Protein identification method, electronic device and storage medium
By extracting and fusing the protein sequences using pre-trained self-supervised models and classification models, the target probability vector is generated, which solves the problem of limited artificial feature extraction capabilities and improves the efficiency and accuracy of protein identification.
Patent Information
- Application Number
- PCT/CN2023/137165
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-22
- Filing Date
- 2023-12-07
- Publication Date
- 2024-11-28
AI Technical Summary
Existing protein identification methods rely on artificially designed complex feature extraction, making it difficult for the extracted key information to fully reflect the internal information of the protein sequence, resulting in low identification efficiency.
The protein sequence is extracted using multiple pre-trained self-supervised models, multiple embedding vectors are generated, and the target probability vector is generated through multiple classification models, and the information of multiple models is fused for prediction, avoiding the limitation of the expression ability of artificial features.
It improves the efficiency and accuracy of protein sequence identification, can more fully reflect the internal information of the protein sequence, and improves the identification efficiency.
Smart Images

Figure CN2023137165_28112024_PF_FP_ABST
Abstract
Description
Protein identification method, electronic device and storage medium Technical Field
[0001] The present application relates to the biological field, and in particular to a protein identification method, electronic device, and storage medium. Background Art
[0002] Current protein identification schemes rely on complex, artificially designed features to extract key information from protein sequences, which is then used to identify the protein sequence. However, artificially designed features often have limited representation capabilities, making it difficult for the extracted key information to fully reflect the information inherent in the protein sequence, resulting in low protein sequence identification efficiency.
[0003] Summary of the Invention
[0004] In view of the above, it is necessary to provide a protein identification method, electronic device and storage medium to solve the technical problem of low efficiency in protein sequence identification due to the limitations of the expression ability of artificially designed features.
[0005] On the one hand, the present application provides a protein identification method, which includes: obtaining a protein sequence, using multiple pre-trained self-supervised models to extract information from the protein sequence to obtain multiple embedding vectors of the protein sequence, generating a target probability vector for the protein sequence based on the multiple embedding vectors and multiple pre-trained classification models, and predicting the target probability vector based on the pre-trained prediction model to obtain a prediction result for the protein category of the protein sequence.
[0006] As can be seen from the above technical solution, the present application uses multiple self-supervised models to extract information from protein sequences. Since the multiple protein models are self-supervised models of multiple different types of network architectures, it can ensure that the multiple embedded vectors extracted are information at different levels in the protein sequence. In addition, the present application relies on multiple pre-trained self-supervised models to extract information from protein sequences, and does not rely on artificially designed features to extract information from protein sequences. Therefore, it can solve the technical problem of limited expressive power of artificially extracted features and can also improve the speed of information extraction. Since the present application inputs each embedding vector into each classification model, the output result of each embedding vector on each classification model can be obtained, and the output results of the multiple embedding vectors on the multiple classification models are spliced to obtain the target probability vector. The target probability vector incorporates the information extracted from the protein sequence by the multiple self-supervised models and the multiple classification models. Therefore, the target probability vector can fully reflect the information contained in the protein sequence. Predicting the target probability vector can improve the efficiency of protein sequence identification.
[0007] In some embodiments, the method further includes: obtaining a data set, dividing the data set to obtain a first training set and a second training set, wherein the first training set includes multiple first protein sequences and a first category label corresponding to each first protein sequence, and the second training set includes multiple second protein sequences and a second category label corresponding to each second protein sequence, and the first category label and the second category label both include preset protein categories, obtaining multiple initial classifiers corresponding to the multiple classification models and a prediction learner corresponding to the prediction model, training each initial classifier according to any one of the multiple self-supervised models, the multiple first protein sequences, and the first category label corresponding to each first protein sequence to obtain a classification model corresponding to each initial classifier, combining any self-supervised model and any classification model to obtain multiple pairs of combination models, extracting information from each second protein sequence according to each pair of combination models to obtain a training probability vector corresponding to each second protein sequence, using the prediction learner to predict each training probability vector to obtain a training prediction result for each second protein sequence, and generating the prediction model according to the training prediction result, the second category label, and the prediction learner.
[0008] In some embodiments, generating the prediction model based on the training prediction results, the second category label and the prediction learner includes: calculating the training loss value based on multiple training prediction results and the second category label corresponding to each training prediction result, adjusting the parameters of the prediction learner according to the training loss value until the training loss value meets the preset conditions, and obtaining the prediction model.
[0009] In some embodiments, calculating the training loss value based on the multiple training prediction results and the second category label corresponding to each training prediction result includes: determining whether each training prediction result is correct based on the second category label corresponding to each training prediction result, and calculating the training loss value based on the correct training prediction results and the number of labels of the second category label.
