding true label are input into the multi-task classifier to perform initial prediction, and initial prediction results of the enzyme class, substrate type and nucleic acid binding property of each specified protein are obtained.

[0058] For example, the specified protein is a deaminase, wherein the corresponding enzyme class is divided into 19 classes, the substrate class is divided into 2 classes, and the nucleic acid binding class is divided into 6 classes. The initial prediction result includes the prediction probability corresponding to the 19 enzyme classes, the prediction probability corresponding to the 2 substrate classes, and the prediction probability corresponding to the 6 nucleic acid binding classes. The prediction results of each classification task are spliced to form a first vector matrix M1.

[0059] In step 302, part of the elements in the first vector matrix are randomly replaced with true labels to obtain a second vector matrix.

[0060] Specifically, a part of the elements in the first vector matrix M1 are randomly masked, and then the masked prediction probability is filled with the true label, and the second vector matrix M2 is obtained.

[0061] In step 303, an error matrix is obtained based on the second vector matrix and the first vector matrix.

[0062] Specifically, the second vector matrix M2 is subtracted from the first vector matrix M1 to obtain the error matrix.

[0063] In step 304, the error matrix and the first feature vector matrix are fused to obtain a second feature vector matrix.

[0064] In step 305, the second feature vector matrix is input into the multi-task classifier to perform prediction, and the enzyme class, substrate type and nucleic acid binding property of each specified protein are obtained.

[0065] In one embodiment, the error matrix is obtained based on the difference between the second vector matrix and the first vector matrix.

[0066] In one embodiment, the error matrix and the first feature vector matrix are fused to obtain a second feature vector matrix, which includes: inputting the error matrix into a self-attention mechanism module to perform feature enhancement to obtain an enhanced error matrix; and fusing the enhanced error matrix and the first feature vector matrix to obtain a second feature vector matrix.

[0067] In the embodiments of the present application, a dynamic attention mechanism is constructed by the error of the known label and the prediction result, the error matrix is input into the self-attention mechanism module for feature enhancement, then the enhanced error matrix and the first feature vector matrix are fused to obtain the second feature vector matrix, and the second feature vector matrix is input into the network of the multi-task classifier, the initial first feature vector matrix and the error information are fused to optimize the feature representation, and finally a more accurate classification prediction is output.

[0068] In one of the embodiments, as shown in Figure 3 The multi-task classification model can be divided into a feature extraction module, a classification module and a feature enhancement module. The feature extraction module performs feature extraction through the ESM protein language model, the classification module includes a first classifier, a second classifier and a third classifier connected in parallel with the ESM protein language model; the first classifier is used to predict the enzyme class of the protein, the second classifier is used to predict the substrate class of the protein, and the third classifier is used to predict the binding property of the protein. The first classifier, the second classifier and the third classifier constitute a multi-task classifier.

[0069] In the training process of the multi-task classification model, the protein sequence data set is input into the ESM protein language model for feature extraction to obtain a high-dimensional feature vector F1, then the high-dimensional feature vector F1 is input into the first classifier, the second classifier and the third classifier respectively to obtain the initial prediction result p1 of the enzyme class, the initial prediction result p2 of the substrate type and the initial prediction result p3 of the nucleic acid binding property. The initial prediction result p1 includes the prediction probability corresponding to each enzyme class, the initial prediction result p2 includes the prediction probability corresponding to each substrate class, and the initial prediction result p3 includes the prediction probability corresponding to each nucleic acid binding class. The initial prediction results of each classification task are spliced to form a first vector matrix M1. A part of elements in the first vector matrix M1 is randomly masked to obtain a third vector matrix M3, then the masked elements are filled with real labels (p'1 / p'2 / p'3) to obtain a second vector matrix M2. The second vector matrix M2 is subtracted from the first vector matrix M1 to obtain an error matrix. The error matrix is input into the self-attention mechanism module for feature enhancement to obtain an enhanced error matrix; the enhanced error matrix and the high-dimensional feature vector F1 are fused for feature fusion to obtain a second feature vector matrix F2, and the second feature vector matrix F2 is input into the first classifier, the second classifier and the third classifier to predict the final classification label to obtain the final prediction result p''1 of the enzyme class, the final prediction result p''2 of the substrate type and the final prediction result p''3 of the nucleic acid binding property.

[0070] In one of the embodiments, the multi-task classifier is used to train the multi-task classification model based on the first feature vector matrix and corresponding labels, and the following includes:

[0071] The first feature vector matrix is input into the first classifier, the second classifier and the third classifier respectively for prediction to obtain the predicted probability of the enzyme class, the predicted probability of the substrate class and the predicted probability of the nucleic acid binding property; based on the predicted probability of the enzyme class and the first label, a first cross-entropy loss is calculated; the first label is the true label of the enzyme class; based on the predicted probability of the substrate class and the second label, a second cross-entropy loss is calculated; the second label is the true label of the substrate class; based on the predicted probability of the nucleic acid binding property and the third label, a third cross-entropy loss is calculated; the third label is the true label of the nucleic acid binding property; based on the average value of the first cross-entropy loss, the second cross-entropy loss and the third cross-entropy loss, the parameters of the multi-task classification model are fine-tuned.

