Brain age prediction method and device based on deep learning and metabolomics data

Abstract

The present invention discloses a method and device for predicting brain age based on deep learning and metabolomics data, which relates to the field of biological data processing, including: obtaining blood samples of healthy people and people with elderly brain diseases and processing them to obtain several metabolite features for feature screening, and determining key metabolites related to brain age; constructing a brain age prediction model and training it to obtain a trained brain age prediction model, the brain age prediction model includes a metabolomics feature representation learning module and a prediction module, the metabolomics feature representation learning module adopts an improved deep residual network, and the prediction module adopts a deep neural network; obtaining a blood sample of a person to be tested and obtaining a metabolomics data matrix composed of metabolite features of corresponding key metabolites, inputting the metabolomics data matrix into the trained brain age prediction model, and predicting the brain age of the person to be tested. The present invention solves the problem that brain age cannot be evaluated comprehensively and accurately.

Description

Technical Field

[0001] The present invention relates to the field of biological data processing, and in particular to a method and device for predicting brain age based on deep learning and metabolomics data. Background Art

[0002] From a pathophysiological perspective, aging is a key risk factor for multiple diseases and a common cause of many chronic diseases, including neurological diseases such as Alzheimer's disease (AD) and Parkinson's disease (PD). Traditionally, chronological age has been used as an indicator to assess the degree of aging and health status. However, there is great heterogeneity in the aging process of individuals, and chronological age cannot accurately reflect the individual's physiological state and aging rate. The heterogeneity of individual aging processes reflects the different effects of genetic and environmental factors on the aging rate. If the biological characteristics of these differentiated aging rates can be accurately measured, it will help improve the prediction of mortality and morbidity. Studies have shown that biological age is superior to chronological age in predicting mortality and health risks, and can more comprehensively reflect the aging changes of multiple tissues and cells, especially in brain health.

[0003] In recent years, researchers have proposed many different biomarkers of aging, including: "epigenetic clock" uses the DNA methylation status of CpG sites in the genome to measure, which can be converted into an age prediction that is highly correlated with the actual age of healthy individuals; leukocyte telomere length shortens as the number of cell divisions increases, which is considered an important sign of cell aging; N-sugar spectrum: changes in glycosylation patterns are associated with aging, and Ink4a / Arf locus expression, etc. These biomarkers reflect the involvement of multiple biological systems and the complexity of the human aging process. These methods also show important value in brain health research, helping to reveal early biomarkers of neurodegenerative diseases.

[0004] At present, the methods for measuring biological age mainly focus on epigenetic clocks (such as DNA methylation levels), telomere length, and proteomics and transcriptomics data. However, these methods all have certain limitations, such as high detection costs, high technical complexity, and insufficient coverage. In addition, most of these methods only reflect biological changes in a certain aspect and cannot comprehensively assess the overall physiological state of an individual. Therefore, there is an urgent need for a new technical means to more comprehensively and accurately assess the biological age of an individual, especially the biological age of the brain. Modern omics platforms provide new opportunities for the systematic and comprehensive assessment of biological aging. Whole-genome DNA methylation, transcriptome, proteome, and metabolomics analysis methods are gradually revealing how complex organisms change at the molecular level with age, and provide important information for future health interventions and treatment plans, especially in the early detection and prevention of neurological diseases.

[0005] Metabolomics is a promising comprehensive technology for assessing biological aging. As the final product of cellular metabolism, metabolites can provide a more complete picture of biological processes and stronger phenotypic representation than other omics maps. In recent years, they have shown significant advantages in predicting biological age. Brain aging is not only closely related to the decline in cognitive function, but also closely related to the increased risk of neurological diseases such as Alzheimer's disease (AD), Parkinson's disease (PD) and stroke. By quantifying the aging process of the brain, individuals who may be at risk of cognitive decline can be identified early, so that more timely and accurate interventions can be carried out. The current defects in the assessment of brain age are as follows:

[0006] 1. Inaccurate assessment: Traditional actual age cannot accurately reflect the biological aging process of the brain, resulting in low accuracy in predicting brain health risks. Some existing biomarkers, such as telomere length and epigenetic clocks, also have certain limitations and cannot fully reflect the overall health status and aging rate of the brain.

[0007] 2. Lack of comprehensiveness: Existing biomarkers such as epigenetic clocks and telomere length can only reflect one aspect of the brain aging process and lack a comprehensive assessment of the overall brain physiological state. Different biomarkers may have complex relationships that are independent or interactive with each other, and a single marker cannot fully reflect the complex physiological and pathological processes in the brain.

[0008] 3. Complexity and high cost: The existing methods for detecting multiple biomarkers are complex and expensive, which limits their application in clinical and large-scale populations. In particular, some high-throughput omics technologies, such as epigenomics and transcriptomics, can provide rich information, but their high cost and technical barriers hinder their widespread application in the field of brain health.

[0009] 4. The prediction model is too simple: Many current brain age prediction models are based on simple linear regression. Regardless of whether they are statistical methods or machine learning methods, these models often have significant errors when processing high-dimensional brain metabolomics data. Simple linear regression models perform poorly when the relationship between features and brain age is complex. Brain omics data typically contain a large number of features, and linear regression models may not be able to capture the nonlinear relationships and interaction effects between these features. Simple linear regression is sensitive to the collinearity of features, which often leads to model instability and increased prediction errors in brain metabolomics data. Brain age prediction models that lack complex models may not provide sufficiently accurate prediction results when processing complex data sets. Summary of the invention

[0010] The purpose of this application is to propose a brain age prediction method and device based on deep learning and metabolomics data to address the above-mentioned technical problems.

