A genetic gene association analysis system based on multi-modal data fusion

Abstract

The application provides a genetic gene correlation analysis system based on multi-modal data fusion. It relates to the technical field of genetic data analysis, and comprises a multi-modal data processing module, a single-modal correlation evidence generation module, a cross-modal evidence conflict evaluation module, a dynamic fusion module, and a gene correlation analysis module. The multi-modal data processing module acquires and pre-processes genetic multi-modal data. The single-modal correlation evidence generation module generates correlation evidence of each modality with respect to a target phenotype. The cross-modal evidence conflict evaluation module constructs a consistency matrix and generates a conflict penalty factor. The dynamic fusion module generates a dynamic fusion weight and forms a fused correlation feature. The gene correlation analysis module outputs candidate genes and a ranking. The genetic gene correlation analysis system based on multi-modal data fusion can reduce the interference of low-quality modalities and conflicting modalities on the results, and improve the accuracy and stability of gene correlation analysis.

Description

Technical Field

[0001] This invention relates to the field of genetic data analysis technology, specifically to a genetic gene association analysis system based on multimodal data fusion. Background Technology

[0002] Genetic association analysis (CAA) is a technique that analyzes genotype data, sequencing data, gene expression data, phenotypic data, and clinical information to determine the association between genes, genetic variants, or gene expression characteristics and diseases, traits, or biological functions. Existing techniques commonly include genome-wide association analysis (GWAS), candidate gene analysis, quantitative trait locus analysis (CTTLA), and multi-omics data-based joint modeling analysis. These methods typically involve preprocessing and feature extraction of single or a few types of data, followed by statistical tests, regression models, machine learning models, or multimodal fusion models to obtain candidate genes, variants, or feature combinations associated with the target phenotype. With the development of high-throughput sequencing, clinical testing, electronic medical records, omics testing, and medical imaging technologies, the sources of data available for genetic analysis are constantly increasing, and multimodal data fusion has gradually become an important technical approach to improve the accuracy and comprehensive interpretability of CAA.

[0003] However, different modal data often differ in terms of completeness, noise level, data scale, and strength of association evidence. Existing technologies often use feature splicing, unified dimensionality reduction, or fixed weight fusion methods, which make it difficult to dynamically adjust the fusion process according to the reliability of each modal data and cross-modal consistency. This results in low-quality data or conflicting data easily interfering with gene association analysis results, reducing the accuracy and stability of candidate gene or key variant site screening. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a genetic association analysis system based on multimodal data fusion. By constructing single-modal association evidence, identifying cross-modal evidence conflicts, and dynamically adjusting the fusion weights using a conflict penalty factor, it solves the problem of low-quality or conflicting modalities interfering with gene association analysis results in multimodal genetic data.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a genetic association analysis system based on multimodal data fusion, comprising: A multimodal data processing module is used to acquire genetic multimodal data corresponding to the object to be analyzed, and to preprocess the genetic multimodal data to obtain a multimodal feature set; wherein, the multimodal feature set includes at least one genetic object to be analyzed, which includes a gene object to be analyzed or a genetic variation object to be analyzed; A single-modal association evidence generation module is used to determine, based on each modal feature, single-modal association evidence of the genetic object to be analyzed relative to the same target phenotype for the same genetic object to be analyzed. The cross-modal evidence conflict evaluation module is used to construct a cross-modal evidence consistency matrix based on multiple single-modal association evidences of the same genetic object to be analyzed relative to the same target phenotype, and to identify conflicting modal features that conflict with other modal features based on the cross-modal evidence consistency matrix, and generate conflict penalty factors corresponding to the conflicting modal features. The dynamic fusion module is used to generate dynamic fusion weights corresponding to each modal feature based on the modal reliability value corresponding to each modal feature, the cross-modal evidence consistency matrix, and the conflict penalty factor, and to fuse multiple modal features corresponding to the same genetic object to be analyzed according to the dynamic fusion weights to obtain the fusion association features of the genetic object to be analyzed relative to the target phenotype; wherein, the conflict penalty factor is used to reduce the dynamic fusion weights corresponding to the conflict modal features; The gene association analysis module is used to determine the gene association results between the genetic object to be analyzed and the target phenotype based on the fusion association features, and output candidate genes, candidate genetic variation sites and corresponding association ranking results according to the gene association results.

