Method and device for extracting cell marker genes in literature

Abstract

The invention provides a method and a device for extracting cell marker genes in literatures, which are characterized in that statement texts are extracted from the literatures, entity statement texts are screened out and cleaned, target entity statement texts are obtained and input into a text classification model, and target statement texts conforming to cell marker gene sentence patterns are screened out; and pairing the core segmented words and the dependent segmented words or the dependent segmented words in the target statement text to obtain segmented word relation pairs, screening and filtering to obtain target segmented word relation pairs meeting conditions, and taking the target segmented word relation pairs as cell marker genes. According to the scheme, the entity statement text containing the cell name entity and the gene name entity at the same time is screened out through the entity library, the target statement text conforming to the cell marker gene sentence pattern is screened out through the text classification model, and the core segmented words and the dependent segmented words or the dependent segmented words in the target statement text are paired; therefore, the purpose of rapidly and efficiently extracting the cell marker gene from the literature is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of text processing, and in particular to a method and a device for extracting cell marker genes from documents. Background Art

[0002] Single-cell sequencing has gradually become the mainstream sequencing analysis technology with its single-cell resolution. The sliced ​​tissue for single-cell sequencing is usually composed of multiple cell types. After single-cell sequencing of the sliced ​​tissue, a gene expression map based on a single cell will be obtained. Since most cell types have their own specific high-level expressed marker genes, and the same cell types from different tissues will have different levels of marker gene expression, for the diverse tissue types currently being studied, after obtaining the gene expression map of a single cell through single-cell sequencing, it is necessary to select appropriate cell marker genes to assist in determining the type of the cell.

[0003] Existing methods for extracting cell marker genes rely on manual reading of literature and extraction and organization, which is not only time-consuming and inefficient, but also results in fewer types of extracted cell marker genes.

[0004] Therefore, the existing methods for extracting cell marker genes are time-consuming and inefficient, which is an urgent problem to be solved. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method for extracting cell marker genes from literature, so as to achieve the purpose of extracting cell marker genes from literature quickly and efficiently.

[0006] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0007] The first aspect of the embodiment of the present invention discloses a method for extracting cell marker genes from literature, comprising:

[0008] Extract sentence text from pre-downloaded documents;

[0009] Based on a pre-built entity library, entity sentence texts that contain both cell name entities and gene name entities are identified and filtered out from the sentence texts;

[0010] Clean the entity sentence text to obtain a target entity sentence text from which interference words are deleted;

[0011] Inputting the target entity sentence text into a pre-built text classification model to filter out the target sentence text that conforms to the cell marker gene sentence pattern;

[0012] Extracting core participles in the target sentence text, wherein the core participle in the target sentence text is one of a predicate, an object, a subject and a conjunction;

[0013] Obtaining dependent participles in the target sentence text that have a dependency relationship with the core participle, and pairing the dependent participles with each other and with the core participle to obtain participle relationship pairs;

[0014] Target word segmentation relationship pairs that contain both cell name entities and gene name entities are screened out, target word segmentation relationship pairs that do not meet the requirements are filtered out, and the remaining target word segmentation relationship pairs are determined to be cell marker genes.

[0015] Preferably, the cleaning of the entity sentence text to obtain the target entity sentence text with the interference words deleted includes:

[0016] For each word in the entity sentence text, if the word does not belong to an entity in the entity library, deleting affixes of words in the word and converting capital letters in the word into lowercase letters; the word is composed of one or more words;

[0017] If the word contains a stop word in a pre-built stop word list, then deleting the stop word in the word, wherein the stop words in the stop word list include at least: a, the, and or;

[0018] If the word contains a symbol in a pre-built symbol list, deleting the symbol from the word, wherein the symbols in the symbol list include at least: a comma and a period;

[0019] If the word is composed of numbers only, the word is deleted to obtain the target entity sentence text with the interference words deleted.

[0020] Preferably, the process of constructing the text classification model includes:

[0021] Obtaining a training set and a validation set, wherein the training set and the validation set contain target entity sentence texts with tags and interference words deleted, wherein the tags are used to indicate whether the target entity sentence texts conform to the cell marker gene sentence pattern;

[0022] Use the natural language text processing library to build a text classification model to be trained;

[0023] Inputting the training set into the text classification model to be trained for iterative training to obtain a trained text classification model;

[0024] The validation set is input into the trained text classification model for verification. If the precision and recall of the target entity sentence text that conforms to the cell marker gene sentence pattern meet the requirements, it is determined that the training of the text classification model is completed.

[0025] Preferably, extracting the core participles in the target sentence text includes:

[0026] Splitting the target sentence text into short sentence texts;

[0027] When the short sentence text contains a first characteristic word, the short sentence text is marked as a short sentence text from which cell marker genes need to be manually extracted and output; the first characteristic word indicates that the short sentence text contains a plurality of cell name entities and a plurality of gene name entities, and there is no corresponding relationship between the cell name entities and the gene name entities;

[0028] When the short sentence text contains a second characteristic word, the short sentence text is removed; the second characteristic word indicates that although the short sentence text contains both the cell name and the gene name, it indicates that the gene is lowly expressed or lacks expression in the cell;

[0029] Extract the core words from the remaining short sentence text.

[0030] Preferably, the step of obtaining dependent participles in the target sentence text that have a dependency relationship with the core participle, and pairing the dependent participles with each other and with the core participle to obtain participle relationship pairs includes:

[0031] Constructing a dependency syntax tree using words in the target sentence text as nodes and dependency relationships or modification relationships between words in the target sentence text as connecting lines;

[0032] From the dependency syntax tree, searching for each sub-node connected to the node corresponding to the core participle, and extracting each dependent participle corresponding to each sub-node;

[0033] Sorting the dependent participles according to the order in which they appear in the target sentence text to obtain the order of the dependent participles;

[0034] Generate each dependent participle list in sequence according to the order of each dependent participle; wherein each dependent participle list stores the corresponding dependent participle;

[0035] For each dependent participle list, if the dependent participle and the core participle meet the prescribed part-of-speech relationship, the dependent participle and the core participle are merged, and the dependent participle list corresponding to the dependent participle is deleted; wherein the prescribed part-of-speech relationship is one of a noun modifier, a numerical modifier, and a compound word relationship;

[0036] For each of the remaining dependent participle lists, if the child node corresponding to the dependent participle in the dependent participle list still has a secondary child node connected to it in the dependency syntax tree, and the word corresponding to the secondary child node meets the prescribed part-of-speech relationship, then merge the word corresponding to the secondary child node with the dependent participle, and delete the secondary child node;

[0037] Taking the words corresponding to the remaining secondary child nodes in the dependency syntax tree as dependent participles, and adding them to the dependent participle list;

[0038] Return to executing the step of merging the dependent participle with the core participle for each dependent participle list, and deleting the dependent participle list corresponding to the dependent participle, if the dependent participle and the core participle meet the prescribed part-of-speech relationship, until all merged dependent participle lists are obtained;

[0039] Add the merged core participle to the first item in the merged dependent participle list;

[0040] Randomly select two dependent word lists from any three adjacent merged dependent word lists, extract a word from each of the two dependent word lists for pairing, and obtain a word relationship pair.

