Bioinformatics analysis system based on large model technology

Abstract

The present application relates to the technical field of bioinformatics, in particular to a biological information analysis system based on large model technology, which comprises a data analysis calibration module, a problem disintegration module, an analysis task arrangement module, a result mapping module and a feedback iteration module.In the present application, dynamic calibration of genome alignment and clinical phenotype timestamp eliminates time sequence misplacement deviation, base complementary pairing combined protein network abnormal screening enhances low-abundance collaborative variation capture, genotype-phenotype discrete distribution quantization unifies multi-modal data benchmark, solves multi-source heterogeneous standardization loss, classifies and integrates pathogenic gene semantic weight, balances statistical threshold and biological function, dynamically optimizes analysis sequence to synchronously cover key mutation area, improves non-coding region function annotation, integrates gene expression clustering and protein network topology in three-dimensional distribution, breaks through two-dimensional space limitation, and reduces genetic heterogeneity false positive rate through closed-loop feedback correction threshold iteration elimination rule.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioinformatics, and particularly relates to a bioinformatics analysis system based on large model technology. BACKGROUND

[0002] The technical field of bioinformatics involves the cross-application of multiple disciplines such as computer science, statistics, and biology, and is mainly used for analyzing and interpreting biological data. The core content of this technical field includes genomics, proteomics, metabolomics, and biological system modeling. Biological data is obtained through high-throughput sequencing technology, and combined with machine learning, mathematical modeling, and statistical analysis methods, biological information is extracted from massive data. The overall technical system of bioinformatics includes data acquisition, data storage, data analysis, data visualization, and other links, and is widely used in the fields of disease research, drug development, personalized medicine, agricultural breeding, etc.

[0003] Among them, the bioinformatics analysis system based on large model technology refers to using deep learning and natural language processing technology to build a biological data analysis model to improve data analysis capability and automation level. This patent subject covers technical matters such as biological sequence alignment, gene function prediction, protein structure analysis, and biological network reasoning, and through the construction of a neural network-based biological data analysis framework, the data mining capability is optimized using large-scale parameter training, the data labeling requirement is reduced using a self-supervised learning strategy, and the feature extraction capability is enhanced using an attention mechanism.

[0004] The existing technology relies on static time window processing of time series data, which is difficult to match the dynamic evolution characteristics of clinical phenotypes, and is prone to cross-period data correlation failure. Traditional variation detection uses a local feature extraction strategy, lacks global correlation rule mining across samples, and causes the omission of synergistic mechanisms. Linear task execution mode ignores the heterogeneity characteristics of variation distribution, causing delays in identifying key biomarkers and wasting resources. Isolated parameter analysis cannot reflect the spatial coupling effect of gene expression and protein interaction in three-dimensional topological relationships, limiting the explanation of molecular pathway mechanisms. Fixed thresholds and artificial rules are difficult to adapt to changes in sample size and type, and population genetic diversity scenarios are prone to cause false filtering or noise residue. Two-dimensional models cannot analyze the spatial clustering characteristics of protein networks, leading to deviations in functional annotation and experimental verification. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art, and a bioinformatics analysis system based on large model technology is proposed.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: the bioinformatics analysis system based on large model technology comprises:

[0007] The data analysis calibration module obtains genome sequence alignment information and clinical phenotype data time stamp, detects the overlapping range of gene variation site coordinates and sequence coding region, extracts cross-sample base complementary pairing mode, performs abnormal identification and removal operation on protein interaction network nodes, calculates the discrete distribution quantity of genotype-phenotype correlation strength, and generates multi-modal data standard value;

[0008] The problem disassembly module calls pathogenic gene annotation semantic weight and mutation function prediction logic dependency relationship based on the multi-modal data standard value, divides structural variation classification boundary numerical difference, and generates analysis task priority sequence;

[0009] The analysis task scheduling module calls the variation classification boundary in the analysis task priority sequence, filters the promoter region mutation and exon splice site mutation distribution range, adjusts the analysis execution sequence, and generates a task execution optimization coefficient;

[0010] The result mapping module extracts gene expression profile clustering center coordinates according to the task execution optimization coefficient, maps the mutation site enrichment area and protein interaction network topology parameters to three-dimensional space distribution, and generates a correlation resolution set;

[0011] The feedback iteration module detects the structural variation hotspot coverage and functional annotation correlation matrix matching degree in the correlation interpretation degree set, adjusts the sequence coding region determination threshold, updates the cross-modal data abnormal sample removal rule, and generates a process optimization parameter set.

[0012] As a further scheme of the application, the multi-modal data standard value includes genome sequence alignment information standard value, clinical phenotype data standard value, gene variation site coordinate standard value, sequence coding region overlapping range standard value, cross-sample base complementary pairing mode standard value, protein interaction network abnormal identification standard value, genotype-phenotype correlation strength discrete distribution quantity standard value, the analysis task priority sequence includes pathogenic gene annotation semantic weight, mutation function prediction logic dependency relationship, structural variation classification boundary numerical difference, the task execution optimization coefficient includes variation classification boundary, promoter region mutation distribution range, exon splice site mutation distribution range, analysis execution sequence adjustment parameter, the correlation resolution set includes gene expression profile clustering center coordinates, mutation site enrichment area, protein interaction network topology parameter three-dimensional space distribution.

[0013] As a further scheme of the application, the data analysis calibration module includes:

[0014] The genome sequence alignment submodule acquires genome sequence alignment information and clinical phenotypic data timestamps, parses genome sequence alignment data, extracts alignment results between samples, calculates alignment scores, obtains mutation sites in the genome alignment sequences, filters regions with high alignment coverage, and associates the genome variation information of each sample with phenotypic data timestamps to generate high-coverage gene sequence alignment data.

[0015] The gene variation functional region detection submodule detects the coordinates of gene variation sites based on the high-coverage gene sequence alignment data, analyzes the overlap between variation sites and genomic sequence coding regions, determines whether the variation site falls into a functional gene region, calculates the overlap ratio of variation sites, filters mutation sites with high overlap ratios, and generates a set of functional region variation sites.

[0016] The genotype-phenotype association calculation submodule calls the set of variant sites in the functional regions, extracts cross-sample base complementarity pairing patterns, analyzes the distribution of differential genotype combinations in phenotypic data, and calculates the discrete distribution of the genotype-phenotype association strength using the formula:

[0017]

[0018] The algorithm calculates the genotype-phenotype association strength, filters association patterns that conform to distribution characteristics, standardizes multimodal data, and obtains standard values ​​for multimodal data.

[0019] Where S represents the distribution value of genotype-phenotype association strength, Representing the i-th g Each genotype variable value, Representative and The corresponding phenotypic variable values, Representing the j-th g The values ​​of irrelevant variable variables, Representative and V jg The corresponding phenotypic variable value, n g m represents the number of genotype variables. g This represents the number of irrelevant variant variables.