[0011] In a first aspect, the present invention provides a method for predicting brain age based on deep learning and metabolomics data, comprising the following steps:

[0012] Blood samples from healthy people and people with elderly brain diseases are obtained and metabolite extraction and chromatography-mass spectrometry analysis are performed to obtain mass spectrometry data of each metabolite in several metabolomes corresponding to the healthy people and the people with elderly brain diseases, and the mass spectrometry data of each metabolite in each metabolome is cleaned and normalized to obtain several metabolite features, which are formed by splicing the chromatographic retention time and mass-to-charge ratio in the mass spectrometry data of the metabolites. Several metabolite features corresponding to the healthy people and the people with elderly brain diseases are screened to determine key metabolites related to brain age;

[0013] A brain age prediction model is constructed and trained using training data consisting of metabolite characteristics of key metabolites corresponding to healthy people to obtain a trained brain age prediction model, wherein the trained brain age prediction model includes a metabolomics feature representation learning module and a prediction module, wherein the metabolomics feature representation learning module includes a first fully connected layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fifth residual module, a sixth residual module and a multi-head self-attention layer connected in sequence, and the prediction module uses a deep neural network;

[0014] A blood sample of the person to be tested is obtained to obtain a metabolomics data matrix composed of the metabolite characteristics of the corresponding key metabolites, the metabolomics data matrix is ​​input into the metabolomics feature representation learning module in the trained brain age prediction model to obtain the metabolite feature representation, the metabolite feature representation is input into the prediction module, the brain age of the person to be tested is predicted, and the brain age acceleration is calculated.

[0015] Preferably, the first residual module, the second residual module, the third residual module, the fourth residual module, the fifth residual module and the sixth residual module all adopt a residual structure, and the residual structure includes a second fully connected layer and a third fully connected layer connected in sequence. The input features of the residual structure pass through the second fully connected layer and the third fully connected layer in sequence to obtain intermediate features, and the intermediate features are added to the input features of the residual structure to form a residual connection and obtain the output features of the residual structure.

[0016] Preferably, the number of neurons in the first fully connected layer, the second fully connected layer and the third fully connected layer are 128, and the ReLU activation function is used.

[0017] Preferably, the multi-head self-attention layer includes 4 attention heads, and the output features of the sixth residual module are subjected to three linear transformation layers to obtain a query vector, a key vector and a value vector, respectively. The query vector, the key vector and the value vector are divided into 4 query sub-vectors, 4 key sub-vectors and 4 value sub-vectors, respectively. Each query sub-vector, key sub-vector and value sub-vector performs a scaled dot product attention operation in each attention head to obtain the output vector of each attention head. The output vectors of the 4 attention heads are concatenated and then linearly transformed through a linear layer to obtain the metabolite feature representation.

[0018] Preferably, the prediction module includes an input layer, three hidden layers and an output layer connected in sequence, the input layer is used to receive the metabolite feature representation, the three hidden layers are respectively the fourth fully connected layer, the fifth fully connected layer and the sixth fully connected layer, the number of neurons in the fourth fully connected layer, the fifth fully connected layer and the sixth fully connected layer are respectively 256, 128 and 64, and all use the ReLU activation function, the output layer has 1 neuron, and uses a linear activation function.

[0019] Preferably, several metabolite features corresponding to healthy people and people with elderly brain diseases are screened to determine key metabolites related to brain age, including:

[0020] Perform differential analysis feature screening on several metabolite features, perform a two-sample t-test on the metabolite features of the healthy population and the metabolite features of the population with elderly brain diseases, obtain t values ​​and p values, use the BH method to correct the p value, obtain the corrected p value, and form a differential metabolite set with metabolites that meet the corrected p value less than 0.05 and the difference multiple greater than 2 or less than 1 / 2. The difference multiple is the multiple ratio of the median value of the metabolite feature in the population with elderly brain diseases to the median value of the metabolite feature in the healthy population for each metabolite;

[0021] A random forest model is constructed, training data is constructed based on the metabolite characteristics of healthy people and the metabolite characteristics of people with elderly brain diseases, the random forest model is trained using the training data, the metabolite characteristics in the training data are input into the random forest model, and the classification results are obtained, and the classification results are healthy people or people with elderly brain diseases. During the training process, the reduction of the Gini index caused by each metabolite characteristic at each node in each decision tree is calculated and accumulated to obtain the total reduction of the Gini index of each decision tree, the total reduction of the Gini index of all decision trees is averaged, the average reduction of the Gini index is calculated and sorted from large to small, and the metabolites corresponding to the metabolite characteristics ranked in the top 500 of the average reduction of the Gini index are selected to form an important metabolite set;

[0022] The intersection of the differential metabolite set and the important metabolite set was taken to obtain the key metabolite features related to brain age.

[0023] In a second aspect, the present invention provides a brain age prediction device based on deep learning and metabolomics data, comprising:

[0024] a feature screening module, configured to obtain blood samples from healthy people and people with elderly brain diseases and perform metabolite extraction and chromatography-mass spectrometry analysis to obtain mass spectrometry data of each metabolite in several metabolomes corresponding to the healthy people and the people with elderly brain diseases, perform data cleaning and normalization on the mass spectrometry data of each metabolite in each metabolome to obtain several metabolite features, wherein the metabolite features are formed by splicing the chromatographic retention time and mass-to-charge ratio in the mass spectrometry data of the metabolites, perform feature screening on several metabolite features corresponding to the healthy people and the people with elderly brain diseases, and determine key metabolites related to brain age;

[0025] A model building module is configured to build a brain age prediction model and train it using training data consisting of metabolite characteristics of key metabolites corresponding to healthy people to obtain a trained brain age prediction model, wherein the trained brain age prediction model includes a metabolomics feature representation learning module and a prediction module, wherein the metabolomics feature representation learning module includes a first fully connected layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fifth residual module, a sixth residual module and a multi-head self-attention layer connected in sequence, and the prediction module uses a deep neural network;

[0026] The inference module is configured to obtain a blood sample from a person to be tested and obtain a metabolomics data matrix consisting of metabolite features of corresponding key metabolites, input the metabolomics data matrix into a metabolomics feature representation learning module in a trained brain age prediction model to obtain a metabolite feature representation, input the metabolite feature representation into a prediction module, predict the brain age of the person to be tested, and calculate the brain age acceleration.

