Method, device and medium for mining copy number variation data for input text

Abstract

The invention relates to a method, equipment and medium for mining copy number variation data for an input text. The method comprises the following steps: splicing acquired context information and query information about copy number variation so as to generate an input text sequence; converting the context vocabulary sequence and the query vocabulary sequence into a word index sequence so as to generate an input vector sequence at least based on the word index sequence; inputting the input vector sequence into a pre-trained language model to generate an output vector through a stacked multi-head attention and feedforward network, the output vector comprising a vector context representation for each vocabulary; and based on the contextual representation of each vocabulary, predicting a probability that each vocabulary starts or ends as an entity to determine copy number variation entities to generate a result representation regarding copy number variation. According to the CNV data extraction method and device, CNV data extraction can be accurately and efficiently carried out on the input text with a complex CNV description form.

Description

Technical Field

[0001] The present invention generally relates to data mining of biological information, and in particular, to methods, computing devices, and computer storage media for mining copy number variation data for input text. Background Art

[0002] Copy number variation (CNV) refers to the deletion or amplification of DNA fragments that are no less than 1 kbp in length compared to the reference genome. CNV detection and normalization provide an important basis for precision medicine.

[0003] Traditional methods for mining CNV data for input text mainly rely on rule matching and / or traditional machine learning, and need to be converted for different genome versions and data formats. Among them, traditional methods include, for example: focusing mainly on the description form of point mutations, and normalizing the chromosome coordinate form of CNV based on machine learning combined with regular expressions. However, through the analysis of the titles and abstracts of more than 30 million relevant documents, it is known that about 6 million documents contain descriptions of CNV, of which no more than 100,000 documents have coordinate descriptions in the titles and abstracts, and most of the documents basically describe CNV in the form of chromosome bands (Cytoband, or "cell band"). The above-mentioned traditional methods that highly rely on regular expression rules for mutation extraction and normalization are difficult to accurately extract complex CNV description forms (including chromosome bands and coordinates), and require manual intervention, which is inefficient.

[0004] In summary, the traditional method for mining copy number variation data from input text has the disadvantage that it is difficult to accurately and efficiently extract input text with complex CNV description forms (including chromosome bands and coordinates). Summary of the invention

[0005] The present invention provides a method, a computing device and a computer storage medium for mining copy number variation data for input text, which can accurately and efficiently extract CNV data for input text with complex CNV description forms.

[0006] According to a first aspect of the present invention, a method for mining copy number variation data for input text is provided. The method comprises: splicing the acquired context information and query information about copy number variation to generate an input text sequence, wherein the input text sequence comprises at least: a context vocabulary sequence in the context information and a query vocabulary sequence in the query information; converting the context vocabulary sequence and the query vocabulary sequence into a word index sequence, so as to generate an input vector sequence based at least on the word index sequence, wherein the input vector sequence comprises a vocabulary embedding vector, a position embedding vector and a segment embedding vector; inputting the input vector sequence into a pre-trained language model to generate an output vector through a stacked multi-head attention and feedforward network, wherein the output vector comprises a vector context representation of each vocabulary; and predicting the probability of each vocabulary as the start or end of an entity based on the context representation of each vocabulary, so as to determine the copy number variation entity, so as to generate a result representation about the copy number variation.

[0007] In some embodiments, determining a copy number variation entity includes: calculating a first probability that each word is the starting position of the copy number variation entity, so as to select a word index corresponding to a maximum value of the first probability as a target starting position of the copy number variation entity; calculating a second probability that each word is the ending position of the copy number variation entity, so as to select a word index corresponding to a maximum value of the second probability as a target ending position of the copy number variation entity; and determining an entity range and entity content of the copy number variation entity based on the determined target starting position and target ending position of the copy number variation entity.

[0008] In some embodiments, generating a result representation about copy number variation includes: extracting variation type entities and inheritance mode entities in an input text based on regular matching; recording a position index of each of the variation type entities, inheritance mode entities, and copy number variation entities in the input text; calculating text distances between entities; and determining matching relationships between entities based at least on the calculated text distances between entities.

[0009] In some embodiments, determining the matching relationship between entities based at least on the calculated text distance between the entities includes: determining candidate matching relationships between the entities based on the text distance between the entities; calculating matching degree representation data for each candidate matching relationship between the entities based on the dependency path weight and the text distance weight; and selecting the candidate matching relationship corresponding to the maximum value of the matching degree representation data as the target matching relationship set between the entities.

[0010] In some embodiments, determining a matching relationship between entities includes: in response to determining that multiple variant type entities or multiple genetic pattern entities have candidate matching relationships with the same copy number variant entity, preferentially selecting a candidate matching relationship with a direct dependency modification relationship as a target matching relationship; in response to determining that there is no candidate matching relationship with a direct dependency modification relationship, calculating matching degree representation correction data based on the dependency relationship representation data and the context matching representation data, so as to select the candidate matching relationship corresponding to the highest value of the matching degree representation correction data as the target matching relationship; and in response to determining that multiple matching degree representation data are the same, selecting a candidate matching relationship with a higher text order as the target matching relationship.

[0011] In some embodiments, the result representation regarding copy number variation includes: the chromosome number, position interval, variation type, inheritance mode, and genome version number of the copy number variation entity, and generating the result representation regarding the copy number variation includes multiple items of the following: normalization for the genome version number; normalization for the variation type entity; normalization for the inheritance mode entity; and normalization for the position interval unit.

[0012] In some embodiments, normalizing the genome version number includes: mapping the copy number variation entity to the first human reference genome and the second human reference genome, respectively; obtaining information of the corresponding reference genome version extracted from the input text to match the entity coordinates of the copy number variation entity of the corresponding reference genome version; and generating a normalized result representation of the copy number variation based on the entity coordinates of the copy number variation entity, wherein the normalized result representation of the copy number variation includes: the chromosome number of the copy number variation, the position interval relative to the first human reference genome, and the position interval relative to the second human reference genome.

[0013] In some embodiments, normalization processing for variant type entities includes: generating a normalized result representation about copy number variation based on entity coordinates of copy number variation entities and a matching relationship between variant type entities and copy number variation entities; the matching relationship between variant type entities and copy number variation entities is generated via the following items: extracting variant type entities from input text using semantic analysis or a natural language processing method based on rule matching; calculating the text distance between variant type entities and copy number variation entities in the input text; determining a matching relationship between variant type entities and copy number variation entities with the shortest text distance; and in response to determining that there are multiple matching relationships between variant type entities and copy number variation entities, selecting multiple matching relationships in combination with context information.

