Protein structural domain ontology knowledge semantic annotation method and online analysis engine system

Abstract

The invention relates to a protein structural domain ontology knowledge semantic annotation method and an online analysis engine system, and provides a protein structural domain semantic annotation map for a user by integrating protein structural domain data types with different definitions. According to the semantic annotation method, a mapping matrix between a structural domain / superstructural domain and an ontology term is constructed by using protein structural domain information and protein ontology annotation; performing overall and relative statistical inference based on the mapping matrix, and retaining the most relevant ontology terms; and obtaining the ontology annotation taking the structural domain as the center by following a real path rule, and for each structural domain / superstructural domain, screening an association relationship between terms according to the obtained overall and relative statistical inference results to form an ontology annotation configuration file. Compared with the prior art, the method has the advantages that a user is supported to carry out semantic annotation on the corresponding protein structural domain, meanwhile, enrichment analysis of ontology knowledge is supported, and powerful support is provided for exploration in the field of semantic annotation of the protein structural domain.

Description

Technical Field

[0001] The present invention relates to the technical field of protein domain ontology knowledge analysis, and in particular to a method for semantic annotation of protein domain data and ontology knowledge and an analysis engine system. Background Art

[0002] In the vast field of molecular biology, accurate annotation of protein function is undoubtedly the key to understanding the essence of life and has great biomedical and pharmaceutical significance. Although molecular experiments can provide the most reliable protein function annotations, due to their relatively low throughput and limited scope, computational prediction has gradually emerged in functional annotation. Accurate computer-generated functional annotations of biological macromolecules provide biologists with a powerful tool to quickly identify the potential role of proteins in biological processes or pathways. In recent years, with the continuous emergence of new technologies, computational prediction of protein structure has become feasible, especially after the advent of the AI ​​tool AlphaFold2, which quickly analyzed almost all known proteins on the earth and generated massive protein data. And the recently released AlphaFold3 successfully predicted protein structure and interaction with unprecedented accuracy. At the same time, with the rapid development of sequencing technology, the demand for genome analysis has grown rapidly, resulting in an urgent need to use computers to annotate the vast majority of proteins. However, the field of protein function annotation still faces huge challenges. How to further improve current computing resources and accurately predict protein functions has become a frontier issue in the current field of computational and molecular biology. To address this challenge, the Critical Assessment of Functional Annotation (CAFA) challenge was born, dedicated to evaluating and improving computational annotation methods for protein functions to narrow the gap between the ever-expanding molecular database and limited understanding resources. As the basic functional unit of proteins, it is more direct and meaningful to associate terms with individual domains rather than only associating ontological terms with full-length proteins. However, most protein sequences still lack biological annotations. In particular, in terms of ontological knowledge annotation, the study of full-length proteins has dominated, while the study of protein domains has lagged behind, which has also led to a lack of corresponding research tools and data resources in this field. In functional genomics, the progress of domain-centered functional annotation lags significantly behind that of protein-level annotation. Given that more and more ontologies are created and used for full-length protein annotations, and protein domains are usually used as functional and evolutionary units of proteins, playing an important role in protein function prediction and phenotype prediction, it is particularly necessary to conduct comprehensive semantic annotations of protein domains in a timely manner. Therefore, how to effectively conduct comprehensive biological annotations of protein domains and deeply mine ontology and domain resources has become an urgent problem to be solved.

[0003] As an indispensable functional unit of proteins, protein domains have been shown to play an important role in protein function prediction and phenotype prediction. However, there is currently a lack of a database that integrates multiple protein domains and their ontological knowledge and a system that can further conduct in-depth data analysis and mining. 1) Most known protein sequences lack biological annotations. Compared with full-length proteins, protein domains have received less attention in ontological annotations, which to a certain extent limits the in-depth understanding of their functions; 2) In current research, there is no systematic work to effectively integrate multiple protein domains and their related ontological knowledge databases. This fragmented information state not only complicates research work, but also hinders researchers from conducting in-depth research in the field of protein domains; 3) Ensuring the accuracy of protein prediction calculations has always been a huge challenge. Due to the complexity and diversity of protein functions, we need more advanced and accurate computational methods to support our prediction work; 4) Although there are already some databases for storing and querying relevant information about protein domains, their performance still needs to be further improved and perfected. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for semantic annotation of protein domain data and ontological knowledge and an analysis engine system.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] As a first aspect of the present invention, a method for semantic annotation of ontological knowledge of protein domains is provided, the method comprising the following steps:

[0007] Use protein domain information and protein ontology annotations to build a mapping matrix between domains / superdomains and ontology terms;

[0008] Based on the resulting mapping matrix, overall and relative statistical inferences are performed, respectively, while respecting the ontology’s directed acyclic graph, and the most relevant ontology terms are retained;

[0009] The true path rule is followed to obtain a complete domain-centered ontology annotation. For each domain / superdomain, the associations between terms are screened according to the obtained overall and relative statistical inference results to form an ontology annotation profile.

