Information screening method based on graph neural network and electronic equipment

Abstract

The invention discloses an information screening method based on a graph neural network and electronic equipment, and relates to the technical field of bioinformatics, and aims to improve the prediction accuracy of microorganisms corresponding to cross-species antibacterial peptides, the information screening method based on the graph neural network comprises the steps of obtaining features of a target substance, the target substance comprising a target antibacterial peptide or a target microorganism; based on the graph neural network and the features of the target substance, obtaining prediction information which is different from the type of the target substance and has a biological guiding relationship; wherein the graph neural network comprises a plurality of different types of nodes and a connection relationship among the plurality of nodes, the connection relationship comprises a first connection relationship among the nodes of the same type and a second connection relationship among the nodes of different types, the first connection relationship is used for representing the similarity among the nodes, and the second connection relationship is used for representing the similarity among the nodes. The second connection relationship is used for representing that the nodes have a biological guiding relationship, and the plurality of nodes at least comprise microbial nodes and antibacterial peptide nodes.

Description

Technical Field

[0001] This disclosure relates to the field of bioinformatics technology, and in particular to an information filtering method and electronic device based on graph neural networks. Background Technology

[0002] Microorganisms (bacteria, fungi, and viruses) are widely distributed both inside and outside the human body, and many diseases are closely related to microbial infections. With the emergence of drug-resistant microorganisms, the efficacy of traditional antibiotics is declining, making the development of alternative therapies urgent. Antimicrobial peptides (AMPs), as biomolecules with broad-spectrum activity, can effectively kill target microorganisms and inhibit the development of drug resistance by inducing membrane permeability through interaction with microbial membranes.

[0003] Currently, AMP research mainly relies on biological experiments for verification, but the long experimental cycles and high costs limit its development speed. With the development of artificial intelligence, more and more computational methods are being applied to AMP prediction. Machine learning-based methods (such as CAMPr3) use sequence features such as amino acid composition, charge, and isoelectric point combined with classifiers to predict antibacterial activity; deep learning methods (such as iAMPCN) are further extended to antiviral and antifungal functions. Despite some success, the diversity of microorganisms presents challenges, making it difficult to improve the prediction accuracy. Summary of the Invention

[0004] Firstly, an information filtering method based on a graph neural network is provided, the method comprising: The characteristics of a target substance are obtained, wherein the target substance includes a target antimicrobial peptide or a target microorganism; Based on graph neural networks and the characteristics of the target substance, predictive information that is different from the type of the target substance and has a biological guidance relationship is obtained; The graph neural network includes multiple nodes of different types and multiple connection relationships between the nodes. The connection relationships include a first connection relationship between nodes of the same type and a second connection relationship between nodes of different types. The first connection relationship is used to characterize the similarity between the nodes, and the second connection relationship is used to characterize the biological guidance relationship between the nodes. Among them, the plurality of nodes include at least microbial nodes and antimicrobial peptide nodes.

[0005] In an exemplary embodiment, obtaining predictive information that is different in type from the target substance and has a biologically guiding relationship based on the graph neural network and the features of the target substance includes: Embed the features of the target substance into the graph neural network; The graph neural network embedded with the features of the target substance is iteratively subjected to T time steps of evolutionary operation to obtain predictive information that has a connection relationship with the features of the target substance; In each evolutionary operation at time step t, the graph neural network obtained at time step t-1 obtains the current information transmission mode, and performs feature transmission on the feature representation of the currently selected partial nodes according to the current information transmission mode. The feature passing is used to enable one of the nodes in the subset of nodes to aggregate the features of its neighboring nodes.

[0006] In an exemplary embodiment, the information transmission mode includes a first transmission mode and a second transmission mode, wherein the first transmission mode is used to characterize feature transmission between multiple nodes of the same type, and the second transmission mode is used to characterize feature transmission between nodes of different types.

[0007] In one exemplary embodiment, the first delivery mode includes a delivery mode between multiple microbial nodes and a delivery mode between multiple antimicrobial peptide nodes; The second delivery mode includes a delivery mode from the microbial node to the antimicrobial peptide node, and a delivery mode from the antimicrobial peptide node to the microbial node.

[0008] In one exemplary embodiment, the time step in which the first transmission mode is located is before the time step in which the second transmission mode is located; Alternatively, the transmission patterns between the microbial nodes and the transmission patterns between the antimicrobial peptide nodes may overlap between the second transmission patterns in different directions.

[0009] In an exemplary embodiment, the graph neural network further includes disease nodes, the target substance includes the target antimicrobial peptide, and the step of obtaining predictive information that is different in type from the target substance and has a biologically guiding relationship based on the graph neural network and the features of the target substance includes: Based on the graph neural network and the characteristics of the target substance, at least one predicted microorganism and at least one predicted disease information corresponding to the target antimicrobial peptide are obtained. Based on the probability of the predicted microorganism and the probability of the predicted disease, information on the predicted microorganisms that the target antimicrobial peptide can disinfect and the predicted diseases that the target antimicrobial peptide can participate in is obtained.

[0010] In one exemplary embodiment, the graph neural network is obtained through the following steps: Basic data is obtained from the database, which includes multiple related data, each of which corresponds to a group of microorganisms and antimicrobial peptides with a disinfection relationship; Based on the multiple sets of associated data, an initial heterogeneous network is constructed, which includes multiple initial nodes; The graph neural network is obtained by training an initial heterogeneous network based on multiple datasets. The dataset includes combinations of antimicrobial peptide samples and microbial samples that have a disinfect relationship, as well as combinations of antimicrobial peptide samples and microbial samples that do not have a disinfect relationship.

[0011] In one exemplary embodiment, the association data includes the sequence of the antimicrobial peptide; the construction of an initial heterogeneous network based on multiple sets of association data, the heterogeneous network including multiple initial nodes, includes: Redundant sequences in the antimicrobial peptide sequence are removed from the associated data to obtain new associated data; Determine a first similarity between the antimicrobial peptides in different new association data, and determine a second similarity between the microorganisms in different new association data; The heterogeneous network is constructed based on the first similarity, the second similarity, and the new associated data.