[0020] As a further aspect of the present invention, the problem decomposition module includes:

[0021] The standard value calculation submodule, based on the standard values ​​of the multimodal data, analyzes the standard range of differential gene characteristic parameters, calls the gene expression level, variation frequency, and epigenetic regulatory factors of the target individual, calculates the deviation of each characteristic parameter from the standard value, and analyzes the overall variation by accumulating the deviation, using the formula:

[0022]

[0023] The operation obtains a standard value calculation deviation, and establishes a standard value deviation result;

[0024] Wherein, D represents a standard value calculation deviation, represents the i s th gene feature parameter value, represents a corresponding multi-modal data standard value, n s represents the number of gene feature parameters, represents an individual environmental factor value, represents an environmental factor reference value, m s represents the number of environmental factors;

[0025] The pathogenic gene annotation submodule analyzes the functional characteristics of the pathogenic gene based on the standard value deviation result, calls a known gene database to extract gene correlation characteristics, analyzes the mutation sites and influence range of the gene sequence, calculates the semantic weight value of the mutation function prediction logic, obtains the gene semantic weight calculation value, and establishes a pathogenic gene function annotation result;

[0026] The gene structural variation classification submodule calls the pathogenic gene function annotation result, filters the structural differences corresponding to the gene variation, calculates the classification boundary value of the multi-structural variation, and compares the standard value deviation result to generate a structural variation classification boundary numerical difference, and establishes an analysis task priority sequence.

[0027] As a further scheme of the application, the analysis task scheduling module comprises:

[0028] The variation classification boundary identification submodule extracts the variation category information corresponding to the multiple analysis tasks based on the analysis task priority sequence, calculates the distribution interval of the mutation type, compares the reference value of the mutation influence degree, identifies the classification boundary of the mutation, calculates the deviation of the multi-category boundary, and generates a variation classification boundary value;

[0029] The mutation distribution screening submodule calls the variation classification boundary value, screens the distribution range of the promoter region mutation and the exon splice site mutation, calculates the distribution proportion of the multiple mutation categories in the differential region, and classifies and sorts the distribution according to the mutation function influence coefficient, and generates a mutation distribution classification interval;

[0030] The task execution order adjustment submodule calls the mutation distribution classification interval, adjusts the analysis task execution order according to the task priority, and adopts the formula:

[0031]

[0032] The task execution optimization coefficient is calculated;

[0033] Wherein, T opt represents the task execution optimization coefficient, represent the adjusted task distribution proportion, represent the classification boundary value before adjustment, represent the priority weight of the task, m x represent the number of analysis tasks.

[0034] As a further scheme of the present application, the result mapping module comprises:

[0035] The task optimization coefficient calculation submodule extracts the mutation site distribution parameter, the protein interaction network connectivity parameter and the gene expression level parameter based on the task execution optimization coefficient, and adopts the formula:

[0036]

[0037] The operation obtains the optimization adjustment coefficient;

[0038] wherein C opt represents the optimization adjustment coefficient, V p represents the distribution density weighting coefficient of the mutation site, represents the mutation site distribution density of the protein interaction network, and a represents the adjustment coefficient, L d represents the gene expression level deviation, V n represents the network connectivity, S c represents the gene expression mean value offset adjustment coefficient;

[0039] The gene expression profile clustering submodule extracts the gene expression profile data based on the optimization adjustment coefficient, calculates the gene expression mean value, identifies the deviation of the mutation gene in the expression profile, filters the genes with the expression offset value exceeding the set threshold, calculates the expression profile clustering center coordinates of the mutation gene, and obtains the gene expression clustering center coordinates;

[0040] The gene network three-dimensional topological mapping submodule calculates the topological parameters of the mutation gene enrichment region in the protein interaction network based on the gene expression clustering center coordinates, maps its three-dimensional space distribution, and generates the correlation resolution set according to the connection of multiple genes in the topological network.

[0041] As a further scheme of the present application, the system further comprises:

[0042] The feedback iteration module detects the structural variation hotspot coverage rate and the functional annotation correlation matrix matching degree in the correlation interpretation degree set, adjusts the sequence coding region determination threshold, updates the cross-modal data abnormal sample elimination rule, and generates a set of process optimization parameters;

[0043] The set of process optimization parameters comprises the structural variation hotspot coverage rate, the functional annotation correlation matrix matching degree, the sequence coding region determination threshold, and the cross-modal data abnormal sample elimination rule.

[0044] As a further scheme of the present application, the feedback iteration module comprises:

[0045] The structural variation hotspot coverage sub-module calculates the coverage of multiple hotspots in the differential sequence coding region based on the structural variation hotspot coverage in the set of correlation explanation degrees, calls a decision threshold to screen hotspot regions with coverage meeting requirements, and uses the formula:

[0046]

[0047] The operation obtains a hotspot coverage metric value;

[0048] Wherein, C hot represents the hotspot coverage metric value, represents the coverage probability of the i t th structural variation hotspot, represents the correlation explanation degree score of the i t th structural variation hotspot, δ s represents an adjustment factor, V max represents the maximum correlation explanation degree score, N s represents the number of hotspots;

[0049] The functional annotation correlation matrix matching sub-module calls the hotspot coverage metric value, calculates the matching degree of multiple coding regions based on the functional annotation correlation matrix, extracts the matching scores of multiple regions, screens regions with matching scores higher than a set threshold, eliminates coding regions with low matching degrees, and obtains a functional matching coefficient;

[0050] The cross-modal data anomaly elimination sub-module calls the functional matching coefficient, adjusts the anomaly elimination rules of cross-modal data, calculates the deviation coefficient of each abnormal sample, screens abnormal samples according to the deviation coefficient threshold, eliminates data with deviation exceeding the standard range, and obtains a set of process optimization parameters.

[0051] Compared with the prior art, the present application has the advantages and positive effects that:

[0052] In the present application, the correlation deviation caused by time sequence misalignment is eliminated by dynamically calibrating the genomic sequence alignment information and the clinical phenotype timestamp data. The capture accuracy of low-abundance collaborative variation is enhanced by extracting the cross-sample base complementary pairing mode combined with protein network abnormal node screening. The unified benchmark of multi-modal data is established by quantifying the discrete distribution of genotype-phenotype correlation, solving the problem of missing standardization of traditional multi-source heterogeneous data. The semantic weight of pathogenic genes is integrated by dividing the numerical difference of structural variation classification boundary, balancing the statistical significance threshold and biological function rationality. The resource allocation is optimized by dynamically adjusting the analysis execution order, synchronously covering the key mutation regions of promoter and exon, and improving the efficiency of non-coding region function annotation. The spatial correlation expression limitation of two-dimensional model is broken through by integrating the gene expression clustering characteristics and protein network topology parameters in three-dimensional space distribution mapping. The false positive rate in the scenario of population genetic heterogeneity is reduced by real-time correction of the decision threshold and iterative elimination of rules through the closed-loop feedback mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 The system flowchart of the present application is shown in Figure 1.

[0054] Figure 2 The data analysis calibration module flowchart of the present application is shown in Figure 2.

