A structural variation filtering method, device and equipment based on multimodal fusion

Abstract

The present application relates to a method, device and equipment for filtering structural variations based on multimodal fusion. The method comprises: encoding the variable information of the structural variation site and generating a characteristic image of the structural variation site. After calculating the ESF value of the variable information, a label is set for the image-text pair consisting of the characteristic image of the structural variation site and the ESF value according to the structural variation type to obtain a labeled image-text pair. The CLIP multimodal model is trained according to preset configuration parameters to obtain a trained multimodal fusion model. The pre-processed structural variation data and the labeled image-text pairs are filtered for variation sites by the trained multimodal fusion model to obtain filtered result data. The use of this method can improve the accuracy of structural variation detection in gene sequences.

Description

Technical Field

[0001] The present invention belongs to the technical field of gene sequence structural variation detection, and in particular relates to a structural variation filtering method, device and equipment based on multimodal fusion. Background Art

[0002] Structural variation generally refers to large-scale sequence variations within the genome, typically involving DNA segments greater than 50 bases in length. These variations include deletions (DELs), insertions (INS), inversions (INVs), duplications (DUPs), and translocations (TRANs). Compared to shorter insertions and deletions (INDELs) and single nucleotide variants (SNVs), structural variations have a higher probability of causing cancer, genetic diseases, neurodevelopmental disorders, and other conditions. For example, a translocation between chromosomes 9 and 22 forms the so-called "Philadelphia chromosome," which causes chronic myeloid leukemia. Deletions on chromosome 7 can cause Verdi syndrome. Trisomy of chromosome 21 can lead to Down syndrome.

[0003] Existing structural variation detection tools can be divided into statistical methods and deep learning methods based on their implementation principles. The former infers structural variation by determining which features in the data are abnormal, while the latter relies on the feature extraction and pattern recognition capabilities of deep neural networks to discover structural variation.

[0004] Early structural variation detection tools, including DELLY, LUMPY, Manta, and SvABA, were primarily based on short-read sequences. These tools typically identify and detect structural variations based on statistical information from alignment results, such as read depth (RD), discordant read pairs (RP), split read alignment (SR), local assembly, or a combination thereof. DELLY utilizes a combination of discordant read pairs and split read alignment signals to detect structural variations in short-read sequences. LUMPY comprehensively detects structural variations in short-read sequences by combining multiple signals, including discordant read pairs, split read alignment, and read depth. Manta uses discordant read pairs and split read alignment to perform local assembly of suspected structural variation regions to verify structural variation events in short-read sequences. SvABA utilizes discordant read pairs and split read alignment signals, combined with local assembly methods, to enhance the detection of complex structural variations in short-read sequences. However, the short read lengths of short-read sequences limit the sensitivity of these tools, resulting in a high number of false positives.

[0005] The rapid development of third-generation sequencing, also known as long-read sequencing technology, such as the Pacific Biosciences (PacBio) and Oxford Nanopore Technology (ONT) platforms, can provide alignment information over a longer span, providing opportunities for more comprehensive and accurate detection of structural variations. However, the higher error rate and longer read length of long-read sequencing technology also make short-read-based structural variation detection algorithms unsuitable, prompting researchers to develop new structural variation detection tools.

[0006] Many existing long-read structural variation detection tools, such as PBSV, Sniffles2, SVIM, and cuteSV, are based on statistical methods. These structural variation detection tools follow the same principle as short-read structural variation detection tools, which rely on statistical information from alignment results. By integrating multiple statistical information and leveraging the advantages of long-read sequencing, they achieve high-precision structural variation detection. PBSV, designed specifically for PacBio long-read data, uses statistical signals in alignment results (such as CIGAR strings and paired-end information) to detect structural variations. Sniffles2 generates candidate regions of potential structural variation events based on discordant read pairs and split read alignment signals in long-read sequences. Further analysis and statistical evaluation verify the authenticity of candidate structural variation events. SVIM analyzes discordant read pairs and split read alignments in long-read sequences to identify potential structural variation breakpoints. It uses probabilistic models to assess the likelihood of different structural variation types and distinguish true structural variation from noise signals. cuteSV is compatible with both short-read and long-read sequencing data, and comprehensively analyzes signals such as inconsistent read pairs, split read alignment, and read depth in the alignment results to identify different types of structural variations.