[0027] In a third aspect, the present invention provides an electronic device comprising one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0028] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0029] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] (1) The brain age prediction method based on deep learning and metabolomics data proposed in this invention has comprehensive detection and high sensitivity. Using non-targeted metabolomics technology, it can simultaneously detect and analyze thousands of metabolites, including small molecules, intermediate metabolites and complex biochemical pathways. Non-targeted metabolomics has extremely high sensitivity and accuracy and can detect low-abundance metabolites. This feature enables the brain age prediction model to capture a variety of subtle but significant physiological and pathological changes associated with brain age, providing a more comprehensive brain age assessment.

[0032] (2) The brain age prediction method based on deep learning and metabolomics data proposed in this invention can provide multi-dimensional biological information. Metabolomics data directly reflects the biochemical activities and material exchanges within and between cells, and can therefore more realistically reflect the physiological state and metabolic health of the body. In contrast, genomic or transcriptomic data can only indirectly reflect the underlying biological processes.

[0033] (3) The brain age prediction method based on deep learning and metabolomics data proposed in the present invention is more convenient for dynamic monitoring, reduces deviations, and makes accurate predictions. Due to differences in individual genes, environment, diet, lifestyle and other factors, the metabolite profile will also be different. An individual's metabolite levels can change dynamically with changes in time, environment and physiological state. The brain age prediction model uses non-targeted metabolomics to monitor changes in brain age over time and the effects of intervention measures through multiple sampling and longitudinal studies, thereby providing a dynamic monitoring method for personalized health management. Combined with improved residual connections and self-attention mechanisms, feature representations can be effectively learned from metabolite features to accurately capture metabolic features related to brain age. Through multiple iterations and optimization algorithms, the brain age prediction model can reduce prediction bias, improve overall prediction performance, and ultimately achieve high-precision brain age prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1A schematic diagram of a process for predicting brain age based on deep learning and metabolomics data according to an embodiment of the present application;

[0036] Figure 2 A schematic diagram of the structure of a brain age prediction model of a brain age prediction method based on deep learning and metabolomics data according to an embodiment of the present application;

[0037] Figure 3 A graph showing the relationship between brain age and chronological age predicted by the brain age prediction method based on deep learning and metabolomics data in an embodiment of the present application;

[0038] Figure 4 A distribution diagram of the Pearson correlation coefficient (r) between the chronological age and the predicted brain age on the Bootstrap test set of the brain age prediction method based on deep learning and metabolomics data in an embodiment of the present application;

[0039] Figure 5 A brain age acceleration distribution diagram of a brain age prediction method based on deep learning and metabolomics data in an embodiment of the present application;

[0040] Figure 6 A schematic diagram of a brain age prediction device based on deep learning and metabolomics data according to an embodiment of the present application;

[0041] Figure 7 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all 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.

[0043] Figure 1 A method for predicting brain age based on deep learning and metabolomics data provided in an embodiment of the present application is shown, comprising the following steps:

[0044] S1. Obtain blood samples from healthy people and people with elderly brain diseases and perform metabolite extraction and chromatography-mass spectrometry analysis to obtain mass spectrometry data of each metabolite in several metabolomes corresponding to the healthy people and the people with elderly brain diseases. Perform data cleaning and normalization on the mass spectrometry data of each metabolite in each metabolome to obtain several metabolite features, which are composed of chromatographic retention time and mass-to-charge ratio in the mass spectrometry data of the metabolites. Perform feature screening on several metabolite features corresponding to the healthy people and the people with elderly brain diseases to determine key metabolites related to brain age.

[0045] In a specific embodiment, several metabolite features corresponding to healthy people and people with elderly brain diseases are screened to determine key metabolites related to brain age, including:

[0046] Perform differential analysis feature screening on several metabolite features, perform a two-sample t-test on the metabolite features of the healthy population and the metabolite features of the population with elderly brain diseases, obtain t values ​​and p values, use the BH method to correct the p value, obtain the corrected p value, and form a differential metabolite set with metabolites that meet the corrected p value less than 0.05 and the difference multiple greater than 2 or less than 1 / 2. The difference multiple is the multiple ratio of the median value of the metabolite feature in the population with elderly brain diseases to the median value of the metabolite feature in the healthy population for each metabolite;

[0047] A random forest model is constructed, training data is constructed based on the metabolite characteristics of healthy people and the metabolite characteristics of people with elderly brain diseases, the random forest model is trained using the training data, the metabolite characteristics in the training data are input into the random forest model, and the classification results are obtained, and the classification results are healthy people or people with elderly brain diseases. During the training process, the reduction of the Gini index caused by each metabolite characteristic at each node in each decision tree is calculated and accumulated to obtain the total reduction of the Gini index of each decision tree, the total reduction of the Gini index of all decision trees is averaged, the average reduction of the Gini index is calculated and sorted from large to small, and the metabolites corresponding to the metabolite characteristics ranked in the top 500 of the average reduction of the Gini index are selected to form an important metabolite set;

[0048] The intersection of the differential metabolite set and the important metabolite set was taken to obtain the key metabolite features related to brain age.

[0049] Specifically, sample collection and metabolite extraction are first performed, and the experimental process includes sample collection, metabolite extraction, LC-MS mass spectrometry analysis and data analysis. The present invention uses advanced data processing and analysis technology, based on the results of ultra-high performance liquid chromatography-mass spectrometry (UPLC-MS) analysis of blood from more than 8,000 adults. During the analysis, the blood sample is first thawed and fully mixed in a low-temperature environment, then treated with pre-cooled methanol, and metabolites are extracted by ultrasonic treatment. Then, the impurities are precipitated at ultra-low temperatures, and the supernatant is separated by high-speed centrifugation. The supernatant is evaporated in a high-speed vacuum concentrator centrifuge for use in mass spectrometry detection. In the mass spectrometry detection stage, the dried sample is re-dissolved with a pre-cooled methanol-water solution and centrifuged again to ensure purity and uniformity. Subsequently, the sample is introduced into an ultra-high performance liquid chromatography system (UHPLC) for separation, and an efficient chromatographic column and a precise gradient elution program are used to achieve optimal separation and detection of metabolites. The entire analysis process is carried out under strictly controlled temperature and flow rate conditions to ensure the accuracy and repeatability of the data. In the raw data analysis stage, MSDIAL software was used to process the raw mass spectrometry data, including peak alignment, retention time correction, and peak area extraction. The metabolite structures were identified by accurate mass matching and secondary spectrum matching, and public databases such as HMDB, MassBank, and GNPS were searched to determine the metabolite names.