[0041] Preferably, the step of screening out target word segmentation relationship pairs that contain both cell name entities and gene name entities, filtering out target word segmentation relationship pairs that do not meet the requirements, and determining that the remaining target word segmentation relationship pairs are cell marker genes comprises:

[0042] Filter out target word segmentation relationship pairs that contain both cell name entities and gene name entities;

[0043] Searching in the dependency syntax tree to obtain the ancestor node set of the two word corresponding nodes in the target word relationship pair, if there is a difference between the ancestor node sets of the nodes corresponding to the two words, deleting the target word relationship pair;

[0044] Searching in the dependency syntax tree to obtain the levels of the two word corresponding nodes in the target word segmentation relationship pair, and if the level difference between the two word corresponding nodes is greater than 3, deleting the target word segmentation relationship pair;

[0045] If the two participle corresponding nodes in the target participle relationship pair both contain subtrees in the dependency syntax tree, then deleting the target participle relationship pair;

[0046] The remaining target-term relationship pairs are determined as cell marker genes.

[0047] Preferably, it also includes:

[0048] Extract the key texts from the literature, including the subject headings, title, abstract, the first paragraph of the method and results;

[0049] Input the key text into a pre-built corpus to extract species information, tissue information and disease information; the corpus includes: a CRAFT corpus for extracting species information, a BioNLP13CG corpus for extracting tissue information and a custom-built corpus, and a BC5CDR corpus for extracting disease information, wherein the custom-built corpus is constructed based on a common tissue list extracted from a CellOntology database and a cancer tissue list obtained from NCI (National Cancer Institute);

[0050] Obtaining a document ID of the document, and associating the document ID with the species information, the tissue information, and the disease information;

[0051] The associated species information, tissue information and disease information are output. A second aspect of an embodiment of the present invention discloses a device for extracting cell marker genes from a document, comprising:

[0052] The first extraction unit is used to extract sentence texts from pre-downloaded documents; based on a pre-built entity library, identify and filter entity sentence texts containing both cell name entities and gene name entities from the sentence texts;

[0053] A cleaning unit, used for cleaning the entity sentence text to obtain a target entity sentence text with interference words removed;

[0054] A first screening unit is used to input the target entity sentence text into a pre-built text classification model to screen out the target sentence text that conforms to the cell marker gene sentence pattern;

[0055] A second extraction unit extracts core participles in the target sentence text, wherein the core participle in the target sentence text is one of a predicate, an object, a subject and a conjunction;

[0056] A pairing unit is used to obtain dependent participles in the target sentence text that have a dependency relationship with the core participle, and pair the dependent participles with each other and with the core participle to obtain participle relationship pairs;

[0057] The second screening unit is used to screen out target word segmentation relationship pairs that contain both cell name entities and gene name entities, filter out target word segmentation relationship pairs that do not meet the requirements, and determine that the remaining target word segmentation relationship pairs are cell marker genes.

[0058] Preferably, the cleaning unit is specifically used for:

[0059] For each word in the entity sentence text, if the word does not belong to an entity in the entity library, the affixes of the words in the word are deleted, and the uppercase letters in the word are converted to lowercase letters; the word consists of one or more words; if the word contains a stop word in a pre-constructed stop word list, the stop word in the word is deleted, and the stop words in the stop word list include at least: a, the and or; if the word contains a symbol in a pre-constructed symbol list, the symbol is deleted from the word, and the symbols in the symbol list include at least: comma and period; if the word is composed of only numbers, the word is deleted to obtain a target entity sentence text with interference words deleted.

[0060] Preferably, it also includes:

[0061] A construction unit is used to obtain a training set and a verification set, wherein the training set and the verification set contain target entity sentence texts with tags and interference words deleted, wherein the tags are used to characterize whether the target entity sentence texts conform to the cell marker gene sentence pattern; a text classification model to be trained is constructed using a natural language text processing library; the training set is input into the text classification model to be trained for iterative training to obtain a trained text classification model; the verification set is input into the trained text classification model for verification, and if the precision and recall of the target entity sentence texts conforming to the cell marker gene sentence pattern meet the requirements, it is determined that the training of the text classification model is completed.

[0062] Based on the method and device for extracting cell marker genes from documents provided by the above-mentioned embodiment of the present invention, sentence texts are extracted from pre-downloaded documents; based on a pre-built entity library, entity sentence texts containing both cell name entities and gene name entities are identified and screened out from the sentence texts; the entity sentence texts are cleaned to obtain target entity sentence texts with interference words deleted; the target entity sentence texts are input into a pre-built text classification model to screen out target sentence texts that conform to the sentence pattern of cell marker genes; core participles in the target sentence texts are extracted, and the components of the core participles in the target sentence texts are one of predicates, objects, subjects and conjunctions; dependent participles in the target sentence texts that have a dependency relationship with the core participles are obtained, and the dependent participles and the dependent participles and the core participles are paired to obtain participle relationship pairs; target participle relationship pairs containing both cell name entities and gene name entities are screened out, target participle relationship pairs that do not meet the requirements are filtered out, and the remaining target participle relationship pairs are determined to be cell marker genes. In this scheme, the entity library is used to screen out entity sentence texts that contain both cell name entities and gene name entities, and the text classification model is used to screen out target sentence texts that conform to the cell marker gene sentence pattern. Finally, the core word segmentations are extracted and paired to obtain the cell marker genes, so as to achieve the purpose of quickly and efficiently extracting cell marker genes from the literature. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0064] Figure 1 A flow chart of a method for extracting cell marker genes from literature disclosed in an embodiment of the present invention;

[0065] Figure 2 A flow chart of extracting species information from documents disclosed in an embodiment of the present invention;

[0066] Figure 3 A flowchart of extracting organizational information from documents disclosed in an embodiment of the present invention;

[0067] Figure 4 This is a structural diagram of a device for extracting cell marker genes from literature disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0070] As can be seen from the background art, the existing methods for manually extracting cell marker genes have the problems of being time-consuming and inefficient.

