Pet health monitoring system based on microbiome gene detection

Abstract

The invention relates to the technical field of pet health monitoring, and discloses a pet health monitoring system based on microbiome gene detection. The system comprises a data acquisition and fluidization module, a cross-level association and preliminary screening module, a dynamic baseline modeling module, an anomaly comparison and background analysis module and a deep threat tracing module. According to the system, a pet gene sequence and metadata are obtained by formulating an acquisition scheme, and a real-time data stream is formed; carrying out cross-level correlation analysis, identifying an abnormal interaction mode and generating an abnormal mark data set with a weight; constructing and dynamically calibrating a personalized baseline model by using the data set and the data stream; comparing real-time data with a baseline to find deviation, and analyzing an abnormal background in combination with metadata; and finally, based on an anomaly identification result, analyzing metadata to trace the deep health threat mode and the root cause. According to the invention, continuous dynamic monitoring and multi-level correlation analysis of the pet microbiome are realized, and the accuracy of early health risk early warning is improved.

Description

Technical Field

[0001] This invention relates to the field of pet health monitoring technology, specifically a pet health monitoring system based on microbiome gene detection. Background Technology

[0002] Current pet health monitoring primarily relies on in vitro symptom observation and regular biochemical tests. While microbiome testing technology has been applied in the pet field, it is mostly limited to single, static analyses of microbial composition or screening for specific pathogens. Existing technologies typically treat gene sequence data as isolated, static snapshots, judging abnormalities by comparing them to fixed databases or universal health standard thresholds. These methods have significant drawbacks. Static analysis cannot capture the dynamic trajectory of an individual pet's microbiome as it changes with age, diet, and environment, making it difficult to detect subtle early abnormalities deviating from an individual's health baseline in a timely manner. Furthermore, existing methods often focus on single taxonomic levels (such as species abundance), neglecting the inherent networked connections and interactions between multi-level data such as gene function and metabolic pathways. This leads to insensitivity to identifying potential health threats caused by imbalances in complex ecological relationships and makes source tracing difficult. Therefore, a technological solution is needed that can continuously and dynamically monitor the pet microbiome, penetrate data hierarchy barriers, and analyze its internal correlation patterns, thereby achieving earlier and more accurate health risk warnings and root cause diagnosis. Summary of the Invention

[0003] The purpose of this invention is to provide a pet health monitoring system based on microbiome gene detection to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a pet health monitoring system based on microbiome gene detection, the system comprising:

[0005] The data acquisition and fluidization module is used to determine the microbiome gene data acquisition plan for pet health monitoring, acquire gene sequence data and sample metadata from pet biological samples according to the plan, and concatenate the acquired data into a real-time gene data stream in chronological order.

[0006] The cross-level correlation and initial screening module is used to conduct cross-level correlation analysis based on the hierarchical logic of real-time gene data streams. It combines the context of the data generation scenario to sort out the correlations at different levels, find the correlations and abnormal interaction patterns of gene data between levels, mark the found results as preliminary abnormal signals, and aggregate them into a weighted abnormal label dataset.

[0007] The dynamic baseline modeling module is used to build a dynamic baseline model based on real-time gene data streams and anomaly marker datasets, calibrate the baseline model using the results of cross-level correlation analysis, and dynamically adjust parameters as the data is updated.

[0008] The anomaly comparison and background analysis module is used to compare real-time collected data with the dynamic baseline model, identify anomalous activities that deviate, mark anomalous behaviors that deviate, and submit the deviation identification results to sample metadata analysis to obtain anomalous background information.

[0009] The Deep Threat Tracing Module is used to analyze sample metadata based on the anomaly identification results of the dynamic baseline model, identify potential deep-seated health threat patterns and abnormal signals, trace the threat patterns and abnormal signals to determine the root cause and abnormal path.

[0010] Preferably, the method for determining the microbiome gene data collection for pet health monitoring involves acquiring gene sequence data and sample metadata from pet biological samples according to the method, and concatenating the acquired data into a real-time gene data stream in chronological order, specifically including:

[0011] First, based on the pet's age, daily diet, and recent health status, determine whether the biological sample to be collected is a fecal sample, oral sample, or skin sample. At the same time, specify the genetic data capture requirements for each type of sample, including whole-genome sequencing depth, functional gene targeted capture range, and sample metadata must include the collection time, collection site, and the pet's recent diet record.

[0012] Gene data capture is performed on real-time collected biological samples according to specific requirements, including extracting gut microbial genes from fecal samples, oral microbial genes from oral samples, and skin microbial genes from skin samples.

[0013] The captured gene data is first classified according to the sample source, then the species classification information is sorted out by microbial species layer, the gene function information is sorted out by functional gene layer, and the sequence base arrangement information is sorted out by gene sequence layer. The gene sequence data and sample metadata including collection time, collection site, pet's recent diet record, and pet breed are then parsed out.

[0014] The parsed gene sequence data are concatenated according to the collection time, and the sample metadata of the corresponding time point is attached to the gene sequence data concatenated at the same time to form a real-time gene data stream.

[0015] Preferably, the method of conducting cross-level correlation analysis based on the hierarchical logic of real-time gene data streams, combining the context of data generation to sort out the associations at different levels, identifying the associations and abnormal interaction patterns of gene data between levels, marking the identified results as preliminary abnormal signals, and compiling them into a weighted abnormal label dataset, specifically includes:

[0016] First, following the generation logic of real-time gene data stream, the gene sequence data is divided into continuous time slices according to the collection time. Within each time slice, the data is organized into species layer, functional gene layer, and gene sequence layer. From the data of each time slice, the environmental conditions of the collection site, the types of components in the pet's diet, and the types of items recently encountered are extracted as contextual information.

[0017] For the species layer and functional gene layer within each time slice, the correlation between the abundance of a certain type of microbial species and the expression abundance of the corresponding functional gene is analyzed. The correlation is then combined with the type of components in the pet diet of that time slice to determine whether it conforms to the normal correspondence logic between species and functional genes under that component. If it does not conform, it is recorded as an abnormal species-function correlation. For the functional gene layer and gene sequence layer, the correlation between the sequence characteristics of a certain functional gene and the gene expression abundance is analyzed. The correlation is then combined with the environmental conditions of the sampling site of that time slice to determine whether it conforms to the normal correspondence logic between functional genes and sequence characteristics under that environment. If it does not conform, it is recorded as an abnormal gene sequence correlation.

[0018] Species function association anomalies and gene sequence association anomalies are marked as preliminary anomalous signals. The anomalous type of each preliminary anomalous signal is recorded as either species function association anomaly or gene sequence association anomaly, the frequency is the number of times the anomaly occurs within the time slice, and the severity is the degree to which the association deviates from normal logic.

[0019] Weights are assigned based on the logical deviation difficulty corresponding to the anomaly type, the frequency of occurrence corresponding to the frequency, and the deviation depth corresponding to the severity of each preliminary anomaly signal. The weighted preliminary anomaly signals are then integrated with the scene context information of the corresponding time slice to form a weighted anomaly label dataset.