[0014] In some embodiments, normalization processing for position interval units includes: using regular expressions to extract position interval descriptions containing units from input text; calculating the standardized position interval length for the extracted position interval descriptions; and combining the standardized position interval length, variation type, and reference genome version to generate a normalized result representation of copy number variation.

[0015] In some embodiments, normalization processing for genetic pattern entities includes: extracting genetic pattern entities from input text; classifying the extracted genetic pattern entities into predetermined genetic pattern classifications, wherein the predetermined classifications are one of new, or rare; calculating the text distance between the genetic pattern entity and the copy number variation entity so as to match the genetic pattern entity corresponding to the minimum text distance with the copy number variation entity; and combining the standardized position interval length, paternal and maternal inheritance, variation type, reference genome version, and predetermined genetic pattern classification so as to generate a normalized result representation of copy number variation.

[0016] According to a second aspect of the present invention, a computing device is also provided, comprising: a memory configured to store one or more computer programs; and a processor coupled to the memory and configured to execute the one or more programs so that the device performs the method of the first aspect of the present invention.

[0017] According to a third aspect of the present invention, a non-transitory computer-readable storage medium is further provided, wherein machine-executable instructions are stored on the non-transitory computer-readable storage medium, and when the machine-executable instructions are executed, the machine executes the method of the first aspect of the present invention.

[0018] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the invention, nor is it intended to limit the scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a system for implementing a method for mining copy number variation data for input text according to an embodiment of the present invention is shown.

[0020] Figure 2 A flowchart of a method for mining copy number variation data based on input text according to an embodiment of the present invention is shown.

[0021] Figure 3 A schematic diagram of a method for mining copy number variation data based on input text according to an embodiment of the present invention is shown.

[0022] Figure 4A flow chart of a method for determining copy number variation entities according to an embodiment of the present invention is shown.

[0023] Figure 5 A flowchart of a method for generating a result representation regarding copy number variation according to an embodiment of the present invention is shown.

[0024] Figure 6 A flow chart of a method for determining a matching relationship between entities according to an embodiment of the present invention is shown.

[0025] Figure 7 A flow chart of a method for correcting and disambiguating matching relationships between entities according to an embodiment of the present invention is shown.

[0026] Figure 8 A flow chart of a method for performing normalization processing on a variation interval unit according to an embodiment of the present invention is shown.

[0027] Fig. 9 A flowchart of a method for normalizing a genetic pattern entity according to an embodiment of the present invention is shown.

[0028] Fig.10 The block diagram of an electronic device suitable for implementing the embodiments of the present invention is schematically shown.

[0029] In the various drawings, the same or corresponding reference numerals represent the same or corresponding parts. DETAILED DESCRIPTION

[0030] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0031] As used herein, the term "including" and its variations mean open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "based at least in part on". The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc. may refer to different or the same objects.

[0032] As described above, the existing technology of traditional methods for mining copy number variation data from input text mainly focuses on the description form of point mutations, which is difficult to fully cover more complex genetic variations such as CNV; in addition, it lacks effective normalization capabilities for the description of chromosome bands, and can only process CNV descriptions in coordinate form, resulting in incomplete extracted information. Furthermore, traditional methods rely heavily on regular expression rules for mutation extraction and normalization, which makes it difficult to capture cross-sentence or paragraph-level contextual information in complex texts, and requires manual intervention, which is inefficient. Therefore, the shortcomings of traditional methods for mining copy number variation data from input text are that it is difficult to accurately and efficiently extract input text with complex CNV description forms (including chromosome bands and coordinates).

[0033] In order to at least partially solve one or more of the above problems and other potential problems, an exemplary embodiment of the present invention proposes a method for mining copy number variation data for input text. The method splices the context information and query information obtained about the copy number variation to generate an input text sequence; converts the context vocabulary sequence and the query vocabulary sequence into a word index sequence, so as to generate an input vector sequence including a vocabulary embedding vector, a position embedding vector and a segment embedding vector based on at least the word index sequence; and inputs the input vector sequence into a pre-trained language model to generate an output vector through a stacked multi-head attention and feedforward network. The present invention can make each output vector contain both the semantic information of the vocabulary itself and the global information of the context and query, and can capture the position information of the vocabulary in the sentence, enhance the model's perception of the order, and thus, even for rare or complex entities, it can effectively reduce noise interference. In addition, by predicting the probability of each vocabulary as the start or end of an entity, and then determining the entity range of the copy number variation entity to generate a result representation of the copy number variation, the present invention can accurately extract the relevant data of the copy number variation entity. Therefore, the present invention can accurately and efficiently extract CNV data for input texts with complex CNV description forms.

[0034] Figure 1 FIG. 1 is a schematic diagram of a system 100 for implementing a method for mining copy number variation data for input text according to an embodiment of the present invention. Figure 1 As shown, the system 100 includes: a computing device 110 and a server 130. In some embodiments, the computing device 110 and the server 130 interact with each other directly or via a network (not shown).

[0035] The server 130 is used, for example, to provide input text, which is, for example, a biomedical text.

[0036] Regarding the computing device 110, it is used to mine copy number variation data for input text. Specifically, the computing device 110 is used to splice the acquired context information and query information about copy number variation to generate an input text sequence, the input text sequence at least including: a context vocabulary sequence in the context information, and a query vocabulary sequence in the query information; and converting the context vocabulary sequence and the query vocabulary sequence into a word index sequence, so as to generate an input vector sequence based at least on the word index sequence, the input vector sequence including a vocabulary embedding vector, a position embedding vector and a segment embedding vector. The computing device 110 is also used to input the input vector sequence into a pre-trained language model to generate an output vector through a stacked multi-head attention and feedforward network, the output vector including a vector context representation of each vocabulary; and based on the context representation of each vocabulary, predicting the probability of each vocabulary as the start or end of an entity, so as to determine the copy number variation entity, to generate a result representation about the copy number variation.

[0037] In some embodiments, the computing device 110 may have one or more processing units, including dedicated processing units such as GPUs, FPGAs, and ASICs, and general-purpose processing units such as CPUs. In addition, one or more virtual machines may also be running on each computing device. The computing device 110 includes, for example, an input text sequence generation unit 112, an input vector sequence generation unit 114, an output vector generation unit 116, and a copy number variation result representation generation unit 118. The above-mentioned input text sequence generation unit 112, input vector sequence generation unit 114, output vector generation unit 116, and copy number variation result representation generation unit 118 may be configured on one or more computing devices 110.

[0038] The input text sequence generating unit 112 is used to concatenate the acquired context information about copy number variation and query information to generate an input text sequence, wherein the input text sequence at least includes: a context vocabulary sequence in the context information and a query vocabulary sequence in the query information.