[0072] In the embodiments of the present application, in the process of training the multi-task classification model, the multi-task classification model is trained by updating the average value of the losses of the three classifiers, and the model parameters of the ESM protein language model and the parameters of each classifier are dynamically fine-tuned.

[0073] In one of the embodiments, before the protein sequence data set is input into the multi-task classification model for training, the method further includes: data preprocessing is performed on each of the specified protein sequences, and all the specified protein sequences are padded to a fixed length.

[0074] It should be noted that the steps shown in the above flow or the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0075] The embodiments of the present application also provide a protein sequence property prediction device, as shown in Figure 4 The device includes:

[0076] The model construction module 10 is configured to construct a multi-task classification model based on an ESM protein language model and a multi-task classifier.

[0077] The model training module 20 is configured to input a protein sequence dataset into the multi-task classification model for training, wherein the protein sequence dataset comprises a plurality of specified protein sequences and corresponding labels, the ESM protein language model is configured to perform feature extraction on each of the specified protein sequences to obtain a first feature vector matrix, and the multi-task classifier is configured to train the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model.

[0078] The model prediction module 30 is configured to input a target protein sequence into the protein property prediction model for prediction to obtain an enzyme class, a substrate type, and a nucleic acid binding property of the target protein.

[0079] In one embodiment, the model training module 20 is further configured to: input the first feature vector matrix and the corresponding labels into the multi-task classifier for initial prediction to obtain initial prediction results of the enzyme class, the substrate type, and the nucleic acid binding property of each of the specified proteins, and construct a first vector matrix based on each of the initial prediction results; randomly replace part of elements in the first vector matrix with real labels to obtain a second vector matrix; obtain an error matrix based on the second vector matrix and the first vector matrix; perform feature fusion on the error matrix and the first feature vector matrix to obtain a second feature vector matrix; and input the second feature vector matrix into the multi-task classifier for prediction to obtain the enzyme class, the substrate type, and the nucleic acid binding property of each of the specified proteins.

[0080] In one embodiment, the model training module 20 is further configured to: obtain an error matrix based on a difference between the second vector matrix and the first vector matrix.

[0081] In one embodiment, the model training module 20 is further configured to: input the error matrix into a self-attention mechanism module for feature enhancement to obtain an enhanced error matrix; and perform feature fusion on the enhanced error matrix and the first feature vector matrix to obtain a second feature vector matrix.

[0082] In one embodiment, the multi-task classifier comprises a first classifier, a second classifier, and a third classifier, the ESM protein language model is connected in parallel with the first classifier, the second classifier, and the third classifier, the first classifier is configured to predict the enzyme class of the protein, the second classifier is configured to predict the substrate class of the protein, and the third classifier is configured to predict the binding property of the protein.

[0083] In one of the embodiments, the model training module 20 is further configured to: input the first feature vector matrix into the first classifier, the second classifier and the third classifier respectively to obtain the predicted probability of the enzyme class, the predicted probability of the substrate class and the predicted probability of the nucleic acid binding property; calculate a first cross-entropy loss based on the predicted probability of the enzyme class and the first label; calculate a second cross-entropy loss based on the predicted probability of the substrate class and the second label; calculate a third cross-entropy loss based on the predicted probability of the nucleic acid binding property and the third label; and fine-tune the parameters of the multi-task classification model based on the average of the first cross-entropy loss, the second cross-entropy loss and the third cross-entropy loss.

[0084] In one of the embodiments, the device further comprises a data preprocessing module configured to perform data preprocessing on each of the specified protein sequences to fill to a fixed length.

[0085] It should be noted that each of the above modules can be a functional module or a program module, which can be implemented by software or hardware. For the modules implemented by hardware, each of the above modules can be located in the same processor; or each of the above modules can also be located in different processors in any combination.

[0086] In one of the embodiments, a computer device is provided, which can be a terminal. An internal structure diagram of the computer device can be as shown in Figure 5 The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved through WIFI, mobile cellular network, NFC (Near Field Communication) or other technologies. The computer program is executed by the processor to implement a protein sequence property prediction method.

[0087] Those skilled in the art can understand that Figure 5 the structure shown in the above figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0088] In one embodiment, a computer readable storage medium is provided, having stored thereon a computer program, which, when executed by a processor, implements the steps of any of the above-described embodiments of the method for predicting protein sequence properties.

[0089] A person of ordinary skill in the art can understand that all or part of the processes in the above-described embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a nonvolatile computer readable storage medium. When the computer program is executed, the processes of the above-described embodiments can be included. Any reference to memory, storage, database or other medium in the embodiments provided by the present application can include at least one of nonvolatile and volatile memory. The nonvolatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, the RAM can be static random access memory (SRAM) or dynamic random access memory (DRAM).