[0071] Therefore, an embodiment of the present invention discloses a method and device for extracting cell marker genes from literature. In this scheme, an entity library is used to screen out entity sentence texts that contain both cell name entities and gene name entities, and a text classification model is used to screen out target sentence texts that conform to the cell marker gene sentence pattern. The core participles and dependent participles in the target sentence text, or the dependent participles, are paired to obtain cell marker genes, so as to achieve the purpose of quickly and efficiently extracting cell marker genes from literature.

[0072] like Figure 1 FIG. 1 is a flow chart of a method for extracting cell marker genes from literature disclosed in an embodiment of the present invention. The method mainly comprises the following steps:

[0073] Step S101: extracting sentence text from pre-downloaded documents.

[0074] The specific implementation process of step S101 is as follows:

[0075] In the PubMed website, we used the keyword "Animals" [MeSH Terms] AND ("Single-Cell Analysis" [MeSH Terms] OR ("single-cell" AND "expression")) to search for literature containing the keyword and obtain the corresponding PMID (PubMed Unique Identifier). PMID is the document ID in the fields of life sciences and medicine included in the PubMed search engine.

[0076] PMID was converted to PMC ID using R package rentrez.

[0077] It should be noted that the difference between PMID and PMC ID lies in the different numbering systems. A document with a PMID number must have a PMC ID number, but a document with a PMCID number does not necessarily have a PMID number.

[0078] Use the R package tidypmc to download the xml text file of the corresponding document according to the PMC ID. After importing the xml text, split it into 6 parts: title, abstract, background introduction, results, methods and discussion, and save it in the PMC.tsv file.

[0079] If the xml file of the document corresponding to the PMC ID is not available (not open source or not in the database), use the software scihub with PMID as the keyword to download the document and store it in PDF format. Then use the python library scipdf to convert PDF into text format, and according to the processing results of scipdf, divide the document text into 5 parts: title, abstract, background introduction, results and methods, and save them in PMID.tsv file.

[0080] Input the sentence text in the PMID.tsv file or the PMC.tsv file into the semantic library model, use the semantic library model to split out the complete sentences ending with a period, and save the complete sentences in the segment.txt file in sequence.

[0081] The semantic library model may be an en_core_sci_scibert model.

[0082] Finally, we use the built-in abbreviation detection method in the Python library scispacy, abbreviation_detector(), to identify each sentence in the segment.txt file, so as to obtain all the abbreviations and their full names contained in each sentence, and save them in the abbreviations_dict.pickle file.

[0083] Step S102: Based on the pre-built entity library, entity sentence texts containing both cell name entities and gene name entities are identified and screened out from the sentence texts.

[0084] In step S102, the entity library is a gene & cell entity library, and the specific construction process is as follows:

[0085] Obtain human and mouse gene annotation files from 10X single-cell sequencing analysis software: Cellranger, extract the human and mouse gene names, and save them in the gene_collect.csv file.

[0086] The cell names involved in each tissue were downloaded and sorted from the CellOntogy database, and the plural cell names were converted into singular cell names using the Python library inflect. All characters were converted to lowercase and saved in the cell_collect.csv file.

[0087] Use the Python library scispacy to create a custom entity library: nlp_model_jnlpba_extra_gene_cell, import the cell names and gene names saved in the gene_collect.csv file and cell_collect.csv file respectively, and obtain the gene & cell entity library.

[0088] In the specific implementation of step S102 , if the sentence text in the segment.txt file contains abbreviations, the abbreviations in the sentence are completed based on the abbreviations_dict.pickle file storing the abbreviations and the corresponding full names.

[0089] Using the Gene & Cell Entity Library and a corpus built based on biomedical data, entity sentences that contain both cell name entities and gene name entities are identified and screened out.

[0090] Among them, the corpora built based on biomedical data include: en_ner_jnlpba_md corpus, nlp_model_craft corpus and nlp_model_bionlp13cg corpus.

[0091] Specifically, the corpus built based on biomedical data stores a large number of cell names and gene names, and the possible cell names and gene names can be inferred based on the positional relationship between words.

[0092] Input the entity sentence into the en_core_sci_scibert semantic library model. If there are cell name entities and gene name entities to be completed in the entity sentence, complete them and save the completed entity sentence to the potentially_valualbe_evidence.txt file.

[0093] It should be noted that the en_core_sci_scibert semantic library model itself has a large training set and an advanced model. It can recognize the full names of most entities involved in the sentence text, and the recognized entity names are relatively complete.

[0094] Step S103: Clean the entity sentence text to obtain the target entity sentence text with interference words deleted.

[0095] In the specific implementation process of step S103,

[0096] Use the Python libraries spacy and scispacy to import the semantic library en_core_sci_scibert, read each sentence in the potentially_valuablebe_evidence.txt file and perform word segmentation to obtain each token, where a token consists of one or more words.

[0097] For each word in the entity sentence text, if the word does not belong to an entity in the entity library, the word is restored to its part of speech, for example, the affixes of the word are deleted and the uppercase letters in the word are converted to lowercase letters.

[0098] In one embodiment, if a word contains a stop word in a pre-built stop word list, the stop word in the word is deleted.

[0099] It should be noted that the stop words in the stop word list include at least: a, the and or, and their components in entity sentences are often articles, prepositions, adverbs or conjunctions. The stop word list can be imported into the stop word list (STOP_WORDS) provided by the spacy library using Python.

[0100] If the word contains a symbol from a pre-built list of symbols, remove the symbol from the word.

[0101] It should be noted that the symbol list can be imported into Python using the symbol list provided by the string library (string.punctuation).

[0102] In one embodiment, symbols from the symbol list except for the four symbols "+", "-", "(" and ")" in the word are deleted.

[0103] If a word consists only of numbers, delete the word.

[0104] In one embodiment, for each pair of parentheses in an entity statement, if the inside is a figure guide text commonly seen in the literature, such as Fig. 1 or Fig. S1, the pair of parentheses and the internal information are deleted; if the inside contains at least one gene name entity from an entity library, the pair of parentheses is retained and the content in the parentheses is deleted.

[0105] Through the above cleaning process, the target entity sentence text with interference words deleted is finally obtained and stored in the potentially_valualbe_evidence_clean.txt file.

[0106] Step S104: input the target entity sentence text into a pre-built text classification model to filter out the target sentence text that conforms to the cell marker gene sentence pattern.

[0107] In step S104, the process of constructing the text classification model is as follows:

[0108] A training set and a validation set are obtained, wherein the training set and the validation set contain target entity sentence texts with tags and interference words removed, and the tags are used to indicate whether the target entity sentence texts conform to the cell marker gene sentence pattern.

[0109] Specifically, at least 655 single-cell documents are downloaded in advance and about 27,000 target entity sentences are obtained based on the processing of the above steps, and the target entity sentence texts are manually marked in advance. Then, the above-mentioned cleaning method is used to obtain the target entity sentences with interference words deleted, and the target entity sentences are divided into a training set and a validation set in proportion, for example, the training set accounts for 80% of the target entity sentences, and the validation set accounts for 20% of the target entity sentences.

