CN114927168A - Method for constructing biomechanically regulated bone reconstruction text mining interactive website - Google Patents

Abstract

The invention relates to a construction method of a biomechanically regulated bone reconstruction text mining interactive website, which comprises the following steps: s1, screening gene information text words in the literature according to the related entries, obtaining gene molecule interaction relation pairs, and constructing a literature database; s2, calculating the correlation between the target retrieval factor and the classical mechanics sensitive path by adopting a weight algorithm based on the gene molecule interaction relation pair in the literature database; and S3, performing visual display on the correlation between the target retrieval factor and the classical mechanical sensitive path, displaying gene molecules in the classical mechanical sensitive path as mutually connected nodes, and linking to corresponding documents in a document database by clicking a connecting line between the nodes. The website provided by the invention is based on open literature resources, utilizes a natural language processing strategy to mine a text database, creates a first bone-related biomechanics text database, innovatively introduces a visualization mode, and establishes a brand-new analysis strategy based on a biological pathway.

Description

A method of constructing a text mining interactive website for biomechanical regulation of bone remodeling

technical field

The invention relates to the technical field of biomechanical website construction, in particular to a method for constructing a text mining interactive website for biomechanical regulation and bone remodeling.

Background technique

Congenital hypoplasia, dysplasia, and defect or absence of bone tissue are relatively common clinical problems, which have a great impact on patients' facial appearance, mental health, and quality of life. In this regard, mechanical stimulation, stress-stretching and other treatment methods based on biomechanical principles are currently relatively safe, reliable, efficient and economical countermeasures. Therefore, clarifying the biomolecular mechanism of bone remodeling under mechanical stimulation is the primary prerequisite for the further development of precise and efficient treatment. At present, there is a large amount of research data in the field of biomechanical regulation of bone remodeling, but the information is scattered and difficult to integrate. Therefore, building a knowledge network technology platform to efficiently obtain important information will provide an important means to rapidly promote the research and development of this field.

Elucidating the response process of bone-related cells to biomechanical stimuli is the basic premise of bone physiology and pathology research. The open and shared knowledge platform has greatly promoted the development of modern science, but the ever-increasing number of publications and massive amounts of information make it more and more difficult for researchers to sort out and mine documents manually. In the era of big data, using machine language processing mode and calling natural language processing tools (NLP) to integrate and sort out biomedical related literature is an efficient, reliable and potential application mode.

At present, computer language tools such as Tagger, iTextMine, and Geneshot can be used to distinguish professional terms and specific expressions in biomedical texts, which provides the possibility for computer language processing strategies for biomedical texts. In recent years, LION LBD, GLAD4U, etc. have used NLP tools to conduct biological text mining, integrate and sort out data, and provide research-related information.

However, in the field of bone-related biomechanics research, the above text research tools are difficult to play an effective role, mainly reflected in the following aspects:

1. Programming ability limitation: Most of the existing text processing tools are aimed at users with certain programming ability, such as Tagger, iTextMine, Geneshot, etc., which require users to master a certain knowledge of natural language processing. For most biomedical researchers, operation hard to accomplish.

2. Background database redundancy: Biological processes are precise and conditional. Although existing NLP tools can extract and store a large amount of data information in a structured way, most of them use unfiltered background databases, which will cause irrelevant information. included, resulting in false positive results. For a specific biological field, especially a relatively small research field such as biomechanics, it is difficult to obtain good search results in the pan-medical research background library. Therefore, researchers need a targeted NLP tool that is more suitable for bone-related biomechanical research.

3. Lack of visual display: For the network structure of complex interaction, it is difficult to provide a clear and logical frame structure for plain text information compared with the graphical display method. Intermolecular connections and interactions are sorted out so that researchers can quickly understand pathway information and locate desired targets.

SUMMARY OF THE INVENTION

In order to solve the above technical problems, the present application provides a method for constructing a text mining interactive website for biomechanical regulation and bone reconstruction.

This application is achieved through the following technical solutions:

A method for constructing a text mining interactive website for biomechanical regulation of bone remodeling, the method comprising:

S1, screen gene information text words in the literature according to related entries, obtain gene-molecule interaction relationship pairs, and construct a literature database;

S2, based on the gene-molecular interaction pairs in the literature database, the weighting algorithm is used to calculate the correlation between the target retrieval factor and the classical mechanics-sensitive pathway;

S3. Visually display the correlation between the target retrieval factor and the classical mechanosensitive pathway, and display the gene molecules in the classical mechanosensitive pathway as interconnected nodes. By clicking the connection between the nodes, you can link to the literature corresponding literature in the database.