[0007] With the rapid development of deep learning technology, a number of deep learning-based structural variation detection tools have emerged, including DeepSVFilter, SVision, SVcnn, and cnnLSV. The core of these methods is to convert various signals in the alignment results into two-dimensional images. These methods then leverage the powerful feature extraction and pattern recognition capabilities of deep neural networks to achieve high-precision structural variation detection. DeepSVFilter encodes read depth, discordant read pairs, and split read alignments in the alignment data into RGB images and uses a CNN network for classification. This allows filtering of DEL and INV types in short-read structural variation results. SVision generates VAR-REF and REF-REF images based on the "variant feature sequence" (VAR) and reference sequence (REF) in long-read sequences. It uses CNNs to learn and extract variant features from these images, enabling detection of complex structural variants. SVcnn designs a candidate variant region algorithm that generates feature images based on the CIGAR string and split read alignment information in the candidate regions. A CNN network is then used to classify the structural variation types in long-read sequences. cnnLSV integrates the detection results of multiple long-read detection tools, proposes an image encoding strategy based on CIGAR string and split read segment alignment information, and uses a CNN network to remove false positives in the results to improve detection performance.

[0008] Statistical methods rely on predefined features, which may not fully capture complex patterns and potential nonlinear relationships in the data, limiting detection capabilities. Furthermore, these methods are often based on fixed detection patterns and have poor generalization capabilities. Deep learning-based methods are often trained on general-purpose CNN models, such as MobileNet and ResNet, which incompletely capture feature signal information. Furthermore, the model's recognition results may be affected by the dataset or comparison tools, potentially leading to performance degradation for cross-platform data models. Summary of the Invention

[0009] Based on this, it is necessary to provide a structural variation filtering method, device and equipment based on multimodal fusion that can improve the accuracy of gene sequence structural variation detection in response to the above technical problems.

[0010] A structural variation filtering method based on multimodal fusion, the method comprising:

[0011] The variable information of the structural variation site is encoded to generate a characteristic image of the structural variation site.

[0012] After calculating the ESF value of the variable information, the image-text pair consisting of the characteristic image of the structural variation site and the ESF value is labeled according to the structural variation type to obtain a labeled image-text pair.

[0013] The CLIP multimodal model is trained according to preset configuration parameters to obtain a trained multimodal fusion model. The preprocessed structural variation data and labeled image text pairs are filtered for variation sites through the trained multimodal fusion model to obtain filtered result data.

[0014] A structural variation filtering device based on multimodal fusion, comprising:

[0015] The feature image generation module is used to encode the variable information of the structural variation site and generate a feature image of the structural variation site.

[0016] The multimodal data annotation module is used to calculate the ESF value of the variable information, and then set labels for the image-text pairs consisting of the characteristic image of the structural variation site and the ESF value according to the structural variation type to obtain labeled image-text pairs.

[0017] The filtering module is used to train the CLIP multimodal model according to preset configuration parameters to obtain a trained multimodal fusion model. The trained multimodal fusion model is used to filter the variation sites of the preprocessed structural variation data and the labeled image text pairs to obtain the filtered result data.

[0018] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0019] The variable information of the structural variation site is encoded to generate a characteristic image of the structural variation site.

[0020] After calculating the ESF value of the variable information, the image-text pair consisting of the characteristic image of the structural variation site and the ESF value is labeled according to the structural variation type to obtain a labeled image-text pair.

[0021] The CLIP multimodal model is trained according to preset configuration parameters to obtain a trained multimodal fusion model. The preprocessed structural variation data and labeled image text pairs are filtered for variation sites through the trained multimodal fusion model to obtain filtered result data.

[0022] The aforementioned multimodal fusion-based structural variation filtering method, device, and apparatus encode the variable information at the structural variation site to generate a feature image, which is then converted into a two-dimensional feature image. This method breaks through the constraints of traditional fixed feature patterns and presents complex relationships and underlying patterns in the data in a visual and structured manner. Compared to relying solely on predefined features, it can more comprehensively capture data features. Simultaneously, the ESF value of the variable information is calculated to form one-dimensional data, which is then combined with the two-dimensional feature image to form an image-text pair. This multimodal data combination characterizes the structural variation site from different dimensions, addressing the shortcomings of general deep learning-based CNN models in capturing feature signals. Furthermore, the image-text pairs are labeled according to the structural variation type. Combined with the preprocessed structural variation data, a multimodal fusion model is constructed using the trained CLIP multimodal model. This trained multimodal fusion model is then used for variant site filtering. This significantly enhances the model's adaptability and robustness to diverse data, enabling high-precision structural variation identification in gene sequences, reducing false positives, improving detection accuracy, and enabling integration with existing detection tools. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 1 is a flow chart of a structural variation filtering method based on multimodal fusion in one embodiment;

