WO2024254739A1 - 蛋白质功能注释的方法及装置、电子设备和存储介质 - Google Patents

蛋白质功能注释的方法及装置、电子设备和存储介质 Download PDF

Info

Publication number
WO2024254739A1
WO2024254739A1 PCT/CN2023/099729 CN2023099729W WO2024254739A1 WO 2024254739 A1 WO2024254739 A1 WO 2024254739A1 CN 2023099729 W CN2023099729 W CN 2023099729W WO 2024254739 A1 WO2024254739 A1 WO 2024254739A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature
sequence
features
training
learning
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2023/099729
Other languages
English (en)
French (fr)
Inventor
苏子栋
赵子裕
郭斐
黎宇翔
董宇亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
BGI Shenzhen Co Ltd
Original Assignee
BGI Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by BGI Shenzhen Co Ltd filed Critical BGI Shenzhen Co Ltd
Priority to CN202380094803.5A priority Critical patent/CN120731460A/zh
Priority to PCT/CN2023/099729 priority patent/WO2024254739A1/zh
Publication of WO2024254739A1 publication Critical patent/WO2024254739A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B15/00—ICT specially adapted for analysing two-dimensional [2D] or three-dimensional [3D] molecular structures, e.g. structural or functional relations or structure alignment
    • G16B15/20—Protein or domain folding

