A protein function prediction method, device, terminal equipment and medium

Abstract

This application is applicable to the field of bioinformatics technology and provides a protein function prediction method, apparatus, terminal device, and medium. The method comprises obtaining the sequence of a target protein, the sequence of an interacting protein, the sequence of a homologous protein, and a structural template of the target protein to construct a multiple sequence alignment; constructing a three-dimensional structure based on the multiple sequence alignment and the structural template; extracting sequence features of the target protein and the interacting protein to obtain sequence features of the target protein and the interacting protein; mixing the sequence features using a protein mixing layer to obtain a first mixed feature; mixing the sequence features of the target protein with the first mixed feature to obtain a second mixed feature; extracting structural features of the three-dimensional structure; obtaining a function confidence score based on the second mixed feature and the structural feature; and predicting the function of the target protein based on the function confidence score. This application can improve the accuracy of protein function prediction.

Description

Technical Field

[0001] The present application belongs to the field of bioinformatics technology, and in particular relates to a protein function prediction method, apparatus, terminal device, and medium. Background Art

[0002] Proteins, as biological macromolecules, are indispensable components of all cells and tissues. They play an important role in the growth, development, movement, heredity, reproduction and other life activities of organisms. Understanding their functions is of great significance for understanding the behavior of biological systems and finding drug targets.

[0003] With the development of genome sequencing and structural genomics technologies, a large number of experimental protein sequences and three-dimensional structures have been accumulated. However, the gap between the large number of identified proteins and the completeness of their annotations continues to widen. As of November 2022, nearly 463,700 (about 84%) protein sequences deposited in the protein database UniProtKB / Swissprot have no manually annotated functions; the same is true for many structures in the Protein Data Bank (PDB), with nearly 156,300 (about 74%) structures lacking manually annotated functions. Even if the functions automatically annotated by computers are included, there are still nearly 111,900 (about 53%) protein structures without functional annotations. Currently, protein function prediction mainly relies on experimental identification and manual work, which greatly reduces the accuracy of protein function prediction and is very cumbersome. Summary of the Invention

[0004] The embodiments of the present application provide a protein function prediction method, apparatus, terminal device, and medium, which can solve the problem of low accuracy of current protein function prediction methods.

[0005] In a first aspect, the present invention provides a method for predicting protein function, comprising:

[0006] Obtaining the sequence of the target protein, the sequence of the interacting protein that interacts with the target protein, the sequence of the homologous protein that is homologous to the target protein, and the structural template of the target protein, and constructing a multiple sequence alignment based on the sequence of the homologous protein;

[0007] Construct the three-dimensional structure of the target protein based on multiple sequence alignment and structural template;

[0008] Performing sequence feature extraction on the target protein sequence and the interacting protein sequence to obtain the sequence features of the target protein and the interacting protein sequence;

[0009] Using a protein mixing layer for mixing protein sequence features, the sequence features of the target protein and the sequence features of the interacting protein are mixed to obtain a first mixed feature of the target protein;

[0010] Mixing the sequence feature of the target protein and the first mixed feature to obtain a second mixed feature of the target protein;

[0011] Extract structural features of the three-dimensional structure to obtain the structural features of the target protein;

[0012] Obtaining a functional confidence score of the target protein based on the second mixed feature and the structural feature;

[0013] The function of the target protein is predicted based on the functional confidence score.

[0014] Optionally, the protein mixing layer comprises a plurality of mixing treatment layers.

[0015] Optionally, each of the multiple hybrid processing layers includes a multi-head attention layer, a feedforward neural network layer, a first regularized residual connection layer, and a second regularized residual connection layer; wherein, the input end of the multi-head attention layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the multi-head attention layer is connected to the first input end of the first regularized residual connection layer, the second input end of the first regularized residual connection layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the first regularized residual connection layer is connected to the input end of the feedforward neural network layer, the output end of the feedforward neural network layer is connected to the first input end of the second regularized residual connection layer, and the second input end of the second regularized residual connection layer receives the data output by the output end of the first regularized residual connection layer.

[0016] Optionally, for the u-th hybrid processing layer, u = 1, 2, ..., O-1, O represents the total number of hybrid processing layers, and the output end of the second regularized residual connection layer of the hybrid processing layer is respectively connected to the input end of the multi-head attention layer of the u+1-th hybrid processing layer and the first input end of the first regularized residual connection layer of the second hybrid processing layer.

[0017] Optionally, for the Oth hybrid processing layer, the output end of the second regularized residual connection layer of the hybrid processing layer outputs the first hybrid feature.