[0024] Figure 2 Schematic diagram of a process for identifying and filtering structural variations in gene sequences based on multimodal technology and integrating structural variation feature image information and ESF values in one embodiment;

[0025] Figure 3 1 is a structural block diagram of a structural variation filtering device based on multimodal fusion in one embodiment;

[0026] Figure 4 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0028] In one embodiment, Figure 1 As shown, a structural variation filtering method based on multimodal fusion is provided, comprising the following steps:

[0029] Step 102: Encode the variable information of the structural variation site to generate a characteristic image of the structural variation site.

[0030] Step 104 , after calculating the ESF value of the variable information, a label is set for the image-text pair consisting of the structural variation site feature image and the ESF value according to the structural variation type to obtain a labeled image-text pair.

[0031] Step 106 : The CLIP multimodal model is trained according to preset configuration parameters to obtain a trained multimodal fusion model. The preprocessed structural variation data and the labeled image-text pairs are filtered for variation sites using the trained multimodal fusion model to obtain filtered result data.

[0032] The aforementioned multimodal fusion-based structural variation filtering method encodes variable information at structural variation sites to generate feature images, which are then converted into two-dimensional feature images. This method breaks the constraints of traditional fixed feature patterns and presents complex relationships and underlying patterns in the data in a visual and structured manner. Compared to relying solely on predefined features, it can more comprehensively capture data characteristics. Simultaneously, the ESF value of the variable information is calculated to form one-dimensional data, which is then combined with the two-dimensional feature images to form image-text pairs. This multimodal data combination characterizes structural variation sites from different dimensions, addressing the shortcomings of general deep learning-based CNN models in capturing feature signals. Furthermore, image-text pairs are labeled according to structural variation type. Combined with preprocessed structural variation data, a multimodal fusion model is constructed using the trained CLIP multimodal model. This trained multimodal fusion model is then used for variant site filtering. This significantly enhances the model's adaptability and robustness to diverse data, enabling high-precision structural variation identification in gene sequences, reducing false positives, improving detection accuracy, and enabling integration with existing detection tools.

[0033] In one embodiment, depth information for structural variation sites in an alignment file is extracted, and the coverage depth at each chromosome location is recorded based on the depth information. The read sequences in the alignment file are then preprocessed to obtain preprocessing information. The coverage depth and preprocessing information are used as variable information for each structural variation site, and a structural variation site feature image is generated based on the variable information. The height of each pixel in the structural variation site feature image is equal to the value of the variable in the variable information.

[0034] In one embodiment, the ESF value of preprocessed information in the variable information is calculated, and the total number of structural variation site feature images and ESF values is checked to see if they are consistent with the total number of structural variation sites. If so, labels are assigned to the image-text pairs consisting of the structural variation site feature images according to the structural variation type to obtain labeled image-text pairs. Otherwise, the variable information is re-acquired to generate the structural variation site feature images.

[0035] In one embodiment, the CLIP multimodal model is trained according to preset configuration parameters, standard image data, and standard text data to obtain a trained multimodal fusion model.

[0036] In one embodiment, the preprocessed structural variation data is obtained by encoding the variable information of the structural variation site of the structural variation data, generating a feature image to be processed, and calculating the ESF value.

[0037] In one embodiment, a trained multimodal fusion model is used to classify the feature images and ESF values corresponding to the pre-processed structural variation data and the labeled image-text pairs at each variation site to be processed to obtain a model classification result, and the false positive results of the structural variation data are filtered according to the model classification result to obtain filtered result data.

[0038] In one embodiment, entries consistent with the model classification results are retained in the structural variation data, and entries inconsistent with the model classification results are deleted or modified to obtain filtered result data.

[0039] In one embodiment, Figure 2 As shown, a method based on multimodal technology and integrating structural variation feature image information and ESF value is provided to realize structural variation identification and filtering of gene sequences. Based on multimodal technology, structural variation feature image information and ESF (Embedding Sequence Feature) text information are integrated to realize structural variation identification and filtering of gene sequences.