[0006] Preferably, the genetic multimodal data includes at least two of the following: genotype data, transcriptome expression data, phenotypic measurement data, clinical laboratory data, proteome data, and metabolome data; the target phenotype includes disease state, quantitative trait, drug response phenotype, or physiological function phenotype.

[0007] Preferably, the multimodal data processing module includes a sample alignment unit, a missing data marker unit, and a standardization unit; the sample alignment unit is used to map different modal data to the same object to be analyzed according to the sample identifier of the object to be analyzed; the missing data marker unit is used to record the missing status of each object to be analyzed in each modal data; the standardization unit is used to perform scale unification processing on different modal data to form the multimodal feature group.

[0008] Preferably, the single-modal association evidence generation module is used to establish a corresponding single-modal association model for the same genetic object to be analyzed, taking each modal feature as an input variable and the target phenotype as a response variable or classification label, and determining the single-modal association evidence based on the output of the single-modal association model; wherein, the single-modal association evidence includes association direction, association strength, and association significance, wherein the association direction is used to indicate a positive or negative association between the corresponding modal feature and the target phenotype, the association strength is used to indicate the degree of influence of the corresponding modal feature on the target phenotype, and the association significance is used to indicate the statistical support of the association relationship between the corresponding modal feature and the target phenotype.

[0009] Preferably, the single-modal association evidence generation module is used to determine the association direction based on the sign of the model coefficients or the direction of the feature contribution value, determine the association strength based on the effect size, correlation coefficient, absolute value of regression coefficient or feature importance, and determine the association significance based on the significance test results, confidence interval results or model stability evaluation results.

[0010] Preferably, the cross-modal evidence conflict evaluation module is used to construct the cross-modal evidence consistency matrix using the modal features corresponding to the same genetic object to be analyzed as matrix rows and matrix columns; the matrix elements in the cross-modal evidence consistency matrix are used to represent the consistency status of the association direction, the degree of deviation of the association strength, and the support status of the association significance between the single-modal association evidence corresponding to two modal features.

[0011] Preferably, the cross-modal evidence conflict evaluation module is used to count the number of directional conflicts between any modal feature and other modal features based on the cross-modal evidence consistency matrix; when the number of directional conflicts reaches a preset conflict number condition, and the degree of deviation of the correlation strength between any modal feature and other modal features that have directional conflicts reaches a preset deviation condition, the any modal feature is determined as the conflicting modal feature.

[0012] Preferably, the cross-modal evidence conflict evaluation module is used to generate the conflict penalty factor based on the number of directional conflicts, the degree of deviation in association strength, and the degree of support for conflict evidence of the conflict modal features; wherein, the degree of support for conflict evidence is jointly determined by the association significance of the conflict modal features and the association significance of other modal features that have directional conflicts with the conflict modal features, and the number of directional conflicts, the degree of deviation in association strength, and the degree of support for conflict evidence are respectively used as positive penalty terms in the generation of the conflict penalty factor.

[0013] Preferably, the dynamic fusion module is used to generate the modal reliability value based on the effective sample coverage, missing proportion, noise level, and scale deviation degree corresponding to each modal feature; wherein, the effective sample coverage is used as a positive evaluation term in the generation of the modal reliability value, and the missing proportion, the noise level, and the scale deviation degree are used as negative evaluation terms in the generation of the modal reliability value.