Definitions

  • the present disclosure relates to the field of data processing technology, and in particular to a method and device for protein function annotation, an electronic device and a storage medium.
  • protein function annotation is to use algorithms to predict the possible functions of a protein.
  • protein function annotation algorithms are mainly classified into three categories, including algorithms based only on sequence information, algorithms based on sequence and structure information, and algorithms based on sequence and interaction networks.
  • the present disclosure provides a method and device for protein function annotation, an electronic device and a storage medium, the main purpose of which is to solve the problem of low prediction accuracy of protein function annotation algorithms based only on sequence information.
  • a method for protein function annotation comprising:
  • sequence features According to the protein sequence, sequence features, pre-trained sequence features and preset annotation features are obtained respectively;
  • the first fusion feature is processed to obtain protein function annotation information.
  • the method further includes:
  • the third learning feature, the fourth learning feature and the fifth learning feature are fused to obtain a second fused feature; wherein the fusion process is an addition process;
  • the second fusion features are classified and calculated to obtain protein function annotation information.
  • the obtaining of sequence features, pre-trained sequence features and preset annotation features according to the protein sequence includes:
  • the preset annotation feature corresponding to the protein sequence is searched.
  • performing feature extraction processing on the preset annotation feature to obtain the first learning feature comprises the following steps:
  • the output features of the first residual fully connected module are used as the input features of the second residual fully connected module for feature extraction
  • the above steps are looped for a first predetermined number of times to obtain the first learning feature, wherein the first predetermined number of times is equal to the first preset number.
  • the first residual fully connected module inputting the preset annotation feature into the residual fully connected module includes:
  • the preset annotation feature and the second preset annotation feature are added together to obtain a first output feature.
  • the sequence feature is processed to obtain a sixth learning feature, and the pre-trained sequence feature is processed to obtain a seventh learning feature;
  • the performing feature extraction processing on the sequence feature and the pre-trained sequence feature to obtain the second learning feature comprises the following steps:
  • the above steps are looped for a second predetermined number of times to obtain the second learning feature, wherein the second predetermined number of times is equal to the second preset number.
  • the step of inputting the third fusion feature into the first residual convolution module in the residual convolution module includes:
  • the third fusion feature and the third output feature are added together to obtain a fourth output feature.
  • the fusing the first learning feature with the second learning feature to obtain a first fused feature includes:
  • the first learning feature and the second learning feature are concatenated to obtain a first fusion feature.
  • the processing of the first fusion feature to obtain protein function annotation information includes:
  • the protein fusion features are classified and processed to obtain the protein function annotation information.
  • the step of inputting the first fusion feature into an integration module for integration processing to obtain a protein fusion feature includes:
  • the first fusion feature and the second protein fusion feature are added together to obtain the protein fusion feature.
  • the method before respectively acquiring sequence features, pre-trained sequence features, and preset annotation features according to the protein sequence, the method further includes:
  • the protein function annotation model is trained based on the training protein sequences.
  • inputting the training protein sequence into the protein function annotation model for training includes:
  • the training protein sequence respectively obtaining a training sequence feature, a training first sequence feature and a training preset annotation feature
  • the first training fusion feature is processed to obtain training protein function annotation information.
  • the method further comprises:
  • feature extraction processing is performed on the training sequence features and the training first sequence features respectively to obtain fourth training learning features and fifth training learning features;
  • the second training fusion features are classified and calculated to obtain training protein function annotation information.
  • a device for protein function annotation comprising:
  • An acquisition unit used for acquiring sequence features, pre-trained sequence features and preset annotation features according to the protein sequence
  • a first processing unit configured to perform feature extraction processing on the preset annotation feature to obtain a first learning feature
  • a second processing unit is used to perform feature extraction processing on the sequence feature and the pre-trained sequence feature to obtain a second learning feature
  • a first fusion unit configured to fuse the first learning feature with the second learning feature to obtain a first fused feature
  • the third processing unit is used to process the first fusion feature to obtain protein function annotation information.
  • the device further comprises:
  • a fourth processing unit configured to, after acquiring sequence features, pre-trained sequence features, and preset annotation features according to the protein sequence, respectively, perform feature extraction processing on the preset annotation features based on a fully connected layer with a first preset number of layers to obtain third learning features;
  • a fifth processing unit configured to perform feature extraction processing on the sequence feature and the pre-trained sequence feature based on a convolution layer with a second preset number of layers, respectively, to obtain a fourth learning feature and a fifth learning feature;
  • a second fusion unit is used to fuse the third learning feature, the fourth learning feature and the fifth learning feature to obtain a second fused feature; wherein the fusion process is an addition process;
  • the sixth processing unit is used to perform classification calculation on the second fusion feature based on a fully connected layer with a third preset number of layers to obtain protein function annotation information.
  • the acquiring unit includes:
  • An encoding module used for encoding the protein sequence based on a preset encoding algorithm to obtain the sequence feature
  • a pre-training module used for performing pre-training processing on the sequence features based on a preset protein language model to obtain the pre-trained sequence features
  • the search module is used to search for the preset annotation feature corresponding to the protein sequence in a preset protein database.
  • the first processing unit includes:
  • a first input module used for inputting the preset annotation features into a first preset number of residual fully connected modules
  • a first processing module used for performing feature extraction processing using the output features of the first residual fully connected module as input features of the second residual fully connected module
  • a first loop module is used to loop and execute a first predetermined number of times according to the processing method of the first processing module to obtain the first learning feature, wherein the first predetermined number of times is equal to the first preset number.
  • the first input module is further used for:
  • the preset annotation feature and the second preset annotation feature are added together to obtain a first output feature.
  • the second processing unit includes:
  • a second processing module is used to process the sequence feature based on a fourth preset number of convolutional layers to obtain a sixth learning feature, and to process the pre-trained sequence feature to obtain a seventh learning feature;
  • a first adding module used for adding the sixth learning feature and the seventh learning feature to obtain a third fusion feature
  • a second input module used for inputting the third fusion feature into a second preset number of residual convolution modules
  • a first extraction module used for performing feature extraction processing using the output features of the first residual convolution module as input features of the second residual convolution module
  • the second loop module is used to loop and execute the processing method of the first extraction module a second predetermined number of times to obtain the second learning feature, wherein the second predetermined number of times is equal to the second preset number.
  • the second input module is further used for:
  • the third fusion feature and the third output feature are added together to obtain a fourth output feature.
  • the first fusion unit is further used for:
  • the first learning feature and the second learning feature are concatenated to obtain a first fusion feature.
  • the third processing unit includes:
  • An integration module used for inputting the first fusion feature into the integration module for integration processing to obtain a protein fusion feature
  • the classification module is used to classify the protein fusion features based on a fifth preset number of fully connected layers to obtain the protein function annotation information.
  • the integration module is further used for:
  • the first fusion feature and the second protein fusion feature are added together to obtain the protein fusion feature.
  • the device further comprises:
  • the training unit is used to train the protein function annotation model based on the training protein sequence before respectively obtaining the sequence features, the pre-trained sequence features and the preset annotation features according to the protein sequence.
  • the training unit includes:
  • a training acquisition module used to respectively acquire training sequence features, training first sequence features and training preset annotation features according to the training protein sequence
  • a first processing module for training used for performing feature extraction processing on the preset annotation feature for training to obtain a first learning feature for training
  • a second training processing module used for performing feature extraction processing on the training sequence features and the first training sequence features to obtain second training learning features
  • a first training fusion module used for fusing the first training learning feature with the second training learning feature to obtain a first training fusion feature
  • the third processing module for training is used to process the first fusion feature for training to obtain protein function annotation information for training.
  • the training unit further includes:
  • a fourth processing module for training configured to, after obtaining a training sequence feature, a training first sequence feature, and a training preset annotation feature respectively according to a training protein sequence, perform feature extraction processing on the training preset annotation feature based on a first preset number of training fully connected layers to obtain a third training learning feature;
  • a fifth training processing module configured to perform feature extraction processing on the training sequence features and the first training sequence features based on the second preset number of training convolutional layers, respectively, to obtain fourth training learning features and fifth training learning features;
  • a second training fusion module used for fusing the third training learning feature, the fourth training learning feature and the fifth training learning feature to obtain a second training fusion feature; wherein the fusion process is an addition process;
  • the sixth processing module for training is used to perform classification calculation on the second training fusion features based on the third preset number of training fully connected layers to obtain training protein function annotation information.
  • an electronic device including:
  • the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.
  • a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the method described in the first aspect.
  • a computer program product comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method as described in the first aspect above.
  • the present invention provides a method and device for protein function annotation, an electronic device and a storage medium, which respectively obtain sequence features, pre-trained sequence features and preset annotation features according to a protein sequence; perform feature extraction processing on the preset annotation features based on a first preset number of residual fully connected modules to obtain a first learning feature; Based on a second preset number of residual convolution modules, feature extraction processing is performed on the sequence features and the pre-trained sequence features to obtain a second learning feature; the first learning feature and the second learning feature are fused to obtain a first fused feature; the first fused feature is processed to obtain protein function annotation information.
  • the present invention obtains sequence features, pre-trained sequence features and preset annotation features based on protein sequences, and by fusion processing of the above three different angles of features, the obtained protein function annotation information is more accurate; the prediction accuracy of protein function annotation is effectively improved.
  • FIG1 is a schematic diagram of a method for protein function annotation provided by an embodiment of the present disclosure
  • FIG2 is a schematic diagram of a process of another method for protein function annotation provided by an embodiment of the present disclosure.
  • FIG3 is a schematic diagram of a network architecture of a protein function annotation model MergeNet algorithm
  • FIG4 is a schematic diagram of a protein function annotation information prediction result provided by the present disclosure.
  • FIG5 is a schematic diagram of a process of another method for protein function annotation provided by an embodiment of the present disclosure.
  • FIG6 is a schematic diagram of a protein sequence feature
  • FIG7 is a schematic diagram of a network architecture of a protein function annotation model ResMergeNet algorithm
  • FIG8 is a schematic diagram of a process for training a protein function annotation model provided by an embodiment of the present disclosure
  • FIG9 is a schematic diagram of a process for training another protein function annotation model provided by an embodiment of the present disclosure.
  • FIG10 is a schematic diagram of the structure of a device for protein function annotation provided by an embodiment of the present disclosure.
  • FIG11 is a schematic diagram of the structure of another apparatus for protein function annotation provided by an embodiment of the present disclosure.
  • FIG. 12 is a schematic block diagram of an example electronic device provided by an embodiment of the present disclosure.
  • sequence features from different angles can now be calculated from protein sequence information.
  • current protein function annotation algorithms based only on sequence information often only use one of the sequence features.
  • the inability to integrate multiple protein sequence features well is the main reason that affects the prediction performance of current protein function annotation algorithms based only on sequence information.
  • FIG1 is a schematic diagram of a flow chart of a method for protein function annotation provided in an embodiment of the present disclosure.
  • the method comprises the following steps:
  • Step 101 according to the protein sequence, respectively obtain sequence features, pre-trained sequence features and preset annotation features.
  • the protein sequence is a protein sequence that needs to be annotated with protein function.
  • the sequence feature is a coding feature (One-Hot feature) generated according to the protein sequence
  • the pre-trained sequence feature is a pre-trained feature calculated by the protein language model ProtBert according to the One-Hot feature of the protein
  • the preset annotation feature is an InterPro feature retrieved from the InterPro database based on the protein sequence.
  • sequence feature described in the disclosed embodiment is described by taking the One-Hot feature obtained by One-Hot encoding as an example
  • the pre-trained sequence feature is described by taking the feature calculated in advance according to the sequence feature using the protein language model as an example
  • preset annotation feature is described by taking the InterPro feature retrieved from the InterPro database according to the sequence using the InterProScan tool as an example, but this does not constitute a limitation on the sequence feature, the pre-trained sequence feature and the preset annotation feature.