[0040] Specifically, feature images are generated based on the RD, RP, and SR statistical information in the alignment (BAM / SAM, Binary / Sequence Alignment / Map format) files, as well as the M (Match), I (Insert), D (Delete), and S (Soft clipping) string information of the CIGAR signal in the alignment file. The ESF features of the CIGAR signal (length, minimum value, first quartile, median, third quartile, maximum value, root mean square value, harmonic mean, mean, standard deviation, and coefficient of variation of each CIGAR signal) are then integrated. The multimodal model CLIP is then used for training to identify structural variations in gene sequences and filter out false positive results in the structural variation result files (VCF format, Variant Call Format). The details are as follows:

[0041] Step 1: Obtain the original data file and preprocess the data file;

[0042] Step 1.1: VCF file preprocessing: structural variation data preprocessing;

[0043] Step 1.2: Alignment file preprocessing: sequence alignment file preprocessing;

[0044] Step 2: Preprocess the data file based on the structural variation sites;

[0045] Step 2.1: Obtaining mutation site information: Traverse the structural variation result file to obtain chromosome mutation site information;

[0046] Step 2.2: Alignment information preprocessing: preprocess the alignment information based on the variant sites;

[0047] Step 2.2.1: Apply for variable spaces all_img, all_img_mids, and all_list to store statistical image feature information, CIGAR image feature information, and ESF variable information;

[0048] Step 2.2.2: Allocate memory space split_read_left, split_read_right, and rd_count for three statistics: left split read, right split read, and read depth;

[0049] Step 2.2.3: Allocate memory space conjugate_m, conjugate_i, conjugate_d, and conjugate_s for the four CIGAR signals Match, Insert, Delete, and Soft Clipping.

[0050] Step 2.2.4: Load the chromosome depth information file, record the read sequence coverage depth of each structural variation position, and store the result in the variable rd_count;

[0051] Step 2.3: read screening;

[0052] Step 2.3.1: Traverse the read sequences in the alignment result file, determine and count the reads on the left (soft clipped) and right (soft clipped), and store the results in the variables split_read_left and split_read_right;

[0053] Step 2.3.2: Traverse the CIGAR signals in the read segment alignment results, count the number of Match, Insert, Delete, and SoftClipping signals, and store them in the variables conjugate_m, conjugate_i, conjugate_d, and conjugate_s;

[0054] Step 3: Encode the characteristic image of the structural variation site based on the preprocessing information and calculate the ESF value;

[0055] Step 3.1: Initialize the image: initialize the preprocessing information;

[0056] Step 3.1.1: Apply for the variable args_list to store the generated image index, start and end positions;

[0057] Step 3.1.2: Prepare variables according to different structural variation types;

[0058] Step 3.1.3: Count the number of different structural variation types;

[0059] Step 3.2: Feature image generation: Generate feature images based on different mutation types and quantities;

[0060] Step 3.2.1: Load the variables split_read_left, split_read_right, rd_count, conjugate_m, conjugate_i, conjugate_d, and conjugate_s.

[0061] Step 3.2.2: Calculate the minimum dimension of each image;

[0062] Step 3.2.3: Define the image height as the maximum value of each dimension minus the minimum value plus 2;

[0063] Step 3.2.4: Fill the initial image (length equal to the length of each dimension and height equal to the image height defined above) with zeros.

[0064] Step 3.2.5: Generate an initial image based on the variable information. The length of the image is the length of the variation region, and the height of each pixel is equal to the value of the variable.

[0065] Step 3.2.5.1: Get the length of each dimension in the variable information and use the length as the horizontal coordinate of the image pixel;

[0066] Step 3.2.5.2: Calculate the difference between the value of each site in the variable information and the minimum value, and use this difference as the vertical coordinate corresponding to each position in the image;

[0067] Step 3.2.5.3: Traverse each site, calculate the values of all sites, and generate the initial image;

[0068] Step 3.2.6: Transform the image and resize it to 244×244;

[0069] Step 3.2.7: Save the generated statistical information image into the all_img.pt file and the generated CIGAR signal image into the all_img_mids file;

[0070] Step 3.3: Calculating the ESF value: The specific embodiment is to calculate the ESF value of the CIGAR signal;

[0071] Step 3.3.1: Allocate memory space for ESF variables conjugate_esf_list;

[0072] Step 3.3.2: Load the variables conjugate_m, conjugate_i, conjugate_d, and conjugate_s;

[0073] Step 3.3.3: Calculate the length of the four CIGAR signals for each variable, and calculate their minimum, first quartile, median, third quartile, maximum, RMS value, harmonic mean, mean, standard deviation, and coefficient of variation. Store the results in the variable conjugate_esf_list.

