Excellent multi-effect gene mining method, device, equipment and medium

Abstract

To provide an excellent multi-effect gene mining method, device, equipment and medium related to the technical field of gene mining.SOLUTION: An excellent multi-effect gene mining method comprises the following steps of: selecting entity classes and determining corresponding data attributes and object attributes to construct a character regulation gene ontology model; extracting various entities and relationships among the entities from a multi-source database to construct a triplet, and generating a character regulation gene knowledge graph through multi-source knowledge fusion; constructing a character regulation gene prediction model based on a connection relationship between different entities; and calculating the similarity between the unknown character gene and the different known character genes based on the prediction model, and determining that the unknown character gene is an excellent multi-effect gene.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to the technical field of gene mining, and more particularly to a method, device, instrument and medium for mining superior pleiotropic genes. [Background technology]

[0002] In scientific research on crop breeding, traits are mainly controlled by important functional genes. Mining and functional analysis of pleiotropic genes (when one gene controls two or more traits, it is a pleiotropic gene) that simultaneously have excellent combined traits such as drought resistance and disease resistance is the key to obtaining new varieties with high yield and quality. Traditional genome-wide association study (GWAS) methods and quantitative trait loci (QTL) mapping methods aim to establish the association between genome data (genetic variation loci) and phenotype data (target traits), and both only consider scientific data at the genome level, but do not consider other related data (including pathway data of candidate genes, protein level data, etc.) together, making it difficult to accurately establish the data association between genes and traits, and the accuracy and interpretability of knowledge discovery results are low. Therefore, traditional methods cannot consider multi-dimensional and multi-type scientific data at the same time, and lack indications on the association between genes between species, which poses a challenge to mining excellent pleiotropic genes. Summary of the Invention [Problem to be solved by the invention]

[0003] In view of this, the present invention provides a method, device, apparatus and medium for mining excellent pleiotropic genes to solve the problem of low accuracy and low interpretability in mining excellent pleiotropic genes. [Means for solving the problem]

[0004] In a first aspect, the present invention provides a method for mining superior pleiotropic genes, the method comprising: Selecting an entity class according to a good pleiotropic gene mining purpose, and determining data attributes of different entities in the entity class and object attributes between entities, and constructing a trait-regulating gene body model according to the entity class, data attributes and object attributes; Based on the trait control gene body model, extract various entities and the relationships between those entities from the multi-source database, construct triplets that characterize the relationships between different entities, and perform association and fusion of multi-source knowledge based on the triplets to generate a trait control gene knowledge graph; Constructing a trait-regulating gene prediction model according to the connection relationships between different entities in the trait-regulating gene knowledge graph, and identifying multi-source data required for calculating the trait-regulating gene prediction model; The method includes obtaining multi-source data and inputting the data into a property-regulating gene prediction model to calculate the similarity between a gene with an unknown property and a gene with a different known property; if the similarity exceeds a predetermined threshold, it is predicted that the gene with the unknown property and the gene with the known property are regulated by the same property; and if the predicted properties exceed a predetermined number, the gene with the unknown property is determined to be an excellent pleiotropic gene.

[0005] The method for mining excellent pleiotropic genes according to the embodiment of the present invention comprises: constructing a property-controlling gene body model by selecting an entity class and determining corresponding data attributes and object attributes; extracting various entities and the relationships between these entities from a pre-defined multi-source database based on the property-controlling gene body model to construct triplets between different entities; generating a property-controlling gene knowledge graph based on the triplets; and constructing a property-controlling gene prediction model based on the relationships between different entities to obtain multi-source data required for calculating the property-controlling gene prediction model, thereby calculating the similarity between the unknown property gene and the known property gene; if the similarity exceeds a predetermined threshold, it is predicted that the unknown property gene and the known property gene are controlled by the same property; if the predicted properties exceed a predetermined number, the unknown property gene is considered to be an excellent pleiotropic gene. The present invention constructs and inputs a body model into it, and then generates a knowledge graph, and predicts property-controlling genes based on the knowledge graph, thereby improving the interpretability of the knowledge discovery result, adding consideration of biological semantic features, and further improving the accuracy of excellent pleiotropic gene mining.

[0006] In an alternative embodiment, the entity classes include proteins, genes, features, signal pathways, gene symbols, protein families, structural domains, subcellular localization, cellular components, molecular functions, biological processes, metabolic pathways and enzymes, with proteins, genes and features as central entities, data attributes being properties of the corresponding entities and object attributes being relationships between different entities.

[0007] The present invention selects entity classes and constructs a property-controlling gene body model based on data attributes and object attributes between entity classes, thereby explaining the relationship between entities at an abstraction level through a logical model, and uses the model framework to organize the value ranges of the entities and the core attributes of the entities in the body model, thereby solving the problem of low interpretability of property-controlling gene discovery results.

[0008] In an alternative embodiment, the multi-source database includes a literature database and a field scientific database.

[0009] The present invention solves the problem of the difficulty in discovering excellent pleiotropic genes by integrating the latest subject knowledge within a field with system chemistry subject knowledge through a data layer construction method for multidimensional scientific data in a tissue-related literature database and a field science knowledge base.