  • the protein sequence is encoded into the One-Hot feature matrix of the protein sequence through the One-Hot encoding technology; the protein language model ProtBert can mine the internal pattern information of the protein sequence according to the One-Hot feature and express the internal pattern information of the protein sequence explicitly; the InterPro feature integrates a variety of features that are strongly related to the protein function.
  • the InterPro feature can be used to classify the protein function with high accuracy.
  • the DeepGOPlus algorithm uses the One-Hot feature of the protein sequence as the only input feature of the algorithm; however, the sequence One-Hot feature is relatively low-level, and it is difficult for the algorithm model to efficiently learn functional category information from it, resulting in low model prediction accuracy.
  • the InterPro feature is used as the only input feature of the DeepGOZero algorithm, the prediction accuracy is relatively high, but not all proteins have InterPro features, and the model using only InterPro features cannot classify all proteins.
  • the disclosed embodiment can effectively improve the accuracy of protein function annotation by using the aforementioned three protein sequence features for learning.
  • Step 102 performing feature extraction processing on the preset annotation feature to obtain a first learning feature.
  • the pre-trained feature is calculated based on the One-Hot feature, and has little difference from the One-Hot feature; while the InterPro feature is very different from the One-Hot feature and the pre-trained feature.
  • Simply adding the three features at the shallow level of the network of the protein function annotation model will destroy the respective characteristics of each feature, thereby affecting the prediction accuracy of the protein function annotation model.
  • simply deepening the number of network layers of the protein function annotation model will cause the problem of gradient vanishing, affect the transmission of information within the network, and further affect the prediction accuracy of the protein function annotation model.
  • the disclosed embodiment can alleviate the problem of gradient vanishing by inputting the preset annotation features (InterPro features) into the first preset number of residual fully connected modules for feature extraction processing, so that the protein function annotation model can deepen the network layer and integrate different protein sequence features well at the high level of the network.
  • preset annotation features InterPro features
  • Step 103 perform feature extraction processing on the sequence feature and the pre-trained sequence feature to obtain a second learning feature.
  • step 102 In order to alleviate the problem of gradient vanishing in the protein function annotation model, the sequence features and the pre-trained sequence features are input into a second preset number of residual convolution modules for feature extraction processing.
  • Step 104 Fusing the first learning feature with the second learning feature to obtain a first fused feature.
  • the InterPro feature is very different from the One-Hot feature and the pre-trained feature
  • the One-Hot feature, the pre-trained feature and the InterPro feature are respectively input into the second preset number of residual convolution modules and the first preset number of residual fully connected modules for feature extraction processing; and the extracted first learning feature and the second learning feature are fused at the high level of the network of the protein function annotation model to obtain the first fused feature. Fusion of these features from different sources can make up for the deficiencies between different features, thereby improving the quality of the overall feature and improving the prediction accuracy of the protein function annotation model.
  • Step 105 Process the first fusion feature to obtain protein function annotation information.
  • the functional annotation information of the protein cannot be directly obtained by simply fusing the first learning feature and the second learning feature. It is also necessary to perform feature extraction on the first fusion feature again to fully exploit the advantages of the One-Hot feature, the pre-trained feature and the InterPro feature to achieve information complementarity between the features. By performing feature extraction on the first fusion feature, protein functional annotation information is generated.
  • the present disclosure provides a method for protein function annotation, which obtains sequence features, pre-trained sequence features and preset annotation features respectively according to the protein sequence; based on a first preset number of residual fully connected modules, performs feature extraction processing on the preset annotation features to obtain a first learning feature; based on a second preset number of residual convolution modules, performs feature extraction processing on the sequence features and the pre-trained sequence features to obtain a second learning feature; fuses the first learning feature with the second learning feature to obtain a first fused feature; processes the first fused feature to obtain protein function annotation information.
  • the present disclosure obtains sequence features, pre-trained sequence features and preset annotation features based on protein sequences, and by fusing the features of the above three different angles, the obtained protein function annotation information has more protein function information; and effectively improves the prediction accuracy of protein function annotation.
  • the embodiments of the present disclosure provide a flowchart of another method for protein function annotation.
  • the method comprises the following steps:
  • Step 201 according to the protein sequence, respectively obtain sequence features, pre-trained sequence features and preset annotation features.
  • step 201 please refer to the above embodiment, so it will not be described in detail here.
  • Step 202 Based on a first preset number of fully connected layers, feature extraction processing is performed on the preset annotation feature to obtain a third learning feature.
  • a protein function annotation model (MergeNet algorithm) is provided that fuses multiple sequence features at the shallow level of the network, which makes up for the defects of the model algorithm using a single feature, improves the prediction accuracy of the algorithm using only One-Hot features, and increases the universality of the algorithm using only InterPro features.
  • MergeNet uses a fully connected layer (MLP) to perform preliminary learning on InterPro features and obtain the learned InterPro features
  • MLP fully connected layer
  • d is the feature dimension of the hidden layer.
  • d is set to 1024, and the feature dimensions d of all hidden layers remain consistent.
  • Step 203 Based on the convolutional layer with a second preset number of layers, feature extraction processing is performed on the sequence feature and the pre-trained sequence feature to obtain a fourth learning feature and a fifth learning feature.
  • the One-Hot feature H and the pre-trained feature E obtained according to the protein sequence are input into a one-dimensional convolutional layer (CNN) for preliminary feature learning, and the obtained Where d is the feature dimension of the convolutional layer, and then H and E retain the largest feature value in the residue dimension for maximum pooling operation, and get and
  • CNN one-dimensional convolutional layer
  • Step 204 fuse the third learning feature, the fourth learning feature and the fifth learning feature to obtain a second fused feature; wherein the fusion process is an addition process.
  • MergeNet merges the three features by summing them up in the shallow layer of the network (i.e., after all three features have passed through a feature learning layer) to obtain the protein feature
  • the summing process is:
  • Step 205 Based on a third preset number of fully connected layers, classify and calculate the second fusion features to obtain protein function annotation information.
  • the protein feature P is sent to the fully connected layer to calculate the logical probability of each functional category.
  • protein function annotation information where class is the total number of protein functions, and in this embodiment, class is 5901.
  • the prediction result schematic diagram is shown in FIG4 , in which three functions of ten proteins predicted by the algorithm are selected, wherein the first column is the predicted protein name, and the following three columns are the predicted functions of each protein, wherein the number of the function in the gene ontology predicted by the algorithm is indicated in front of “
  • the embodiments of the present disclosure provide a flowchart of another method for protein function annotation.
  • the method comprises the following steps:
  • Step 301 Encode the protein sequence based on a preset encoding algorithm to obtain the sequence feature.
  • Step 302 pre-train the sequence features based on a preset protein language model to obtain the pre-trained sequence features.
  • the preset encoding algorithm is a One-Hot encoding algorithm
  • the One-Hot encoding algorithm is used to encode the protein sequence to be annotated to generate a One-Hot feature.
  • the One-Hot feature is pre-trained using the protein language model ProtBert.
  • the protein language model can mine the internal pattern information of the protein sequence and explicitly express the internal pattern information of the protein sequence.
  • N is the number of amino acids in the protein
  • the pre-trained feature E is not a binary feature matrix composed of 0 and 1, but a feature matrix composed of continuous floating-point numbers.
  • the protein SKSP4_ASCTR composed of eight residues in SwissProt has the sequence: CHYIFNTC, The One-Hot features and pre-trained features of the sequence are shown in Figure 6.
  • the dimension of the One-Hot features of the sequence is The dimension of its sequence pre-training feature is
  • Step 303 Searching for the preset annotation feature corresponding to the protein sequence in a preset protein database.
  • FIG. 7 is a schematic diagram of the network architecture of the ResMergeNet model algorithm.
  • the InterProScan tool is used to retrieve the InterPro features of proteins that are strongly correlated with the functional categories from the InterPro database.
  • M is the dimension of the protein InterPro feature, that is, the total number of categories of the InterPro feature.
  • Step 304 input the preset annotation features into a first preset number of residual fully connected modules for feature extraction processing; and obtain the first learning features.
  • the steps specifically include: inputting the preset annotation features into the first residual fully connected module in the residual fully connected module; using the output features of the first residual fully connected module as input features of the second residual fully connected module for feature extraction processing; and looping the above steps for a first predetermined number of times to obtain the first learning feature, wherein the first predetermined number is equal to the first preset number.
  • the first residual fully connected module in which the preset annotation features are input into the residual fully connected module includes: inputting the preset annotation features into a normalization layer for normalization processing to obtain the normalized preset annotation features; inputting the normalized preset annotation features into a nonlinear layer for nonlinear mapping to obtain the nonlinearly mapped preset annotation features; inputting the nonlinearly mapped preset annotation features into a fully connected layer for feature extraction to obtain a first preset annotation feature; inputting the first preset annotation feature into the normalization layer, the nonlinear layer and the fully connected layer in sequence for processing to obtain a second preset annotation feature; and summing the preset annotation features with the second preset annotation features to obtain a first output feature.
  • the InterPro feature (M is the total number of InterPro feature categories) is sent to four sequentially connected residual fully connected modules (ResMLP) for high-level feature extraction.
  • the input features are first normalized by the normalization layer (Norm), then nonlinearly mapped by the nonlinear layer (ReLU), and then learned by the fully connected layer (MLP). Then, the features are learned by the normalization layer, nonlinear layer, and fully connected layer in sequence. At this time, the initial input features of the residual fully connected module are residually summed with the learned features to obtain the high-level InterPro features. (First learning feature).
  • Step 305 Based on a fourth preset number of convolutional layers, the sequence features are processed to obtain a sixth learning feature, and the pre-trained sequence features are processed to obtain a seventh learning feature.
  • the One-Hot feature and the pre-trained feature are respectively input into a one-dimensional convolution layer for preliminary feature extraction, and the obtained and Wherein N is the number of amino acids in the protein, and d is the characteristic dimension of the convolution layer. It should be noted that the embodiment of the present disclosure takes a one-dimensional convolution layer as an example for explanation, which does not constitute a limitation on the fourth preset number of layers.
  • Step 306 Add the sixth learning feature and the seventh learning feature to obtain a third fusion feature.
  • the model after learning a one-dimensional convolutional layer, the model performs sum processing on these two features, and the feature after the sum processing is the third fusion feature P s , that is,
  • Step 307 input the third fusion feature into a second preset number of residual convolution modules for feature extraction processing to obtain the second learning feature.
  • the steps specifically include: inputting the third fusion feature into the first residual convolution module in the residual convolution module; using the output feature of the first residual convolution module as the input feature of the second residual convolution module for feature extraction processing; and looping the above steps for a second predetermined number of times to obtain the second learning feature, wherein the second predetermined number is equal to the second preset number.
  • the step of inputting the third fused feature into the first residual convolution module in the residual convolution module includes: inputting the third fused feature into a normalization layer for normalization processing to obtain the third fused feature after normalization processing; inputting the third fused feature after normalization processing into a nonlinear layer for nonlinear mapping to obtain the third fused feature after nonlinear mapping; inputting the third fused feature after nonlinear mapping into a convolution layer for feature learning to obtain a second output feature; inputting the second output feature into the normalization layer, the nonlinear layer and the convolution layer in sequence for processing to obtain a third output feature; and adding the third fused feature and the third output feature to obtain a fourth output feature.
  • the third fused feature Ps is sent to 8 sequentially connected residual convolution modules to extract high-level features.
  • the input feature is first normalized by the normalization layer (Norm), then nonlinearly mapped by the nonlinear layer (ReLU), and then learned by the convolution layer (CNN), and then sequentially learned by the normalization layer, nonlinear layer and convolution layer.
  • the fourth output feature retains the largest eigenvalue in the residue dimension for maximum pooling operation to obtain a high-level sequence feature. (Second learning characteristic).
  • Step 308 concatenate the first learned feature and the second learned feature to obtain a first fused feature.
  • the high-level sequence feature Ps (second learning feature) and the high-level InterPro feature I (first learning feature) become more abstract, and the feature distribution becomes similar, both of which are highly correlated with the function of the protein.
  • ResMergeNet concatenates the high-level sequence feature Ps and the high-level InterPro feature I, and the resulting feature is called the high-level protein feature (first fusion feature), i.e. Among them, concatenate(,) means concatenating two vectors according to the feature dimension.
  • Step 309 input the first fusion feature into an integration module for integration processing to obtain a protein fusion feature.
  • the steps specifically include: inputting the first fusion feature into a normalization layer for normalization processing to obtain a first fusion feature after normalization processing; inputting the first fusion feature after normalization processing into a nonlinear layer for nonlinear mapping to obtain a first fusion feature after nonlinear mapping; inputting the first fusion feature after nonlinear mapping into a fully connected layer for feature extraction to obtain a first protein fusion feature; inputting the first protein fusion feature into the normalization layer, the nonlinear layer and the fully connected layer in sequence for processing to obtain a second protein fusion feature; and summing the first fusion feature and the second protein fusion feature to obtain the protein fusion feature.
  • the high-level protein feature (the first fusion feature)
  • the input residual integration module (Integrate) is used for further protein feature extraction.