Further, between the step S1 and the step S2, it also includes performing deep neural network training on the PMC database, screening text keywords with biological information, and constructing a corpus.

Further, the biological information includes mechanical type, research species, and cell type.

Preferably, the related entries in the step S1 include biomechanics and bone related entries.

Further, the step S1 includes computer language normalization and preprocessing on the gene information text words.

Further, the step S1 also includes using PubTator to identify the gene information text words, and converting the gene information text words into official names by calling the API of the NCBI gene database.

Further, the formula of the weight algorithm in the step S2 is:

In the formula, r(g,p) is the correlation coefficient between gene g and the classical mechanosensitive pathway p, Ni represents the total number of related entities in the literature database of the _ith gene in the classical mechanosensitive pathway _p , and Np is the classical mechanosensitive pathway The total number of related entities of all genes of p in the literature database, Ω _g and Ω _p represent the set of gene g and classical mechanosensitive pathway p, respectively.

Preferably, the classical mechanosensitive pathway includes at least one of Hippo, BMP, TGFβ, Wnt, Notch, PI3K/Akt, MAPK, and Ras.

Further, in the step S3, it also includes visually displaying the path information of the target retrieval factor in the KEGG database.

Further, in the step S3, it also includes visually displaying the gene-molecule interaction relationship pairs of the target retrieval factor in the String data.

Compared with the prior art, the present application has the following beneficial effects:

1. Use web tools to provide an open search port, which is convenient for users to customize the search range, and does not require users to master complex computer programming skills.

2. Bone-related biomechanical information was clarified by setting strict inclusion criteria for literature databases. For complex bone-related biomechanical regulatory networks, false positive information can be filtered to a large extent, making the results more credible and effective.

3. Adopt the mode of visual network diagram to ensure the user's interactive operation, make computer literature mining available to researchers, and promote information dissemination and understanding in a more user-friendly mode.

4. Combining classical mechanosensitive pathways with text mining results enables users to locate target genes or gene sets through biological pathways. Based on the literature database and weighting algorithm, the correlation between the target retrieval factor and each classical mechanosensitive pathway is calculated, and the pathway and gene interactive search are provided at the same time, which makes gene navigation more convincing and meaningful.

Description of drawings

The accompanying drawings described herein are used to provide a further understanding of the embodiments of the present application, and constitute a part of the present application, and do not constitute a limitation on the embodiments of the present invention.

Fig. 1 is the flow chart of the present invention;

Fig. 2 is the deep neural network training schematic diagram of corpus;

Fig. 3 is the retrieval window interface diagram of the present invention;

Fig. 4 is the retrieval result interface diagram of the present invention;

Fig. 5 is the schematic diagram of plate 1-3 in Fig. 4;

Fig. 6 is the schematic diagram of plate 4-6 in Fig. 4;

FIG. 7 is a schematic diagram of the plate 7 in FIG. 4 .

Detailed ways

In order to make the purpose, technical solutions and beneficial effects of the present application clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments. Obviously, the described embodiments are some, but not all, embodiments of the present invention. The components of the embodiments of the invention generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations.

1. Database Construction and Text Markup

The website of the present invention is centered on the related research of "biomechanics" and "bone", and takes the keyword set as the collection standard of literature resources. The screening included 34,937 articles published between January 1, 2010 and December 31, 2020. After computer language normalization and preprocessing of text words, the genetic information in each article was first processed by PubTator The identification is then converted to the official name by calling the API of the NCBI gene database.

Ncbi provides the e-utilities api for the ncbi entrez system and allows access to all entrez databases, including pubmed, pmc, gene and protein, which facilitates batch processing and bulk text word retrieval (https://www.ncbi.nlm.nih. gov/home/develop/api/). Text data, consisting of the title and abstract of each article, are first tagged, parsed, and normalized by the text processing library Natural Language Toolkit (nltk, http://www.nltk.org/) to avoid ambiguous descriptions and ensure Identifiability for subsequent processing. Name entity recognition (ner) is then performed to extract the required details of each paper. On the one hand, pubtator (https://www.ncbi.nlm.nih.gov/research/pubtator/), as a mature biomedical term recognition tool, has achieved good results in identifying vague and complex biomedical term names The effect of , was used to annotate the genes and proteins present in the text database. Gene IDs were then converted to standard names based on access to the ncbi gene database by biopython (https://biopython.org/).