[0010] In an alternative embodiment, the process of extracting various entities and the relationships between them from a multi-source database based on a property-regulating gene body model, and constructing a triplet that characterizes the relationship between different entities, includes: using the property entity as a search term, obtaining a protein entity related to the property entity based on a literature database, and checking the relationship between the property entity and the protein entity, and then constructing a protein-related-property triplet; obtaining heterogeneous protein sequences based on each type of field science database, and extracting the homologous proteins and corresponding genes of the protein entity based on the protein sequences, and constructing a protein-homology-protein triplet and a protein-corresponding-gene triplet; obtaining structured data related to the protein entity and gene entity based on each type of field science database, and cleaning the structured data, and then constructing a triplet between proteins, genes, and other entities other than properties based on the shared protein unique identifier attributes in different field science databases.

[0011] The present invention extracts relationships between various entities based on existing data to construct corresponding triplets, and performs multi-source knowledge association and fusion for the triplets, thereby resolving the descriptive information of the same entity or concept from multiple sources to achieve low-redundancy and high-precision merging. Based on the triplets, a trait-regulating gene knowledge graph including all entities is generated by inputting it into the main model, providing data support for the mining of excellent pleiotropic genes.

[0012] In an alternative embodiment, the property-regulating gene prediction model is for calculating the similarity between genes of unknown properties and genes of known properties, and the required multi-source data includes the number of nodes commonly connected between different genes, the number of node types commonly connected between different genes, protein sequence similarity values ​​between proteins corresponding to different genes, the number of nodes commonly connected between proteins corresponding to different genes, and the number of node types commonly connected between proteins corresponding to different genes, and the calculation formula is: S(g1,g2)=C(k)·D(k)·S(p1,p2)·C(m)·D(m), where g1 is a gene with known properties, p1 is a protein corresponding to the gene with known properties, g2 is a gene with unknown properties, p2 is a protein corresponding to the gene with unknown properties, S(g1, g2) is the similarity between the gene with unknown properties and the gene with known properties, C(k) is the number of nodes to which different genes are commonly connected, D(k) is the number of node types to which different genes are commonly connected, S(p1, p2) is the protein sequence similarity value between the proteins corresponding to the different genes, C(m) is the number of nodes to which proteins corresponding to different genes are commonly connected, and D(m) is the number of node types to which proteins corresponding to different genes are commonly connected.

[0013] The present invention proposes a regulatory gene mining method based on the relationship structure between different entities in the knowledge graph, using the attribute information of nodes in the fusion graph and the topological structure information, which can solve the problem of low accuracy of regulatory gene prediction results and ensure the reliability of excellent pleiotropic gene mining results.

[0014] In an optional embodiment, the process of obtaining multi-source data required for calculating the property control gene prediction model includes: querying a first entity node set N(g1) connected to a gene g1 of a known property and a second entity node set N(g2) connected to a gene of an unknown property based on the property control gene knowledge graph; selecting a common entity node in the first entity node set N(g1) and the second entity node set N(g2) as a node commonly connected to a gene of a known property and a gene of an unknown property; and the judgment formula is: k = N(g1) ∩ N(g2), and determining the number C(k) of nodes in which genes of known properties and genes of unknown properties are commonly connected, and dividing the number C(k) based on the entity type of the node to obtain the number D(k) of node types in which genes of known properties and genes of unknown properties are commonly connected; querying a third entity node set N(p1) connected to a protein p1 corresponding to a gene g1 of known properties and a fourth entity node set N(p2) connected to a protein p2 corresponding to a gene g2 of unknown properties based on the property control gene knowledge graph; selecting entity nodes that are commonly present in the third entity node set N(p1) and the fourth entity node set N(p2) as nodes in which proteins corresponding to genes of known properties and genes of unknown properties are commonly connected; m = N(p1) ∩ N(p2), and The method includes determining the number C(m) of nodes in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected, and dividing the number C(m) of node types in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected, and obtaining protein sequences of different proteins, and calculating the similarity S(p1, p2) between the protein sequences.

[0015] The present invention uses a trait-controlling gene knowledge graph to obtain the multi-source data required for the prediction model, and the knowledge graph contains all the node information of the entities. The similarity between genes is calculated based on the common connection nodes of different gene entities. The more common connection nodes there are, the more types of common connection nodes there are, which represents the higher the similarity between genes. This can realize the fusion of inter-species multidimensional scientific data and increase the interpretability of knowledge discovery results.

[0016] In one alternative embodiment, the step of determining the predetermined threshold comprises: The method includes: selecting known gene-association-property triplets from a literature database for a specified year period as a dataset, and calculating similarity scores between genes; dividing the dataset in a specified ratio based on the year in which the genes were first discovered to obtain a training set and a validation set; sorting the gene similarity scores in the training set in ascending order, and traversing from smallest to largest; taking the similarity scores obtained by the traversal as thresholds in order to calculate corresponding Macro-F1 values; and when the Macro-F1 value is selected as the maximum value, taking the corresponding gene entity similarity score as a specified threshold, and performing validation based on the validation set to obtain a final specified threshold.

[0017] The present invention judges the degree of similarity and determines a predetermined threshold for the existing gene-association-characteristic data determined, thereby ensuring that genes with unknown characteristics whose similarity exceeds the predetermined threshold have the same characteristic control effect as genes with known characteristics, thereby improving the accuracy and reliability of excellent pleiotropic gene mining.