[0110] Use the natural language text processing library to build a text classification model to be trained.

[0111] Specifically, the Python library spacy (natural language text processing library) is used to build a text classification model. The text classification model built using spacy combines the common bag-of-words model and the neural network. Then, based on the configuration file provided by spacy, the text classification model is modified to binary classification.

[0112] The training set is input into the text classification model to be trained for iterative training to obtain the trained text classification model.

[0113] The validation set is input into the trained text classification model for verification. If the precision and recall of the target entity sentence text that conforms to the cell marker gene sentence pattern meet the requirements, the training of the text classification model is determined to be completed.

[0114] Specifically, the entity sentence text from the potentially_valualbe_evidence_clean.txt file is input into a pre-built text classification model, and the text classification model outputs the probability that each entity sentence text is a cell marker gene sentence, and then the entity sentence and the corresponding probability are stored in the potentially_alualbe_evidence_AddPredict.txt file.

[0115] Remove entity sentence texts with probabilities less than 0.7 from the potentially_alualbe_evidence_AddPredict.txt file, and use the remaining entity sentence texts with probabilities equal to or greater than 0.7 as target entity sentence texts.

[0116] Step S105: extracting core words in the target sentence text.

[0117] In step S105, the component of the core participle in the target sentence text is one of a predicate, an object, a subject and a conjunction.

[0118] In the specific implementation process of step S105, the target sentence text is split into short sentence texts according to the user-defined rules.

[0119] In one embodiment, when the first characteristic word exists in the short sentence text, it is marked as a short sentence text that needs to be manually extracted with cell marker genes, output and saved, and the cell marker genes are subsequently manually extracted.

[0120] Among them, the first characteristic word represents that there are multiple cell name entities and multiple cell names in the short sentence text, and there is no corresponding relationship between the cell name entity and the gene name entity. The first characteristic word is such as "neither", "not", "respectively" and so on.

[0121] It should be noted that the appearance of the first characteristic word is usually accompanied by multiple cell name entities and multiple gene name entities in a sentence. There is no one-to-one correspondence between the cell name entities and the gene name entities, and a diagram is needed to make a judgment. Therefore, once such words are detected, the entire sentence is saved in the summary_puzzle.csv file, which needs to be manually extracted later.

[0122] When the second characteristic word exists in the short sentence text, the short sentence text is eliminated.

[0123] Among them, the second characteristic word represents that although the cell name and gene name exist in the short sentence text at the same time, it means that the gene is lowly expressed or lacks expression in the cell. The second characteristic word includes "decreased", "downregulated", "absented" and so on.

[0124] It should be noted that the appearance of the second characteristic word usually indicates that although the sentence contains both cell names and gene names, they are not cell marker genes (the gene expression is reduced or lacks, not highly expressed). To avoid erroneous extraction, the short sentence is discarded.

[0125] Extract the core words from the remaining short sentences without extracting cell marker genes.

[0126] Specifically, the Python library spacy is used to import the semantic library en_core_sci_scibert, segment the short sentences, and obtain the core segmentations in the short sentence text.

[0127] In one embodiment, if there are parentheses in the short sentence, the word before the parentheses is completed by the Python custom function fill_token_name(), and then forms a group of word relationship pairs with the word in the parentheses, which are added to the word relationship pairs obtained in step S106. Specifically, the Python library spacy is used to import the semantic library en_core_sci_scibert. The word in the short sentence is tagged with part of speech. For the word before the parentheses, other word that meets the prescribed part of speech relationship is found, and the other word is added to the word before the word as a completion, wherein the prescribed part of speech relationship is one of a nominal modification, a numerical modifier, and a compound word relationship.

[0128] Step S106: Obtain dependent participles in the target sentence text that have a dependency relationship with the core participle, and pair the dependent participles with each other and with the core participle to obtain participle relationship pairs.

[0129] The specific implementation process of step S106 includes the following steps:

[0130] Step S11: construct a dependency syntax tree using the words in the target sentence text as nodes and the dependencies or modification relationships between the words in the target sentence text as connecting lines.

[0131] In step S11, there is a direct or indirect modification relationship between words in the target sentence text, and the dependency syntax tree focuses on the grammatical connection between words in the sentence and constrains it into a tree structure.

[0132] In the specific implementation process of step S11, for each short sentence, the Python library spacy is used to import the semantic library en_core_sci_scibert. When the short sentence is segmented, part-of-speech tagging is performed at the same time. The part-of-speech tagging distinguishes whether each word in the sentence is an adjective, adverb or noun, etc., and obtains the dependency or modification relationship between words.

[0133] Step S12: from the dependent syntactic tree, search for each child node connected to the node corresponding to the core word, and extract each dependent word corresponding to each child node.

[0134] Step S13: sorting the dependent participles according to the order in which they appear in the target sentence text to obtain the sorting of the dependent participles.

[0135] For example, the order of the dependent participles connected to the core participle 1 is a, b, c, and d.

[0136] Step S14: Generate each dependent word list in sequence according to the order of each dependent word.

[0137] In step S14, each dependent word list stores the corresponding dependent word.

[0138] For example, according to the order of the dependent participles being a, b, c and d, dependent participle lists A, B, C and D are generated, and dependent participle lists A, B, C and D respectively store the dependent participles a, b, c and d.

[0139] Step S15: For each dependent participle list, if the dependent participle and the core participle meet the prescribed part-of-speech relationship, the dependent participle and the core participle are merged, and the dependent participle list corresponding to the dependent participle is deleted.

[0140] In step S15, the prescribed part-of-speech relationship is one of a noun modifier (nmod), a numeral modifier (nummod), and a compound word relationship (compound).

[0141] For example, for dependent participle a in dependent participle list A, if the part-of-speech relationship between it and core participle 1 meets the specified requirements, then the dependent participle a is merged with the core participle 1 to obtain the merged core participle a1, and the dependent participle list A is deleted, and the remaining dependent participle lists B, C and D continue to execute the following steps.

[0142] Step S16: For each remaining dependent participle list, if the child node corresponding to the dependent participle in the dependent participle list also has a secondary child node connected to it in the dependency syntax tree, and the word corresponding to the secondary child node meets the specified part-of-speech relationship, then the word corresponding to the secondary child node is merged with the dependent participle, and the secondary child node is deleted.

[0143] For example, for the remaining dependent participle lists B, C, and D, assuming that the child node corresponding to the dependent participle b in the dependent participle list B has a secondary child node connected to it in the dependency syntax tree, and the word x corresponding to the secondary child node meets the specified part-of-speech relationship, then the word x is merged with the dependent participle b to obtain the dependent participle xb, and saved in the dependent participle list B. The above example is also used for the dependent participle lists C and D.