  • the input features are first normalized by the normalization layer (Norm), then nonlinearly mapped by the nonlinear layer (ReLU), and then learned by the fully connected layer (MLP). Then, the features are learned by the normalization layer, nonlinear layer, and fully connected layer in sequence. At this time, the initial input features of the residue fully connected module are residually summed with the learned features to obtain (Protein fusion signature).
  • Step 310 Based on a fifth preset number of fully connected layers, the protein fusion features are classified to obtain the protein function annotation information.
  • the obtained protein fusion feature P is sent to the fully connected layer (MLP) to classify the protein function, and the logical probability of the functional category of the protein is obtained. (Protein function annotation information).
  • the method before respectively acquiring sequence features, pre-trained sequence features, and preset annotation features according to the protein sequence, the method further includes:
  • the protein function annotation model is trained based on the training protein sequences.
  • the protein sequence for training is input into the protein function annotation model for training to improve the prediction accuracy of the model.
  • the embodiments of the present disclosure provide a flowchart of a method for training a protein function annotation model.
  • the method comprises the following steps:
  • Step 401 acquiring training sequence features, training first sequence features and training preset annotation features respectively according to the training protein sequence.
  • Step 402 Perform feature extraction processing on the preset annotation feature for training to obtain a first learning feature for training.
  • Step 403 Perform feature extraction processing on the training sequence feature and the first training sequence feature to obtain a second training learning feature.
  • Step 404 Fusing the first training learning feature with the second training learning feature to obtain a first training fused feature.
  • Step 405 Process the first training fusion feature to obtain training protein function annotation information.
  • steps 101 to 105 please refer to the detailed description in the above embodiment, so they will not be described one by one here.
  • the present disclosed embodiment after respectively obtaining training sequence features, training first sequence features and training preset annotation features according to the training protein sequence, the present disclosed embodiment provides a flow chart of another method for training a protein function annotation model.
  • the method comprises the following steps:
  • Step 501 acquiring training sequence features, training first sequence features and training preset annotation features respectively according to the training protein sequence.
  • Step 502 Based on the first preset number of training fully connected layers, feature extraction processing is performed on the training preset annotation features to obtain third training learning features.
  • Step 503 Based on the second preset number of training convolutional layers, feature extraction processing is performed on the training sequence features and the training first sequence features to obtain fourth training learning features and fifth training learning features.
  • Step 504 fuse the third training learning feature, the fourth training learning feature, and the fifth training learning feature to obtain a second training fused feature; wherein the fusion process is an addition process.
  • Step 505 Based on a third preset number of training fully connected layers, classify and calculate the second training fusion features to obtain training protein function annotation information.
  • steps 201 to 205 please refer to the detailed description in the above embodiment, so they will not be described one by one here.
  • the present invention also provides a protein function annotation device. Since the device embodiment of the present invention corresponds to the above-mentioned method embodiment, details not disclosed in the device embodiment can be referred to the above-mentioned method embodiment, and will not be repeated in the present invention.
  • FIG10 is a schematic diagram of the structure of a protein function annotation device provided in an embodiment of the present disclosure. As shown in FIG10 , the device includes: an acquisition unit 61 , a first processing unit 62 , a second processing unit 63 , a first fusion unit 64 and a third processing unit 65 .
  • An acquisition unit 61 is used to acquire sequence features, pre-trained sequence features and preset annotation features according to the protein sequence;
  • a first processing unit 62 is used to perform feature extraction processing on the preset annotation feature to obtain a first learning feature
  • a second processing unit 63 is used to perform feature extraction processing on the sequence feature and the pre-trained sequence feature to obtain a second learning feature
  • a first fusion unit 64 configured to fuse the first learning feature with the second learning feature to obtain a first fused feature
  • the third processing unit 65 is used to process the first fusion feature to obtain protein function annotation information.
  • the present disclosure provides a device for protein function annotation, which obtains sequence features, pre-trained sequence features and preset annotation features according to protein sequences; performs feature extraction processing on the preset annotation features to obtain a first learning feature; performs feature extraction processing on the sequence features and the pre-trained sequence features to obtain a second learning feature; fuses the first learning feature with the second learning feature to obtain a first fused feature; processes the first fused feature to obtain protein function annotation information.
  • the present disclosure obtains sequence features, pre-trained sequence features and preset annotation features based on protein sequences, and by fusing the features from the above three different angles, the obtained protein function annotation information is more accurate; and the prediction accuracy of protein function annotation is effectively improved.
  • the device further includes:
  • a fourth processing unit 66 is used for, after acquiring sequence features, pre-trained sequence features and preset annotation features according to the protein sequence, performing feature extraction processing on the preset annotation features based on a first preset number of fully connected layers to obtain third learning features;
  • a fifth processing unit 67 is used to perform feature extraction processing on the sequence feature and the pre-trained sequence feature based on the convolution layer with a second preset number of layers to obtain a fourth learning feature and a fifth learning feature;
  • a second fusion unit 68 is used to fuse the third learning feature, the fourth learning feature and the fifth learning feature to obtain a second fused feature; wherein the fusion process is an addition process;
  • the sixth processing unit 69 is used to perform classification calculation on the second fusion features based on a third preset number of fully connected layers to obtain protein function annotation information.
  • the acquisition unit 61 includes:
  • An encoding module 611 is used to encode the protein sequence based on a preset encoding algorithm to obtain the sequence feature;
  • a pre-training module 612 is used to perform pre-training processing on the sequence features based on a preset protein language model to obtain the pre-trained sequence features;
  • the search module 613 is used to search for the preset annotation feature corresponding to the protein sequence in a preset protein database.
  • the first processing unit 62 includes:
  • a first input module 621 used to input the preset annotation features into a first preset number of residual fully connected modules
  • a first processing module 622 configured to perform feature extraction processing using the output features of the first residual fully connected module as input features of a second residual fully connected module;
  • the first loop module 623 is used to loop and execute the processing method of the first processing module for a first predetermined number of times to obtain the first learning feature, wherein the first predetermined number of times is equal to the first preset number.
  • the first input module 621 is further used to:
  • the preset annotation feature and the second preset annotation feature are added together to obtain a first output feature.
  • the second processing unit 63 includes:
  • a second processing module 631 is used to process the sequence feature based on a fourth preset number of convolutional layers to obtain a sixth learning feature, and to process the pre-trained sequence feature to obtain a seventh learning feature;
  • a first adding module 632 configured to add the sixth learning feature and the seventh learning feature to obtain a third fusion feature
  • a second input module 633 used to input the third fusion feature into a second preset number of residual convolution modules
  • a first extraction module 634 configured to perform feature extraction processing using the output features of the first residual convolution module as input features of the second residual convolution module;
  • the second loop module 635 is used to loop and execute the processing method of the first extraction module for a second predetermined number of times to obtain the second learning feature, wherein the second predetermined number of times is equal to the second preset number.
  • the second input module 633 is further configured to:
  • the third fusion feature and the third output feature are added together to obtain a fourth output feature.
  • the first fusion unit 64 is further configured to:
  • the first learning feature and the second learning feature are concatenated to obtain a first fusion feature.
  • the third processing unit 65 includes:
  • Integration module 651 used for inputting the first fusion feature into an integration module for integration processing to obtain a protein fusion feature
  • the classification module 652 is used to classify the protein fusion features based on a fifth preset number of fully connected layers to obtain the protein function annotation information.
  • the integration module 651 is further used to:
  • the first fusion feature and the second protein fusion feature are added together to obtain the protein fusion feature.
  • the device further includes:
  • the training unit 610 is used to train the protein function annotation model based on the training protein sequence before respectively obtaining the sequence features, the pre-trained sequence features and the preset annotation features according to the protein sequence.
  • the training unit 610 includes:
  • a training acquisition module 6101 is used to respectively acquire a training sequence feature, a training first sequence feature and a training preset annotation feature according to a training protein sequence;
  • a first training processing module 6102 is used to perform feature extraction processing on the preset annotation features for training to obtain first learning features for training;
  • a second training processing module 6103 is used to perform feature extraction processing on the training sequence features and the first training sequence features to obtain second training learning features;
  • a first training fusion module 6104 is used to fuse the first training learning feature with the second training learning feature to obtain a first training fusion feature
  • the third training processing module 6105 is used to process the first training fusion feature to obtain training protein function annotation information.
  • the training unit 610 further includes:
  • the fourth processing module 6106 for training is used to obtain the training sequence features, the first sequence features for training and the preset annotation features for training respectively according to the training protein sequence, and then perform feature extraction processing on the preset annotation features for training based on the first preset number of fully connected layers for training to obtain third learning features for training;
  • a fifth training processing module 6107 configured to perform feature extraction processing on the training sequence features and the training first sequence features based on the second preset number of training convolutional layers, respectively, to obtain fourth training learning features and fifth training learning features;
  • the second training fusion module 6108 is used to fuse the third training learning feature, the fourth training learning feature and the fifth training learning feature to obtain a second training fusion feature; wherein the fusion process is an addition process;
  • the sixth processing module 6109 for training is used to perform classification calculation on the second training fusion features based on the third preset number of training fully connected layers to obtain training protein function annotation information.
  • the present disclosure also provides an electronic device, a readable storage medium and a computer program product.
  • FIG. 12 shows a schematic block diagram of an example electronic device 700 that can be used to implement an embodiment of the present disclosure.
  • the electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.
  • the electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices.
  • the components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and/or required herein.
  • the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 702 or a computer program loaded from a storage unit 708 into a RAM (Random Access Memory) 703.
  • ROM Read-Only Memory
  • RAM Random Access Memory
  • various programs and data required for the operation of the device 700 can also be stored.
  • the computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704.
  • An I/O (Input/Output) interface 705 is also connected to the bus 704.
  • a number of components in the device 700 are connected to the I/O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc.
  • the communication unit 709 allows the device 700 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.
  • the computing unit 701 may be a variety of general and/or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc.
  • the computing unit 701 performs the various methods and processes described above, such as the method of protein function annotation.
  • the method of protein function annotation may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708.
  • part or all of the computer program may be loaded and/or installed on the device 700 via the ROM 702 and/or the communication unit 709.
  • the computer program When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the method described above may be performed.
  • the computing unit 701 may be configured to execute the aforementioned protein function annotation method in any other appropriate manner (eg, by means of firmware).
  • Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Array), ASICs (Application-Specific Integrated Circuit), ASSPs (Application Specific Standard Product), SOCs (System On Chip), CPLDs (Complex Programmable Logic Device), computer hardware, firmware, software, and/or combinations thereof.
  • FPGAs Field Programmable Gate Array
  • ASICs Application-Specific Integrated Circuit
  • ASSPs Application Specific Standard Product
  • SOCs System On Chip
  • CPLDs Complex Programmable Logic Device
  • These various embodiments may include: being implemented in one or more computer programs that are executable and/or interpreted on a programmable system including at least one programmable processor that may be a special purpose or general purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
  • a programmable processor that may be a special purpose or general purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
  • the program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, implements the functions/operations specified in the flow chart and/or block diagram.
  • the program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.
  • a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device.
  • a machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium.
  • a machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing.
  • machine-readable storage media would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, RAM, ROM, EPROM (Electrically Programmable Read-Only-Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
  • the systems and techniques described herein may be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer.
  • a display device e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor
  • a keyboard and pointing device e.g., a mouse or trackball
  • Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).
  • the systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components.
  • the components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.
  • a computer system may include a client and a server.
  • the client and the server are generally remote from each other and usually interact through a communication network.
  • the relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other.
  • the server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short).
  • the server may also be a server for a distributed system, or a server combined with a blockchain.
  • artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies.
  • Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning/deep learning, big data processing technology, knowledge graph technology, and other major directions.