[0018] In a second aspect, the present invention provides an apparatus for mining superior pleiotropic genes, the apparatus comprising: A body model construction module for selecting an entity class according to a superior pleiotropic gene mining purpose, determining data attributes of different entities in the entity class and object attributes between entities, and constructing a trait-regulating gene body model according to the entity class, the data attributes and the object attributes; a knowledge graph generating module for extracting various entities and relationships between those entities from a multi-source database based on the trait control gene body model, constructing triplets that characterize the relationships between different entities, and associating and fusing multi-source knowledge based on the triplets to generate a trait control gene knowledge graph; A prediction model construction module for constructing a trait control gene prediction model according to the connection relationships between different entities in the trait control gene knowledge graph, and for identifying multi-source data required for calculating the trait control gene prediction model; and a pleiotropic gene mining module for acquiring multi-source data, inputting the data into a property-controlling gene prediction model to calculate the similarity between a gene with an unknown property and a gene with a different known property, predicting that the gene with the unknown property and the gene with the known property are controlled by the same property if the similarity exceeds a predetermined threshold, and determining that the gene with the unknown property is an excellent pleiotropic gene if the predicted properties exceed a predetermined number.

[0019] The excellent pleiotropic gene mining device according to the embodiment of the present invention selects an entity class and determines corresponding data attributes and object attributes to construct a property-controlling gene body model, extracts various entities and the relationships between them from a pre-defined multi-source database based on the property-controlling gene body model to construct triplets between different entities, generates a property-controlling gene knowledge graph based on the triplets, and constructs a property-controlling gene prediction model based on the relationships between different entities to obtain multi-source data required for calculating the property-controlling gene prediction model, calculates the similarity between genes, and if the similarity exceeds a predetermined threshold, it is predicted that the gene with unknown property and the gene with known property are controlled by the same property, and if the predicted property exceeds a predetermined number, the gene with unknown property is considered to be an excellent pleiotropic gene. The present invention constructs and inputs a body model into it, and then generates a knowledge graph, and predicts property-controlling genes based on the knowledge graph, thereby improving the interpretability of knowledge discovery results, adding consideration of biological semantic features, and further improving the accuracy of excellent pleiotropic gene mining.

[0020] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, the memory and the processor being communicatively connected to each other, computer instructions stored in the memory, and the processor executing the computer instructions to perform the method for mining superior pleiotropic genes according to the first aspect or any one of the corresponding embodiments thereof.

[0021] In a fourth aspect, the present invention provides a computer readable storage medium having computer instructions stored thereon for causing a computer to perform the method for mining superior pleiotropic genes according to the first aspect or any one of the embodiments corresponding thereto. [Brief description of the drawings]

[0022] In order to more clearly describe the specific embodiments of the present invention or the technical solutions of the prior art, the following will briefly describe the drawings necessary for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are only part of the embodiments of the present invention, and those skilled in the art can also obtain other drawings based on these drawings without creative efforts.

[0023] [Figure 1] 1 is a schematic flow chart of a method for mining superior pleiotropic genes according to an embodiment of the present invention. [Diagram 2] 1 is a schematic diagram of a property-regulating gene body model in a method for mining excellent pleiotropic genes according to an embodiment of the present invention; [Diagram 3] FIG. 2 is a schematic diagram of gene similarity calculation in the method for mining superior pleiotropic genes according to an embodiment of the present invention. [Figure 4] FIG. 2 is a schematic diagram of the mining of superior pleiotropic genes in the method for mining superior pleiotropic genes according to an embodiment of the present invention; [Diagram 5] 1 is a block diagram of an apparatus for mining superior pleiotropic genes according to an embodiment of the present invention; [Figure 6] FIG. 2 is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] In order to make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be described clearly and completely below with reference to the drawings of the embodiments of the present invention, and obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Any other embodiments obtained by those skilled in the art without creative efforts based on the embodiments of the present invention are within the protection scope of the present invention.

[0025] The embodiment of the present invention is applied to the scene of pleiotropic trait mining for genes with unknown traits. The embodiment of the present invention provides a method for mining excellent pleiotropic genes, constructs a trait control gene knowledge graph, and performs excellent pleiotropic gene mining based on the knowledge graph, thereby improving gene mining accuracy and increasing interpretability. It should be noted that the steps shown in the flowcharts of the drawings may be performed, for example, in a computer system with a set of computer executable instructions, and although the flowcharts show a logical order, in some cases the steps shown or described may be performed in an order different from that shown here.

[0026] In this embodiment, a method for mining excellent pleiotropic genes is provided, which can be used in the above-mentioned computer. FIG. 1 is a flowchart of the method for mining excellent pleiotropic genes according to an embodiment of the present invention. As shown in FIG. 1, the process includes the following steps S101 to S104.

[0027] In step S101, an entity class is selected according to the objective of excellent pleiotropic gene mining, and data attributes of different entities in the entity class and object attributes between entities are determined. A trait-regulating gene body model is constructed based on the entity class, data attributes and object attributes.