[0018] Optionally, using a protein mixing layer for mixing protein sequence features to mix the sequence features of the target protein and the sequence features of the interacting protein to obtain a first mixed feature of the target protein includes:

[0019] For each mixed processing layer, the formula

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027] Get the first mixed feature in, A set of sequence features consisting of the sequence features of the target protein and the sequence features of the interacting proteins, p'0 represents the fixed-dimensional sequence feature of the target protein, p' b represents the fixed-dimensional sequence feature of the b-th interacting protein, b=1,2,...,m, m represents the total number of interacting proteins, a i represents the i-th sequence of the b-th interacting protein, i=1,2,...,n, n represents the length of the protein sequence, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the query space Q, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the key space K, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the value space V, W o Indicates that the feature dimension of the head attention is transformed to Learnable linear projections of the same feature dimension, W1 means A learnable linear projection to a high-dimensional feature space, W2 means A learnable linear projection that recovers from the high-dimensional feature space to the original feature space, h represents the number of heads in the multi-head attention layer, d k =d v =d model / h=1280, b1 and b2 are both bias terms, MultiHead(·) represents the multi-head attention mechanism calculation function, FusionLayer(·) represents the protein mixing layer calculation function, AddAndNorm(·) represents the regularized residual connection layer, LayerNorm(·) represents the layer normalization calculation function, FFN(·) represents the feedforward neural network calculation function, ReLU(·) represents the linear rectification function, Concat(·) represents the feature connection calculation function, Attention(·) represents the attention mechanism calculation function, Represents the field of real numbers.

[0028] Optionally, the sequence feature of the target protein and the first mixed feature are mixed to obtain a second mixed feature of the target protein, including:

[0029] By calculating the formula Get the second mixed feature v of the target protein sp ;

[0030] Optionally, structural feature extraction is performed on the three-dimensional structure to obtain the structural features of the target protein, including:

[0031] Based on the three-dimensional structure, a k-nearest neighbor graph is constructed; the k-nearest neighbor graph is composed of the coordinate nodes of the Cα atoms of each residue in the three-dimensional structure of the target protein connected to the adjacent nodes; the three-dimensional structure includes multiple residues;

[0032] For each residue in the k-nearest neighbor graph, an initial scalar feature s and an initial vector feature V are constructed to obtain multiple groups of initial residue features; each group of initial residue features includes an initial scalar feature and an initial vector feature corresponding to a residue; wherein, n represents the number of constructed scalar features, and v represents the number of constructed vector features:

[0033] For each residue, the formula

[0034] s′=σ((Concat(s,||V h ||2)W m +b)

[0035] V′=σ + (||V h W μ ||2)⊙V h W μ

[0036] V h =VW h

[0037] Obtain the intermediate scalar feature s′ and intermediate vector feature V′ corresponding to the residue; W h represents a learnable linear projection of the feature space transformation for the initial vector features, W m represents a learnable linear projection for feature space transformation of fused features containing scalar features and vector features, W μ represents a learnable linear projection for the feature space transformation of vector features, b represents the bias term, σ and σ + Both represent activation functions, ||·||2 represents L2 regularization, and μ represents the number of new vector features obtained;

[0038] Based on the graph neural network model, the intermediate scalar feature s′ and the intermediate vector feature V′ are combined to obtain the structural feature v of the target protein. s .

[0039] Optionally, a functional confidence score of the target protein is obtained based on the second mixed feature and the structural feature, including:

[0040] By calculating the formula

[0041] v target =Concat(v sp , GMP(v s ))

[0042]

[0043] s p =Sigmoid((v target W1+b1)W2+b2)

[0044] Get the functional confidence score s of the target protein p ;in, s k represents the confidence score of the kth function of the target protein, k = 1, 2, ..., n, n represents the total number of protein functions, b1 and b2 both represent bias terms, and GMP(·) represents the global averaging layer calculation function.

[0045] Optionally, before using a protein mixing layer for mixing protein sequence features to mix the sequence features of the target protein and the sequence features of the interacting protein to obtain a first mixed feature of the target protein, the protein function prediction method provided by the present application further includes:

[0046] For any protein among the target protein and the interacting protein, the formula

[0047]

[0048] Get the fixed dimension sequence feature p′ of the protein a , a=0,1,2,...,b,...,m, b=1,2,...,m, m represents the total number of interacting proteins, p′0 represents the fixed-dimensional sequence features of the target protein, p′ b represents the fixed-dimensional sequence features of the b-th interacting protein.