[0020] Preferably, the step of constructing a dynamic baseline model based on real-time gene data streams and anomaly marker datasets, calibrating the baseline model using the results of cross-level correlation analysis, and dynamically adjusting parameters as data is updated specifically includes:

[0021] First, extract the species-level microbial species abundance distribution, functional gene expression abundance distribution, and gene sequence feature distribution of each time slice from the real-time gene data stream. Then, extract the preliminary anomalous signal weights and scene context information of each time slice from the weighted anomaly label dataset.

[0022] Using the extracted species layer abundance distribution, functional gene layer expression abundance distribution, and gene sequence layer feature distribution as basic features, and the weights of the abnormal signals in the weighted abnormal label dataset as correction features, an initial dynamic baseline model is constructed. The basic features are used to train the model to learn the stable features of each layer distribution under normal conditions, and the correction features are used to train the model to identify the boundary between normal and abnormal distributions.

[0023] For each new time-slice of real-time gene data stream and corresponding weighted anomaly marker dataset, the basic features of the time-slice are first input into the initial dynamic baseline model to obtain the basic predicted values ​​of the distribution of each layer of the time-slice. Then, the anomaly signal weights in the weighted anomaly marker dataset of the time-slice are used to adjust the basic predicted values. The adjusted results are used as the parameters of the baseline model of the time-slice. The overall parameters of the dynamic baseline model are updated with the parameters of the new time-slice. At the same time, the threshold for the model to identify the boundary between normal and abnormal is adjusted according to the distribution of the anomaly signal weights of the new time-slice, so as to achieve dynamic parameter tuning as the data is updated.

[0024] Preferably, the step of comparing the real-time collected data with the dynamic baseline model to identify abnormal activities and mark the abnormal behaviors, and then submitting the deviation identification results to sample metadata analysis to obtain abnormal background information, specifically includes:

[0025] The real-time pet gene data is divided into time slices consistent with the dynamic baseline model according to the collection time. The species layer abundance, functional gene layer expression abundance, and gene sequence layer features in the gene sequence data of each time slice are respectively input into the dynamic baseline model. The model outputs the normal range of the distribution of each layer in that time slice.

[0026] Compare the distribution of each layer of real-time data with the normal range of the model output. If the abundance of species layer, the expression abundance of functional gene layer, or the characteristics of gene sequence layer in a certain time slice exceed the normal range, it is judged as a potential abnormal activity.

[0027] For potential abnormal activity markers that deviate from the marked abnormal behavior, the degree of deviation of the abnormal behavior is recorded as the difference between the actual distribution and the normal range, and the relevant context information is the scene context information of that time slice;

[0028] The severity of the deviation is assessed based on the degree of deviation and the frequency of the abnormal activity. The greater the degree of deviation and the higher the frequency, the higher the abnormality level. Each abnormal activity is assigned an abnormality level.

[0029] Extract the collection time, collection location, and recent pet diet records from the relevant contextual information of the abnormal behavior. Use the extracted information to retrieve sample metadata and read the abnormal background information, such as the environmental conditions at the time of sample collection, the specific components of the pet's diet, and the specific types of items the pet has recently come into contact with.

[0030] Preferably, the analysis of sample metadata based on the anomaly identification results using the dynamic baseline model to identify potential deep-seated health threat patterns and abnormal signals, tracing the threat patterns and abnormal signals to determine the root cause and abnormal path, specifically includes:

[0031] First, collect sample metadata that matches the abnormal background information in the deviation identification results. The sample metadata includes the collection environment status of the time slice corresponding to the abnormal behavior, the specific ingredients of the pet's diet, the types of items recently contacted, and the pet's historical health records. Obtain complete scene information when the abnormal behavior occurred from the sample metadata.

[0032] Using complete scene information as an analytical framework, we conduct correlation analysis on the collection environment status, specific dietary components, contact item types, and historical health records in the sample metadata to identify abnormal dietary components, environmental conditions, or contact items that occur simultaneously with abnormal behavior. By combining the abnormal hierarchical correlation in the genetic data, we identify potential deep-seated health threat patterns as species functional association anomalies caused by dietary components, gene sequence association anomalies caused by environmental conditions, or cross-layer association anomalies caused by contact items. At the same time, we identify the abnormal signals as the specific data manifestations corresponding to these threat patterns.

[0033] Health cues related to anomalous signals are extracted from sample metadata. These health cues are dietary, environmental, or contact items that occurred when similar anomalies appeared in historical health records. Threat types are identified as digestive health threats, skin health threats, or immune health threats based on the matching degree between health cues and current anomalous signals.

[0034] For the identified abnormal signals, trace back from the current time slice to the historical time slices. First, find the direct source of the abnormal signal in the current time slice, which is the intake of a certain type of food component, a change in a certain type of environmental state, or contact with a certain type of contact item in the previous time slice. Then continue to trace back to the source in the previous time slice until the time slice that initially caused the abnormality and the corresponding factor are found, and the initial source of the abnormality is determined.

[0035] The report integrates the threat patterns, threat types, and abnormal paths corresponding to the abnormal signals with their original sources to generate a combined report.

[0036] Preferably, the step of concatenating the parsed gene sequence data according to the collection time, and attaching the sample metadata of the corresponding time point to the concatenated gene sequence data to form a real-time gene data stream, specifically includes:

[0037] The parsed gene sequence data is divided into time units based on the hour of collection time. The gene sequence data within each time unit are concatenated from the earliest to the latest collection time. The metadata of the samples collected within the same time unit is appended to the end of the concatenated gene sequence data in that time unit according to the order of collection time. If samples from multiple sites are collected at the same time point, the corresponding metadata is appended according to the site priority, forming a real-time gene data stream concatenated by time units with metadata appended.

[0038] Preferably, the extraction of contextual information from each time slice of data, including the environmental conditions of the collection site, the types of ingredients in the pet's diet, and the types of items recently encountered, specifically includes:

[0039] The environmental temperature and humidity records and cleanliness records of the collection site at the corresponding collection time point are retrieved from the sample metadata to determine the environmental status. The protein source, carbohydrate source, and additive type in the pet's diet in the 24 hours prior to the time segment are retrieved to determine the dietary component type. The toy material, bedding cleanliness, and type of external plants that the pet came into contact with in the 72 hours prior to the time segment are retrieved to determine the recently contacted item type.

[0040] Preferably, the step of updating the overall parameters of the dynamic baseline model with the parameters of the new time slice, and simultaneously adjusting the threshold for identifying the normal and abnormal boundaries of the model based on the distribution of the abnormal signal weights in the new time slice, specifically includes:

[0041] The parameters of the new time slice include the basic predicted values ​​and corrected parameters of each layer distribution of the time slice. The basic predicted values ​​of the new time slice are added to the historical basic feature library of the model. The average value of the historical basic feature library is used to update the basic weights of each layer distribution of the model. The corrected parameters of the new time slice are added to the historical corrected feature library of the model. The median of the historical corrected feature library is used to update the weights of the corrected features of the model.

[0042] The distribution of abnormal signal weights in the new time slice is statistically analyzed. If the proportion of high-weight abnormal signals increases, the threshold for the model to identify the boundary between normal and abnormal signals is increased, making the model more strict in identifying abnormalities. If the proportion of low-weight abnormal signals increases, the threshold is decreased, making the model more lenient in identifying abnormalities. This achieves the adjustment of the boundary threshold according to the distribution of abnormal signal weights.