[0050] Next, data cleaning and normalization are performed: Based on the above data, the metabolome data is cleaned and normalized, which is a key step in metabolomics analysis, aiming to ensure the quality and consistency of the data for subsequent analysis and interpretation. The KNN interpolation method is used to process the missing values ​​in the mass spectrometry data of each metabolite in the metabolome. If the number of missing values ​​in the mass spectrometry data of a metabolite in the metabolome is greater than the threshold, the metabolite is eliminated to obtain the mass spectrometry data of each metabolite after data cleaning to avoid bias in the analysis results. The mass spectrometry data of each metabolite after data cleaning is logarithmically transformed and each value is unit added to allow zero value conversion, and then centered on the mean and scaled to unit variance to reduce the skewed distribution of the data so that the mean of each feature is 0 and the standard deviation is 1. The normalization formula is:

[0051] ;

[0052] in, and are the mean and standard deviation of each feature, Represents the original data value, Indicates the normalized value.

[0053] Finally, the chromatographic retention time and mass-to-charge ratio in the mass spectral data of each metabolite were extracted to construct the metabolite signature.

[0054] In order to find the metabolite features that provide the best brain age prediction performance, the embodiments of the present application use blood samples from two different types of people, including healthy people without obvious major diseases and patients diagnosed with elderly brain diseases (Alzheimer's disease, Parkinson's disease) whose probability of disease increases with age. And three metabolomics data sets were merged, including: MS HPOS in serum (hydrophilic interaction chromatography HILIC can enhance the separation of highly polar small molecules ionized in positive mode), MS LNEG in serum (lipid-targeted reversed-phase chromatography can provide the maximum resolution of fatty acids, triglycerides and phospholipids ionized in negative mode), and MS LPOS in serum (lipid-targeted reversed-phase chromatography, ionized in positive mode). After peak alignment, retention time correction and peak area extraction, the offline data was cleaned and normalized, and a total of 14,549 metabolite features were obtained for feature screening of model construction.

[0055] Feature screening is an important step in building an efficient prediction model, which can significantly improve the performance and interpretability of the model. Therefore, in the feature screening stage, difference analysis feature screening and machine learning feature screening were performed respectively, and the intersection of the feature results was obtained, which is the feature set of the input model.

[0056] (1) Differential analysis feature screening: Through statistical analysis, the difference in metabolite concentration between healthy people and people with elderly brain diseases is compared. The specific method includes: taking all metabolite features of healthy people and all metabolite features of people with elderly brain diseases as two groups of samples, each sample is a metabolite feature, and performing a two-sample t-test on the two groups of samples. The t-test statistic (usually denoted as t-value) is a core value in the t-test calculation, which is used to measure the difference between the means of the two groups of sample sets relative to the size of the internal variation of the sample sets. The calculation formula of the t-test statistic is:

[0057] ;

[0058] in, and are the sample means of healthy people and people with elderly brain diseases, and are the sample variances of the two sample sets, respectively, and n1 and n2 are the sample sizes. The t-test statistic measures the ratio of the mean difference to the intra-group variation. A larger t-value indicates that the difference in the mean between the groups is larger relative to the intra-group variation, and vice versa. The p-value is a probability value obtained in the t-test, which indicates the probability of observing the current data or more extreme data when the null hypothesis is true (i.e., assuming that the means of the two groups are equal). The p-value is used to help determine whether the t-value is large enough to consider that there is a significant difference between the two sets of data. The p-value is adjusted by the Benjamini-Hochberg (BH) method to obtain the corrected p-value to control the false positive rate. The discriminant standard is set to have a corrected p-value less than 0.05, and the difference multiple (the ratio of the median value of the samples of the elderly brain disease population to the median value of the samples of the healthy population for each metabolite) is greater than 2 or the difference multiple is less than 1 / 2, so as to screen out differential metabolites and form a differential metabolite set. A difference multiple greater than 1 means that the content of a certain metabolite in the samples of people with elderly brain diseases is increased relative to the samples of healthy people. A difference multiple equal to 2 means that the content of a certain metabolite in the samples of people with elderly brain diseases is twice that of the samples of healthy people. Metabolites that meet this condition are considered to be metabolites that undergo significant changes with age, that is, differential metabolites, and there are 366 differential metabolites in total. Through this detailed differential analysis process, metabolites that change significantly with age can be identified, providing important biomarkers for subsequent brain age prediction models. This method not only improves the accuracy of metabolomics data analysis, but also provides new insights into understanding metabolic changes during brain aging.

[0059] (2) Machine learning feature screening:

[0060] The metabolomics data of healthy people and people with elderly brain diseases were grouped, and the random forest model was used for feature screening. The healthy people were labeled as 0 and the people with elderly brain diseases were labeled as 1 to construct training data.

[0061] First, a random forest model is trained using training data so that the random forest model can distinguish between healthy people and people with elderly brain diseases. In order to perform feature selection through the random forest method, it is first necessary to calculate the contribution of each metabolite feature to the reduction in impurity of the random forest model during the training of the random forest model. The feature importance of the random forest is calculated by examining the contribution of the metabolite feature's split node in the tree to the reduction in impurity of the model during the training of the random forest model. This process involves multiple steps and ultimately helps to screen out the most representative metabolites. Specifically, each metabolite feature is input into the random forest model, and the Gini index before and after the metabolite feature is split on all nodes of all decision trees is calculated to obtain the reduction in the Gini index (i.e., the reduction in impurity). These impurity reductions are accumulated in the entire decision tree and then averaged in the entire random forest model to obtain the average reduction in the Gini index of the metabolite feature. The average reduction in the Gini index of all metabolite features is sorted from large to small. The higher the ranking, the more important the metabolite feature. The metabolites corresponding to the metabolite features ranked in the top 500 in terms of average reduction in the Gini index are selected as important metabolites to form a set of important metabolites.

