Using artificial intelligence to provide early warning of adolescent depression

Abstract

The present invention relates to the field of mental health analysis and early warning technology, and specifically to an early warning method for adolescent depression tendency using artificial intelligence, comprising the following steps: collecting adolescent microbiome data, analyzing to obtain initial intestinal microbiome data, and detecting to obtain abnormal intestinal microbiome data; analyzing adolescent psychological behavior data and identifying behavior patterns; identifying biomarkers related to depression tendency; comparing the overlap between biomarkers and abnormal intestinal microbiome data to determine potential depression tendency; constructing a depression tendency verification model, the model outputs the behavior pattern corresponding to the biomarker to determine whether there is a depression tendency; setting an early warning threshold, and automatically triggering an early warning mechanism when the depression tendency index exceeds the threshold. The present invention helps to reveal the complex biological and psychological mechanisms behind depression, thereby providing deeper insights into the prevention and intervention of adolescent depression.

Description

Technical Field

[0001] The present invention relates to the field of mental health analysis and early warning technology, and in particular to an early warning method for adolescent depression tendency using artificial intelligence. Background Art

[0002] In the current field of mental health, especially in research focusing on adolescent depression, a persistent challenge is how to identify and predict depressive tendencies in a timely and accurate manner. Traditional methods rely on face-to-face clinical assessments, self-report questionnaires, and observations by parents or teachers. Although these methods are effective in certain situations, they also have certain limitations. For example, self-report questionnaires may be affected by the subject's current emotional state, self-perception ability, and the authenticity of the answers, while the accessibility and frequency of clinical assessments are limited by resource availability.

[0003] With the deepening of biomarker research and the rapid development of information technology, new methods have begun to be explored to supplement traditional methods of detecting depressive tendencies. In recent years, scientific research has revealed that there is a close connection between the human microbiome, especially the intestinal microbiome, and human mental health, which has triggered interest in using microbiome data as a potential biomarker to predict mental health status; in recent years, scientists have found that the microbiome is closely related to human health and disease, including digestive system diseases, immune system function, and mental health. Regarding whether microbiome data can be used to predict adolescent depressive tendencies, current studies have shown that there is a certain correlation between the composition of the intestinal microbiome and human mental health. Some studies have found that specific changes in intestinal microorganisms are associated with psychological states such as mood regulation, anxiety, and depression. This is called the "gut-brain axis," which refers to a two-way communication system between the intestine and the brain, in which changes in the intestinal microbiome can affect brain function through neural, endocrine, and immune pathways, thereby affecting emotions and behaviors.

[0004] In addition, with the popularization of social media and wearable devices, a large amount of behavioral data has become possible, providing a new perspective to observe and analyze individual psychological behavioral patterns.

[0005] In this context, it is feasible and necessary to develop an artificial intelligence model that integrates biomarkers and psychological behavioral data, especially using deep learning techniques such as multi-layer perceptron (MLP) to predict adolescent depression tendencies. This method can make full use of multi-dimensional data and improve the accuracy of prediction by learning complex patterns in the data. Summary of the invention

[0006] Based on the above objectives, the present invention provides an early warning method for adolescent depression tendency using artificial intelligence.

[0007] The method for early warning of adolescent depression tendency using artificial intelligence includes the following steps:

[0008] S1: Collect microbiome data of adolescents, obtain initial intestinal microbiome data through non-invasive biological sample analysis, and test the initial intestinal microbiome data to obtain abnormal intestinal microbiome data;

[0009] S2: Analyze adolescents’ psychological and behavioral data, including emotional self-assessment, behavioral questionnaires, and social interaction frequency, and identify behavioral patterns;

[0010] S3: Use machine learning techniques to conduct in-depth analysis of abnormal microbiome data and identify biomarkers associated with depressive tendencies;

[0011] S4: Compare the overlap between biomarkers and abnormal gut microbiome data. If the overlap exceeds the preset threshold, it is judged as having potential depression tendency.

[0012] S5: Construct a depression tendency verification model, which takes into account the interactive effects of biomarkers and behavioral patterns. The model outputs the behavioral pattern corresponding to the biomarker, and compares the corresponding behavioral pattern with the behavioral pattern identified by S2 to determine whether there is a depression tendency;

[0013] S6: Set a warning threshold. When the depression tendency index exceeds the threshold, the warning mechanism is automatically triggered.

