Risk prediction system for postmenopausal female hyperuricemia / gout

Abstract

The invention discloses a risk prediction system for postmenopausal female hyperuricemia / gout, which mainly performs risk prediction and screening based on clinical phenotype data, genomics and metabonomics, and specifically comprises: a data acquisition module, which is responsible for collecting and processing gene detection data of a subject and collecting and arranging clinical data of the subject; and the data preprocessing module is responsible for ensuring the quality and consistency of the input data and providing accurate basic data for risk prediction. The invention belongs to the technical field of biological gene detection, and aims to provide a comprehensive solution based on gene detection and machine learning for risk assessment of postmenopausal female hyperuricemia, and by collecting clinical data and gene detection data of a subject, an accurate individualized risk prediction and prevention scheme is realized, so that the risk assessment of postmenopausal female hyperuricemia is realized. The accuracy and individuation degree of screening are improved, a reliable decision basis is provided for clinicians and health managers, and the method has important clinical application prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of biological gene detection, and in particular to a risk prediction system for hyperuricemia / gout in postmenopausal women. Background Art

[0002] As women age, their hormone levels change significantly after menopause, especially the decrease of estrogen, which can lead to changes in metabolic function, thereby increasing the risk of hyperuricemia and gout. Suffering from hyperuricemia and gout not only seriously impairs the quality of life, but is also associated with a series of serious cardiovascular and metabolic diseases, such as hypertension, coronary heart disease, diabetes and chronic kidney disease. Therefore, in the context of an aging population, early screening and prevention of hyperuricemia and gout in postmenopausal women is particularly important for improving women's quality of life and reducing the burden of disease.

[0003] Currently, screening methods for hyperuricemia and gout in postmenopausal women still have many limitations, such as insufficient accuracy and low personalization. These traditional screening methods often fail to accurately identify high-risk groups, resulting in some women failing to receive effective prevention and intervention at an early stage. Therefore, it is particularly urgent to develop a more accurate and personalized risk prediction system.

[0004] The rapid development of genetic testing technology has provided a new opportunity to solve this problem. Genetic testing can reveal an individual's risk of hyperuricemia and gout at the genetic level, thereby achieving more accurate early screening. Uric acid transport gene mutations are closely related to the onset of hyperuricemia, and the ABCG2 and SLC2A9 genes are most closely related to hyperuricemia in Asian populations. Previous studies have shown that single nucleotide polymorphisms (SNPs) of the SLC2A9 gene are associated with hyperuricemia in Asian populations; the results of research by Chinese scholars also showed that the blood uric acid level of carriers of the rs3733591 mutation genotype was significantly higher than that of the control group. Carrying rs3775948, rs3733589, and rs1014290 mutation genes can also reduce the risk of asymptomatic hyperuricemia. ABCG2 gene SNPs are closely related to the occurrence and development of hyperuricemia and gout, and the rs2231142 site has been confirmed by many studies to significantly increase the risk of hyperuricemia and gout.

[0005] In addition, from the perspective of metabolites, the study of serum metabolomics provides a new perspective for the prediction of hyperuricemia and gout. By analyzing the metabolite patterns in the blood, metabolomics can reveal biochemical processes closely related to the occurrence of the disease, help identify specific metabolic pathway disorders and potential metabolic biomarkers. These biomarkers help to identify high-risk individuals at an early stage, thereby achieving more effective prevention and intervention measures. In the study of hyperuricemia, amino acid metabolism disorders are particularly significant. Using nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS) techniques, researchers can analyze the molecular composition of samples, while liquid chromatography and gas chromatography are used to separate metabolites. Studies have shown that there are significant differences in the plasma free amino acid profile (PFAA) of patients with hyperuricemia, especially the levels of cysteine, glutamine, phenylalanine and threonine are affected, and these amino acids are closely related to blood uric acid levels. These findings demonstrate that certain specific metabolite changes are closely related to the development of hyperuricemia, providing an important tool for clinical prediction.