[0028] Specifically, in the embodiment of the present invention, the attribute control gene body model is constructed by, but not limited to, 13 kinds of entity classes, 16 kinds of data attributes and 14 object attributes, where the 13 kinds of entity classes include protein, gene, trait, signal pathway, gene symbol, protein family, structural domain, subcellular location, cellular component, molecular function, biological process, metabolic pathway and enzyme, and protein, gene and trait are the central entities, data attributes are the characteristics of the corresponding entities, and object attributes are the relationships between different entities, and the attribute control gene body model constructed in this way is shown in FIG. Taking the protein type entity as an example, the attribute "associates with" is used to connect the property type with the protein type of known properties, and to establish the relationship between the protein of known properties and the properties. At the same time, attributes describing the data of the protein itself, namely protein ID (protein identity), species, date of creation, function description, phenotype disruption, and PubMed identity (PMID), are added. In addition, the association between two proteins is established as an important object attribute in the body model by the object attribute "homologous to", which is also an important basis for realizing the fusion of inter-species multidimensional scientific data.Based on this, a corresponding relationship between proteins and genes is established, and attributes describing the data of the gene itself, namely, gene ID (gene identity), species, physical location, PANTHER database number (panther identity), and transcript name are added. Establishing a corresponding relationship between proteins and gene symbols by identifying with object attributes is considered to be a key point of cross-species gene function knowledge discovery, but is merely illustrative and not limiting.

[0029] Step S102: Based on the trait control gene body model, various entities and the relationships between those entities are extracted from the multi-source database, triplets characterizing the relationships between different entities are constructed, and multi-source knowledge association and fusion are performed based on the triplets to generate a trait control gene knowledge graph.

[0030] Specifically, in the embodiment of the present invention, it is only described that the property control gene body model has different types of object attributes between different entity types, but it is also necessary to input the property control gene body model to obtain the property control gene knowledge graph. The embodiment of the present invention extracts different entities and the relationships between those entities from the determined existing data in the literature database and the field science database, but is not limited thereto. Here, the literature in the literature database contains the latest achievements in the field, and the extraction of related knowledge in the literature represents the latest knowledge and research progress of the field. This is because the latest achievements and discoveries of scientists are generally published in the literature first. Therefore, the knowledge extracted from the literature is the most novel knowledge in the field. The field science database contains systemized and standardized field knowledge, and reorganizing and integrating this knowledge can improve the efficiency of knowledge extraction and expand the knowledge relationship of the field knowledge graph, which is of great significance to the discovery of field knowledge. The embodiment of the present invention selects the Pubmed literature database and the Unipro field science database, respectively, but in actual operation, tracing back to the Pubmed literature database through the Unipro field science database is merely exemplary and not limited thereto. The fusion of the two types of databases combines the most novel knowledge with standardized and systematized knowledge, which not only improves the efficiency of knowledge extraction, but also expands the knowledge relationships in the field knowledge graph, thereby enabling the discovery of new subject knowledge.

[0031] In an optional embodiment, the embodiment of the present invention uses traits description keywords as search terms to connect to the Pubmed literature database through the Uniprot field scientific database to obtain protein IDs, and further manually checks the relationship between the literature and the traits to establish protein-related-trait triplets.

[0032] In an alternative embodiment, the present invention downloads heterogeneous protein sequences from the Uniport database, and then uses the BLAST calculation tool to calculate the similarity between the heterogeneous protein sequences, thereby obtaining protein-protein homology relationships and protein-gene correspondence relationships, and constructing protein-homology-protein triplets and protein-correspondence-gene triplets.

[0033] In an alternative embodiment, the present invention downloads structured data on related genes and proteins from each type of field scientific database, and performs data cleaning through pandas. Based on the shared protein unique identifier registration numbers in different scientific databases, the association of protein unique identifier attributes is used to construct triplets between proteins, genes, and other entities other than properties.

[0034] In an alternative embodiment, the embodiment of the present invention performs multi-source knowledge association and fusion on the above extracted entity triplets, mainly resolving the description information about the same entity or concept from multiple sources to achieve low redundancy and high accuracy merging. Finally, a trait control gene knowledge graph is formed, which includes 13 entity classes, 16 data attributes, and 14 object attributes, and a part of the trait control gene knowledge graph is shown in Figure 3.

[0035] Step S103: construct a property-regulating gene prediction model according to the connection relationships between different entities in the property-regulating gene knowledge graph, and identify multi-source data required for calculating the property-regulating gene prediction model.

[0036] Specifically, in the embodiment of the present invention, as shown in Figure 3, the connection relationship between different entities is shown in the property control gene knowledge graph, among which the knowledge level structure between genes and properties contains many dimensions of scientific data. Therefore, the embodiment of the present invention summarizes the related features of the multi-dimensional scientific data between genes and properties in the knowledge graph, and proposes a property control gene prediction model for calculating the similarity between genes of unknown properties and genes of known properties, and the required multi-source data includes the number of nodes commonly connected between different genes, the number of node types commonly connected between different genes, the protein sequence similarity value between proteins corresponding to different genes, the number of nodes commonly connected between proteins corresponding to different genes, and the number of node types commonly connected between proteins corresponding to different genes, and the calculation formula is: S(g1,g2)=C(k)·D(k)·S(p1,p2)·C(m)·D(m), where g1 is a gene with known properties, p1 is a protein corresponding to the gene with known properties, g2 is a gene with unknown properties, p2 is a protein corresponding to the gene with unknown properties, S(g1, g2) is the similarity between the gene with unknown properties and the gene with known properties, C(k) is the number of nodes to which different genes are commonly connected, D(k) is the number of node types to which different genes are commonly connected, S(p1, p2) is the protein sequence similarity value between the proteins corresponding to the different genes, C(m) is the number of nodes to which proteins corresponding to different genes are commonly connected, and D(m) is the number of node types to which proteins corresponding to different genes are commonly connected.