[0014] Furthermore, the S1 specifically includes:

[0015] S11: Collect biological samples from adolescents using non-invasive sample collection techniques, including saliva, skin swabs, or stool samples;

[0016] S12: Use high-throughput sequencing technologies, including 16S rRNA sequencing and whole genome sequencing (WGS), to sequence the collected samples. 16S rRNA sequencing focuses on the areas of bacteria and archaea, providing information on the relative abundance of microbial species, and WGS provides information on the composition of microorganisms, including bacteria, viruses, fungi, and archaea.

[0017] S13: Use bioinformatics methods to process sequencing data, including data quality control, sequence splicing, operational taxonomic unit clustering, and species annotation. Data quality control includes removing low-quality and short sequences to ensure data quality. Sequence splicing assembles sequencing reads into longer sequences. Operational taxonomic unit clustering groups sequences based on sequence similarity to represent specific microbial communities. Species annotation compares operational taxonomic unit clusters with known microbial databases to determine their species identities. Analyze the composition and diversity of the intestinal microbiome, identify microbial communities that deviate from the normal intestinal microbiome, that is, abnormal intestinal microbiome data, including analyzing changes in the relative abundance of microorganisms, the disappearance or appearance of specific microbial communities, and changes in overall diversity.

[0018] Furthermore, the S2 specifically includes:

[0019] S21: Collect adolescents’ self-rating data on their emotions, including asking them to rate their own emotional states through online or paper questionnaires;

[0020] S22: Collect data on adolescents’ behavioral questionnaires, which include a series of questions designed to assess adolescents’ daily behavioral habits, participation in social activities, interests and hobbies, sleep quality, and learning status;

[0021] S23: Collect data on the frequency of adolescents' social interactions through social media platforms, school records, or parent feedback, including the frequency of their activities on social media, the number of times they participate in social activities, and the frequency of communication with friends and family;

[0022] S24: Use principal component analysis (PCA) to comprehensively analyze the collected emotional self-assessment data, behavioral questionnaire data, and social interaction frequency data to identify behavioral patterns, including social avoidance behavior, emotional fluctuation patterns, and lack of interest.

[0023] Furthermore, the S3 specifically includes:

[0024] S31, constructing a feature matrix: quality control of the microbiome data obtained by high-throughput sequencing, including removal of low-quality sequences and noise, performing the ASVs (amplified sequence variants) method on the sequences, and standardizing their abundance data to form an initial feature matrix, where each ASV represents a feature and each ASV abundance constitutes a feature vector;

[0025] S32, feature extraction: training a recurrent neural network (RNN) model and automatically extracting microbiome patterns from the initial feature matrix through the RNN model;

[0026] S33, using recursive feature elimination algorithm to screen out the most influential microbial features for predicting depression tendency;

[0027] S34, validate and test the RNN model, using cross-validation techniques to evaluate the accuracy, sensitivity, and specificity of the model to ensure that the identified biomarkers are statistically relevant to adolescent depression tendencies;

[0028] S35, based on the results of the RNN model, determine the biomarkers associated with adolescent depression tendency.

[0029] Furthermore, the RNN model uses internal hidden states to capture the dependencies between time points in the sequence. For a given sequence X = (x 1 , x 2 , ..., x T ), the RNN model gradually updates its hidden state h through the following calculation steps t :

[0030] h t =f(W xh x t +W hh h t-1 +b h ), where h t is the hidden state at time step t, x t is the input at time step t, W xh is the weight matrix input to the hidden layer, W hh is the weight matrix from hidden layer to hidden layer, b h is the bias term of the hidden layer, and f is the activation function, such as tanh or ReLU;

[0031] Output calculation:

[0032] o t =W ho h t +b o ;

[0033]

[0034] Among them, t is the original output of the output layer at time step t, W ho is the weight matrix from the hidden layer to the output layer, b o is the bias term of the output layer, is the predicted output at time step t, and the softmax function is used for the probability output of multi-classification problems.

[0035] Furthermore, the RNN model selects a cross entropy loss function and an Adam optimizer for training optimization, the loss function is used to evaluate the difference between the model prediction and the true label, and the optimizer is used to adjust the model parameters to minimize the loss function;

[0036] The cross entropy loss function is expressed as: Among them, y i is the true label of sample i, is the probability predicted by the model;

[0037] The model is trained using the training dataset, the model weights are updated through the back-propagation algorithm, and in each training cycle, the model performance is evaluated using the validation dataset, and the model parameters are adjusted to optimize the performance.