Landscapes

  • Spectroscopy & Molecular Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biotechnology (AREA)
  • Evolutionary Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

本公开提供了一种蛋白质功能注释的方法及装置、电子设备和存储介质，根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；对预设注释特征进行特征提取处理，得到第一学习特征；对序列特征、预训练序列特征进行特征提取处理，得到第二学习特征；将第一学习特征与第二学习特征进行融合处理，得到第一融合特征；对第一融合特征进行处理，得到蛋白质功能注释信息。与相关技术相比，本公开基于蛋白质序列获取序列特征、预训练序列特征以及预设注释特征，通过对前述三种不同角度的特征的融合处理，使得获得的蛋白质功能注释信息更加准确。

Description

蛋白质功能注释的方法及装置、电子设备和存储介质 技术领域
本公开涉及数据处理技术领域，尤其涉及一种蛋白质功能注释的方法及装置、电子设备和存储介质。
背景技术
蛋白质中氨基酸的顺序决定了蛋白质的空间结构，蛋白质的结构又决定了蛋白质的在生物学上的功能。蛋白质功能注释的目的是用算法对蛋白质可能具有的功能进行预测。当前蛋白质功能注释算法主要分类三类，包括仅基于序列信息的算法，基于序列和结构信息的算法以及基于序列和交互网络的算法。
由于蛋白质的结构信息和交互网络信息需要通过复杂实验获得，相对难以获取，而蛋白质的氨基酸序列信息是最易获得的蛋白质的信息，因此仅基于序列信息算法应用最为广泛。但是相对于其他使用了难以从序列中获得的额外信息的算法，仅基于序列信息的蛋白质功能注释算法的预测准确率较低。
发明内容
本公开提供了一种蛋白质功能注释的方法及装置、电子设备和存储介质。其主要目的在于解决仅基于序列信息的蛋白质功能注释算法的预测准确率较低的问题。
根据本公开的第一方面，提供了一种蛋白质功能注释的方法，其中，包括：
根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；
对所述预设注释特征进行特征提取处理，得到第一学习特征；
对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；
将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；
对所述第一融合特征进行处理，得到蛋白质功能注释信息。
可选的，在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之后，所述方法还包括：
基于第一预设层数的全连接层，对所述预设注释特征进行特征提取处理，得到第三学习特征；
基于第二预设层数的卷积层，分别对所述序列特征、所述预训练序列特征进行特征提取处理，得到第四学习特征与第五学习特征；
将所述第三学习特征、所述第四学习特征以及所述第五学习特征进行融合处理，得到第二融合特征；其中，所述融合处理为加和处理；
基于第三预设层数的全连接层，对所述第二融合特征进行分类计算，得到蛋白质功能注释信息。
可选的，所述根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征包括：
基于预设编码算法，对所述蛋白质序列进行编码得到所述序列特征；
基于预设蛋白质语言模型，对所述序列特征进行预训练处理，得到所述预训练序列特征；
在预设蛋白质数据库中，查找与所述蛋白质序列对应的所述预设注释特征。
可选的，所述对所述预设注释特征进行特征提取处理，得到第一学习特征包括如下步骤：
将所述预设注释特征输入第一预设数量的残差全连接模块；
将第一个残差全连接模块的输出特征作为第二个残差全连接模块的输入特征进行特征提取处理；
按照如上步骤循环执行第一预定次数，以获取所述第一学习特征，其中，所述第一预定次数等于所述第一预设数量。
可选的，所述将所述预设注释特征输入所述残差全连接模块中的第一个残差全连接模块包括：
将所述预设注释特征输入归一化层进行归一化处理，得到归一化处理后的所述预设注释特征；
将所述归一化处理后的预设注释特征输入非线性层进行非线性映射，得到非线性映射后的预设注释特征；
将所述非线性映射后的预设注释特征输入全连接层进行特征提取，得到第一预设注释特征；
将所述第一预设注释特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二预设注释特征；
将所述预设注释特征与所述第二预设注释特征进行加和处理，得到第一输出特征。
可选的，基于第四预设层数的卷积层，对所述序列特征进行处理，得到第六学习特征、对所述预训练序列特征进行处理，得到第七学习特征；
将所述第六学习特征与所述第七学习特征进行加和处理，得到第三融合特征；
所述对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征包括如下步骤：
将所述第三融合特征输入第二预设数量的残差卷积模块；
将第一个所述残差卷积模块的输出特征作为第二个所述残差卷积模块的输入特征进行特征提取处理；
按照如上步骤循环执行第二预定次数，得到所述第二学习特征，其中，所述第二预定次数等于所述第二预设数量。
可选的，所述将所述第三融合特征输入所述残差卷积模块中的第一个残差卷积模块包括：
将所述第三融合特征输入归一化层进行归一化处理，得到归一化处理后的第三融合特征；
将所述归一化处理后的所述第三融合特征输入非线性层进行非线性映射，得到非线性映射后的第三融合特征；
将所述非线性映射后的第三融合特征输入卷积层进行特征学习，得到第二输出特征；
将所述第二输出特征依次输入所述归一化层、所述非线性层以及所述卷积层进行处理，得到第三输出特征；
将所述第三融合特征与所述第三输出特征进行加和处理，得到第四输出特征。
可选的，所述将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征包括：
将所述第一学习特征与所述第二学习特征进行拼接处理，得到第一融合特征。
可选的，所述对所述第一融合特征进行处理，得到蛋白质功能注释信息包括：
将所述第一融合特征输入整合模块进行整合处理，得到蛋白质融合特征；
基于第五预设层数的全连接层，对所述蛋白质融合特征进行分类处理，得到所述蛋白质功能注释信息。
可选的，所述将所述第一融合特征输入整合模块进行整合处理，得到蛋白质融合特征包括：
将所述第一融合特征输入归一化层进行归一化处理，得到归一化处理后的第一融合特征；
将所述归一化处理后的第一融合特征输入非线性层进行非线性映射，得到非线性映射后的第一融合特征；
将所述非线性映射后的第一融合特征输入全连接层进行特征提取，得到第一蛋白质融合特征；
将所述第一蛋白质融合特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二蛋白质融合特征；
将所述第一融合特征与所述第二蛋白质融合特征进行加和处理，得到所述蛋白质融合特征。
可选的，在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之前，所述方法还包括：
基于训练用蛋白质序列，对蛋白质功能注释模型进行训练。
可选的，所述将训练用蛋白质序列输入蛋白质功能注释模型进行训练包括：
根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征；
对所述训练用预设注释特征进行特征提取处理，得到第一训练用学习特征；
对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第二训练用学习特征；
将所述第一训练用学习特征与所述第二训练用学习特征进行融合处理，得到第一训练用融合特征；
对所述第一训练用融合特征进行处理，得到训练用蛋白质功能注释信息。
可选的，在根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征之后，所述方法还包括：
基于第一预设层数的训练用全连接层，对所述训练用预设注释特征进行特征提取处理，得到第三训练用学习特征；
基于第二预设层数的训练用卷积层，分别对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第四训练用学习特征与第五训练用学习特征；
将所述第三训练用学习特征、所述第四训练用学习特征以及所述第五训练用学习特征进行融合处理，得到第二训练用融合特征；其中，所述融合处理为加和处理；
基于第三预设层数的训练用全连接层，对所述第二训练用融合特征进行分类计算，得到训练用蛋白质功能注释信息。
根据本公开的第二方面，提供了一种蛋白质功能注释的装置，包括：
获取单元，用于根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；
第一处理单元，用于对所述预设注释特征进行特征提取处理，得到第一学习特征；
第二处理单元，用于对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；
第一融合单元，用于将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；
第三处理单元，用于对所述第一融合特征进行处理，得到蛋白质功能注释信息。
可选的，所述装置还包括：
第四处理单元，用于在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之后，基于第一预设层数的全连接层，对所述预设注释特征进行特征提取处理，得到第三学习特征；
第五处理单元，用于基于第二预设层数的卷积层，分别对所述序列特征、所述预训练序列特征进行特征提取处理，得到第四学习特征与第五学习特征；
第二融合单元，用于将所述第三学习特征、所述第四学习特征以及所述第五学习特征进行融合处理，得到第二融合特征；其中，所述融合处理为加和处理；
第六处理单元，用于基于第三预设层数的全连接层，对所述第二融合特征进行分类计算，得到蛋白质功能注释信息。
可选的，所述获取单元包括：
编码模块，用于基于预设编码算法，对所述蛋白质序列进行编码得到所述序列特征；
预训练模块，用于基于预设蛋白质语言模型，对所述序列特征进行预训练处理，得到所述预训练序列特征；
查找模块，用于在预设蛋白质数据库中，查找与所述蛋白质序列对应的所述预设注释特征。
可选的，所述第一处理单元包括：
第一输入模块，用于将所述预设注释特征输入第一预设数量的残差全连接模块；
第一处理模块，用于将第一个残差全连接模块的输出特征作为第二个残差全连接模块的输入特征进行特征提取处理；
第一循环模块，用于按照第一处理模块的处理方法循环执行第一预定次数，以获取所述第一学习特征，其中，所述第一预定次数等于所述第一预设数量。
可选的，所述第一输入模块还用于：
将所述预设注释特征输入归一化层进行归一化处理，得到归一化处理后的所述预设注释特征；
将所述归一化处理后的预设注释特征输入非线性层进行非线性映射，得到非线性映射后的预设注释特征；
将所述非线性映射后的预设注释特征输入全连接层进行特征提取，得到第一预设注释特征；
将所述第一预设注释特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二预设注释特征；
将所述预设注释特征与所述第二预设注释特征进行加和处理，得到第一输出特征。
可选的，所述第二处理单元包括：
第二处理模块，用于基于第四预设层数的卷积层，对所述序列特征进行处理，得到第六学习特征、对所述预训练序列特征进行处理，得到第七学习特征；
第一加和模块，用于将所述第六学习特征与所述第七学习特征进行加和处理，得到第三融合特征；
第二输入模块，用于将所述第三融合特征输入第二预设数量的残差卷积模块；
第一提取模块，用于将第一个所述残差卷积模块的输出特征作为第二个所述残差卷积模块的输入特征进行特征提取处理；
第二循环模块，用于按照第一提取模块的处理方法循环执行第二预定次数，得到所述第二学习特征，其中，所述第二预定次数等于所述第二预设数量。
可选的，所述第二输入模块还用于：
将所述第三融合特征输入归一化层进行归一化处理，得到归一化处理后的第三融合特征；
将所述归一化处理后的所述第三融合特征输入非线性层进行非线性映射，得到非线性映射后的第三融合特征；
将所述非线性映射后的第三融合特征输入卷积层进行特征学习，得到第二输出特征；
将所述第二输出特征依次输入所述归一化层、所述非线性层以及所述卷积层进行处理，得到第三输出特征；
将所述第三融合特征与所述第三输出特征进行加和处理，得到第四输出特征。
可选的，所述第一融合单元还用于：
将所述第一学习特征与所述第二学习特征进行拼接处理，得到第一融合特征。
可选的，所述第三处理单元包括：
整合模块，用于将所述第一融合特征输入整合模块进行整合处理，得到蛋白质融合特征；
分类模块，用于基于第五预设层数的全连接层，对所述蛋白质融合特征进行分类处理，得到所述蛋白质功能注释信息。
可选的，所述整合模块还用于：
将所述第一融合特征输入归一化层进行归一化处理，得到归一化处理后的第一融合特征；
将所述归一化处理后的第一融合特征输入非线性层进行非线性映射，得到非线性映射后的第一融合特征；
将所述非线性映射后的第一融合特征输入全连接层进行特征提取，得到第一蛋白质融合特征；
将所述第一蛋白质融合特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二蛋白质融合特征；
将所述第一融合特征与所述第二蛋白质融合特征进行加和处理，得到所述蛋白质融合特征。
可选的，所述装置还包括：
训练单元，用于在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之前，基于训练用蛋白质序列，对蛋白质功能注释模型进行训练。
可选的，所述训练单元包括：
训练用获取模块，用于根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征；
训练用第一处理模块，用于对所述训练用预设注释特征进行特征提取处理，得到第一训练用学习特征；
训练用第二处理模块，用于对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第二训练用学习特征；
训练用第一融合模块，用于将所述第一训练用学习特征与所述第二训练用学习特征进行融合处理，得到第一训练用融合特征；
训练用第三处理模块，用于对所述第一训练用融合特征进行处理，得到训练用蛋白质功能注释信息。
可选的，所述训练单元还包括：
训练用第四处理模块，用于在根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征之后，基于第一预设层数的训练用全连接层，对所述训练用预设注释特征进行特征提取处理，得到第三训练用学习特征；
训练用第五处理模块，用于基于第二预设层数的训练用卷积层，分别对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第四训练用学习特征与第五训练用学习特征；
训练用第二融合模块，用于将所述第三训练用学习特征、所述第四训练用学习特征以及所述第五训练用学习特征进行融合处理，得到第二训练用融合特征；其中，所述融合处理为加和处理；
训练用第六处理模块，用于基于第三预设层数的训练用全连接层，对所述第二训练用融合特征进行分类计算，得到训练用蛋白质功能注释信息。
根据本公开的第三方面，提供了一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行前述第一方面所述的方法。
根据本公开的第四方面，提供了一种存储有计算机指令的非瞬时计算机可读存储介质，其中，所述计算机指令用于使所述计算机执行前述第一方面所述的方法。
根据本公开的第五方面，提供了一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现如前述第一方面所述的方法。
本公开提供了一种蛋白质功能注释的方法及装置、电子设备和存储介质，根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；基于第一预设数量的残差全连接模块，对所述预设注释特征进行特征提取处理，得到第一学习特征； 基于第二预设数量的残差卷积模块，对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；对所述第一融合特征进行处理，得到蛋白质功能注释信息。与相关技术相比，本公开基于蛋白质序列获取序列特征、预训练序列特征以及预设注释特征，通过对前述三种不同角度的特征的融合处理，使得获得的蛋白质功能注释信息更加准确；有效提升了蛋白质功能注释的预测准确率。
应当理解，本部分所描述的内容并非旨在标识本申请的实施例的关键或重要特征，也不用于限制本申请的范围。本申请的其它特征将通过以下的说明书而变得容易理解。
附图说明
附图用于更好地理解本方案，不构成对本公开的限定。其中：
图1为本公开实施例提供的一种蛋白质功能注释的方法的流程示意图；
图2为本公开实施例提供的另一种蛋白质功能注释的方法的流程示意图；
图3为一种蛋白质功能注释模型MergeNet算法网络架构示意图；
图4为本公开提供的一种蛋白质功能注释信息预测结果示意图；
图5为本公开实施例提供的另一种蛋白质功能注释的方法的流程示意图；
图6为一种蛋白质序列特征的示意图；
图7为一种蛋白质功能注释模型ResMergeNet算法网络架构示意图；
图8为本公开实施例提供的一种蛋白质功能注释模型训练的流程示意图；