[0006] Combined with the above research foundation, the use of machine learning methods to conduct in-depth analysis and pattern recognition of large amounts of genetic data can further improve the accuracy and efficiency of screening. Machine learning algorithms can extract valuable information from huge amounts of genomic data, predict an individual's risk of hyperuricemia and gout, and provide personalized prevention and treatment recommendations. Therefore, the development of a risk prediction system for hyperuricemia / gout in postmenopausal women is of great significance for improving the quality of life and health of postmenopausal women. Summary of the invention

[0007] To this end, the present invention provides a risk prediction system for hyperuricemia / gout in postmenopausal women to solve the above-mentioned problems in the prior art.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A risk prediction system for hyperuricemia / gout in postmenopausal women, mainly based on clinical phenotypic data, genomics and metabolomics for risk prediction and screening, including:

[0010] The data acquisition module is responsible for collecting and processing the genetic test data of the subjects, and collecting and collating the clinical information of the subjects;

[0011] Data preprocessing module, responsible for ensuring the quality and consistency of input data and providing accurate basic data for risk prediction;

[0012] The feature extraction module uses the TabNet model to convert the basic data provided by the data preprocessing module into high-dimensional feature representations to obtain SNP features, metabolite measurement features, and clinical features;

[0013] The feature fusion module combines SNP features, metabolite measurement features and clinical features to form a feature vector, and then performs fusion processing on the feature vector;

[0014] The risk prediction module predicts whether an individual has the risk of hyperuricemia and gout based on the fused feature vector;

[0015] The result feedback module scores and classifies the individual's risk of hyperuricemia and gout based on the prediction results.

[0016] As a further optimized solution of the present invention, the data acquisition module includes a gene data acquisition submodule and a clinical information acquisition submodule;

[0017] The gene data acquisition submodule is responsible for collecting and processing the genetic test data of the subjects, including the blood sample collection identification and storage unit, the blood sample processing and preservation unit, and the DNA extraction unit;

[0018] The clinical information acquisition submodule is responsible for collecting and organizing the clinical data of the subjects, including but not limited to age, medical history, blood pressure and blood lipid information. The clinical information acquisition submodule is connected to the interface of the physical examination system to realize automatic data collection and synchronization.

[0019] As a further optimized solution of the present invention, the data acquisition module also includes a metabolite data acquisition submodule for sample pre-treatment, high-resolution mass spectrometry analysis, and targeted analysis of multiple reaction monitoring.

[0020] As a further optimized solution of the present invention, the data preprocessing module includes a data cleaning submodule and a data standardization submodule;

[0021] The data cleaning submodule is used to

[0022] Integrity check: ensure the integrity and accuracy of gene single nucleotide polymorphism (SNP) information and clinical data;

[0023] Consistency verification: Check data logic and biological plausibility to ensure consistency between gene expression data and clinical characteristics;

[0024] The data standardization submodule is used to

[0025] SNP data encoding: SNP data is categorical data, so One-hot encoding is performed to convert SNP categorical variable data into a numerical format that the model can effectively process;

[0026] Clinical characteristic data: The numerical data was standardized using the mean and standard deviation of the training data.

[0027] As a further optimized solution of the present invention, the feature extraction module includes

[0028] Feature transformation block: performs nonlinear transformation on input features to generate high-dimensional feature representation;

[0029] Mask generation block: assigns a weight to each input feature through the attention mechanism and dynamically selects features;

[0030] Decision step: Multiple decision steps generate and fuse feature representations, allowing the model to gradually aggregate information for accurate predictions.

[0031] As a further optimized solution of the present invention, the feature fusion module includes:

[0032] Feature splicing unit: splicing SNP features, metabolite measurement features and clinical features to form a 1×06 feature vector;

[0033] Fusion layer processing unit: The 1×06 feature vectors are fused to obtain a fully connected layer.