[0049] In a second aspect, the present invention provides a protein function prediction device, comprising:

[0050] An acquisition module is used to obtain the sequence of the target protein, the sequence of the interacting protein that interacts with the target protein, the sequence of the homologous protein that is homologous to the target protein, and the structural template of the target protein, and to construct a multiple sequence alignment based on the sequence of the homologous protein;

[0051] The three-dimensional structure module is used to construct the three-dimensional structure of the target protein based on multiple sequence alignment and structure templates;

[0052] A sequence feature module is used to extract sequence features of the target protein sequence and the interacting protein sequence to obtain the sequence features of the target protein and the interacting protein sequence;

[0053] a first mixing module, configured to mix the sequence features of the target protein and the sequence features of the interacting protein using a protein mixing layer for mixing protein sequence features, to obtain a first mixed feature of the target protein;

[0054] A second mixing module is used to mix the sequence feature of the target protein and the first mixed feature to obtain a second mixed feature of the target protein;

[0055] Structural feature module, used to extract structural features of three-dimensional structures and obtain the structural features of target proteins;

[0056] a confidence score calculation module, configured to obtain a functional confidence score of the target protein based on the second mixed feature and the structural feature;

[0057] The prediction module is used to predict the function of the target protein based on the function confidence score.

[0058] In a third aspect, an embodiment of the present application provides a terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned protein function prediction method when executing the computer program.

[0059] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned protein function prediction method.

[0060] The above solution of the present application has the following beneficial effects:

[0061] In some embodiments of the present application, the sequence of the target protein, the sequence of the interacting protein, the sequence of the homologous protein and the structural template of the target protein are obtained, and a multiple sequence alignment is constructed based on the sequence of the homologous protein. Then, the three-dimensional structure of the target protein is constructed based on the multiple sequence alignment and the structural template. Then, sequence feature extraction is performed on the target protein sequence and the interacting protein sequence to obtain the sequence feature of the target protein and the sequence feature of the interacting protein. Then, the sequence feature of the target protein and the sequence feature of the interacting protein are mixed using a protein mixing layer to obtain a first mixed feature of the target protein. Then, the sequence feature of the target protein and the first mixed feature are mixed to obtain a second mixed feature of the target protein. Subsequently, structural feature extraction is performed on the three-dimensional structure of the target protein to obtain the structural feature of the target protein. Then, based on the second mixed feature and the structural feature, a target protein function confidence score is obtained. Finally, based on the target protein function confidence score, the target protein function is predicted. Among them, the sequence features of the target protein and the sequence features of the interacting protein are mixed using a protein mixing layer to obtain a first mixed feature, and then the sequence features of the target protein and the first mixed feature are mixed to obtain a second mixed feature. This can avoid the target protein sequence features being diluted by the model, and at the same time combine multiple protein features to improve the accuracy of protein function prediction.

[0062] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0064] Figure 1A flowchart of a protein function prediction method provided in one embodiment of the present application;

[0065] Figure 2 This is a graph showing protein function prediction results provided in one embodiment of the present application;

[0066] Figure 3 A schematic diagram of the structure of a protein function prediction device provided in one embodiment of the present application;

[0067] Figure 4 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0068] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0069] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0070] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0071] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0072] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0073] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0074] In response to the problem of low accuracy of current protein function prediction methods, the present application provides a protein function prediction method, apparatus, terminal device and medium, which obtains the sequence of the target protein, the sequence of the interacting protein, the sequence of the homologous protein and the structural template of the target protein, and constructs a multiple sequence alignment based on the sequence of the homologous protein, and then constructs the three-dimensional structure of the target protein based on the multiple sequence alignment and the structural template, and then extracts sequence features of the target protein sequence and the interacting protein sequence to obtain the sequence features of the target protein and the sequence features of the interacting protein, and then uses a protein mixing layer to mix the sequence features of the target protein and the sequence features of the interacting protein to obtain a first mixed feature of the target protein, and then mix the sequence features of the target protein and the first mixed feature to obtain a second mixed feature of the target protein, and then extracts structural features of the three-dimensional structure of the target protein to obtain the structural features of the target protein, and then obtains the target protein function confidence score based on the second mixed feature and the structural feature, and finally predicts the target protein function based on the target protein function confidence score. Among them, the sequence features of the target protein and the sequence features of the interacting protein are mixed using a protein mixing layer to obtain a first mixed feature, and then the sequence features of the target protein and the first mixed feature are mixed to obtain a second mixed feature. This can avoid the target protein sequence features being diluted by the model, and at the same time combine multiple protein features to improve the accuracy of protein function prediction.

[0075] like Figure 1 As shown, the protein function prediction method provided in this application mainly includes the following steps:

[0076] Step 11: Obtain the sequence of the target protein, the sequence of the interacting protein that interacts with the target protein, the sequence of the homologous protein that is homologous to the target protein, and the structural template of the target protein, and construct a multiple sequence alignment based on the sequence of the homologous protein.