[0012] In one exemplary embodiment, training an initial heterogeneous network based on multiple datasets to obtain the graph neural network includes: The features of antimicrobial peptide samples in the dataset are embedded into the heterogeneous network; wherein, the heterogeneous network is used to iteratively perform T time-step evolutionary operations on nodes in the network with the embedded features of the antimicrobial peptide samples. In each t-th time-step evolutionary operation, based on the heterogeneous network obtained at the (t-1)-th timestamp, the current information transmission mode is obtained, and feature transmission is performed on the feature representations of the currently selected subset of nodes according to the current information transmission mode; wherein, the feature transmission is used to enable one node in the subset of nodes to aggregate the features of its neighboring nodes; Obtain the predicted microorganisms that have a disinfect relationship with the antimicrobial peptide sample output by the heterogeneous network; Based on the predicted microorganism and the microbial sample corresponding to the antimicrobial peptide sample, the learning parameters of the heterogeneous network are updated to obtain the graph neural network.

[0013] In a second aspect, an electronic device is provided, comprising: a memory and a processor, the memory being coupled to the processor; wherein the memory stores program instructions, which, when executed by the processor, cause the electronic device to perform the information filtering method based on a graph neural network as described in any one of the first aspects.

[0014] The information filtering method based on graph neural networks in this disclosure can acquire the characteristics of target substances and, based on the graph neural network and the characteristics of target substances, acquire predictive information that is different in type from the target substances and has a biologically oriented relationship. The target substances include target antimicrobial peptides or target microorganisms. The graph neural network includes multiple nodes of different types and the connection relationships between multiple nodes. The connection relationships include a first connection relationship between nodes of the same type and a second connection relationship between nodes of different types. The first connection relationship is used to characterize the similarity between nodes, and the second connection relationship is used to characterize the biologically oriented relationship between nodes. Furthermore, the multiple nodes in the graph neural network include at least microbial nodes and antimicrobial peptide nodes.

[0015] In this embodiment, since the connections between multiple nodes in the graph neural network include not only first connections but also second connections, and the types of multiple nodes can be different, the graph neural network is a heterogeneous network. It constructs a relationship map between antimicrobial peptides and microorganisms. Therefore, when embedding the features of the target substance into the graph neural network, based on the similarity between antimicrobial peptides and microorganisms, as well as the biological guidance relationship between antimicrobial peptides and microorganisms (such as a disinfect relationship), predictive information with a biological guidance relationship with the target substance can be found through corresponding feature aggregation paths. For example, when the target substance is a target antimicrobial peptide, the microorganisms that can be disinfected by the target antimicrobial peptide can be predicted, thus achieving drug-microorganism prediction. Similarly, when the target substance is a target microorganism, the antimicrobial peptides that can disinfect the target microorganism can be predicted, thus achieving microorganism-drug prediction. Thus, this embodiment fully utilizes the heterogeneous interactions between multiple heterogeneous nodes in the graph neural network, as well as the interactions between nodes of the same type, improving the prediction accuracy of the disinfect relationship between antimicrobial peptides and microorganisms to address the challenges of microbial diversity.

[0016] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, specific embodiments of this disclosure are described below. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments or related technologies of this disclosure, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the scale in the drawings is for illustration only and does not represent the actual scale.

[0018] Figure 1 A schematic diagram illustrating the process of acquiring a graph neural network in an embodiment of this disclosure is shown; Figure 2 and Figure 3 Schematic diagrams of two heterogeneous networks in embodiments of this disclosure are shown respectively; Figure 4 This illustration shows a flowchart of the steps for training a graph neural network in an embodiment of the present disclosure. Figure 5 A flowchart illustrating the steps of the information filtering method based on graph neural networks in an embodiment of this disclosure is shown. Figure 6 A schematic diagram of an example of an information filtering method based on a graph neural network in an embodiment of this disclosure is shown. Detailed Implementation

[0019] To make the above-mentioned objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0020] In related technologies, AMP research mainly relies on biological experiments for verification, but the long experimental cycle and high cost limit the speed of development. With the development of artificial intelligence, machine learning methods have been applied to AMP prediction. However, machine learning-based methods (such as CAMPr3) use sequence features such as amino acid composition, charge, and isoelectric point combined with classifiers to predict antibacterial activity, while deep learning methods (such as iAMPCN) have been further extended to antiviral and antifungal functions.

[0021] However, current predictions are mostly limited to the phylum or kingdom level. In the face of the challenges brought by microbial diversity, they lack precise AMP identification for specific species, which reduces the hit rate between antimicrobial peptides and microorganisms. In other words, the prediction accuracy of the disinfect relationship between antimicrobial peptides and microorganisms is not high.

[0022] In related technologies, with the development of GNNs, metapaths have shown excellent performance in capturing the topology of heterogeneous networks. Metapaths in biological heterogeneous networks may include specific metabolic pathways or biological principles, which helps to interpret AMP predictions. However, related technologies rely on handcrafted metapaths, which are limited by domain experience and thus difficult to transfer, restricting the flexibility of feature transfer. Therefore, the improvement in the prediction accuracy of the disinfect relationship between antimicrobial peptides and microorganisms is limited.

[0023] In view of this, the inventors of this application provide an automatic evolutionary information aggregation mechanism. Under this mechanism, an automatic evolutionary representation method is adopted, which divides feature transmission into multiple timestamps. In each timestamp, the feature transmission mode of the current timestamp is automatically selected based on the node representation of the graph neural network in the previous timestamp, thereby dynamically generating the feature representation of the nodes in the graph neural network. This enables adaptive and dynamic inter-node feature transmission for complex heterogeneous networks, which helps the representation learning of the graph neural network and improves the prediction accuracy of the disinfect relationship between antimicrobial peptides and microorganisms.