[0062] Finally, all the analysis results were integrated to determine a set of metabolites that were highly correlated with brain age, and they were applied to the construction and optimization of the brain age prediction model. The final metabolite set was selected based on the significance of differential analysis and machine learning screening, and the two sets were intersected to obtain a total of 304 key metabolites related to brain age, which were optimized and applied to the brain age prediction model to provide more accurate brain age estimates. This process not only helps to understand the role of metabolic changes in brain aging, but also provides potential biomarkers for personalized health management and disease intervention.

[0063] S2, constructing a brain age prediction model and training it with training data composed of metabolite characteristics of key metabolites corresponding to healthy people to obtain a trained brain age prediction model, the trained brain age prediction model includes a metabolomics feature representation learning module and a prediction module, the metabolomics feature representation learning module includes a first fully connected layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fifth residual module, a sixth residual module and a multi-head self-attention layer connected in sequence, and the prediction module adopts a deep neural network.

[0064] In a specific embodiment, the first residual module, the second residual module, the third residual module, the fourth residual module, the fifth residual module and the sixth residual module all adopt a residual structure, and the residual structure includes a second fully connected layer and a third fully connected layer connected in sequence. The input features of the residual structure pass through the second fully connected layer and the third fully connected layer in sequence to obtain intermediate features, and the intermediate features are added to the input features of the residual structure to form a residual connection and obtain the output features of the residual structure.

[0065] In a specific embodiment, the number of neurons in the first fully connected layer, the second fully connected layer, and the third fully connected layer are 128, and the ReLU activation function is used.

[0066] In a specific embodiment, the multi-head self-attention layer includes 4 attention heads, and the output features of the sixth residual module are subjected to three linear transformation layers to obtain a query vector, a key vector and a value vector, respectively. The query vector, the key vector and the value vector are divided into 4 query sub-vectors, 4 key sub-vectors and 4 value sub-vectors, respectively. Each query sub-vector, key sub-vector and value sub-vector performs a scaled dot product attention operation in each attention head to obtain the output vector of each attention head. The output vectors of the 4 attention heads are concatenated and then linearly transformed through a linear layer to obtain the metabolite feature representation.

[0067] In a specific embodiment, the prediction module includes an input layer, three hidden layers and an output layer connected in sequence, the input layer is used to receive the metabolite feature representation, the three hidden layers are respectively a fourth fully connected layer, a fifth fully connected layer and a sixth fully connected layer, the number of neurons in the fourth fully connected layer, the fifth fully connected layer and the sixth fully connected layer are respectively 256, 128 and 64, and all use the ReLU activation function, the output layer has 1 neuron, and uses a linear activation function.

[0068] Specifically, the brain age prediction model combines deep learning technology with metabolomics data analysis, aiming to provide accurate biological age estimation through efficient feature learning and prediction algorithms. The model consists of two main modules: (1) a metabolomics feature representation learning module that uses an improved deep residual network to learn feature representations of metabolomics data; (2) a prediction module that predicts an individual's brain age based on the metabolite feature representations obtained from the metabolomics feature representation learning module.

[0069] Metabolomics feature representation learning module: In the metabolomics feature representation learning module of the brain age prediction model, an improved deep residual network is constructed through residual connections and self-attention mechanisms, which can efficiently learn metabolite feature representations from metabolite features. The core of this model is to build a deep residual network and combine it with an attention mechanism to capture the complex relationship between features. The entire training process includes steps such as preparation of input data, design of model architecture, model training and optimization.

[0070] First, metabolomics training data is obtained, and the brain age prediction model is trained by the metabolomics training data. The metabolomics data matrix consisting of the key metabolite features corresponding to 304 key metabolites obtained after data processing of the blood samples of healthy people with known actual ages in the metabolomics training data is used as the input data of the brain age prediction model. The metabolomics data matrix is ​​a matrix of size n×304, where n represents the number of samples. The number of samples in the training data is the total number of healthy people, and the actual age of the healthy people is used as the label. In the embodiments of the present application, the actual age is based on the age on the ID card. In one embodiment, the metabolomics training data is divided into a training set and a validation set = 7:3, the training set is used for model training, and the validation set is used to evaluate model performance and prevent overfitting. In other embodiments, the ratio of the training set to the validation set can be determined according to demand. The embodiment of the present application selects the key metabolite features corresponding to the healthy population to construct the metabolomics training data to avoid the key metabolite features of people with elderly brain diseases affecting the accuracy of the prediction of the brain age prediction model.

[0071] Next, refer to Figure 2 , construct a metabolomics feature representation learning module. The deep residual network in the metabolomics feature representation learning module includes a first fully connected layer, 6 residual modules and a multi-head self-attention layer. Each residual module contains two fully connected layers and a skip connection. The input of the residual module is directly passed to the output and added to the output after passing through the two fully connected layers to form a residual connection. The two fully connected layers are the second fully connected layer and the third fully connected layer. The number of neurons in each fully connected layer is set to 128. The introduction of the skip connection is to alleviate the gradient vanishing problem in the deep residual network. This design allows information to flow through the network more efficiently and promotes deeper network training. In the residual module, each fully connected layer uses a nonlinear activation function, the ReLU activation function (Rectified Linear Unit), to introduce nonlinear transformations to capture complex feature relationships. The output of each residual module is the sum of the input and the result of the two-layer fully connected transformation. This residual connection design enables the network to better learn and transmit information while reducing the risk of gradient vanishing.

[0072] In order to further enhance the feature learning ability, the metabolomics feature representation learning module introduces a multi-head self-attention layer, which captures the complex relationship between features through a 4-head attention mechanism. The core idea of ​​the attention mechanism is to highlight the interaction between important features by assigning different weights to different features. In the multi-head self-attention layer, the query matrix, key matrix, and value matrix of the input features are first calculated, and the importance of each feature is determined by the interaction of these matrices. Specifically, the similarity between features is used to calculate weights, which are used to weighted sum the input features to generate new feature representations. The self-attention mechanism can capture the complex interactions between metabolites, which is particularly important in metabolomics data because different metabolites may be highly correlated and influence each other. The dimension of the output of the multi-head self-attention layer is consistent with the input of the multi-head self-attention layer. The input of the multi-head self-attention layer comes from the output of the last residual module. The dimension of the output of the last residual module is 128, and the dimension of the output of the multi-head self-attention layer is also 128.