[0090] Any combination of the above-described technical features of the embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above-described embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0091] The above-described embodiments only express several embodiments of the present application, which are described in detail and specifically, but should not be understood as a limitation on the scope of the patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method of protein sequence property prediction, characterized by, The method comprises: building a multi-task classification model based on an ESM protein language model and a multi-task classifier; inputting a protein sequence dataset into the multi-task classification model for training, wherein the protein sequence dataset comprises a plurality of specified protein sequences and corresponding labels, the ESM protein language model is used for feature extraction of each specified protein sequence to obtain a first feature vector matrix, and the multi-task classifier is used for training the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model; wherein the multi-task classifier is used for training the multi-task classification model based on the first feature vector matrix and the corresponding labels, comprising: inputting the first feature vector matrix and the corresponding labels into the multi-task classifier for initial prediction to obtain initial prediction results of the enzyme class, substrate type and nucleic acid binding property of each specified protein, and constructing a first vector matrix based on each initial prediction result; randomly replacing part of elements in the first vector matrix with real labels to obtain a second vector matrix; obtaining an error matrix based on the second vector matrix and the first vector matrix; performing feature fusion on the error matrix and the first feature vector matrix to obtain a second feature vector matrix; inputting the second feature vector matrix into the multi-task classifier for prediction to obtain the enzyme class, substrate type and nucleic acid binding property of each specified protein. inputting a target protein sequence into the protein property prediction model for prediction to obtain the enzyme class, substrate type and nucleic acid binding property of the target protein.

2. The method of claim 1, wherein, The error matrix is obtained based on the difference between the second vector matrix and the first vector matrix. The error matrix is obtained based on the difference between the second vector matrix and the first vector matrix.

3. The method of claim 1, wherein, The error matrix is obtained based on the difference between the second vector matrix and the first vector matrix. The error matrix is obtained based on the difference between the second vector matrix and the first vector matrix. The error matrix is obtained based on the difference between the second vector matrix and the first vector matrix.

4. The method of claim 1, wherein, The multi-task classifier comprises a first classifier, a second classifier and a third classifier, and the ESM protein language model is connected in parallel with the first classifier, the second classifier and the third classifier; the first classifier is used for predicting the enzyme class of a protein, the second classifier is used for predicting the substrate class of a protein, and the third classifier is used for predicting the binding property of a protein.

5. The method of claim 4, wherein, The multi-task classifier is used for training the multi-task classification model based on the first feature vector matrix and the corresponding labels, comprising: inputting the first feature vector matrix into the first classifier, the second classifier and the third classifier respectively for prediction to obtain the prediction probability of the enzyme class, the prediction probability of the substrate class and the prediction probability of the nucleic acid binding property; a first cross-entropy loss is calculated based on the predicted probability of the enzyme class and the first label; a second cross-entropy loss is calculated based on the predicted probability of the substrate class and the second label; a third cross-entropy loss is calculated based on the predicted probability of the nucleic acid binding property and the third label; parameters of the multi-task classification model are fine-tuned based on an average value of the first cross-entropy loss, the second cross-entropy loss and the third cross-entropy loss.

6. The method of claim 1, wherein, Before the protein sequence dataset is input into the multi-task classification model for training, the method further comprises: data preprocessing is performed on each of the specified protein sequences, and the specified protein sequences are padded to a fixed length.

7. A protein sequence property prediction device, characterized by, The device comprises: a model construction module configured to construct a multi-task classification model based on an ESM protein language model and a multi-task classifier; a model training module configured to input a protein sequence dataset into the multi-task classification model for training, wherein the protein sequence dataset comprises a plurality of specified protein sequences and corresponding labels, the ESM protein language model is configured to perform feature extraction on each of the specified protein sequences to obtain a first feature vector matrix, and the multi-task classifier is configured to train the multi-task classification model based on the first feature vector matrix and the corresponding labels to obtain a protein property prediction model; wherein the multi-task classifier is configured to train the multi-task classification model based on the first feature vector matrix and the corresponding labels comprises: inputting the first feature vector matrix and the corresponding labels into the multi-task classifier for initial prediction to obtain initial prediction results of enzyme classes, substrate types and nucleic acid binding properties of each of the specified proteins, and constructing a first vector matrix based on each of the initial prediction results; randomly replacing part of elements in the first vector matrix with real labels to obtain a second vector matrix; obtaining an error matrix based on the second vector matrix and the first vector matrix; performing feature fusion on the error matrix and the first feature vector matrix to obtain a second feature vector matrix; inputting the second feature vector matrix into the multi-task classifier for prediction to obtain enzyme classes, substrate types and nucleic acid binding properties of each of the specified proteins; a model prediction module configured to input a target protein sequence into the protein property prediction model for prediction to obtain enzyme classes, substrate types and nucleic acid binding properties of the target protein.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 6. The computer program is executed by the processor to implement the method of any one of claims 1 to 6.

Images