[0037] Step S104: Obtain multi-source data and input it into the property-regulating gene prediction model to calculate the similarity between the gene with unknown property and the gene with different known property. If the similarity exceeds a predetermined threshold, it is predicted that the gene with unknown property and the gene with known property are regulated by the same property. If the predicted properties exceed a predetermined number, the gene with unknown property is determined to be an excellent pleiotropic gene.

[0038] Specifically, in the embodiment of the present invention, multi-source data required for calculating the property control gene prediction model is obtained from the property control gene knowledge graph, that is, the number C(k) of nodes commonly connected between genes of known properties and genes of unknown properties and the number D(k) of commonly connected node types, the number C(m) of nodes commonly connected between proteins corresponding to genes of known properties and genes of unknown properties and the number D(m) of commonly connected node types, and the protein sequence similarity value S(p1, p2) between the corresponding proteins are obtained, and the obtaining process is as follows: First, query a first entity node set N(g1) connected to a gene g1 of a known property and a second entity node set N(g2) connected to a gene of an unknown property according to a property control gene knowledge graph; Secondly, a common entity node in the first entity node set N(g1) and the second entity node set N(g2) is selected as a node to which a gene of a known property and a gene of an unknown property are commonly connected, and the judgment formula is k = N(g1) ∩ N(g2), and Third, determine the number C(k) of nodes in which genes with known properties and genes with unknown properties are commonly connected, and divide the number C(k) based on the entity type of the node to obtain the number D(k) of node types in which genes with known properties and genes with unknown properties are commonly connected; Fourth, query a third entity node set N(p1) connected to a protein p1 corresponding to a gene g1 of a known property and a fourth entity node set N(p2) connected to a protein p2 corresponding to a gene g2 of an unknown property based on the property control gene knowledge graph; Fifth, a common entity node in the third entity node set N(P1) and the fourth entity node set N(P2) is selected as a node in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected, and the judgment formula is m = N(p1) ∩ N(p2), and Sixth, determining the number C(m) of nodes in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected, and dividing the number C(m) of node types in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected, based on the entity type of the node; Seventh, obtain the protein sequences of different proteins and calculate the similarity S(p1, p2) between the protein sequences.

[0039] In an optional embodiment, the similarity between a gene with an unknown property and a gene with a known property is calculated based on a property control gene prediction model, and the higher the score, the higher the closeness between the gene and the gene with a known property, and the higher the possibility of being regulated by the same property. In an embodiment of the present invention, when the similarity exceeds a predetermined threshold, the gene with an unknown property is preset to be regulated by the same property as the gene with a known property. After multiple mining, when the mined properties exceed a predetermined number, the gene with the unknown property is regarded as a superior pleiotropic gene, and in an embodiment of the present invention, a gene with an unknown property having two or more properties is regarded as a superior pleiotropic gene, but is not limited thereto. Here, the process of determining the predetermined threshold is as follows.

[0040] First, a known gene-association-trait triplet is selected as a dataset from a literature database for a specific year period, and a similarity score between genes is calculated. In the embodiment of the present invention, an experiment was performed by selecting a gene-trait association dataset extracted from literature data from 1988 to 2023, but the present invention is not limited to this.

[0041] Second, the dataset is divided in a predetermined ratio based on the year the gene was first discovered to obtain a training set and a validation set. In the embodiment of the present invention, the dataset is divided in a predetermined ratio of 8:2 based on the results of previous research, that is, the gene-trait association dataset extracted from the literature from 1988 to 2017 is selected as the training set, and the gene-trait association dataset from 2018 to 2023 is selected as the validation set.

[0042] Third, the similarity scores of the genes in the training set are sorted in ascending order, and then traversed from smallest to largest. The similarity scores obtained by traversing are used as thresholds in order to calculate the corresponding Macro-F1 values. The formula for calculating the Macro-F1 value is: JPEG2025036347000002.jpg3960, where TP represents an actual positive sample that is predicted as a positive sample, FP represents an actual negative sample that is predicted as a positive sample, FN represents an actual positive sample that is predicted as a negative sample, TN represents an actual negative sample that is predicted as a negative sample, Precision represents the precision rate, and Recall represents the recall rate.

[0043] Fifth, when the Macro-F1 value is selected as the maximum value, the corresponding gene entity similarity score is set as the predetermined threshold, and the verification is performed based on the verification set to obtain the final predetermined threshold. By performing verification using the literature dataset from 2018 to 2023, the genes whose weight scores in the verification set are equal to or greater than the threshold are screened and predicted as characteristic control genes, and the accuracy is evaluated to verify the scientific validity and validity of the model, so that the predetermined threshold in the embodiment of the present invention is determined to be 2009.44, but is not limited thereto.

[0044] In an alternative embodiment, the present invention performs similarity calculation for gene entity LOC_Os05g12260 and gene entity LOC_Os02g15640 based on the trait-regulating gene prediction model. As can be seen from the part of the trait-regulating gene knowledge graph shown in Figure 3, the number of overlapping nodes between the two gene entities is 13, the number of categories of the overlapping nodes is 4, the corresponding proteins are Q6I5C3 and Q6EN42, the protein sequence similarity between the proteins is 79.191, the number of overlapping nodes between the two protein entities is 2, and the number of categories of the overlapping nodes is 2. Based on the trait-regulating gene prediction model, the similarity score between the two gene entities is obtained to be 13*4*79.19*2*2=16471.72, which is significantly larger than the gene similarity threshold value of 2009.44. Therefore, it was speculated that the gene LOC_Os02g15640 might also be related to the trait of drought tolerance, and in 2019, scientists proved that the rice gene LOC_Os02g15640 is definitely related to the trait of drought tolerance (Haneta l., 2019), which further demonstrated the scientific validity of the trait control gene prediction model.