[0034] The present invention has the following advantages:

[0035] This system aims to provide a comprehensive solution based on genetic testing and machine learning for the risk assessment of hyperuricemia and gout in postmenopausal women. By collecting the clinical information and genetic testing data of the subjects, especially the ABCG2 gene and SLC2A9 gene, a risk assessment model for hyperuricemia and gout is constructed to achieve accurate individualized risk prediction and prevention plans. The main functions of the system include data collection, genetic testing, risk assessment model construction, model verification and optimization, and full-process management implemented by the system, which not only improves the accuracy and personalization of screening, but also provides reliable decision-making basis for clinicians and health managers, and has important clinical application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A principle block diagram of a risk prediction system for hyperuricemia / gout in postmenopausal women provided by the present invention;

[0037] Figure 2 This is a structural framework diagram of the TabNet model in the risk prediction system for hyperuricemia / gout in postmenopausal women provided by the present invention. DETAILED DESCRIPTION

[0038] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] like Figure 1 As shown in the figure, a risk prediction system for hyperuricemia / gout in postmenopausal women is mainly based on clinical phenotypic data, genomics and metabolomics for risk prediction and screening, including:

[0040] The data acquisition module is responsible for collecting and processing the genetic test data of the subjects, and collecting and collating the clinical information of the subjects;

[0041] Data preprocessing module, responsible for ensuring the quality and consistency of input data and providing accurate basic data for risk prediction;

[0042] The feature extraction module uses the TabNet model to convert the basic data provided by the data preprocessing module into high-dimensional feature representations to obtain SNP features, metabolite measurement features, and clinical features;

[0043] TabNet is a deep learning model based on the attention mechanism, which is designed for processing tabular data. It can dynamically select features, process high-dimensional sparse data, and has powerful feature representation capabilities. The structural framework of the TabNet model is as follows: Figure 2 As shown;

[0044] The feature fusion module combines SNP features, metabolite measurement features and clinical features to form a feature vector, and then performs fusion processing on the feature vector;

[0045] The risk prediction module predicts whether an individual has a risk of hyperuricemia based on the fused feature vector;

[0046] The result feedback module scores and classifies the individual's risk of hyperuricemia and gout based on the prediction results.

[0047] In this embodiment:

[0048] The data acquisition module includes a gene data acquisition submodule and a clinical information acquisition submodule;

[0049] Specifically, the gene data acquisition submodule is one of the core components of this system, which is mainly responsible for collecting and processing the genetic test data of the subjects, including the blood sample collection identification and storage unit, the blood sample processing and preservation unit, and the DNA extraction unit;

[0050] The specific steps are as follows:

[0051] (1) Blood sample collection, labeling and storage:

[0052] Collection method: Subjects go to designated medical institutions or laboratories for venous blood sampling by professional medical staff. 5 ml of venous blood is collected to ensure sufficient samples for subsequent DNA extraction and analysis. Purple-capped blood collection tubes containing EDTA anticoagulant are used to prevent blood coagulation and ensure sample quality.

[0053] Sample identification: Each blood sample is identified with a unique identification code (barcode) to ensure the accuracy and traceability of the sample throughout the entire processing and analysis process. The identification code label contains the subject's basic information (such as subject number, collection date and time, etc.) and is linked to the electronic records in the system;

[0054] After the collection is completed, the identification code and sample information are entered into the sample management module of the system to ensure the digital storage and management of all sample information;

[0055] (2) Blood sample processing and storage: The collected blood samples are preliminarily processed to ensure the quality and integrity of the samples and stored under appropriate conditions for further testing and analysis. The specific steps include:

[0056] (a) Blood centrifugation: The collected blood sample was placed in a centrifuge and centrifuged at 3000 rpm for 10 min;

[0057] (b) Separation of plasma and cells: After centrifugation, the blood sample is separated into an upper plasma layer and a lower cell pellet. The upper plasma layer is removed and the lower cell pellet is retained.