[0010] In some embodiments, generating a target probability vector for the protein sequence based on the multiple embedding vectors and the pre-trained multiple classification models includes: inputting each embedding vector into the multiple classification models to obtain multiple initial probability vectors for each embedding vector, and concatenating all the initial probability vectors of the multiple embedding vectors to obtain the target probability vector.
[0011] In some embodiments, the protein category includes preset protein categories and non-preset protein categories, and the predicting of the target probability vector based on the pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence includes: encoding the preset protein category and the non-preset protein category respectively to obtain a category vector, classifying the target probability vector based on the category vector, determining the predicted probability that the protein sequence belongs to the preset protein category, and determining the prediction result based on a comparison result of the predicted probability and a preset threshold.
[0012] In some embodiments, determining the prediction result based on the comparison result of the prediction probability and the preset threshold includes: if the prediction probability is greater than or equal to the preset threshold, determining the prediction result as that the protein sequence belongs to the preset protein category, or, if the prediction probability is less than the preset threshold, determining the prediction result as that the protein sequence does not belong to the preset protein category.
[0013] In some embodiments, the predetermined protein classes include thermophilic proteins and psychrophilic proteins.
[0014] On the other hand, the present application provides a protein identification device that runs on an electronic device, the device comprising: an acquisition unit for acquiring a protein sequence, an extraction unit for extracting information from the protein sequence using multiple pre-trained self-supervised models to obtain multiple embedding vectors of the protein sequence, a generation unit for generating a target probability vector of the protein sequence based on the multiple embedding vectors and multiple pre-trained classification models, and a prediction unit for predicting the target probability vector based on the pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence.
[0015] On the other hand, the present application provides an electronic device, comprising: a memory storing at least one instruction; and a processor executing the at least one instruction to implement the protein identification method.
[0016] On the other hand, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the protein identification method. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] FIG1 is a structural diagram of an electronic device provided in one embodiment of the present application.
[0018] FIG2 is a flow chart of a protein identification method provided in one embodiment of the present application.
[0019] FIG3 is a flowchart of a method for generating prediction results provided in an embodiment of the present application.
[0020] FIG4 is a flowchart of a method for generating multiple self-supervisory models, multiple classification models, and a prediction model provided in an embodiment of the present application.
[0021] FIG5 is a functional module diagram of a protein identification device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0022] It should be noted that in this application, "at least one" means one or more, and "more than one" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of this application are used to distinguish similar objects, rather than to describe a specific order or sequence.
[0023] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.
[0024] Thermophilic proteins, also commonly known as thermostable proteins, are a class of proteins that can maintain their structure and function under high temperature conditions. Thermophilic proteins are usually derived from thermophilic bacteria, especially archaea. Due to the feasibility of thermophilic proteins in enhancing protein stability, enhancing enzyme activity and being suitable for high-temperature industrial processes, they have been widely used in various fields such as biotechnology, food processing and pharmaceuticals. For example, thermostable cellulases from thermophiles have been used to efficiently convert lignocellulosic biomass into biofuels. However, the identification and screening of thermophilic proteins is a laborious and tedious process. In order to accelerate the development of related fields and better understand the mechanism of thermophilic proteins, it is crucial to develop high-throughput screening methods to achieve rapid identification of thermophilic proteins.
[0025] Numerous computational methods have been developed for identifying thermophilic proteins. However, these methods often rely on complex, artificially designed features to extract key information from protein sequences, which are then used to identify the protein sequences. However, artificially designed features often have limited representation capabilities, making it difficult for the extracted key information to fully reflect the information inherent in the protein sequence, resulting in low protein sequence identification efficiency.
[0026] To solve the above problems, the present application provides a protein identification method, electronic device, and storage medium. To make the purpose, technical solution, and advantages of the present application more clear, the present application is described in detail below with reference to the accompanying drawings and specific embodiments.
[0027] The protein identification method provided in the embodiments of the present application can be applied to one or more electronic devices.
[0028] FIG1 is a block diagram of an electronic device according to an embodiment of the present application. In FIG1 , the electronic device 1 includes, but is not limited to, a memory 12 , a processor 13 , and a computer program stored in the memory 12 and executable on the processor 13 , such as a protein identification program.
[0029] The electronic device 1 is a device that can automatically calculate parameter values and / or process information according to pre-set or stored instructions. Its hardware includes, but is not limited to: a microprocessor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), an embedded device, etc.
[0030] The electronic device 1 can be any electronic product that can perform human-computer interaction with a user, such as a personal computer, a server, a tablet computer, a smart phone, etc.
[0031] The electronic device 1 may further include a network device and / or a user device. The network device includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers. FIG1 is merely an example of the electronic device 1 and does not limit the electronic device 1. The electronic device 1 may include more or fewer components than shown, or a combination of certain components, or different components. For example, the electronic device 1 may also include input and output devices, network access devices, buses, etc.