[0055] Figure 3 The problem disassembly module flowchart of the present application is shown in Figure 3.

[0056] Figure 4 The analysis task scheduling module flowchart of the present application is shown in Figure 4.

[0057] Figure 5 The result mapping module flowchart of the present application is shown in Figure 5.

[0058] Figure 6 The feedback iteration module flowchart of the present application is shown in Figure 6. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical scheme and advantages of the present application clearer and more understandable, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0060] In the description of the present application, it should be understood that the terms "length", "width", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise explicitly and specifically limited.

[0061] Embodiment one

[0062] Please refer to Figure 1 The bioinformatics analysis system based on large model technology comprises:

[0063] The data analysis calibration module obtains genome sequence alignment information and clinical phenotype data time stamp, detects the overlapping range of gene variation site coordinates and sequence coding region, extracts cross-sample base complementary pairing mode, performs abnormal identification and removal operation on protein interaction network nodes, calculates the discrete distribution quantity of genotype-phenotype correlation strength, and generates multi-modal data standard value;

[0064] The problem decomposition module calls the pathogenic gene annotation semantic weight and mutation function prediction logical dependence relationship based on the multi-modal data standard value, divides the structural variation classification boundary value difference, and generates an analysis task priority sequence;

[0065] The analysis task scheduling module calls the variation classification boundary in the analysis task priority sequence, filters the distribution range of promoter region mutation and exon splice site mutation, adjusts the analysis execution order, and generates a task execution optimization coefficient;

[0066] The result mapping module extracts the gene expression profile cluster center coordinates according to the task execution optimization coefficient, maps the mutation site enrichment region and protein interaction network topology parameters to three-dimensional space distribution, and generates a correlation resolution set;

[0067] The feedback iteration module detects the structural variation hotspot coverage rate and functional annotation correlation matrix matching degree in the correlation explanation degree set, adjusts the sequence coding region judgment threshold, updates the cross-modal data abnormal sample removal rule, and generates a flow optimization parameter set.

[0068] The multi-modal data standard values include genomic sequence alignment information standard values, clinical phenotype data standard values, gene mutation site coordinate standard values, sequence coding region overlap range standard values, cross-sample base complementary pairing mode standard values, protein interaction network anomaly identification standard values, genotype-phenotype correlation strength discrete distribution quantity standard values, the analysis task priority sequence includes pathogenic gene annotation semantic weight, mutation function prediction logic dependency relationship, structural variation classification boundary numerical difference, the task execution optimization coefficient includes variation classification boundary, promoter region mutation distribution range, exon splice site mutation distribution range, analysis execution order adjustment parameter, the correlation resolution set includes gene expression profile clustering center coordinates, mutation site enrichment region, protein interaction network topology parameter three-dimensional space distribution, the process optimization parameter set includes structural variation hotspot coverage, functional annotation correlation matrix matching degree, sequence coding region judgment threshold, cross-modal data abnormal sample rejection rule.

[0069] Please refer to Figure 2 , the data analysis calibration module includes:

[0070] The genomic sequence alignment submodule obtains genomic sequence alignment information and clinical phenotype data timestamps, analyzes genomic sequence alignment data, extracts alignment results between samples, calculates alignment scores, obtains mutation sites in genomic alignment sequences, filters regions with high alignment coverage, and associates genomic variation information for each sample according to the phenotype data timestamp to generate high-coverage genomic sequence alignment data;

[0071] First, the genomic sequencing data of multiple samples is obtained and standardized pretreated. The sequence data of each sample is filtered for quality, removing low-quality base regions to ensure data reliability. In the sample alignment process, the input genomic sequence is divided into fragments using a sliding window method, with each window size set to 100 bp. Each fragment is aligned with the reference genome in turn, and the alignment score of each fragment is recorded. The alignment score is calculated based on the weighted processing of matching base number and mismatching conditions, where the matching base score is set to +2, the mismatching score is set to -3, and the indel penalty is set to -5. For example, if a 100 bp fragment has 80 matching bases, 15 mismatching bases, and 5 indels, the alignment score is calculated as follows:

[0072] S = (80 x 2) + (15 x (-3)) + (5 x (-5)) = 160 - 45 - 25 = 90;

[0073] After all the alignment results are summarized, the alignment consistency score is calculated based on the differences in alignment scores between different samples, and the overall alignment level between samples is calculated using the mean. For each sample's genomic alignment sequence, extract the variant sites, including single nucleotide variants (SNPs) and insertions and deletions (Indels). For each variant site, record its base substitution information (e.g., A→T) and count the frequency of the variant at that site in all samples. Variant site screening is based on coverage, where coverage is defined as:

[0074]

[0075] When the coverage is higher than 0.8, the site is considered a high-confidence site. For example, if a site has a sequencing depth of 200X and 160 sequences contain the mutation, the coverage is calculated as follows:

[0076]

[0077] The results show that the mutation site has a high alignment coverage in the sequencing sample, so it has a high credibility and can be used for subsequent gene variant functional region detection and genotype-phenotype association analysis.

[0078] The gene variant functional region detection submodule detects gene variant site coordinates based on high-coverage gene sequence alignment data, analyzes the overlap between variant sites and coding regions of the genome sequence, determines whether the variant site falls within a functional gene region, calculates the overlap ratio of the variant site, and selects high-overlap-ratio mutation sites to generate a functional region variant site set.

[0079] For each mutation site, query whether the coordinate is located in a known gene functional region. Compare the coordinate position of each mutation site with the genome annotation database to determine whether the variant falls within a coding region, promoter region, or regulatory region. For mutations that fall within a coding region, further calculate the amino acid substitution caused by the mutation. For example, if the variant site chr1:12345 (G→A) corresponds to the codon GAG encoding glutamic acid (E), and the mutation changes it to AAG encoding lysine (K), the mutation causes the amino acid substitution E→K. Record the change and calculate the proportion of variant sites that affect protein function. The functional impact of the variant site is calculated based on the overlap ratio:

[0080]

[0081] If 80 out of 100 variant sites are located in the functional region, the calculation is as follows:

[0082]

[0083] The results indicate that most of the variant sites in this gene alignment data significantly overlap with known gene functional regions, suggesting that these variants may directly affect gene function. Therefore, they can be further used for genotype-phenotype association calculations to determine whether they are related to clinical phenotypes.

[0084] The genotype-phenotype association calculation submodule calls upon the set of functional region variant sites, extracts cross-sample base complementarity patterns, analyzes the distribution of differential genotype combinations in phenotypic data, and calculates the discrete distribution of genotype-phenotype association strength using the formula:

[0085]

[0086] The algorithm calculates the genotype-phenotype association strength, filters association patterns that conform to distribution characteristics, standardizes multimodal data, and obtains standard values ​​for multimodal data.

[0087] Where S represents the distribution value of genotype-phenotype association strength, Representing the i-th g Each genotype variable value, Representative and The corresponding phenotypic variable values, Representing the j-th g The values ​​of irrelevant variable variables, Representative and The corresponding phenotypic variable value, n g m represents the number of genotype variables. g This represents the number of irrelevant variant variables.