[0144] Step S17: taking the words corresponding to the remaining secondary child nodes in the dependency syntax tree as dependent participles and adding them to the dependent participle list.

[0145] For example, after dependent word list B is processed by step S16, there are two secondary child nodes remaining in the dependency syntax tree, corresponding to words y and z respectively. Word y and word z are added to dependent word list B as dependent word list, and dependent word list C and D are also executed according to the above example.

[0146] Step S18: Return to step S15 and execute until all merged dependent word lists are obtained.

[0147] In step S18, the process returns to step S15 and performs three cycles to obtain the dependent word segmentation lists after each merging.

[0148] Step S19: Add the merged core participle to the first item in the merged dependent participle list.

[0149] For example, the merged core participle a1 is added as the first item to the merged dependent participle lists B, C, and D.

[0150] Step S110: randomly selecting two dependent segmentation lists from any three adjacent merged dependent segmentation lists, extracting a segmentation from each of the two dependent segmentation lists for pairing, and obtaining a segmentation relationship pair.

[0151] It can be understood that the dependent word segmentation list is generated according to the order in which the dependent word segmentation lists appear in the target entity sentence, and each dependent word segmentation list has an arrangement order. Therefore, in step S110, based on the arrangement order of the merged dependent word segmentation lists, two dependent word segmentation lists are randomly selected from any three adjacent merged dependent word segmentation lists.

[0152] For example, core word lists B, C, and D are three adjacent dependent word lists that have been merged. Core word lists B and D are randomly selected, and one word is randomly selected from each of the core word lists B and D for pairing.

[0153] It should be noted that there can only be one core participle in a participle relationship pair, that is, there are only pairings between dependent participles, and between core participles and dependent participles.

[0154] In one embodiment, after obtaining each word segmentation relationship pair, a filtering process is further included, specifically including:

[0155] Step S107: Screen out target word segmentation relationship pairs that contain both cell name entities and gene name entities, filter out target word segmentation relationship pairs that do not meet the requirements, and determine that the remaining target word segmentation relationship pairs are cell marker genes.

[0156] In step S107, the cell marker genes are specifically cell-marker genes, which are described as N marker genes of a certain cell type.

[0157] In the specific implementation process of step S107, the entity statement text of the potentially_valualbe_evidence_clean.txt file extracts cell name entities and gene name entities, and generates a cell name entity list and a gene name entity list. Based on the cell name entity list and the gene name entity list, each word segmentation relationship pair is judged in turn to see whether it contains both the cell name entity and the gene name entity from the cell name entity list and the gene name entity list. If so, it is determined to be the target word segmentation relationship pair.

[0158] After obtaining each target word segmentation relationship pair, it is necessary to filter out the target word segmentation relationship pairs that do not meet the requirements, as follows:

[0159] First, for each word in a word pair, extract its ancestor node. If there is a difference between the ancestor node sets of the two words in the pair, filter out the word pair.

[0160] Second, for each word in a word pair, calculate its node level in the tree structure. If the level difference between the two words is greater than 3, filter out the word pair.

[0161] Third, for each word in the word pair, determine whether it contains a subtree consisting of child nodes based on the corresponding dependency syntax tree. If both words in the word pair contain subtrees, filter out the word pair.

[0162] In the third point, the subtree is composed of the descendant nodes of the nodes corresponding to the participles in the dependency syntax tree, wherein the words corresponding to the descendant nodes must meet relevant conditions, that is, the components of the words corresponding to the descendant nodes in the original sentence are noun subjects, passive noun subjects or noun modifiers, etc.

[0163] Based on the method for extracting cell marker genes from documents disclosed in the above-mentioned embodiment of the present invention, sentence text is extracted from pre-downloaded documents, entity sentence texts containing both cell name entities and gene name entities are identified and screened out from the sentence texts based on a pre-built entity library, the entity sentence texts are cleaned to obtain target entity sentence texts with interference words deleted, the target entity sentence texts are input into a text classification model, target sentence texts that conform to the sentence pattern of cell marker genes are screened out, the core participles and dependent participles in the target sentence texts, or the dependent participles, are paired, and the target participle relationship pair is determined to be a cell marker gene. In this scheme, the entity library is used to screen out entity sentence texts containing both cell name entities and gene name entities, the text classification model is used to screen out target sentence texts that conform to the sentence pattern of cell marker genes, and finally the core participles are extracted and paired to obtain cell marker genes, so as to achieve the purpose of extracting cell marker genes from documents quickly and efficiently.

[0164] Based on the method disclosed in the above embodiment of the present invention for extracting cell marker genes from literature, Figure 2 As shown, it is a flow chart of extracting species information from documents disclosed in an embodiment of the present invention.

[0165] It should be noted that the extraction of species information is divided into two parts. The first part is the situation where there are characteristic sentences or keywords in the literature. The specific extraction process is as follows:

[0166] 1. Extract characteristic sentences or keywords from the literature.

[0167] Specifically, the R package easyPubMed was used to download the corresponding literature with PMID as the keyword, and characteristic sentences or subject words were extracted.

[0168] 2. Input the keywords into the CRAFT corpus used to extract species information and extract information of multiple species.

[0169] 3. If the number of species information is 1, the species name is output directly based on the species information.

[0170] If the number of species information is greater than 1, each species information constitutes a species information set, and the species name corresponding to each species information cannot be distinguished, then the existing species information is used to compare with the species information set.

[0171] If there is an intersection, and the species information in the intersection is consistent with the existing species information, then the species name corresponding to the species information in the intersection is output, that is, the species name corresponding to the existing species information.

[0172] For example, Figure 2In the example of the embodiment of the present invention, the existing species information-1, species information-2 and species information-3 are respectively compared with the species information set.

[0173] If the number of species information is 0, the second part of the species information extraction process is used for extraction.

[0174] The second part is the case where there is no subject word in the literature, or the number of species information extracted by the subject word in the first part is 0. The specific extraction process is as follows:

[0175] 1. Extract the title, abstract, background introduction, methods and the first paragraph of the results section of the literature.

[0176] Specifically, use Python to read the segment.txt file corresponding to the document ID and split it into the title, abstract, method, and the first paragraph of the results.

[0177] 2. Input the title, abstract, methods and the first paragraph of the results into the CRAFT corpus for extracting species information, and extract species information-1 corresponding to the title, species information-2 corresponding to the method, species information-3 corresponding to the first paragraph of the results, and species information-4 corresponding to the collection of the first paragraph of the abstract, methods and results.