[0010] As a preferred technical solution, the overall and relative statistical inferences described above include the following specific steps:

[0011] Fisher test is used to infer the association between the annotation information of the row ontology terms and the specific domains / superdomains in the mapping matrix, wherein the ontology terms are presented in the form of a directed acyclic graph;

[0012] Determine the best association using the true path rule of the ontology directed acyclic graph;

[0013] The P value measuring the probability of association between ontology and domain / superdomain was calculated, the false discovery rate of multiple hypothesis testing was used to assess the significance of the association between statistical inference results of mapping matrices, and the hypergeometric score was used to quantify the strength of association.

[0014] As a preferred technical solution, the method of determining the best association using the true path rule of the ontology directed acyclic graph is specifically as follows: comparing the importance of each term in two different contexts, one context containing all UniProt proteins available for analysis, and the other context containing only UniProt proteins directly annotated by the parent term;

[0015] If an ontology term and its parent terms are significantly associated with a domain / superdomain in the first context, the overall P value P overall is greater than the threshold; but is not significantly associated with the parent term in the second context, i.e., the relative P value P relative is less than the threshold, only the parent terms are associated with the domain / superdomain.

[0016] As a preferred technical solution, the P value for measuring the probability of association between the ontology and the domain / superdomain is calculated as follows:

[0017] P = max{P overall ,P relative}

[0018] Where P represents the overall P value of the overall enrichment test statistical inference overall and the relative P value P of the relative enrichment test statistic inference relative The maximum P value in represents the statistical significance of the prediction results;

[0019] The overall P value of the overall enrichment test statistic inference is calculated as follows:

[0020]

[0021] Where N represents the number of protein domains that contain at least one ontology and one domain, M represents the number of protein domains that contain at least one domain, K represents the number of protein domains annotated to the ontology, X represents the number of protein domains annotated to the ontology and containing the domain, and P represents the number of protein domains annotated to the ontology and containing the domain. overall represents the expected probability of observing X or more protein domains under the first background hypergeometric distribution;

[0022] The relative P value of the relative enrichment test statistical inference is calculated as follows:

[0023]

[0024] Among them, N pa is the number of protein domains annotated by all direct parents in the ontology in the directed acyclic graph structure, M pa Indicates that N pa The number of protein domains that contain domains after the intersection of protein domain sequences in the protein domain structure, K represents the number of protein domains annotated to the ontology, X represents the number of observed protein domains annotated to the ontology and containing domains, P relative represents the expected probability of observing x or more protein domains under the second background hypergeometric distribution.

[0025] As a preferred technical solution, when a parent node has multiple child node annotation information, the associated information with the minimum FDR and maximum hypergeometric score is selected, and the associated ontology terms are extended to all parent terms to form a complete ontology annotation profile.

[0026] As a preferred technical solution, the method steps are to further optimize the annotation configuration file, and divide the directed acyclic graph structure of the ontology according to the information content of the ontology term, that is, the negative logarithmic transformation of the domain / superdomain frequency annotated with the term, so that each partition is located in a different path but reflects the same or similar specificity:

[0027] Through multiple seeds with increasing information content, a GO slim is generated, which is a subset created by collapsing terms upward into more general and higher-level terms related to a specific context;

[0028] The specificity of the ontology is measured by IC and divided into multiple ontology term levels with increasing granularity. The specific steps are as follows:

[0029] All ontology terms in the directed acyclic graph are unlabeled and the IC values ​​are initialized;

[0030] Identify the terms that are closest to the initialized IC, label the identified terms and their parent and child nodes, and exclude them from further analysis, repeat the first two steps, iteratively identify unlabeled GO terms, until all terms are labeled and have an IC value.

[0031] As a preferred technical solution, the method provides two versions of ontology and protein domain association information: high-quality association information and high-coverage association information; the high-quality association information is based on single-domain proteins with known functions and is supported by causal domain evidence; the high-coverage association information further includes associations supported by statistical disambiguation of multi-domain proteins with known functions.

[0032] As a second aspect of the present invention, there is provided a protein domain ontology knowledge online analysis engine system, the analysis engine system comprising:

[0033] Protein domain and ontology knowledge base, including semantic annotation datasets of ontology knowledge related to different definitions of protein domains and protein domain enrichment;

[0034] The main engine for online analysis of protein domains and ontological knowledge is used to provide online web page data mining of ontological knowledge and protein domains, and to execute the semantic annotation method of ontological knowledge of protein domains as described above;

[0035] The main engine of protein domain prediction based on large language model machine learning obtains the questions that users input, extracts the main information in the text, performs learning and reasoning tasks, and generates text to answer users' questions;

[0036] The human-computer interaction module converts the search terms into data that can be recognized by each engine and displays the results visually.