[0045] In an alternative embodiment, as shown in FIG. 4, the present invention is exemplified by mining a superior pleiotropic wheat gene, TraesCS2A02G338300. According to literature reports, it has been proven that the Arabidopsis gene AT5G39610 is related to the trait of salt resistance (He XJ, 2010), the rice gene LOC_Os11g03370 is related to the traits of drought resistance and salt tolerance (Zheng X, 2009), the rice gene LOC_Os04g38720 is related to the traits of grain weight and pant height (Jiang, Dagang, Chen, 2018; Chen X, Lu S, Wang Y, 2015), and the rice gene LOC_Os12g03050 is related to the traits of disease resistance (Chromosomes TR, Consortia S, 2005). Based on the trait control gene prediction model, gene similarity calculation was performed, and the calculation results showed that the similarity scores between the wheat gene TraesCS5B02G054200 and these four genes all significantly exceeded the threshold, so it was predicted that the gene may have the function of a pair of genes affecting multiple traits, and is an excellent pleiotropic gene, that is, the gene may control five traits, including grain weight, salt tolerance, drought resistance, disease resistance and stem length. Currently, literature reports have proven that the wheat gene TraesCS2A02G338300 is definitely related to the trait of drought tolerance (Lv S, Guo H, Zhang M, 2020).

[0046] In an alternative embodiment, the present invention further verifies the accuracy of the discovery results of the discovered superior pleiotropic genes through molecular biology experimental methods such as gene expression levels at the transcription level, protein content measurements at the protein level, and haplotype analysis.

[0047] The method for mining excellent pleiotropic genes according to the embodiment of the present invention comprises: constructing a property-controlling gene body model by selecting an entity class and determining corresponding data attributes and object attributes; extracting various entities and the relationships between these entities from a pre-defined multi-source database based on the property-controlling gene body model to construct triplets between different entities; generating a property-controlling gene knowledge graph based on the triplets; constructing a property-controlling gene prediction model based on the relationships between different entities; obtaining multi-source data required for calculating the property-controlling gene prediction model to calculate the similarity between genes; if the similarity exceeds a predetermined threshold, it is predicted that the gene with unknown property and the gene with known property are controlled by the same property; if the predicted property exceeds a predetermined number, the gene with unknown property is regarded as an excellent pleiotropic gene. The present invention constructs and inputs a body model into it, and then generates a knowledge graph, and predicts property-controlling genes based on the knowledge graph, thereby improving the interpretability of knowledge discovery results, adding consideration of biological semantic features, and further improving the accuracy of excellent pleiotropic gene mining.

[0048] This embodiment provides an excellent pleiotropic gene mining device, which includes a main model construction module 501, a knowledge graph generation module 502, a prediction model construction module 503, and a pleiotropic gene mining module 504, as shown in FIG.

[0049] The body model construction module 501 is for selecting an entity class according to an excellent pleiotropic gene mining objective, determining data attributes of different entities in the entity class and object attributes between entities, and constructing a trait-regulating gene body model based on the entity class, data attributes and object attributes.

[0050] The knowledge graph generation module 502 is for extracting various entities and the relationships between those entities from the multi-source database based on the trait control gene body model, constructing triplets characterizing the relationships between different entities, and performing association and fusion of multi-source knowledge based on the triplets to generate a trait control gene knowledge graph.

[0051] The prediction model construction module 503 is for constructing a trait-regulating gene prediction model based on the connection relationships between different entities in the trait-regulating gene knowledge graph, and for identifying multi-source data required for the calculation of the trait-regulating gene prediction model.

[0052] The pleiotropic gene mining module 504 acquires multi-source data and inputs it into a trait-regulating gene prediction model to calculate the similarity between a gene with an unknown trait and a gene with a different known trait; if the similarity exceeds a predetermined threshold, it is predicted that the gene with an unknown trait and the gene with a known trait are regulated by the same trait; if the predicted traits exceed a predetermined number, the gene with an unknown trait is determined to be an excellent pleiotropic gene.

[0053] The further functional description of each of the above modules and units is similar to the corresponding embodiment above, and detailed description thereof is omitted here.

[0054] The superior pleiotropic gene mining apparatus in this embodiment is shown in the form of a functional unit, where a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory executing one or more software or fixed programs, and / or other devices that can provide the above functions.

[0055] An embodiment of the present invention further provides a computer device, which comprises an excellent pleiotropic gene mining device as shown in FIG.

[0056] Please refer to FIG. 6, which is a structural schematic diagram of a computer device according to an alternative embodiment of the present invention. As shown in FIG. 6, the computer device includes one or more processors 10, a memory 20, and an interface for connecting each component, the interface including a high-speed interface and a low-speed interface. Each component is communicatively connected to each other via different buses and can be mounted on a common motherboard, or can be mounted in other manners as needed. The processor can process instructions executed in the computer device, including instructions stored in the memory or instructions for displaying graphical information of a GUI on an external input / output device in the memory (e.g., a display device coupled to the interface). In some alternative embodiments, multiple processors and / or multiple buses may be used with multiple memories as needed. Similarly, multiple computer devices may be connected, each device providing a part of the required operations (e.g., as a server array, a set of blade servers, or a multi-processor system). In FIG. 6, one processor 10 is taken as an example.