[0039] Regarding the input vector sequence generating unit 114, it is used to convert the context vocabulary sequence and the query vocabulary sequence into a word index sequence, so as to generate an input vector sequence based on at least the word index sequence, and the input vector sequence includes a vocabulary embedding vector, a position embedding vector and a segment embedding vector.

[0040] As for the output vector generating unit 116, it is used to input the input vector sequence into the pre-trained language model so as to generate an output vector through a stacked multi-head attention and feed-forward network, wherein the output vector includes a vector context representation of each vocabulary word.

[0041] The copy number variation result representation generating unit 118 is used to predict the probability of each word being the start or end of an entity based on the context representation of each word, so as to determine the copy number variation entity, so as to generate the copy number variation result representation.

[0042] Figure 2 FIG. 2 is a flowchart of a method 200 for mining copy number variation data based on input text according to an embodiment of the present invention. It should be understood that the method 200 can be implemented in Fig.10 The electronic device 1000 described above may also be used. Figure 1 The method 200 is executed at the described computing device 110. It should be understood that the method 200 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present invention is not limited in this respect.

[0043] At step 202 , the computing device 110 concatenates the acquired context information about copy number variation and query information to generate an input text sequence, wherein the input text sequence includes at least: a context word sequence in the context information and a query word sequence in the query information.

[0044] Contextual information about copy number variation, which is, for example, context sentences in a biomedical text about copy number variation.

[0045] The query information about copy number variation is, for example, a query sentence about copy number variation. The query word sequence is, for example, a word sequence in the query sentence, which is used to guide the model to focus on a specific entity (for example, a copy number variation entity).

[0046] Regarding the input text sequence, it at least includes: a context vocabulary sequence in the context information, and a query vocabulary sequence in the query information. In some embodiments, the input text sequence includes: position identification information, a context vocabulary sequence, segmentation identification information, and a query vocabulary sequence. For example, the following expression (1) illustrates an example of an input text sequence.

[0047] {[CLS], x1, x2, ..., xN, [SEP], q1, q2, ..., qM, [SEP]} (1) In the above expression (1), [CLS] represents the position identification information of the input text sequence. [SEP] represents the segmentation identification information used to separate the context information and the query information. x1, x2, ..., xN represent the context vocabulary sequence. q1, q2, ..., qM represent the query vocabulary sequence.

[0048] At step 204, computing device 110 converts the context vocabulary sequence and the query vocabulary sequence into a word index sequence, so as to generate an input vector sequence based at least on the word index sequence, the input vector sequence including a vocabulary embedding vector, a position embedding vector, and a segment embedding vector. For example, computing device 110 maps each vocabulary in the context vocabulary sequence and the query vocabulary sequence into a discrete word index sequence through a vocabulary table.

[0049] For example, Figure 3 As shown, the computing device 110 converts the context vocabulary sequence 310 (eg, Token_X1, Token_X2, ... Token_XN) and the query vocabulary sequence 312 (eg, Token_Q1, Token_Q2, ... Token_QN)) into a word index sequence. Figure 3 As shown, the word index sequence also includes: position identification information 314 and segmentation identification information 316.

[0050] Regarding the input vector sequence, for example, Figure 3 As indicated by the mark 320 in the figure. The input vector sequence is generated based on the word index sequence, and the input vector sequence includes, for example, a vocabulary embedding vector, a position embedding vector, and a segment embedding vector. For example, the following expression (2) illustrates an example of the input vector sequence.

[0051] Embeddings=[E [CLS] ,E Tx1 ,E Tx2 …E TxN ,E [SEP] ,E Tq1 ,E Tq2 …E TqM ,E [SEP] ] (2) In the above expression (2), E Tx1 ,E Tx2 …E TxN and E Tq1 ,E Tq2 …E TqM stands for word embedding, which is used to provide semantic information of words. [SEP] E stands for segment embedding, which is used to distinguish the context part and the query part in the word embedding. [CLS] Represents position embedding, which is used to capture the position information of words in a sentence to enhance the model's perception of order. Embeddings represents the input vector sequence.

[0052] At step 206 , the computing device 110 inputs the input vector sequence into the pre-trained language model to generate an output vector through a stacked multi-head attention and feed-forward network, wherein the output vector includes a vector context representation of each vocabulary word.

[0053] By adopting the above methods, each output vector contains not only the semantic information of the word itself, but also the global information of the context and query. In addition, the multi-head attention mechanism can capture long-distance dependencies, which is conducive to improving the model's ability to understand complex sentence structures.

[0054] Regarding the pre-trained language model, it is, for example but not limited to, a bidirectional encoder representation Transformers (Bidirectional Encoder Representations from Transformers for Biomedical Text Mining, BioBERT) model for biomedical text mining. The BioBER model is mainly a pre-trained language representation model built on the BERT (Bidirectional Encoder Representations from Transformers) model. The BioBERT model uses a stack of multiple layers of Transformer encoders. Each layer contains a self-attention mechanism and a feedforward neural network, which can help the BioBERT model to capture complex dependencies in the text. It should be understood that the BioBERT model can simultaneously consider the contextual information on the left and right sides during pre-training, and therefore, it is conducive to a more comprehensive understanding of the context of each word. The present invention generates an output vector by using a pre-trained BioBERT model, which is conducive to further improving the representation ability of biomedical terms in the input text.

[0055] In some embodiments, the pre-trained language model may also be one of the ClinicalBERT model, the SciBERT model, the BlueBERT model, and the MedBERT model. Among them, the ClinicalBERT model is a model pre-trained with large-scale medical data, which can more accurately capture and understand the semantic information in medical texts. The SciBERT model is a BERT model pre-trained based on 1.14 million scientific papers (covering the fields of biomedicine and computer science), which focuses on scientific text processing. The BlueBERT model is based on the pre-trained parameters of BERT, and is further optimized and trained on biomedical corpora such as Public Medicine (PubMed) abstracts and clinical notes. The MedBERT model is a pre-trained model designed specifically for medical text processing, and has been deeply trained on medical text datasets such as PubMed and the US Internet Medical Health Information Service Platform (WebMD).

[0056] For example, the computing device 110 uses the BioBERT model to process the input vector sequence 320 encoded at step 204, and generates an output vector 322 through a stacked multi-head attention and feedforward network, and the output vector 322 includes a vector context representation of each word. It should be understood that each output vector contains both the semantic information of the vocabulary itself and the global information of the context and query. It should be understood that in a small sample scenario, the generalization ability and positioning accuracy of the model are improved through the BioBERT and query enhancement methods.

[0057] Regarding the output vector, it is illustrated as, for example, the following expression (3).