[0077] In an embodiment of the present application, the sequence of the above-mentioned target protein can be obtained from the Uniprot (Universal Protein, currently the protein database with the richest information and the widest resources) database; the above-mentioned interacting proteins can be obtained from the STRING (a database for online search of known protein interactions) database, and the interacting proteins with a confidence score of more than 300 are retained; the structural template of the above-mentioned target protein can be obtained from the AlphaFoldDB database (an artificial intelligence system developed by DeepMind (a cutting-edge artificial intelligence company) that can provide high-accuracy protein structure prediction); the above-mentioned homologous proteins can be obtained from the Uniref90 database (a component of the Uniprot database).

[0078] It should be noted that constructing a multiple sequence alignment is common knowledge, and its construction process will not be described in detail here.

[0079] Step 12: construct the three-dimensional structure of the target protein based on the multiple sequence alignment and the structural template.

[0080] In some embodiments of the present application, AlphaFold2 is selected as the protein structure model (the model's prediction of most protein structures is only one atom away from the actual structure, reaching the level of human observation and prediction using complex instruments such as cryo-electron microscopes).

[0081] Specifically, after the multiple sequence alignment and the structural template of the target protein are input into AlphaFold2, the model outputs the three-dimensional structure of the target protein with atomic precision.

[0082] Step 13: extract sequence features from the target protein sequence and the interacting protein sequence to obtain the sequence features of the target protein and the interacting protein sequence.

[0083] In some embodiments of the present application, a pre-trained ESM-1b language model is selected to extract sequence features of the target protein and features of the interacting proteins, respectively.

[0084] For example, for any protein among the target protein and the interacting protein, after ESM-1b feature extraction, a sequence feature is obtained. Here, n represents the length of the protein sequence.

[0085] Since the residues of each protein (the remaining structural parts after the amino and carboxyl groups between amino acids are dehydrated to form bonds) are different, it is also necessary to average the sequence feature p in the residue dimension to facilitate subsequent calculations.

[0086] Specifically, for any protein among the target protein and the interacting protein, the formula

[0087]

[0088] Get the fixed dimension sequence feature p′ of the protein a , a=0,1,2,...,b,...,m, b=1,2,...,m, m represents the total number of interacting proteins, p′0 represents the fixed-dimensional sequence features of the target protein, p′ b represents the fixed-dimensional sequence features of the b-th interacting protein.

[0089] Step 14: using a protein mixing layer for mixing protein sequence features, the sequence features of the target protein and the sequence features of the interacting proteins are mixed to obtain a first mixed feature of the target protein.

[0090] The above-mentioned protein mixing layer includes multiple mixing processing layers, each mixing processing layer includes a multi-head attention layer, a feedforward neural network layer, a first regularized residual connection layer and a second regularized residual connection layer; wherein, the input end of the multi-head attention layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the multi-head attention layer is connected to the first input end of the first regularized residual connection layer, the second input end of the first regularized residual connection layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the first regularized residual connection layer is connected to the input end of the feedforward neural network layer, the output end of the feedforward neural network layer is connected to the first input end of the second regularized residual connection layer, and the second input end of the second regularized residual connection layer receives the data output by the output end of the first regularized residual connection layer.

[0091] For the u-th hybrid processing layer, u = 1, 2, ..., O-1, O represents the total number of hybrid processing layers, and the output end of the second regularized residual connection layer of the hybrid processing layer is respectively connected to the input end of the multi-head attention layer of the u+1-th hybrid processing layer and the first input end of the first regularized residual connection layer of the second hybrid processing layer.

[0092] For the Oth hybrid processing layer, the output end of the second regularized residual connection layer of the hybrid processing layer outputs the first hybrid feature.

[0093] It should be noted that the above-mentioned regularized residual connection layer is obtained by combining the residual connection layer and the regularization layer. In some embodiments of the present application, the result output by the residual connection layer is input into the regularization layer and then subjected to regularization processing, which can reduce the parameter fluctuation of the model and improve the performance of the model.

[0094] Specifically, for each mixed processing layer, the formula

[0095]

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] Get the first mixed feature in, A set of sequence features consisting of the sequence features of the target protein and the sequence features of the interacting proteins, p'0 represents the fixed-dimensional sequence feature of the target protein, p' b represents the fixed-dimensional sequence feature of the b-th interacting protein, b = 1, 2, ..., m, m represents the total number of interacting proteins, a i represents the i-th sequence of the b-th interacting protein, i = 1, 2, ..., n, n represents the length of the protein sequence, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the query space Q, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the key space K, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the value space V, W o Indicates that the feature dimension of the head attention is transformed to Learnable linear projections of the same feature dimension, W1 means A learnable linear projection to a high-dimensional feature space, W2 means A learnable linear projection that recovers from the high-dimensional feature space to the original feature space, h represents the number of heads in the multi-head attention layer, d k =d v =d model / h=1280, b1 and b2 are both bias terms, MultiHead(·) represents the multi-head attention mechanism calculation function, FusionLayer(·) represents the protein mixing layer calculation function, AddAndNorm(·) represents the regularized residual connection layer, LayerNorm(·) represents the layer normalization calculation function, FFN(·) represents the feedforward neural network calculation function, ReLU(·) represents the linear rectification function, Concat(·) represents the feature connection calculation function, Attention(·) represents the attention mechanism calculation function, Represents the field of real numbers.