[0057] The processor 10 may be a central processing unit, a network processor, or a combination thereof. Wherein, the processor 10 may further include a hardware chip. The hardware chip may be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0058] Here, the memory 20 stores instructions executable by the at least one processor 10, thereby causing the at least one processor 10 to execute and realize the methods shown in the above embodiments.

[0059] The memory 20 may include a storage program area and a storage data area, where the storage program area may store an operating system, an application program necessary for at least one function, and the storage data area may store data generated based on the use of the computer device, etc. The memory 20 may also include high-speed random access memory and may further include non-transitory memory, such as at least one magnetic disk memory device, flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 may optionally include memory located remotely relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0060] The memory 20 may include volatile memory, such as random access memory, or the memory may include non-volatile memory, such as flash memory, a hard disk, or a solid state drive, or the memory 20 may further include a combination of the above types of memory.

[0061] The computing device further comprises a communications interface 30 through which the computing device can communicate with other devices or with a communications network.

[0062] The embodiments of the present invention further provide a computer readable storage medium, and the above-mentioned methods according to the embodiments of the present invention may be implemented in hardware, firmware, or implemented so as to be recorded in a storage medium, or implemented as computer code that may be downloaded from a network, initially stored in a remote storage medium or a non-transitory machine-readable storage medium and then stored in a local storage medium, so that the methods described herein may be processed in software as stored in a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware, where the storage medium may be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid state drive, and further, the storage medium may further include a combination of the above types of memory. As can be understood, the computer, processor, microprocessor controller, or programmable hardware may include a storage component capable of storing or receiving the software or computer code, and when the software or computer code is accessed and executed by the computer, processor, or hardware, it realizes the methods shown in the above-mentioned embodiments.

[0063] Although the embodiments of the present invention have been described with reference to the drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and all such modifications and variations are included in the scope defined by the appended claims.

Claims

1. A method for mining superior pleiotropic genes, comprising: Select an entity class according to the purpose of excellent pleiotropic gene mining, and determine the data attributes of different entities in the entity class and the object attributes between entities. Construct an attribute-regulated gene body model based on the entity class, the data attributes and the object attributes, the entity class includes protein, gene, attribute, signal pathway, gene symbol, protein family, structural domain, subcellular localization, cellular component, molecular function, biological process, metabolic pathway and enzyme, and the protein, gene and attribute are the central entities, the data attributes are the characteristics of the corresponding entities, and the object attributes are the relationships between different entities; According to the property control gene body model, extract various entities and relationships between those entities from a multi-source database to construct triplets that characterize the relationships between different entities, and perform association and fusion of multi-source knowledge according to the triplets to generate a property control gene knowledge graph; A property control gene prediction model is constructed according to the connection relationship between different entities in the property control gene knowledge graph, and multi-source data required for calculation of the property control gene prediction model is confirmed, and the property control gene prediction model is for calculating the similarity between genes of unknown properties and genes of known properties, and the required multi-source data includes the number of nodes commonly connected between different genes, the number of node types commonly connected between different genes, protein sequence similarity values ​​between proteins corresponding to different genes, the number of nodes commonly connected between proteins corresponding to different genes, and the number of node types commonly connected between proteins corresponding to different genes, and the calculation formula is: S(g1, g2)=C(k) D(k) S(p1, p2) C(m) D(m), where g1 is a gene with known properties, p1 is a protein corresponding to the gene with known properties, g2 is a gene with unknown properties, p2 is a protein corresponding to the gene with unknown properties, S(g1, g2) is the similarity between the gene with unknown properties and the gene with known properties, C(k) is the number of nodes to which different genes are commonly connected, D(k) is the number of node types to which different genes are commonly connected, S(p1, p2) is a protein sequence similarity value between the proteins corresponding to the different genes, C(m) is the number of nodes to which proteins corresponding to different genes are commonly connected, and D(m) is the number of node types to which proteins corresponding to different genes are commonly connected; obtaining the multi-source data, inputting the data into the property-controlling gene prediction model to calculate the similarity between the gene with the unknown property and the gene with a different known property; if the similarity exceeds a predetermined threshold, predicting that the gene with the unknown property and the gene with the known property are controlled by the same property; and if the predicted properties exceed a predetermined number, determining that the gene with the unknown property is an excellent pleiotropic gene.

2. 2. The method of claim 1, wherein the multi-source databases include a bibliographic database and a field scientific database.

3. The step of extracting various entities and relationships between those entities from a multi-source database based on the property control gene body model, and constructing triplets that characterize relationships between different entities, comprises: Using the property entity as a search term, obtain protein entities related to the property entity based on a literature database, and construct a triplet of protein-association-property after examining the relationship between the property entity and the protein entity; Obtaining heterogeneous protein sequences according to various types of field scientific databases, and extracting homologous proteins and corresponding genes of protein entities according to the protein sequences to construct protein-homology-protein triplet and protein-correspondence-gene triplet; The method according to claim 2, further comprising: obtaining structured data related to protein entities and gene entities based on each type of field scientific database; cleaning the structured data; and constructing triplets between proteins, genes and other entities other than properties based on shared protein unique identifier attributes in different field scientific databases.