[0024] First, the process of obtaining the graph neural network in the embodiments of this application will be explained.

[0025] Please refer to Figure 1 As shown, Figure 1 A schematic diagram illustrating the acquisition process of a graph neural network is shown, such as... Figure 1 As shown, it includes the following steps: Step S1: Retrieve basic data from the database; The basic data includes multiple related data, each of which corresponds to a group of microorganisms and antimicrobial peptides with a disinfection relationship.

[0026] The database can be a publicly available database in this field, such as DBAASP and Disbiome. Generally, publicly available databases contain information on antimicrobial peptides that have actual disinfecting effects on microorganisms. Typically, there is data on a single microorganism that can be disinfected by multiple antimicrobial peptides, and there is also data on a single antimicrobial peptide that can disinfect multiple microorganisms.

[0027] In this embodiment, data on a group of microorganisms and antimicrobial peptides that have a disinfect relationship will be recorded as a set of associated data, which can then be used as basic data.

[0028] Different correlation data can correspond to different disinfection relationships between microorganisms and antimicrobial peptides.

[0029] Step S2: Construct an initial heterogeneous network based on multiple related data; The heterogeneous network includes multiple initial nodes and a first connection relationship and a second connection relationship between the multiple nodes. The types of the multiple initial nodes can be different. The multiple initial nodes can include microbial nodes and antimicrobial peptide nodes obtained from basic data. The first connection relationship can be used to characterize the similarity between nodes of the same type, and the second connection relationship can be used to characterize the biological guidance relationship between nodes of different types.

[0030] In this embodiment, an initial heterogeneous network can be constructed based on associated data.

[0031] Specifically, please conclude Figure 2 As shown, the antimicrobial peptide in the associated data can be regarded as the antimicrobial peptide node T1 in the heterogeneous network, and the microorganism in the associated data can be regarded as the microorganism node T2 in the heterogeneous network. The elimination relationship between the antimicrobial peptide and the microorganism is characterized by connecting the antimicrobial peptide node and the microorganism node. This connection relationship between the antimicrobial peptide node and the microorganism node can be called the second connection relationship B1.

[0032] like Figure 2 As shown, a microorganism can be disinfected by multiple antimicrobial peptides, and there are also cases where one antimicrobial peptide can disinfect multiple microorganisms. Therefore, an antimicrobial peptide node can connect to at least one microorganism node, and a microorganism node can connect to at least one antimicrobial peptide node.

[0033] In this embodiment, nodes of the same type can also be connected based on their similarity.

[0034] For example, the similarity between the characteristics of two microorganisms can be determined, and the nodes of the two microorganisms can be connected according to the similarity to obtain a first connection relationship between the microorganism nodes; or the similarity between the characteristics of two antimicrobial peptides can be determined, and the nodes of the two antimicrobial peptides can be connected according to the similarity to obtain a first connection relationship between the antimicrobial peptide nodes.

[0035] In some examples of this embodiment, when constructing multiple initial nodes of a heterogeneous network, redundant sequences in the antimicrobial peptide sequences in the associated data can be removed first to obtain new associated data; then, a first similarity between antimicrobial peptides in different new associated data and a second similarity between microorganisms in different new associated data can be determined; based on the first similarity, the second similarity, and the new associated data, a heterogeneous network is constructed.

[0036] In this example, the sequence length of the antimicrobial peptide can be limited, such as to less than 50, and CD-HIT can be used to remove redundant sequences with a threshold of 0.7 to ensure the non-redundancy of the associated data and the efficiency of the analysis.

[0037] like Figure 2As shown, after limiting the sequence length of the antimicrobial peptide in each associated data, new associated data can be obtained. Then, the first similarity B2 between the antimicrobial peptides in different new associated data can be calculated. Specifically, the first similarity between different antimicrobial peptides that disinfect the same microorganism can be calculated, and the first similarity between different antimicrobial peptides that disinfect different microorganisms can be calculated.

[0038] For example, if the antimicrobial peptides in new associated data A and new associated data B are different but the microorganisms are the same, then a first similarity between the antimicrobial peptides in new associated data A and new associated data B can be calculated.

[0039] For example, if the antimicrobial peptides in new associated data A and new associated data C are different and the microorganisms are different, then a first similarity between the antimicrobial peptides in new associated data A and new associated data C can be calculated.

[0040] like Figure 2 As shown, a second similarity B3 between microorganisms in different new associated data can be calculated. Specifically, a second similarity can be calculated between different species of microorganisms that can be disinfected by the same antimicrobial peptide, and a second similarity can be calculated between different species of microorganisms that can be disinfected by different antimicrobial peptides.

[0041] For example, if the microorganisms in new associated data D and new associated data E are different but have the same antimicrobial peptides, then a second similarity can be calculated between the microorganisms in new associated data D and new associated data E.

[0042] For example, if the microorganisms in new associated data D and new associated data F are different and their antimicrobial peptides are different, then a second similarity between the microorganisms in new associated data D and new associated data F can be calculated.

[0043] Therefore, it is possible to establish the similarity between different antimicrobial peptides in various disinfection relationships, thereby achieving similarity across microbial species and antimicrobial peptide species.

[0044] In practice, two antimicrobial peptide nodes with a first similarity greater than a first preset similarity can be connected, as can two microbial nodes with a second similarity greater than a second preset similarity. This can eliminate the influence of nodes with similarity on subsequent information transmission and avoid information interference.

[0045] In this example, the first similarity refers to the sequence similarity of the antimicrobial peptides, which can be calculated using the Smith-Waterman algorithm.

[0046] In this example, the Gaussian kernel method can be used to calculate the second similarity between microorganisms.

[0047] In some embodiments, the associated data may also include data on the relationship between microorganisms and diseases, characterizing diseases caused by microorganisms.

