A syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning

Abstract

The present invention provides a syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning, belonging to the field of bioinformation processing technology based on deep learning. The present invention uses the dual-stream fine-tuning structure in deep transfer learning as the core framework. By establishing two transfer learning tasks: 1) the source domain is disease gene prediction; the target domain is syndrome gene prediction; 2) the source domain is symptom gene prediction; the target domain is syndrome gene prediction. Through training on the source domain task, the problem is transferred to the syndrome gene prediction problem in the target domain, realizing syndrome-gene relationship prediction in the target domain zero-sample scenario. In the main network of transfer learning, through our designed syndrome knowledge graph embedding learning, meta-path semantic embedding learning, and multi-order meta-path embedding aggregation, semantic dependency learning of relational meta-paths is realized. At the same time, a prediction scoring of relationships based on tensor decomposition is designed to realize prediction scoring of syndrome genes.

Description

Technical Field

[0001] The present invention relates to the field of biological information processing technology based on deep learning, and specifically to a syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning. Background Art

[0002] Deciphering the molecular mechanisms of syndromes is a key topic in Traditional Chinese Medicine (TCM) research. In recent years, researchers have conducted research on the biological basis of kidney yang deficiency, spleen qi deficiency, and blood stasis syndromes, primarily through experimental approaches to uncover molecular relationships between syndromes. These approaches, primarily based on systems biology, encompass functional genomics, proteomics, and metabolomics. For example, Zhao Xiaoshan et al. used modern molecular biology techniques to screen differentially expressed genes in patients with kidney deficiency and investigated the intrinsic relationship between these genes and the kidney deficiency phenotype. Jiang et al. analyzed lncRNA expression profiles in patients with pulmonary tuberculosis and found significant differences in lncRNA expression across syndromes. Li Jing et al. conducted controlled studies on patients with different TCM syndromes to explore the relationship between TCM syndromes and molecular biology in patients with chronic atrophic gastritis. They found that patients with both qi and yin deficiency had upregulated EGF levels and EGFR mRNA levels, significantly different from those in other syndrome groups.

[0003] Some scholars have conducted research combining data mining with artificial intelligence algorithms to predict syndrome genes, but this is relatively rare. For example, Zhou Xuezhong et al. proposed a gene relationship mining method to effectively mine syndrome-gene relationships. Dong Hanyang used the Fisher test to generate some syndrome-gene relationships and used consistency analysis and other methods to verify the reliability of the syndrome-gene data. He then proposed PSGene, a syndrome-gene relationship prediction method that combines pre-training and knowledge completion.

[0004] Systems biology research approaches based on syndrome-based animal model experiments require high technical, methodological, and equipment requirements, as well as significant human and material resources, which has limited the further development of molecular syndrome research. Syndrome-based gene methods based on machine learning and deep learning typically require high-quality syndrome-gene relationship data to train the models. However, high-quality syndrome-gene relationship data is currently unavailable, which also limits the performance of such methods. Summary of the Invention

[0005] The purpose of the present invention is to provide a syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning to solve at least one technical problem existing in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning, comprising:

[0008] Obtain the syndrome to be predicted;

[0009] The syndromes are processed based on a pre-trained relationship prediction model to obtain the syndrome-gene relationship; wherein, the training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, finding the multi-order meta-path of the entity pairs from the pre-obtained multi-order meta-path dictionary, and obtaining the embedded representation of the entities in the meta-path, wherein the source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs; the entity representations in the multi-order meta-path are input into the shared weight layer to realize the sharing of neural network parameters in the source domain and target domain; in the bottleneck layer of the neural network, a more compact feature expression is obtained, and the bottleneck layer is followed by an adaptive layer, which is used to calculate the distance loss between the source domain and the target domain as a migration regularization term; and the classification task is completed in the output layer.

[0010] Furthermore, the multi-order meta-path of the entity pair is found from the pre-obtained multi-order meta-path dictionary, including: referencing the attention mechanism to adaptively learn the weights of different meta-paths to obtain the quantitative attention level of each path; wherein, the key K and value V in the attention mechanism are both semantic representation vectors of the meta-path, and Q is the first-order embedding representation of the sample entity pair; in the calculation process of the attention mechanism, the similarity between the query Q and each path representation K is first calculated, and then the score value with the value V is calculated, and the weight of each meta-path is obtained through the Softmax activation function, and then the semantic embedding representations of all meta-paths are weighted summed to obtain the final embedding representation of the sample entity pair.