[0037] As a preferred technical solution, the main engine for online analysis of protein domains and ontological knowledge includes:

[0038] The protein domain and ontological knowledge page search sub-engine is used to receive keywords input by users from the human-computer interaction module to query related protein domain and ontological knowledge resources; perform resource query and matching on the keywords input by users based on the data provided by the ontological knowledge sub-base related to protein domain enrichment; and return the query results such as protein domain or ontological knowledge to the human-computer interaction module in the form of a visually displayed page;

[0039] The ontological knowledge directed acyclic graph annotation information and annotated protein domain browsing sub-engine are used to retrieve the ontological knowledge type selected by the user from the human-computer interaction module, present and data mine the ontological knowledge type selected by the user according to the differently defined protein domain knowledge sub-bases and protein domain enrichment-related ontological knowledge sub-bases, and return the selected results to the human-computer interaction module in the form of visually displayed data;

[0040] The ontological knowledge annotation enrichment analysis sub-engine of the structural domain is used to receive the protein domain list input by the user from the human-computer interaction module; retrieve and perform ontological enrichment analysis on the data of the protein domain list input by the user according to the ontological knowledge sub-base related to protein domain enrichment; and return the retrieval results and ontological enrichment analysis results to the human-computer interaction module in the form of visually displayed data.

[0041] As a preferred technical solution, the large language model machine learning protein domain prediction main engine is equipped with an Evoformer module and a Diffusion Network. Starting from the original atomic coordinates, the diffusion model is learned and trained to simulate the interaction of biological molecules and expand the cycle to the disordered region. By integrating and training the above-mentioned module networks, protein function prediction is finally achieved.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1) The present invention provides a semantic annotation method for the biomedical ontology of protein domains, which is intended to perform functional annotations on protein domains, including superdomains in multiple domains. While respecting the directed acyclic graph of gene ontology, two types of statistical inferences, overall and relative, are performed, and only parent terms are associated with domains / superdomains. This dual constraint ensures that only the gene ontology terms that are most significantly associated with domains / superdomains are retained.

[0044] 2) The present invention follows the true path rule to obtain a complete domain-centric gene ontology annotation significance measure. For each domain / superdomain, the associated gene ontology term (i.e., direct annotation) will be extended to all parent ancestor terms (i.e., inherited annotations), together forming a complete gene ontology annotation profile.

[0045] 3) In order to further optimize the annotation profile, the present invention divides the directed acyclic graph structure of the gene ontology according to the information content of the gene ontology term (i.e., the negative logarithmic transformation of the domain / superdomain frequency annotated with the term), and each partition is located in a different path but can reflect the same or similar specificity. Rather than relying solely on the depth of the ontology graph to define term specificity, the actual usage of the term is fully considered when determining the four-level deep classification of domains / superdomains.

[0046] 4) In addition to the single domain, the present invention further associates the gene ontology terms with superdomains. A superdomain is a combination of two or more continuous domains that can work together. In the present invention, only those fully assigned superdomains without significant gaps between domains are included, and superdomains that are not assigned to known domain regions are excluded, thereby ensuring the accuracy of the annotation. The association of gene ontology with superdomains not only helps us to deeply understand how domain combinations promote functional diversification, but also provides strong support for predicting the functions of multi-domain proteins.

[0047] 5) The main engine of the large language model machine learning protein domain prediction of the present invention will extract the main information in the text based on the principle of machine learning, automatically perform learning and reasoning tasks, and use the previously learned patterns and structures to generate new texts to answer users' questions, thereby timely feedback the results to users; it can provide a convenient, fast and simple protein domain information mining tool for relevant researchers. In addition, the Evoformer module and the Diffusion Network diffusion network are added, starting from the original atomic coordinates, learning and training diffusion models, simulating the interaction of biomolecules, and expanding the cycle for disordered regions to better discover the flexibility and variability of these regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the logical architecture of the semantic annotation engine system for protein domain biomedical ontology of the present invention;

[0049] Figure 2 A flowchart of the specific operation process of an implementation case of the protein domain and ontology knowledge page search sub-engine of the present invention;

[0050] Figure 3 A specific operation process flow chart of an implementation case of the ontological knowledge directed acyclic graph annotation information and annotated protein domain browsing sub-engine of the present invention;

[0051] Figure 4 The present invention is a flowchart of the specific operation process of the implementation case of the ontological knowledge annotation enrichment analysis sub-engine of the structural domain. DETAILED DESCRIPTION

[0052] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0053] Example 1

[0054] As one embodiment of the present invention, this embodiment provides a protein domain and ontological knowledge online analysis engine system, which provides users with a corrected protein domain semantic annotation map by integrating different defined protein domain data types (such as SCOP, Pfam and InterPro). The protein domain and ontological knowledge knowledge base of the present invention contains protein domain data with different definitions and ontological knowledge data related to protein domain enrichment. The enrichment relationship map constructed by various types of annotation data supports users to perform semantic annotations on corresponding protein domains, and supports ontological enrichment analysis.

[0055] like Figure 1As shown, the semantic annotation engine system of the protein domain biomedical ontology includes:

[0056] Protein domain and ontology knowledge base;

[0057] The main online analysis engine of protein domain and ontological knowledge is based on protein domain and ontological knowledge. The first main analysis engine is connected to the protein domain and ontological knowledge base;

[0058] The main prediction engine of protein domains based on large language model machine learning is used to provide online web data mining of ontological knowledge and protein domains based on large language model machine learning knowledge; the second prediction main engine is connected to the protein domain and ontological knowledge base;

[0059] The human-computer interaction module receives the search terms and inputs the search terms into the first annotation main engine or the second annotation main engine, and receives and displays the interpretation results of each annotation main engine.