[0058] [H [CLS] ,H Tx1 ,H Tx2 …H TxN ,H [SEP] ,H Tq1 ,H Tq2 …H TqM ,H [SEP] ] (3) In the above expression (3), H Tx1 ,H Tx2 …H TxN H represents the vector context representation of the context vocabulary. Tq1 ,H Tq2 …H TqM Represents the vector context representation of the query term. [SEP] Represents the segment vector. [CLS] Represents the position vector. It should be understood that the output vector contains both the semantic information of the vocabulary itself and the global information of the context and query. In addition, the multi-head attention mechanism captures long-distance dependencies and improves the ability of the pre-trained language model to understand complex sentence structures. The query sentence can guide the model to focus on specific entity types, thereby enhancing the recognition ability of rare or ambiguous entities.

[0059] At step 208 , the computing device 110 predicts the probability of each word being the start or end of an entity based on the context representation of each word, so as to determine the copy number variation entity, so as to generate a result representation about the copy number variation.

[0060] Regarding the method for determining a copy number variation entity, it includes, for example: calculating a first probability that each word is the starting position of the copy number variation entity, so as to select the word index corresponding to the maximum value of the first probability as the target starting position of the copy number variation entity; calculating a second probability that each word is the ending position of the copy number variation entity, so as to select the word index corresponding to the maximum value of the second probability as the target ending position of the copy number variation entity; and determining the entity range and entity content of the copy number variation entity based on the determined target starting position and target ending position of the copy number variation entity. Figure 4 The method 400 for determining the entity range and entity content of the copy number variation entity is specifically described, and will not be repeated here.

[0061] It should be understood that normalizing the expression of copy number variation results is a crucial task, and its goal is to convert the diverse expression formats of copy number variation results into a consistent and standard format. This normalization process can not only improve the usability and comparability of data, but also lay a solid foundation for subsequent functional analysis and biological interpretation.

[0062] The result representation of the copy number variation entity includes, for example, the chromosome number, position interval, variation type, inheritance mode, and genome version number of the copy number variation entity. For example, the result representation of the copy number variation is: [chromosome number: position interval_variation type_inheritance mode_hg19, chromosome number: position interval_variation type_inheritance mode_hg38], where hg19 and hg38 represent different genome version numbers, respectively.

[0063] The method for generating a result representation about copy number variation includes, for example: normalization of the result representation about copy number variation. It specifically includes multiple of the following items: normalization processing for the genome version number; normalization processing for the variant type entity; normalization processing for the genetic pattern entity; and normalization processing for the position interval unit. It should be understood that the standardization and multi-dimensional (for example, including: genome version number, variant type, variant genetic pattern, and variant interval unit) normalization of the result representation about copy number variation not only improves the ability to share and reuse data, but also promotes cross-domain collaborative research, providing important support for precision medicine and population genetics research.

[0064] The following is an explanation of the method for normalizing the genome version number. It should be understood that the genome version (such as hg19 or hg38) may not be explicitly mentioned in the biomedical text, and even the description of some copy number variations is far away from the genome version description. In addition, there are significant differences in the coordinate systems of different genome versions. The coordinates of the same genomic region in the coordinate systems of different genome versions may change, which will have a direct impact on the precise positioning of the variation. Therefore, it is necessary to normalize the genome version number for copy number variation entities.

[0065] Regarding the method for normalizing the genome version number, it includes, for example: mapping the copy number variation entity to the first human reference genome and the second human reference genome respectively; obtaining the information of the corresponding reference genome version extracted from the input text to match the entity coordinates of the copy number variation entity of the corresponding reference genome version; and generating a normalized result representation of the copy number variation based on the entity coordinates of the copy number variation entity, wherein the normalized result representation of the copy number variation includes: the chromosome number of the copy number variation, the position interval relative to the first human reference genome, and the position interval relative to the second human reference genome. In some embodiments, the first human reference genome is, for example, hg19. The second human reference genome is, for example, hg38. It should be understood that for the copy number variation entity in the chromosome band description format, it describes the genomic region rather than the precise entity coordinates, and does not rely on the genome version information. Therefore, during the normalization process, the identified CNV entities need to be mapped to two commonly used genome versions, hg19 and hg38, respectively, to support CNV analysis across genome versions. With respect to the chromosome coordinate format, since the chromosome coordinates are directly dependent on the genome version information, the computing device 110 preferentially matches the chromosome coordinate format of the corresponding version based on the reference genome version mentioned in the literature (such as clearly indicated as hg19 or hg38). It should be understood that by adopting the above means, the present invention can simultaneously support hg19 and hg38 genome versions, avoiding the result representation deviation of copy number variation caused by inconsistent reference genome versions, thereby enhancing the versatility of the result representation of copy number variation after normalization, and, through the normalization processing of dual reference genome versions, the present invention can also ensure data integrity, and provide reliable positioning support even in the absence of version information.

[0066] For example, the normalized result after the genome version number normalization is expressed as: [chromosome number: position interval_hg19, chromosome number: position interval_hg38]. In some embodiments, the extracted CNV entity is, for example: 5q35.2 --> q35.3. Then the normalized result after the genome version number normalization is expressed as: ['chr5:172800000-180915260_hg19', 'chr5:173300000-181538259_hg38'].

[0067] The following describes the method for normalizing the variation types.

[0068] Regarding the method for normalizing the variant type, it includes, for example: generating a normalized result representation about the copy number variation based on the entity coordinates of the copy number variation entity and the matching relationship between the variant type entity and the copy number variation entity; the matching relationship between the variant type entity and the copy number variation entity is generated through the following items: extracting the variant type entity from the input text using a semantic analysis or a natural language processing method based on rule matching; calculating the text distance between the variant type entity and the copy number variation entity in the input text; determining the matching relationship between the variant type entity and the copy number variation entity with the closest text distance; and in response to determining that there are multiple matching relationships between the variant type entity and the copy number variation entity, selecting multiple matching relationships in combination with context information. For example, the normalized result representation in the form of the variant type normalization process is, for example, [chromosome number: position interval_variation type_hg19, chromosome number: position interval_variation type_hg38]. In some embodiments, the input text is, for example, "Adeletion was detected in chromosome region 5q35.2 --> q35.3.". The extracted CNV entity is, for example, "5q35.2 --> q35.3". The variant type is, for example, deletion. The normalized result after variant type normalization is, for example, ['chr5:172800000-180915260_del_hg19', 'chr5:173300000-181538259_del_hg38'].