On the other hand, for other special terms, such as type of force, cell type, and species, a self-compiled corpus was built to extract information about mechanics type, study species and cell type. A comparison is made to identify the name entity. Through self-built corpus, self-built library-based classification and data retrieval, users can change the search range within the web page options to specify a given force condition or set cell line, which helps to obtain more specific results.

The self-built corpus is realized in the following ways: As shown in Figure 2, a deep neural network based on the pre-trained language model BERT is designed, and the network parameters are optimized and improved. The main parameters are as follows: batch size: 32; epochs: 4 ; learning rate: 5e-5; hidden_size: 128. After training 13.5 million words in the full English corpus PMC and 4.5 million words in the biological literature corpus PubMed, a text keyword extraction model with biological information was obtained.

2. Interactions between mechanobiological pathways

The way cells and tissues perceive and transmit mechanical information depends on the interaction between genes, and the cascade network of interactions constitutes biological signaling pathways. As shown in Fig. 2, in the route navigation section, this embodiment first shows the typical pathways regulating this process and their interactions.

As shown in Figure 3, in the pathway navigation section, the website in this example displays classical mechanosensitive pathways, such as Hippo, BMP, TGFβ, Wnt, Notch, PI3K/Akt, MAPK, Ras signaling pathway, etc. Interactions in mechanotransduction, providing users with background information in the field of mechanobiology.

In this mode, this embodiment sorts out trusted paths and their interactions, and provides general background information for users. Makes the navigation of genes in mechanosensory and mechanotransduction more convincing and meaningful by combining credible pathways associated with hippo, bmp, wnt, gPCR, tgf-beta, igf, integrins and cell junctions.

Second, the understanding of a single molecule is usually one-sided and limited. In contrast, linking molecules and pathways is more beneficial for researchers to understand and further explore their mechanisms of action. Thus, in one possible design, combining normal pathways with text mining results enables users to locate their target genes or gene sets through biological processes. Correlations between genes and mechanistically related pathways were scored by matching submitted genes to annotated gene sets for each pathway, and possible connections were provided based on text mining techniques.

In order to obtain a reasonable scoring system, this application calculates the correlation with the target retrieval factor and each classical mechanosensitive pathway based on the molecular interaction relationship pairs in the literature database, which can help researchers to quickly locate relevant biological signal transduction patterns. The score is calculated as follows:

In the above formula, r(g,p) is the correlation coefficient between the gene g and the classical mechanosensitive pathway p, Ni represents the total number of related entities in the literature database of the _ith gene in the classical mechanosensitive pathway _p , and Np is the classical mechanosensitive pathway The total number of related entities of all genes of pathway p in the literature database, Ω _g and Ω _p represent the set of gene g and classical mechanosensitive pathway p, respectively.

Using the weight algorithm can highlight the importance of pathway star molecules, which conforms to the logic of text data mining. When the target search molecule and pathway star molecules co-occur, it can be considered that the target search molecule is more likely to be associated with the pathway.

3. Visual website architecture

To support cross-platform visualization, the web architecture of the website is based on the Django framework, the back-end database is implemented using MySQL, and the semantic UI is used for the front-end architecture.

As an NLP web tool, the present invention website combines demonstration and prediction strategies to propose an efficient and plausible method to tease out connections and crosstalk between molecules that perform mechanosensory and mechanotransduction in bone.

The website of the present invention uses a graph network to display the molecules in all mechanical pathways as interconnected nodes, and the corresponding original documents can be linked to by clicking the connection between the nodes. The prior art is not repeated here.

Entities retrieved from literature databases can be sub-classified through the self-built corpus described above, allowing users to choose to focus on specific force types or specific cell lines, thereby facilitating more precise and targeted literature-based discovery. At the same time, the website of the present invention creatively adopts the pathway fitting method. Based on the weighting algorithm, the system can display the correlation score between the target search molecule and the classical mechanical pathway according to the NLP result, and link the user's target molecule with the components of the classical pathway. , making it more suitable for biomedical research.