[0048] like Figure 3 As shown, when constructing a heterogeneous network based on associated data, a disease can be used as a disease node T3 in the heterogeneous network, and a third similarity B4 between different disease nodes can be calculated. Two disease nodes with a third similarity greater than a preset third similarity can be connected to form a first connection relationship between nodes of the same type. Microbial nodes and disease nodes T3 of the diseases caused by microbial nodes T2 can be connected to form a second connection relationship B5.

[0049] Step S3: Train the initial heterogeneous network based on multiple datasets to obtain a graph neural network; The dataset includes combinations of antimicrobial peptide samples and microbial samples with a disinfect relationship, as well as combinations of antimicrobial peptide samples and microbial samples without a disinfect relationship.

[0050] In this embodiment, the combination of antimicrobial peptide samples and microbial samples with a disinfect relationship can be called a positive sample, and the combination of antimicrobial peptide samples and microbial samples without a disinfect relationship can be called a negative sample.

[0051] In this embodiment, the features of antimicrobial peptide samples in the dataset can be embedded into a heterogeneous network, and feature aggregation can be performed based on the node feature transfer mode of dynamic path, thereby obtaining a heterogeneous network after feature aggregation. In this heterogeneous network, the connection relationship between the antimicrobial peptide sample and other multiple nodes is constructed. Each connection relationship can be regarded as the connection edge between the antimicrobial peptide sample and other multiple nodes. Based on the probability of the connection edge, the microbial node that can be eliminated by the antimicrobial peptide sample can be predicted, and the microbial node is used as the predicted microorganism.

[0052] Next, based on the predicted microorganisms corresponding to the antimicrobial peptide samples and the actual microorganisms that can be eliminated corresponding to the antimicrobial peptide samples, a loss function is constructed to update the parameters of the heterogeneous network.

[0053] In one example of this embodiment, such as Figure 4 As shown, it includes the following steps: Step S31: Embed the features of the antimicrobial peptide samples in the dataset into a heterogeneous network; The heterogeneous network is used to iteratively perform T time-step evolutionary operations on nodes in a network that embeds features of antimicrobial peptide samples. In each t-th time-step evolutionary operation, based on the heterogeneous network obtained at the (t-1)-th timestamp, the current information transmission mode is obtained, and feature transmission is performed on the feature representations of the currently selected partial nodes according to the current information transmission mode. Feature transmission is used to enable one node in the partial nodes to aggregate the features of its neighboring nodes. Step S32: Obtain the predicted microorganisms with a disinfect relationship to the antimicrobial peptide sample from the heterogeneous network output; Step S33: Based on the predicted microorganisms and the microbial samples corresponding to the antimicrobial peptide samples, update the learning parameters of the heterogeneous network to obtain a graph neural network.

[0054] In this example, the features of the antimicrobial peptide sample can be used to encode the sequence of the antimicrobial peptide sample.

[0055] In this example, the features of the antimicrobial peptide sample can be embedded into the heterogeneous network. For example, the features of the antimicrobial peptide sample can be used as new nodes in the heterogeneous network, and the heterogeneous network can learn the association between the new node and the original multiple nodes.

[0056] In this embodiment, after embedding the features of the antimicrobial peptide sample into the heterogeneous network, the evolution operation can be performed iteratively for T time steps. In each evolution operation at the t-th time step, the information transmission mode required at the t-th time step can be obtained based on the heterogeneous network obtained at the (t-1)-th timestamp.

[0057] Here, the information transmission mode can be understood as the transmission path of the currently being calculated nodes. This transmission path is a directional path, for example, such as... Figure 3 As shown, the node propagation path can be T21-T12, which represents the aggregation of features from T21 onto T12.

[0058] In this context, at each time step t, the information transmission pattern at that time step can be dynamically determined based on the heterogeneous network obtained at time step (t-1). In this way, the feature transmission between nodes is no longer a fixed meta-path.

[0059] In some examples, the information transfer pattern may include a first transfer pattern and a second transfer pattern. The first transfer pattern is used to characterize feature transfer between multiple nodes of the same type, and the second transfer pattern is used to characterize feature transfer between nodes of different types.

[0060] In this example, the information transmission pattern may include not only feature transmission between different types of nodes, such as not only node transmission paths T21-T12, but also feature transmission between nodes of the same type, such as transmission paths between different antimicrobial peptide nodes, transmission paths between different microbial nodes, and transmission paths between nodes with different functional diseases.

[0061] In the evolutionary operation at each time step, the first and second transmission modes mentioned above can be dynamically selected.

[0062] In one embodiment, the first delivery mode includes a delivery mode between multiple microbial nodes and a delivery mode between multiple antimicrobial peptide nodes; the second delivery mode includes a delivery mode from microbial nodes to antimicrobial peptide nodes and a delivery mode from antimicrobial peptide nodes to microbial nodes.

[0063] In the case of disease nodes, the first transmission mode also includes a transmission mode between multiple disease nodes, and the second transmission mode also includes a transmission mode from microbial nodes to disease nodes.

[0064] In one embodiment, the time step in which the first transmission mode is located is between the time steps in which the second transmission mode is located.

[0065] In this embodiment, feature transfer is first performed between nodes of the same type in T time steps. After the transfer is completed, feature transfer can be performed between nodes of different types. This enables hierarchical information aggregation within the same species, and then cross-species information aggregation, thereby improving the gradient of information aggregation.

[0066] In one embodiment, the delivery patterns between microbial nodes and between antimicrobial peptide nodes overlap between second delivery patterns in different directions.

[0067] In this embodiment, since the heterogeneous network includes multiple nodes of different types, feature transfer between nodes of the same type can occur intermittently between feature transfers between nodes of different types.