[0069] It should be understood that the CNV entity and its variant type entity (e.g., "deletion" or "duplication") described in the biomedical text are often far apart, and directly matching the CNV entity with its variant type may lead to incorrect association. In addition, in some biomedical texts, the variant type entity is not explicitly stated, but is expressed through context or implicitly, which increases the difficulty of extracting the variant type entity. In large-scale biomedical text data processing, an efficient and automated method is needed to accurately associate the variant type entity with the CNV entity.

[0070] In the above means, by independently extracting variant type entities and determining the distance between variant type and CNV entity, the present invention significantly reduces the probability of incorrectly associating variant type and CNV entity due to the dispersion or implicit expression of variant type information. In addition, when variant type information is missing or there are multiple possibilities, optimization combined with contextual information can ensure the biological relevance and accuracy of the results.

[0071] Regarding the method of normalizing the variant genetic mode and the variant interval unit, the following will combine Figure 8 , Fig. 9 This has been explained and I will not go into details here.

[0072] In the above scheme, the context information and query information about the copy number variation obtained are spliced ​​to generate an input text sequence; the context vocabulary sequence and the query vocabulary sequence are converted into a word index sequence, so as to generate an input vector sequence including a vocabulary embedding vector, a position embedding vector and a segment embedding vector based on at least the word index sequence; and the input vector sequence is input into a pre-trained language model, so as to generate an output vector through a stacked multi-head attention and feedforward network. The present invention can make each output vector contain both the semantic information of the vocabulary itself and the global information of the context and query, and can capture the position information of the vocabulary in the sentence, enhance the model's perception of the order, and thus, even for rare or complex entities, it can effectively reduce noise interference. In addition, by predicting the probability of each vocabulary as the start or end of an entity, and then determining the entity range of the copy number variation entity to generate a result representation of the copy number variation, the present invention can accurately extract the relevant data of the copy number variation entity. Therefore, the present invention can accurately and efficiently extract CNV data for input texts with complex CNV description forms.

[0073] Figure 4 FIG. 4 is a flowchart of a method 400 for determining a copy number variation entity according to an embodiment of the present invention. It should be understood that the method 400 may be implemented, for example, in Fig.10 The electronic device 1000 described above may also be used. Figure 1 The method 400 is executed at the described computing device 110. It should be understood that the method 400 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present invention is not limited in this respect.

[0074] At step 402 , the computing device 110 calculates a first probability of each word being a starting position of a copy number variation entity, so as to select a word index corresponding to a maximum value of the first probability as a target starting position of the copy number variation entity.

[0075] like Figure 3 As shown, the computing device 110 predicts the first probability of each word as the starting position of the copy number variation entity based on the vector context representation of each word in the output vector 322, via the linear connection network layer 340 and the Softmax network layer 350; and selects the word index corresponding to the maximum value of the first probability as the target starting position of the copy number variation entity. For example, the output vector includes the vector context representation of N words. The corresponding predicted N words are respectively used as the N first probabilities corresponding to the starting position of the copy number variation entity (the N first probabilities are, for example, L(start)1, L(start)2…L(start) N ). For example, the value of the second first probability (ie, L(start)2) is determined to be the largest by the argmax function. Then the computing device 110 selects the vocabulary index corresponding to L(start)2 as the target starting position of the copy number variation entity. The argmax function is a function for finding a parameter (set) of a function.

[0076] The function of predicting the first probability of each word as the starting position of the copy number variation entity will be described below in conjunction with expression (4).

[0077]

[0078] In the above expression (4), Represents the contextual representation of the i-th token in the output vector. The weight of the function representing the predicted first probability. Represents the bias function for predicting the first probability function. Represents a function of the first probability of the i-th marker in the predicted output vector as the starting position of the copy number variation entity. It should be understood that the i-th marker corresponds to the i-th vocabulary.

[0079] At step 404 , the computing device 110 calculates a second probability of each word being the end position of the copy number variation entity, so as to select a word index corresponding to a maximum value of the second probability as the target end position of the copy number variation entity.

[0080] For example, the computing device 110 predicts the second probability of each word as the end position of the copy number variation entity via the linear connection network layer 340 and the softmax network layer 350; and selects the word index corresponding to the maximum value of the second probability as the target starting position of the copy number variation entity. For example, the predicted N words are respectively used as the N second probabilities corresponding to the starting positions of the copy number variation entity (the N second probabilities are, for example, L(end)1, L(end)2…L(end) N ). For example, the Nth second probability (ie, L(end)) is selected by the argmax function. N ) as the target end position of the copy number variation entity.

[0081] The function for predicting the second probability of each word being the end position of the copy number variation entity will be described below in conjunction with expression (12).

[0082]

[0083] In the above expression (12), Represents the contextual representation of the j-th token in the output vector. Represents the weight of the function that predicts the second probability. Represents the bias function for predicting the second probability function. A function representing the second probability of the jth marker in the predicted output vector being the end position of a copy number variation entity.

[0084] At step 406 , the computing device 110 determines an entity range and entity content of the copy number variation entity based on the determined target start position and target end position of the copy number variation entity.

[0085] For example, the computing device 110 can obtain the entity scope and entity content of the copy number variation entity based on the vocabulary index corresponding to the target starting position and the vocabulary index corresponding to the target ending position of the copy number variation entity. In some embodiments, the prediction of the starting position and the ending position can be jointly trained by sharing parameters to reduce overfitting.

[0086] Regarding the entity scope of the copy number variation entity, it is composed of the vocabulary index corresponding to the predicted target start position (e.g., ) and the vocabulary index corresponding to the target end position (e.g. ) is determined by Less than or equal to .

[0087] Regarding the entity content, it is the corresponding substring in the input sequence .

[0088] In some embodiments, the input sequence is, for example, "{[CLS], 'A', 'deletion', 'in', 'chromosome', '5q35.2', '-->', 'q35.3', [SEP], 'What', 'is', 'the', 'CNV','?', [SEP]}". The vocabulary index corresponding to the target starting position is, for example, 5 (marked as '5q35.2'). The vocabulary index corresponding to the target ending position is, for example, 7 (marked as 'q35.3'). The entity range is: '5q35.2 --> q35.3'.

[0089] By adopting the above means, the present invention can achieve accurate boundary identification and precise positioning of copy number variation entities.

[0090] Figure 5 FIG. 5 is a flowchart of a method 500 for generating a result representation about a copy number variation according to an embodiment of the present invention. It should be understood that the method 500 can be used, for example, in Fig.10 The electronic device 1000 described above may also be used. Figure 1 The method 500 is executed at the described computing device 110. It should be understood that the method 500 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present invention is not limited in this respect.

[0091] At step 502 , the computing device 110 extracts variant type entities and genetic pattern entities from the input text based on regular matching.