[0032] The network where the electronic device 1 is located includes but is not limited to: the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (VPN), etc.
[0033] FIG2 is a flowchart of a protein identification method according to an embodiment of the present application. The order of the steps in the flowchart may be adjusted based on different needs, and some steps may be omitted. The method is performed by an electronic device, such as the electronic device 1 shown in FIG1 .
[0034] S11, obtain protein sequence.
[0035] In one embodiment, the protein sequence is a protein sequence whose category needs to be identified. The electronic device can obtain the protein sequence from a dataset. For example, the dataset can be part of or all of the data in the Uniprot database, the BRENDA database, and other common databases containing protein sequences.
[0036] In one embodiment, the protein sequence may be a sequence in a dataset whose category needs to be identified.
[0037] In another embodiment, the protein sequence can be obtained by other means, and this application does not limit the protein sequence.
[0038] S12, using multiple pre-trained self-supervised models to extract information from the protein sequence to obtain multiple embedding vectors of the protein sequence.
[0039] In one embodiment, the plurality of self-supervised models include, but are not limited to, a SeqVec model, a ProtCNN model, a ProtTrans model, and a CPCProt model, wherein the plurality of self-supervised models are pre-trained to convergence models.
[0040] In another embodiment, the multiple self-supervisory models may also be other models, and this application does not limit the multiple self-supervisory models.
[0041] In some embodiments, the electronic device uses multiple pre-trained self-supervised models to extract information from the protein sequence to obtain multiple embedding vectors of the protein sequence, including: the electronic device inputs each embedding vector into each self-supervised model to obtain the embedding vector corresponding to the output of each embedding vector on each self-supervised model.
[0042] The information extraction process of each protein model for the protein sequence may refer to the process of the model performing representation learning on the protein sequence to obtain the protein embedded representation in the related art.
[0043] In this embodiment, information such as the dimension and amount of information of each embedding vector is related to the self-supervised model corresponding to the embedding vector. For example, if the multiple protein models are the SeqVec model, the ProtCNN model, the ProtTrans model, and the CPCProt model, and the multiple embedding vectors are four vectors, the dimension of the embedding vector extracted by the SeqVec model is 1024, the dimension of the embedding vector extracted by the ProtCNN model is 1100, the dimension of the embedding vector extracted by the ProtTrans model is 1024, and the dimension of the embedding vector extracted by the CPCProt model is 1536.
[0044] In this embodiment, multiple self-supervised models are used to extract features from the protein sequence, rather than relying on manually extracted features to extract information from the protein sequence. This overcomes the technical issue of the limited expressive power of manually extracted features and improves the speed of information extraction. Because each self-supervised model has a different architecture, it can extract embedding vectors at different levels and dimensions within the protein sequence.
[0045] In other embodiments of the present application, in order to enable the protein sequence to have more embedded representations, the number of the multiple self-supervisory models may be greater.
[0046] S13: Generate a target probability vector for the protein sequence based on the multiple embedding vectors and the multiple pre-trained classification models.
[0047] In some embodiments, the multiple classification models include, but are not limited to: Random Forest model (RF), Adaptive Boosting model (AdaBoost, AB), Guided Aggregation model or Bagging model (BA), Gradient Boosting Decision Tree Model XGBoost (eXtremeGradient Boosting, XGB) and (LightGBM, LGB), etc.
[0048] In one embodiment, the electronic device generates a target probability vector for the protein sequence based on the multiple embedding vectors and the pre-trained multiple classification models, including: the electronic device inputs each embedding vector into the multiple classification models to obtain multiple initial probability vectors for each embedding vector, and then the electronic device splices all the initial probability vectors of the multiple embedding vectors to obtain the target probability vector.
[0049] Among them, since the multiple classification models are different models, the process of each classification model generating each embedding vector according to each input embedding vector can refer to the classification method or classification formula of each classification model in the relevant technology.
[0050] For example, continuing with the above embodiment, if the multiple embedding vectors are 4 embedding vectors output by the SeqVec model, the ProtCNN model, the ProtTrans model and the CPCProt model, and the multiple classification models are the AdaBoost model, the Bagging model, the XGBoost model and the LightGBM model, the 4 embedding vectors are respectively input into 5 classification models to obtain 20 initial probability vectors. If the dimension of each initial probability vector is 1 dimension, the 20 1-dimensional initial probability vectors are spliced to obtain a target probability vector with a dimension of 20 dimensions.
[0051] In another embodiment of the present application, the multiple classification models may be other models, which are not limited in the present application. The above examples of the multiple classification models are merely examples and do not constitute a limitation on the multiple classification models.
[0052] In this embodiment, the multiple classification models can be cascaded, and each embedding vector can be classified using a model integrated from the multiple classification models to obtain the target probability vector, thereby reducing the overfitting problem of a single model. Furthermore, because the target probability vector is obtained by concatenating multiple initial probability vectors, it incorporates the information from these multiple initial probability vectors, allowing it to fully reflect the information inherent in the protein sequence.