[0043] Preferably, the step of retrieving sample metadata using the extracted information and reading abnormal background information such as the environmental conditions at the time of sample collection, the specific components of the pet's diet, and the specific types of items recently contacted, specifically includes:

[0044] The environmental temperature and humidity records and the cleanliness records of the collection site at the same collection time point are retrieved from the sample metadata using the extracted collection time as the environmental status. The dietary composition records of the corresponding collection time for that location are retrieved from the sample metadata using the extracted collection site. The specific dietary components for that time range are retrieved from the sample metadata using the time range of the extracted recent dietary records. The contact item type records for that time range are retrieved from the sample metadata using the range of the seventy-two hours prior to the extracted collection time.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] By conducting cross-level correlation analysis based on the hierarchical logic of real-time gene data streams, and combining this with contextual analysis to clarify relationships between different levels, this approach identifies and initially labels abnormal interaction patterns in gene data across levels. This method overcomes the limitations of traditional single-level analysis. It can identify potential risk states that appear normal at a single level but have disrupted cross-level synergistic relationships, such as situations where species composition has not changed significantly but key metabolic pathway activity has been dysregulated. This network-based analysis method can capture complex, early-stage abnormal signals missed by traditional threshold comparison methods, significantly improving the sensitivity and specificity of health risk warnings and enabling earlier detection of hidden, systemic health problems.

[0047] By constructing a dynamic baseline model based on real-time data streams and anomaly-labeled datasets generated from initial screening, and using cross-level analysis results for calibration and dynamic parameter tuning, this approach overcomes the shortcomings of static or population-wide baselines, which are unsuitable for long-term individual monitoring. The dynamic baseline model adaptively learns and tracks the normal fluctuation range of an individual pet's microbiome over time, internalizing its unique physiological state, lifestyle habits, and other factors into model parameters. This personalized dynamic modeling approach effectively distinguishes between normal individual fluctuations and true pathological deviations, significantly reducing false alarms caused by individual differences or physiological changes. This makes health status assessment more accurate and reliable, providing a truly personalized health monitoring benchmark for each pet. Attached Figure Description

[0048] Figure 1 This is a timeline diagram of the pet health monitoring system based on microbiome gene detection described in this invention;

[0049] Figure 2 Flowchart for the dynamic baseline modeling module;

[0050] Figure 3 This is a flowchart for the deep threat attribution module. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1This invention provides a pet health monitoring system based on microbiome gene detection. The system includes: a data acquisition and fluidization module, a cross-level correlation and initial screening module, a dynamic baseline modeling module, an anomaly comparison and background analysis module, and a deep threat tracing module. The data acquisition and fluidization module determines the microbiome gene data acquisition plan for pet health monitoring, acquires gene sequence data and sample metadata from pet biological samples according to the plan, and concatenates the acquired data into a real-time gene data stream in chronological order. The cross-level correlation and initial screening module conducts cross-level correlation analysis based on the hierarchical logic of the real-time gene data stream, analyzes the correlations at different levels in conjunction with the context of data generation, identifies the correlations and anomaly interaction patterns between gene data levels, marks the identified results as preliminary anomaly signals, and aggregates them into a weighted anomaly label dataset. The dynamic baseline modeling module constructs a dynamic baseline model based on the real-time gene data stream and the anomaly label dataset, calibrates the baseline model using the results of cross-level correlation analysis, and dynamically adjusts parameters as data is updated. The anomaly comparison and background analysis module compares real-time collected data with the dynamic baseline model to identify anomalous activities and behaviors, and then analyzes the sample metadata to obtain anomaly background information. The deep threat tracing module analyzes sample metadata based on the anomaly identification results from the dynamic baseline model to identify potential deep-seated health threat patterns and anomalous signals. It then traces these threat patterns and anomalous signals to determine the root cause and anomalous pathways.

[0053] In one embodiment of the invention, the required biological samples are determined by considering the pet's age, daily diet, and recent health status: fecal samples, oral samples, or skin samples. The genetic data capture requirements for each sample type are specified as whole-genome sequencing depth, functional gene targeting capture range, and mandatory inclusion of sample metadata including collection time, collection site, and the pet's recent dietary records. Genetic data capture is performed on the real-time collected biological samples according to these specified requirements. Gut microbial genes are extracted from fecal samples, oral microbial genes from oral samples, and skin microbial genes from skin samples. The captured genetic data is first classified by sample origin, then sorted by microbial species level, gene function level, and gene sequence base arrangement level to extract the gene sequence data and sample metadata including collection time, collection site, the pet's recent dietary records, and the pet breed. The parsed gene sequence data is divided into time units based on the hour of collection time. The gene sequence data within each time unit are concatenated from the earliest to the latest collection time. The metadata of the samples collected within the same time unit is appended to the end of the concatenated gene sequence data in that time unit according to the order of collection time. If samples from multiple sites are collected at the same time point, the corresponding metadata is appended according to the site priority, forming a real-time gene data stream concatenated by time units with metadata appended.

[0054] In practice, the types of biological samples to be collected and the requirements for capturing genetic data are determined by considering the pet's age, daily diet, and recent health status. For example, for a three-year-old Labrador Retriever whose daily diet is commercial dry food and whose recent health status is characterized by occasional soft stools, the biological sample to be collected must be a fecal sample. At the same time, the requirements for capturing genetic data for this fecal sample must be that shotgun metagenomic sequencing is used to achieve a sequencing depth of 10x, the functional gene targeting capture range must cover carbohydrate active enzyme genes and antibiotic resistance genes, and the sample metadata must include the collection time, the collection site "rectum", and the pet's diet record within the last 72 hours.

[0055] In some embodiments, gene data capture is performed on biological samples collected in real time according to specific requirements. For the fecal samples of Labrador Retrievers mentioned above, total deoxyribonucleic acid is extracted from the samples using a nucleic acid extraction kit. Then, a sequencing library is prepared using a library preparation kit and sequenced on a sequencing platform to extract intestinal microbial gene sequence data from the fecal samples. This process is also applicable to extracting oral microbial genes from oral swab samples or skin microbial genes from skin swab samples.

[0056] In practice, the captured gene sequence data is first classified according to the sample source, such as biological samples from feces, oral cavity, and skin. Then, the species classification information is sorted out according to the microbial species level. This process involves comparing the sequencing reads with the reference microbial genome database and calculating the relative abundance. Gene function information is sorted out according to the functional gene level. This process involves annotating the reads into a functional gene database such as KEGG. Sequence base arrangement information is sorted out according to the gene sequence level. This process involves extracting single nucleotide polymorphism features. Once the analysis is completed, structured gene sequence data and sample metadata including collection time, collection site, recent pet diet records, and pet breed are obtained.

[0057] It is understandable that concatenating the parsed gene sequence data according to the collection time requires setting specific time unit division rules. For example, time units can be divided by the hour of collection time. All sample data collected between 2 pm and 3 pm belong to the same time unit. The gene sequence data in each time unit are concatenated from morning to night according to the collection time. If samples from multiple sites, such as feces and oral cavity, are collected at the same time point, the metadata of the feces sample is arranged before the metadata of the oral cavity sample according to the preset rule that the gut microbiota has a higher priority than the oral microbiota. Finally, the metadata of all samples collected in the same time unit is appended to the end of the concatenated gene sequence data in that time unit in this order.