[0038] Furthermore, the S4 specifically includes:

[0039] Based on the identification of a list of biomarkers associated with depression tendency, a reference biomarker set B is constructed;

[0040] For each adolescent sample tested, an abnormal microbiome data set A is extracted from the intestinal microbiome data obtained by high-throughput sequencing, which includes microbial species with abnormally increased or decreased abundance;

[0041] Calculate the overlap R between the reference biomarker set B and the abnormal microbiome data set A for each sample: Where |A∩B| represents the number of elements in the intersection of sets A and B, |B| is the total number of elements in the reference biomarker set, and the overlap R reflects the degree of match between the abnormal microbiome components in the sample and the biomarkers known to be associated with depression;

[0042] A preset threshold T is set. If the overlap R of the sample exceeds the threshold T, the adolescent corresponding to the sample is judged to have potential depressive tendencies; otherwise, the possibility of his or her depressive tendencies is considered to be low.

[0043] Furthermore, the S5 specifically includes:

[0044] Integrate the identified biomarkers and behavioral patterns associated with depression proneness to form a comprehensive signature set;

[0045] The comprehensive feature set is processed using data encoding and standardization to ensure that the input data of the model is in a uniform format and scale;

[0046] Multilayer Perceptron (MLP) was selected as the basic architecture of the depression tendency validation model, and the interaction between biomarkers and behavioral patterns was considered by introducing interaction terms into the depression tendency validation model.

[0047] Use a labeled dataset (where the label indicates whether there is a tendency to depression) to train and verify the model. The model training process includes parameter optimization and overfitting control (regularization).

[0048] Model output interpretation: Explain the behavioral patterns corresponding to biomarkers by analyzing the weights and feature contributions in the trained model. For example, the model may reveal that changes in certain microbiome components are highly correlated with social avoidance behavior patterns.

[0049] Furthermore, the depression tendency verification model specifically includes:

[0050] Feature engineering: Before the input layer, feature engineering is performed to create interaction terms. There are two sets of features, one of which is biomarkers B = {b 1 , b 2 , ..., b m}, the other group is behavior pattern P = {p 1 , p 2 , ..., p n}, the interaction term is generated by calculating the product of each pair of features between the two sets of features: I ij =b i ×p j , where I ij It is a biomarker b i and behavior patterns j In this way, we get m×n interaction terms.

[0051] Construct an MLP where the input layer receives the original features and interaction terms as input. Suppose the total number of original features and interaction terms is d, then the input layer has d neurons.

[0052] Hidden layer and output layer: One or more hidden layers are designed to process the input data. Each neuron in the hidden layer calculates the weighted sum of all neurons in the previous layer and passes through the ReLU activation function. The output layer contains one neuron for regression problems.

[0053] For the hidden layer l, the output h of each neuron l，k The calculation is as follows:

[0054] Among them, f is the activation function, w l，kj is the weight from the jth neuron in the l-1th layer to the kth neuron in the lth layer, b l，k is the bias term of the kth neuron in the lth layer, D l-1 is the number of neurons in the l-1th layer.

[0055] Furthermore, the model output explanation specifically includes:

[0056] Weight explanation: Each input feature is multiplied by the corresponding weight, and the product reflects the contribution of the feature to the neuron activation value. The larger the absolute value of the weight, the more important the feature is in the model. A positive weight indicates a positive correlation between the feature and the target variable, and a negative weight indicates a negative correlation.

[0057] Feature contribution: The contribution of a feature to the model prediction result is determined by observing the weight of the feature and its propagation in the neural network. If a biomarker feature has a high weight in multiple nodes and these multiple nodes contribute significantly to the final prediction, then the biomarker is a high feature importance biomarker. Subsequently, the relationship between the high feature importance biomarker and the psychological behavior pattern data is analyzed by observing the performance of the biomarker with high feature importance in a certain behavior pattern. If the high expression of a biomarker appears consistently with a certain behavior pattern, it is inferred that the biomarker is associated with the behavior pattern and is determined to have a tendency to depression.

[0058] Beneficial effects of the present invention:

[0059] The present invention provides a comprehensive mental health assessment method by comprehensively analyzing the microbiome data and psychological behavioral data of adolescents. This method not only focuses on traditional psychological assessment indicators, but also introduces microbiome data as biomarkers, providing a new dimension for early warning of depressive tendencies. By identifying specific microorganisms and psychological behavioral patterns associated with depressive tendencies, it helps to reveal the complex biological and psychological mechanisms behind depression, thereby providing deeper insights into the prevention and intervention of adolescent depression.