图9为本公开实施例提供的另一种蛋白质功能注释模型训练的流程示意图；
图10为本公开实施例提供的一种蛋白质功能注释的装置的结构示意图；
图11为本公开实施例提供的另一种蛋白质功能注释的装置的结构示意图；
图12为本公开实施例提供的示例电子设备的示意性框图。
具体实施方式
以下结合附图对本公开的示范性实施例做出说明，其中包括本公开实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本公开的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。
随着蛋白质计算工具的发展，目前可以从蛋白质序列信息中计算出不同角度的序列特征。然而，当前仅基于序列信息的蛋白质功能注释算法往往只使用了其中一种序列特征，无法很好地融合多种蛋白质序列特征是影响当前仅基于序列信息的蛋白质功能注释算法预测性能的主要原因。
下面参考附图描述本公开实施例的蛋白质功能注释的方法及装置、电子设备和存储介质。
图1为本公开实施例所提供的一种蛋白质功能注释的方法的流程示意图。
如图1所示，该方法包含以下步骤：
步骤101，根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征。
在本公开实施例中，蛋白质序列为需要进行蛋白质功能注释的蛋白质序列。所述序列特征为根据蛋白质序列生成的编码特征(One-Hot特征)，所述预训练序列特征为蛋白质语言模型ProtBert根据蛋白质的One-Hot特征计算出预训练特征，所述预设注释特征为基于蛋白质的序列在InterPro数据库中检索的InterPro特征。需要说明的是，本公开实施例中所述序列特征以One-Hot编码得到的One-Hot特征为例进行说明，所述预训练序列特征以用蛋白质语言模型根据序列特征预先进行计算出的特征为例进行说明，所述预设注释特征以用InterProScan工具根据序列在InterPro数据库中检索出的InterPro特征为例进行说明，但这并不构成对所述序列特征、所述预训练序列特征以及所述预设注释特征的限定。
根据需要进行蛋白质功能注释的蛋白质序列，通过One-Hot编码技术蛋白质序列编码成蛋白质序列的One-Hot特征矩阵；蛋白质语言模型ProtBert根据One-Hot特征可以挖掘蛋白质序列的内部模式信息，并将蛋白质序列的内部模式信息显式地表达出来；InterPro特征整合了多种与蛋白质功能强相关的特征，利用InterPro特征可以以较高的精度对蛋白质的功能进行分类。
DeepGOPlus算法使用蛋白质序列的One-Hot特征作为算法唯一的输入特征；但是序列One-Hot特征较为底层，算法模型很难高效的从中学到功能类别信息，导致模型预测准确率不高。虽然使用InterPro特征作为DeepGOZero算法唯一的输入特征，预测准确率相对较高，但是并非所有蛋白质都具有InterPro特征，仅利用InterPro特征的模型无法对所有蛋白质进行分类。本公开实施例通过使用前述三种蛋白质序列特征进行学习，可以有效地提高蛋白质功能注释的准确率。
步骤102，对所述预设注释特征进行特征提取处理，得到第一学习特征。
在本公开实施例中，预训练特征基于One-Hot特征计算得到，与One-Hot特征差异较小；而InterPro特征与One-Hot特征、预训练特征的差异性很大。在蛋白质功能注释模型的网络浅层简单的将三种特征相加，会破坏每个特征的各自的特性，进而影 响蛋白质功能注释模型的预测准确率。但是简单地加深蛋白质功能注释模型的网络层数会造成梯度消失的问题，影响信息在网络内部的传递，进而也会影响影响蛋白质功能注释模型的预测准确率。本公开实施例通过将预设注释特征(InterPro特征)输入第一预设数量的残差全连接模块进行特征提取处理，可以减缓梯度消失的问题，使得蛋白质功能注释模型可以加深网络层，并在网络的高层很好地融合不同的蛋白质序列特征。
在蛋白质功能注释模型中，使用数量更多的残差全连接模块会出现过拟合的特征，影响预测的准确率，而数量较少的残差全连接模块也会影响预测的准确率；具体的，所述第一预设数量设置在4个左右更为合适；需要说明的是，本公开实施例以4个残差全连接模块进行说明，但这并不构成对所述第一预设数量的限定。
步骤103，对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征。
在本公开实施例中，请参考步骤102的说明，为了缓解蛋白质功能注释模型中梯度消失的问题，将所述序列特征、所述预训练序列特征输入第二预设数量的残差卷积模块进行特征提取处理。
在蛋白质功能注释模型中，使用数量更多的残差卷积模块会出现过拟合的特征，影响预测的准确率，而数量较少的残差卷积模块也会影响预测的准确率；具体的，所述第二预设数量设置在8个左右更为合适；需要说明的是，本公开实施例以8个残差卷积模块进行说明，但这并不构成对所述第而预设数量的限定。
步骤104，将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征。
在本公开实施例中，由于InterPro特征与One-Hot特征、预训练特征的差异性很大，因此将One-Hot特征、预训练特征以及InterPro特征分别输入第二预设数量的残差卷积模块、第一预设数量的残差全连接模块进行特征提取处理；并在蛋白质功能注释模型的网络高层将提取到的第一学习特征和第二学习特征进行融合处理，得到第一融合特征。将这些来自不同特征进行融合，可以弥补不同特征之间的不足，从而提高整体特征的质量，提高蛋白质功能注释模型的预测准确率。
步骤105，对所述第一融合特征进行处理，得到蛋白质功能注释信息。
在本公开实施例中，将所述第一学习特征和所述的第二学习特征简单的融合后，并不能够直接的得到蛋白质的功能注释信息。还需要对第一融合特征再进行一次特征提取，充分挖掘One-Hot特征、预训练特征以及InterPro特征之间的优势，实现特征之间的信息互补。通过对第一融合特征进行特征提取，生成蛋白质功能注释信息。
本公开提供了一种蛋白质功能注释的方法，根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；基于第一预设数量的残差全连接模块，对所述预设注释特征进行特征提取处理，得到第一学习特征；基于第二预设数量的残差卷积模块，对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；对所述第一融合特征进行处理，得到蛋白质功能注释信息。与相关技术相比，本公开基于蛋白质序列获取序列特征、预训练序列特征以及预设注释特征，通过对前述三种不同角度的特征的融合处理，使得获得的蛋白质功能注释信息具有更多的蛋白质功能信息；有效提升了蛋白质功能注释的预测准确率。
为了清楚说明本公开实施例，本公开实施例提供了另一种蛋白质功能注释的方法的流程示意图。
如图2所示，该方法包含以下步骤：
步骤201，根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征。
有关步骤201可参阅上述实施例中的详细说明，故在此不再进行一一赘述。
步骤202，基于第一预设层数的全连接层，对所述预设注释特征进行特征提取处理，得到第三学习特征。
具体在本公开实施例中，提供了一种在网络浅层将多种序列特征融合的蛋白质功能注释模型(MergeNet算法),弥补了使用单一特征的模型算法的缺陷，提高了仅使用One-Hot特征的算法预测准确率，增加了仅使用InterPro特征的算法应用普适性。
具体的，请参阅图3；MergeNet用一层全连接层(MLP)对InterPro特征进行初步的学习，得到学习后的InterPro特征其中d为隐藏层的特征维度，在本实施例中将d设置成1024，并且所有隐藏层的特征维度d均保持一致。
步骤203，基于第二预设层数的卷积层，分别对所述序列特征、所述预训练序列特征进行特征提取处理，得到第四学习特征与第五学习特征。
具体在本公开实施例中，请继续参阅图3；将根据蛋白质序列得到的One-Hot特征H和预训练特征E输入一层一维卷积层(CNN)进行特征的初步学习，得到 其中d为卷积层的特征维度，然后将H和E在残基维度保留最大的特征值以进行最大池化操作，得到和
步骤204，将所述第三学习特征、所述第四学习特征以及所述第五学习特征进行融合处理，得到第二融合特征；其中，所述融合处理为加和处理。
具体在本公开实施例中，MergeNet在网络的浅层(即三种特征均经过一个特征学习层后)将三种特征通过加和处理的方式进行融合，得到蛋白质特征示例性的，加和处理的方式为
步骤205，基于第三预设层数的全连接层，对所述第二融合特征进行分类计算，得到蛋白质功能注释信息。
具体在本公开实施例中，将蛋白质特征P送入全连接层计算出各个功能类别的逻辑概率(蛋白质功能注释信息),其中class为蛋白质功能的总数，在本实施例中class为5901。预测结果示意图如图4所示，图中选取了算法预测出的十个蛋白质的三个功能，其中第一列为预测的蛋白质名称，后面三列为对每个蛋白质所预测出的功能，其中用“|”前面表示算法所预测的基因本体论中功能的编号，“|”后面表示蛋白质具有该功能的概率值。
为了清楚说明本公开实施例，本公开实施例提供了另一种蛋白质功能注释的方法的流程示意图。
如图5所示，该方法包含以下步骤：
步骤301，基于预设编码算法，对所述蛋白质序列进行编码得到所述序列特征。
步骤302，基于预设蛋白质语言模型，对所述序列特征进行预训练处理，得到所述预训练序列特征。
具体在本公开实施例中，所述预设编码算法为One-Hot编码算法，利用One-Hot编码算法对待注释的蛋白质序列进行编码生成One-Hot特征。需要说明的是，本公开实施例并不对所述预设编码算法进行限定。利用蛋白质语言模型ProtBert对One-Hot特征进行预训练处理。蛋白质语言模型可以挖掘蛋白质序列的内部模式信息，并将蛋白质序列的内部模式信息显式地表达出来。
示例性的，请参阅图6，利用One-Hot编码基于序列计算出蛋白质序列的One-Hot特征其中N为蛋白质中氨基酸的数量，20为每个残基的特征维度，也是氨基酸的类别总数。如果该氨基酸对应的是20种氨基酸类别中的第j个，则Hi,j＝1，Hi其他19个位置上的值均为0；将序列的OneHot特征送入预训练模型ProtBert得到预训练特征其中30是预训练特征中残基的特征维度。与One-Hot特征不一样的是，预训练特征E不是由0和1组成的二进制特征矩阵，而是由连续的浮点数组成的特征矩阵，以SwissProt中由八个残基组成的蛋白质SKSP4_ASCTR为例，其序列为：CHYIFNTC， 序列的One-Hot特征和预训练特征如图6所示，其序列OneHot特征的维度为其序列预训练特征的维度为
步骤303，在预设蛋白质数据库中，查找与所述蛋白质序列对应的所述预设注释特征。
具体在本公开实施例中，请参阅图7，图7为ResMergeNet模型算法网络架构示意图。使用InterProScan工具从InterPro数据库检索出与功能类别强相关的蛋白质InterPro特征其中M为蛋白质InterPro特征的维度，即InterPro特征的类别总数。
步骤304，将所述预设注释特征输入第一预设数量的所述残差全连接模块中进行特征提取处理；得到所述第一学习特征。
所述步骤具体包括：将所述预设注释特征输入所述残差全连接模块中的第一个残差全连接模块；将所述第一个残差全连接模块的输出特征作为第二个残差全连接模块的输入特征进行特征提取处理；按照如上步骤循环执行第一预定次数，以获取所述第一学习特征，其中，所述第一预定次数等于所述第一预设数量。
所述将所述预设注释特征输入所述残差全连接模块中的第一个残差全连接模块包括：将所述预设注释特征输入归一化层进行归一化处理，得到归一化处理后的所述预设注释特征；将所述归一化处理后的预设注释特征输入非线性层进行非线性映射，得到非线性映射后的预设注释特征；将所述非线性映射后的预设注释特征输入全连接层进行特征提取，得到第一预设注释特征；将所述第一预设注释特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二预设注释特征；将所述预设注释特征与所述第二预设注释特征进行加和处理，得到第一输出特征。
具体在本公开实施例中，请继续参阅图7，将InterPro特征(M为InterPro特征的类别总数)送入4个顺序连接的残差全连接模块(ResMLP)中进行高层特征的提取。在每个残差全连接模块中，输入特征首先通过归一化层(Norm)进行特征的归一化，然后通过非线性层(ReLU)进行非线性映射，接着通过全连接层(MLP)进行特征的学习，再接着顺序通过归一化层、非线性层和全连接层进行特征的学习，此时将该残差全连接模块的初始输入特征与经过学习的特征进行残差加和，得到高层InterPro特征(第一学习特征)。
步骤305，基于第四预设层数的卷积层，对所述序列特征进行处理，得到第六学习特征、对所述预训练序列特征进行处理，得到第七学习特征。
具体在本公开实施例中，请继续参阅图7，将One-Hot特征与预训练特征分别输入一层一维卷积层进行特征的初步提取，得到和其中N为蛋白质中氨基酸的数量，d为卷积层的特征维度。需要说明的是，本公开实施例以一层一维卷积层为例进行说明，这并不构成对第四预设层数的限定。
步骤306，将所述第六学习特征与所述第七学习特征进行加和处理，得到第三融合特征。
具体在本公开实施例中，在经过一层一维卷积层的学习后，模型将这两个特征进行加和处理，加和处理后的特征为第三融合特征Ps，即
步骤307，将所述第三融合特征输入第二预设数量的所述残差卷积模块中进行特征提取处理，得到所述第二学习特征。
所述步骤具体包括：将所述第三融合特征输入所述残差卷积模块中的第一个残差卷积模块；将所述第一个所述残差卷积模块的输出特征作为第二个所述残差卷积模块的输入特征进行特征提取处理；按照如上步骤循环执行第二预定次数，得到所述第二学习特征，其中，所述第二预定次数等于所述第二预设数量。
所述将所述第三融合特征输入所述残差卷积模块中的第一个残差卷积模块包括：将所述第三融合特征输入归一化层进行归一化处理，得到归一化处理后的第三融合特征；将所述归一化处理后的所述第三融合特征输入非线性层进行非线性映射，得到非线性映射后的第三融合特征；将所述非线性映射后的第三融合特征输入卷积层进行特征学习，得到第二输出特征；将所述第二输出特征依次输入所述归一化层、所述非线性层以及所述卷积层进行处理，得到第三输出特征；将所述第三融合特征与所述第三输出特征进行加和处理，得到第四输出特征。
具体在本公开实施例中，请继续参阅图7，将第三融合特征Ps送入8个顺序连接的残差卷积模块中进行高层特征的提取。在每个残差卷积模块中，输入特征首先通过归一化层(Norm)进行特征的归一化，然后通过非线性层(ReLU)进行非线性映射，接着通过卷积层(CNN)进行特征的学习，再接着顺序通过归一化层、非线性层和卷积层进行特征的学习。在经过所有残差卷积模块的学习后，将第四输出特征在残基维度保留最大的特征值以进行最大池化操作，得到高层序列特征(第二学习特征)。
步骤308，将所述第一学习特征与所述第二学习特征进行拼接处理，得到第一融合特征。
具体在本公开实施例中，高层序列特征Ps(第二学习特征)和高层InterPro特征I(第一学习特征)变得更为抽象，并且特征分布变得相似，均与蛋白质的功能高度相关。ResMergeNet将高层序列特征Ps和高层InterPro特征I进行拼接，得到的特征称为高层蛋白质特征(第一融合特征)，即其中concatenate(,)表示将两个向量按照特征维度进行拼接。
步骤309，将所述第一融合特征输入整合模块进行整合处理，得到蛋白质融合特征。
所述步骤具体包括：将所述第一融合特征输入归一化层进行归一化处理，得到归一化处理后的第一融合特征；将所述归一化处理后的第一融合特征输入非线性层进行非线性映射，得到非线性映射后的第一融合特征；将所述非线性映射后的第一融合特征输入全连接层进行特征提取，得到第一蛋白质融合特征；将所述第一蛋白质融合特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二蛋白质融合特征；将所述第一融合特征与所述第二蛋白质融合特征进行加和处理，得到所述蛋白质融合特征。
具体在本公开实施例中，请继续参阅图7，将高层蛋白质特征(第一融合特征)，输入残差整合模块(Integrate)中进行蛋白质特征的进一步提取。在残差整合模块中，输入特征首先通过归一化层(Norm)进行特征的归一化，然后通过非线性层(ReLU)进行非线性映射，接着通过全连接层(MLP)进行特征的学习，再接着顺序通过归一化层、非线性层和全连接层进行特征的学习，此时将该残基全连接模块的初始输入特征与经过学习的特征进行残差加和，得到(蛋白质融合特征)。
步骤310，基于第五预设层数的全连接层，对所述蛋白质融合特征进行分类处理，得到所述蛋白质功能注释信息。
具体在本公开实施例中，请继续参阅图7，将得到的蛋白质融合特征P送入全连接层(MLP)中进行蛋白质功能的分类，得到蛋白质的功能类别的逻辑概率(蛋白质功能注释信息)。
作为本公开实施例一种可能的实现方式中，在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之前，所述方法还包括：
基于训练用蛋白质序列，对蛋白质功能注释模型进行训练。
具体在本公开实施例中，在使用蛋白质功能注释模型进行蛋白质功能注释之前，将训练用的蛋白质序列输入蛋白质功能注释模型进行训练，以提高模型的预测准确率。
为了清楚说明本公开实施例，本公开实施例提供了一种蛋白质功能注释模型地训练的方法的流程示意图。
如图8所示，该方法包含以下步骤：
步骤401，根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征。