[0011] Furthermore, the set of all entities is denoted as E * ={e s ,e p ,e d ,e g}, where e s Represents all syndrome entities, e p Represents all symptom entities, e d represents all disease entities, e g Represents all gene entities; the set of all relationship types is denoted as R * ={r s-p ,r s-d ,r p-d ,r p-g ,r d-g ,r g-g}, r s-p 、r s-d 、r p-d 、r p-g 、rd-g 、r g-g They represent syndrome-symptom, syndrome-disease, symptom-disease, symptom-gene, disease-gene, and gene-gene relationships respectively; after obtaining the embedded representation of the meta-path, the contextual semantic learning of the meta-path is performed, that is, the recurrent neural network RNN ​​is used to learn the deep semantic embedding representation of the meta-path.

[0012] Furthermore, we obtained symptom-disease, symptom-gene, and disease-gene relationship data from the SympGAN website, syndrome-symptom and syndrome-disease relationship data from traditional Chinese medicine books, and protein interaction data from the STRING11 database. We integrated the above six relationships and performed standardization checks, removed duplicate data, and encoded corresponding and unified data to obtain a syndrome knowledge graph. We used the knowledge graph embedding learning algorithm TransE to learn the complex relationships between the four entities of disease, symptom, syndrome, and gene in the syndrome knowledge graph, and finally mapped the syndrome, disease, symptom, and gene into a 50-dimensional embedding vector for downstream prediction tasks.

[0013] Furthermore, the loss function of the migration neural network is:

[0014] L u =L c (D s )+λL(D s ,d t )

[0015] Among them, L c (D s ) is the classification loss of the source domain dataset for disease gene prediction, λL(D s ,D t ) is the migration regularization term, λ is an adjustable parameter, L(D s ,D t ) is the loss of the adaptive layer.

[0016] Furthermore, the tensor decomposition algorithm is used to predict and score the relationship:

[0017]

[0018] in, is the head entity embedding of the first-order path, For its tail entity embedding, e r is the meta-path embedding representation after multi-order meta-path aggregation, × n Represents a tensor product.

[0019] In a second aspect, the present invention provides a syndrome-gene relationship prediction system that integrates meta-path semantic dependency and transfer learning, comprising:

[0020] An acquisition module, used to obtain the syndrome to be predicted;

[0021] A processing module is used to process syndromes based on a pre-trained relationship prediction model to obtain syndrome-gene relationships; wherein, the training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, finding the multi-order meta-path of the entity pair from a pre-obtained multi-order meta-path dictionary, and obtaining the embedded representation of the entity in the meta-path, wherein the source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs; inputting the entity representation in the multi-order meta-path into the shared weight layer to realize the sharing of neural network parameters in the source domain and target domain; obtaining a more compact feature expression in the bottleneck layer of the neural network, and the adaptive layer after the bottleneck layer is used to calculate the distance loss between the source domain and the target domain as a migration regularization term; completing the classification task in the output layer.

[0022] In a third aspect, the present invention provides a non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning as described in the first aspect is implemented.

[0023] In a fourth aspect, the present invention provides a computer device comprising a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning as described in the first aspect.

[0024] In a fifth aspect, the present invention provides an electronic device comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for implementing the syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning as described in the first aspect.

[0025] The beneficial effects of the present invention are as follows: a syndrome gene prediction method integrating meta-path semantic dependency and transfer learning is proposed, which migrates the models trained on source domain tasks such as disease gene prediction and symptom gene prediction to the target domain, and realizes syndrome gene relationship prediction in the zero-sample scenario of the target domain; a design including syndrome knowledge graph embedding learning, meta-path semantic embedding learning and multi-order meta-path embedding aggregation is designed, and a transfer learning neural network framework for syndrome gene prediction is constructed, which realizes the semantic dependency learning of relational meta-paths.

[0026] Additional advantages of the present invention will be more clearly given in the following description or learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 This is a flow chart of the syndrome gene prediction method that integrates meta-path semantic dependency and transfer learning as described in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.

[0030] Those skilled in the art will understand that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which this invention belongs.

[0031] It should also be understood that terms, such as those defined in commonly used dictionaries, should be understood to have a meaning consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless as defined herein.

[0032] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a," "an," "said," and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the stated 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, and / or groups thereof.