[0060] The present invention adopts a multi-layer perceptron (MLP) as the basic framework of the depression tendency verification model, introduces interaction terms to consider the interactive effects between biomarkers and behavioral patterns, and explains the psychological behavioral patterns corresponding to specific biomarkers by analyzing model weights and feature contributions. This method provides a highly comprehensive analysis method. By comparing the behavioral patterns identified by the model with the actual observed behavioral patterns, individuals with depression tendencies can be more accurately identified and verified. This comprehensive analysis can not only deepen our understanding of the interaction between the complex biological and psychological factors behind depression, but also provide a more scientific and accurate basis for early identification and intervention, thereby helping to improve the prevention and treatment of depression tendencies in adolescents. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0062] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0063] Figure 2 A schematic diagram of a multilayer perceptron structure with interaction terms introduced according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0065] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0066] like Figure 1-2 As shown, the method for early warning of adolescent depression tendency using artificial intelligence includes the following steps:

[0067] S1: Collect microbiome data of adolescents, obtain initial intestinal microbiome data through non-invasive biological sample analysis, and test the initial intestinal microbiome data to obtain abnormal intestinal microbiome data;

[0068] S2: Analyze adolescents’ psychological and behavioral data, including emotional self-assessment, behavioral questionnaires, and social interaction frequency, and identify behavioral patterns;

[0069] S3: Use machine learning techniques to conduct in-depth analysis of abnormal microbiome data and identify biomarkers associated with depressive tendencies;

[0070] S4: Compare the overlap between biomarkers and abnormal gut microbiome data. If the overlap exceeds the preset threshold, it is judged as having potential depression tendency.

[0071] S5: Construct a depression tendency verification model, which takes into account the interactive effects of biomarkers and behavioral patterns. The model outputs the behavioral pattern corresponding to the biomarker, and compares the corresponding behavioral pattern with the behavioral pattern identified by S2 to determine whether there is a depression tendency;

[0072] S6: Set a warning threshold. When the depression tendency index exceeds the threshold, the warning mechanism is automatically triggered.

[0073] S1 specifically includes:

[0074] S11: Collect biological samples from adolescents using non-invasive sample collection techniques, including saliva, skin swabs, or stool samples;

[0075] S12: Use high-throughput sequencing technologies, including 16S rRNA sequencing and whole genome sequencing (WGS), to sequence the collected samples. 16S rRNA sequencing focuses on the areas of bacteria and archaea, providing information on the relative abundance of microbial species, and WGS provides information on the composition of microorganisms, including bacteria, viruses, fungi, and archaea.

[0076] S13: Use bioinformatics methods to process sequencing data, including data quality control, sequence splicing, operational taxonomic unit clustering, and species annotation. Data quality control includes removing low-quality and short sequences to ensure data quality. Sequence splicing assembles sequencing reads into longer sequences. Operational taxonomic unit clustering groups sequences based on sequence similarity to represent specific microbial communities. Species annotation compares operational taxonomic unit clusters with known microbial databases to determine their species identities. Analyze the composition and diversity of the intestinal microbiome, identify microbial communities that deviate from the normal intestinal microbiome, that is, abnormal intestinal microbiome data, including analyzing changes in the relative abundance of microorganisms, the disappearance or appearance of specific microbial communities, and changes in overall diversity.

[0077] S2 specifically includes:

[0078] S21: Collect adolescents’ self-rating data on their emotions, including asking them to rate their own emotional states through online or paper questionnaires;

[0079] S22: Collect data on adolescents’ behavioral questionnaires, which include a series of questions designed to assess adolescents’ daily behavioral habits, participation in social activities, interests and hobbies, sleep quality, and learning status;

[0080] S23: Collect data on the frequency of adolescents' social interactions through social media platforms, school records, or parent feedback, including the frequency of their activities on social media, the number of times they participate in social activities, and the frequency of communication with friends and family;

[0081] S24: Use principal component analysis (PCA) to comprehensively analyze the collected emotional self-assessment data, behavioral questionnaire data, and social interaction frequency data to identify behavioral patterns, including social avoidance behavior, emotional fluctuation patterns, and lack of interest.

[0082] The specific calculation steps of PCA are:

[0083] Standardized data: For each type of psychological behavioral data (such as emotional self-assessment, behavioral questionnaires, social interaction frequency, etc.), the data is first standardized to ensure that the mean of each variable is 0 and the standard deviation is 1. This is because PCA is sensitive to the scale of the data.

[0084] Construct the covariance matrix: Calculate the covariance matrix of the standardized data, the elements of the covariance matrix ∑ ij Defined as: Among them, x ki is the value of the i-th variable of the k-th sample, is the mean of the ith variable, and n is the number of samples.

[0085] Calculate the eigenvalues ​​and eigenvectors of the covariance matrix: each eigenvector represents the direction of a principal component, and the corresponding eigenvalue represents the variance in that direction, that is, the importance of the principal component.