步骤402，对所述训练用预设注释特征进行特征提取处理，得到第一训练用学习特征。
步骤403，对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第二训练用学习特征。
步骤404，将所述第一训练用学习特征与所述第二训练用学习特征进行融合处理，得到第一训练用融合特征。
步骤405，对所述第一训练用融合特征进行处理，得到训练用蛋白质功能注释信息。
有关步骤101至步骤105可参阅上述实施例中的详细说明，故在此不再进行一一赘述。
进一步地，在本实施例一种可能的实现方式中，在根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征之后，本公开实施例提供了另一种蛋白质功能注释模型地训练的方法的流程示意图。
如图9所示，该方法包含以下步骤：
步骤501，根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征。
步骤502，基于第一预设层数的训练用全连接层，对所述训练用预设注释特征进行特征提取处理，得到第三训练用学习特征。
步骤503，基于第二预设层数的训练用卷积层，分别对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第四训练用学习特征与第五训练用学习特征。
步骤504，将所述第三训练用学习特征、所述第四训练用学习特征以及所述第五训练用学习特征进行融合处理，得到第二训练用融合特征；其中，所述融合处理为加和处理。
步骤505，基于第三预设层数的训练用全连接层，对所述第二训练用融合特征进行分类计算，得到训练用蛋白质功能注释信息。
有关步骤201至步骤205可参阅上述实施例中的详细说明，故在此不再进行一一赘述。
与上述的蛋白质功能注释的方法相对应，本发明还提出一种蛋白质功能注释的装置。由于本发明的装置实施例与上述的方法实施例相对应，对于装置实施例中未披露的细节可参照上述的方法实施例，本发明中不再进行赘述。
图10为本公开实施例提供的一种蛋白质功能注释的装置的结构示意图，如图10所示，包括：获取单元61、第一处理单元62、第二处理单元63、第一融合单元64和第三处理单元65。
获取单元61，用于根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；
第一处理单元62，用于对所述预设注释特征进行特征提取处理，得到第一学习特征；
第二处理单元63，用于对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；
第一融合单元64，用于将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；
第三处理单元65，用于对所述第一融合特征进行处理，得到蛋白质功能注释信息。
本公开提供了一种蛋白质功能注释的装置，根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；对所述预设注释特征进行特征提取处理，得到第一学习特征对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；对所述第一融合特征进行处理，得到蛋白质功能注释信息。与相关技术相比，本公开基于蛋白质序列获取序列特征、预训练序列特征以及预设注释特征，通过对前述三种不同角度的特征的融合处理，使得获得的蛋白质功能注释信息更加准确；有效提升了蛋白质功能注释的预测准确率。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述装置还包括：
第四处理单元66，用于在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之后，基于第一预设层数的全连接层，对所述预设注释特征进行特征提取处理，得到第三学习特征；
第五处理单元67，用于基于第二预设层数的卷积层，分别对所述序列特征、所述预训练序列特征进行特征提取处理，得到第四学习特征与第五学习特征；
第二融合单元68，用于将所述第三学习特征、所述第四学习特征以及所述第五学习特征进行融合处理，得到第二融合特征；其中，所述融合处理为加和处理；
第六处理单元69，用于基于第三预设层数的全连接层，对所述第二融合特征进行分类计算，得到蛋白质功能注释信息。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述获取单元61包括：
编码模块611，用于基于预设编码算法，对所述蛋白质序列进行编码得到所述序列特征；
预训练模块612，用于基于预设蛋白质语言模型，对所述序列特征进行预训练处理，得到所述预训练序列特征；
查找模块613，用于在预设蛋白质数据库中，查找与所述蛋白质序列对应的所述预设注释特征。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述第一处理单元62包括：
第一输入模块621，用于将所述预设注释特征输入第一预设数量的残差全连接模块；
第一处理模块622，用于将所述第一个残差全连接模块的输出特征作为第二个残差全连接模块的输入特征进行特征提取处理；
第一循环模块623，用于按照第一处理模块的处理方法循环执行第一预定次数，以获取所述第一学习特征，其中，所述第一预定次数等于所述第一预设数量。
进一步地，在本实施例一种可能的实现方式中，所述第一输入模块621还用于：
将所述预设注释特征输入归一化层进行归一化处理，得到归一化处理后的所述预设注释特征；
将所述归一化处理后的预设注释特征输入非线性层进行非线性映射，得到非线性映射后的预设注释特征；
将所述非线性映射后的预设注释特征输入全连接层进行特征提取，得到第一预设注释特征；
将所述第一预设注释特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二预设注释特征；
将所述预设注释特征与所述第二预设注释特征进行加和处理，得到第一输出特征。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述第二处理单元63包括：
第二处理模块631，用于基于第四预设层数的卷积层，对所述序列特征进行处理，得到第六学习特征、对所述预训练序列特征进行处理，得到第七学习特征；
第一加和模块632，用于将所述第六学习特征与所述第七学习特征进行加和处理，得到第三融合特征；
第二输入模块633，用于将所述第三融合特征输入第二预设数量的残差卷积模块；
第一提取模块634，用于将所述第一个所述残差卷积模块的输出特征作为第二个所述残差卷积模块的输入特征进行特征提取处理；
第二循环模块635，用于按照第一提取模块的处理方法循环执行第二预定次数，得到所述第二学习特征，其中，所述第二预定次数等于所述第二预设数量。
进一步地，在本实施例一种可能的实现方式中，所述第二输入模块633还用于：
将所述第三融合特征输入归一化层进行归一化处理，得到归一化处理后的第三融合特征；
将所述归一化处理后的所述第三融合特征输入非线性层进行非线性映射，得到非线性映射后的第三融合特征；
将所述非线性映射后的第三融合特征输入卷积层进行特征学习，得到第二输出特征；
将所述第二输出特征依次输入所述归一化层、所述非线性层以及所述卷积层进行处理，得到第三输出特征；
将所述第三融合特征与所述第三输出特征进行加和处理，得到第四输出特征。
进一步地，在本实施例一种可能的实现方式中，所述第一融合单元64还用于：
将所述第一学习特征与所述第二学习特征进行拼接处理，得到第一融合特征。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述第三处理单元65包括：
整合模块651，用于将所述第一融合特征输入整合模块进行整合处理，得到蛋白质融合特征；
分类模块652，用于基于第五预设层数的全连接层，对所述蛋白质融合特征进行分类处理，得到所述蛋白质功能注释信息。
进一步地，在本实施例一种可能的实现方式中，所述整合模块651还用于：
将所述第一融合特征输入归一化层进行归一化处理，得到归一化处理后的第一融合特征；
将所述归一化处理后的第一融合特征输入非线性层进行非线性映射，得到非线性映射后的第一融合特征；
将所述非线性映射后的第一融合特征输入全连接层进行特征提取，得到第一蛋白质融合特征；
将所述第一蛋白质融合特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二蛋白质融合特征；
将所述第一融合特征与所述第二蛋白质融合特征进行加和处理，得到所述蛋白质融合特征。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述装置还包括：
训练单元610，用于在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之前，基于训练用蛋白质序列，对蛋白质功能注释模型进行训练。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述训练单元610包括：
训练用获取模块6101，用于根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征；
训练用第一处理模块6102，用于对所述训练用预设注释特征进行特征提取处理，得到第一训练用学习特征；
训练用第二处理模块6103，用于对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第二训练用学习特征；
训练用第一融合模块6104，用于将所述第一训练用学习特征与所述第二训练用学习特征进行融合处理，得到第一训练用融合特征；
训练用第三处理模块6105，用于对所述第一训练用融合特征进行处理，得到训练用蛋白质功能注释信息。
进一步地，在本实施例一种可能的实现方式中，如图11所示，所述训练单元610还包括：
训练用第四处理模块6106，用于在根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征之后，基于第一预设层数的训练用全连接层，对所述训练用预设注释特征进行特征提取处理，得到第三训练用学习特征；
训练用第五处理模块6107，用于基于第二预设层数的训练用卷积层，分别对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第四训练用学习特征与第五训练用学习特征；
训练用第二融合模块6108，用于将所述第三训练用学习特征、所述第四训练用学习特征以及所述第五训练用学习特征进行融合处理，得到第二训练用融合特征；其中，所述融合处理为加和处理；
训练用第六处理模块6109，用于基于第三预设层数的训练用全连接层，对所述第二训练用融合特征进行分类计算，得到训练用蛋白质功能注释信息。
需要说明的是，前述对方法实施例的解释说明，也适用于本实施例的装置，原理相同，本实施例中不再限定。
根据本公开的实施例，本公开还提供了一种电子设备、一种可读存储介质和一种计算机程序产品。
图12示出了可以用来实施本公开的实施例的示例电子设备700的示意性框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本公开的实现。
如图12所示，设备700包括计算单元701，其可以根据存储在ROM(Read-Only Memory，只读存储器)702中的计算机程序或者从存储单元708加载到RAM(Random Access Memory，随机访问/存取存储器)703中的计算机程序，来执行各种适当的动作和处理。在RAM 703中，还可存储设备700操作所需的各种程序和数据。计算单元701、ROM 702以及RAM 703通过总线704彼此相连。I/O(Input/Output，输入/输出)接口705也连接至总线704。
设备700中的多个部件连接至I/O接口705，包括：输入单元706，例如键盘、鼠标等；输出单元707，例如各种类型的显示器、扬声器等；存储单元708，例如磁盘、光盘等；以及通信单元709，例如网卡、调制解调器、无线通信收发机等。通信单元709允许设备700通过诸如因特网的计算机网络和/或各种电信网络与其他设备交换信息/数据。
计算单元701可以是各种具有处理和计算能力的通用和/或专用处理组件。计算单元701的一些示例包括但不限于CPU(Central Processing Unit，中央处理单元)、GPU(Graphic Processing Units，图形处理单元)、各种专用的AI(Artificial Intelligence，人工智能)计算芯片、各种运行机器学习模型算法的计算单元、DSP(Digital Signal Processor，数字信号处理器)、以及任何适当的处理器、控制器、微控制器等。计算单元701执行上文所描述的各个方法和处理，例如蛋白质功能注释的方法。例如，在一些实施例中，蛋白质功能注释的方法可被实现为计算机软件程序，其被有形地包含于机器可读介质，例如存储单元708。在一些实施例中，计算机程序的部分或者全部可以经由ROM 702和/或通信单元709而被载入和/或安装到设备700上。当计算机程序加载到RAM 703并由计算单元701执行时，可以执行上文描述的方法的一个或多个步骤。备选地，在其他实施例中，计算单元701可以通过其他任何适当的方式(例如，借助于固件)而被配置为执行前述蛋白质功能注释的方法。
本文中以上描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、FPGA(Field Programmable Gate Array，现场可编程门阵列)、ASIC(Application-Specific Integrated Circuit，专用集成电路)、ASSP(Application Specific Standard Product，专用标准产品)、SOC(System On Chip，芯片上系统的系统)、CPLD(Complex Programmable Logic Device，复杂可编程逻辑设备)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至该存储系统、该至少一个输入装置、和该至少一个输出装置。
用于实施本公开的方法的程序代码可以采用一个或多个编程语言的任何组合来编写。这些程序代码可以提供给通用计算机、专用计算机或其他可编程数据处理装置的处理器或控制器，使得程序代码当由处理器或控制器执行时使流程图和/或框图中所规定的功能/操作被实施。程序代码可以完全在机器上执行、部分地在机器上执行，作为独立软件包部分地在机器上执行且部分地在远程机器上执行或完全在远程机器或服务器上执行。
在本公开的上下文中，机器可读介质可以是有形的介质，其可以包含或存储以供指令执行系统、装置或设备使用或与指令执行系统、装置或设备结合地使用的程序。机器可读介质可以是机器可读信号介质或机器可读储存介质。机器可读介质可以包括但不限于电子的、磁性的、光学的、电磁的、红外的、或半导体系统、装置或设备，或者上述内容的任何合适组合。机器可读存储介质的更具体示例会包括基于一个或多个线的电气连接、便携式计算机盘、硬盘、RAM、ROM、EPROM(Electrically Programmable Read-Only-Memory，可擦除可编程只读存储器)或快闪存储器、光纤、CD-ROM(Compact Disc Read-Only Memory，便捷式紧凑盘只读存储器)、光学储存设备、磁储存设备、或上述内容的任何合适组合。
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(Cathode-Ray Tube，阴极射线管)或者LCD(LiquidCrystal Display，液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如，作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：LAN(Local Area Network，局域网)、WAN(Wide Area Network，广域网)、互联网和区块链网络。
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。服务器可以是云服务器，又称为云计算服务器或云主机，是云计算服务体系中的一项主机产品，以解决了传统物理主机与VPS服务("Virtual Private Server"，或简称"VPS")中，存在的管理难度大，业务扩展性弱的缺陷。服务器也可以为分布式系统的服务器，或者是结合了区块链的服务器。
其中，需要说明的是，人工智能是研究使计算机来模拟人的某些思维过程和智能行为(如学习、推理、思考、规划等)的学科，既有硬件层面的技术也有软件层面的技术。人工智能硬件技术一般包括如传感器、专用人工智能芯片、云计算、分布式存储、大数据处理等技术；人工智能软件技术主要包括计算机视觉技术、语音识别技术、自然语言处理技术以及机器学习/深度学习、大数据处理技术、知识图谱技术等几大方向。
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本公开中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本公开公开的技术方案所期望的结果，本文在此不进行限制。
上述具体实施方式，并不构成对本公开保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本公开的精神和原则之内所作的修改、等同替换和改进等，均应包含在本公开保护范围之内。