[0058] In some embodiments, for the concatenation process, when samples from different parts of the same pet are acquired at the same acquisition time point, such as skin samples and oral samples are acquired at the same time, the site priority rule stipulates that the skin sample metadata is attached before the oral sample metadata. Then the structure of the resulting real-time gene data stream fragment is: time unit identifier, gene sequence data block sorted by time, followed by skin sample metadata block, and finally oral sample metadata block.

[0059] Optionally, the granularity of the time unit can be adjusted according to the monitoring frequency. For high-frequency monitoring scenarios, the time unit can be set to the minute level. In this case, the gene sequence data is concatenated in the order of collection within each minute, and the corresponding sample metadata is also appended at the minute level, which forms a real-time gene data stream with higher time resolution.

[0060] In practice, a complete real-time gene data stream consists of continuous time-series data segments. Each data segment has a consistent internal structure and follows the format of first gene sequence data and then corresponding sample metadata. This structure ensures the temporal order of the data and the complete attachment of contextual information, providing a time-indexed basis for subsequent cross-level analysis.

[0061] In one embodiment of the present invention, following the generation logic of real-time gene data streams, gene sequence data is divided into continuous time slices according to the collection time. Within each time slice, data is organized into species layer, functional gene layer, and gene sequence layer. Contextual information such as the environmental state of the collection site, the types of ingredients in the pet's diet, and the types of items the pet has recently come into contact with are extracted from the data of each time slice. The environmental state is determined by searching the sample metadata for the environmental temperature and humidity records and cleanliness records of the collection site at the corresponding collection time point of the time slice. The protein source, carbohydrate source, and additive type in the pet's diet in the 24 hours prior to the time slice are determined as the dietary ingredient type. The types of items the pet has recently come into contact with are determined by searching the toy material, bedding cleanliness, and types of external plants the pet has come into contact with in the 72 hours prior to the time slice. For each time slice, the abundance of a certain microbial species and the expression abundance of its corresponding functional gene are analyzed at the species and functional gene layers. The correlation is then assessed based on the component types in the pet diet of that time slice to determine if it conforms to the normal correspondence between species and functional genes under that component. If not, it is recorded as an abnormal species-functional association. For the functional gene and gene sequence layers, the correlation between the sequence characteristics of a functional gene and its gene expression abundance is analyzed. The correlation is then assessed based on the environmental conditions of the sampling site of that time slice to determine if it conforms to the normal correspondence between functional genes and sequence characteristics under that environment. If not, it is recorded as an abnormal gene-sequence association. Species-functional association anomalies and gene-sequence association anomalies are marked as preliminary anomaly signals. The anomaly type (species-functional association anomaly or gene-sequence association anomaly), frequency (number of times the anomaly occurs within the time slice), and severity (degree of deviation from normal logic) of each preliminary anomaly signal are recorded. Weights are assigned based on the logical deviation difficulty corresponding to the anomaly type, the frequency (occurrence frequency), and the deviation depth (deepness of deviation) corresponding to the severity of each preliminary anomaly signal. These weighted preliminary anomaly signals are then integrated with the scene context information of the corresponding time slice to form a weighted anomaly-labeled dataset.

[0062] In practice, gene sequence data is divided into continuous time slices according to the collection time, based on the generation logic of real-time gene data stream. For example, a natural day is considered as a time slice, and all gene sequence data collected on that day are grouped into the same time slice for processing. Within each time slice, the data is organized into species layer, functional gene layer, and gene sequence layer. Species layer data is represented by a list of microbial species and their abundance, functional gene layer data is represented by a list of functional gene classifications and their expression abundance, and gene sequence layer data is represented by a list of characteristic single nucleotide polymorphism sites and their frequencies.

[0063] In some embodiments, contextual information such as the environmental state of the sampling site, the types of ingredients in the pet's diet, and the types of items recently encountered are extracted from each time slice data. This extraction operation specifically involves: searching the sample metadata for the environmental temperature and humidity records of the sampling site at the corresponding sampling time point of the time slice, for example, the environmental temperature of 25 degrees Celsius and the relative humidity of 60% when sampling the anus; searching for the cleanliness records of the sampling site, for example, the record for the skin sampling site "back" is "clean and dry", which together constitute the environmental state; searching for the protein sources in the pet's diet in the 24 hours prior to the time slice, such as "chicken", the carbohydrate sources, such as "rice", and the types of additives, such as "prebiotic fructooligosaccharides", which constitute the types of ingredients in the diet; searching for the toy materials that the pet has been in contact with in the 72 hours prior to the time slice, such as "rubber", the cleanliness of the bedding, such as "not cleaned within a week", and the types of external plants, such as "pothos", which constitute the types of items recently encountered.

[0064] In practice, for each time slice, the species layer and functional gene layer are analyzed to determine the correlation between the abundance of a certain type of microbial species and the expression abundance of the corresponding functional gene. For example, the Pearson correlation coefficient between the abundance of Firmicutes and the expression abundance of amylase gene in a time slice is analyzed. Combined with the type of pet diet in that time slice, it is judged whether the correlation conforms to the normal correspondence logic between species and functional genes under that component. If the carbohydrate source in the diet is "rice", the normal correspondence logic expects that the abundance of Firmicutes and the expression of amylase gene should be positively correlated. However, if the actual calculated correlation coefficient is significantly negative, this situation is recorded as an abnormal species-functional association.

[0065] It is understandable that analyses are performed at the functional gene layer and gene sequence layer to determine the association between the sequence characteristics of a functional gene and the gene expression abundance. For example, the association between the allele frequency of the single nucleotide polymorphism site "SNP_A234T" of the β-lactamase gene and the gene expression abundance is analyzed within a time slice. The environmental conditions of the sampling site in that time slice are considered to determine whether the association conforms to the normal correspondence logic between functional genes and sequence characteristics under that environment. If the environmental conditions of the sampling site are "clean and dry", the normal correspondence logic expects that the frequency of the mutation site and the gene expression abundance will not be strongly associated. However, the actual analysis shows that there is a strong positive correlation. This situation is recorded as an abnormal gene sequence association.

[0066] Optionally, when marking species function association anomalies and gene sequence association anomalies as preliminary anomalous signals, it is necessary to record the attributes of each preliminary anomalous signal, record the anomalous type as "species function association anomaly" or "gene sequence association anomaly", record the frequency as the number of times the anomaly occurs in the corresponding time slice. For example, if three independent species function association anomalies are detected in a time slice, record the severity as the degree to which the association deviates from the normal logic. This degree can be quantified by calculating the absolute difference between the actual correlation coefficient and the expected correlation coefficient.

[0067] In some embodiments, weights are assigned based on the logical deviation difficulty corresponding to the anomaly type, the frequency of occurrence corresponding to the frequency, and the deviation depth corresponding to the severity of each preliminary anomaly signal. The assignment process can be calculated according to the following formula:

[0068]

[0069] in, This represents the final weight assigned to the initial anomaly signal. The difficulty level of the logical deviation is represented by a value predefined by the anomaly type, such as "Species Functional Association Anomaly". The value is higher than "gene sequence association disorder". This represents the frequency of an anomaly occurring within the time slice. This represents the calculated severity quantification value. , , Using a preset adjustment factor, the calculated weighted preliminary anomaly signal is integrated with the extracted scene context information of the corresponding time slice to form a weighted anomaly label dataset.