[0014] Preferably, the dynamic fusion module is used to generate initial reliability weights corresponding to each modality feature based on the modality reliability value, generate consistency correction coefficients based on the cross-modality evidence consistency matrix, and reduce the initial reliability weights based on the conflict penalty factor to obtain corrected weights; normalize the corrected weights corresponding to each modality feature to obtain the dynamic fusion weights; the gene association analysis module is used to calculate the fusion association score corresponding to each genetic object to be analyzed based on the fusion association feature, and generate the association ranking result according to the fusion association score.

[0015] This invention provides a genetic association analysis system based on multimodal data fusion. It has the following beneficial effects:

[0016] This genetic association analysis system based on multimodal data fusion, by setting up a single-modal association evidence generation module and a cross-modal evidence conflict evaluation module, first generates single-modal association evidence for each modal feature relative to the same target phenotype for the same genetic object to be analyzed. Then, it constructs a cross-modal evidence consistency matrix based on multiple single-modal association evidences, thereby identifying conflicting modal features between different modalities that are inconsistent in association direction, association strength, or significance support. Compared with methods that directly perform feature splicing, uniform dimensionality reduction, or fixed-weight fusion of multimodal data, this invention can determine whether the association evidence provided by each modality supports each other before fusion, avoiding the indiscriminate inclusion of conflicting modal features in the fusion process, thereby reducing the interference of contradictory modal evidence on the genetic association analysis results.

[0017] This invention also utilizes a dynamic fusion module to generate dynamic fusion weights based on modality reliability values, cross-modality evidence consistency matrices, and conflict penalty factors. The conflict penalty factor is then used to reduce the dynamic fusion weights corresponding to conflicting modality features, allowing the contributions of different modality features during the fusion process to be adjusted according to data quality and evidence consistency. Therefore, even when different modality data exhibit differences in integrity, noise, scale, and the strength of association evidence, more stable fusion association features can still be obtained. Based on these fusion association features, candidate genes, candidate genetic variation sites, and association ranking results are output, thereby improving the accuracy, stability, and interpretability of gene association analysis results. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of the genetic gene association analysis system based on multimodal data fusion of the present invention; Figure 2 This is a flowchart of the cross-modal evidence conflict identification and conflict penalty factor generation process of the present invention; Figure 3 This is a flowchart of the dynamic fusion weight generation and associated sorting output of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 like Figure 1 As shown, this embodiment of the invention provides a genetic gene association analysis system based on multimodal data fusion, including a multimodal data processing module for acquiring genetic multimodal data corresponding to the object to be analyzed, and preprocessing the genetic multimodal data to obtain a multimodal feature set; wherein, the multimodal feature set includes at least one genetic object to be analyzed corresponding to multiple modal features, and the genetic object to be analyzed includes a gene object to be analyzed or a genetic variation object to be analyzed.

[0021] Among them, multiple modal features used for association analysis of the genetic object to be analyzed refer to different modal features that participate in the association analysis of the target phenotype together with the genetic object to be analyzed in the same batch of objects to be analyzed; features in clinical test data, proteomic data and metabolomic data that cannot be directly mapped to a single gene or genetic variant can be used as auxiliary modal features or covariate features in the association analysis of the genetic object to be analyzed to participate in subsequent evidence generation.

[0022] Genetic multimodal data include at least two of the following: genotype data, transcriptome expression data, phenotypic measurement data, clinical laboratory data, proteome data, and metabolome data; target phenotypes include disease states, quantitative traits, drug response phenotypes, or physiological function phenotypes.

[0023] The phenotypic measurement data used as input are phenotypic auxiliary data other than the target phenotypic; the target phenotypic is used as a response variable or classification label in the single-modal association model and does not directly participate in dynamic fusion as an input modal feature to be fused.

[0024] The multimodal data processing module includes a sample alignment unit, a missing data marker unit, and a standardization unit. The sample alignment unit is used to map different modal data to the same object to be analyzed according to the sample identifier of the object to be analyzed. The missing data marker unit is used to record the missing status of each object to be analyzed in each modal data. The standardization unit is used to perform scale unification processing on different modal data to form multimodal feature groups.