[0058] (3) DNA extraction: Use the DNA extraction kit of TIANGEN to extract genomic DNA from the leukocyte precipitate according to the operating manual. The specific steps are as follows:

[0059] (a) Lysing cells: Add lysis buffer and mix the cell pellet thoroughly to lyse the cells;

[0060] (b) Protein precipitation: Add proteinase K and appropriate buffer to perform protein precipitation reaction;

[0061] (c) DNA purification: DNA is purified by adsorption column or magnetic bead method to remove impurities and proteins, and finally obtain high-purity genomic DNA. The extracted DNA samples are stored at -20°C or lower to prevent DNA degradation;

[0062] (4) Genetic testing: Gene chip technology is used to test samples, especially focusing on the mutation of ABCG2 gene and SLC2A9 gene. The specific steps include:

[0063] (a) DNA quantification and quality control: Use a spectrophotometer (such as NanoDrop) and agarose gel electrophoresis to quantify and control the quality of the extracted DNA to ensure that the DNA sample is suitable for subsequent testing;

[0064] (b) Gene chip detection:

[0065] The Illumina BeadChip platform was used to perform genome-wide scanning of DNA samples, focusing on the variation of specific genes (ABCG2 and SLC2A9). The quantified DNA samples were labeled and hybridized according to the platform requirements, followed by chip scanning and data acquisition to obtain detailed sequence information of the target genes.

[0066] (c) Gene sequencing data entry and storage: The test results are converted into electronic data format and entered into the system database to ensure the accuracy and completeness of the data. The specific steps include: converting the original gene chip data (such as CEL files) into a usable genotype data format (VCF files), and performing data cleaning and quality control. The sorted gene data is entered into the system's gene database through the data import interface to ensure the data is accurate. The gene data management platform (BaseSpace Sequence Hub) is used to store, manage and back up the gene data to ensure data security and traceability.

[0067] Specifically, the clinical information acquisition submodule is responsible for collecting and organizing the clinical data of the subjects, including but not limited to age, medical history, blood pressure and blood lipid information. The clinical information acquisition submodule is connected to the interface of the physical examination system to realize automatic data collection and synchronization;

[0068] Furthermore, routine information of participants was obtained from participating medical institutions through the physical examination system, including age, systolic blood pressure, diastolic blood pressure, body mass index, waist circumference, white blood cell count, hemoglobin, blood glucose, total cholesterol, aspartate aminotransferase (GOT), alanine aminotransferase (GPT), γ-glutamyl transpeptidase (GGT), creatinine, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, urine albumin positivity, smoking status, drinking status, diabetes, dyslipidemia, a total of 22 items.

[0069] Specifically, the data acquisition module also includes a metabolite data acquisition submodule for sample pre-treatment, high-resolution mass spectrometry analysis, and targeted analysis of multiple reaction monitoring;

[0070] (a) Sample pretreatment: serum samples stored at −80°C were thawed on ice, 50 μL of each sample was placed in a 96-well plate, and 150 μL of cold methanol was added. After vortexing for 30 seconds, the 96-well plate was placed in a −20°C freezer for 1 hour to precipitate proteins. The supernatant was transferred to an LC-MS vial and stored at −80°C for subsequent LC-MS analysis. Quality control samples (QC) were prepared by mixing a small aliquot (10 μL) of each sample.

[0071] (b) UPLC-QTOF / MS analysis: The samples were analyzed in positive (+ESI) and negative (-ESI) modes using a Nexera UHPLC LC-30A system from SHIMADZU Technologies, Japan, and an AB 6600TripleTOF quadruple time-of-flight (QTOF) mass spectrometer from SCIEX, Canada. The samples (2 μL) were eluted at a flow rate of 0.5 mL / min at 25°C on a Waters ACQUITY UPLC BEH Amide column (100 mm × 2.1 mm, 1.7 μm particle size). The mobile phase A was ultrapure water containing 25 mM ammonium hydroxide (NH4OH) and 25 mM ammonium acetate (NH4OAc). The mobile phase B was pure ACN. The samples were eluted according to the following gradient: 0-0.5 min, 95% B; 0.5-7 min, 95% to 65% B; 7-8 min, 65% to 40% B; 8-9 minutes, hold 40% B; 9-9.1 minutes, 40% to 95% B; 9.1-12 minutes, 95% B;

[0072] (c) Multiple reaction monitoring (MRM) mode was used for targeted analysis, with specific settings including specific precursor and product ion pairs, mass spectra acquired at a rate of 4 spectra per second, and the TOF mass spectrum range set to m / z 60–1200 Da;

[0073] The source parameters were set as follows: GAS1 and GAS2 were both 60; CUR was 30; TEM was 650°C; ISVF was 5000V (positive) and -4000V (negative), and the product ion scan parameters were set as m / z 25-1200, collision energy was 30eV, and accumulation time was 30ms.