[0088] First, for all samples, extract the alleles for each locus and form a genotype matrix, with the values ​​of each genotype variable... It is calculated from allele frequencies. For example, if the allele frequency of a certain mutation site is 0.4 in 100 samples, then... Phenotypic variable values Obtained from normalized clinical phenotypic data, using min-max normalization:

[0089]

[0090] Let the phenotypic variable P have a minimum value of 50 and a maximum value of 200 in the dataset. If the phenotypic value of a certain sample is 100, the normalization calculation is as follows:

[0091]

[0092] Then, the genotype-phenotype association strength S was calculated for all genotype variables. and phenotypic variables Calculate using the following formula:

[0093]

[0094] Let n g = 3 genotypic variables, m g = 2 independent variant variables, with data as follows:

[0095] G1 = 0.4, P1 = 0.33, G2 = 0.5, P2 = 0.50, G3 = 0.3, P3 = 0.20, V1 = 0.2, Q1 = 0.15, V2 = 0.1, Q2 = 0.10;

[0096] Calculate the numerator part:

[0097] (0.4 x 0.33) + (0.5 x 0.50) + (0.3 x 0.20) - [(0.2 x 0.15) + (0.1 x 0.10)] = 0.132 + 0.25 + 0.06 - (0.03 + 0.01) = 0.402;

[0098] Calculate the denominator part:

[0099]

[0100] Final calculation:

[0101]

[0102] The result shows that the genotypic-phenotypic correlation strength distribution value is 0.54, which indicates that there is a certain correlation between the genotypic variables and the phenotypic variables, but the correlation strength is at a medium level. In subsequent analysis, the genotypic classification can be further optimized based on the distribution of different phenotypic data to improve the correlation between genotypes and phenotypes, thereby improving the accuracy of disease prediction or functional research.

[0103] Please refer to Figure 3 , the problem decomposition module includes:

[0104] The standard value calculation submodule analyzes the standard range of the differentiated gene feature parameters based on the standard values of the multi-modal data, calls the gene expression level, variant frequency and epigenetic regulatory factors of the target individual, calculates the deviation of each feature parameter from the standard value, and analyzes the overall variation by accumulating the deviation, using the formula:

[0105]

[0106] The operation obtains the standard value calculation deviation, and establishes a standard value deviation result;

[0107] Where D represents the standard value calculation deviation, represents the i s th gene feature parameter value, representing the corresponding multi-modal data standard value, n s representing the number of gene signature parameters, representing the individual environmental factor value, representing the environmental factor benchmark value, m s representing the number of environmental factors;

[0108] The main calculation steps are as follows:

[0109] 1. Obtain standard values and target individual data

[0110] (1) Extract standard values from the database The standard values are usually from healthy population gene expression datasets, such as TCGA, GTEx or 1,000 Genomes.

[0111] (2) Obtain the gene signature parameter values of the target individual and environmental factor values These data can be obtained by high-throughput sequencing (RNA-seq), whole genome sequencing (WGS), or epigenetic detection (such as ATAC-seq).

[0112] For example, set the partial gene data of an individual as follows:

[0113] Table 1 Gene expression values of target individual and standard values

[0114] Gene Target individual expression (TPM) Standard value (TPM) A 30 50 B 80 90 C 25 20

[0115] Table 2 Target individual environmental factor deviation:

[0116] Environmental factor Target individual value Standard value Air pollution index (pg / m3) 80 50

[0117] 2. Calculate the gene expression deviation

[0118] The expression level deviation of each gene is calculated as follows:

[0119]

[0120] That is:

[0121] Gene A:

[0122] Gene B:

[0123] Gene C:

[0124] This calculation indicates that the expression level of gene A is 40% lower than the standard value, the expression level of gene B is 11.1% lower, and the expression level of gene C is 25% higher.

[0125] 3. Calculate environmental factor bias

[0126] The calculation method of environmental factors is as follows:

[0127]

[0128] That is:

[0129] Air pollution index: |80-50| = 30

[0130] 4. Cumulative bias calculation

[0131] Square all the gene bias and add the bias of environmental factors:

[0132]

[0133] Put in the data:

[0134] D = (-0.4) 2 +(-0.111) 2 +(0.25) 2 +30;

[0135] = 0.16 + 0.0123 + 0.0625 + 30;

[0136] = 30.2348;

[0137] 5. Result analysis

[0138] The final calculation D = 30.2348, which represents the overall variation of the target individual. In order to judge whether this value is abnormal, it can be compared with the data of healthy individuals. Assuming that the mean D healthy of healthy individuals is 3.2, and the standard deviation σ = 1.5, then:

[0139] If D > D healthy + 2σ = 3.2 + 2(1.5) = 6.2, it is considered that the individual has significant genomic abnormalities.

[0140] The calculated value D = 30.2348 is far beyond 6.2, indicating that the target individual's gene expression and environmental factors deviate from the normal range.

[0141] Finally, the gene expression level and environmental factors of this individual show significant bias, and further analysis is needed to determine whether it involves specific pathogenic gene variation or epigenetic regulation abnormalities, and enter the pathogenic gene annotation submodule.

[0142] The pathogenic gene annotation submodule analyzes the functional characteristics of the pathogenic gene based on the standard value deviation result, calls a known gene database to extract gene association characteristics, analyzes the mutation site and influence range of the gene sequence, calculates the semantic weight value of the mutation function prediction logic, obtains the gene semantic weight calculation value, and establishes the pathogenic gene function annotation result.

[0143] For example, if the mRNA expression level of a certain gene B is abnormal, and the mutation site rsID234567 is annotated as "pathogenic" in the ClinVar database, further analysis of the mutation influence range of the gene is required. When analyzing the gene sequence, first locate the exon or intron region where the gene mutation occurs. For example, if a mutation occurs in the 3rd exon of gene B, and the mutation causes a non-synonymous substitution (such as alanine A replaced by valine V) in the protein amino acid sequence, the potential functional impact needs to be calculated. When calculating the mutation function prediction, multiple bioinformatics scoring indicators are used, such as PolyPhen-2 and SIFT scores. These scores are predicted by machine learning models to predict the impact of mutations on protein structure and function. For example, if the PolyPhen-2 score is 0.95 (the value range is 0-1, and the closer the value is to 1, the more harmful the mutation), the mutation may have a greater functional impact. In addition, the semantic weight value of the gene also needs to be calculated. The semantic weight value can be obtained by calculating the functional relevance of the gene in the Gene Ontology (GO) database. For example, in the GO database, if a gene is involved in multiple key biological processes, its semantic weight can be calculated by calculating its frequency in different GO categories. For example, the semantic weight value of gene B is calculated as 2.3. Finally, the calculated semantic weight value and pathogenic mutation information are integrated to establish the pathogenic gene function annotation result.

[0144] The gene structure variation classification submodule calls the pathogenic gene function annotation result, filters the structural differences corresponding to the gene variation, calculates the classification boundary value of multiple structural variations, and compares the standard value deviation result to generate the structural variation classification boundary value difference, and establishes the analysis task priority sequence.