[0070] See Figure 2In one embodiment of the present invention, the species-level microbial species abundance distribution, the functional gene expression abundance distribution, and the gene sequence feature distribution for each time slice are extracted from the real-time gene data stream. Preliminary anomalous signal weights and scene context information for each time slice are extracted from a weighted anomaly marker dataset. Using the extracted species-level abundance distribution, functional gene expression abundance distribution, and gene sequence feature distribution as basic features, and the anomalous signal weights from the weighted anomaly marker dataset as correction features, an initial dynamic baseline model is constructed. The basic features are used to train the model to learn the stable features of each layer's distribution under normal conditions, and the correction features are used to train the model to identify the boundary between normal and anomalous distributions. For each new time-slice of real-time gene data and its corresponding weighted anomaly marker dataset, the basic features of that time-slice are first input into the initial dynamic baseline model to obtain the basic predicted values ​​of the distribution at each layer of that time-slice. Then, the basic predicted values ​​are adjusted using the anomaly signal weights from the weighted anomaly marker dataset of that time-slice. The adjusted results are used as the parameters of the baseline model for that time-slice. The overall parameters of the dynamic baseline model are updated using the parameters of the new time-slice. Simultaneously, the threshold for identifying the normal and anomaly boundaries is adjusted based on the distribution of the anomaly signal weights in the new time-slice, achieving dynamic parameter tuning as data updates. The parameters of the new time-slice include the basic predicted values ​​and corrected parameters of the distribution at each layer of that time-slice. The basic predicted values ​​of the new time-slice are added to the model's historical basic feature library, and the average value of the historical basic feature library is used to update the basic weights of the distribution at each layer of the model. The corrected parameters of the new time-slice are added to the model's historical corrected feature library, and the median of the historical corrected feature library is used to update the weights of the model's corrected features. The distribution of abnormal signal weights in the new time slice is statistically analyzed. If the proportion of high-weight abnormal signals increases, the threshold for the model to identify the boundary between normal and abnormal signals is increased, making the model more strict in identifying abnormalities. If the proportion of low-weight abnormal signals increases, the threshold is decreased, making the model more lenient in identifying abnormalities. This achieves the adjustment of the boundary threshold according to the distribution of abnormal signal weights.

[0071] In practice, the species abundance distribution of microorganisms, the functional gene expression abundance distribution of functional genes, and the gene sequence feature distribution of gene sequences are extracted from the real-time gene data stream for each time slice. The extraction operation is performed on a time slice basis. For example, for a cycle with a seven-day time slice, the species abundance vector, functional gene expression vector, and gene sequence feature vector of each day are extracted from the gene sequence data from the first to the seventh day. At the same time, the preliminary anomalous signal weights and scene context information of the corresponding time slice are extracted from the weighted anomaly label dataset. The preliminary anomalous signal weights are stored in the form of a list, and the scene context information includes the dietary components and environmental conditions recorded each day.

[0072] In some embodiments, the extracted species layer abundance distribution, functional gene layer expression abundance distribution, and gene sequence layer feature distribution are used as basic features, and the weights of the abnormal signals in the weighted abnormal label dataset are used as correction features to construct an initial dynamic baseline model. The initial dynamic baseline model adopts a multivariate state estimation framework, uses the basic feature matrix of historical normal time slices to train the model and generate an initial estimation matrix, and the model simultaneously learns an initial abnormal signal weight threshold to distinguish between normal and abnormal distributions.

[0073] In practice, for each new time-slice real-time gene data stream and corresponding weighted anomaly marker dataset, such as the dataset for the eighth day, the basic features of the eighth day time-slice are first input into the initial dynamic baseline model. The initial dynamic baseline model calculates the basic predicted values ​​of the distribution of each layer on the eighth day based on the historical basic feature matrix. The basic predicted values ​​include the predicted species abundance vector, functional gene expression vector, and gene sequence feature vector. Then, the basic predicted values ​​are adjusted using the anomaly signal weights in the weighted anomaly marker dataset of the eighth day time-slice. The adjustment method is to perform a weighted fusion of the basic predicted vector and the anomaly signal weight vector, and use the adjusted result as the parameter of the baseline model for the eighth day time-slice.

[0074] It is understandable that updating the overall parameters of the dynamic baseline model with the parameters of the new time slice involves adding the basic predicted value of the eighth-day time slice to the model's historical basic feature library. The historical basic feature library stores the basic feature vectors of the past seven days. The average value of all vectors in the historical basic feature library on each feature dimension is used to update the basic weights of the distribution of each layer of the dynamic baseline model. At the same time, the corrected parameters of the eighth-day time slice are added to the model's historical corrected feature library. The median of all corrected parameter values ​​in the historical corrected feature library is used to update the weights of the model's corrected features. This achieves the iterative evolution of the model parameters.

[0075] Optionally, the threshold for identifying the boundary between normal and abnormal signals in the model can be adjusted based on the distribution of abnormal signal weights in the new time slice. The distribution of weights of all preliminary abnormal signals in the eighth-day time slice can be statistically analyzed, and the proportion of high-weight signals to the total number of signals can be calculated. If this proportion increases by 10% compared to the previous time slice, the anomaly determination threshold set in the dynamic baseline model can be increased, so that the model needs a greater deviation to determine the state as abnormal. If this proportion decreases, the threshold can be decreased accordingly. The logic of dynamic parameter tuning can be expressed by the following formula:

[0076]

[0077] in, This represents the adjusted anomaly detection threshold for the model. This represents the threshold before adjustment. This represents the proportion of high-weighted anomalous signals in the new time slice (day eight). This represents the proportion of high-weighted anomalous signals in the previous time slice (day 7). Adjust the sensitivity coefficient to the preset threshold.

[0078] In some embodiments, the updating of the overall parameters of the dynamic baseline model and the adjustment of the boundary threshold are carried out simultaneously. After the data of the eighth day are included, the basic weight matrix, the corrected feature weights, and the anomaly detection threshold of the dynamic baseline model are all updated. The model uses the updated full set of parameters to perform state estimation and anomaly detection on the basic features input on the subsequent ninth day, thereby realizing the dynamic evolution of the model as the data stream continues to be input.

[0079] In one embodiment of the present invention, real-time acquired pet genetic data is divided into time slices consistent with the dynamic baseline model according to the acquisition time. The species abundance, functional gene expression abundance, and gene sequence features of the gene sequence data in each time slice are input into the dynamic baseline model, and the model outputs the normal range of the distribution of each layer in that time slice. The distribution of each layer in the real-time data is compared with the normal range output by the model. If the species abundance, functional gene expression abundance, or gene sequence features in a certain time slice exceed the normal range, it is determined to be a potential abnormal activity. For the determined potential abnormal activity, the deviation is marked as an abnormal behavior, and the degree of deviation is recorded as the difference between the actual distribution and the normal range, along with the relevant context information as the scene context information of that time slice. The severity of the deviation behavior is assessed based on the magnitude of the deviation and the frequency of the abnormal activity; the greater the deviation and the higher the frequency, the higher the abnormality level. Each abnormal activity is assigned an abnormality level. Extracting data such as collection time, collection site, and recent pet diet records from the contextual information of the abnormal behavior, and retrieving sample metadata using the extracted information, we can obtain abnormal background information such as the environmental conditions at the time of sample collection, the specific components of the pet's diet, and the specific types of items recently contacted. Using the extracted collection time, we retrieve environmental temperature and humidity records and collection site cleanliness records from the sample metadata at the same collection time point as the environmental conditions. Using the extracted collection site, we retrieve the diet composition records for that location at the corresponding collection time from the sample metadata. Using the time range from the extracted recent diet records, we retrieve the specific diet components for that time range from the sample metadata. Using the extracted 72-hour range prior to the collection time, we retrieve the types of items contacted for that time range from the sample metadata.