[0074] In this embodiment,

[0075] The data preprocessing module includes a data cleaning submodule and a data standardization submodule;

[0076] The data cleaning submodule is used to

[0077] Integrity check: ensure the integrity and accuracy of gene single nucleotide polymorphism (SNP) information and clinical data;

[0078] Consistency verification: Check data logic and biological plausibility to ensure consistency between gene expression data and clinical characteristics;

[0079] Handling missing values: For missing data, use the median of the training data set to fill it;

[0080] The data standardization submodule is used to

[0081] SNP data encoding: SNP data is categorical data, so One-hot encoding is performed to convert SNP categorical variable data into a numerical format that the model can effectively process;

[0082] For example: AA, AG, GG, one-hot encoding creates a new binary feature for each possible genotype:

[0083] AA: [1, 0, 0]

[0084] AG: [0, 1, 0]

[0085] GG: [0, 0, 1]

[0086] Clinical characteristic data: Standardize numerical data such as age, blood pressure, and blood lipids using the mean and standard deviation of the training data;

[0087] For binary clinical variables such as diabetes, 0 and 1 are used to represent them. For example, if a subject has diabetes, the variable can be coded as 1, and if not, it can be coded as 0. For multi-categorical clinical variables, one-hot coding is used, such as smoking status: never, occasionally, often, very often;

[0088] In this embodiment, the feature extraction module includes

[0089] Feature transformation block: performs nonlinear transformation on input features to generate high-dimensional feature representation;

[0090] Mask generation block: assigns a weight to each input feature through the attention mechanism and dynamically selects features;

[0091] Decision step: Multiple decision steps generate and fuse feature representations, allowing the model to gradually aggregate information for accurate predictions;

[0092] The following three TabNet models were used to process SNP data and clinical data respectively;

[0093] (I) TabNet model for SNP data: This model converts SNP data into 1×20 feature representation. The specific steps are as follows:

[0094] (1) Feature Transformation Block

[0095] Function: Perform nonlinear transformation on input features to generate high-dimensional feature representation;

[0096] Composition: Multiple fully connected layers and BatchNorm layers (BatchNormalization);

[0097] Specific parameters:

[0098] Input dimension (input_dim): 30 (the number of SNP features input);

[0099] Feature dimension (n_d): 8 (feature dimension in each decision step);

[0100] BatchNorm layer: The BatchNorm layer is used after the fully connected layer to ensure the stability and effectiveness of the features;

[0101] (2) Mask Generating Block

[0102] Function: Assign weights to each input feature through the attention mechanism and dynamically select features;

[0103] Composition: Sparse Attention Mechanism;

[0104] Specific parameters:

[0105] Attention dimensions (n_a): 8 (the number of attention dimensions in each decision step);

[0106] Sparse regularization coefficient (sparsity_coef): set to 0.0001 to control the sparsity of the attention mechanism;

[0107] (3) Decision Steps

[0108] Function: The TabNet model consists of multiple decision steps, each of which generates a feature representation and fuses it with the previous decision step;

[0109] Composition: Each decision step consists of a feature transformation block and a mask generation block;

[0110] Specific parameters:

[0111] Number of decision steps (n_steps): 3 (the number of decision steps);

[0112] γ parameter (gamma): 1.3 (controls the adjustment strength of each decision step);

[0113] (4) Regularization

[0114] Function: To prevent overfitting, the TabNet model introduces sparse regularization and masking mechanisms to enhance the generalization ability of the model;

[0115] Composition: Sparse Regularization and Masking Mechanism;