[0092] Regarding the input text, in some embodiments, it is, for example, "Maternally derived terminal deletion of chromosome 7 (q35-qter) and a de novo duplication of 11q12." Regarding variant type entities, they include, for example, "deletion" and "duplication".

[0093] Regarding inheritance mode entities, for example, they include: De Novo, Paternal Inheritance, and Maternal Inheritance.

[0094] For example, copy number variation entities are 7(q35-qter) (which indicates the portion of the long arm (q) of human chromosome 7 from the q35 region to the end (qter)), 11q12, and the like.

[0095] At step 504 , the computing device 110 records the position index of each of the variation type entity, the inheritance mode entity, and the copy number variation entity in the input text.

[0096] The position index of the entity in the input text is represented as follows in combination with expression (5).

[0097]

[0098] In the above expression (5), Represents the index of the entity in the input text. Represents the starting position index. Represents the end position index.

[0099] For example, in the above input text, the extracted variant type entities and their position indexes are “deletion (position index: [28, 36]), duplication (position index: [77, 88])”. The copy number variant entities and their position indexes are “7(q35-qter) (position index: [51, 62]), 11q12 (position index: [92, 97])”. The inheritance mode entities and their position indexes are “Maternally (position index: [0, 10]), de novo (position index: [69, 76])” At step 506 , computing device 110 calculates text distances between entities.

[0100] Regarding the text distance between entities, the algorithm is explained below in conjunction with expression (6).

[0101]

[0102] In the above expression (6), Represents the textual distance between entities x and y. Represents the starting index position of entity x. Indicates the end index position of the entity y.

[0103] At step 508 , the computing device 110 determines a matching relationship between the entities based at least on the calculated text distances between the entities.

[0104] Regarding the method for determining the matching relationship between entities, in some embodiments, it includes, for example: if the computing device 110 determines that the matching relationship between the entities is less than a predetermined text distance threshold, then determining that there is a matching relationship between the entities. For example, if the following expression (7) is satisfied, it indicates that there is a matching relationship between the corresponding entities.

[0105]

[0106] In the above expression (7), Represents the text distance between entities x and y. In some embodiments, entity x represents, for example, a copy number variation entity; entity y represents, for example, a variation type entity or a genetic mode entity. δ Represents a predetermined text distance threshold.

[0107] By adopting the above means, the present invention can determine the matching relationship between entities through the text order and index position of the entities.

[0108] Figure 6 FIG. 6 is a flowchart of a method 600 for determining a matching relationship between entities according to an embodiment of the present invention. It should be understood that the method 600 may be implemented, for example, in Fig.10 The electronic device 1000 described above may also be used. Figure 1 The method 600 is executed at the described computing device 110. It should be understood that the method 600 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present invention is not limited in this respect.

[0109] At step 602 , computing device 110 determines candidate matching relationships between entities based on text distances between the entities.

[0110] At step 604 , the computing device 110 calculates matching degree representation data for each candidate matching relationship between entities based on the dependency path weight and the text distance weight.

[0111] Regarding dependency path weights, the algorithm for calculating dependency path weights is described below in conjunction with expression (8).

[0112]

[0113] In the above expression (8), represents the dependency path weight. In the above expression (8), Represents the dependency path weight.

[0114] Regarding the text distance weight, the algorithm for calculating the dependency path weight is explained below in conjunction with expression (9).

[0115]

[0116] In the above expression (9), Represents the text distance weight. Represents the textual distance between entities x and y.

[0117] The following describes an algorithm for calculating the matching degree characterization data in conjunction with expression (10).

[0118]

[0119] In the above expression (10), Represents the matching degree characterization data. Represents the dependency path weight. Represents the text distance weight. Represents the weight hyperparameter, and the two satisfy .

[0120] For example, the input text is "Maternally derived terminal deletion of chromosome 7(q35-qter) and a de novo duplication of 11q12". The target matching relationship between the copy number variation entity and the variation type entity is: "- deletion <-> 7(q35-qter)" and "- duplication<-> 11q12". The target matching relationship between the inheritance mode entity and the copy number variation entity is: "- maternally <->7(q35-qter)" and "- de novo <-> 11q12". That is, "- maternally inherited <-> 7(q35-qter)" and "- de novo <-> 11q12".

[0121] At step 606 , the computing device 110 selects the candidate matching relationship corresponding to the maximum value of the matching degree representation data as the target matching relationship between the entities.

[0122] By adopting the above means, the present invention can utilize dependency syntax parsing to identify grammatical modifications and entity dependencies, and optimize the matching relationship between entities.

[0123] Figure 7 1 shows a flow chart of a method 700 for correcting and disambiguating matching relationships between entities according to an embodiment of the present invention, that is, a method for further determining matching relationships between entities. It should be understood that the method 700 can be, for example, Fig.10 The electronic device 1000 described above may also be used. Figure 1The method 700 is executed at the described computing device 110. It should be understood that the method 700 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present invention is not limited in this respect.

[0124] At step 702, if the computing device 110 determines that there are multiple variant type entities or multiple genetic pattern entities that have candidate matching relationships with the same copy number variant entity, and there are candidate matching relationships with direct dependency modification relationships, the candidate matching relationships with direct dependency modification relationships are preferentially selected as the target matching relationships.

[0125] At step 704, if the computing device 110 determines that there is no candidate matching relationship of the direct dependency modification relationship, the matching degree representation correction data is calculated based on the dependency representation data and the context matching representation data to select the candidate matching relationship corresponding to the highest value of the matching degree representation correction data as the target matching relationship.

[0126] The calculation method of the matching degree characterization correction data is explained below in conjunction with expression (11).

[0127]

[0128] In the above expression (11), Represents dependency representation data. Represents context matching representation data. represents the degree of match characterizing the calibration data. Represents the balance coefficient.

[0129] At step 706 , if the computing device 110 determines that there are multiple matching degree representation data that are the same, a candidate matching relationship with a higher text order is selected as a target matching relationship.

[0130] In the above scheme, the algorithm based on entity position, dependency syntactic analysis and contextual relationship can not only ensure the accurate matching relationship between entities, but also improve the robustness of the matching algorithm to adapt to the entity relationship matching of complex text descriptions.

[0131] Figure 8 FIG. 8 is a flowchart of a method 800 for normalizing a variation interval unit according to an embodiment of the present invention. It should be understood that the method 800 may be implemented, for example, in Fig.10 The electronic device 1000 described above may also be used. Figure 1 The method 800 is executed at the described computing device 110. It should be understood that the method 800 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present invention is not limited in this respect.