[0068] For example, at time step t, the transmission pattern between microbial nodes can be performed; at time step t+1, the feature transmission from microorganisms to antimicrobial peptides can be performed; at time step t+2, the feature transmission between antimicrobial peptides can be performed; and at time step t+3, the feature transmission from antimicrobial peptides to microorganisms can be performed. This process can be repeated in an alternating manner to improve the cross-species information aggregation capability and avoid the loss of underlying information.

[0069] The specific feature transfer path, such as which nodes should perform adjustments and transfers at the t-th time step, and the path used for feature transfer within these nodes, can be random or determined based on the feature transfer results of the previous time step. In the latter case, the feature transfer path for each time step can be continuously learned and optimized.

[0070] For example, a timestamp is set. The heterogeneous network characteristics at a given location are expressed as follows (1): Equation (1) in, For timestamps Characteristics of all nodes, For the previous timestamp up to the current timestamp The information aggregation pattern set. The features of each node are generated by inheriting or propagating information from previous timestamps through the currently selected information transmission pattern, as shown in equation (2): Equation (2) in, This represents an update function based on the characteristics of the information transmission pattern.

[0071] Information transmission patterns can include “microorganism → peptide”, “peptide → microorganism”, “microorganism → disease”, etc., or they can choose not to affect or directly inherit the previous timestamp features.

[0072] In this example, each timestamp The node characteristics are affected by previous timestamps The characteristics of the influence, through possible information transmission patterns According to probability Dissemination. Specifically, timestamps. Timestamp The characteristics of the influence are calculated as follows (3): Equation (3) in, Let the node degree matrix be... For timestamps The node characteristics. In the peptide-microbe association prediction task, the focus is on peptide and microbe nodes.

[0073] As described in the foregoing embodiments, timestamp The possible information transmission modes are limited to equation (4): Equation (4) In this context, MP represents the microbial-antimicrobial peptide information transmission mode, PM represents the antimicrobial peptide-microbial information transmission mode, DM represents the disease-microbial information transmission mode, MM represents the microbial-microbial information transmission mode, and PP represents the antimicrobial peptide-antimicrobial peptide information transmission mode.

[0074] In some embodiments, the feature representation of each node in a graph neural network can be understood as an "evolutionary" process, with each timestamp... The node feature is represented by the previous timestamp. The influence of feature identifiers, specifically, on information transmission patterns in each evolutionary operation of timestamps. By weight Propagation, where i represents a node and t represents a timestamp.

[0075] In this embodiment, a binary gate can be defined for each candidate transmission path in each information transmission mode. .when The candidate propagation path is selected when the value is 0, and is blocked when the value is 0. Therefore, in the evolutionary operation at each timestamp, it can be based on... The information transmission mode of the current timestamp is obtained, and then the transmission path for information transmission is selected according to the binary values ​​corresponding to the candidate transmission paths, thereby aggregating the feature representations of the nodes according to the transmission path.

[0076] In some examples of this embodiment, candidate propagation paths can be sampled independently and with equal probability in each hyperedge of the graph neural network, and the hyperparameters can be used to... Controlling the proportion of activated candidate propagation paths, so that The proportion is From this, we can obtain... arrive All activation path sets Specifically, timestamps Timestamp The characteristics of the influence are calculated as shown in equations (5) and (6): Equation (5) Equation (6) in, Let be the degree matrix of the supernodes. For timestamps Supernode features, These are the structural parameters for the corresponding candidate propagation paths. These are the path weights obtained through softmax.

[0077] The aforementioned hyperedge can refer to the connecting edge between nodes.

[0078] Ultimately, timestamp That is, the node features at time step t are obtained by accumulating the effects of the preceding timestamps and processing them through the GELU (Gaussian Error Linear Unit) activation function, as shown in equation (7): Equation (7) Finally, the evolutionary heterogeneity diagram is represented by the following equation (8): Equation (8) This results in a skip structure between timestamps, enabling heterogeneous networks to adaptively capture complex semantic information and achieve dynamic modeling and efficient feature extraction of peptide-microbe-disease associations.

[0079] Through the iterative information transmission described above, the predicted microorganisms corresponding to the antimicrobial peptide samples can be iteratively obtained. Then, based on the microbial samples corresponding to the antimicrobial peptide samples and the predicted microorganisms, a loss function is constructed to obtain the loss value, and the parameters of the heterogeneous grid are updated based on the loss value.

[0080] The updated parameters can be the information transmission mode, as well as the nodes and feature transmission paths targeted in each evolutionary operation.

[0081] Therefore, by continuously learning from positive and negative samples, key high-order semantic structures in peptide-microbe association prediction can be captured, which greatly improves the interpretability and generalization ability of peptide-microbe association prediction in heterogeneous networks.

[0082] After obtaining the above graph neural network, it can be applied to information filtering methods, such as to obtain the microorganisms corresponding to the target antimicrobial peptides, or to obtain the antimicrobial peptides corresponding to the target microorganisms. In this way, the relationship between drugs and microorganisms, as well as the relationship between microorganisms and drugs, can be accurately predicted.

[0083] Please refer to Figure 5 As shown, Figure 5 A flowchart illustrating the steps of an information filtering method based on graph neural networks is shown, as follows: Figure 5 As shown, it includes the following steps: Step S100: Obtain the characteristics of the target substance; The target substances include target antimicrobial peptides or target microorganisms.

[0084] In this embodiment, the target substance can be a target antimicrobial peptide, so that the microorganisms that the target antimicrobial peptide can disinfect can be obtained based on a graph neural network.

[0085] In this embodiment, the target substance can also be the target microorganism. Thus, an antimicrobial peptide capable of eliminating the target microorganism can be obtained based on a graph neural network. Of course, in this case, the sample input to the graph neural network during its training process can be a microbial sample, and the training process is the same as when inputting an antimicrobial peptide sample, which will not be elaborated here.

[0086] The characteristics of the target antimicrobial peptide can be obtained by encoding the sequence of the target antimicrobial peptide.