[0080] In practice, the real-time pet gene data is divided into time slices consistent with the dynamic baseline model according to the collection time. The dynamic baseline model is updated in 24-hour time slices. Therefore, the real-time data is also processed in 24-hour time slices. The species layer abundance, functional gene layer expression abundance, and gene sequence layer features in the gene sequence data of each time slice are input into the updated dynamic baseline model. The dynamic baseline model outputs the normal range of the distribution of each layer in the time slice. The normal range is given in the form of numerical intervals. For example, for the abundance of the species "Lactobacillus acidophilus", the normal range output is [0.15%, 0.30%].

[0081] In some embodiments, the distribution of each layer of real-time data is compared with the normal range output by the dynamic baseline model to determine potential abnormal activities. For example, for a certain time slice, the abundance detection value of the species "Clostridium difficile" in the real-time data is 0.45%, while the normal range for this species output by the dynamic baseline model is [0.01%, 0.20%]. The detection value exceeds the upper limit of the normal range, so the abundance of species layer in this time slice is judged as a potential abnormal activity. Similarly, if the expression abundance of "tetracycline resistance gene tetM" in the functional gene layer or the frequency of a specific single nucleotide polymorphism site in the gene sequence layer exceeds its corresponding normal range, a corresponding judgment is also made.

[0082] In practice, for any abnormal behavior that deviates from the identified potential abnormal activity markers, the degree of deviation is recorded as the absolute difference between the actual distribution value and the boundary value of the normal range. For example, the degree of deviation of the abundance of "Clostridium difficile" is |0.45% - 0.20%| = 0.25%. The relevant context information is recorded as the scene context information of that time slice. The scene context information includes the date corresponding to that time slice, the collection site information "feces", and the dietary component type extracted from the sample metadata "dog food containing novel protein sources".

[0083] It is understandable that the severity of deviant behavior is assessed based on the magnitude of the deviation and the frequency of anomalous activity, and an anomalous level is assigned to each anomalous activity. The assessment process follows a pre-defined level mapping rule, which quantifies the degree of deviation into discrete levels and counts the number of times the same anomalous type (such as anomalous abundance of a specific species) occurs within a continuous time slice as the frequency. The greater the deviation and the higher the frequency, the higher the mapped anomalous level. The anomalous level can be calculated using the following formula:

[0084]

[0085] in, The calculated anomaly level is represented by a positive integer. The normalized value representing the degree of deviation. This represents the frequency of the same anomaly type occurring within the most recent five time slices. and These are preset weighting coefficients. The table below shows an example of anomaly activity comparison and rating:

[0086] Table 1: Abnormal Activity Comparison and Level Assessment Table

[0087] Time slice Anomaly Level Detection value Normal range Degree of deviation (D) Recent frequency (F) Anomaly level ( ) 2025-10-05 Species layer (Clostridium difficile) 0.45% [0.01%, 0.20%] 0.25 2 3 2025-10-05 Functional gene layer (tetM) 120 RPKM [10, 50] RPKM 70 1 2

[0088] Optionally, the collection time, collection site, and recent pet diet records can be extracted from the relevant contextual information of the abnormal behavior. For example, the collection time can be extracted as "2025-10-05", the collection site as "feces", and the recent pet diet record as "mainly consuming brand A dog food in the past 24 hours". The environmental temperature and humidity records and collection site cleanliness records at the same collection time point can be retrieved from the sample metadata using the extracted collection time "2025-10-05" as the environmental status. The retrieval results are: environmental temperature 23 degrees Celsius, humidity 55 percent, and collection site cleanliness "normal".

[0089] In some embodiments, the extracted collection site "feces" is used to retrieve the dietary component records of the corresponding collection time in the sample metadata. The time range "2025-10-04 to 2025-10-05" in the extracted recent dietary records is used to retrieve the specific dietary components of the sample metadata within that time range. The specific search results are "chicken powder, pea protein, corn starch". The seventy-two hours before the extracted collection time "2025-10-05" are used to retrieve the contact item type records of the sample metadata within that time range. The search results are "rubber toys, synthetic fiber carpets, park lawns". All of this information constitutes the abnormal background information of the abnormal behavior.

[0090] See Figure 3In one embodiment of the present invention, sample metadata matching the abnormal background information in the deviation identification results is collected. The sample metadata includes the collection environment status of the time slice corresponding to the abnormal behavior, the specific components of the pet's diet, the types of recently contacted items, and the pet's historical health records. Complete scene information when the abnormal behavior occurred is obtained from the sample metadata. Using the complete scene information as an analysis framework, correlation analysis is performed on the collection environment status, specific components of the diet, types of contacted items, and historical health records in the sample metadata to identify abnormalities in diet components, environmental status, or contacted items that occur simultaneously with the abnormal behavior. Combining the abnormal hierarchical correlation in the genetic data, potential deep-seated health threat patterns are identified as species functional association anomalies caused by diet components, gene sequence association anomalies caused by environmental status, or cross-level association anomalies caused by contacted items. At the same time, abnormal signals are identified as the specific data manifestations corresponding to these threat patterns. Health clues related to the abnormal signals are extracted from the sample metadata. Health clues are the diet, environment, or contacted items when similar abnormalities occurred in historical health records. The threat type is identified as a digestive health threat, a skin health threat, or an immune health threat based on the matching degree between the health clues and the current abnormal signal. For identified anomalous signals, the process traces back from the current time slice to historical time slices. First, it identifies the direct source of the anomalous signal in the current time slice—whether it was the intake of a certain dietary component, a change in a certain environmental state, or contact with a certain type of object in a previous time slice. Then, it continues tracing back to the source in previous time slices until the initial time slice and corresponding factor that triggered the anomaly are found, thus determining the initial source of the anomaly. The threat pattern, threat type, anomaly path, and initial source corresponding to the anomalous signal are integrated to generate a combined report.

[0091] In practice, sample metadata that matches the abnormal background information in the deviation identification results is collected. For example, for an abnormal behavior marked as "abnormal abundance of Clostridium difficile", its abnormal background information includes the collection time "2025-10-05" and the collection site "feces". Based on this, all records containing the fields "2025-10-05" and "feces" in the sample metadata are retrieved and collected. The sample metadata includes the collection environment status of the time slice corresponding to the abnormal behavior, such as "indoor temperature of 23 degrees Celsius", the specific ingredients of the pet's diet, such as "chicken meal, pea protein, corn starch", the types of items recently contacted, such as "rubber toys, synthetic fiber carpets", and the pet's historical health records, such as "two diarrhea records in the past six months". The complete scene information when the abnormal behavior occurred is obtained from these collected sample metadata entries.