[0074] Step 3.3.4: After generation, save the result of the variable conjugate_esf_list into the all_list.pt file;

[0075] Step 3.4: After generation is complete, check whether the number of feature images and ESF values is consistent with the number of variant sites. If not, return to step 3.2. If consistent, proceed to step 4.

[0076] Step 4: Set labels for corresponding feature images according to the structural variation type;

[0077] Step 4.1: Create a new image storage folder: Create a folder for each structural variation type;

[0078] Step 4.2: Set generated images / ESF training labels: Traverse each image and label it with the structural variation type according to the index and variation type in args_list for later model training;

[0079] Step 4.3: Save related files into folders: Save the labeled images into folders corresponding to the variant types according to the variant types;

[0080] Step 5: Use the CLIP multimodal model for training to obtain a trained model;

[0081] Step 5.1: Configure model parameters: Configure model training parameters;

[0082] Step 5.2: Dataset preparation: Dataset preparation, random selection;

[0083] Step 5.2.1: Set the random number generator seed;

[0084] Step 5.2.2: Randomly shuffle the elements of the training list according to the seed sequence of the random number generator;

[0085] Step 5.2.3: Randomly select 80% of the data as training data and 20% of the data as validation and test data;

[0086] Step 5.3: Load CLIP model training;

[0087] Step 5.4: Save the trained model;

[0088] Step 6: Process the VCF to be filtered to generate a feature image / ESF: Obtain the structural variation results to be processed, generate a feature image based on the structural variation sites, and calculate the ESF value;

[0089] Step 6.1: Process the VCF file of the structural variant to be processed in the same way as step 1.2;

[0090] Step 6.2: Process the (BAM / SAM) file in the same way as steps 1.2-1.3 to extract depth information.

[0091] Step 6.3: Preprocess the data file in the same way as step 2;

[0092] Step 6.4: Generate the feature image and calculate the ESF value in the same way as step 3;

[0093] Step 7: Model result classification and filter result writing: Use the trained model to filter and obtain the filtered VCF file;

[0094] Step 7.1: Load the trained model to classify the image and ESF of each variant site to be processed;

[0095] Step 7.1.1: Load the trained model and set the model parameters;

[0096] Step 7.1.2: Use the model to classify the images and ESF of the variant sites to be processed;

[0097] Step 7.1.3: Keep the entries that are consistent with the model classification results and delete / modify the inconsistent entries;

[0098] Step 7.2: Write the filtered results into a VCF file.

[0099] It is worth noting that based on multimodal technology, it integrates structural variation feature image information and ESF text information, and adopts CLIP model for training. It is suitable for second-generation and third-generation sequencing, and can achieve high-precision structural variation identification of gene sequences. It can also be integrated with existing structural variation detection tools to significantly reduce false positive results.

[0100] It should be understood that although Figure 1-Figure 2 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 1-Figure 2 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0101] In one embodiment, Figure 3 As shown, a structural variation filtering device based on multimodal fusion is provided, comprising: a feature image generation module 302, a multimodal data annotation module 304 and a filtering module 306, wherein:

[0102] The feature image generation module 302 is used to encode the variable information of the structural variation site and generate a feature image of the structural variation site.

[0103] The multimodal data labeling module 304 is configured to, after calculating the ESF value of the variable information, set a label for the image-text pair consisting of the structural variation site feature image and the ESF value according to the structural variation type to obtain a labeled image-text pair.

[0104] The filtering module 306 is used to train the CLIP multimodal model according to preset configuration parameters to obtain a trained multimodal fusion model, and filter the preprocessed structural variation data and labeled image text pairs for variation sites using the trained multimodal fusion model to obtain filtered result data.