[0086] Based on the size of the eigenvalues, the most important principal components are selected, which can explain most of the variability of the data.

[0087] Transform to a new feature space: Project the original data onto the selected principal components to obtain a new feature space. These new features are the coordinates of the original data in the direction of the principal components.

[0088] In the present invention, by applying PCA, specific behavioral patterns associated with adolescent depressive tendencies can be identified, as follows:

[0089] Dimensionality reduction: PCA simplifies complex psychological and behavioral data into several principal components by extracting the main sources of variation, which helps to reveal the underlying behavioral patterns. For example, one principal component may represent the social avoidance behavior pattern of adolescents, while another principal component may represent the emotional fluctuation pattern.

[0090] Visualization: Project the adolescents’ data into the space composed of principal components and visualize the data in a lower dimension (such as two-dimensional or three-dimensional space). By observing the positions of different adolescents in this new feature space, potential groups with depressive tendencies can be identified.

[0091] Pattern recognition: By analyzing the relationship between each principal component and the original variables, we can interpret the behavioral pattern represented by each principal component. For example, if a principal component is highly correlated with sadness scores and a decrease in the frequency of social interactions in the self-assessment of mood, then this principal component may reflect a behavioral pattern related to depression.

[0092] S3 specifically includes:

[0093] S31, constructing a feature matrix: quality control of the microbiome data obtained by high-throughput sequencing, including removal of low-quality sequences and noise, performing the ASVs (amplified sequence variants) method on the sequences, and standardizing their abundance data to form an initial feature matrix, where each ASV represents a feature and each ASV abundance constitutes a feature vector;

[0094] S32, feature extraction: training a recurrent neural network (RNN) model and automatically extracting microbiome patterns from the initial feature matrix through the RNN model;

[0095] S33, using recursive feature elimination algorithm to screen out the most influential microbial features for predicting depression tendency;

[0096] S34, validate and test the RNN model, using cross-validation techniques to evaluate the accuracy, sensitivity, and specificity of the model to ensure that the identified biomarkers are statistically relevant to adolescent depression tendencies;

[0097] S35, based on the results of the RNN model, determine the biomarkers associated with adolescent depression tendency.

[0098] The RNN model uses internal hidden states to capture the dependencies between time points in a sequence. For a given sequence X = (x 1 ,x 2 ,…,x T ), the RNN model gradually updates its hidden state h through the following calculation steps t :

[0099] h t =f(W xh x t +W hh h t-1 +b h ), where h t is the hidden state at time step t, x tis the input at time step t, W xh is the weight matrix input to the hidden layer, W hh is the weight matrix from hidden layer to hidden layer, b h is the bias term of the hidden layer, and f is the activation function, such as tanh or ReLU;

[0100] Output calculation:

[0101] o t =W ho h t +b o ;

[0102]

[0103] Among them, t is the original output of the output layer at time step t, W ho is the weight matrix from the hidden layer to the output layer, b o is the bias term of the output layer, is the predicted output at time step t, and the softmax function is used for the probability output of multi-classification problems.

[0104] In the present invention, the input x at each time step is t is the feature vector of preprocessed microbiome data. By capturing the temporal dependencies between these inputs, RNN can identify microbiome patterns associated with depressive tendencies.

[0105] The above RNN model is trained by combining known depression-related data (psychological assessment, behavioral data, etc.) as training labels.

[0106] The RNN model uses the cross entropy loss function and Adam optimizer for training optimization. The loss function is used to evaluate the difference between the model prediction and the true label, and the optimizer is used to adjust the model parameters to minimize the loss function.

[0107] The cross entropy loss function is expressed as: Among them, y i is the true label of sample i, is the probability predicted by the model;

[0108] The model is trained using the training dataset, the model weights are updated through the back-propagation algorithm, and in each training cycle, the model performance is evaluated using the validation dataset, and the model parameters are adjusted to optimize the performance.

[0109] The purpose of recursive feature elimination is to find the most influential features by recursively reducing the size of the feature set, including:

[0110] Train a model and get the importance score for each feature, which can be the model weight or other relevant metrics.

[0111] Based on the importance score, remove the features with the lowest score.

[0112] After removing the feature from the feature set, retrain the model and calculate the importance of the remaining features.

[0113] Repeatedly remove and recalculate importance until the desired number of features is reached or a stopping condition is met. In the case of RNNs, feature importance is calculated by observing the W associated with each feature. x The absolute value of the h weight is used to determine the feature. A higher absolute value of the weight means that the feature has a greater impact on the model output.