Claims (17)

  1. 一种蛋白质功能注释的方法，其特征在于，包括：
    根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；
    对所述预设注释特征进行特征提取处理，得到第一学习特征；
    对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；
    将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；
    对所述第一融合特征进行处理，得到蛋白质功能注释信息。
  2. 根据权利要求1所述的方法，其特征在于，在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之后，所述方法还包括：
    基于第一预设层数的全连接层，对所述预设注释特征进行特征提取处理，得到第三学习特征；
    基于第二预设层数的卷积层，分别对所述序列特征、所述预训练序列特征进行特征提取处理，得到第四学习特征与第五学习特征；
    将所述第三学习特征、所述第四学习特征以及所述第五学习特征进行融合处理，得到第二融合特征；其中，所述融合处理为加和处理；
    基于第三预设层数的全连接层，对所述第二融合特征进行分类计算，得到蛋白质功能注释信息。
  3. 根据权利要求1所述的方法，其特征在于，所述根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征包括：
    基于预设编码算法，对所述蛋白质序列进行编码得到所述序列特征；
    基于预设蛋白质语言模型，对所述序列特征进行预训练处理，得到所述预训练序列特征；
    在预设蛋白质数据库中，查找与所述蛋白质序列对应的所述预设注释特征。
  4. 根据权利要求1所述的方法，其特征在于，所述对所述预设注释特征进行特征提取处理，得到第一学习特征包括如下步骤：
    将所述预设注释特征输入第一预设数量的残差全连接模块；
    将第一个残差全连接模块的输出特征作为第二个残差全连接模块的输入特征进行特征提取处理；
    按照如上步骤循环执行第一预定次数，以获取所述第一学习特征，其中，所述第一预定次数等于所述第一预设数量。
  5. 根据权利要求4所述的方法，其特征在于，所述将所述预设注释特征输入所述残差全连接模块中的第一个残差全连接模块包括：
    将所述预设注释特征输入归一化层进行归一化处理，得到归一化处理后的所述预设注释特征；
    将所述归一化处理后的预设注释特征输入非线性层进行非线性映射，得到非线性映射后的预设注释特征；
    将所述非线性映射后的预设注释特征输入全连接层进行特征提取，得到第一预设注释特征；
    将所述第一预设注释特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二预设注释特征；
    将所述预设注释特征与所述第二预设注释特征进行加和处理，得到第一输出特征。
  6. 根据权利要求3所述的方法，其特征在于，基于第四预设层数的卷积层，对所述序列特征进行处理，得到第六学习特征、对所述预训练序列特征进行处理，得到第七学习特征；
    将所述第六学习特征与所述第七学习特征进行加和处理，得到第三融合特征；
    所述对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征包括如下步骤：
    将所述第三融合特征输入第二预设数量的残差卷积模块；
    将第一个所述残差卷积模块的输出特征作为第二个所述残差卷积模块的输入特征进行特征提取处理；
    按照如上步骤循环执行第二预定次数，得到所述第二学习特征，其中，所述第二预定次数等于所述第二预设数量。
  7. 根据权利要求6所述的方法，其特征在于，所述将所述第三融合特征输入所述残差卷积模块中的第一个残差卷积模块包括：
    将所述第三融合特征输入归一化层进行归一化处理，得到归一化处理后的第三融合特征；
    将所述归一化处理后的所述第三融合特征输入非线性层进行非线性映射，得到非线性映射后的第三融合特征；
    将所述非线性映射后的第三融合特征输入卷积层进行特征学习，得到第二输出特征；
    将所述第二输出特征依次输入所述归一化层、所述非线性层以及所述卷积层进行处理，得到第三输出特征；
    将所述第三融合特征与所述第三输出特征进行加和处理，得到第四输出特征。
  8. 根据权利要求1所述的方法，其特征在于，所述将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征包括：
    将所述第一学习特征与所述第二学习特征进行拼接处理，得到第一融合特征。
  9. 根据权利要求1所述的方法，其特征在于，所述对所述第一融合特征进行处理，得到蛋白质功能注释信息包括：
    将所述第一融合特征输入整合模块进行整合处理，得到蛋白质融合特征；
    基于第五预设层数的全连接层，对所述蛋白质融合特征进行分类处理，得到所述蛋白质功能注释信息。
  10. 根据权利要求9所述的方法，其特征在于，所述将所述第一融合特征输入整合模块进行整合处理，得到蛋白质融合特征包括：
    将所述第一融合特征输入归一化层进行归一化处理，得到归一化处理后的第一融合特征；
    将所述归一化处理后的第一融合特征输入非线性层进行非线性映射，得到非线性映射后的第一融合特征；
    将所述非线性映射后的第一融合特征输入全连接层进行特征提取，得到第一蛋白质融合特征；
    将所述第一蛋白质融合特征依次输入所述归一化层、所述非线性层以及所述全连接层进行处理，得到第二蛋白质融合特征；
    将所述第一融合特征与所述第二蛋白质融合特征进行加和处理，得到所述蛋白质融合特征。
  11. 根据权利要求1所述的方法，其特征在于，在根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征之前，所述方法还包括：
    基于训练用蛋白质序列，对蛋白质功能注释模型进行训练。
  12. 根据权利要求11所述方法，其特征在于，所述将训练用蛋白质序列输入蛋白质功能注释模型进行训练包括：
    根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征；
    对所述训练用预设注释特征进行特征提取处理，得到第一训练用学习特征；
    对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第二训练用学习特征；
    将所述第一训练用学习特征与所述第二训练用学习特征进行融合处理，得到第一训练用融合特征；
    对所述第一训练用融合特征进行处理，得到训练用蛋白质功能注释信息。
  13. 根据权利要求12所述的方法，其特征在于，在根据训练用蛋白质序列，分别获取训练用序列特征、训练用第一序列特征以及训练用预设注释特征之后，所述方法还包括：
    基于第一预设层数的训练用全连接层，对所述训练用预设注释特征进行特征提取处理，得到第三训练用学习特征；
    基于第二预设层数的训练用卷积层，分别对所述训练用序列特征、所述训练用第一序列特征进行特征提取处理，得到第四训练用学习特征与第五训练用学习特征；
    将所述第三训练用学习特征、所述第四训练用学习特征以及所述第五训练用学习特征进行融合处理，得到第二训练用融合特征；其中，所述融合处理为加和处理；
    基于第三预设层数的训练用全连接层，对所述第二训练用融合特征进行分类计算，得到训练用蛋白质功能注释信息。
  14. 一种蛋白质功能注释的装置，其特征在于，包括：
    获取单元，用于根据蛋白质序列，分别获取序列特征、预训练序列特征以及预设注释特征；
    第一处理单元，用于对所述预设注释特征进行特征提取处理，得到第一学习特征；
    第二处理单元，用于对所述序列特征、所述预训练序列特征进行特征提取处理，得到第二学习特征；
    第一融合单元，用于将所述第一学习特征与所述第二学习特征进行融合处理，得到第一融合特征；
    第三处理单元，用于对所述第一融合特征进行处理，得到蛋白质功能注释信息。
  15. 一种电子设备，其特征在于，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-13中任一项所述的方法。
  16. 一种存储有计算机指令的非瞬时计算机可读存储介质，其特征在于，所述计算机指令用于使所述计算机执行根据权利要求1-13中任一项所述的方法。
  17. 一种计算机程序产品，其特征在于，包括计算机程序，所述计算机程序在被处理器执行时实现根据权利要求1-13中任一项所述的方法。
PCT/CN2023/099729 2023-06-12 2023-06-12 蛋白质功能注释的方法及装置、电子设备和存储介质 Ceased WO2024254739A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202380094803.5A CN120731460A (zh) 2023-06-12 2023-06-12 蛋白质功能注释的方法及装置、电子设备和存储介质
PCT/CN2023/099729 WO2024254739A1 (zh) 2023-06-12 2023-06-12 蛋白质功能注释的方法及装置、电子设备和存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2023/099729 WO2024254739A1 (zh) 2023-06-12 2023-06-12 蛋白质功能注释的方法及装置、电子设备和存储介质

Publications (1)

Publication Number Publication Date
WO2024254739A1 true WO2024254739A1 (zh) 2024-12-19

Family

ID=93851191

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/099729 Ceased WO2024254739A1 (zh) 2023-06-12 2023-06-12 蛋白质功能注释的方法及装置、电子设备和存储介质

Country Status (2)

Country Link
CN (1) CN120731460A (zh)
WO (1) WO2024254739A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110070909A (zh) * 2019-03-21 2019-07-30 中南大学 一种基于深度学习的融合多特征的蛋白质功能预测方法
CN112562790A (zh) * 2020-12-09 2021-03-26 中国石油大学(华东) 基于深度学习调控疾病靶点的中药分子推荐系统、计算机设备、存储介质
US20220172055A1 (en) * 2019-04-11 2022-06-02 Google Llc Predicting biological functions of proteins using dilated convolutional neural networks
CN114898811A (zh) * 2022-05-26 2022-08-12 清华大学 蛋白质训练模型的训练方法及装置、电子设备和存储介质
WO2022185179A1 (en) * 2021-03-02 2022-09-09 Glaxosmithkline Biologicals Sa Natural language processing to predict properties of proteins
CN116092577A (zh) * 2023-01-09 2023-05-09 中国海洋大学 一种基于多源异质信息聚合的蛋白质功能预测方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110070909A (zh) * 2019-03-21 2019-07-30 中南大学 一种基于深度学习的融合多特征的蛋白质功能预测方法
US20220172055A1 (en) * 2019-04-11 2022-06-02 Google Llc Predicting biological functions of proteins using dilated convolutional neural networks
CN112562790A (zh) * 2020-12-09 2021-03-26 中国石油大学(华东) 基于深度学习调控疾病靶点的中药分子推荐系统、计算机设备、存储介质
WO2022185179A1 (en) * 2021-03-02 2022-09-09 Glaxosmithkline Biologicals Sa Natural language processing to predict properties of proteins
CN114898811A (zh) * 2022-05-26 2022-08-12 清华大学 蛋白质训练模型的训练方法及装置、电子设备和存储介质
CN116092577A (zh) * 2023-01-09 2023-05-09 中国海洋大学 一种基于多源异质信息聚合的蛋白质功能预测方法

Also Published As

Publication number Publication date
CN120731460A (zh) 2025-09-30

Similar Documents

Publication Publication Date Title
JP7403605B2 (ja) マルチターゲット画像テキストマッチングモデルのトレーニング方法、画像テキスト検索方法と装置
CN114612759B (zh) 视频处理方法、查询视频的方法和模型训练方法、装置
WO2023138188A1 (zh) 特征融合模型训练及样本检索方法、装置和计算机设备
CN112800774B (zh) 基于注意力机制的实体关系抽取方法、装置、介质和设备
JP7495543B2 (ja) 事前トレーニングモデルのプロンプトベクトルの決定方法、装置及び電子機器
WO2025020611A1 (zh) 会话应答方法和装置、电子设备、存储介质
CN117931858B (zh) 数据查询方法、装置、计算机设备和存储介质
CN117874234A (zh) 基于语义的文本分类方法、装置、计算机设备及存储介质
JP2022117941A (ja) イメージ検索方法、装置、電子機器、及びコンピュータ読み取り可能な記憶媒体
CN116662633A (zh) 搜索方法、模型训练方法、装置、电子设备及存储介质
WO2025152309A1 (zh) 一种数据检索的方法及网络附加存储设备
CN117194710A (zh) 多粒度视频检索方法及装置
CN116796038A (zh) 遥感数据检索方法、装置、边缘处理设备及存储介质
WO2025162037A1 (zh) 语音识别模型的训练方法、装置、电子设备及存储介质
WO2024152550A1 (zh) 图片处理方法、装置、电子设备及存储介质
US20220318503A1 (en) Method and apparatus for identifying instruction, and screen for voice interaction
CN114419327B (zh) 图像检测方法和图像检测模型的训练方法、装置
CN118823489A (zh) 基于多尺度注意力机制的图像分类方法、装置和设备
CN115994922B (zh) 运动分割方法、装置、电子设备及存储介质
WO2024254739A1 (zh) 蛋白质功能注释的方法及装置、电子设备和存储介质
CN114943217B (zh) 一种合同风险识别方法、装置、设备及存储介质
WO2021072892A1 (zh) 基于神经网络混合模型的法律条文检索方法及相关设备
CN118840756A (zh) 文本检测及文本检测模型训练方法、装置和设备
CN114494782B (zh) 图像处理方法、模型训练方法、相关装置及电子设备
CN116303917A (zh) 一种数据的检索方法、装置及设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23940977

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 202380094803.5

Country of ref document: CN

WWP Wipo information: published in national office

Ref document number: 202380094803.5

Country of ref document: CN

NENP Non-entry into the national phase

Ref country code: DE