[0060] Among them, the protein domain and ontological knowledge base includes semantic annotation datasets such as knowledge sub-bases of protein domains with different definitions and ontological knowledge related to protein domain enrichment. The semantic annotation knowledge base specifically includes: knowledge sub-bases of protein domains with different definitions and ontological knowledge sub-bases related to protein domain enrichment;

[0061] The protein domain knowledge sub-base with different definitions includes protein domain data from SCOP superfamily domain, Pfam domain and InterPro domain databases, and integrates multiple protein domain databases to perform comprehensive and complete protein domain queries. It is suitable for the retrieval and data mining functions of protein domain data types.

[0062] The protein domain enrichment-related ontology knowledge sub-base is used to provide ontology knowledge of semantic annotations of protein domains. These ontology knowledge include functions, pathways, transcriptional regulatory factors, molecular markers, phenotypes, diseases, drugs, etc. The ontology sub-base currently supports the following:

[0063] (1) Function: Gene Ontology (GO) (June 2024 version), including the ontology of biological processes (GOBP), the ontology of molecular functions (GOMF), and the ontology of cellular composition (GOCC).

[0064] (2) Pathways: KEGG (version 110.1), REACTOME (version V89), PANTHER (version 19.0), WikiPathways (January 2024 version) and MitoCarta (MitoCarta version 3.0).

[0065] (3) Transcriptional regulatory factors: ENRICHR Consensus TFs (version dated February 2023) and TRRUST (version dated November 23, 2021).

[0066] (4) Molecular markers: MSigDB (October 2023 version).

[0067] (5) Phenotype: Human Phenotype Ontology HPO (June 2024 version), Mammalian Phenotype Ontology MPO (June 2024 version) and other phenotypic and anatomical ontologies of model organisms, such as WormBase (WS292 version), FlyBase (FB2024_03), ZFIN (April 2024 version) and TAIR (May 2024 version).

[0068] (6) Disease: Mondo disease ontology (version June 4, 2024), experimental factor ontology EFO of GWAS disease characteristics (version 3.67.0).

[0069] (7) Drugs: DGIdb Druggable Categories (June 2024 version) and Open TargetsTractability Buckets (June 2024 version).

[0070] The ontological knowledge sub-base related to protein domain enrichment can be applied to the ontological knowledge semantic annotation of protein domain data types.

[0071] The main engine for online analysis of protein domains and ontological knowledge can be used to provide online web page data mining functions for ontological knowledge and protein domains. The main engine for online analysis of protein domains and ontological knowledge specifically includes: a sub-engine for searching protein domains and ontological knowledge by page, a sub-engine for browsing the annotation information of ontological knowledge directed acyclic graphs and annotated protein domains, and a sub-engine for enrichment analysis of ontological knowledge annotations of domains. The above sub-engines complement each other to provide users with a comprehensive, fast and effective query and analysis function.

[0072] The protein domain and ontological knowledge page search sub-engine is used to receive user-input keywords from the human-computer interaction module to query related protein domain and ontological knowledge resources. The user enters keywords of interest, such as protein domain information, SCOP / Pfam / InterPro ID or ontological terms, and the keyword input by the user is queried and matched based on the data provided by the ontological knowledge sub-base related to protein domain enrichment. Finally, the query results such as protein domain or ontological knowledge are returned to the human-computer interaction module in the form of a visually displayed page.

[0073] The ontological knowledge directed acyclic graph annotation information and annotated protein domain browsing sub-engine are used to retrieve the ontological knowledge type selected by the user from the human-computer interaction module, present and data mine the ontological knowledge type selected by the user according to the differently defined protein domain knowledge sub-bases and protein domain enrichment-related ontological knowledge sub-bases, and finally return the selected results to the human-computer interaction module in the form of visually displayed data.

[0074] The ontological knowledge annotation enrichment analysis sub-engine of the structural domain is used to receive the protein domain list input by the user from the human-computer interaction module, and to retrieve and perform ontological enrichment analysis on the data of the protein domain list input by the user according to the ontological knowledge sub-base related to protein domain enrichment, and finally to return the retrieval results and ontological enrichment analysis results to the human-computer interaction module in the form of visually displayed data.

[0075] The human-computer interaction module converts search terms into data that can be recognized by various engines for the web page interaction system, unifies the data format for data analysis, and finally visualizes the results in the form of HTML web pages. Furthermore, the human-computer interaction module is composed of the interaction layer, front-end UI and access layer, supports all browsers on PC and mobile terminals, and supports Window, Linux and MacOS systems.

[0076] The interactive layer is used to convert search terms into data that can be recognized by various engines, unify the data format, and input the converted search terms into the protein domain and ontology knowledge online analysis main engine or the large language model machine learning protein domain prediction main engine located at the back end connected to the interactive layer; the front-end UI layer is used to recognize the input search terms.