4. The step of acquiring the multi-source data includes: Querying a first entity node set N(g1) connected to a gene g1 of a known property and a second entity node set N(g2) connected to a gene of an unknown property according to the property-controlling gene knowledge graph; A common entity node in the first entity node set N(g1) and the second entity node set N(g2) is selected as a node to which a gene of a known property and a gene of an unknown property are commonly connected, and the judgment formula is k = N(g1) ∩ N(g2), and determining the number C(k) of nodes to which genes with known properties and genes with unknown properties are commonly connected, and dividing the number C(k) based on the entity type of the node to obtain the number D(k) of node types to which genes with known properties and genes with unknown properties are commonly connected; Querying a third entity node set N(p1) connected to a protein P1 corresponding to a gene g1 of a known property and a fourth entity node set N(p2) connected to a protein p2 corresponding to a gene g2 of an unknown property based on the property-regulating gene knowledge graph; A common entity node in the third entity node set N (P1) and the fourth entity node set N (P2) is selected as a node in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected, and the judgment formula is m = N(p1) ∩ N(p2), and determining the number C(m) of nodes in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected, and dividing the number C(m) based on the entity type of the node to obtain the number D(m) of node types in which proteins corresponding to genes with known properties and genes with unknown properties are commonly connected; 2. The method of claim 1, further comprising: obtaining protein sequences of different proteins; and calculating a similarity S(p1, p2) between the protein sequences.

5. The step of determining the predetermined threshold value includes: Selecting a dataset of known gene-association-trait triplets from a literature database for a given year period, and calculating a similarity score between genes; Dividing the dataset into a predetermined ratio based on the year the gene was first discovered to obtain a training set and a validation set; Sorting the similarity scores of the genes in the training set in ascending order, and traversing from smallest to largest, and thresholding the similarity scores obtained by the traversal in order to calculate the corresponding Macro-F1 values; The method according to claim 3, further comprising: when the Macro-F1 value is selected as the maximum value, a corresponding Gene Entity Similarity score is set as a predetermined threshold, and validation is performed based on the validation set to obtain a final predetermined threshold value.

6. A mining apparatus for excellent pleiotropic genes, the apparatus comprising: a body model construction module for selecting an entity class according to the purpose of excellent pleiotropic gene mining, determining data attributes of different entities in the entity class and object attributes between entities, and constructing a property-regulating gene body model based on the entity class, data attributes and object attributes, the entity class including protein, gene, property, signal pathway, gene symbol, protein family, structural domain, subcellular localization, cellular component, molecular function, biological process, metabolic pathway and enzyme, the protein, gene and property being the central entity, the data attributes being the characteristics of the corresponding entity, and the object attributes being the relationship between different entities; a knowledge graph generating module for extracting various entities and relationships between those entities from a multi-source database based on the property-controlling gene body model, constructing triplets that characterize relationships between different entities, and associating and fusing multi-source knowledge based on the triplets to generate a property-controlling gene knowledge graph; The property control gene prediction model is constructed based on the connection relationship between different entities in the property control gene knowledge graph, and multi-source data required for calculating the property control gene prediction model is confirmed. The property control gene prediction model is for calculating the similarity between a gene of unknown property and a gene of known property, and the required multi-source data includes the number of nodes commonly connected between different genes, the number of node types commonly connected between different genes, protein sequence similarity values ​​between proteins corresponding to different genes, the number of nodes commonly connected between proteins corresponding to different genes, and the number of node types commonly connected between proteins corresponding to different genes, and the calculation formula is: S(g1, g2)=C(k) D(k) S(p1, p2) C(m) D(m), a prediction model construction module, wherein g1 is a gene with known properties, p1 is a protein corresponding to the gene with known properties, g2 is a gene with unknown properties, p2 is a protein corresponding to the gene with unknown properties, S(g1, g2) is the similarity between the gene with unknown properties and the gene with known properties, C(k) is the number of nodes at which different genes are commonly connected, D(k) is the number of node types at which different genes are commonly connected, S(p1, p2) is a protein sequence similarity value between the proteins corresponding to the different genes, C(m) is the number of nodes at which proteins corresponding to different genes are commonly connected, and D(m) is the number of node types at which proteins corresponding to different genes are commonly connected; and a pleiotropic gene mining module for acquiring the multi-source data, inputting the data into the property-controlling gene prediction model to calculate the similarity between the gene with the unknown property and the gene with a different known property, predicting that the gene with the unknown property and the gene with the known property are controlled by the same property if the similarity exceeds a predetermined threshold, and determining that the gene with the unknown property is an excellent pleiotropic gene if the predicted properties exceed a predetermined number.

7. 1. A computing device comprising: A computer device comprising a memory and a processor, the memory and the processor being communicatively connected to each other, computer instructions being stored in the memory, and the processor executing the computer instructions to perform the method for mining excellent pleiotropic genes described in any one of claims 1 to 5.

8. 1. A computer-readable storage medium, comprising: A computer-readable storage medium, characterized in that computer instructions are stored in the computer-readable storage medium, and the computer instructions are for causing a computer to execute the method for mining excellent pleiotropic genes described in any one of claims 1 to 5.

Images