[0178] 3. Determine whether species information-1 is a common species.

[0179] If so, output the species name corresponding to species information -1.

[0180] If not, determine whether species information-2 is consistent with species information-1.

[0181] If they are consistent, output the species names corresponding to species information-1 and species information-2.

[0182] If they are inconsistent, select the species with the highest frequency in species information-1, species information-2 and species information-3.

[0183] When the frequency of the species with the highest frequency in species information-1, species information-2 and species information-3 is greater than 1, the species name of the species is directly output.

[0184] When the frequency of the species with the highest frequency in species information-1, species information-2 and species information-3 is not greater than 1, the species frequency is compared with the frequency of each species in species information-4, and the species name corresponding to the species with the highest frequency is output.

[0185] Based on the method disclosed in the above embodiment of the present invention for extracting cell marker genes from literature, Figure 3As shown, it is a flowchart of extracting organizational information from documents disclosed in an embodiment of the present invention. The specific extraction process is as follows:

[0186] 1. Extract the subject words, title, abstract, method, first paragraph of the results section and specific sentences from the literature.

[0187] The specific sentence is a sentence containing single-cell sequencing keywords, including "singlecell", "dissociation", "isolation", "singlenucleus", etc.

[0188] For specific document downloading methods and keyword extraction methods, please refer to the above embodiments of the present invention.

[0189] 2. Input the subject words, title, abstract, methods, the first paragraph of the results section and specific sentences into the BioNLP13CG corpus used to extract tissue information, and extract the tissue types corresponding to the above parts. In addition, the intersection of the cell entities identified in the methods and specific sentences is also extracted. The cell entity extraction method is consistent with the method used to identify the cell name entity part mentioned above.

[0190] 3. The organization type corresponding to the text of the subject words, titles and specific sentences is taken as the candidate organization 1.

[0191] From candidate organizations 1, select the organization types that exist in the subject words, titles, and specific sentence texts, and output the organization names corresponding to the organization types.

[0192] 4. If the above-mentioned process 3 does not exist, combine the tissue type in candidate tissue 1 with the tissue type corresponding to the first paragraph of the results section and the tissue type corresponding to the methods text to obtain candidate tissue 2.

[0193] 5. Count the frequencies of the following organizational types:

[0194] A. Count the rankings of the frequencies of the tissue types in candidate tissue 2 in the subject terms, title, abstract, methods, the first paragraph of the results section, and specific sentences.

[0195] B. Count the rank of the frequency of tissue types in candidate tissue 2 coexisting with single-cell sequencing keywords in the same sentence.

[0196] C. Count the tissue types in candidate tissue 2 and the rank of the frequency of the cell types in the intersection of the method and the cell types corresponding to the specific statement existing in the same statement.

[0197] 6. Filter out the top two organization types in terms of cumulative ranking among the statistical results of A, B, and C, and output the corresponding organization names.

[0198] In one embodiment, for the extraction of disease information, the title of the document is input into the BC5CDR corpus used to extract disease information. If disease information is detected in the title, it is output, otherwise it is judged as a normal sample.

[0199] In one embodiment, the extracted species information, tissue information, and disease information are associated with the corresponding document ID, and the associated species information, tissue information, and disease information are output.

[0200] Based on the process of extracting species, tissue and disease information from documents disclosed in the above embodiment of the present invention, various parts of the document are extracted and input into the corresponding corpus to extract species, tissue and disease information, and species names and tissue names are extracted according to corresponding rules. In this solution, the purpose of efficiently extracting species, tissue and disease information from documents is achieved by using the corpus and extraction rules.

[0201] Based on the method for extracting cell marker genes from literature disclosed in the above embodiment of the present invention, correspondingly, Figure 4 As shown, an apparatus for extracting cell marker genes from literature disclosed in an embodiment of the present invention includes: a first extraction unit 401, a cleaning unit 402, a first screening unit 403, a second extraction unit 404, a pairing unit 405 and a second screening unit 406.

[0202] The first extraction unit 401 is used to extract sentence texts from pre-downloaded documents, and based on a pre-built entity library, identify and filter entity sentence texts containing both cell name entities and gene name entities from the sentence texts.

[0203] The cleaning unit 402 is used to clean the entity sentence text to obtain the target entity sentence text with the interference words deleted.

[0204] In one embodiment, the cleaning unit 402 is specifically used for:

[0205] For each word in the entity sentence text, if the word does not belong to an entity in the entity library, the affixes of the words in the word are deleted, and the uppercase letters in the word are converted to lowercase letters. The word consists of one or more words. If the word contains stop words in a pre-built stop word list, the stop words in the word are deleted. The stop words in the stop word list include at least: a, the and or. If the word contains symbols in a pre-built symbol list, the symbols are deleted from the word. If the word consists only of numbers, the word is deleted to obtain the target entity sentence text with interference words deleted.

[0206] The first screening unit 403 is used to input the target entity sentence text into a pre-built text classification model to screen out the target sentence text that conforms to the cell marker gene sentence pattern.

[0207] The second extraction unit 404 extracts core words in the target sentence text.

[0208] In one embodiment, the second extraction unit 404 includes:

[0209] The split subunit is used to split the target sentence text into short sentence texts.

[0210] The extraction subunit is used to mark and output the short sentence text as a short sentence text that requires manual extraction of cell marker genes when the first feature word exists in the short sentence text. The first feature word represents that there are multiple cell name entities and multiple gene name entities in the short sentence text, and there is no corresponding relationship between the cell name entity and the gene name entity. When the second feature word exists in the short sentence text, the short sentence text is eliminated. The second feature word represents that although the short sentence text contains both cell names and gene names, it indicates that the gene is lowly expressed or lacks expression in the cell. The core participles in the remaining short sentence texts where the cell marker genes have not been extracted are extracted.

[0211] Specifically, the component of the core participle in the target sentence text is one of the predicate, object, subject and conjunction.

[0212] The pairing unit 405 is used to pair the words related to the core participle in the target sentence text with the core participle to obtain a participle relationship pair.

[0213] In one embodiment, the pairing unit 405 includes:

[0214] The construction subunit is used to construct a dependency syntactic tree using the words in the target sentence text as nodes and the dependency or modification relationship between the words in the target sentence text as connecting lines.

[0215] The search subunit is used to search for each sub-node connected to the node corresponding to the core word in the dependency syntax tree, and extract each dependent word corresponding to each sub-node.

[0216] A generating subunit is used to sort the dependent participles according to the order in which they appear in the target sentence text to obtain the sorting of the dependent participles, and to generate the dependent participle lists in sequence according to the sorting of the dependent participles; wherein each dependent participle list stores the corresponding dependent participle.