[0116] Specific parameters:

[0117] Sparsity regularization coefficient (sparsity_coef): 0.0001;

[0118] Masking probability (masking_prob): 0.1 (the probability of randomly masking some input features during training);

[0119] The final SNP data TabNet model converts the SNP data from the original 30 features into a 1×20 high-dimensional feature representation through a series of nonlinear transformations, attention mechanisms, and decision steps;

[0120] (II) TabNet model for metabolite determination data: This model converts metabolic data into a 1×20 feature representation. The specific steps are as follows:

[0121] (1) Feature Transformation Block

[0122] Function: Perform nonlinear transformation on the input metabolite features to generate richer high-dimensional feature representations.

[0123] Composition: It consists of multiple fully connected layers and BatchNorm layers.

[0124] Specific parameters:

[0125] Input dimension (input_dim): 100 (number of input metabolite features).

[0126] Feature Dimension (n_d): 16 (the number of feature dimensions in each decision step), considering the increase in the number of features, the feature dimension is increased to better capture information.

[0127] BatchNorm layer: used after each fully connected layer to improve model stability and effectiveness.

[0128] (2) Mask Generating Block

[0129] Function: Assign weights to each input feature through the attention mechanism and dynamically select important metabolite features.

[0130] Composition: Adopt sparse attention mechanism.

[0131] Specific parameters:

[0132] Attention dimension (n_a): 16, the same as n_d to keep the model consistent.

[0133] Sparse regularization coefficient (sparsity_coef): 0.0001, controls the sparsity of the attention mechanism.

[0134] (3) Decision Steps

[0135] Function: It consists of multiple decision steps, each of which further refines the feature representation based on the previous one.

[0136] Composition: Each decision step consists of a feature transformation block and a mask generation block.

[0137] Specific parameters:

[0138] Number of decision steps (n_steps): 5. Increase the number of decision steps to accommodate more input features and complexity.

[0139] γ parameter (gamma): 1.5, appropriately increased to strengthen the adjustment strength of each decision step.

[0140] (4) Regularization

[0141] Function: Enhance the generalization ability of the model and prevent overfitting caused by the increase in the number of features;

[0142] Composition: includes sparse regularization and masking mechanism;

[0143] Specific parameters:

[0144] Sparsity regularization coefficient (sparsity_coef): 0.0001;

[0145] Masking probability (masking_prob): 0.15. Considering the increase in the number of features, the masking probability is appropriately increased to increase the randomness of training;

[0146] (III) Clinical Data TabNet Model: The clinical data TabNet model is specifically used to process clinical disease history information. The model converts clinical data into a 1×20 feature representation. The specific steps are as follows:

[0147] (1) Feature Transformation Block

[0148] Function: Perform nonlinear transformation on input features to generate high-dimensional feature representation;

[0149] Composition: Multiple fully connected layers and BatchNorm layers (BatchNormalization);

[0150] Specific parameters:

[0151] Input dimension (input_dim): 22 (the number of clinical features input);

[0152] Feature dimension (n_d): 8 (feature dimension in each decision step);

[0153] BatchNorm layer: The BatchNorm layer is used after the fully connected layer to ensure the stability and effectiveness of the features;

[0154] (2) Mask Generating Block

[0155] Function: Assign weights to each input feature through the attention mechanism and dynamically select features;

[0156] Composition: Sparse Attention Mechanism;

[0157] Specific parameters:

[0158] Attention dimensions (n_a): 4 (the number of attention dimensions in each decision step);

[0159] Sparse regularization coefficient (sparsity_coef): set to 0.0001 to control the sparsity of the attention mechanism;

[0160] (3) Decision Steps

[0161] Function: The TabNet model consists of multiple decision steps, each of which generates a feature representation and fuses it with the previous decision step;

[0162] Composition: Each decision step consists of a feature transformation block and a mask generation block;

[0163] Specific parameters:

[0164] Number of decision steps (n_steps): 3 (the number of decision steps);

[0165] γ parameter (gamma): 1.1 (controls the adjustment strength of each decision step);