[0025] The single-modal association evidence generation module is used to determine the single-modal association evidence of the same genetic object relative to the same target phenotype based on each modal feature.

[0026] The unimodal association evidence generation module is used to establish corresponding unimodal association models for the same genetic object to be analyzed, taking each modal feature as an input variable and the target phenotype as a response variable or classification label. The module then determines the unimodal association evidence based on the output of the unimodal association model. The unimodal association evidence includes association direction, association strength, and association significance. Association direction indicates whether the corresponding modal feature and the target phenotype are positive or negatively associated. Association strength indicates the degree of influence of the corresponding modal feature on the target phenotype. Association significance indicates the statistical support for the association between the corresponding modal feature and the target phenotype.

[0027] Specifically, for the same genetic object to be analyzed, a single-modal association model is established with the target phenotype using its genotype characteristics, expression characteristics, clinical auxiliary characteristics or omics auxiliary characteristics. Each single-modal association model outputs the association direction, association strength and association significance under the corresponding modality, so that each modality first independently forms association evidence for the same target phenotype.

[0028] The single-modal association evidence generation module is used to determine the association direction based on the sign of the model coefficients or the direction of the feature contribution value, determine the association strength based on the effect size, correlation coefficient, absolute value of regression coefficient or feature importance, and determine the association significance based on the significance test results, confidence interval results or model stability evaluation results.

[0029] The cross-modal evidence conflict assessment module is used to construct a cross-modal evidence consistency matrix based on multiple single-modal association evidences of the same genetic object relative to the same target phenotype, and to identify conflicting modal features that conflict with other modal features based on the cross-modal evidence consistency matrix, and generate conflict penalty factors corresponding to the conflicting modal features.

[0030] The cross-modal evidence conflict assessment module is used to construct a cross-modal evidence consistency matrix with each modal feature corresponding to the same genetic object to be analyzed as the matrix rows and columns. The matrix elements in the cross-modal evidence consistency matrix are used to represent the consistency of association direction, the degree of deviation of association strength, and the support status of association significance between the single-modal association evidence corresponding to two modal features.

[0031] In one implementation, each matrix element in the cross-modal evidence consistency matrix corresponds to the evidence comparison result between two modal features; when the two modal features have the same association direction, the difference in association strength does not reach the preset deviation condition, and both have statistical support, it is determined that the cross-modal evidence consistency between the two is high; when the two have opposite association directions or the difference in association strength reaches the preset deviation condition, it is determined that there is a cross-modal evidence conflict between the two.

[0032] The cross-modal evidence conflict assessment module is used to count the number of directional conflicts between any modal feature and other modal features based on the cross-modal evidence consistency matrix. When the number of directional conflicts reaches the preset conflict quantity condition, and the degree of deviation of the correlation strength between any modal feature and other modal features that have directional conflicts reaches the preset deviation condition, the modal feature is identified as a conflicting modal feature.

[0033] The cross-modal evidence conflict assessment module is used to generate conflict penalty factors based on the number of directional conflicts, the degree of deviation in association strength, and the degree of support for conflict evidence of conflict modal features. The degree of support for conflict evidence is jointly determined by the association significance of the conflict modal features and the association significance of other modal features that have directional conflicts with the conflict modal features. The number of directional conflicts, the degree of deviation in association strength, and the degree of support for conflict evidence are used as positive penalty terms in the generation of the conflict penalty factors.

[0034] The dynamic fusion module is used to generate dynamic fusion weights for each modal feature based on the modal reliability value, cross-modal evidence consistency matrix, and conflict penalty factor. It then fuses multiple modal features corresponding to the same genetic object to be analyzed according to the dynamic fusion weights to obtain the fusion association features of the genetic object to be analyzed relative to the target phenotype. The conflict penalty factor is used to reduce the dynamic fusion weights corresponding to conflict modal features.

[0035] The dynamic fusion module is used to generate modal reliability values ​​based on the effective sample coverage, missing proportion, noise level, and scale deviation corresponding to each modal feature. Among them, the effective sample coverage is used as a positive evaluation term in the generation of modal reliability values, while the missing proportion, noise level, and scale deviation are used as negative evaluation terms in the generation of modal reliability values.