[0033] In the description of this specification, reference to the terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless otherwise contradictory.

[0034] To facilitate understanding of the present invention, the present invention is further explained below with reference to specific embodiments in conjunction with the accompanying drawings. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0035] Those skilled in the art should understand that the drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily necessary for implementing the present invention.

[0036] The systems biology research methods based on syndrome animal model experiments have high requirements for technology, methods and equipment, and also require a large amount of manpower and material resources, which limits the further development of syndrome molecular research. Syndrome gene methods based on machine learning and deep learning usually require high-quality syndrome gene relationship data to train the model. However, there is currently no high-quality syndrome gene relationship data, which also limits the performance of such methods. The purpose of the present invention is to establish a syndrome gene prediction method that integrates meta-path semantic dependency and transfer learning, that is, using the dual-stream fine-tuning structure in deep transfer learning as the core framework, by establishing two transfer learning tasks, the model trained on source domain tasks such as disease gene prediction and symptom gene prediction is transferred to the target domain, and syndrome gene relationship prediction under the zero-sample scenario of the target domain is realized. In other words, even without syndrome gene relationship data, model training and syndrome gene prediction can be achieved, solving the zero-sample problem of syndrome gene prediction.

[0037] Example 1

[0038] In this embodiment 1, a syndrome-gene relationship prediction system that integrates meta-path semantic dependency and transfer learning is first provided, including: an acquisition module for acquiring the syndrome to be predicted; a processing module for processing the syndrome based on a pre-trained relationship prediction model to obtain the syndrome-gene relationship; wherein the training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, searching for multi-order meta-paths of the entity pairs from a pre-obtained multi-order meta-path dictionary, and obtaining embedded representations of the entities in the meta-paths, wherein the source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs; inputting the entity representations in the multi-order meta-paths into a shared weight layer to achieve parameter sharing of the source domain and target domain neural network; obtaining a more compact feature expression in the bottleneck layer of the neural network, and the bottleneck layer is followed by an adaptive layer for calculating the distance loss between the source domain and the target domain as a transfer regularization term; and completing the classification task in the output layer.

[0039] In this example, the aforementioned system was utilized to implement a syndrome-gene relationship prediction method that integrates meta-path semantic dependencies with transfer learning. Specifically, using the dual-stream fine-tuning architecture of deep transfer learning as the core framework, two transfer learning tasks were established to transfer models trained on source domain tasks, such as disease gene prediction and symptom gene prediction, to the target domain, enabling syndrome-gene relationship prediction in a zero-shot scenario in the target domain. Furthermore, when constructing the transfer learning main network, syndrome knowledge graph embedding learning, semantic meta-path embedding learning, and multi-order meta-path embedding aggregation were designed to achieve semantic dependency learning of relational meta-paths. Furthermore, a tensor-decomposition-based prediction and scoring method for syndrome genes was designed.

[0040] like Figure 1 As shown, in this embodiment, a syndrome gene prediction method is integrated with meta-path semantic dependency and transfer learning. With the dual-stream fine-tuning structure in deep transfer learning as the core framework, two transfer learning tasks are established: 1) the source domain is disease gene prediction; the target domain is syndrome gene prediction; 2) the source domain is symptom gene prediction; the target domain is syndrome gene prediction. Through the training on the source domain task, it is transferred to the syndrome gene prediction problem in the target domain, and the syndrome gene relationship prediction in the zero-sample scenario of the target domain is realized. In the main network of transfer learning, through the syndrome knowledge graph embedding learning designed by us, the semantic embedding learning of meta-path, and the embedding aggregation of multi-order meta-path, the semantic dependency learning of the relational meta-path is realized. At the same time, the prediction scoring of the relationship based on tensor decomposition is designed to realize the prediction scoring of syndrome genes. The details are as follows.

[0041] (1) Syndrome knowledge graph embedding learning

[0042] First, we obtained symptom-disease, symptom-gene, and disease-gene relationship data from the SympGAN website, syndrome-symptom and syndrome-disease relationship data from Traditional Chinese Medicine books, and protein interaction data from the STRING11 database. By integrating these six relationships and performing standardization checks, deduplication, and encoding alignment and unification, we generated a syndrome knowledge graph. We then used the knowledge graph embedding learning algorithm TransE to learn the complex relationships between the four entities in the syndrome knowledge graph. Ultimately, we mapped syndromes, diseases, symptoms, and genes into 50-dimensional embedding vectors for downstream prediction tasks.