[0114] S4 specifically includes:

[0115] Based on the identification of a list of biomarkers associated with depression tendency, a reference biomarker set B is constructed;

[0116] For each adolescent sample tested, an abnormal microbiome data set A is extracted from the intestinal microbiome data obtained by high-throughput sequencing, which includes microbial species with abnormally increased or decreased abundance;

[0117] Calculate the overlap R between the reference biomarker set B and the abnormal microbiome data set A for each sample: Where |A∩B| represents the number of elements in the intersection of sets A and B, |B| is the total number of elements in the reference biomarker set, and the overlap R reflects the degree of match between the abnormal microbiome components in the sample and the biomarkers known to be associated with depression;

[0118] A preset threshold T is set. If the overlap R of the sample exceeds the threshold T, the adolescent corresponding to the sample is judged to have potential depressive tendencies; otherwise, the possibility of his or her depressive tendencies is considered to be low.

[0119] S5 specifically includes:

[0120] Integrate the identified biomarkers and behavioral patterns associated with depression proneness to form a comprehensive signature set;

[0121] The comprehensive feature set is processed using data encoding and standardization to ensure that the input data of the model is in a uniform format and scale;

[0122] Multilayer Perceptron (MLP) was selected as the basic architecture of the depression tendency validation model, and the interaction between biomarkers and behavioral patterns was considered by introducing interaction terms into the depression tendency validation model.

[0123] Use a labeled dataset (where the label indicates whether there is a tendency to depression) to train and verify the model. The model training process includes parameter optimization and overfitting control (regularization).

[0124] Model output interpretation: Explain the behavioral patterns corresponding to biomarkers by analyzing the weights and feature contributions in the trained model. For example, the model may reveal that changes in certain microbiome components are highly correlated with social avoidance behavior patterns.

[0125] The model was evaluated on an independent test set to verify its accuracy and generalization ability. Appropriate evaluation indicators, such as accuracy and area under the ROC curve (AUC), were used to measure the effectiveness of the model in predicting adolescent depression tendencies.

[0126] The depression tendency validation model specifically includes:

[0127] Feature engineering: Before the input layer, feature engineering is performed to create interaction terms. There are two sets of features, one of which is biomarkers B = {b 1 , b 2 , ..., b m}, the other group is behavior pattern P = {p 1 , p 2 , ..., p n}, the interaction term is generated by calculating the product of each pair of features between the two sets of features: I ij =b i ×p j , where I ij It is a biomarker b i and behavior patterns j In this way, we get m×n interaction terms.

[0128] Construct an MLP where the input layer receives the original features and interaction terms as input. Suppose the total number of original features and interaction terms is d, then the input layer has d neurons.

[0129] Hidden layer and output layer: One or more hidden layers are designed to process the input data. Each neuron in the hidden layer calculates the weighted sum of all neurons in the previous layer and passes through the ReLU activation function. The output layer contains one neuron for regression problems.

[0130] For the hidden layer l, the output h of each neuron l，k The calculation is as follows:

[0131] Among them, f is the activation function, w l，kj is the weight from the jth neuron in the l-1th layer to the kth neuron in the lth layer, b l，k is the bias term of the kth neuron in the lth layer, Dl-1 is the number of neurons in the l-1th layer.

[0132] Model output explanations include:

[0133] Weight explanation: Each input feature is multiplied by the corresponding weight, and the product reflects the contribution of the feature to the neuron activation value. The larger the absolute value of the weight, the more important the feature is in the model. A positive weight indicates a positive correlation between the feature and the target variable, and a negative weight indicates a negative correlation.

[0134] Feature contribution: The contribution of a feature to the model prediction result is determined by observing the weight of the feature and its propagation in the neural network. If a biomarker feature has a high weight in multiple nodes and these multiple nodes contribute significantly to the final prediction, then the biomarker is a high feature importance biomarker. Subsequently, the relationship between the high feature importance biomarker and the psychological behavior pattern data is analyzed by observing the performance of the biomarker with high feature importance in a certain behavior pattern. If the high expression of a biomarker appears consistently with a certain behavior pattern, it is inferred that the biomarker is associated with the behavior pattern and is determined to have a tendency to depression.

[0135] Suppose we have a biomarker Biomarker_A and a psychological behavioral feature Behavior_Feature_A. Through weighted activation analysis, we find that high weighted activation values ​​of Biomarker_A are closely associated with low (or high) values ​​of Behavior_Feature_A. This analysis suggests that the expression level of Biomarker_A may be related to the specific psychological behavioral pattern represented by Behavior_Feature_A. In this way, we can link specific biomarkers with corresponding psychological behavioral patterns, thereby providing insights into a deeper understanding of adolescent depressive tendencies.