[0103] Step 15: Mix the sequence feature of the target protein and the first mixed feature to obtain a second mixed feature of the target protein.

[0104] Specifically, by calculating the formula Get the second mixed feature v of the target protein sp ;

[0105] It should be noted that mixing the sequence features of the target protein and the first mixed features can avoid forgetting the sequence feature model of the target protein, which helps to improve the accuracy of protein function prediction.

[0106] Step 16: extract structural features of the three-dimensional structure to obtain the structural features of the target protein.

[0107] Step 17: Obtain a functional confidence score of the target protein based on the second mixed feature and the structural feature.

[0108] Specifically, by calculating the formula

[0109] v target =Concat(v sp , GMP(v s ))

[0110]

[0111] s p =Sigmoid((v target W1+b1)W2+b2)

[0112] Get the functional confidence score s of the target protein p ;in, s k represents the confidence score of the kth function of the target protein, k = 1, 2, ..., n, n represents the total number of protein functions, b1 and b2 both represent bias terms, GMP(·) represents the global average layer calculation function, because vs After step 13, the residue level is obtained, so GMP is needed to convert it into protein level GMP (v s ), Sigmoid(·) represents the Sigmoid function.

[0113] Step 18: predict the function of the target protein based on the function confidence score.

[0114] For example, the confidence scores of each function of the target protein are obtained as follows: Figure 2 As shown, Figure 2 In the expression, GO (geneontology): 0044260=86 indicates that the probability of the target protein having the cellular macromolecular metabolism function (GO: 0044260) is 86%.

[0115] The specific process of step 16 (extracting structural features from the three-dimensional structure to obtain the structural features of the target protein) is exemplified below.

[0116] Step 16.1: Construct a k-nearest neighbor graph based on the three-dimensional structure.

[0117] The above three-dimensional structure includes a plurality of residues.

[0118] For example, the coordinate nodes of the Cα atom of each residue are connected to adjacent nodes to form a k-nearest neighbor graph.

[0119] Step 16.2: construct an initial scalar feature s and an initial vector feature V for each residue in the k-nearest neighbor graph to obtain multiple groups of initial residue features; each group of initial residue features includes an initial scalar feature and an initial vector feature corresponding to a residue; wherein, n represents the number of constructed scalar features, and v represents the number of constructed vector features.

[0120] Exemplarily, an initial scalar feature s and an initial vector feature V are constructed for each residue and edge source in the k-nearest neighbor graph constructed in step 16.1.

[0121] Taking the i-th node as an example, suppose the C atom of the i-th node is represented by C i , whose initial scalar features include the sine and cosine of the dihedral angle. The dihedral angle is determined by the nitrogen atom N at the i-th node. i , the alpha carbon atom Cα on the i-th node i , the carbon atom C on the i-th node i , the carbon atom C on the i-1-th node i-1 , nitrogen atom N on the i+1-th node i+1 Calculated; its initial vector features include Cα i+1 -Cαi and Cα i-1 -Cα i The forward and reverse unit vectors in the direction, and the beta carbon atom Cβ at the i-th node estimated by assuming tetrahedral geometry and normalization i -Cα i The unit vector in the direction of . Among them, the alpha carbon is the carbon position closest to the parent functional group in the main chain of the organic compound, and the beta carbon is the second carbon position connected to the parent functional group.

[0122] In some embodiments of the present application, a Gaussian radial basis function is used to encode the edge length, and a sinusoidal encoding of the relative position information is used as the scalar feature of the edge. The vector feature of the edge is the direction formed by the Cα atoms at both ends of the edge.

[0123] Step 16.3, for each residue, calculate the formula

[0124] s′=σ((Concat(s,||Vh||2)W m +b)

[0125] V′=σ + (||V h W μ ||2)⊙V h W μ

[0126] V h =VW h

[0127] Obtain the intermediate scalar feature s′ and intermediate vector feature V′ corresponding to the residue; W h represents a learnable linear projection of the feature space transformation for the initial vector features, W m represents a learnable linear projection for feature space transformation of fused features containing scalar features and vector features, W μ represents a learnable linear projection for the feature space transformation of vector features, b represents the bias term, σ and σ + Both represent activation functions, ||·||2 represents L2 regularization, and μ represents the number of new vector features obtained.