[0073] Through these steps, the metabolomics feature representation learning module is able to learn high-quality feature representations from the metabolomics data matrix consisting of metabolite features corresponding to 304 key metabolites. These feature representations are input into the prediction module to further predict the individual's brain age using a deep neural network (DNN).

[0074] First, the metabolite feature representations output from the metabolomics feature representation learning module are used as the input of the prediction module. These metabolite feature representations are optimized to accurately capture the metabolite features associated with brain age and have high representation capabilities. The core of the prediction module is a deep neural network (DNN) composed of multiple fully connected layers. The specific model architecture design is as follows:

[0075] Input layer: receives the metabolite feature representation output from the metabolomics feature representation learning module, represented as an n×k matrix, where n represents the number of samples and the input dimension k is the output dimension 128 of the last layer of the metabolomics feature representation learning module.

[0076] Three hidden layers: These hidden layers are designed to further process input features and extract higher-level feature representations. Each layer uses the ReLU activation function. The first hidden layer is the fourth fully connected layer, which contains 256 neurons and uses the ReLU activation function to capture complex feature relationships; the second hidden layer is the fifth fully connected layer, which contains 128 neurons and uses the ReLU activation function to further extract and refine features; the third hidden layer is the sixth fully connected layer, which contains 64 neurons and uses the ReLU activation function to compress feature dimensions and improve the generalization ability of the model.

[0077] Output layer: Contains one neuron, which outputs the predicted brain age of each person through a linear activation function. The single neuron is designed to directly regress the prediction target, that is, the brain age of each person.

[0078] In the training process of the brain age prediction model, the mean absolute error (MSE) is used as the loss function, and the model parameters are adjusted by the gradient descent method and the optimization algorithm. The training set is divided into 32 small batches, and one batch of data is used for parameter update in each iteration to reduce memory usage and increase training speed. For each training batch, the mean absolute error (MSE) between the predicted brain age and the actual age is calculated, the gradient is calculated by the back propagation algorithm, and the model parameters are updated using the optimization algorithm. At the end of each epoch, the validation set is used to evaluate the model performance and the MSE of the validation set is calculated. At the same time, an early stopping mechanism is introduced. When the loss of the validation set no longer decreases within several epochs, the training is terminated early to prevent the model from overfitting.

[0079] The final results showed that the brain age prediction model had significant advantages in predicting brain age, with a Pearson correlation coefficient of 0.73, a determination coefficient of 0.73, and an MSE of 13.55. Figure 3 As shown in the figure, the Pearson correlation coefficient (0.73) and the coefficient of determination (R²=0.73) between the predicted brain age and the actual age both indicate that there is a strong linear relationship between brain age and actual age, while the mean absolute error of 13.55 reflects the small average deviation between the predicted and actual values. The low value of MSE indicates that the brain age prediction model has high prediction accuracy and small error, and can provide reliable brain age estimates. This result shows that the brain age prediction model can not only accurately predict the brain age of an individual, but also effectively capture the metabolic characteristics related to brain aging, thereby providing an important biomarker for personalized health management and disease intervention.

[0080] refer to Figure 4 To define a stable metabolic brain age score for each participant, the BrainAgingNet model was built 100 times using the bootstrap procedure on the resampled training and test sets of the dataset. Age predictions in the independent training set were highly accurate and stable across all resamples, with an average correlation coefficient of 0.74 (95% confidence interval, 0.72 to 0.76) between predicted brain age and actual age in the test set.

[0081] S3, obtaining a blood sample from the person to be tested and obtaining a metabolomics data matrix consisting of metabolite characteristics of corresponding key metabolites, inputting the metabolomics data matrix into a metabolomics characteristic representation learning module in a trained brain age prediction model to obtain a metabolite characteristic representation, inputting the metabolite characteristic representation into a prediction module, predicting the brain age of the person to be tested, and calculating the brain age acceleration.

[0082] Specifically, after the key metabolites related to brain age are determined in step S1, metabolite extraction and chromatography-mass spectrometry analysis are performed on the obtained blood samples of the person to be tested to obtain mass spectrometry data of each metabolite in several metabolomes corresponding to the person to be tested, and data cleaning and normalization are performed on the mass spectrometry data of each metabolite in each metabolome to obtain several metabolite features, and metabolite features corresponding to the key metabolites are selected from the several metabolite features, and a metabolomics data matrix of a size of 1×304 is constructed, and the metabolomics data matrix corresponding to the person to be tested is input into the trained brain age prediction model to predict brain age.

[0083] Based on the brain age prediction model, the brain age of the person to be tested can be predicted, so as to evaluate the individual's physiological state and health risks. Through further analysis of the prediction results, key metabolite markers can be identified, providing important clues for subsequent biological research and clinical applications. At the same time, the embodiment of the present application adopts the concept of Brain Age Acceleration (BAA), which refers to the difference between an individual's predicted brain age and his actual age, that is, the part of the predicted brain age that exceeds or is lower than the actual age. The specific definitions are as follows:

[0084] Brain age acceleration = predicted brain age - chronological age;

[0085] If the predicted brain age is higher than the chronological age, the brain age acceleration is positive, indicating that the individual's brain physiological state may be "older" than the actual age; conversely, if the predicted brain age is lower than the chronological age, the acceleration is negative, indicating that the individual's brain physiological state may be "younger" than the actual age. BAA can be used as an important indicator to assess an individual's health status and brain aging rate.

[0086] The application of this indicator provides a new tool for clinicians to help identify potential health risks at an early stage and provide a reference for individualized health management. By evaluating the acceleration of brain age through the brain age prediction model, it is possible to more accurately judge the individual's brain health status and promote early prevention and intervention of brain aging-related diseases.