[0043] (2) Semantic Embedding Learning of Meta-paths

[0044] The semantic embedding learning module of meta-path includes multi-order meta-path embedding and meta-path semantic learning. For each entity pair sample, its multi-order meta-path needs to be obtained as the supplementary information of the entity pair. A1 first-order paths (entity pairs themselves), a2 second-order paths, and a3 third-order paths are selected respectively. Formally, the set of all entities is denoted as E * ={es ,e p ,e d ,e g}, where e s Represents all syndrome entities, e p Represents all symptom entities, e d represents all disease entities, e g Represents all gene entities. At the same time, the set of all relationship types is recorded as R * ={r s-p ,r s-d ,r p-d ,r p-g ,r d-g ,r g-g}, where the elements represent the syndrome-symptom, syndrome-disease, symptom-disease, symptom-gene, disease-gene, and gene-gene relationships from left to right. Taking the disease-gene relationship as an example, for the input triple (d1, r d-g ,g1),d1∈e d , g1∈e g , the path set corresponding to its first-order element path is itself, that is, S 1 ={(d1,g1)}, the set of second-order meta-paths includes disease-symptom-gene and disease-gene-gene, as follows.

[0045] S 2 ={(d1,p i ,g1)∪(d1,g j ,g1)|p i ∈e p ,g j ∈e g}

[0046] The set of third-order meta-paths includes relationships such as disease-syndrome-symptom-gene, disease-syndrome-disease-gene, etc. Here, taking disease-syndrome-symptom-gene and disease-syndrome-disease-gene as examples, they are represented as follows.

[0047] S 3 ={(d1,s i ,p j ,g1)∪(d1, s m,d n ,g1)|s i ∈e s ,p j ∈e p ,d n ∈e d}

[0048] In addition, S i The path in is represented as The kth entity in the path is represented as Where k = {1, 2, 3, 4}. After obtaining the embedded representation of the meta-path, the contextual semantic learning of the meta-path is performed, that is, the deep semantic embedding representation of the meta-path is learned using the recurrent neural network (RNN). Specifically, for the set S 2 、S 3 Each path element in And their embedded representations are used as the input of RNN, and after passing through the recurrent neural network module of RNN, the contextual semantic embedding representation of each meta-path is obtained.

[0049] (3) Embedded aggregation of multi-order meta-paths

[0050] The embedding aggregation part of the multi-order meta-path uses the attention mechanism to adaptively learn the weights of different meta-paths and obtain the quantitative attention level of each path. Specifically, the key K and value V in the attention mechanism are the semantic representation vectors of the meta-path, and Q is the first-order embedding representation of the sample entity pair. In the calculation process of the attention mechanism, the similarity between the query Q and each path representation K is first calculated, and then the score value with the value V is calculated. The weight of each meta-path is obtained through the Softmax activation function, and then the semantic embedding representations of all meta-paths are weighted and summed to obtain the final embedding representation of the sample entity pair.

[0051] (4) Prediction and scoring of relationships

[0052] The relationship prediction and scoring part requires scoring the predicted relationship. Here, we use the tensor decomposition algorithm to perform relationship prediction and scoring. The scoring function is calculated as follows.

[0053]

[0054] in, is the head entity embedding of the first-order path, For its tail entity embedding, e r is the meta-path embedding representation after multi-order meta-path aggregation, × n Represents a tensor product (shorthand notation along the pattern n).

[0055] (5) Deep Transfer Learning

[0056] In the deep transfer learning phase, we will use the example of disease-gene data as the source domain and syndrome-gene data as the target domain. First, the multi-order meta-paths of entity pairs in both the source and target domains are retrieved from a pre-derived dictionary of multi-order meta-paths, and the embedded representations of the entities in the meta-paths are obtained. The entity representations in the multi-order meta-paths are then fed into the pre-trained backbone network, which requires fine-tuning. This is the shared weight layer of the transfer network, which is composed of source and target domain data. During transfer training, the shared weight layer is fine-tuned at a low learning rate. Sharing weights means that the parameters of this portion of the neural network are fully shared between the source and target domains. Next, the bottleneck layer of the neural network is used to obtain a more compact feature representation, improving training speed. Finally, the output layer completes the classification task. The parameters of the bottleneck and output layers are retrained with random initialization and updated at a high learning rate. Specifically, an adaptive layer is added after the bottleneck layer to calculate the distance loss between the source and target domains, which serves as a transfer regularization term. The loss function of the transfer neural network can be expressed as follows.