[0128] Step 16.4: Based on the graph neural network model, the intermediate scalar feature s′ and the intermediate vector feature V′ are combined to obtain the target protein structural feature v s .

[0129] The following is an example of the process of updating graph nodes in a graph neural network.

[0130] For each graph node, the vectors of the graph node and the edges adjacent to the graph node are concatenated, and then the geometric vector perceptrons (GVP) are used to update the features.

[0131] Specifically, assuming and denote the embedding of node j and edge (j→i) respectively. The message transmitted from node j to node i can be expressed as g represents a function with a geometric vector perceptron, and the process of graph propagation can be expressed as follows:

[0132]

[0133] Here, k′ represents the number of incoming messages.

[0134] During graph propagation, the graph neural network uses GVP to update node embeddings, including the scalar features and vector features of all nodes. The update process can be expressed as follows:

[0135]

[0136] For a protein containing n residues, after the above processing, we get Where n represents the feature representation of each residue, d s Represents the dimension of each residue node in the output.

[0137] The protein function prediction device provided by the present application is exemplarily described below with reference to specific examples.

[0138] like Figure 3 As shown, an embodiment of the present application provides a protein function prediction device, the protein function prediction device 300 comprising:

[0139] An acquisition module 301 is used to obtain the sequence of the target protein, the sequence of the interacting protein that interacts with the target protein, the sequence of the homologous protein that is homologous to the target protein, and the structural template of the target protein, and to construct a multiple sequence alignment based on the sequence of the homologous protein;

[0140] A three-dimensional structure module 302 is used to construct a three-dimensional structure of a target protein based on multiple sequence alignment and a structure template;

[0141] A sequence feature module 303 is used to extract sequence features from the target protein sequence and the interacting protein sequence to obtain the sequence features of the target protein and the interacting protein sequence;

[0142] A first mixing module 304 is configured to mix the sequence features of the target protein and the sequence features of the interacting protein using a protein mixing layer for mixing protein sequence features to obtain a first mixed feature of the target protein;

[0143] The second mixing module 305 is used to mix the sequence feature of the target protein and the first mixed feature to obtain a second mixed feature of the target protein;

[0144] Structural feature module 306, for extracting structural features of the three-dimensional structure to obtain structural features of the target protein;

[0145] A confidence score calculation module 307 is used to obtain a functional confidence score of the target protein based on the second mixed feature and the structural feature;

[0146] The prediction module 308 is used to predict the function of the target protein according to the function confidence score.

[0147] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0148] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0149] like Figure 4 As shown, an embodiment of the present application provides a terminal device, such as Figure 4 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above method embodiments when executing the computer program D102.

[0150] Specifically, when the processor D100 executes the computer program D102, it obtains the sequence of the target protein, the sequence of the interacting protein, the sequence of the homologous protein and the structural template of the target protein, and constructs a multiple sequence alignment based on the sequence of the homologous protein. Then, it constructs the three-dimensional structure of the target protein based on the multiple sequence alignment and the structural template, and then extracts sequence features of the target protein sequence and the interacting protein sequence to obtain the sequence features of the target protein and the sequence features of the interacting protein. Then, the sequence features of the target protein and the sequence features of the interacting protein are mixed using a protein mixing layer to obtain a first mixed feature of the target protein. Then, the sequence features of the target protein and the first mixed feature are mixed to obtain a second mixed feature of the target protein. Subsequently, structural feature extraction is performed on the three-dimensional structure of the target protein to obtain the structural features of the target protein. Then, based on the second mixed feature and the structural feature, a target protein function confidence score is obtained. Finally, the function of the target protein is predicted based on the target protein function confidence score. Among them, the sequence features of the target protein and the sequence features of the interacting protein are mixed using a protein mixing layer to obtain a first mixed feature, and then the sequence features of the target protein and the first mixed feature are mixed to obtain a second mixed feature. This can avoid the target protein sequence features being diluted by the model, and at the same time combine multiple protein features to improve the accuracy of protein function prediction.

[0151] The processor D100 may be a central processing unit (CPU), or may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0152] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card, etc. equipped on the terminal device D10. Furthermore, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0153] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned various method embodiments can be implemented.