[0087] Step S200: Based on the graph neural network and the characteristics of the target substance, obtain predictive information that is different from the type of the target substance and has a biological guidance relationship; The graph neural network includes multiple nodes of different types and the connection relationships between multiple nodes. The connection relationships include a first connection relationship between nodes of the same type and a second connection relationship between nodes of different types. The first connection relationship is used to characterize the similarity between nodes, and the second connection relationship is used to characterize the biological guidance relationship between nodes. Among them, multiple nodes include at least microbial nodes and antimicrobial peptide nodes.

[0088] In this embodiment, the graph neural network is a network obtained through the aforementioned training. This network is a heterogeneous network. It is understood that the first connection relationship and the second connection relationship included in the graph neural network can be referred to the foregoing embodiments, and will not be repeated here.

[0089] In this embodiment, a graph neural network is used to predict information that is different from the type of the target substance and has a biologically guiding relationship.

[0090] For example, the predictive information can be microbial information, in which case the target substance can be a target antimicrobial peptide.

[0091] For example, the prediction information can be antimicrobial peptide information, in which case the target substance can be the target microorganism.

[0092] In this embodiment, the features of the target substance can be embedded into the graph neural network. For example, it can be added as a new node in the graph neural network. The graph neural network needs to perform feature aggregation on the multiple nodes currently included based on the optimal information transmission mode and the optimal feature transmission path learned during the training process, so as to find the node with the highest correlation with the newly added node.

[0093] For example, if the target substance is a target antimicrobial peptide, the features of the target antimicrobial peptide are embedded into a graph neural network as a newly added antimicrobial peptide node. Based on the optimal information transmission pattern and optimal feature transmission path learned during training, the graph neural network performs feature aggregation on the multiple nodes currently included, thereby finding the microbial node with the highest correlation to the new antimicrobial peptide node. This microbial node is the microorganism that the target antimicrobial peptide can disinfect, thus realizing the target microorganism for the drug development.

[0094] For example, if the target substance is a target microorganism, the features of the target microorganism are embedded into a graph neural network as a newly added microbial node. Based on the optimal information transmission pattern and optimal feature transmission path learned during training, the graph neural network performs feature aggregation on the multiple nodes currently included, thereby finding the antimicrobial peptide node with the highest correlation to the new microbial node. This antimicrobial peptide node can be considered to be able to disinfect the target microorganism, thus realizing drug prediction for newly emerging microorganisms.

[0095] In this embodiment, the biological guiding relationship can refer to the disinfecting relationship between antimicrobial peptides and microorganisms, such as the inactivation of microorganisms by antimicrobial peptides.

[0096] In some examples, the graph neural network may also include disease nodes, so the graph neural network can also be used to predict the types of diseases that antimicrobial peptides can treat.

[0097] For example, the features of the target antimicrobial peptide can be embedded into a graph neural network as a newly added antimicrobial peptide node. Based on the optimal information transmission pattern and optimal feature transmission path learned during training, the graph neural network aggregates features from multiple nodes to find the disease node with the highest correlation to the new antimicrobial peptide node. This disease node can be considered as the disease that the target antimicrobial peptide can treat, thus realizing the prediction of drugs and diseases.

[0098] Of course, the prediction of diseases requires the participation of microbial nodes, using microorganisms as bridging nodes between antimicrobial peptides and diseases. In the information transmission model, the following seven modes are needed: Information transmission patterns between microorganisms and antimicrobial peptides, information transmission patterns between antimicrobial peptides and microorganisms, information transmission patterns between diseases and microorganisms, information transmission patterns between microorganisms, information transmission patterns between antimicrobial peptides and antimicrobial peptides, information transmission patterns between diseases, and transmission patterns from disease nodes to microbial nodes.

[0099] In this example, when obtaining the prediction information corresponding to the target substance, at least one predicted microorganism and at least one predicted disease information corresponding to the target antimicrobial peptide can be obtained based on the graph neural network and the characteristics of the target substance; and, based on the probability of the predicted microorganism and the probability of the predicted disease, the predicted microorganism that the target antimicrobial peptide can disinfect and the predicted disease information that the target antimicrobial peptide can participate in can be obtained.

[0100] In this example, when the target substance is a target antimicrobial peptide, the graph neural network can simultaneously predict multiple predicted microorganisms that can be disinfected by the target antimicrobial peptide, as well as multiple diseases that can be treated by the target antimicrobial peptide. Each predicted microorganism corresponds to a probability, which represents the probability that the microorganism can be disinfected by the target antimicrobial peptide, or its confidence level. Similarly, each predicted disease corresponds to a probability, which represents the probability that the target antimicrobial peptide can treat the predicted disease, or its confidence level.

[0101] In practice, the predicted microorganism with the highest probability can be used as the target antimicrobial peptide to disinfect and output, and the predicted disease information with the highest probability can be used as the target antimicrobial peptide to treat and output.

[0102] In some embodiments, when obtaining predictive information that is different from the type of the target substance and has a biological guidance relationship based on the graph neural network and the characteristics of the target substance, the characteristics of the target substance can be embedded in the graph neural network; and the graph neural network with the embedded characteristics of the target substance can be iteratively performed on the graph neural network for T time steps to obtain predictive information that has a connection relationship with the characteristics of the target substance. In each evolutionary operation at time step t, the graph neural network obtained at time step t-1 acquires the current information transmission pattern and performs feature transmission on the feature representation of the currently selected partial nodes according to the current information transmission pattern; wherein, feature transmission is used to enable one of the partial nodes to aggregate the features of its neighboring nodes.

[0103] The process in this embodiment can be referred to the training process of the graph neural network described above, and will not be repeated here.

[0104] In one example of this embodiment, the information transmission mode includes a first transmission mode and a second transmission mode. The first transmission mode is used to characterize feature transmission between multiple nodes of the same type, and the second transmission mode is used to characterize feature transmission between nodes of different types.