[0077] The main engine of protein domain prediction based on large language model machine learning is based on the principle of machine learning. It collects manually annotated protein domains as training sets, systematically learns data to improve performance, finds patterns and rules hidden in the data, and thus predicts new data. It also uses a large amount of text data for training to deepen the model's understanding of human language so that effective conversations and question answering can be carried out. This main engine is a generative question answering engine. Users input the questions they want to query into the main engine, and the model can extract the main information in the text, automatically perform learning and reasoning tasks, and use previously learned patterns and structures to generate new text to answer users' questions, thereby providing timely feedback to users. This main engine can provide relevant researchers with a convenient, fast and simple protein domain information mining tool to achieve functional prediction of protein domains, which is very important for studying molecular mechanisms and structure-based drug design.

[0078] The main engine of protein domain prediction based on large language model machine learning will add Evoformer module (a protein language model) and Diffusion Network. It will start from the original atomic coordinates, learn and train the diffusion model, simulate the interaction of biological molecules, and expand the loop to the disordered regions to better discover the flexibility and variability of these regions. By integrating and training the above functions, protein function prediction will be finally achieved.

[0079] 1) The protein domain semantic annotation database constructed by the present invention is not only large in scale, but also detailed in content. It covers a variety of protein domain types, such as SCOP, Pfam and InterPro, and incorporates rich ontological knowledge data. This comprehensive and detailed data coverage provides a solid foundation for mapping from protein domains to ontological knowledge, ensuring the integrity and comprehensiveness of the semantic annotations of protein domains. It is particularly worth mentioning that the knowledge base of the present invention has the characteristics of open source and scalability, which means that its data resources are not only rich, but also can be continuously expanded and optimized according to actual needs, further meeting the researchers' needs for in-depth exploration of the semantic annotations of protein domains.

[0080] 2) In addition to providing data retrieval and browsing functions, the present invention also gives users the ability to deeply explore the functions and phenotypes of domain data. Through the platform of the present invention, users can examine and analyze personal data from a more comprehensive perspective, thereby more accurately grasping the mechanism of action of protein domains in biological processes. This in-depth data mining function not only provides strong data support for their research work, but also helps users discover new scientific laws.

[0081] 3) The present invention provides a more integrated and user-friendly approach to resource mining. Through an intuitive and easy-to-use interface design and a detailed user manual, users can easily get started and efficiently utilize the various functions of the present invention. Whether it is a novice using it for the first time or an experienced old user, they can easily browse and effectively utilize the resources of the present invention through step-by-step instructions and clear guidance. This user-friendly design not only improves the user experience, but also promotes the integration and efficient discovery of knowledge, providing strong support for researchers' exploration in the field of protein domain semantic annotation.

[0082] Example 2

[0083] As a second aspect of the present invention, the ontological knowledge annotation enrichment analysis sub-engine of the present invention also develops a semantic annotation method for biomedical ontology of protein domains, and the algorithm aims to perform functional annotation on protein domains (including superdomains in multiple domains). Taking gene ontology as an example, the application of other ontologies such as pathway ontology, phenotype ontology, disease ontology and other ontological knowledge in the knowledge sub-base is the same in principle.

[0084] Step 1: Use the protein domain information of SUPERFAMILY and the protein gene ontology annotation of UniProtKB-GOA to construct a mapping matrix between domains / superdomains and gene ontology terms: the columns are specific domains / superdomains, the rows are gene ontology term annotation information, and the annotation information corresponds to the protein domain and ontology knowledge base.

[0085] Step 2: Based on the mapping matrix obtained in step 1, two types of statistical inference, overall and relative, are performed while respecting the directed acyclic graph of the gene ontology. This dual inference ensures that only the most relevant gene ontology terms are retained. The specific steps are as follows: First, the Fisher exact test is used to infer the association between rows (gene ontology term annotation information) and columns (specific domains / superdomains). The terms in the gene are presented in the form of a directed acyclic graph, where a single term is a node and its relationship with the parent term (multiple parent terms are allowed) is regarded as a directed edge. Then, the true path rule of the gene ontology directed acyclic graph is used to determine the best association, that is, to compare the importance of each term in two different backgrounds: one background contains all UniProt proteins available for analysis (which can be annotated by the gene ontology), and the other background contains only UniProt proteins directly annotated by the parent term. If a gene ontology term and its parent term were significantly associated with a domain / superdomain in the first context (overall P value, formula 1), but not significantly associated with the parent term in the second context (relative P value, formula 2), we only associated the parent term with the domain / superdomain. This double constraint ensured that only the gene ontology terms most significantly associated with the domain / superdomain were retained. Then, a P value was calculated to measure the probability of association between the gene ontology and the domain / superdomain, with higher P values ​​indicating stronger predictive power. The false discovery rate (FDR) for multiple hypothesis testing was used, with a screening threshold of less than 10 -3 , FDR is used to evaluate the significance of the association between the above statistical inference results of matrices (Formula 3), and the hypergeometric score (h-score) is used to quantify the strength of the association (Formula 4).