4. Correlation identification and visualization

According to the user-defined scope, the website of the present invention can automatically retrieve and visualize the entities related to the target searched molecule and the correlations with the pathways. Interactive operations are suitable for graphic illustrations, enabling user-defined desirable layouts and details for each entity. After clicking on the edge between entities, a pop-up window can display confirmation and the resource article with the corresponding sentence highlighted in red. The collection of raw text enables users to decide the significance and reliability of the connections discovered by the AI, which may be valid and accurate. Hierarchical search enables second- and third-layer relation extraction to amplify the network, favoring the development of new molecules.

Governing pathway maps and mechanism biology of bone localization depend heavily on successive responses and interactions of several pathways, as described above. With this in mind, this example identifies a classical pathway involving mechanosensitivity and mechanotransduction with plausible evidence. Overview paths and their interaction with standing proofs are visualized via diagrams and svg.js, a lightweight library for manipulating and animating svg files. The elements of each path are searched on KEGG (Kyoto Encyclopedia of Genes and Genomes) and then compared with the datasets of this example, which contribute to the list of items contained in each path, which can be determined by the correlation coefficient. Correlations between target genes and pathways are ranked and visualized.

In addition to scoring, the Biomechanical Regulation of Bone Remodeling Text Mining Interactive website provides an interactive option to add all/selective components of the target pathway to the NLP network to form a molecule-to-pathway network to uncover more indirect connections.

The content of the result interface will be described in detail below:

Combined with Figures 4 to 7, the left side of the result interface displays the search results of mechanical pathway associations, as follows: As shown in Figure 5, panel 1 shows the correlation between the target search molecule and each classical mechanical pathway; panel 2, the user The pathway of interest can be selected, and the pathway molecules can be added to the network for combined search; in section 3, the user can quickly understand the pathway information of the target search molecule in the KEGG database, so as to have a more comprehensive understanding of the action pathway of the molecule. As shown in Figure 6, the middle part of the result page mainly visualizes the molecular interaction information, and provides a variety of "String" button options. String is a database containing protein interaction information based on research evidence and algorithmic predictions. Integrating String's results with the original NLP network can provide ideas for the study of emerging molecules. By clicking the hierarchical search function provided by the website of the present invention, the user can enlarge the relational network to the second and third layers, thereby expanding the network search range, which is beneficial to the discovery of new molecules in the pathway. The corresponding icons in the website provide functions to change the display mode, download pictures, and reset pictures. You can change the display mode of the molecular correlation diagram, download and save the current correlation diagram and restore the display of the previous version of the correlation diagram. For the retrieval of the corresponding source documents, the user can click the connection between the nodes with the mouse, as shown in Figure 7, the pop-up window will display its relevance and the corresponding literature, and the corresponding sentence is also highlighted in red.

This embodiment chooses to use co-occurrence to define correlation, instead of using machine learning method to identify grammatical data for relation extraction. Not a machine. Based on relation retrieval and visualization of each normalized sentence, entities are associated by co-occurrence, and then corresponding sentences are labeled. The co-occurrence score records the number of items corresponding to the co-occurrence tag. These sentences and corresponding entities are stored in a relational database and implemented by sqlite (https://www.sqlite.org/index.html). String (https://string-db.org/) is used for a comprehensive search of the target, and this embodiment provides secondary and/or string search options that can contribute to more results.

The method of using the website of the present invention: the user enters the target retrieval factor in the retrieval window interface of the website of the present invention, and some classical mechanics-sensitive pathways are displayed on the retrieval window interface, providing the user with background information in the field of mechanobiology, which can help the user to quickly determine Classical mechanosensitive pathways. When the input is completed, the interface will switch to the search results interface, which is divided into 7 sections for display. The left side of the interface is plate 1-3, the middle of the interface is plate 4-6, and there is a plate 7 on the right side of the interface.

Panel 1 uses a histogram to show the correlation between target search molecules and various classical mechanosensitive pathways; Panel 2 displays user-selectable pathways of interest, which can be added to the network for combined search; in Panel 3, users can quickly understand Pathway information of target search molecules in the KEGG database, in order to more comprehensively understand the action pathways of target search molecules.

By clicking the layered search function provided by panel 5, the user can enlarge the relationship network to the second and third layers to expand the network search scope. Section 6 provides functional icons for changing the display mode, downloading pictures, and resetting pictures. Clicking the icons can change the display mode of the molecular correlation diagram, download and save the current correlation diagram, and restore the display of the previous version of the correlation diagram.