[0105] Specifically, the first delivery mode includes delivery modes between multiple microbial nodes and delivery modes between multiple antimicrobial peptide nodes; The second delivery mode includes a delivery mode from microbial nodes to antimicrobial peptide nodes, and a delivery mode from antimicrobial peptide nodes to microbial nodes.

[0106] The first transmission mode may further include transmission modes between multiple disease nodes, and the second transmission mode may further include transmission modes from microbial nodes to disease nodes, and transmission modes from disease nodes to microbial nodes.

[0107] In one example of this embodiment, the time step in which the first delivery mode is located is before the time step in which the second delivery mode is located; or, the delivery modes between the microbial nodes and the delivery modes between the antimicrobial peptide nodes intersect between the second delivery modes in different directions.

[0108] This example can be referred to the training process of the graph neural network described above, and will not be repeated here.

[0109] The following is an exemplary description of the information filtering method based on graph neural networks in this embodiment.

[0110] Please refer to Figure 6 As shown, Figure 6 A schematic diagram of the overall process is shown, such as Figure 6 As shown: First, basic data is obtained from the database to construct a heterogeneous network.

[0111] like Figure 6 As shown, the basic data includes data on microorganisms with a disinfect relationship and antimicrobial peptides, as well as data on the pathogenicity relationship between microorganisms and diseases. The sequences of antimicrobial peptides were limited to less than 50. Then, the first similarity between different antimicrobial peptides (Peptid), the second similarity between different microorganisms (Microbe), and the third similarity between different diseases (Disease) were determined.

[0112] like Figure 6 As shown, multiple nodes are created based on multiple associated data, such as microbial node M, antimicrobial peptide node P, and disease node D. Then, based on the associated data, microbial node M and antimicrobial peptide node P with a disinfection relationship are connected, and microbial node M and disease node D with a pathogenic relationship are connected.

[0113] Furthermore, based on the first similarity, antimicrobial peptide nodes with high similarity are connected; based on the second similarity, microbial nodes with high similarity are connected; and based on the third similarity, disease nodes with high similarity are connected.

[0114] This forms a heterogeneous network.

[0115] Next, the heterogeneous network is trained.

[0116] Features of antimicrobial peptide samples are embedded into a heterogeneous network as representations of antimicrobial peptide nodes in the network. The heterogeneous network is based on evolutionary heterogeneity and performs feature aggregation through multiple time steps T iterations.

[0117] The specific process is as described in the foregoing embodiments. In this example, using... Figure 6 As shown, an illustrative example is provided: Specifically, in each timestamp evolution operation, the information transmission pattern By weight Propagation, where i represents a node and t represents a timestamp.

[0118] Define corresponding binary gates for candidate transmission paths F1, F2, and F3 in each information transmission mode. .when The candidate transmission path is selected when the value is 0, and the candidate transmission path is blocked when the value is 0.

[0119] In the evolution operation at timestamp t-1, it can be based on The information transmission mode of the current timestamp is obtained, and then the transmission path for information transmission is selected according to the binary values ​​corresponding to the candidate transmission paths, thereby aggregating the feature representations of the nodes according to the transmission path.

[0120] During the aggregation process, candidate propagation paths can be sampled independently and with equal probability in each hyperedge of the graph neural network, and the hyperparameters can be used to perform the aggregation. Controlling the proportion of activated candidate propagation paths, so that The proportion is For example, such as Figure 6 As shown, we can obtain Figure 6 The aggregation path and aggregation result are shown in section 601.

[0121] Among them, timestamp Timestamp The characteristics of the influence are calculated as shown in equations (5) and (6) above.

[0122] By aggregating information across multiple time steps, key high-order semantic structures in peptide-microbe association prediction can be captured. Specifically, refer to... Figure 6 As shown in Figure 602.

[0123] Subsequently, based on the predicted microorganisms corresponding to the antimicrobial peptide samples and the microbial samples, a loss function is constructed to update the parameters of the heterogeneous network, thereby obtaining a graph neural network.

[0124] The resulting graph neural network is published for inference purposes.

[0125] In reasoning scenarios, the features of new antimicrobial peptides can be embedded into graph neural networks to obtain the microorganisms that the new antimicrobial peptides can disinfect.

[0126] The method in this embodiment has the following advantages: ① By constructing and utilizing heterogeneous networks formed by peptide-microbe association information, cross-species, integrated feature representation learning of antimicrobial peptides and microbial characteristics can be achieved. This framework enables highly robust identification of novel antimicrobial peptides and solves the problem of poor predictive ability of existing methods in cross-species or unrelated microbial environments.

[0127] ② An evolutionary graph strategy is used to automatically search for and mine higher-order semantic relationships in heterogeneous networks. This strategy allows the model to adaptively learn optimal information aggregation patterns and paths (i.e., meta-paths), thereby capturing key higher-order semantic structures in peptide-microbe association prediction. This greatly improves the interpretability and generalization ability of antimicrobial peptide-microbe association prediction in heterogeneous networks.

[0128] ③ By applying a graph neural network to the heterogeneous network and performing cross-species feature fusion operations, microbial-specific information can be incorporated into the representation of antimicrobial peptide nodes, thereby obtaining a robust peptide node embedding representation.

[0129] ④ Based on the robust antimicrobial peptide node embedding representation, the association between antimicrobial peptides and microorganisms is predicted, thereby achieving robust cross-species identification of novel antimicrobial peptides.

[0130] In some embodiments, an electronic device is also provided, comprising: a memory and a processor, the memory being coupled to the processor; wherein the memory stores program instructions that, when executed by the processor, cause the electronic device to perform any of the graph neural network-based information filtering methods described in the foregoing embodiments.

[0131] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0132] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0133] The above provides a detailed description of a display method, apparatus, system, and medium provided by this disclosure. Specific examples have been used to illustrate the principles and implementation methods of this disclosure. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this disclosure. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this disclosure. Therefore, the content of this specification should not be construed as a limitation of this disclosure.