[0036] The dynamic fusion module generates initial reliability weights for each modal feature based on modal reliability values, generates consistency correction coefficients based on the cross-modal evidence consistency matrix, and reduces the initial reliability weights based on the conflict penalty factor to obtain corrected weights. The corrected weights for each modal feature are then normalized to obtain dynamic fusion weights. After normalization, the sum of the dynamic fusion weights for each modal feature is 1. The dynamic fusion module then performs weighted fusion of multiple modal features of the same genetic object to be analyzed according to each dynamic fusion weight, generating fusion association features for subsequent gene association analysis. The gene association analysis module calculates the fusion association score for each genetic object to be analyzed based on the fusion association features and generates association ranking results according to the fusion association scores.

[0037] The gene association analysis module is used to determine the gene association results between the genetic object to be analyzed and the target phenotype based on the fusion association characteristics, and output candidate genes, candidate genetic variation sites and corresponding association ranking results according to the gene association results.

[0038] In this way, this embodiment can identify conflicting evidence between different modalities before multimodal genetic data fusion, and reduce the fusion weight of conflict modal features by using a conflict penalty factor, thereby reducing the interference of low-quality modalities or conflict modalities on the ranking results of candidate genes and candidate genetic variation sites.

[0039] Example 2 like Figure 2 As shown in Example 1, this example further provides a method for cross-modal evidence conflict identification and conflict penalty factor generation, which is used to determine whether there is a conflict in the single-modal association evidence formed by the same genetic object under different modalities before multimodal feature fusion, and to generate corresponding conflict penalty factors for the modal features that have conflicts.

[0040] Specifically, the same genetic object G1 is selected as the analysis object, and the target phenotype is the disease state. For this genetic object G1, the single-modal association evidence generation module establishes single-modal association models based on genotype modality features, transcriptome expression modality features, proteome modality features, and clinical auxiliary modality features, and outputs the single-modal association evidence of each modality feature relative to the target phenotype. As an example, the association direction corresponding to the genotype modality feature is positive, with an association strength of 0.42 and an association significance value of 0.018; the association direction corresponding to the transcriptome expression modality feature is positive, with an association strength of 0.39 and an association significance value of 0.026; the association direction corresponding to the proteome modality feature is negative, with an association strength of 0.86 and an association significance value of 0.009; and the association direction corresponding to the clinical auxiliary modality feature is positive, with an association strength of 0.47 and an association significance value of 0.041. The above values ​​are only used to illustrate the implementation of this embodiment and do not constitute a limitation on the scope of protection of this invention.

[0041] The cross-modal evidence conflict assessment module constructs a cross-modal evidence consistency matrix corresponding to the genetic object G1 to be analyzed, using genotype modality features, transcriptome expression modality features, proteome modality features, and clinical auxiliary modality features as matrix rows and columns. The matrix elements in this cross-modal evidence consistency matrix are used to record the evidence comparison results between any two modality features, including the consistency status of association direction, the degree of deviation in association strength, and the support status of association significance.

[0042] In one implementation, if the association directions of two modal features are the same, their association direction consistency is recorded as consistent; if the association directions of two modal features are opposite, their association direction consistency is recorded as conflicting. The degree of association strength deviation is the absolute difference between the corresponding association strengths of the two modal features; if the degree of association strength deviation reaches a preset deviation condition, it is determined that there is an association strength deviation between the two. If the association significance values ​​corresponding to the two modal features both meet the preset significance condition, it is determined that the conflict or consistency result between the two has significant support.