[0087] refer to Figure 5The data in the figure show that the distribution of brain age acceleration (BAA) is roughly normal, and most individuals have BAA values ​​between -8 and 8. This means that most people's brain age is close to their actual age, but a few people have larger BAA values, indicating that their brain age deviates significantly from their actual age. This result helps to identify those whose brain age is significantly accelerated or slowed down, providing an important reference for personalized health management and early intervention.

[0088] Based on the established model, the metabolomics data matrix of the person to be tested is input into the brain age prediction model to obtain the predicted brain age and BAA value, which can evaluate the individual's brain age more comprehensively and accurately, and can achieve a comprehensive assessment of the individual's brain physiological state, thereby improving the accuracy of brain health risk prediction and providing a scientific basis for personalized brain health management and early intervention.

[0089] Further references Figure 6 As an implementation of the methods shown in the above figures, the present application provides an embodiment of a brain age prediction device based on deep learning and metabolomics data. Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0090] The present application embodiment provides a brain age prediction device based on deep learning and metabolomics data, comprising:

[0091] The feature screening module 1 is configured to obtain blood samples from healthy people and people with elderly brain diseases and perform metabolite extraction and chromatography-mass spectrometry analysis to obtain mass spectrometry data of each metabolite in several metabolomes corresponding to the healthy people and the people with elderly brain diseases, perform data cleaning and normalization on the mass spectrometry data of each metabolite in each metabolome to obtain several metabolite features, wherein the metabolite features are formed by splicing the chromatographic retention time and mass-to-charge ratio in the mass spectrometry data of the metabolites, perform feature screening on several metabolite features corresponding to the healthy people and the people with elderly brain diseases, and determine key metabolites related to brain age;

[0092] A model building module 2 is configured to build a brain age prediction model and train it using training data consisting of metabolite characteristics of key metabolites corresponding to healthy people to obtain a trained brain age prediction model, wherein the trained brain age prediction model includes a metabolomics feature representation learning module and a prediction module, wherein the metabolomics feature representation learning module includes a first fully connected layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fifth residual module, a sixth residual module and a multi-head self-attention layer connected in sequence, and the prediction module uses a deep neural network;

[0093] The reasoning module 3 is configured to obtain a blood sample from the person to be tested and obtain a metabolomics data matrix composed of metabolite characteristics of corresponding key metabolites, input the metabolomics data matrix into the metabolomics characteristic representation learning module in the trained brain age prediction model to obtain the metabolite characteristic representation, input the metabolite characteristic representation into the prediction module, predict the brain age of the person to be tested, and calculate the brain age acceleration.

[0094] Figure 7 Schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present invention. Figure 7 As shown, the electronic device of this embodiment includes: a processor 701 and a memory 702; wherein the memory 702 is used to store computer-executable instructions; the processor 701 is used to execute the computer-executable instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant description in the above method embodiment.

[0095] Optionally, the memory 702 may be independent or integrated with the processor 701 .

[0096] When the memory 702 is independently provided, the electronic device further includes a bus 703 for connecting the memory 702 and the processor 701 .

[0097] The embodiment of the present invention further provides a computer storage medium, in which computer execution instructions are stored. When the processor 701 executes the computer execution instructions, the above method is implemented.

[0098] The embodiment of the present invention further provides a computer program product, including a computer program. When the computer program is executed by the processor 701, the above method is implemented.

[0099] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0100] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

[0101] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned module-composed unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0102] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor 701 to perform some steps of the methods of various embodiments of the present application.

[0103] It should be understood that the processor 701 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or the processor 701 may be any conventional processor 701, etc. The steps of the method disclosed in the invention may be directly embodied in the execution of the hardware processor 701, or may be executed by a combination of hardware and software modules in the processor 701.

[0104] The memory 702 may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk.

[0105] The bus 703 may be an Industry Standard Architecture (ISA), a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus 703 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus 703 in the drawings of the present application is not limited to only one bus 703 or one type of bus 703.

[0106] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0107] An exemplary storage medium is coupled to the processor 701, so that the processor 701 can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor 701. The processor 701 and the storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as ASIC). Of course, the processor 701 and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0108] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A brain age prediction method based on deep learning and metabolomics data, characterized in that: The following steps are involved: Blood samples from healthy people and people with elderly brain diseases are obtained and metabolite extraction and chromatography-mass spectrometry analysis are performed to obtain mass spectrometry data of each metabolite in several metabolomes corresponding to the healthy people and the people with elderly brain diseases. The mass spectrometry data of each metabolite in each metabolome is cleaned and normalized to obtain several metabolite features, which are spliced ​​by chromatographic retention time and mass-to-charge ratio in the mass spectrometry data of the metabolites. Several metabolite features corresponding to the healthy people and the people with elderly brain diseases are screened to determine key metabolites related to brain age, including: Perform differential analysis feature screening on several metabolite features, perform a two-sample t-test on the metabolite features of a healthy population and the metabolite features of a population suffering from elderly brain diseases, obtain a t value and a p-value, correct the p-value using the BH method to obtain a corrected p-value, and form a differential metabolite set with metabolites corresponding to metabolite features satisfying the corrected p-value being less than 0.05 and having a difference multiple greater than 2 or less than 1 / 2, wherein the difference multiple is the multiple ratio of the median value of the metabolite feature in the population suffering from elderly brain diseases to the median value of the metabolite feature in the healthy population for each metabolite; Constructing a random forest model, the random forest model includes several decision trees, constructing training data based on the metabolite characteristics of healthy people and the metabolite characteristics of people with elderly brain diseases, using the training data to train the random forest model, inputting the metabolite characteristics in the training data into the random forest model, and obtaining a classification result, wherein the classification result is a healthy person or a person with elderly brain diseases, calculating and accumulating the reduction of the Gini index caused by each metabolite characteristic at each node in each decision tree during the training process, and obtaining the total reduction of the Gini index of each decision tree, averaging the total reduction of the Gini index of all decision trees, calculating the average reduction of the Gini index and sorting them from large to small, and selecting metabolites corresponding to the metabolite characteristics ranked in the top 500 in terms of the average reduction of the Gini index to form an important metabolite set; Taking the intersection of the differential metabolite set and the important metabolite set to obtain key metabolite features related to brain age; Constructing a brain age prediction model and training it with training data consisting of metabolite characteristics of key metabolites corresponding to healthy people to obtain a trained brain age prediction model, wherein the trained brain age prediction model includes a metabolomics feature representation learning module and a prediction module, wherein the metabolomics feature representation learning module includes a first fully connected layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fifth residual module, a sixth residual module and a multi-head self-attention layer connected in sequence, and the prediction module uses a deep neural network; A blood sample of the person to be tested is obtained to obtain a metabolomics data matrix composed of metabolite characteristics of corresponding key metabolites, the metabolomics data matrix is ​​input into the metabolomics characteristic representation learning module in the trained brain age prediction model to obtain a metabolite characteristic representation, the metabolite characteristic representation is input into the prediction module, the brain age of the person to be tested is predicted, and the brain age acceleration is calculated.