[0057] L u =L c (D s )+λL(D s ,D t )

[0058] Among them, L c (D s ) is the classification loss of the source domain dataset for disease gene prediction, λL(D s ,D t ) is the migration regularization term, λ is an adjustable parameter, L(D s ,D t ) is the loss of the adaptive layer.

[0059] Example 2

[0060] This embodiment 2 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning as described above is implemented. The method includes:

[0061] Obtain the syndrome to be predicted;

[0062] The syndromes are processed based on a pre-trained relationship prediction model to obtain the syndrome-gene relationship; wherein, the training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, finding the multi-order meta-path of the entity pairs from the pre-obtained multi-order meta-path dictionary, and obtaining the embedded representation of the entities in the meta-path, wherein the source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs; the entity representations in the multi-order meta-path are input into the shared weight layer to realize the sharing of neural network parameters in the source domain and target domain; in the bottleneck layer of the neural network, a more compact feature expression is obtained, and the bottleneck layer is followed by an adaptive layer, which is used to calculate the distance loss between the source domain and the target domain as a migration regularization term; and the classification task is completed in the output layer.

[0063] Example 3

[0064] This embodiment 3 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning as described above, the method comprising:

[0065] Obtain the syndrome to be predicted;

[0066] The syndromes are processed based on a pre-trained relationship prediction model to obtain the syndrome-gene relationship; wherein, the training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, finding the multi-order meta-path of the entity pairs from the pre-obtained multi-order meta-path dictionary, and obtaining the embedded representation of the entities in the meta-path, wherein the source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs; the entity representations in the multi-order meta-path are input into the shared weight layer to realize the sharing of neural network parameters in the source domain and target domain; in the bottleneck layer of the neural network, a more compact feature expression is obtained, and the bottleneck layer is followed by an adaptive layer, which is used to calculate the distance loss between the source domain and the target domain as a migration regularization term; and the classification task is completed in the output layer.

[0067] Example 4

[0068] This embodiment 4 provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning as described above, the method comprising:

[0069] Obtain the syndrome to be predicted;

[0070] The syndromes are processed based on a pre-trained relationship prediction model to obtain the syndrome-gene relationship; wherein, the training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, finding the multi-order meta-path of the entity pairs from the pre-obtained multi-order meta-path dictionary, and obtaining the embedded representation of the entities in the meta-path, wherein the source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs; the entity representations in the multi-order meta-path are input into the shared weight layer to realize the sharing of neural network parameters in the source domain and target domain; in the bottleneck layer of the neural network, a more compact feature expression is obtained, and the bottleneck layer is followed by an adaptive layer, which is used to calculate the distance loss between the source domain and the target domain as a migration regularization term; and the classification task is completed in the output layer.

[0071] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide the functions for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0075] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solutions disclosed in the present invention without the need for creative work should be included in the scope of protection of the present invention.

Claims

1. A syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning, characterized in that: include: Obtain the syndrome to be predicted; The syndromes are processed based on a pre-trained relationship prediction model to obtain syndrome-gene relationships. The training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, searching for multi-order meta-paths of the entity pairs from a pre-obtained multi-order meta-path dictionary and obtaining embedded representations of the entities in the meta-paths. The source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs. The entity representations in the multi-order meta-paths are input into a shared weight layer to achieve parameter sharing of the neural network in the source and target domains. A more compact feature expression is obtained in the bottleneck layer of the neural network. The bottleneck layer is followed by an adaptive layer to calculate the distance loss between the source and target domains as a migration regularization term. The classification task is completed in the output layer. Among them, the multi-order meta-path of the entity pair is found from the pre-obtained multi-order meta-path dictionary, including: referencing the attention mechanism to adaptively learn the weights of different meta-paths to obtain the quantitative attention level of each path; wherein, the key K and value V in the attention mechanism are both semantic representation vectors of the meta-path, and Q is the first-order embedding representation of the sample entity pair; in the calculation process of the attention mechanism, the similarity between the query Q and each path representation K is first calculated, and then the score value with the value V is calculated. The weight of each meta-path is obtained through the Softmax activation function, and then the semantic embedding representations of all meta-paths are weighted and summed to obtain the final embedding representation of the sample entity pair.