[0145] First, extract the mutation data of the target individual, such as copy number variation (CNV), insertion and deletion variation (Indel), and chromosome translocation, and call the reference genome for alignment. For example, if the individual's gene C has a deletion at chromosome location chr12:345678-345890, compared with the reference sequence of the genome, it is confirmed that it belongs to the CNV type variation. Then, calculate the impact of the variation on gene function, and use the mutation frequency data for classification,

[0146] As in the 1,000 Genomes database, the population frequency of the CNV is 0.001, which means it is a rare variation, then calculate the classification boundary value of the multi-structure variation, the calculation of the classification boundary value depends on the statistical distribution in the genomic data, for example, in the TCGA database, if the structure variation of a gene affects the function of the 0.05, 0.5, 0.95 quantile limit, the variation sample can be divided into low impact (<0.05), moderate impact (0.05-0.95) and high impact (≥0.95), for example, the gene C variation impact score of the target individual is calculated as 0.98, which exceeds the 95% quantile, and is determined as a high impact variation, finally, compare the standard deviation result, generate the structural variation classification boundary value difference, and establish the analysis task priority sequence, for example, if the target individual carries 3 high impact variations, 2 moderate impact variations and 5 low impact variations, the function impact of high impact variation is analyzed first, and the mutation site with the highest priority is sorted out.

[0147] Table 1: Example data

[0148] Gene Target individual expression (TPM) Standard value (TPM) Deviation calculation A 30 50 -0.4 B 80 90 -0.111 C 25 20 0.25

[0149] As shown in Table 1, the expression level of gene A is low, gene B is close to the standard value, and the expression level of gene C is slightly high, combined with the deviation calculation value, the expression pattern deviation of these genes in the target individual can be further analyzed.

[0150] Assuming that the deviations of genes A, B and C of the target individual are -0.4, -0.111 and 0.25 respectively, and the environmental factor air pollution exposure index is 80, the standard value is 50, then:

[0151] D = (-0.4) 2 +(-0.111) 2 +(0.25) 2 +|80-50|;

[0152] = 0.16+0.0123+0.0625+30;

[0153] = 30.2348;

[0154] The result shows that the comprehensive variation degree of the individual is large, and there may be significant genomic abnormalities, which needs to be further analyzed for the specific pathogenic gene impact.

[0155] Please refer to Figure 4 , the analysis task scheduling module includes:

[0156] The mutation classification boundary identification submodule extracts mutation category information corresponding to multiple analysis tasks based on the analysis task priority sequence, calculates the distribution interval of mutation types, compares the benchmark value of mutation impact degree, identifies the classification boundary of mutations, calculates the deviation of multi-classification boundaries, and generates mutation classification boundary values;

[0157] First, based on the analysis task priority sequence, the mutation category information involved in the current task is extracted, including the specific type of mutation (such as point mutation, insertion and deletion mutation, copy number variation, etc.), the corresponding chromosome region number, and the affected gene sequence segment. According to the known mutation distribution characteristics recorded in the database, data cleaning is performed, and only the mutation categories within the scope of the current analysis task are retained. For mutation categories not within the scope of the task, they are excluded. Second, the distribution interval of mutation types is calculated, that is, the occurrence frequency of the same mutation category in the sample is counted, and the relative distribution proportion in the whole genome is calculated. For example, if the number of occurrences of a certain mutation category detected in a certain gene region is 1200 times, and the total number of mutations of this category in the whole genome is 18000 times, then the relative distribution proportion in this region can be calculated as:

[0158]

[0159] For each type of mutation, its impact degree is calculated, and the impact classification benchmark value in the reference database or experimental data is used, such as the mutation impact classification standard provided by SIFT, PolyPhen, etc. For example, SIFT value less than 0.05 is considered to have a greater impact, and PolyPhen score greater than 0.85 is considered to have a greater impact. By comparing the benchmark value of mutation impact degree, it is determined whether the current mutation belongs to the high impact category, and the classification boundary is set accordingly, for example, setting the SIFT value in [0, 0.05] as high impact, and the PolyPhen score in [0.85, 1.0] as high risk mutation. In addition, to quantify the deviation of multi-classification boundaries, the deviation between the mutation distribution of the current task and the benchmark value needs to be calculated, and the calculation formula is as follows:

[0160]

[0161] Where P baseline is the average distribution value of the corresponding mutation category in the reference data set. For example, if the distribution proportion of this category in the benchmark database is 0.08, then:

[0162]

[0163] The results show that the mutation category distribution in the current sample has a deviation of 16.6% compared to the benchmark data, indicating that the mutation characteristics of the current sample have certain individual differences. If the deviation value exceeds the set threshold (such as 0.15), the variation classification boundary needs to be adjusted to ensure that the classification standard used in subsequent analysis tasks can accurately reflect the mutation characteristics of the current sample, thereby reducing classification errors caused by benchmark deviation. Finally, the calculated classification boundary value is stored in the task scheduling module for subsequent task calls.

[0164] The mutation distribution screening submodule calls the variation classification boundary value to screen the distribution range of promoter region mutations and exon splice site mutations, calculates the distribution proportion of multiple mutation categories in the differentiated region, and classifies and sorts the distribution according to the mutation functional influence coefficient, generating a mutation distribution classification interval.

[0165] First, the classified mutation categories are obtained from the variation classification boundary values generated in the previous module, and mutations located in the promoter region and exon splice site are screened out. The screening standard is based on the genome coordinate database. For example, the promoter region is usually defined as the range of 2000bp upstream of the transcription start site, while the exon splice site mutation mainly focuses on the splice donor and splice acceptor mutations within ±2bp range. Therefore, mutations that meet this range are extracted, and their distribution proportion in different mutation categories is calculated. For example, in the promoter region of a certain gene, mutation category A appears 300 times, and mutation category B appears 500 times. In the exon splice site, mutation category A appears 100 times, and category B appears 400 times. The distribution ratio in the differentiated region is calculated as follows:

[0166]

[0167] The results show that the mutation proportion of category B in the promoter region and splice site is higher, which may have a more significant regulatory effect. Therefore, in subsequent task scheduling, the mutation of category B should be given priority for detailed functional analysis.

[0168] The task execution order adjustment submodule calls the mutation distribution classification interval, adjusts the analysis task execution order according to the task priority, and uses the formula:

[0169]

[0170] The task execution optimization coefficient is calculated;

[0171] Where T opt represents the task execution optimization coefficient, represents the adjusted task distribution proportion, represents the classification boundary value before adjustment, represents the priority weight of the task, mx represent the number of analysis tasks.