[0053] S14, predicting the target probability vector according to the pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence.
[0054] In one embodiment, the protein categories include preset protein categories and non-preset protein categories. The preset protein categories include, but are not limited to, thermophilic proteins and psychrophilic proteins. The non-preset protein categories are protein categories other than the preset protein categories. For example, if the preset protein category is thermophilic proteins, the non-preset protein category may be non-thermophilic proteins. Alternatively, if the preset protein category is psychrophilic proteins, the non-preset protein category may be non-psychrophilic proteins.
[0055] In some embodiments, the prediction result includes that the protein sequence belongs to the preset protein category and that the protein sequence does not belong to the preset protein category.
[0056] In some embodiments, the prediction model may be a machine learning model, for example, a support vector machine (SVM) model.
[0057] In this embodiment, since the target probability vector can fully reflect the information contained in the protein sequence, predicting the target probability vector can improve the accuracy of protein sequence identification.
[0058] As can be seen from the above technical solution, the present application uses multiple self-supervised models to extract information from protein sequences. Since the multiple protein models are self-supervised models of multiple different types of network architectures, it can ensure that the multiple embedded vectors extracted are information at different levels in the protein sequence. In addition, the present application relies on multiple pre-trained self-supervised models to extract information from protein sequences, and does not rely on artificially designed features to extract information from protein sequences. Therefore, it can solve the technical problem of limited expressive power of artificially extracted features and can also improve the speed of information extraction. Since the present application inputs each embedding vector into each classification model, the output result of each embedding vector on each classification model can be obtained, and the output results of the multiple embedding vectors on the multiple classification models are spliced to obtain the target probability vector. The target probability vector incorporates the information extracted from the protein sequence by the multiple self-supervised models and the multiple classification models. Therefore, the target probability vector can fully reflect the information contained in the protein sequence. Predicting the target probability vector can improve the efficiency of protein sequence identification.
[0059] In some embodiments of the present application, the electronic device inputs the target probability vector into the prediction model to obtain a prediction result for the protein class of the protein sequence. FIG3 is a flowchart of a method for generating prediction results provided in one embodiment of the present application, comprising the following steps:
[0060] S141 : The electronic device encodes the preset protein category and the non-preset protein category respectively to obtain a category vector.
[0061] In some embodiments, the predetermined protein category includes, but is not limited to, thermophilic proteins and psychrophilic proteins. The non-preset protein category is a category other than the predetermined protein category. For example, if the predetermined protein category is thermophilic proteins, the non-preset protein category may be non-thermophilic proteins. Alternatively, if the predetermined protein category is psychrophilic proteins, the non-preset protein category may be non-psychrophilic proteins.
[0062] In some embodiments, the electronic device may perform one-hot encoding on the preset protein category and the non-preset protein category to obtain the category vector. For example, the electronic device may encode the preset protein category as 1 and encode the non-preset protein category as 0, thereby obtaining a category vector consisting of 1 and 0.
[0063] In other embodiments of the present application, the electronic device may encode the preset protein category and the non-preset protein category in other ways, and the present application does not limit the encoding method.
[0064] S142: The electronic device classifies the target probability vector based on the category vector to determine a predicted probability that the protein sequence belongs to the preset protein category.
[0065] In some embodiments, the electronic device classifies the target probability vector based on the category vector, and the method for determining the predicted probability that the protein sequence belongs to the preset protein category can refer to the classification formula corresponding to the sigmoid function or softmax function in the relevant technology.
[0066] S143: The electronic device determines the prediction result according to a comparison result of the prediction probability and a preset threshold.
[0067] In some embodiments, the electronic device determines the prediction result based on the comparison result of the prediction probability and the preset threshold, including: if the prediction probability is greater than or equal to the preset threshold, the electronic device determines the prediction result as the protein sequence belongs to the preset protein category, or, if the prediction probability is less than the preset threshold, the electronic device determines the prediction result as the protein sequence does not belong to the preset protein category, or the protein sequence belongs to the non-preset protein category.
[0068] In one embodiment of the present application, if the electronic device determines that the prediction result is that the protein sequence belongs to the preset protein category, the electronic device outputs first preset data to indicate that the protein sequence belongs to the preset protein category, or, if the electronic device determines that the prediction result is that the protein sequence does not belong to the preset protein category (the electronic device determines that the prediction result is that the protein sequence belongs to the non-preset protein category), the electronic device outputs second preset data to indicate that the protein sequence does not belong to the preset protein category.
[0069] The first preset data and the second preset data can be in the form of a numerical value, a letter, or a combination of a numerical value and a letter. The first preset data and the second preset data can be set arbitrarily, and this application does not impose any restrictions on this. For example, when the first preset data and the second preset data are both numerical values, the first preset value can be 1, and the second preset value can be 0.