[0043] Taking the example data above, the association direction of the proteomic modality feature is negative, while the association directions of the genotype modality feature, transcriptome expression modality feature, and clinical auxiliary modality feature are all positive. Therefore, the proteomic modality feature forms a directional conflict with the genotype modality feature, transcriptome expression modality feature, and clinical auxiliary modality feature, respectively. Furthermore, the deviation in association strength between the proteomic modality feature and the genotype modality feature is 0.44, the deviation is 0.47 between the proteomic modality feature and the transcriptome expression modality feature, and the deviation is 0.39 between the proteomic modality feature and the clinical auxiliary modality feature. If the preset deviation condition is that the deviation in association strength is not less than 0.30, then the proteomic modality feature satisfies the correlation strength deviation condition with all three modality features mentioned above.

[0044] The cross-modal evidence conflict assessment module further counts the number of directional conflicts between any modal feature and other modal features. For proteomics modal features, the number of directional conflicts is 3; for genotype modal features, transcriptome expression modal features, and clinical auxiliary modal features, they only have directional conflicts with proteomics modal features. If the preset conflict quantity condition is that the number of directional conflicts is not less than 2, then the proteomics modal feature meets the preset conflict quantity condition. At the same time, the deviation of the association strength between the proteomics modal feature and other modal features with directional conflicts all meet the preset deviation condition. Therefore, the cross-modal evidence conflict assessment module identifies the proteomics modal feature as a conflicting modal feature.

[0045] After identifying conflict modal features, the cross-modal evidence conflict assessment module generates a conflict penalty factor based on the number of directional conflicts, the degree of deviation in association strength, and the degree of support for conflict evidence. The number of directional conflicts indicates the number of times a modal feature experiences directional conflicts with other modal features; the degree of deviation in association strength indicates the difference in association strength between the modal feature and other modal features that experience directional conflicts; and the degree of support for conflict evidence is determined jointly by the association significance value of the conflict modal feature and the association significance values ​​of other modal features that experience directional conflicts with it.

[0046] As an example, a proteomic modality feature has 3 directional conflicts and an average association strength deviation of 0.43. Furthermore, all modality features that conflict with it meet the preset significance criteria. Based on this, the cross-modal evidence conflict assessment module determines that the proteomic modality feature meets the preset significance support criteria and generates a corresponding conflict penalty factor. The number of directional conflicts, the association strength deviation, and the degree of conflict evidence support are all used as positive penalty terms in the generation of the conflict penalty factor, enabling the conflict penalty factor to reflect the degree of interference of the conflicting modality feature on the subsequent dynamic fusion process.

[0047] With the processing method described above in this embodiment, when different modal data can form single-modal association evidence, the system does not directly splice or fuse each modal feature with fixed weights. Instead, it first determines whether the evidence from different modalities is consistent with each other under the same genetic object to be analyzed and the same target phenotype. For conflict modal features whose association direction is inconsistent with most modalities and whose association strength deviates from the preset deviation conditions, a corresponding conflict penalty factor is generated so that the subsequent dynamic fusion module can reduce the dynamic fusion weight corresponding to the conflict modal feature, thereby reducing the interference of conflict modal features on the gene association analysis results.

[0048] Example 3 like Figure 3 As shown, based on Examples 1 and 2, this example further provides a dynamic fusion weight generation and association ranking output method, which is used to dynamically fuse multiple modal features corresponding to the same genetic object to be analyzed after identifying conflict modal features and generating conflict penalty factors, and output the association ranking results of candidate genes or candidate genetic variation sites.

[0049] Specifically, the dynamic fusion module first generates modal reliability values ​​based on the effective sample coverage, missing proportion, noise level, and scale deviation corresponding to each modal feature. Effective sample coverage is used as a positive evaluation term in the generation of modal reliability values, while missing proportion, noise level, and scale deviation are used as negative evaluation terms. Therefore, modal features with sufficient sample coverage, few missing samples, low noise, and small scale deviation can obtain relatively higher modal reliability values; while modal features with poor data quality have reduced initial contributions in subsequent fusion processes.