[0086]

[0087] Where N represents the number of proteins that contain at least one gene ontology and one domain in the protein domain, M represents the number of proteins that contain at least one domain in the protein domain, K represents the number of protein domains annotated to gene ontology, X represents the number of protein domains annotated to gene ontology and containing domains, and P represents the number of proteins annotated to gene ontology and containing domains. overall represents the expected probability of observing X or more protein domains under the first background hypergeometric distribution.

[0088]

[0089] Among them, N pa is the number of protein domains annotated by all direct parents of gene ontology in the directed acyclic graph structure, M pa Indicates that N pa The number of protein domains that contain domains after the intersection of protein domain sequences in the protein domains, K represents the number of protein domains annotated to gene ontology, X represents the number of observed protein domains annotated to gene ontology and containing domains, P relative represents the expected probability of observing X or more protein domains under the second background hypergeometric distribution.

[0090] P = max{P overall ,P relative} (3)

[0091] Among them, P represents the maximum P value of the overall enrichment test and the relative enrichment test. The corresponding FDR is calculated to represent the statistical significance of the prediction results.

[0092] h-score=min{S overall ,S relative} (4)

[0093] Among them, h-score represents the strength of association and is used to evaluate the prediction results. overall and S relative Indicates that the scores calculated using z-score are all calculated with the minimum value.

[0094] Step 3: Follow the true path rule to obtain a complete domain-centric GO annotation. For each domain / superdomain, the association relationship between terms is screened according to the FDR and h-score obtained in step 2. For example, when a parent node has multiple child node annotation information, the association information with the smallest FDR and the largest h-score is selected. Finally, the associated GO terms (i.e., direct annotations) will be extended to all parent terms (i.e., inherited annotations) to form a complete GO annotation profile.

[0095] In addition, to further optimize this annotation profile, the directed acyclic graph structure of the gene ontology was partitioned according to the information content of the gene ontology term (i.e., the negative logarithmic transformation of the domain / superdomain frequency annotated to the term), with each partition located in a different path but reflecting the same or similar specificity. Through four seeds with increasing information content, we generated a "GO slim" (a subset created by collapsing terms upward into more general and higher-level terms related to specific contexts in order to improve the ease of operation and efficiency of use of terms, and the terms are manually determined subjectively), and used IC (Information content) to measure the specificity of the gene ontology, which was divided into four levels of gene ontology terms with increasing granularity: highly general, general, specific, and highly specific. The specific steps are as follows: First, all gene ontology terms in the directed acyclic graph are unlabeled, and the IC value is initialized (such as 0.5, 1, 1.5 or 2). Then, the term closest to the initialized IC is identified, and the identified terms and their parent nodes and child nodes are labeled and excluded from further analysis. The first two steps are repeated to iteratively identify unlabeled GO terms until all terms are labeled and correspond to an IC value. Among them, IC = 0.5 is defined as a highly general level, IC = 1 is defined as a general level, IC = 1.5 is defined as a specific level, and IC = 2 is defined as a highly specific level. Since the gene ontology is created for annotating proteins, some parts of its structure may not be as effective as other parts when annotating domains / superdomains. Therefore, the algorithm does not rely solely on the depth of the ontology graph to define term specificity, but fully considers the actual usage of the terms when determining the four-level deep classification of domains / superdomains.

[0096] Relying on the above algorithm, the present invention provides users with two versions of gene ontology and protein domain association information: high-quality association information and high-coverage association information. Among them, high-quality association information is based on single-domain proteins with known functions and is supported by clear causal domain evidence. High-coverage association information further incorporates associations supported by statistical disambiguation of multi-domain proteins with known functions. Although its reliability is slightly inferior to the former, it provides a wider range of functional coverage for large-scale studies. From the perspective of domain center, high-quality association information is undoubtedly more reliable, but high-coverage association information is equally reliable for extensive functional exploration and analysis, and provides users with richer functional annotations. In practical applications, considering the positive impact of annotation coverage on enrichment analysis results, we strongly recommend using high-coverage association information for gene ontology annotation in large-scale studies. For domain-based enrichment analysis, it is strongly recommended to limit the annotation to "GO slim". For example, the ontological annotation enrichment analysis sub-engine of the domain involved in this engine will follow this rule.

[0097] In addition to the single domain, the present invention further associates the gene ontology terms with superdomains. A superdomain is a combination of two or more continuous domains that can work together, which in the present invention only includes fully assigned superdomains without significant gaps between domains, excluding superdomains that are not assigned to known domain regions, thereby ensuring the accuracy of the annotation. The association of gene ontology with superdomains not only helps us to deeply understand how domain combinations promote functional diversification, but also provides strong support for predicting the functions of multi-domain proteins.

[0098] Example 3

[0099] Figure 2 As shown, it is a specific operation process flow chart of the implementation case of the protein domain and ontology knowledge faceted search sub-engine of the present invention, and the implementation steps are as follows:

[0100] Step 3.1: The user enters keywords on the web page (human-computer interaction module). The keyword type can be a protein domain name (such as SCOP, Pfam and InterPro domains) or ontological knowledge;

[0101] Step 3.2: Based on the input keyword type, the protein domain and ontology knowledge faceted search sub-engine starts to perform the search function and feeds the search results back to the front-end layer;

[0102] Step 3.3: Users can browse the search results at the access level. An expansion box will be generated under the search box, which will provide a hyperlink to a specific page of the search results. Clicking on the hyperlink will automatically redirect the user to a web page that not only provides keyword information, but also provides protein domains and ontology knowledge tables, ontology knowledge annotation information, and hyperlink information to a specific page.