[0166] (4) Regularization

[0167] Function: In order to prevent overfitting during the training phase, the TabNet model introduces sparse regularization and masking mechanisms to enhance the generalization ability of the model;

[0168] Composition: Sparse Regularization and Masking Mechanism;

[0169] Specific parameters:

[0170] Sparsity regularization coefficient (sparsity_coef): 0.001;

[0171] Masking probability (masking_prob): 0.1 (the probability of randomly masking some input features during training);

[0172] The final clinical data TabNet model converts the SNP data from the original 15 features to a 1×20 high-dimensional feature representation through a series of nonlinear transformations, attention mechanisms, and decision steps.

[0173] In this embodiment,

[0174] The feature fusion module includes

[0175] Feature splicing unit: splicing SNP features, metabolite measurement features and clinical features to form a 1×06 feature vector;

[0176] Input: SNP features, metabolite assay features, and clinical features after TabNet processing;

[0177] SNP features: 1×20 dimensions;

[0178] Metabolite features: 1×20 dimensions;

[0179] Clinical features: 1×20 dimension;

[0180] Operation: Concatenate the above three types of features together to form a 1×06 feature vector;

[0181] Fusion layer processing unit: fuses the 1×06 feature vectors to obtain a fully connected layer;

[0182] The concatenated high-dimensional features are further processed to extract more meaningful feature representations, which consists of a fully connected layer. Fully Connected Layer 1

[0183] Reduce the 1×60 dimension feature to 1×20 feature space;

[0184] Fully connected layer parameters: input dimension 60, output dimension 20;

[0185] Batch Normalization Layer

[0186] Normalize the output of the fully connected layer to ensure the stability and effectiveness of the features;

[0187] Activation Function

[0188] The ReLU activation function is used to introduce nonlinear transformation.

[0189] In this embodiment,

[0190] The goal of the risk prediction module is to predict whether an individual has a risk of hyperuricemia based on the fused features. Specifically, the input features are 1×20 high-dimensional feature representations generated by the aforementioned TabNet fusion module. A fully connected layer is used to map the features to the final output space, outputting a single node (for binary classification tasks), and the result is mapped to a probability value between 0 and 1 through the Sigmoid activation function.

[0191] In this embodiment,

[0192] The goal of the outcome feedback module is to score and classify an individual's risk of hyperuricemia based on the model's predicted results and provide personalized living and dietary advice.

[0193] enter

[0194] Prediction probability: the probability value of hyperuricemia output by the model (between 0 and 1);

[0195] Output

[0196] Risk scoring: converting predicted probabilities into specific risk scores;

[0197] Risk classification: Categorize individuals as high-risk or low-risk based on the score;

[0198] Lifestyle and diet advice: Personalized advice based on risk classification;

[0199] Main steps

[0200] Risk score generation

[0201] Input: model prediction probability;

[0202] Operation: Convert the predicted probability into a score between 0 and 100;

[0203] Formula: Risk score = predicted probability * 100;

[0204] Output: Risk score;

[0205] Risk Classification

[0206] Input: Risk score;

[0207] Action: Categorize individuals as high-risk or low-risk based on predefined thresholds;

[0208] High risk: risk score>=50;

[0209] Low risk: risk score <50;

[0210] Output: risk classification results;

[0211] Life and diet advice

[0212] Input: risk classification results;

[0213] Action: Provide personalized recommendations based on risk classification;

[0214] High risk:

[0215] It is recommended to maintain a healthy weight and avoid obesity;

[0216] Control your diet and reduce the intake of high-purine foods, such as animal offal, seafood, etc.;

[0217] Drink more water to promote uric acid excretion;

[0218] Limit alcohol intake, especially beer;

[0219] Do moderate exercise and maintain good living habits;

[0220] Check uric acid levels regularly and take uric acid-lowering medications as prescribed by your doctor;

[0221] Low risk:

[0222] Continue to maintain a healthy diet and lifestyle;

[0223] Pay attention to moderate exercise and maintain weight;

[0224] Eat a balanced diet with adequate intake of protein, vitamins and minerals;

[0225] Avoid excessive intake of high-purine foods and alcohol;

[0226] Have regular health checks to stay healthy.