[0172] Firstly, the mutation distribution classification interval is called to obtain the distribution order of different mutation categories in the target area, and the tasks are adjusted in combination with the analysis task priority. The priority weight According to the importance of the mutation category, for example, for high-impact mutation categories, a higher priority is set, such as U=5, and for medium-impact categories, U=3. The optimization coefficient T is calculated opt The following formula is used:

[0173]

[0174] Assume that the classification boundary value before adjustment The adjusted task distribution ratio Then calculate the deviation:

[0175]

[0176] Assume that the current number of analysis tasks m x =3, and the sum of the priority weights is:

[0177]

[0178] Substitute the calculation:

[0179] T opt =0.166xlog(12)=0.166x1.079=0.179;

[0180] The results show that the degree of adjustment of the task execution order calculated based on the optimization coefficient is small, indicating that the priority order of the current task is reasonable. If the optimization coefficient exceeds a certain set threshold (such as 0.3), it indicates that the task execution order needs to be further adjusted to improve the overall execution efficiency of mutation screening and analysis tasks.

[0181] Please refer to Figure 5 , the result mapping module includes:

[0182] The task optimization coefficient calculation submodule extracts the mutation site distribution parameter, protein interaction network connectivity parameter and gene expression level parameter based on the task execution optimization coefficient, and uses the formula:

[0183]

[0184] The optimization adjustment coefficient is obtained by operation;

[0185] Where C opt represents the optimization adjustment coefficient, V p represents the distribution density weighting coefficient of the mutation site, The mutation site distribution density represents the protein-protein interaction network, α represents the adjustment coefficient, and L... d V represents the deviation of gene expression levels. n S represents network connectivity. c The mean shift adjustment factor for representative gene expression;

[0186] First, the distribution of mutation sites needs to be statistically analyzed. This is done by interpreting the gene sequencing results of the sample data, calculating the distribution frequency of each mutation site across the entire genome, and assigning a distribution density weighting coefficient V. p For example, if a gene mutation occurs 150 times in 1000 gene samples, its mutation frequency can be expressed as 150 / 1000 = 0.15. The distribution density weight V is then set accordingly. p =0.15×10 (set 10 as the standardization factor), i.e., V p =1.5, followed by the distribution density of mutation sites in the protein-protein interaction network. To perform the calculations, first construct a protein-protein interaction network and count the total number N protein nodes connected to the gene. p And calculate its mutation frequency in the network. Assuming that the gene involves 50 protein nodes and the mutation occurs 10 times, then... Then, the deviation of gene expression level L was measured. d The calculation, based on gene expression profile data, statistically analyzes the difference between the expression level of the target gene and its average expression level. For example, assuming the average normal expression level of a gene is 8.0, while the expression level of a mutant sample is 12.0, then L... d =|12.0-8.0|=4.0. After determining the adjustment coefficient α (for example, setting α=0.5), calculate α·L. d =0.5 × 4.0 = 2.0, next calculate the network connectivity V. n This value can be obtained by counting the number of adjacent nodes of the target gene in the protein-protein interaction network. For example, if the target gene has 20 adjacent nodes in the network, then V n =20, and finally calculate the gene expression mean shift adjustment factor S. c The calculation method is the mean of gene expression offset values. For example, in 5 different samples, the expression offset values ​​of the target gene are 3.0, 4.0, 2.5, 3.5, and 4.2, respectively, then S c = (3.0 + 4.0 + 2.5 + 3.5 + 4.2) / 5 = 3.44. Substitute the calculated value into the optimization adjustment coefficient calculation formula:

[0187]

[0188] The result indicates that the optimal adjustment coefficient C is... optThe value is negative, indicating that the distribution density of the gene mutation is low, the connectivity in the protein interaction network is high, and the gene expression level deviation is large. The calculation result of the optimization adjustment coefficient will be used to further judge whether the gene belongs to the mutation gene group with greater impact and whether it needs to be given higher weight or screening processing in the subsequent calculation process. The value can be used to determine the gene screening standard or construct the risk assessment model.

[0189] The gene expression profile clustering submodule extracts gene expression profile data based on the optimization adjustment coefficient, calculates the gene expression mean, and identifies the deviation of the mutation gene in the expression profile. Genes with expression deviation values exceeding the set threshold are screened, and the expression profile clustering center coordinates of the mutation gene are calculated to obtain the gene expression clustering center coordinates.

[0190] First, the gene expression profile data matrix is obtained, which records the expression values of different genes under different experimental conditions. For example, the expression values of the target gene in 100 samples are extracted to form a 100x1 expression matrix. The gene expression mean is calculated. Assuming that the expression mean of a certain gene in 10 normal samples is 8.5, and the expression mean in 10 mutant samples is 12.1, the deviation of the expression level is calculated as follows:

[0191] ΔE = 12.1 - 8.5 = 3.6;

[0192] Then, gene screening is performed, and the expression deviation threshold T is set to screen genes with expression deviation values exceeding the threshold. For example, set T = 2.0, then the expression deviation of the gene is 3.6 > 2.0, which meets the screening condition. Then, the expression profile clustering center coordinates of the mutation gene are calculated. The expression data is divided into multiple clustering centers by the K-means clustering method. Assuming that the expression data clustering center coordinates of the gene are (8.4, 12.2), the expression profile clustering center coordinates of the gene are (8.4, 12.2). The result shows that the expression data of the gene shows a clear separation trend between the mutant and non-mutant states, indicating that it may play a key role in the regulatory network and can be used as a candidate biomarker for further research or disease prediction. The value can be used for subsequent topological mapping analysis of the gene network.

[0193] The gene network three-dimensional topological mapping submodule calculates the topological parameters of the mutation gene enrichment region in the protein interaction network based on the gene expression clustering center coordinates, maps its three-dimensional space distribution, and generates a correlation resolution set based on the connection of multiple genes in the topological network.

[0194] First, the protein interaction network data is obtained, and the connection number of the target gene with other genes is counted. Assuming that the connection number of the target gene is 15, the local clustering coefficient is calculated as follows:

[0195]

[0196] Assuming that the adjacent genes of the target gene have 10 genes, and the number of edges formed by mutual connection is 25, then

[0197]

[0198] Then the betweenness centrality B of the gene in the network is calculated by counting the frequency of being an intermediate node in the shortest path. Assuming that the betweenness centrality of the target gene in the network shortest path is 0.35, the three-dimensional topological parameters of the target gene can be represented as (15, 0.56, 0.35). Subsequently, the connection of multiple genes is analyzed to generate a set of correlation resolution, for example, the connection characteristics of multiple mutant genes are counted to form a data set

[0199] {(15, 0.56, 0.35), (20, 0.63, 0.42), (18, 0.59, 0.39)}, and finally the gene correlation resolution data of the topological network is obtained. The results show that the target gene has a relatively tight topological structure in the protein-protein interaction network, with a low betweenness centrality but a high local clustering degree, which means that the gene is more likely to be related to local regulation rather than global information transmission. Further analysis can be combined with the three-dimensional topological parameters of other genes to construct a gene regulation network and explore its role in a specific biological process. This data can be used for subsequent bioinformatics analysis to optimize the screening strategy of mutant genes.