[0103] Example 4

[0104] Figure 3 As shown, it is a specific operation process flow chart of an implementation case of the ontological knowledge directed acyclic graph annotation information and annotated protein domain browsing sub-engine of the present invention, and the implementation steps are as follows:

[0105] Step 4.1: The user selects the type of ontological knowledge on the web page (human-computer interaction module);

[0106] Step 4.2: The ontological knowledge directed acyclic graph annotation information and annotated protein domain browsing sub-engine begin to analyze the ontological knowledge selected by the user and feed back the search results to the front-end layer;

[0107] Step 4.3: The user can browse at the access level and can further select and browse the hierarchical pages of sub-category names. SCOP, Pfam and InterPro information are displayed separately;

[0108] Example 5

[0109] like Figure 4 As shown, it is a specific operation process flow chart of an implementation case of the ontological knowledge annotation enrichment analysis sub-engine of the structural domain of the present invention, and the implementation steps are as follows:

[0110] Step 5.1: The user inputs protein domain data on the web page (human-computer interaction module);

[0111] Step 5.2: The user selects the corresponding domain database according to the requirements. Please ensure that the identifier of the input domain is consistent with the level of the domain.

[0112] Step 5.3: The user selects the appropriate knowledge sub-base type from the provided protein domain and ontology knowledge bases;

[0113] Step 5.4: The user selects the parameters required for ontology enrichment analysis and the ontology knowledge sub-base related to protein domain enrichment;

[0114] Step 5.5: The user clicks the SUBMIT button on the web page to call the ontological knowledge annotation enrichment analysis sub-engine of the structural domain for analysis.

[0115] Step 5.6: The ontological knowledge annotation enrichment analysis sub-engine of the structural domain begins to perform enrichment analysis on the data provided by the user and feeds back the enrichment analysis results to the front-end layer.

[0116] Step 5.7: Users can easily browse and download detailed analysis results in the access layer. The analysis results include three interactive tables and high-resolution images of enrichment results: ① Input domain information module, which provides an interactive table that lists the protein domain information entered by the user in detail. This table not only helps users to perform data mining, but also clearly marks the running time in the table header, making it easier for users to understand the analysis. ② Enriched ontology terms module, which provides an enriched ontology knowledge table. Each ontology knowledge provides a hyperlink to a specific information page. The table provides annotation information for the protein domain, including ontology knowledge type information and corresponding regulatory interaction information. ③ Dotplot for enriched ontology terms module, which provides an enrichment result dot plot, which globally displays the enrichment of ontology knowledge and provides users with intuitive visual analysis results. In addition, users can also download high-quality PDF files for further analysis and display locally. Through these three modules, users can fully and deeply understand the function and enrichment of protein domains, providing strong support for scientific research.

[0117] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for semantic annotation of ontological knowledge of protein domains, characterized in that: The method steps include: Use protein domain information and protein ontology annotations to build a mapping matrix between domains / superdomains and ontology terms; Based on the resulting mapping matrix, overall and relative statistical inferences are performed, respectively, while respecting the ontology’s directed acyclic graph, and the most relevant ontology terms are retained; The true path rule is followed to obtain a complete domain-centered ontology annotation. For each domain / superdomain, the associations between terms are screened according to the obtained overall and relative statistical inference results to form an ontology annotation profile.

2. A protein domain ontology knowledge semantic annotation method according to claim 1, characterized in that: The specific steps of the overall and relative statistical inference are as follows: Fisher test is used to infer the association between the annotation information of the row ontology terms and the specific domains / superdomains in the mapping matrix, wherein the ontology terms are presented in the form of a directed acyclic graph; Determine the best association using the true path rule of the ontology directed acyclic graph; The P value measuring the probability of association between ontology and domain / superdomain was calculated, the false discovery rate of multiple hypothesis testing was used to assess the significance of the association between statistical inference results of mapping matrices, and the hypergeometric score was used to quantify the strength of association.

3. A method for semantic annotation of ontological knowledge of protein domains according to claim 2, characterized in that: The method of determining the best association using the true path rule of the ontology directed acyclic graph is as follows: comparing the importance of each term in two different contexts, one context containing all UniProt proteins available for analysis, and the other context containing only UniProt proteins directly annotated by the parent term; If an ontology term and its parent terms are significantly associated with a domain / superdomain in the first context, the overall P value P overall is greater than the threshold; but is not significantly associated with the parent term in the second context, i.e., the relative P value P relative If the value is less than the threshold, only the parent terms are associated with the domain / superdomain.