2. The method for predicting brain age based on deep learning and metabolomics data according to claim 1, characterized in that: The first residual module, the second residual module, the third residual module, the fourth residual module, the fifth residual module and the sixth residual module all adopt a residual structure, and the residual structure includes a second fully connected layer and a third fully connected layer connected in sequence. The input features of the residual structure pass through the second fully connected layer and the third fully connected layer in sequence to obtain intermediate features, and the intermediate features are added to the input features of the residual structure to form a residual connection and obtain the output features of the residual structure.

3. The method for predicting brain age based on deep learning and metabolomics data according to claim 2, characterized in that: The number of neurons in the first fully connected layer, the second fully connected layer and the third fully connected layer are 128, and the ReLU activation function is used.

4. The method for predicting brain age based on deep learning and metabolomics data according to claim 1, characterized in that: The multi-head self-attention layer includes 4 attention heads. The output features of the sixth residual module are subjected to three linear transformation layers to obtain a query vector, a key vector and a value vector respectively. The query vector, the key vector and the value vector are divided into 4 query sub-vectors, 4 key sub-vectors and 4 value sub-vectors respectively. Each query sub-vector, key sub-vector and value sub-vector performs a scaled dot product attention operation in each attention head to obtain the output vector of each attention head. The output vectors of the 4 attention heads are concatenated and then linearly transformed through a linear layer to obtain the metabolite feature representation.

5. The method for predicting brain age based on deep learning and metabolomics data according to claim 1, characterized in that: The prediction module includes an input layer, three hidden layers and an output layer connected in sequence, the input layer is used to receive the metabolite feature representation, the three hidden layers are respectively a fourth fully connected layer, a fifth fully connected layer and a sixth fully connected layer, the number of neurons in the fourth fully connected layer, the fifth fully connected layer and the sixth fully connected layer are respectively 256, 128 and 64, and all use the ReLU activation function, the output layer has 1 neuron and uses a linear activation function.

6. A brain age prediction device based on deep learning and metabolomics data, characterized in that: include: The feature screening module is configured to obtain blood samples from healthy people and people with elderly brain diseases and perform metabolite extraction and chromatography-mass spectrometry analysis to obtain mass spectrometry data of each metabolite in several metabolomes corresponding to the healthy people and the people with elderly brain diseases, perform data cleaning and normalization on the mass spectrometry data of each metabolite in each metabolome to obtain several metabolite features, wherein the metabolite features are spliced ​​by the chromatographic retention time and mass-to-charge ratio in the mass spectrometry data of the metabolites, perform feature screening on several metabolite features corresponding to the healthy people and the people with elderly brain diseases, and determine key metabolites related to brain age, specifically including: Perform differential analysis feature screening on several metabolite features, perform a two-sample t-test on the metabolite features of a healthy population and the metabolite features of a population suffering from elderly brain diseases, obtain a t value and a p-value, correct the p-value using the BH method to obtain a corrected p-value, and form a differential metabolite set with metabolites corresponding to metabolite features satisfying the corrected p-value being less than 0.05 and having a difference multiple greater than 2 or less than 1 / 2, wherein the difference multiple is the multiple ratio of the median value of the metabolite feature in the population suffering from elderly brain diseases to the median value of the metabolite feature in the healthy population for each metabolite; Constructing a random forest model, the random forest model includes several decision trees, constructing training data based on the metabolite characteristics of healthy people and the metabolite characteristics of people with elderly brain diseases, using the training data to train the random forest model, inputting the metabolite characteristics in the training data into the random forest model, and obtaining a classification result, wherein the classification result is a healthy person or a person with elderly brain diseases, calculating and accumulating the reduction of the Gini index caused by each metabolite characteristic at each node in each decision tree during the training process, and obtaining the total reduction of the Gini index of each decision tree, averaging the total reduction of the Gini index of all decision trees, calculating the average reduction of the Gini index and sorting them from large to small, and selecting metabolites corresponding to the metabolite characteristics ranked in the top 500 in terms of the average reduction of the Gini index to form an important metabolite set; Taking the intersection of the differential metabolite set and the important metabolite set to obtain key metabolite features related to brain age; A model building module is configured to build a brain age prediction model and train it using training data composed of metabolite characteristics of key metabolites corresponding to healthy people to obtain a trained brain age prediction model, wherein the trained brain age prediction model includes a metabolomics feature representation learning module and a prediction module, wherein the metabolomics feature representation learning module includes a first fully connected layer, a first residual module, a second residual module, a third residual module, a fourth residual module, a fifth residual module, a sixth residual module and a multi-head self-attention layer connected in sequence, and the prediction module uses a deep neural network; The inference module is configured to obtain a blood sample of a person to be tested and obtain a metabolomics data matrix composed of metabolite features of corresponding key metabolites, input the metabolomics data matrix into the metabolomics feature representation learning module in the trained brain age prediction model to obtain a metabolite feature representation, input the metabolite feature representation into the prediction module, predict the brain age of the person to be tested, and calculate the brain age acceleration.

7. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images