For the retrieval of the corresponding source documents, the user can click the connection between the nodes in the panel 4 with the mouse, and the pop-up window in the panel 7 will display its relevance and the corresponding literature, and the corresponding sentence is also highlighted in red.

To sum up, the website of the present invention is based on the open document resources, uses the natural language processing (NLP) strategy, mines the text database, and constructs a web page interaction tool. Created the first bone-related biomechanical text database, innovatively introduced visualization mode to visualize complex and obscure text information; used self-created weighting algorithm to calculate the correlation between target retrieval factors and classical mechanics-sensitive pathways, and established a A new analysis strategy based on biological pathways; at the same time, the introduction of web interactive tools to visualize and simplify the process of biological literature exploration can greatly promote the research on the molecular mechanism of bone-related mechanobiology and promote more effective data processing. and knowledge sharing.

The present application no longer relies on unfiltered sources, but instead specifies targets in bone mechanobiology processes and retrieves taxonomic information from self-organizing libraries. In this way, users can choose to explore across all mechanistically related articles, and can even specify their target to enforce type, cell line or species, a discussion-based biomedical platform that facilitates knowledge sharing and provides researchers with unprecedented access to The range and tricky mass of information. At the same time, the present application adopts a path-centric strategy, using weighted scoring and combinatorial algorithms, to enable the navigation and exploration of individual genes or collections in mechanobiological processes.

The above specific embodiments further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above are only specific embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Within the spirit and principle of the present invention, any modifications, equivalent replacements, improvements, etc. made should be included within the protection scope of the present invention.

Claims (10)

1. A biomechanical regulation bone reconstruction text mining interactive website construction method is characterized by comprising the following steps: the method comprises the following steps:

s1, screening gene information text words in the literature according to the related entries, obtaining gene molecule interaction relation pairs, and constructing a literature database;

s2, based on the gene-molecule interaction relation pair in the literature database, calculating the correlation between the target retrieval factor and the classical mechanics sensitive path by adopting a weight algorithm;

and S3, performing visual display on the correlation between the target retrieval factor and the classical mechanical sensitive path, displaying gene molecules in the classical mechanical sensitive path as mutually connected nodes, and linking to corresponding documents in a document database by clicking a connecting line between the nodes.

2. The biomechanically regulated bone remodeling text mining interaction website building method of claim 1, wherein: and between the step S1 and the step S2, deep neural network training is carried out on the PMC database, text keywords with biological information are screened, and a corpus is constructed.

3. The method for constructing a biomechanically controlled bone remodeling text-mining interaction website of claim 2, wherein: the biological information includes mechanical type, research species, cell type.

4. The method for constructing a biomechanically controlled bone remodeling text-mining interaction website according to claim 1 or 2, wherein: the related entries in the step S1 include biomechanical and bone related entries.

5. The biomechanically regulated bone remodeling text mining interaction website building method of claim 4, wherein: said step S1 includes computer language normalization and preprocessing of the gene information text words.

6. The biomechanically regulated bone remodeling text mining interaction website building method of claim 5, wherein: the step S1 further includes recognizing the gene information text word using pubtat, and converting the gene information text word into a formal name by calling an API of the NCBI gene database.

7. The method for constructing a biomechanically controlled bone remodeling text-mining interaction website according to claim 1 or 2, wherein: the formula of the weighting algorithm in step S2 is:

wherein r (g, p) is the correlation coefficient of gene g and classical mechanics sensitive path p, N _i Represents the total number of related entities of the ith gene in the classical mechanical sensitive pathway p in the literature database, N _p The total number of related entities in the literature database, omega, for all genes of the classical mechanosensitive pathway p _g 、Ω _p Representing gene g and the set of classical mechanosensitive pathways p, respectively.

8. The biomechanically regulated bone remodeling text mining interaction website building method of claim 7, wherein: the classical mechanosensitive pathway includes at least one of Hippo, BMP, TGF β, Wnt, Notch, PI3K/Akt, MAPK, Ras.

9. The biomechanically regulated bone remodeling text mining interaction website building method of claim 1, wherein: in step S3, the method further includes visually displaying the path information of the target search factor in the KEGG database.

10. The biomechanically regulated bone remodeling text-mining interaction website building method of claim 1 or 9, wherein: in step S3, the method further includes visually displaying the gene-molecule interaction relationship pairs of the target search factors in the String data.

Images