[0070] In some embodiments of the present application, before using the multiple self-supervised models, the multiple classification models, and the prediction model, it is necessary to generate the multiple self-supervised models, the multiple classification models, and the prediction model. As shown in Figure 4, it is a flowchart of a method for generating multiple self-supervised models, multiple classification models, and prediction models provided in an embodiment of the present application, comprising the following steps:
[0071] S21, the electronic device obtains a data set, and divides the data set to obtain a first training set and a second training set.
[0072] In some embodiments, the dataset includes multiple protein sequences, each protein sequence in the dataset has a corresponding category label, and the dataset may include, but is not limited to, partial or all data from the Uniprot database, the BRENDA database, and common databases containing protein sequences. The first training set includes multiple first protein sequences and a first category label corresponding to each first protein sequence, and the second training set includes multiple second protein sequences and a second category label corresponding to each second protein sequence, wherein both the first category label and the second category label include preset protein categories.
[0073] In some embodiments, each first protein sequence has a corresponding first category label, and each second protein sequence has a corresponding second category label, wherein the first category label is used to indicate the category of each first protein sequence, and the second category label is used to indicate the category of each second protein sequence.
[0074] In this embodiment, the electronic device can set its own method for dividing the data set, and this application does not limit this. For example, the electronic device can use 80% of the data in the data set as the first training set, and use the category label of the 80% of the data as the first category label. Then, the electronic device uses 20% of the data in the data set as the second training set, and uses the category label of the 20% of the data as the second category label. The above example is only an example of a method for dividing the data set, and does not constitute a limitation on the method for dividing the data set.
[0075] S22: The electronic device obtains a plurality of initial classifiers corresponding to the plurality of classification models and a prediction learner corresponding to the prediction model.
[0076] In some embodiments, the multiple initial classifiers may include a classifier corresponding to the random forest model Random Forest, a classifier corresponding to the adaptive boosting model AdaBoost, a classifier corresponding to the bagging model Bagging, a classifier corresponding to the gradient boosting decision tree model XGBoost, and a classifier corresponding to the gradient boosting decision tree model LightGBM, and the prediction learner may be a learner of the support vector machine model SVM.
[0077] S23, the electronic device trains each initial classifier according to any one of the multiple self-supervised models, the multiple first protein sequences, and the first category label corresponding to each first protein sequence to obtain a classification model corresponding to each initial classifier.
[0078] In some embodiments, the electronic device inputs each first protein sequence into each self-supervised model to obtain an initial embedding vector corresponding to the output of each first protein sequence on each self-supervised model, wherein the multiple initial embedding vectors corresponding to each first protein sequence correspond to the first category label of the first protein sequence. The electronic device then inputs the multiple initial embedding vectors corresponding to each first protein sequence into the multiple initial classifiers respectively to obtain a classification vector corresponding to the output of each initial embedding vector of each first protein sequence on each initial classifier, splices the multiple classification vectors corresponding to each first protein sequence to obtain a splicing probability vector corresponding to each first protein sequence, and predicts the splicing probability vector corresponding to each first protein sequence to obtain a predicted category of each first protein sequence. The electronic device calculates an initial loss value based on the predicted categories of the multiple first proteins and the first category labels, and adjusts the parameters of the multiple initial classifiers based on the initial loss value until the initial loss value meets the configuration conditions, thereby obtaining a classification model corresponding to the initial classifier.
[0079] In this embodiment, the multiple initial embedding vectors are generated in a manner substantially identical to the multiple embedding vectors described above, and the splicing probability vector is generated in a manner substantially identical to the target probability vector described above, so this application will not repeat the description. Furthermore, the electronic device's prediction of each splicing probability vector may refer to the prediction process for the target probability vector described below, and the calculation method for the initial loss value may refer to the calculation method for the training loss value described below. The configuration conditions are substantially identical to the preset conditions described below.
[0080] S24, the electronic device combines any self-supervision model and any classification model to obtain multiple pairs of combination models.
[0081] For example, if the multiple self-supervisory models are represented by A, B, C, and D, respectively, and the multiple classification models are represented by a, b, c, d, and e, respectively, the electronic device combines any self-supervisory model with any classification model to obtain multiple pairs of combined models as shown in Table 1.
[0082] Table 1 Examples of multiple pairwise combination models
[0083] S25 , the electronic device extracts information from each second protein sequence according to each pair of combination models to obtain a training probability vector corresponding to each second protein sequence.
[0084] In some embodiments, the generation method of each training probability vector is substantially the same as the generation method of the target probability vector, and thus will not be described again in this application.
[0085] S26, the electronic device uses the prediction learner to predict each training probability vector to obtain a training prediction result for each second protein sequence.