[0134] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0135] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

[0136] The terms "an embodiment," "embodiment," or "one or more embodiments" as used herein mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of this disclosure. Furthermore, please note that the examples of the phrase "in one embodiment" do not necessarily all refer to the same embodiment.

[0137] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this disclosure may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0138] In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This disclosure can be implemented by means of hardware comprising a plurality of different elements and by means of a suitably programmed computer. In a unit claim enumerating a plurality of means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words may be interpreted as names.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure.

Claims

1. An information filtering method based on graph neural networks, characterized in that, The method includes: The characteristics of a target substance are obtained, wherein the target substance includes a target antimicrobial peptide or a target microorganism; Based on graph neural networks and the characteristics of the target substance, predictive information that is different from the type of the target substance and has a biological guidance relationship is obtained; The graph neural network includes multiple nodes of different types and multiple connection relationships between the nodes. The connection relationships include a first connection relationship between nodes of the same type and a second connection relationship between nodes of different types. The first connection relationship is used to characterize the similarity between the nodes, and the second connection relationship is used to characterize the biological guidance relationship between the nodes. Among them, the plurality of nodes include at least microbial nodes and antimicrobial peptide nodes.

2. The information filtering method based on graph neural networks according to claim 1, characterized in that, The step of obtaining predictive information that is different in type from the target substance and has a biologically guiding relationship based on the graph neural network and the features of the target substance includes: Embed the features of the target substance into the graph neural network; The graph neural network embedded with the features of the target substance is iteratively subjected to T time steps of evolutionary operation to obtain predictive information that has a connection relationship with the features of the target substance; In each evolutionary operation at time step t, the graph neural network obtained at time step t-1 obtains the current information transmission mode, and performs feature transmission on the feature representation of the currently selected partial nodes according to the current information transmission mode. The feature passing is used to enable one of the nodes in the subset of nodes to aggregate the features of its neighboring nodes.

3. The information filtering method based on graph neural networks according to claim 2, characterized in that, The information transmission mode includes a first transmission mode and a second transmission mode. The first transmission mode is used to characterize feature transmission between multiple nodes of the same type, and the second transmission mode is used to characterize feature transmission between nodes of different types.

4. The information filtering method based on graph neural networks according to claim 3, characterized in that, The first delivery mode includes a delivery mode between multiple microbial nodes and a delivery mode between multiple antimicrobial peptide nodes; The second delivery mode includes a delivery mode from the microbial node to the antimicrobial peptide node, and a delivery mode from the antimicrobial peptide node to the microbial node.

5. The information filtering method based on graph neural networks according to claim 4, characterized in that, The time step in which the first transmission mode is located is before the time step in which the second transmission mode is located. Alternatively, the transmission patterns between the microbial nodes and the transmission patterns between the antimicrobial peptide nodes may overlap between the second transmission patterns in different directions.

6. The information filtering method based on graph neural networks according to claims 1-4, characterized in that, The graph neural network also includes disease nodes, the target substance includes the target antimicrobial peptide, and the step of obtaining predictive information that is different in type from the target substance and has a biologically guiding relationship based on the graph neural network and the characteristics of the target substance includes: Based on the graph neural network and the characteristics of the target substance, at least one predicted microorganism and at least one predicted disease information corresponding to the target antimicrobial peptide are obtained. Based on the probability of the predicted microorganism and the probability of the predicted disease, information on the predicted microorganisms that the target antimicrobial peptide can disinfect and the predicted diseases that the target antimicrobial peptide can participate in is obtained.

7. The information filtering method based on graph neural networks according to claim 1, characterized in that, The graph neural network is obtained through the following steps: Basic data is obtained from the database, which includes multiple related data, each of which corresponds to a group of microorganisms and antimicrobial peptides with a disinfection relationship; Based on the multiple sets of associated data, an initial heterogeneous network is constructed, which includes multiple initial nodes; The graph neural network is obtained by training an initial heterogeneous network based on multiple datasets. The dataset includes combinations of antimicrobial peptide samples and microbial samples that have a disinfect relationship, as well as combinations of antimicrobial peptide samples and microbial samples that do not have a disinfect relationship.

8. The information filtering method based on graph neural networks according to claim 7, characterized in that, The associated data includes the sequence of the antimicrobial peptide; Based on the multiple sets of associated data, an initial heterogeneous network is constructed, comprising multiple initial nodes, including: Redundant sequences in the antimicrobial peptide sequence are removed from the associated data to obtain new associated data; Determine a first similarity between the antimicrobial peptides in different new association data, and determine a second similarity between the microorganisms in different new association data; The heterogeneous network is constructed based on the first similarity, the second similarity, and the new associated data.

9. The information filtering method based on graph neural networks according to claim 7, characterized in that, The process of training an initial heterogeneous network based on multiple datasets to obtain the graph neural network includes: The features of antimicrobial peptide samples in the dataset are embedded into the heterogeneous network; wherein, the heterogeneous network is used to iteratively perform T time-step evolutionary operations on nodes in the network with the embedded features of the antimicrobial peptide samples. In each t-th time-step evolutionary operation, based on the heterogeneous network obtained at the (t-1)-th timestamp, the current information transmission mode is obtained, and feature transmission is performed on the feature representations of the currently selected subset of nodes according to the current information transmission mode; wherein, the feature transmission is used to enable one node in the subset of nodes to aggregate the features of its neighboring nodes; Obtain the predicted microorganisms that have a disinfect relationship with the antimicrobial peptide sample output by the heterogeneous network; Based on the predicted microorganism and the microbial sample corresponding to the antimicrobial peptide sample, the learning parameters of the heterogeneous network are updated to obtain the graph neural network.

10. An electronic device, characterized in that, include: A memory and a processor, wherein the memory is coupled to the processor; The memory stores program instructions, which, when executed by the processor, cause the electronic device to perform the information filtering method based on graph neural networks as described in any one of claims 1-9.

Images