[0217] The merging subunit is used for merging the dependent participle with the core participle for each dependent participle list, and deleting the dependent participle list corresponding to the dependent participle; wherein the specified part-of-speech relationship is one of a nominal modifier, a numerical modifier, and a compound word relationship; for each remaining dependent participle list, if the child node corresponding to the dependent participle in the dependent participle list also has a secondary child node connected to it in the dependency syntax tree, and the word corresponding to the secondary child node meets the specified part-of-speech relationship, then the word corresponding to the secondary child node is merged with the dependent participle, and the secondary child node is deleted.

[0218] The return sub-unit is used to take the words corresponding to the remaining secondary child nodes in the dependency syntax tree as dependent participles and add them to the dependent participle list; the merging sub-unit is used to execute for each dependent participle list, if the dependent participle and the core participle meet the prescribed part-of-speech relationship, then the dependent participle is merged with the core participle, and the dependent participle list corresponding to the dependent participle is deleted, until each merged dependent participle list is obtained; the merged core participle is added to the first item in the merged dependent participle list.

[0219] The pairing subunit is used to randomly select two dependent segmentation lists from any three adjacent merged dependent segmentation lists, extract a segmentation from each of the two dependent segmentation lists for pairing, and obtain a segmentation relationship pair.

[0220] The second screening unit 406 is used to screen out target word segmentation relationship pairs that contain both cell name entities and gene name entities, filter out target word segmentation relationship pairs that do not meet the requirements, and determine that the remaining target word segmentation relationship pairs are cell marker genes.

[0221] In one embodiment, the second screening unit 406 is specifically used for:

[0222] For each word in the entity sentence text, if the word does not belong to an entity in the entity library, delete the affixes of the words in the word, and convert the uppercase letters in the word to lowercase letters; the word consists of one or more words; if the word contains a stop word in the pre-built stop word list, delete the stop word in the word; if the word contains a symbol in the pre-built symbol list, delete the symbol from the word; if the word is composed of only numbers, delete the word, and obtain the target entity sentence text with the interference words deleted. Among them, the stop words in the stop word list include at least: a, the and or, and the symbols in the symbol list include at least: comma and period.

[0223] In one embodiment, the device further comprises:

[0224] A construction unit is used to obtain a training set and a validation set, wherein the training set and the validation set contain target entity sentence texts with tags and interference words deleted, wherein the tags are used to characterize whether the target entity sentence texts conform to the cell marker gene sentence pattern, a text classification model to be trained is constructed using a natural language text processing library, the training set is input into the text classification model to be trained for iterative training to obtain a trained text classification model, the validation set is input into the trained text classification model for verification, and if the precision and recall of the target entity sentence texts conforming to the cell marker gene sentence pattern meet the requirements, it is determined that the training of the text classification model is completed.

[0225] In one embodiment, the device further comprises:

[0226] The third extraction unit is used to extract key texts from the document, input the key texts into a pre-built corpus, extract species information, tissue information and disease information, the corpus includes: CRAFT corpus for extracting species information, BioNLP13CG corpus and custom-built corpus for extracting tissue information, BC5CDR corpus for extracting disease information, obtain the document ID of the document, associate the document ID with species information, tissue information and disease information, and output the associated species information, tissue information and disease information.

[0227] Among them, the key text includes: subject headings, title, abstract, methods and the first paragraph of results.

[0228] The custom-built corpus was constructed based on the common tissue list extracted from the Cell Ontology database and the cancer tissue list extracted from the NCI (National Cancer Institute).

[0229] Based on the above-mentioned device for extracting cell marker genes from documents disclosed in the embodiment of the present invention, sentence text is extracted from pre-downloaded documents, entity sentence texts containing both cell name entities and gene name entities are identified and screened out from the sentence texts based on the pre-built entity library, the entity sentence texts are cleaned to obtain target entity sentence texts with interference words deleted, the target entity sentence texts are input into the text classification model, the target sentence texts that conform to the sentence pattern of cell marker genes are screened out, the core participles in the target sentence texts are extracted, the core participles and dependent participles in the target sentence texts are paired, or the dependent participles are paired, to obtain a participle relationship pair, the target participle relationship pair containing both cell name entities and gene name entities is screened out, and the target participle relationship pair is determined to be a cell marker gene. In this scheme, the entity library is used to screen out entity sentence texts containing both cell name entities and gene name entities, the text classification model is used to screen out the target sentence text that conforms to the sentence pattern of cell marker genes, the core participles and dependent participles in the target sentence texts are paired, or the dependent participles are paired, to obtain cell marker genes, so as to achieve the purpose of extracting cell marker genes from documents quickly and efficiently.

[0230] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0231] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0232] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting cell marker genes from literature, characterized in that: include: Extract sentence text from pre-downloaded documents; Based on a pre-built entity library, entity sentence texts that contain both cell name entities and gene name entities are identified and filtered out from the sentence texts; Clean the entity sentence text to obtain a target entity sentence text from which interference words are deleted; Inputting the target entity sentence text into a pre-built text classification model to screen out the target sentence text that conforms to the cell marker gene sentence pattern; Extracting core participles in the target sentence text, wherein the core participle in the target sentence text is one of a predicate, an object, a subject and a conjunction; Obtaining dependent participles in the target sentence text that have a dependency relationship with the core participle, and pairing the dependent participles with each other and with the core participle to obtain participle relationship pairs; Target word segmentation relationship pairs that contain both cell name entities and gene name entities are screened out, target word segmentation relationship pairs that do not meet the requirements are filtered out, and the remaining target word segmentation relationship pairs are determined to be cell marker genes.

2. The method according to claim 1, characterized in that The step of cleaning the entity sentence text to obtain a target entity sentence text with interference words deleted includes: For each word in the entity sentence text, if the word does not belong to an entity in the entity library, deleting affixes of words in the word and converting capital letters in the word into lowercase letters; the word is composed of one or more words; If the word contains a stop word in a pre-built stop word list, then deleting the stop word in the word, wherein the stop words in the stop word list include at least: a, the, and or; If the word contains a symbol in a pre-built symbol list, deleting the symbol from the word, the symbols in the symbol list at least include: a comma and a period; If the word is composed of numbers only, the word is deleted to obtain the target entity sentence text with the interference words deleted.

3. The method according to claim 1, characterized in that: The construction process of the text classification model includes: Obtaining a training set and a validation set, wherein the training set and the validation set contain target entity sentence texts with tags and interference words deleted, wherein the tags are used to indicate whether the target entity sentence texts conform to the cell marker gene sentence pattern; Use the natural language text processing library to build a text classification model to be trained; Inputting the training set into the text classification model to be trained for iterative training to obtain a trained text classification model; The validation set is input into the trained text classification model for verification. If the precision and recall of the target entity sentence text that conforms to the cell marker gene sentence pattern meet the requirements, it is determined that the training of the text classification model is completed.