[0086] In some embodiments, the training prediction results and the protein sequence prediction results are generated in substantially the same manner, and thus are not described again in this application.
[0087] S27, the electronic device generates the prediction model according to the training prediction result, the second category label and the prediction learner.
[0088] In some embodiments, the electronic device generates the prediction model based on the training prediction results, the second category label and the prediction learner, including: the electronic device calculates the training loss value based on multiple training prediction results and the second category label corresponding to each training prediction result, adjusts the parameters of the prediction learner according to the training loss value until the training loss value meets the preset conditions, and obtains the prediction model.
[0089] The electronic device can adjust parameters such as weights and biases of the multiple self-supervised models, the multiple initial classifiers, and the predictive learner. The preset conditions can be set voluntarily, and this application does not impose any restrictions on this. For example, the preset conditions correspond to the training prediction results. For example, the preset conditions can be that the training loss value drops to a preset interval, the training loss value drops to a minimum, or the training loss value no longer changes, wherein the preset interval can be set voluntarily, and this application does not impose any restrictions on this.
[0090] In some embodiments, the electronic device calculates the training loss value based on the multiple training prediction results and the second category label corresponding to each training prediction result, including: the electronic device determines whether each training prediction result is correct based on the second category label corresponding to each training prediction result, and then the electronic device calculates the training loss value based on the correct training prediction results and the number of labels of the second category label.
[0091] In which, the electronic device can compare each training prediction result with the corresponding second category label. If the protein indicated by each training prediction result is the same as the corresponding second category label, the electronic device determines that each training prediction result is correct; or, if each training prediction result is different from the corresponding second category label, the electronic device determines that each training prediction result is wrong.
[0092] For example, when the training prediction result of any second protein sequence is that the second protein sequence belongs to the preset protein category, if the second category label of the second protein sequence is the preset protein category, the electronic device determines that the training prediction result of the second protein sequence is correct. Alternatively, when the training prediction result of any second protein sequence is that the second protein sequence belongs to the preset protein category, if the second category label of the second protein sequence is not the preset protein category, the electronic device determines that the training prediction result of the second protein sequence is incorrect.
[0093] In one embodiment, the electronic device uses an integration algorithm (such as a bagging algorithm or a boosting algorithm) to perform integration training on the multiple initial classifiers to obtain a model formed by cascading the multiple classification models.
[0094] In this embodiment, the electronic device calculates the number of correct training prediction results and determines the ratio between the correct number and the number of labels as the training loss value. When the training loss value no longer changes, the electronic device stops adjusting to obtain the multiple self-supervised models, the multiple classification models, and the prediction model.
[0095] In other embodiments of the present application, the training loss value may also be calculated in other ways. For example, the electronic device may determine the ratio between the number of errors in the erroneous training prediction results and the number of labels as the training loss value. When the training loss value no longer decreases or decreases to a preset range, the electronic device stops adjusting to obtain the multiple self-supervised models, the multiple classification models, and the prediction model.
[0096] Figure 5 shows a functional block diagram of a protein identification device according to one embodiment of the present application. The protein identification device 11 includes an acquisition unit 110, an extraction unit 111, a generation unit 112, and a prediction unit 113. A module / unit as referred to herein refers to a series of computer-readable instruction segments capable of being acquired by the processor 13 in Figure 1 and performing a fixed function, stored in the memory 12 in Figure 1. The functions of each module / unit in this embodiment will be described in detail in subsequent embodiments.
[0097] In some embodiments, the acquisition unit 110 is used to acquire protein sequences.
[0098] The extraction unit 111 is configured to extract information from the protein sequence using a plurality of pre-trained self-supervised models to obtain a plurality of embedding vectors of the protein sequence.
[0099] The generating unit 112 is configured to generate a target probability vector for the protein sequence based on the multiple embedding vectors and the multiple pre-trained classification models.
[0100] In some embodiments, the generation unit 112 is further used to input each embedding vector into the multiple classification models to obtain multiple initial probability vectors for each embedding vector, and concatenate all the initial probability vectors of the multiple embedding vectors to obtain the target probability vector.
[0101] The prediction unit 113 is configured to predict the target probability vector according to a pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence. The preset protein categories include thermophilic proteins and psychrophilic proteins.
[0102] In some embodiments, the protein category includes a preset protein category and a non-preset protein category, and the prediction unit 113 is further used to predict the target probability vector based on a pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence, including: encoding the preset protein category and the non-preset protein category respectively to obtain a category vector, classifying the target probability vector based on the category vector, determining the predicted probability that the protein sequence belongs to the preset protein category, and determining the prediction result based on the comparison result of the predicted probability and a preset threshold.