[0136] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention is limited to these examples. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0137] The present invention is intended to cover all such substitutions, modifications and variations that fall within the broad scope of the claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An early warning method for adolescent depression tendency using artificial intelligence, characterized in that: The following steps are involved: S1: Collect microbiome data of adolescents, obtain initial intestinal microbiome data through non-invasive biological sample analysis, and test the initial intestinal microbiome data to obtain abnormal intestinal microbiome data; S2: Analyze adolescents’ psychological and behavioral data, including emotional self-assessment, behavioral questionnaires, and social interaction frequency, and identify behavioral patterns; S3: Use machine learning techniques to conduct in-depth analysis of abnormal microbiome data and identify biomarkers associated with depressive tendencies, including: S31, construct feature matrix: perform quality control on the microbiome data obtained by high-throughput sequencing, including removing low-quality sequences and noise, performing ASVs method on the sequences, and standardizing their abundance data to form an initial feature matrix, where each ASV represents a feature and each ASV abundance constitutes a feature vector; S32, feature extraction: training a recurrent neural network (RNN) model and automatically extracting microbiome patterns from the initial feature matrix through the RNN model; S33, using recursive feature elimination algorithm to screen out the most influential microbial features for predicting depression tendency; S34, validate and test the RNN model, using cross-validation techniques to evaluate the accuracy, sensitivity, and specificity of the model to ensure that the identified biomarkers are statistically relevant to adolescent depression tendencies; S35, based on the results of the RNN model, determine the biomarkers associated with adolescent depression tendency; S4: Compare the overlap between biomarkers and abnormal gut microbiome data. If the overlap exceeds the preset threshold, it is judged as having potential depression tendency. S5: Construct a depression tendency verification model, which takes into account the interactive effects of biomarkers and behavioral patterns. The model outputs the behavioral pattern corresponding to the biomarker, and compares the corresponding behavioral pattern with the behavioral pattern identified by S2 to determine whether there is a depression tendency. Specifically, it includes: Integrate the identified biomarkers and behavioral patterns associated with depression proneness to form a comprehensive signature set; The comprehensive feature set is processed using data encoding and standardization to ensure that the input data of the model is in a uniform format and scale; Multilayer Perceptron (MLP) was selected as the basic architecture of the depression tendency validation model, and the interaction between biomarkers and behavioral patterns was considered by introducing interaction terms into the depression tendency validation model. Use labeled datasets to train and validate the model. The model training process includes parameter optimization and overfitting control. Model output interpretation: explaining the behavior patterns corresponding to biomarkers by analyzing the weights and feature contributions in the trained model; S6: Set a warning threshold. When the depression tendency index exceeds the threshold, the warning mechanism is automatically triggered.

2. The method for early warning of adolescent depression tendency using artificial intelligence according to claim 1 is characterized in that: The S1 specifically includes: S11: Collect biological samples from adolescents using non-invasive sample collection techniques, including saliva, skin swabs, or stool samples; S12: Use high-throughput sequencing technologies, including 16S rRNA sequencing and whole genome sequencing (WGS), to sequence the collected samples. 16S rRNA sequencing focuses on the areas of bacteria and archaea, providing information on the relative abundance of microbial species, and WGS provides information on the composition of microorganisms, including bacteria, viruses, fungi, and archaea. S13: Use bioinformatics methods to process sequencing data, including data quality control, sequence splicing, operational taxonomic unit clustering, and species annotation. Data quality control includes removing low-quality and short sequences to ensure data quality. Sequence splicing assembles sequencing reads into longer sequences. Operational taxonomic unit clustering groups sequences based on sequence similarity to represent specific microbial communities. Species annotation compares operational taxonomic unit clusters with known microbial databases to determine their species identities, analyze the composition and diversity of the intestinal microbiome, and identify microbial communities that deviate from the normal intestinal microbiome, that is, abnormal intestinal microbiome data.

3. The method for early warning of adolescent depression tendency using artificial intelligence according to claim 2 is characterized in that: The S2 specifically includes: S21: Collect adolescents’ self-rating data on their emotions, including asking them to rate their own emotional states through online or paper questionnaires; S22: Collect data on adolescents’ behavioral questionnaires, which include a series of questions designed to assess adolescents’ daily behavioral habits, participation in social activities, interests and hobbies, sleep quality, and learning status; S23: Collect data on the frequency of adolescents' social interactions through social media platforms, school records, or parent feedback, including the frequency of their activities on social media, the number of times they participate in social activities, and the frequency of communication with friends and family; S24: Use principal component analysis (PCA) to comprehensively analyze the collected emotional self-assessment data, behavioral questionnaire data, and social interaction frequency data to identify behavioral patterns, including social avoidance behavior, emotional fluctuation patterns, and lack of interest.