[0200] Please refer to Figure 6 , the feedback iteration module includes:

[0201] The structural variation hotspot coverage sub-module calculates the coverage of multiple hotspots in the differential sequence coding region based on the structural variation hotspot coverage in the correlation resolution set, calls the determination threshold to filter the hotspot regions with coverage meeting the requirements, and uses the formula:

[0202]

[0203] to obtain the hotspot coverage metric value;

[0204] wherein, C hot represents the hotspot coverage metric value, represents the coverage probability of the i t th structural variation hotspot, represents the correlation resolution score of the i t th structural variation hotspot, δ s represents the adjustment factor, V max represents the maximum correlation resolution score, and N s represents the number of hotspots;

[0205] First, the corresponding coverage probability value is extracted The value represents the distribution probability of the hotspot in the current sequence coding region, and the correlation interpretation score is extracted The score is given according to historical data or experimental results, and represents the significance of the hotspot in gene variation influence or functional annotation. In order to ensure that the coverage calculation can effectively measure the importance of the hotspot region, all N s hotspots are called to calculate the coverage probability and interpretation score To obtain the cumulative weighted coverage value of the hotspot, after completing the cumulative value calculation, divide it by the total number of hotspots N s , obtain the normalized hotspot coverage benchmark value, which is used to adjust the coverage measurement range in subsequent calculation, further call adjustment factor δ s for bias correction, the adjustment factor can be determined by historical data, for example, by analyzing multiple sets of sequence hotspot distribution, taking the standard deviation of the coverage rate as δ s value, the corrected result is multiplied by the maximum interpretation score V max , which is determined by the maximum value of all hotspots , and finally divided by the highest coverage probability value of all hotspots The value is used to normalize the coverage measurement value C hot , ensure that the measurement result is applicable to hotspots of different sizes, and finally obtain the coverage measurement value of the structural variation hotspot, and screen the hotspot region with coverage rate higher than the determination threshold according to the value.

[0206] Formula parameter assignment and calculation:

[0207] Suppose there are 4 structural variation hotspots, their coverage probability and interpretation score are as follows:

[0208]

[0209] Calculate

[0210] Let δ s = 10 (average adjustment factor calculated based on multiple data sets), then:

[0211]

[0212] Let V max = 75, the final calculation is:

[0213]

[0214] The result shows that the overall metric value of the current structural variation hotspot coverage is 210.94, which can be used to screen hot spot regions with higher coverage.

[0215] The function annotation correlation matrix matching submodule calls the hotspot coverage metric value, calculates the matching degree of the multi-coding region based on the function annotation correlation matrix, extracts the matching score of the multi-region, screens the regions with a matching score higher than the set threshold, eliminates the coding regions with low matching degree, and obtains the function matching coefficient;

[0216] First, the corresponding function annotation score is obtained from the function annotation matrix. The function annotation score is derived from database annotation information or experimental measurement data. The function score of each coding region represents its importance in structural variation impact or gene expression. Then, the matching score of the coding region with all structural variation hotspots is calculated. The matching score is calculated by iterating all hotspot regions, extracting the related score in the function annotation matrix, and calculating the cosine similarity or correlation score between the coding region and the hotspot region. For example, assuming that a coding region has 3 hotspot associations, the matching score is calculated as follows:

[0217] Assuming that the coding region function score is 60, 75, and 80, and the corresponding hotspot coverage score is 130, 150, and 180, the matching score is calculated as follows:

[0218]

[0219] Calculate the numerator:

[0220] (60x130) + (75x150) + (80x180) = 7800 + 11250 + 14400 = 33450;

[0221] Calculate the denominator:

[0222]

[0223] Final matching score:

[0224]

[0225] The result indicates that the coding region has a high degree of matching with the hotspot region. If the matching score exceeds the preset threshold (e.g., 0.95), the coding region is retained, otherwise it is eliminated.

[0226] The cross-modal data anomaly elimination submodule calls the function matching coefficient, adjusts the anomaly elimination rules of the cross-modal data, calculates the deviation coefficient of each abnormal sample, screens the abnormal samples according to the deviation coefficient threshold, eliminates the data with a deviation beyond the standard range, and obtains the process optimization parameter set.

[0227] First, the deviation coefficient of each abnormal sample is calculated, and the calculation of the deviation coefficient adopts the standardization method of sample deviation from the matching mean value, for example, for a sample, the matching scores are 0.85, 0.88, 0.92, and the matching mean value is calculated as:

[0228]

[0229] The deviation is calculated:

[0230]

[0231] Suppose the deviation threshold is set to 0.05, then since 0.0287 is less than the threshold, the sample is retained, otherwise it is rejected.

[0232] Finally, the data with deviation exceeding the standard range is removed, and the process optimization parameter set is obtained to optimize the final result.

[0233] The above is only a preferred embodiment of the present application, and is not intended to limit the present application in other forms. Any skilled person in the art can use the disclosed technical content to make changes or modifications to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made to the above embodiments without departing from the technical solution content of the present application, according to the technical essence of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. A bioinformatics analysis system based on large model technology, characterized in that, The system includes: The data parsing and calibration module acquires genome sequence alignment information and clinical phenotype data timestamps, detects the overlap range between gene variant site coordinates and sequence coding regions, extracts cross-sample base complementary pairing patterns, performs anomaly identification and removal operations on protein interaction network nodes, calculates the discrete distribution of genotype-phenotype association strength, and generates multimodal data standard values. Based on the standard values ​​of the multimodal data, the problem decomposition module calls the semantic weight of pathogenic gene annotation and the logical dependency of mutation function prediction, divides the numerical differences of structural variation classification boundaries, and generates a priority sequence of analysis tasks. The analysis task orchestration module calls the variant classification boundary in the analysis task priority sequence, filters the distribution range of promoter region mutations and exon splicing site mutations, adjusts the analysis execution order, and generates task execution optimization coefficients. The result mapping module extracts the coordinates of the gene expression profile cluster centers based on the task execution optimization coefficients, maps the mutation site enrichment region and protein interaction network topology parameters into a three-dimensional spatial distribution, and generates a set of associated resolutions.

2. The bioinformatics analysis system based on large model technology according to claim 1, characterized in that, The multimodal data standard values ​​include standard values ​​for genome sequence alignment information, clinical phenotype data, gene variant site coordinates, sequence coding region overlap range, cross-sample base pairing pattern, protein interaction network anomaly markers, and the discrete distribution of genotype-phenotype association strength. The analysis task priority sequence includes semantic weights for pathogenic gene annotations, logical dependencies for mutation function prediction, and numerical differences in structural variant classification boundaries. The task execution optimization coefficients include variant classification boundaries, promoter region mutation distribution range, exon splicing site mutation distribution range, and analysis execution order adjustment parameters. The association resolution set includes the coordinates of gene expression profile cluster centers, mutation site enrichment regions, and the three-dimensional spatial distribution of protein interaction network topology parameters.