[0103] In some embodiments, the prediction unit 113 is further used to determine the prediction result based on the comparison result of the prediction probability and the preset threshold, including: if the prediction probability is greater than or equal to the preset threshold, determining that the prediction result is that the protein sequence belongs to the preset protein category, or if the prediction probability is less than the preset threshold, determining that the prediction result is that the protein sequence does not belong to the preset protein category.
[0104] In some embodiments, the generation unit 112 is further used to obtain a data set, divide the data set to obtain a first training set and a second training set, wherein the first training set includes multiple first protein sequences and a first category label corresponding to each first protein sequence, and the second training set includes multiple second protein sequences and a second category label corresponding to each second protein sequence, and the first category label and the second category label both include preset protein categories, obtain multiple initial classifiers corresponding to the multiple classification models and a prediction learner corresponding to the prediction model, train each initial classifier according to any one of the multiple self-supervised models, the multiple first protein sequences, and the first category label corresponding to each first protein sequence to obtain a classification model corresponding to each initial classifier, combine any self-supervised model and any classification model to obtain multiple pairs of combination models, extract information from each second protein sequence according to each pair of combination models to obtain a training probability vector corresponding to each second protein sequence, use the prediction learner to predict each training probability vector to obtain a training prediction result for each second protein sequence, and generate the prediction model according to the training prediction result, the second category label, and the prediction learner.
[0105] In some embodiments, the generation unit 112 is further used to generate the prediction model based on the training prediction results, the second category label and the prediction learner, including: calculating the training loss value based on multiple training prediction results and the second category label corresponding to each training prediction result, adjusting the parameters of the prediction learner according to the training loss value until the training loss value meets the preset conditions, and obtaining the prediction model.
[0106] In some embodiments, the generation unit 112 is further used to calculate the training loss value based on the multiple training prediction results and the second category label corresponding to each training prediction result, including: determining whether each training prediction result is correct based on the second category label corresponding to each training prediction result, and calculating the training loss value based on the correct training prediction results and the number of labels of the second category label.
[0107] As can be seen from the above technical solution, the present application uses multiple self-supervised models to extract information from protein sequences. Since the multiple protein models are self-supervised models of multiple different types of network architectures, it can ensure that the multiple embedded vectors extracted are information at different levels in the protein sequence. In addition, the present application relies on multiple pre-trained self-supervised models to extract information from protein sequences, and does not rely on artificially designed features to extract information from protein sequences. Therefore, it can solve the technical problem of limited expressive power of artificially extracted features and can also improve the speed of information extraction. Since the present application inputs each embedding vector into each classification model, the output result of each embedding vector on each classification model can be obtained, and the output results of the multiple embedding vectors on the multiple classification models are spliced to obtain the target probability vector. The target probability vector incorporates the information extracted from the protein sequence by the multiple self-supervised models and the multiple classification models. Therefore, the target probability vector can fully reflect the information contained in the protein sequence. Predicting the target probability vector can improve the efficiency of protein sequence identification.
[0108] In one embodiment, continuing with the previous description of the electronic device in FIG. 1 , the processor 13 may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor 13 is the computing core and control center of the electronic device 1, connecting various parts of the entire electronic device 1 using various interfaces and lines, and obtaining the operating system of the electronic device 1 and various installed applications, program codes, etc.
[0109] The processor 13 obtains the operating system and various installed applications of the electronic device 1. The processor 13 obtains the applications to implement the steps in the above-mentioned various protein identification method embodiments, such as the steps shown in Figures 2, 3 and 4.
[0110] For example, the computer program can be divided into one or more modules / units, such as an acceleration unit. One or more modules / units are stored in the memory 12 and retrieved by the processor 13 to complete the present application. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions. The instruction segments are used to describe the acquisition process of the computer program in the electronic device 1.
[0111] The memory 12 can be used to store computer programs and / or modules. The processor 13 implements various functions of the electronic device 1 by running or accessing the computer programs and / or modules stored in the memory 12, as well as accessing data stored in the memory 12. The memory 12 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area may store data created based on the use of the server. In addition, the memory 12 may include non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device.
[0112] The memory 12 may be an external memory and / or an internal memory of the electronic device 1. Furthermore, the memory 12 may be a physical memory, such as a memory stick, a TF card (Trans-flash Card), and the like.
[0113] If the modules / units integrated in the electronic device 1 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is obtained by the processor, it can implement the steps of the above-mentioned various method embodiments.
[0114] The computer program includes computer program code, which may be in source code form, object code form, an accessible file, or some intermediate form. Computer-readable media may include any entity or device capable of carrying computer program code, recording media, USB flash drives, removable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).
[0115] 2 , the memory 12 in the electronic device 1 stores a plurality of instructions to implement a protein identification method, and the processor 13 can obtain the plurality of instructions to implement: obtaining a protein sequence; extracting information from the protein sequence using a plurality of pre-trained self-supervised models to obtain a plurality of embedding vectors of the protein sequence; generating a target probability vector of the protein sequence based on the plurality of embedding vectors and a plurality of pre-trained classification models; and predicting the target probability vector based on the pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence.