[0092] In some embodiments, complete scene information is used as an analytical framework to perform correlation analysis on the collection environment status, specific dietary components, contact item types, and historical health records in the sample metadata. For example, the analysis found that "pea protein" appeared for the first time in the specific dietary components of the time slice "2025-10-05" where the abnormal behavior occurred, and that the dietary records of the two previous episodes of diarrhea contained "legume components". At the same time, species functional association analysis in the genetic data showed that the abundance of Clostridium difficile was abnormally positively correlated with the expression of some virulence genes, thereby identifying a potential deep-seated health threat pattern as "abnormal species functional association caused by dietary components (pea protein)". The identified abnormal signal is the specific data performance corresponding to this threat pattern, namely "the abundance of Clostridium difficile increased from the baseline of 0.1% to 0.45% and the expression of virulence genes was upregulated".

[0093] In practice, health cues related to the abnormal signal are extracted from the sample metadata. For example, entries corresponding to "two previous diarrhea records" are extracted from historical health records. The entries record the diet, environment, or contact with items at that time, and the health cues obtained are "the diet before diarrhea both contained legumes" and "the environment was humid during diarrhea". The threat type is identified based on the matching degree between the health cues and the current abnormal signal. The matching degree is calculated by comparing the similarity between the features of the current abnormal signal and the features of the historical health cues. The calculation formula can be expressed as:

[0094]

[0095] in, Represents the degree of threat type matching. This represents the number of feature dimensions being compared. Representing the The weights of each feature The first one representing the current abnormal signal 1 eigenvalue, The first representing historical health clues 1 eigenvalue, It is a comparison function that returns 1 when two feature values ​​are identical or fall within the same preset range, and 0 otherwise. The calculated matching degree... If the clue "the diet before diarrhea contained legumes" is higher than the preset threshold, the threat type is identified as "digestive health threat".

[0096] It is understandable that identifying abnormal signals involves tracing back from the current time slice to historical time slices to determine the abnormal path and initial source. Taking the "abnormal abundance of Clostridium difficile" signal as an example, tracing back from the current time slice "2025-10-05" to the previous time slice "2025-10-04", the analysis found that "pea protein" had been introduced into the dietary records of the "2025-10-04" time slice, and the abundance of Clostridium difficile began to show a slight upward trend. This was determined to be the direct source of the current abnormal signal. Continuing to trace back to the previous time slice "2025-10-03", the dietary records showed no legume components and the microbiome was normal. Therefore, the time slice that initially triggered the abnormality was determined to be "2025-10-04", and the corresponding factor was "the introduction of pea protein into the diet".

[0097] Optionally, the threat pattern, threat type, anomalous path, and original source corresponding to the anomalous signal can be integrated to generate a merged report. For example, the merged report can be structured to include the following parts: the threat pattern is "species function association anomaly caused by dietary component (pea protein)", the threat type is "digestive health threat", the anomalous path is "2025-10-04 dietary introduction of pea protein -> 2025-10-05 anomalous increase in Clostridium difficile abundance accompanied by upregulation of virulence gene expression", and the original source is "dietary changes on 2025-10-04".

[0098] In some embodiments, if an abnormal signal is found to be associated with multiple factors across multiple time slices during the tracing process, the abnormal path will contain multiple nodes. For example, the abnormal path may be recorded as "2025-10-01 Decreased environmental cleanliness -> 2025-10-03 Skin microbiome dysbiosis -> 2025-10-05 Abnormal expression of immune-related functional genes". The merged report will fully describe this multi-step abnormal evolution path and the factors corresponding to each node.

[0099] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A pet health monitoring system based on microbiome gene detection, characterized in that, The system includes: The data acquisition and fluidization module is used to determine the microbiome gene data acquisition plan for pet health monitoring, acquire gene sequence data and sample metadata from pet biological samples according to the plan, and concatenate the acquired data into a real-time gene data stream in chronological order. The cross-level correlation and initial screening module is used to conduct cross-level correlation analysis based on the hierarchical logic of real-time gene data streams. It combines the context of the data generation scenario to sort out the correlations at different levels, find the correlations and abnormal interaction patterns of gene data between levels, mark the found results as preliminary abnormal signals, and aggregate them into a weighted abnormal label dataset. The dynamic baseline modeling module is used to build a dynamic baseline model based on real-time gene data streams and anomalous marker datasets, calibrate the baseline model using the results of cross-level correlation analysis, and dynamically adjust parameters as the data is updated. The anomaly comparison and background analysis module is used to compare real-time collected data with the dynamic baseline model, identify anomalous activities that deviate, mark anomalous behaviors that deviate, and submit the deviation identification results to sample metadata analysis to obtain anomalous background information. The Deep Threat Tracing Module is used to analyze sample metadata based on the anomaly identification results of the dynamic baseline model, identify potential deep-seated health threat patterns and abnormal signals, trace the threat patterns and abnormal signals to determine the root cause and abnormal path.

2. The pet health monitoring system based on microbiome gene detection according to claim 1, characterized in that, The aforementioned microbiome gene data acquisition scheme for pet health monitoring involves obtaining gene sequence data and sample metadata from pet biological samples according to the scheme, and concatenating the acquired data in chronological order into a real-time gene data stream. Specifically, this includes: First, based on the pet's age, daily diet, and recent health status, determine whether the biological sample to be collected is a fecal sample, oral sample, or skin sample. At the same time, specify the genetic data capture requirements for each type of sample, including whole-genome sequencing depth, functional gene targeted capture range, and sample metadata must include the collection time, collection site, and the pet's recent diet record. Gene data capture is performed on real-time collected biological samples according to specific requirements, including extracting gut microbial genes from fecal samples, oral microbial genes from oral samples, and skin microbial genes from skin samples. The captured gene data is first classified according to the sample source, then the species classification information is sorted out by microbial species layer, the gene function information is sorted out by functional gene layer, and the sequence base arrangement information is sorted out by gene sequence layer. The gene sequence data and sample metadata including collection time, collection site, pet's recent diet record, and pet breed are then parsed out. The parsed gene sequence data are concatenated according to the collection time, and the sample metadata of the corresponding time point is attached to the gene sequence data concatenated at the same time to form a real-time gene data stream.

3. The pet health monitoring system based on microbiome gene detection according to claim 2, characterized in that, The method involves conducting cross-level correlation analysis based on the hierarchical logic of real-time gene data streams, combining the context of data generation to analyze the relationships between different levels, identifying the correlations and abnormal interaction patterns of gene data between levels, marking the identified results as preliminary anomaly signals, and compiling them into a weighted anomaly label dataset. Specifically, this includes: First, following the generation logic of real-time gene data stream, the gene sequence data is divided into continuous time slices according to the collection time. Within each time slice, the data is organized into species layer, functional gene layer, and gene sequence layer. From the data of each time slice, the environmental conditions of the collection site, the types of components in the pet's diet, and the types of items recently encountered are extracted as contextual information. For the species layer and functional gene layer within each time slice, the correlation between the abundance of a certain type of microbial species and the expression abundance of the corresponding functional gene is analyzed. The correlation is then combined with the type of components in the pet diet of that time slice to determine whether it conforms to the normal correspondence logic between species and functional genes under that component. If it does not conform, it is recorded as an abnormal species-function correlation. For the functional gene layer and gene sequence layer, the correlation between the sequence characteristics of a certain functional gene and the gene expression abundance is analyzed. The correlation is then combined with the environmental conditions of the sampling site of that time slice to determine whether it conforms to the normal correspondence logic between functional genes and sequence characteristics under that environment. If it does not conform, it is recorded as an abnormal gene sequence correlation. Species function association anomalies and gene sequence association anomalies are marked as preliminary anomalous signals. The anomalous type of each preliminary anomalous signal is recorded as either species function association anomaly or gene sequence association anomaly, the frequency is the number of times the anomaly occurs within the time slice, and the severity is the degree to which the association deviates from normal logic. Weights are assigned based on the logical deviation difficulty corresponding to the anomaly type, the frequency of occurrence corresponding to the frequency, and the deviation depth corresponding to the severity of each preliminary anomaly signal. The weighted preliminary anomaly signals are then integrated with the scene context information of the corresponding time slice to form a weighted anomaly label dataset.