3. The bioinformatics analysis system based on large model technology according to claim 2, characterized in that, The data parsing and calibration module includes: The genome sequence alignment submodule acquires genome sequence alignment information and clinical phenotype data timestamps, parses genome sequence alignment data, extracts alignment results between samples, calculates alignment scores, obtains mutation sites in the genome alignment sequence, filters regions with high alignment coverage, and associates the genome variation information of each sample with phenotype data timestamps to generate high-coverage gene sequence alignment data. The gene variation functional region detection submodule detects the coordinates of gene variation sites based on the high-coverage gene sequence alignment data, analyzes the overlap between variation sites and genomic sequence coding regions, determines whether the variation site falls into a functional gene region, calculates the overlap ratio of variation sites, filters mutation sites with high overlap ratios, and generates a set of functional region variation sites. The genotype-phenotype association calculation submodule calls the set of variant sites in the functional regions, extracts cross-sample base complementarity pairing patterns, analyzes the distribution of differential genotype combinations in phenotypic data, and calculates the discrete distribution of the genotype-phenotype association strength using the formula: The algorithm calculates the genotype-phenotype association strength, filters association patterns that conform to distribution characteristics, standardizes multimodal data, and obtains standard values ​​for multimodal data. Where S represents the distribution value of genotype-phenotype association strength, Representing the i-th g Each genotype variable value, Representative and The corresponding phenotypic variable values, Representing the j-th g The values ​​of irrelevant variable variables, Representative and The corresponding phenotypic variable value, n g m represents the number of genotype variables. g This represents the number of irrelevant variant variables.

4. The bioinformatics analysis system based on large model technology according to claim 3, characterized in that, The problem decomposition module includes: The standard value calculation submodule, based on the standard values ​​of the multimodal data, analyzes the standard range of differential gene characteristic parameters, calls the gene expression level, variation frequency, and epigenetic regulatory factors of the target individual, calculates the deviation of each characteristic parameter from the standard value, and analyzes the overall variation by accumulating the deviation, using the formula: The standard value is obtained through calculation, the deviation is calculated, and the standard value deviation result is established. Where D represents the deviation in the calculation of the standard value. Representing the i-th s Individual gene characteristic parameter values, n represents the corresponding standard value of multimodal data. s The number of representative gene characteristic parameters Represents individual environmental factor values. m represents the baseline value of environmental factors. s Represents the quantity of environmental factors; Based on the standard value deviation results, the pathogenic gene annotation submodule analyzes the functional characteristics of pathogenic genes, calls the known gene database to extract gene association features, analyzes the variation sites and influence range of gene sequences, calculates the semantic weight value of the mutation function prediction logic, obtains the gene semantic weight calculation value, and establishes the pathogenic gene functional annotation results. The gene structure variation classification submodule calls the functional annotation results of the pathogenic gene, filters the structural differences corresponding to gene variations, calculates the classification boundary values ​​of multiple structural variations, compares the deviation results with the standard values, generates the numerical difference of the structural variation classification boundary, and establishes the priority sequence of analysis tasks.

5. The bioinformatics analysis system based on large model technology according to claim 4, characterized in that, The analysis task orchestration module includes: The mutation classification boundary identification submodule extracts mutation category information corresponding to multiple analysis tasks based on the analysis task priority sequence, calculates the distribution range of mutation types, compares with the baseline value of mutation impact, identifies the mutation classification boundary, calculates the deviation of multiple category boundaries, and generates mutation classification boundary values. The mutation distribution screening submodule calls the mutation classification boundary value to screen the distribution range of promoter region mutations and exon splicing site mutations, calculates the distribution ratio of multiple mutation categories in the differential region, and classifies and sorts the distribution according to the mutation functional influence coefficient to generate mutation distribution classification intervals. The task execution order adjustment submodule calls the mutation distribution classification interval and adjusts the execution order of the analysis tasks according to task priority, using the formula: The task execution optimization coefficients are calculated. Among them, T opt Represents the task execution optimization coefficient. This represents the adjusted task distribution ratio. This represents the classification boundary value before adjustment. The priority weight of the task, m x This represents the number of analysis tasks.

6. The bioinformatics analysis system based on large model technology according to claim 5, characterized in that, The result mapping module includes: The task optimization coefficient calculation submodule extracts mutation site distribution parameters, protein-protein interaction network connectivity parameters, and gene expression level parameters based on the task execution optimization coefficient, using the following formula: Calculate and obtain the optimization adjustment coefficients; Among them, C opt V represents the optimization adjustment coefficient. p The distribution density weighting coefficient representing the mutation site. The mutation site distribution density represents the protein-protein interaction network, α represents the adjustment coefficient, and L... d V represents the deviation of gene expression levels. n S represents network connectivity. c The mean shift adjustment factor for representative gene expression; The gene expression profile clustering submodule extracts gene expression profile data based on the optimized adjustment coefficient, calculates the mean gene expression, identifies the deviation of mutated genes in the expression profile, filters genes whose expression offset exceeds a set threshold, calculates the cluster center coordinates of the expression profile of mutated genes, and obtains the cluster center coordinates of gene expression. The gene network three-dimensional topology mapping submodule calculates the topological parameters of the mutated gene enrichment region in the protein interaction network based on the coordinates of the gene expression cluster center, maps its three-dimensional spatial distribution, and generates an association resolution set based on the connection of multiple genes in the topology network.

7. The bioinformatics analysis system based on large model technology according to claim 6, characterized in that, The system also includes: The feedback iteration module detects the matching degree between the structural variation hotspot coverage and the functional annotation association matrix in the association interpretation set, adjusts the sequence coding region determination threshold, updates the cross-modal data abnormal sample removal rules, and generates a set of process optimization parameters. The set of process optimization parameters includes structural variation hotspot coverage, functional annotation association matrix matching degree, sequence coding region determination threshold, and cross-modal data abnormal sample removal rules.

8. The bioinformatics analysis system based on large model technology according to claim 7, characterized in that, The feedback iteration module includes: The structural variation hotspot coverage submodule calculates the coverage of multiple hotspots in the differential sequence coding region based on the structural variation hotspot coverage rate in the associated interpretability set, and uses a judgment threshold to filter hotspot regions with coverage rates that meet the requirements, using the formula: Calculate and obtain hotspot coverage metrics; Among them, C hot Represents a hotspot coverage metric. Representing the i-th t The coverage probability of a structural variation hotspot. Representing the i-th t The explanatory power score for the association of structural variation hotspots, δ s V represents the adjustment factor. max N represents the maximum explanatory power score. s Represents the number of hot topics; The Functional Annotation Association Matrix Matching Submodule calls the hotspot coverage metric, calculates the matching degree of multiple coding regions based on the functional annotation association matrix, extracts the matching score of multiple regions, filters regions with matching scores higher than a set threshold, removes coding regions with low matching degree, and obtains the functional matching coefficient. The cross-modal data anomaly removal submodule calls the function matching coefficient, adjusts the cross-modal data anomaly removal rules, calculates the deviation coefficient of each anomaly sample, filters anomaly samples according to the deviation coefficient threshold, removes data with deviations exceeding the standard range, and obtains a set of process optimization parameters.

Images