[0116] Specifically, the specific implementation method of the processor 13 for the above instructions can refer to the description of the relevant steps in the corresponding embodiment of Figure 2, which will not be repeated here.
[0117] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules is merely a logical function division, and other division methods may be used in actual implementation.
[0118] Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of these modules may be selected based on actual needs to achieve the purpose of this embodiment.
[0119] In addition, the functional modules in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional modules.
[0120] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the present application is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.
[0121] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in this application may also be implemented by a single unit or device through software or hardware. Terms such as first and second are used to indicate names and do not imply any particular order.
[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.
Claims
1. A protein identification method, characterized in that: The method comprises: Obtain protein sequences; Using a plurality of pre-trained self-supervised models to extract information from the protein sequence to obtain a plurality of embedding vectors of the protein sequence; Generating a target probability vector for the protein sequence according to the multiple embedding vectors and the multiple pre-trained classification models; The target probability vector is predicted according to the pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence.
2. The protein identification method according to claim 1, characterized in that The method further comprises: Acquire a data set, divide the data set into a first training set and a second training set, wherein the first training set includes a plurality of first protein sequences and a first category label corresponding to each first protein sequence, and the second training set includes a plurality of second protein sequences and a second category label corresponding to each second protein sequence, and the first category label and the second category label both include a preset protein category; Obtaining a plurality of initial classifiers corresponding to the plurality of classification models and a prediction learner corresponding to the prediction model; Training each initial classifier according to any one of the multiple self-supervised models, the multiple first protein sequences, and a first category label corresponding to each first protein sequence to obtain a classification model corresponding to each initial classifier; Combine any self-supervisory model and any classification model to obtain multiple pairs of combined models; Extract information from each second protein sequence according to each pair of combined models to obtain a training probability vector corresponding to each second protein sequence; Using the prediction learner to predict each training probability vector to obtain a training prediction result for each second protein sequence; The prediction model is generated according to the training prediction result, the second category label and the prediction learner.
3. The protein identification method according to claim 2, characterized in that: Generating the prediction model according to the training prediction result, the second category label and the prediction learner includes: Calculate a training loss value according to the plurality of training prediction results and a second category label corresponding to each training prediction result; The prediction model is obtained by adjusting the parameters of the prediction learner according to the training loss value until the training loss value meets a preset condition.
4. The protein identification method according to claim 2, characterized in that: Calculating the training loss value according to the plurality of training prediction results and the second category label corresponding to each training prediction result includes: Determine whether each training prediction result is correct according to the second category label corresponding to each training prediction result; The training loss value is calculated according to the correct training prediction result and the number of labels of the second category label.
5. The protein identification method according to claim 1, characterized in that: Generating the target probability vector of the protein sequence according to the multiple embedding vectors and the pre-trained multiple classification models includes: Input each embedding vector into the multiple classification models to obtain multiple initial probability vectors for each embedding vector; All initial probability vectors of the multiple embedding vectors are concatenated to obtain the target probability vector.
6. The protein identification method according to claim 1, characterized in that: The protein category includes a preset protein category and a non-preset protein category, and the prediction of the target probability vector according to the pre-trained prediction model to obtain a prediction result of the protein category of the protein sequence includes: Encoding the preset protein category and the non-preset protein category respectively to obtain a category vector; Classifying the target probability vector based on the category vector to determine the predicted probability that the protein sequence belongs to the preset protein category; The prediction result is determined according to a comparison result between the prediction probability and a preset threshold.
7. The protein identification method according to claim 6, characterized in that: Determining the prediction result according to the comparison result between the prediction probability and the preset threshold value includes: If the prediction probability is greater than or equal to a preset threshold, determining that the prediction result is that the protein sequence belongs to the preset protein category; or If the prediction probability is less than the preset threshold, it is determined that the prediction result is that the protein sequence does not belong to the preset protein category.
8. The protein identification method according to claim 6, characterized in that: The preset protein categories include thermophilic proteins and psychrophilic proteins.
9. An electronic device, characterized in that: The electronic device comprises: a memory storing at least one instruction; and A processor executes the at least one instruction to implement the protein identification method according to any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the protein identification method according to any one of claims 1 to 8.
Citation Information
Patent Citations
Training method and device of self-supervised learning model, equipment and storage medium
CN112420123A
Protein expression model pre-training and protein interaction prediction method and device
CN114333982A
Drug target general prediction method and device based on self-supervised learning, and medium
CN116013428A
Protein identification method, electronic equipment and storage medium
CN116646009A
Protein database search using learned representations
US20220165356A1
Cited By
Protein function annotation method and device based on pre-trained large language model
CN119479836A