2. The syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning according to claim 1, characterized in that: The set of all entities is denoted as E * ={e s ,e p ,e d ,e g }, where e s Represents all syndrome entities, e p Represents all symptom entities, e d represents all disease entities, e g Represents all gene entities; the set of all relationship types is denoted as R * ={r s-p ,r s-d ,r p-d ,r p-g ,r d-g ,r g-g }, r s-p 、r s-d 、r p-d 、r p-g 、r d-g 、r g-g They represent syndrome-symptom, syndrome-disease, symptom-disease, symptom-gene, disease-gene, and gene-gene relationships respectively; after obtaining the embedded representation of the meta-path, the contextual semantic learning of the meta-path is performed, that is, the recurrent neural network RNN ​​is used to learn the deep semantic embedding representation of the meta-path.

3. The syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning according to claim 2, characterized in that: Symptom-disease, symptom-gene, and disease-gene relationship data were obtained from the SympGAN website, syndrome-symptom and syndrome-disease relationship data were obtained from Traditional Chinese Medicine books, and protein interaction data were obtained from the STRING11 database. The syndrome knowledge graph was obtained by integrating the above six relationships and performing standardization checks, removing duplicate data, and encoding correspondence and unification. The knowledge graph embedding learning algorithm TransE was used to learn the complex relationships between the four entities of disease, symptom, syndrome, and gene in the syndrome knowledge graph, and finally mapped the syndrome, disease, symptom, and gene into a 50-dimensional embedding vector for downstream prediction tasks.

4. The syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning according to claim 1, characterized in that: The loss function of the transfer neural network is: L u =L c (D s )+λL(D s ,D t ) Among them, L c (D s ) is the classification loss of the source domain dataset for disease gene prediction, λL(D s ,D t ) is the migration regularization term, λ is an adjustable parameter, L(D s ,D t ) is the loss of the adaptive layer.

5. The syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning according to claim 1, characterized in that: Use tensor decomposition algorithm to predict and score relationships: in, is the head entity embedding of the first-order path, For its tail entity embedding, e r is the meta-path embedding representation after multi-order meta-path aggregation, × n Represents a tensor product.

6. A syndrome-gene relationship prediction system integrating meta-path semantic dependency and transfer learning, characterized by: include: An acquisition module, used to obtain the syndrome to be predicted; A processing module is used to process syndromes based on a pre-trained relationship prediction model to obtain syndrome-gene relationships; wherein the training of the pre-trained relationship prediction model includes: for source domain entity pairs and target domain entity pairs, searching for multi-order meta-paths of the entity pairs from a pre-obtained multi-order meta-path dictionary, and obtaining embedded representations of the entities in the meta-paths, wherein the source domain entity pairs include disease-gene entity pairs and symptom-gene entity pairs, and the target domain entity pairs include syndrome-gene entity pairs; inputting the entity representations in the multi-order meta-paths into a shared weight layer to achieve parameter sharing of the source domain and target domain neural network; obtaining a more compact feature expression in the bottleneck layer of the neural network, and the adaptive layer after the bottleneck layer is used to calculate the distance loss between the source domain and the target domain as a migration regularization term; and completing the classification task in the output layer; Among them, the multi-order meta-path of the entity pair is found from the pre-obtained multi-order meta-path dictionary, including: referencing the attention mechanism to adaptively learn the weights of different meta-paths to obtain the quantitative attention level of each path; wherein, the key K and value V in the attention mechanism are both semantic representation vectors of the meta-path, and Q is the first-order embedding representation of the sample entity pair; in the calculation process of the attention mechanism, the similarity between the query Q and each path representation K is first calculated, and then the score value with the value V is calculated. The weight of each meta-path is obtained through the Softmax activation function, and then the semantic embedding representations of all meta-paths are weighted and summed to obtain the final embedding representation of the sample entity pair.

7. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning as described in any one of claims 1 to 5 is implemented.

8. A computer device, characterized in that: It includes a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the syndrome-gene relationship prediction method that integrates meta-path semantic dependency and transfer learning as described in any one of claims 1 to 5.

9. An electronic device, characterized in that: include: A processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to execute instructions for implementing the syndrome-gene relationship prediction method integrating meta-path semantic dependency and transfer learning as described in any one of claims 1 to 5.

Images