4. The method according to claim 1, characterized in that The extracting core participles in the target sentence text includes: Splitting the target sentence text into short sentence texts; When the short sentence text contains a first characteristic word, the short sentence text is marked as a short sentence text from which cell marker genes need to be manually extracted and output; the first characteristic word indicates that the short sentence text contains a plurality of cell name entities and a plurality of gene name entities, and there is no corresponding relationship between the cell name entities and the gene name entities; When the short sentence text contains a second characteristic word, the short sentence text is removed; the second characteristic word indicates that although the short sentence text contains both the cell name and the gene name, it indicates that the gene is lowly expressed or lacks expression in the cell; Extract the core words from the remaining short sentence text.

5. The method according to claim 1, characterized in that The step of obtaining dependent participles in the target sentence text that have a dependency relationship with the core participle, and pairing the dependent participles with each other and with the core participle to obtain participle relationship pairs includes: Constructing a dependency syntax tree using words in the target sentence text as nodes and dependency relationships or modification relationships between words in the target sentence text as connecting lines; From the dependency syntax tree, searching for each sub-node connected to the node corresponding to the core participle, and extracting each dependent participle corresponding to each sub-node; Sorting the dependent participles according to the order in which they appear in the target sentence text to obtain the order of the dependent participles; Generate each dependent participle list in sequence according to the order of each dependent participle; wherein each dependent participle list stores the corresponding dependent participle; For each dependent participle list, if the dependent participle and the core participle meet the prescribed part-of-speech relationship, the dependent participle and the core participle are merged, and the dependent participle list corresponding to the dependent participle is deleted; wherein the prescribed part-of-speech relationship is one of a noun modifier, a numerical modifier, and a compound word relationship; For each of the remaining dependent participle lists, if the child node corresponding to the dependent participle in the dependent participle list still has a secondary child node connected to it in the dependency syntax tree, and the word corresponding to the secondary child node meets the prescribed part-of-speech relationship, then merge the word corresponding to the secondary child node with the dependent participle, and delete the secondary child node; Taking the words corresponding to the remaining secondary child nodes in the dependency syntax tree as dependent participles, and adding them to the dependent participle list; Return to executing the step of merging the dependent participle with the core participle for each dependent participle list, and deleting the dependent participle list corresponding to the dependent participle, if the dependent participle and the core participle meet the prescribed part-of-speech relationship, until all merged dependent participle lists are obtained; Add the merged core participle to the first item in the merged dependent participle list; Randomly select two dependent word lists from any three adjacent merged dependent word lists, extract a word from each of the two dependent word lists for pairing, and obtain a word relationship pair.

6. The method according to claim 5, characterized in that The method of screening out target word segmentation relationship pairs containing both cell name entities and gene name entities, filtering out target word segmentation relationship pairs that do not meet the requirements, and determining that the remaining target word segmentation relationship pairs are cell marker genes, comprises: Filter out target word segmentation relationship pairs that contain both cell name entities and gene name entities; Searching in the dependency syntax tree to obtain the ancestor node set of the two word corresponding nodes in the target word relationship pair, if there is a difference between the ancestor node sets of the nodes corresponding to the two words, deleting the target word relationship pair; Searching in the dependency syntax tree to obtain the levels of the two word corresponding nodes in the target word segmentation relationship pair, and if the level difference between the two word corresponding nodes is greater than 3, deleting the target word segmentation relationship pair; If the two participle corresponding nodes in the target participle relationship pair both contain subtrees in the dependency syntax tree, then deleting the target participle relationship pair; The remaining target-term relationship pairs are determined as cell marker genes.

7. The method according to any one of claims 1 to 6, characterized in that: Also includes: Extract the key texts from the literature, including the subject headings, title, abstract, the first paragraph of the method and results; Input the key text into a pre-built corpus to extract species information, tissue information and disease information; the corpus includes: a CRAFT corpus for extracting species information, a BioNLP13CG corpus for extracting tissue information and a custom-built corpus, and a BC5CDR corpus for extracting disease information, wherein the custom-built corpus is constructed based on a common tissue list extracted from a Cell Ontology database and a cancer tissue list obtained from NCI (National Cancer Institute); Obtaining a document ID of the document, and associating the document ID with the species information, the tissue information, and the disease information; The associated species information, tissue information and disease information are output.

8. A device for extracting cell marker genes from literature, characterized in that: include: A first extraction unit, used for extracting sentence text from pre-downloaded documents; Based on a pre-built entity library, entity sentence texts that contain both cell name entities and gene name entities are identified and filtered out from the sentence texts; A cleaning unit, used for cleaning the entity sentence text to obtain a target entity sentence text with interference words removed; A first screening unit is used to input the target entity sentence text into a pre-built text classification model to screen out the target sentence text that conforms to the cell marker gene sentence pattern; A second extraction unit extracts core participles in the target sentence text, wherein the core participle in the target sentence text is one of a predicate, an object, a subject and a conjunction; A pairing unit is used to obtain dependent participles in the target sentence text that have a dependency relationship with the core participle, and pair the dependent participles with each other and with the core participle to obtain participle relationship pairs; The second screening unit is used to screen out target word segmentation relationship pairs that contain both cell name entities and gene name entities, filter out target word segmentation relationship pairs that do not meet the requirements, and determine that the remaining target word segmentation relationship pairs are cell marker genes.

9. The device according to claim 8, characterized in that The cleaning unit is specifically used for: For each word in the entity sentence text, if the word does not belong to an entity in the entity library, the affixes of the words in the word are deleted, and the uppercase letters in the word are converted to lowercase letters; the word consists of one or more words; if the word contains a stop word in a pre-constructed stop word list, the stop word in the word is deleted, and the stop words in the stop word list include at least: a, the and or; if the word contains a symbol in a pre-constructed symbol list, the symbol is deleted from the word, and the symbols in the symbol list include at least: comma and period; if the word is composed of only numbers, the word is deleted to obtain a target entity sentence text with interference words deleted.

10. The device according to claim 8, characterized in that Also includes: A construction unit is used to obtain a training set and a validation set, wherein the training set and the validation set contain target entity sentence texts with tags and interference words deleted, wherein the tags are used to indicate whether the target entity sentence texts conform to the cell marker gene sentence pattern; Use the natural language text processing library to build a text classification model to be trained; The training set is input into the text classification model to be trained for iterative training to obtain a trained text classification model; the verification set is input into the trained text classification model for verification, and if the precision and recall of the target entity sentence text that conforms to the cell marker gene sentence pattern meet the requirements, it is determined that the training of the text classification model is completed.

Images