4. The ontological knowledge semantic annotation method of a protein domain according to claim 3, characterized in that: The P value, which measures the probability of association between an ontology and a domain / superdomain, is calculated as follows: P=max{P overall ,P relative } Where P represents the overall P value of the overall enrichment test statistical inference overall and the relative P value P of the relative enrichment test statistic inference relative The maximum P value in represents the statistical significance of the prediction results; The overall P value of the overall enrichment test statistic inference is calculated as follows: Where N represents the number of protein domains that contain at least one ontology and one domain, M represents the number of protein domains that contain at least one domain, K represents the number of protein domains annotated to the ontology, X represents the number of protein domains annotated to the ontology and containing the domain, and P represents the number of protein domains annotated to the ontology and containing the domain. overall represents the expected probability of observing X or more protein domains under the first background hypergeometric distribution; The relative P value of the relative enrichment test statistical inference is calculated as follows: Among them, N pa is the number of protein domains annotated by all direct parents in the ontology in the directed acyclic graph structure, M pa Indicates that N pa The number of protein domains that contain domains after the intersection of protein domain sequences in the protein domain structure, K represents the number of protein domains annotated to the ontology, X represents the number of observed protein domains annotated to the ontology and containing domains, P relative represents the expected probability of observing X or more protein domains under the second background hypergeometric distribution.

5. The ontology-based knowledge semantic annotation method of a protein domain according to claim 2, characterized in that: When a parent node has multiple child node annotation information, the associated information with the minimum FDR and maximum hypergeometric score is selected, and the associated ontology terms are expanded to all parent terms together to form a complete ontology annotation profile.

6. A method for semantic annotation of ontological knowledge of protein domains according to claim 5, characterized in that: The method steps are to further optimize the annotation configuration file, and partition the directed acyclic graph structure of the ontology according to the information content of the ontology terms, that is, the negative logarithmic transformation of the domain / superdomain frequency annotated with the term, so that each partition is located in a different path but reflects the same or similar specificity: Through multiple seeds with increasing information content, a GO slim is generated, which is a subset created by collapsing terms upward into more general and higher-level terms related to a specific context; IC is used to divide the specificity of the ontology into multiple ontology term levels with increasing granularity. The specific steps are as follows: All ontology terms in the directed acyclic graph are unlabeled and the IC values ​​are initialized; Identify the terms that are closest to the initialized IC, label the identified terms and their parent and child nodes, and exclude them from further analysis, repeat the first two steps, and iteratively identify unlabeled GO terms until all terms are labeled and have unique IC values.

7. The ontology-based knowledge semantic annotation method of a protein domain according to claim 1, characterized in that: The method provides two versions of ontology and protein domain association information: high-quality association information and high-coverage association information; the high-quality association information is based on single-domain proteins with known functions and is supported by causal domain evidence; the high-coverage association information further includes associations supported by statistical disambiguation of multi-domain proteins with known functions.

8. A protein domain ontology knowledge online analysis engine system, characterized in that: The analysis engine system comprises: Protein domain and ontology knowledge base, including semantic annotation datasets of ontology knowledge related to different definitions of protein domains and protein domain enrichment; A main online analysis engine for protein domains and ontological knowledge, used to provide online web data mining of ontological knowledge and protein domains, and to execute the semantic annotation method for ontological knowledge of protein domains as described in any one of claims 1 to 7; The main engine of protein domain prediction based on large language model machine learning obtains the questions that users input, extracts the main information in the text, performs learning and reasoning tasks, and generates text to answer users' questions; The human-computer interaction module converts the search terms into data that can be recognized by each engine and displays the results visually.

9. The protein domain ontology knowledge online analysis engine system according to claim 8, characterized in that: The main engine for online analysis of protein domains and ontological knowledge includes: The protein domain and ontological knowledge page search sub-engine is used to receive keywords input by users from the human-computer interaction module to query related protein domain and ontological knowledge resources; perform resource query and matching on the keywords input by users based on the data provided by the ontological knowledge sub-base related to protein domain enrichment; and return the query results such as protein domain or ontological knowledge to the human-computer interaction module in the form of a visually displayed page; The ontological knowledge directed acyclic graph annotation information and annotated protein domain browsing sub-engine are used to retrieve the ontological knowledge type selected by the user from the human-computer interaction module, present and data mine the ontological knowledge type selected by the user according to the differently defined protein domain knowledge sub-bases and protein domain enrichment-related ontological knowledge sub-bases, and return the selected results to the human-computer interaction module in the form of visually displayed data; The ontological knowledge annotation enrichment analysis sub-engine of the structural domain is used to receive the protein domain list input by the user from the human-computer interaction module; retrieve and perform ontological enrichment analysis on the data of the protein domain list input by the user according to the ontological knowledge sub-base related to protein domain enrichment; and return the retrieval results and ontological enrichment analysis results to the human-computer interaction module in the form of visually displayed data.

10. The protein domain ontology knowledge online analysis engine system according to claim 8, characterized in that: The large language model machine learning protein domain prediction main engine is configured with an Evoformer module and a Diffusion Network. Starting from the original atomic coordinates, the diffusion model is learned and trained to simulate the interaction of biological molecules and to expand the cycle to the disordered region. By integrating and training the above-mentioned module networks, protein function prediction is finally achieved.

Images