[0227] The present invention has the following advantages:

[0228] This system aims to provide a comprehensive solution based on genetic testing and machine learning for the risk assessment of hyperuricemia and gout in postmenopausal women. By collecting the clinical information and genetic testing data of the subjects, especially the ABCG2 gene and SLC2A9 gene, a risk assessment model for hyperuricemia and gout is constructed to achieve accurate individualized risk prediction and prevention plans. The main functions of the system include data collection, genetic testing, risk assessment model construction, model verification and optimization, and full-process management implemented by the system, which not only improves the accuracy and personalization of screening, but also provides reliable decision-making basis for clinicians and health managers, and has important clinical application prospects.

Claims

1. A risk prediction system for hyperuricemia / gout in postmenopausal women, characterized in that: include: The data acquisition module is responsible for collecting and processing the genetic test data of the subjects, and collecting and collating the clinical information of the subjects; Data preprocessing module, responsible for ensuring the quality and consistency of input data and providing accurate basic data for risk prediction; The feature extraction module uses the TabNet model to convert the basic data provided by the data preprocessing module into high-dimensional feature representations to obtain SNP features, metabolite measurement features, and clinical features; The feature fusion module combines SNP features, metabolite measurement features and clinical features to form a feature vector, and then performs fusion processing on the feature vector; The risk prediction module predicts whether an individual has the risk of hyperuricemia / gout based on the fused feature vector; The result feedback module scores and classifies the individual's hyperuricemia / gout risk based on the prediction results.

2. A risk prediction system for hyperuricemia / gout in postmenopausal women according to claim 1, characterized in that: The data acquisition module includes a gene data acquisition submodule and a clinical information acquisition submodule; The gene data acquisition submodule is responsible for collecting and processing the genetic test data of the subjects, including the blood sample collection identification and storage unit, the blood sample processing and preservation unit, and the DNA extraction unit; The clinical information acquisition submodule is responsible for collecting and organizing the clinical data of the subjects, including but not limited to age, reproductive history and menstrual status, medical history, blood pressure and blood lipid information. The clinical information acquisition submodule is connected to the interface of the physical examination system to realize automatic data collection and synchronization.

3. A risk prediction system for hyperuricemia / gout in postmenopausal women according to claim 2, characterized in that: The data acquisition module also includes a metabolite data acquisition submodule, which is used for sample pre-treatment, high-resolution mass spectrometry analysis, and targeted analysis of multiple reaction monitoring.

4. A risk prediction system for hyperuricemia / gout in postmenopausal women according to claim 1, characterized in that: The data preprocessing module includes a data cleaning submodule and a data standardization submodule; The data cleaning submodule is used to Completeness check: Ensure the completeness and accuracy of SNP information and clinical data; Consistency verification: Check data logic and biological plausibility to ensure consistency between gene expression data and clinical characteristics; The data standardization submodule is used to SNP data encoding: SNP data is categorical data, so One-hot encoding is performed to convert SNP categorical variable data into a numerical format that the model can effectively process; Clinical characteristic data: The numerical data was standardized using the mean and standard deviation of the training data.

5. A risk prediction system for hyperuricemia / gout in postmenopausal women according to claim 1, characterized in that: The feature extraction module includes Feature transformation block: performs nonlinear transformation on input features to generate high-dimensional feature representation; Mask generation block: assigns a weight to each input feature through the attention mechanism and dynamically selects features; Decision step: Multiple decision steps generate and fuse feature representations, allowing the model to gradually aggregate information for accurate predictions.

6. A risk prediction system for hyperuricemia / gout in postmenopausal women according to claim 1, characterized in that: The feature fusion module includes Feature splicing unit: splicing SNP features, metabolite measurement features and clinical features to form a 1×06 feature vector; Fusion layer processing unit: The 1×06 feature vectors are fused to obtain a fully connected layer.

Images