[0154] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0155] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the protein function prediction device / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0156] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0157] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0158] In the embodiments provided in this application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0159] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0160] The protein function prediction method provided by this application only requires the input of the target protein sequence. The model can automatically search for proteins that interact with the target protein, automatically predict the target protein's structure, and achieve function prediction based on sequence, structure, and interaction information. The present invention can dynamically integrate existing experimental data to achieve high-performance protein function prediction, which can help biological experiments reduce blindness and experimental costs, effectively narrow the gap between protein sequence and function, and provide an effective means to improve the prevention, diagnosis and treatment of major diseases.

[0161] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting protein function, characterized in that: include: Obtaining the sequence of a target protein, the sequence of an interacting protein that interacts with the target protein, the sequence of a homologous protein that is homologous to the target protein, and a structural template of the target protein, and constructing a multiple sequence alignment based on the sequences of the homologous proteins; constructing the three-dimensional structure of the target protein according to the multiple sequence alignment and the structural template; performing sequence feature extraction on the sequence of the target protein and the sequence of the interacting protein to obtain sequence features of the target protein and the sequence features of the interacting protein; Using a protein mixing layer for mixing protein sequence features, the sequence features of the target protein and the sequence features of the interacting protein are mixed to obtain a first mixed feature of the target protein; performing a mixing process on the sequence feature of the target protein and the first mixed feature to obtain a second mixed feature of the target protein; Extracting structural features of the three-dimensional structure to obtain structural features of the target protein; Obtaining a functional confidence score of the target protein according to the second mixed feature and the structural feature; Predicting the function of the target protein according to the functional confidence score; Wherein, the protein mixing layer includes a plurality of mixing treatment layers; Each of the multiple hybrid processing layers includes a multi-head attention layer, a feedforward neural network layer, a first regularized residual connection layer, and a second regularized residual connection layer; wherein, the input end of the multi-head attention layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the multi-head attention layer is connected to the first input end of the first regularized residual connection layer, the second input end of the first regularized residual connection layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the first regularized residual connection layer is connected to the input end of the feedforward neural network layer, the output end of the feedforward neural network layer is connected to the first input end of the second regularized residual connection layer, and the second input end of the second regularized residual connection layer receives the data output by the output end of the first regularized residual connection layer; For the u-th hybrid processing layer, u = 1, 2, ..., O-1, O represents the total number of hybrid processing layers, the output end of the second regularized residual connection layer of the hybrid processing layer is connected to the input end of the multi-head attention layer of the u+1-th hybrid processing layer and the first input end of the first regularized residual connection layer of the second hybrid processing layer; For the Oth hybrid processing layer, the output end of the second regularized residual connection layer of the hybrid processing layer outputs the first hybrid feature.

2. The prediction method according to claim 1, characterized in that The method of mixing the sequence features of the target protein and the sequence features of the interacting protein using a protein mixing layer for mixing protein sequence features to obtain a first mixed feature of the target protein includes: For each mixed processing layer, the formula Get the first mixed feature in, A sequence feature set consisting of the sequence features of the target protein and the sequence features of the interacting protein, p'0 represents the fixed-dimensional sequence feature of the target protein, p' b represents the fixed-dimensional sequence feature of the b-th interacting protein, b=1,2,...,m, m represents the total number of interacting proteins, a i represents the i-th sequence of the b-th interacting protein, i=1,2,...,n, n represents the length of the protein sequence, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the query space Q, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the key space K, Indicates that the i-th head in the multi-head attention layer will A learnable linear projection to the value space V, W o Indicates that the feature dimension of the head attention is transformed to Learnable linear projections of the same feature dimension, W1 means A learnable linear projection to a high-dimensional feature space, W2 means A learnable linear projection that recovers from the high-dimensional feature space to the original feature space, h represents the number of heads in the multi-head attention layer, d k =d v =d model / h=1280, b1 and b2 are both bias terms, MultiHead(·) represents the multi-head attention mechanism calculation function, FusionLayer(·) represents the protein mixing layer calculation function, AddANdNorm(·) represents the regularized residual connection layer, LayerNorm(·) represents the layer normalization calculation function, FFN(·) represents the feedforward neural network calculation function, ReLU(·) represents the linear rectification function, Concat(·) represents the feature connection calculation function, Attention(·) represents the attention mechanism calculation function, Represents the field of real numbers.