[0105] The specific definitions of the multimodal fusion-based structural variation filtering device can be found in the above-mentioned definitions of the multimodal fusion-based structural variation filtering method, and will not be repeated here. The various modules in the multimodal fusion-based structural variation filtering device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above modules can be embedded in or independent of the processor of the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0106] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a structural variation filtering method based on multimodal fusion is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0107] Those skilled in the art will understand that Figure 3-Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0108] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0109] The variable information of the structural variation site is encoded to generate a characteristic image of the structural variation site.

[0110] After calculating the ESF value of the variable information, the image-text pair consisting of the characteristic image of the structural variation site and the ESF value is labeled according to the structural variation type to obtain a labeled image-text pair.

[0111] The CLIP multimodal model is trained according to preset configuration parameters to obtain a trained multimodal fusion model. The preprocessed structural variation data and labeled image text pairs are filtered for variation sites through the trained multimodal fusion model to obtain filtered result data.

[0112] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0113] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0114] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and such modifications and improvements are intended to fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A structural variation filtering method based on multimodal fusion, characterized in that: The method comprises: Encode the variable information of the structural variation site and generate a characteristic image of the structural variation site; After calculating the ESF value of the variable information, a label is set for the image-text pair consisting of the structural variation site feature image and the ESF value according to the structural variation type to obtain a labeled image-text pair; the ESF value of the preprocessed information in the variable information is calculated, and a check is made to see whether the total number of the structural variation site feature images and the ESF values is consistent with the total number of structural variation sites. If they are consistent, a label is set for the image-text pair consisting of the structural variation site feature images according to the structural variation type to obtain a labeled image-text pair; otherwise, the variable information is re-acquired to generate the structural variation site feature images; the ESF value is a statistical feature set calculated based on the CIGAR signal; The CLIP multimodal model is trained according to preset configuration parameters, standard image data, and standard text data to obtain a trained multimodal fusion model. The preprocessed structural variation data and the labeled image-text pairs are filtered for variation sites using the trained multimodal fusion model to obtain filtered result data.

2. The method according to claim 1, characterized in that Encode the variable information of the structural variation site and generate a characteristic picture of the structural variation site, including: Extracting depth information of the structural variation sites in the alignment file, and recording the coverage depth of each chromosome location according to the depth information; and preprocessing the read sequence of the alignment file to obtain preprocessing information; The coverage depth and the preprocessing information are used as variable information of each of the structural variation sites, and a structural variation site feature image is generated according to the variable information; the height of each pixel in the structural variation site feature image is equal to the value of the variable in the variable information.

3. The method according to claim 2, characterized in that The pre-processed structural variation data is obtained by encoding the variable information of the structural variation site of the structural variation data, generating a feature image to be processed, and calculating the ESF value.

4. The method according to claim 3, characterized in that The pre-processed structural variation data and the labeled image-text pairs are filtered for variation sites using the trained multimodal fusion model to obtain filtered result data, including: The trained multimodal fusion model is used to classify the feature images and ESF values corresponding to the preprocessed structural variation data and the labeled image-text pairs at each variation site to be processed to obtain a model classification result. The false positive results of the structural variation data are filtered according to the model classification result to obtain filtered result data.

5. The method according to claim 4, characterized in that Filtering false positive results of the structural variation data according to the model classification result to obtain filtered result data, including: In the structural variation data, entries consistent with the model classification result are retained, and entries inconsistent with the model classification result are deleted or modified to obtain filtering result data.

6. A structural variation filtering device based on multimodal fusion, characterized in that: The device comprises: A feature image generation module is used to encode the variable information of the structural variation site and generate a feature image of the structural variation site; a multimodal data annotation module configured to, after calculating the ESF value of the variable information, label the image-text pair consisting of the structural variation site feature image and the ESF value according to the structural variation type to obtain a labeled image-text pair; calculate the ESF value of the preprocessed information in the variable information, check whether the total number of the structural variation site feature images and the ESF values is consistent with the total number of structural variation sites; if so, label the image-text pair consisting of the structural variation site feature images according to the structural variation type to obtain a labeled image-text pair; otherwise, reacquire the variable information to generate the structural variation site feature image; the ESF value is a statistical feature set calculated based on the CIGAR signal; The filtering module is used to train the CLIP multimodal model according to preset configuration parameters, standard image data and standard text data to obtain a trained multimodal fusion model. The trained multimodal fusion model is used to filter the variation sites of the preprocessed structural variation data and the labeled image text pairs to obtain the filtered result data.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

Images