4. The method for early warning of adolescent depression tendency using artificial intelligence according to claim 1 is characterized in that: The RNN model uses internal hidden states to capture the dependencies between time points in a sequence. For a given sequence , the RNN model gradually updates its hidden state through the following calculation steps : ,in, is at the time step The hidden state of is at the time step Input, is the weight matrix input to the hidden layer, is the weight matrix from hidden layer to hidden layer, is the bias term of the hidden layer, is the activation function; Output calculation: ； ； in, is the output layer at time step The original output of is the weight matrix from the hidden layer to the output layer, is the bias term of the output layer, is at the time step The prediction output of , the softmax function is used for the probability output of multi-classification problems.

5. The method for early warning of adolescent depression tendency using artificial intelligence according to claim 4 is characterized in that: The RNN model selects the cross entropy loss function and the Adam optimizer for training optimization. The loss function is used to evaluate the difference between the model prediction and the true label, and the optimizer is used to adjust the model parameters to minimize the loss function. The cross entropy loss function is expressed as: ,in, It is a sample The real label, is the probability predicted by the model; The model is trained using the training dataset, the model weights are updated through the back-propagation algorithm, and in each training cycle, the model performance is evaluated using the validation dataset, and the model parameters are adjusted to optimize the performance.

6. The method for early warning of adolescent depression tendency using artificial intelligence according to claim 5 is characterized in that: The S4 specifically includes: Construct a reference biomarker panel based on identifying a list of biomarkers associated with depression proneness ; For each adolescent sample tested, an abnormal microbiome data set was extracted from the intestinal microbiome data obtained from high-throughput sequencing. , which include microbial species with abnormally increased or decreased abundance; Computation of reference biomarker panels Anomalous microbiome data sets for each sample The overlap between : ,in, Representing a collection and The number of elements in the intersection of is the total number of elements in the reference biomarker set, the overlap reflects the degree of match between abnormal microbiome components in the sample and biomarkers known to be associated with depression; Set a preset threshold , if the sample overlap Exceeding this threshold , then the adolescents corresponding to the sample are judged to have potential depressive tendencies; otherwise, the possibility of their depressive tendencies is considered to be low.

7. The method for early warning of adolescent depression tendency using artificial intelligence according to claim 1 is characterized in that: The depression tendency verification model specifically includes: Feature Engineering: Before the input layer, feature engineering is performed to create interaction terms. There are two sets of features, one of which is biomarkers. , the other group is behavioral patterns , the interaction term is generated by calculating the product of each pair of features between the two sets of features: ,in, It is a biomarker and behavior patterns In this way, we get interaction terms; Construct an MLP where the input layer receives the original features and interaction terms as input. Let the total number of original features and interaction terms be , then the input layer has neurons; Hidden layer and output layer: One or more hidden layers are designed to process the input data. Each neuron in the hidden layer calculates the weighted sum of all neurons in the previous layer and passes through the ReLU activation function. The output layer contains one neuron for regression problems. For the hidden layer , the output of each neuron The calculation is as follows: ,in, is the activation function, It is from Layer Neuron to Layer The weight of a neuron, It is Layer The bias term of a neuron, It is The number of neurons in the layer.

8. The method for early warning of adolescent depression tendency using artificial intelligence according to claim 7 is characterized in that: The model output explanation specifically includes: Weight explanation: Each input feature is multiplied by the corresponding weight, and the product reflects the contribution of the feature to the neuron activation value. The larger the absolute value of the weight, the more important the feature is in the model. A positive weight indicates a positive correlation between the feature and the target variable, and a negative weight indicates a negative correlation. Feature contribution: The contribution of a feature to the model prediction result is determined by observing the weight of the feature and its propagation in the neural network. If a biomarker feature has a high weight in multiple nodes and these multiple nodes contribute significantly to the final prediction, then the biomarker is a high feature importance biomarker. Subsequently, the relationship between the high feature importance biomarker and the psychological behavior pattern data is analyzed by observing the performance of the biomarker with high feature importance in a certain behavior pattern. If the high expression of a biomarker appears consistently with a certain behavior pattern, it is inferred that the biomarker is associated with the behavior pattern and is determined to have a tendency to depression.

Images