3. The prediction method according to claim 2, characterized in that The mixing process of the sequence feature of the target protein and the first mixed feature to obtain the second mixed feature of the target protein includes: By calculating the formula Get the second mixed feature v of the target protein sp ; 4. The prediction method according to claim 3, characterized in that The extracting structural features of the three-dimensional structure to obtain the structural features of the target protein includes: Constructing a k-nearest neighbor graph based on the three-dimensional structure; the k-nearest neighbor graph is composed of coordinate nodes of the Cα atoms of each residue in the three-dimensional structure of the target protein connected to adjacent nodes; the three-dimensional structure includes multiple residues; Constructing an initial scalar feature s and an initial vector feature V for each residue in the k-nearest neighbor graph to obtain multiple groups of initial residue features; each group of initial residue features includes an initial scalar feature and an initial vector feature corresponding to a residue; wherein, n represents the number of constructed scalar features, and v represents the number of constructed vector features; For each residue, the formula s'=σ((Concat(s,||V h ||2)W m +b) V'=σ + (||V h IN μ ||2)⊙V h IN μ V h =VW h Obtain the intermediate scalar feature s' and intermediate vector feature V' corresponding to the residue; W h represents a learnable linear projection of the feature space transformation for the initial vector features, W m represents a learnable linear projection for feature space transformation of fused features containing scalar features and vector features, W μ represents a learnable linear projection for the feature space transformation of vector features, b represents the bias term, σ and σ + Both represent activation functions, ||·||2 represents L2 regularization, and μ represents the number of new vector features obtained; Based on the graph neural network model, the intermediate scalar feature s' and the intermediate vector feature V' are combined to obtain the structural feature v of the target protein s .

5. The prediction method according to claim 4, characterized in that Obtaining a function confidence score of the target protein according to the second mixed feature and the structural feature includes: By calculating the formula v target =Concat(v sp ,GMP(v s )) s p =Sigmoid((v target W1+b1)W2+b2) Get the functional confidence score s of the target protein p ;in, s k represents the confidence score of the kth function of the target protein, k = 1, 2, ..., n, n represents the total number of protein functions, b1 and b2 both represent bias terms, and GMP(·) represents the global averaging layer calculation function.

6. The prediction method according to claim 1, characterized in that Before using the protein mixing layer for mixing protein sequence features to mix the sequence features of the target protein and the sequence features of the interacting protein to obtain the first mixed features of the target protein, the prediction method further includes: For any one of the target protein and the interacting protein, the formula Get the fixed dimension sequence feature p' of the protein a , a=0,1,2,...,b,...,m,b=1,2,...,m,m represents the total number of interacting proteins, p'0 represents the fixed-dimensional sequence features of the target protein, p' b represents the fixed-dimensional sequence features of the b-th interacting protein.

7. A protein function prediction device, characterized in that: include: an acquisition module, configured to acquire the sequence of a target protein, the sequence of an interacting protein interacting with the target protein, the sequence of a homologous protein homologous to the target protein, and a structural template of the target protein, and construct a multiple sequence alignment based on the sequences of the homologous proteins; A three-dimensional structure module, used to construct the three-dimensional structure of the target protein based on the multiple sequence alignment and the structure template; A sequence feature module, used to extract sequence features from the sequence of the target protein and the sequence of the interacting protein to obtain the sequence features of the target protein and the sequence features of the interacting protein; a first mixing module, configured to mix the sequence features of the target protein and the sequence features of the interacting protein using a protein mixing layer for mixing protein sequence features, to obtain a first mixed feature of the target protein; a second mixing module, configured to mix the sequence feature of the target protein and the first mixed feature to obtain a second mixed feature of the target protein; A structural feature module is used to extract structural features of the three-dimensional structure to obtain the structural features of the target protein; a confidence score calculation module, configured to obtain a functional confidence score of the target protein based on the second mixed feature and the structural feature; A prediction module, configured to predict the function of the target protein according to the function confidence score; The first mixing module is specifically used to implement: The protein mixing layer includes a plurality of mixing treatment layers; Each of the multiple hybrid processing layers includes a multi-head attention layer, a feedforward neural network layer, a first regularized residual connection layer, and a second regularized residual connection layer; wherein, the input end of the multi-head attention layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the multi-head attention layer is connected to the first input end of the first regularized residual connection layer, the second input end of the first regularized residual connection layer receives the sequence features of the target protein and the sequence features of the interacting protein, the output end of the first regularized residual connection layer is connected to the input end of the feedforward neural network layer, the output end of the feedforward neural network layer is connected to the first input end of the second regularized residual connection layer, and the second input end of the second regularized residual connection layer receives the data output by the output end of the first regularized residual connection layer; For the u-th hybrid processing layer, u = 1, 2, ..., O-1, O represents the total number of hybrid processing layers, the output end of the second regularized residual connection layer of the hybrid processing layer is connected to the input end of the multi-head attention layer of the u+1-th hybrid processing layer and the first input end of the first regularized residual connection layer of the second hybrid processing layer; For the Oth hybrid processing layer, the output end of the second regularized residual connection layer of the hybrid processing layer outputs the first hybrid feature.

8. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the protein function prediction method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the protein function prediction method according to any one of claims 1 to 6 is implemented.

Images