[0050] Subsequently, the dynamic fusion module generates initial reliability weights based on the modal reliability values ​​corresponding to each modal feature, and generates consistency correction coefficients based on the cross-modal evidence consistency matrix. The consistency correction coefficients represent the degree of consistency of the association evidence between the corresponding modal feature and other modal features. When a modal feature has a consistent association direction, a small deviation in association strength, and consistent association significance support status with other modal features, the consistency correction coefficient corresponding to that modal feature is used to retain or increase its initial reliability weight. When a modal feature is identified as a conflicting modal feature, the dynamic fusion module further reduces and corrects its initial reliability weight based on the conflict penalty factor corresponding to that conflicting modal feature, obtaining the corrected weights.

[0051] In one implementation, the dynamic fusion module normalizes the modified weights corresponding to each modality feature, ensuring that the sum of the dynamic fusion weights for multiple modality features corresponding to the same genetic object being analyzed is 1. After normalization, the dynamic fusion module performs weighted fusion of multiple modality features corresponding to the same genetic object being analyzed according to the dynamic fusion weights for each modality feature, obtaining the fusion association features of the genetic object being analyzed relative to the target phenotype. Therefore, conflicting modality features are not directly eliminated, but their fusion contribution is reduced based on their degree of conflict and data reliability, thus preserving usable information and reducing the interference of conflict evidence on the final result.

[0052] The gene association analysis module calculates the fusion association score for each genetic object to be analyzed based on the fusion association feature, and generates association ranking results according to the fusion association score. When the genetic object to be analyzed is a gene object, the gene association analysis module outputs candidate genes and their corresponding association ranking results; when the genetic object to be analyzed is a genetic variant object, the gene association analysis module outputs candidate genetic variant sites and their corresponding association ranking results.

[0053] In this embodiment, modal reliability, cross-modal evidence consistency, and conflict penalty factor are used together to generate dynamic fusion weights, thereby limiting the influence of low-reliability or conflicting modalities in the fusion process, thus improving the stability of fusion association features and the reliability of candidate gene or candidate genetic variation site ranking results.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A genetic association analysis system based on multimodal data fusion, characterized in that, include: A multimodal data processing module is used to acquire genetic multimodal data corresponding to the object to be analyzed, and to preprocess the genetic multimodal data to obtain a multimodal feature set; wherein, the multimodal feature set includes at least one genetic object to be analyzed, which includes a gene object to be analyzed or a genetic variation object to be analyzed; A single-modal association evidence generation module is used to determine, based on each modal feature, single-modal association evidence of the genetic object to be analyzed relative to the same target phenotype for the same genetic object to be analyzed. The cross-modal evidence conflict evaluation module is used to construct a cross-modal evidence consistency matrix based on multiple single-modal association evidences of the same genetic object to be analyzed relative to the same target phenotype, and to identify conflicting modal features that conflict with other modal features based on the cross-modal evidence consistency matrix, and generate conflict penalty factors corresponding to the conflicting modal features. The dynamic fusion module is used to generate dynamic fusion weights corresponding to each modal feature based on the modal reliability value corresponding to each modal feature, the cross-modal evidence consistency matrix, and the conflict penalty factor, and to fuse multiple modal features corresponding to the same genetic object to be analyzed according to the dynamic fusion weights to obtain the fusion association features of the genetic object to be analyzed relative to the target phenotype; wherein, the conflict penalty factor is used to reduce the dynamic fusion weights corresponding to the conflict modal features; The gene association analysis module is used to determine the gene association results between the genetic object to be analyzed and the target phenotype based on the fusion association features, and output candidate genes, candidate genetic variation sites and corresponding association ranking results according to the gene association results.

2. The genetic association analysis system based on multimodal data fusion according to claim 1, characterized in that: The genetic multimodal data includes at least two of the following: genotype data, transcriptome expression data, phenotypic measurement data, clinical laboratory data, proteome data, and metabolome data; the target phenotype includes disease state, quantitative trait, drug response phenotype, or physiological function phenotype.