[0132] It should be understood that the length of the variation interval describing CNV entities in biomedical texts may use a variety of units, such as KB, MB. These units are converted inconsistently in different data sources. In addition, the inconsistency of units will lead to incomparability between data sources, which will have a negative impact on the accuracy of the analysis results. Therefore, an efficient method is needed to standardize the interval lengths of different units into a unified format.

[0133] At step 802 , computing device 110 extracts location interval descriptions containing units from input text using regular expressions.

[0134] In some embodiments, the input text is, for example, “chr19:54.6-54.7 Mb del”, and the extracted location interval is, for example, 54.6-54.7 Mb.

[0135] At step 804, the computing device 110 calculates the standardized position interval length for the extracted position interval description. For example, all units are converted to BP (base pairs), which are the most commonly used units in genomic data. For example, the following conversion rules are used: 1 KB = 1,000 BP; 1 MB = 1,000,000 BP. It should be understood that by uniformly converting to BP units, data deviations caused by unit differences are eliminated, thereby ensuring the reliability of the analysis results.

[0136] For example, the starting position is: 54.6 MB × 1,000,000 = 54,600, 000BP. The ending position is: 54.7MB × 1,000,000 = 54,700,000BP.

[0137] At step 806, the computing device 110 combines the standardized position interval length, the variation type, and the reference genome version to generate a normalized result representation of the copy number variation. For example, the normalized result of the variation interval unit normalization is identified as: [chromosome number: position interval_variation type_hg19, chromosome number: position interval_variation type_hg38]. Specifically, "['chr19:54600000-54700000_del_hg19', 'chr19:54600000-54700000_del_hg38']".

[0138] In the above scheme, efficient and automatic normalization of variation interval units is achieved through the combination of regular expressions and mathematical calculations, reducing manual intervention; and it is applicable to variation interval descriptions in a variety of units and formats, and can handle complex literature data scenarios.

[0139] Fig. 9FIG. 9 is a flowchart of a method 900 for normalizing a genetic pattern entity according to an embodiment of the present invention. It should be understood that the method 900 can be implemented, for example, in Fig.10 The electronic device 1000 described above may also be used. Figure 1 The method 900 is executed at the described computing device 110. It should be understood that the method 900 may also include additional actions not shown and / or may omit the actions shown, and the scope of the present invention is not limited in this respect.

[0140] It should be understood that CNVs are usually closely related to specific inheritance patterns (such as de novo, paternal inheritance, and maternal inheritance), and the standardization of inheritance patterns is crucial to clarify their biological significance. However, the genetic patterns in biomedical texts may be scattered in different parts of the text, and the direct association description with CNV entities is low, which increases the difficulty of extracting genetic pattern entities and matching CNV entities with genetic pattern entities. Therefore, it is necessary to normalize the genetic pattern entities.

[0141] At step 902 , computing device 110 extracts genetic pattern entities from input text.

[0142] At step 904, the computing device 110 classifies the extracted inheritance pattern entity into a predetermined inheritance pattern classification, which is one of de novo, paternal inheritance, and maternal inheritance. For example, the computing device 110 normalizes the inheritance pattern description extracted from the text into a predetermined inheritance pattern classification, which is predefined and is one of de novo (i.e., De Novo), paternal inheritance (i.e., Paternal Inheritance), and maternal inheritance (i.e., Maternal Inheritance).

[0143] At step 906 , the computing device 110 calculates the text distance between the genetic pattern entity and the copy number variation entity, so as to match the genetic pattern entity corresponding to the minimum text distance with the copy number variation entity.

[0144] For example, the computing device 110 preferentially selects the genetic pattern entity corresponding to the closest distance to associate with the CNV entity based on the distance between the genetic pattern entity and the CNV entity in the text. If there are multiple genetic pattern entities in the same text, the genetic pattern entity with a higher semantic relevance to the CNV entity is preferentially associated. For example: If a text describes both "maternal inheritance" and "newly developed" genetic pattern entities, but "newly developed" is semantically closer to the CNV entity, "newly developed" is selected as the genetic pattern entity matching result associated with the CNV entity. By adopting the above means, the present invention can combine contextual information and use semantic similarity to further optimize the matching results to ensure that the pattern and entity have biological relevance.

[0145] At step 908 , the computing device 110 combines the normalized position interval length, the variation type, the reference genome version, and the predetermined genetic pattern classification to generate a normalized result representation regarding the copy number variation.

[0146] The normalized result of copy number variation processed by genetic pattern entity normalization is expressed, for example, as: [chromosome number: position interval_variation type_genetic pattern_hg19, chromosome number: position interval_variation type_genetic pattern_hg38]. In some embodiments, the input text is, for example, "Molecular cytogenetic analysis showed apaternally derived 5q35.2 --> q35.3 direct duplication". The extracted CNV entity is, for example, "5q35.2 --> q35.3". The genetic pattern entity is, for example, "paternally". The normalized result processed by genetic pattern entity normalization is expressed, for example, as: "['chr5:172800000-180915260_dup_Paternal Inheritance_hg19', 'chr5:173300000-181538259_dup_Paternal Inheritance_hg38']".

[0147] In the above scheme, by extracting and standardizing the genetic pattern, the present invention can further enrich the annotation of CNV and improve the accuracy of its functional classification. And the standardized genetic pattern information provides reliable data support for subsequent analysis such as disease association research and genetic risk assessment. In addition, by combining the distance principle with semantic analysis, the matching efficiency is improved while ensuring the biological relevance of the results; and through consistent formats and predefined classifications, the complexity of multi-source data integration is simplified, which is suitable for the diverse needs of cross-research.

[0148] Fig.10The block diagram of the electronic device 1000 suitable for implementing the embodiment of the present invention is schematically shown. The electronic device 1000 may be used to implement the execution Figure 2 , Figures 4 to 9 The methods 200, 400 to 900 shown in FIG. Fig.10 As shown, the electronic device 1000 includes a central processing unit (i.e., CPU 1001), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (i.e., ROM 1002) or computer program instructions loaded from a storage unit 1008 to a random access memory (i.e., RAM 1003). In RAM 1003, various programs and data required for the operation of the electronic device 1000 can also be stored. CPU 1001, ROM 1002, and RAM 1003 are connected to each other via a bus 1004. An input / output interface (i.e., I / O interface 1005) is also connected to the bus 1004.

[0149] Multiple components in the electronic device 1000 are connected to the I / O interface 1005, including: an input unit 1006, an output unit 1007, and a storage unit 1008. The CPU 1001 performs the various methods and processes described above, such as performing methods 200, 400 to 900. For example, in some embodiments, methods 200, 400 to 900 may be implemented as a computer software program, which is stored in a machine-readable medium, such as a storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the CPU 1001, one or more operations of the methods 200, 400 to 900 described above may be performed. Alternatively, in other embodiments, the CPU 1001 may be configured to perform one or more actions of the methods 200, 400 to 900 by any other appropriate means (e.g., by means of firmware).