3. The genetic association analysis system based on multimodal data fusion according to claim 1, characterized in that: The multimodal data processing module includes a sample alignment unit, a missing data marker unit, and a standardization unit. The sample alignment unit is used to map different modal data to the same object to be analyzed according to the sample identifier of the object to be analyzed. The missing data marker unit is used to record the missing status of each object to be analyzed in each modal data. The standardization unit is used to perform scale unification processing on different modal data to form the multimodal feature group.

4. The genetic association analysis system based on multimodal data fusion according to claim 1, characterized in that: The single-modal association evidence generation module is used to establish a corresponding single-modal association model for the same genetic object to be analyzed, taking each modal feature as an input variable and the target phenotype as a response variable or classification label, and determining the single-modal association evidence based on the output of the single-modal association model. The single-modal association evidence includes association direction, association strength, and association significance. The association direction indicates a positive or negative association between the corresponding modal feature and the target phenotype; the association strength indicates the degree of influence of the corresponding modal feature on the target phenotype; and the association significance indicates the statistical support for the association between the corresponding modal feature and the target phenotype.

5. A genetic association analysis system based on multimodal data fusion according to claim 4, characterized in that: The single-modal association evidence generation module is used to determine the association direction based on the sign of the model coefficients or the direction of the feature contribution value, determine the association strength based on the effect size, correlation coefficient, absolute value of regression coefficient or feature importance, and determine the association significance based on the significance test results, confidence interval results or model stability evaluation results.

6. The genetic association analysis system based on multimodal data fusion according to claim 1, characterized in that: The cross-modal evidence conflict evaluation module is used to construct the cross-modal evidence consistency matrix using the modal features corresponding to the same genetic object to be analyzed as matrix rows and matrix columns; the matrix elements in the cross-modal evidence consistency matrix are used to represent the consistency status of the association direction, the degree of deviation of the association strength, and the support status of the association significance between the single-modal association evidence corresponding to two modal features.

7. A genetic association analysis system based on multimodal data fusion according to claim 6, characterized in that: The cross-modal evidence conflict evaluation module is used to count the number of directional conflicts between any modal feature and other modal features based on the cross-modal evidence consistency matrix. When the number of directional conflicts reaches a preset conflict number condition, and the degree of deviation of the correlation strength between any modal feature and other modal features that have directional conflicts reaches a preset deviation condition, the any modal feature is determined as the conflicting modal feature.

8. A genetic association analysis system based on multimodal data fusion according to claim 7, characterized in that: The cross-modal evidence conflict evaluation module is used to generate the conflict penalty factor based on the number of directional conflicts, the degree of deviation in association strength, and the degree of support for conflict evidence of the conflict modal features; wherein, the degree of support for conflict evidence is jointly determined by the association significance of the conflict modal features and the association significance of other modal features that have directional conflicts with the conflict modal features, and the number of directional conflicts, the degree of deviation in association strength, and the degree of support for conflict evidence are respectively used as positive penalty terms in the generation of the conflict penalty factor.

9. A genetic association analysis system based on multimodal data fusion according to claim 1, characterized in that: The dynamic fusion module is used to generate the modal reliability value based on the effective sample coverage, missing proportion, noise level, and scale deviation corresponding to each modal feature; wherein, the effective sample coverage is used as a positive evaluation term in the generation of the modal reliability value, and the missing proportion, the noise level, and the scale deviation are used as negative evaluation terms in the generation of the modal reliability value.

10. A genetic association analysis system based on multimodal data fusion according to claim 1, characterized in that: The dynamic fusion module is used to generate initial reliability weights corresponding to each modal feature based on the modal reliability value, generate consistency correction coefficients based on the cross-modal evidence consistency matrix, and reduce and correct the initial reliability weights based on the conflict penalty factor to obtain corrected weights. The modified weights corresponding to each modal feature are normalized to obtain the dynamic fusion weights; The gene association analysis module is used to calculate the fusion association score corresponding to each genetic object to be analyzed based on the fusion association characteristics, and generate the association ranking result according to the fusion association score.

Images