[0150] It should be further noted that the present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0151] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer readable storage medium used here is not interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.

[0152] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0153] The computer program instructions for performing the operation of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit may execute the computer-readable program instructions, thereby implementing various aspects of the present invention.

[0154] Various aspects of the present invention are described herein with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to an embodiment of the present invention. It should be understood that each box of the flow chart and / or block diagram and the combination of each box in the flow chart and / or block diagram can be implemented by computer-readable program instructions.

[0155] These computer-readable program instructions can be provided to a processor in a voice interaction device, a general-purpose computer, a special-purpose computer, or a processing unit of other programmable data processing devices, thereby producing a machine, so that when these instructions are executed by a processing unit of a computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions enable the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0156] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0157] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the equipment, method and computer program product according to multiple embodiments of the present invention. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of this module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0158] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

[0159] The above are only optional embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for mining copy number variation data for input text, characterized in that: include: splicing the obtained context information about the copy number variation and the query information to generate an input text sequence, wherein the input text sequence at least includes: a context word sequence in the context information and a query word sequence in the query information; Converting the context vocabulary sequence and the query vocabulary sequence into a word index sequence, so as to generate an input vector sequence based at least on the word index sequence, the input vector sequence comprising a vocabulary embedding vector, a position embedding vector, and a segment embedding vector; Inputting the sequence of input vectors into a pre-trained language model to generate an output vector through a stacked multi-head attention and feed-forward network, the output vector including a vector context representation of each vocabulary word; and Based on the context representation of each word, the probability of each word being the start or end of an entity is predicted so as to determine the copy number variation entity to generate a result representation about the copy number variation.

2. The method according to claim 1, characterized in that The copy number variation entities identified include: calculating a first probability of each word being a starting position of a copy number variation entity, so as to select a word index corresponding to a maximum value of the first probability as a target starting position of the copy number variation entity; calculating a second probability of each word being the end position of the copy number variation entity, so as to select a word index corresponding to a maximum value of the second probability as the target end position of the copy number variation entity; and Based on the determined target start position and target end position of the copy number variation entity, an entity range and an entity content of the copy number variation entity are determined.

3. The method according to claim 1, characterized in that Generating a representation of copy number variation results includes: Extract variant type entities and genetic pattern entities from input text based on regular matching; Record the position index of each entity in the variation type entity, the inheritance mode entity, and the copy number variation entity in the input text; Calculating textual distances between entities; and Based at least on the calculated text distances between the entities, a matching relationship between the entities is determined.

4. The method according to claim 3, characterized in that Determining the matching relationship between entities based at least on the calculated text distance between the entities includes: Based on the text distance between entities, determine the candidate matching relationships between entities; Based on the dependency path weight and the text distance weight, for each candidate matching relationship between entities, calculate matching degree representation data; and The candidate matching relationship corresponding to the maximum value of the matching degree representation data is selected as the target matching relationship between entities.

5. The method according to claim 4, characterized in that Determining the matching relationship between entities includes: In response to determining that multiple variant type entities or multiple genetic pattern entities have candidate matching relationships with the same copy number variant entity, preferentially selecting a candidate matching relationship of a direct dependency modification relationship as a target matching relationship; In response to determining that there is no candidate matching relationship with a direct dependency modification relationship, calculating matching degree representation correction data based on the dependency representation data and the context matching representation data, so as to select the candidate matching relationship corresponding to the highest value of the matching degree representation correction data as the target matching relationship; and In response to determining that the plurality of matching degree representation data are the same, a candidate matching relationship with a higher text order is selected as a target matching relationship.

6. The method according to claim 1, characterized in that The result representation about the copy number variation includes: the chromosome number, position interval, variation type, inheritance mode, and genome version number of the copy number variation entity, and generating the result representation about the copy number variation includes multiple items of the following items: Normalize the genome version number; Normalize the variant entities; Normalization is performed for inheritance mode entities; and Normalization is performed on the position interval unit.

7. The method according to claim 6, characterized in that Normalization for genome version numbers includes: mapping the copy number variation entities to a first human reference genome and a second human reference genome, respectively; Obtaining information of the corresponding reference genome version extracted from the input text so as to match entity coordinates of the copy number variation entity of the corresponding reference genome version; and Based on the entity coordinates of the copy number variation entity, a normalized result representation of the copy number variation is generated, wherein the normalized result representation of the copy number variation includes: the chromosome number of the copy number variation, the position interval relative to the first human reference genome, and the position interval relative to the second human reference genome.

8. The method according to claim 6, characterized in that Normalization processing for variant type entities includes: Generate a normalized result representation of the copy number variation based on the entity coordinates of the copy number variation entity and the matching relationship between the variation type entity and the copy number variation entity; The matching relationship between the variation type entity and the copy number variation entity is generated by: Extract variant entities from input text using semantic analysis or rule-based natural language processing methods; Calculate the text distance between variant type entities and copy number variant entities in the input text; Determine a matching relationship between a variant type entity and a copy number variant entity that has the closest text distance; and In response to determining that there are multiple matching relationships between the variation type entity and the copy number variation entity, a selection is performed for the multiple matching relationships in combination with the context information.

9. The method according to claim 6, characterized in that The normalization process for the position interval unit includes: Use regular expressions to extract position interval descriptions containing units from the input text; For the extracted location interval description, calculating the standardized location interval length; and Normalized position interval lengths, variant types, and reference genome versions are combined to generate a normalized representation of copy number variation.

10. The method according to claim 9, characterized in that Normalization of genetic pattern entities includes: Extracting genetic pattern entities from input text; Classifying the extracted genetic pattern entity into a predetermined genetic pattern classification, wherein the predetermined classification is one of de novo, paternal inheritance, and maternal inheritance; Calculating the text distance between the genetic pattern entity and the copy number variation entity, so as to match the genetic pattern entity corresponding to the minimum text distance with the copy number variation entity; and Normalized position interval lengths, variant type, reference genome version, and predetermined inheritance mode classification are combined to generate a normalized result representation for copy number variation.

11. A computing device, characterized in that: include: at least one processing unit; At least one memory, the at least one memory being coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the device to perform the steps of the method according to any one of claims 1 to 10.

12. A computer-readable storage medium, characterized in that: A computer program is stored on a computer-readable storage medium, and when the computer program is executed by a machine, the method according to any one of claims 1 to 10 is implemented.

Images