4. The pet health monitoring system based on microbiome gene detection according to claim 3, characterized in that, The process involves constructing a dynamic baseline model based on real-time gene data streams and anomaly marker datasets, calibrating the baseline model using the results of cross-level correlation analysis, and dynamically adjusting parameters as the data is updated. Specifically, this includes: First, extract the species-level microbial species abundance distribution, functional gene expression abundance distribution, and gene sequence feature distribution of each time slice from the real-time gene data stream. Then, extract the preliminary anomalous signal weights and scene context information of each time slice from the weighted anomaly label dataset. Using the extracted species layer abundance distribution, functional gene layer expression abundance distribution, and gene sequence layer feature distribution as basic features, and the weights of the abnormal signals in the weighted abnormal label dataset as correction features, an initial dynamic baseline model is constructed. The basic features are used to train the model to learn the stable features of each layer distribution under normal conditions, and the correction features are used to train the model to identify the boundary between normal and abnormal distributions. For each new time-slice of real-time gene data stream and corresponding weighted anomaly marker dataset, the basic features of the time-slice are first input into the initial dynamic baseline model to obtain the basic predicted values ​​of the distribution of each layer of the time-slice. Then, the anomaly signal weights in the weighted anomaly marker dataset of the time-slice are used to adjust the basic predicted values. The adjusted results are used as the parameters of the baseline model of the time-slice. The overall parameters of the dynamic baseline model are updated with the parameters of the new time-slice. At the same time, the threshold for the model to identify the boundary between normal and abnormal is adjusted according to the distribution of the anomaly signal weights of the new time-slice, so as to achieve dynamic parameter tuning as the data is updated.

5. The pet health monitoring system based on microbiome gene detection according to claim 4, characterized in that, The process involves comparing real-time collected data with a dynamic baseline model to identify and flag abnormal activities and behaviors. The deviation identification results are then fed into sample metadata analysis to obtain abnormal background information. Specifically, this includes: The real-time pet gene data is divided into time slices consistent with the dynamic baseline model according to the collection time. The species layer abundance, functional gene layer expression abundance, and gene sequence layer features in the gene sequence data of each time slice are respectively input into the dynamic baseline model. The model outputs the normal range of the distribution of each layer in that time slice. Compare the distribution of each layer of real-time data with the normal range of the model output. If the abundance of species layer, the expression abundance of functional gene layer, or the characteristics of gene sequence layer in a certain time slice exceed the normal range, it is judged as a potential abnormal activity. For potential abnormal activity markers that deviate from the marked abnormal behavior, the degree of deviation of the abnormal behavior is recorded as the difference between the actual distribution and the normal range, and the relevant context information is the scene context information of that time slice; The severity of the deviation is assessed based on the degree of deviation and the frequency of the abnormal activity. The greater the degree of deviation and the higher the frequency, the higher the abnormality level. Each abnormal activity is assigned an abnormality level. Extract the collection time, collection location, and recent pet diet records from the relevant contextual information of the abnormal behavior. Use the extracted information to retrieve sample metadata and read the abnormal background information, such as the environmental conditions at the time of sample collection, the specific components of the pet's diet, and the specific types of items the pet has recently come into contact with.

6. The pet health monitoring system based on microbiome gene detection according to claim 5, characterized in that, The anomaly identification results based on the dynamic baseline model analyze sample metadata to identify potential deep-seated health threat patterns and abnormal signals. The source of these threat patterns and abnormal signals is traced to determine the root cause and abnormal path, specifically including: First, collect sample metadata that matches the abnormal background information in the deviation identification results. The sample metadata includes the collection environment status of the time slice corresponding to the abnormal behavior, the specific ingredients of the pet's diet, the types of items recently contacted, and the pet's historical health records. Obtain complete scene information when the abnormal behavior occurred from the sample metadata. Using complete scene information as an analytical framework, we conduct correlation analysis on the collection environment status, specific dietary components, contact item types, and historical health records in the sample metadata to identify abnormal dietary components, environmental conditions, or contact items that occur simultaneously with abnormal behavior. By combining the abnormal hierarchical correlation in the genetic data, we identify potential deep-seated health threat patterns as species functional association anomalies caused by dietary components, gene sequence association anomalies caused by environmental conditions, or cross-layer association anomalies caused by contact items. At the same time, we identify the abnormal signals as the specific data manifestations corresponding to these threat patterns. Health cues related to anomalous signals are extracted from sample metadata. These health cues are dietary, environmental, or contact items that occurred when similar anomalies appeared in historical health records. Threat types are identified as digestive health threats, skin health threats, or immune health threats based on the matching degree between health cues and current anomalous signals. For the identified abnormal signals, trace back from the current time slice to the historical time slices. First, find the direct source of the abnormal signal in the current time slice, which is the intake of a certain type of food component, a change in a certain type of environmental state, or contact with a certain type of contact item in the previous time slice. Then continue to trace back to the source in the previous time slice until the time slice that initially caused the abnormality and the corresponding factor are found, and the initial source of the abnormality is determined. The report integrates the threat patterns, threat types, and abnormal paths corresponding to the abnormal signals with their original sources to generate a combined report.

7. The pet health monitoring system based on microbiome gene detection according to claim 2, characterized in that, The process of concatenating the parsed gene sequence data according to the collection time, and attaching the sample metadata of the corresponding time point to the concatenated gene sequence data to form a real-time gene data stream, specifically includes: The parsed gene sequence data is divided into time units based on the hour of collection time. The gene sequence data within each time unit are concatenated from the earliest to the latest collection time. The metadata of the samples collected within the same time unit is appended to the end of the concatenated gene sequence data in that time unit according to the order of collection time. If samples from multiple sites are collected at the same time point, the corresponding metadata is appended according to the site priority, forming a real-time gene data stream concatenated by time units with metadata appended.

8. The pet health monitoring system based on microbiome gene detection according to claim 3, characterized in that, The extraction of contextual information from each time-slice of data, including the environmental conditions at the collection location, the types of ingredients in the pet's diet, and the types of items recently encountered, specifically includes: The environmental temperature and humidity records and cleanliness records of the collection site at the corresponding collection time point are retrieved from the sample metadata to determine the environmental status. The protein source, carbohydrate source, and additive type in the pet's diet in the 24 hours prior to the time segment are retrieved to determine the dietary component type. The toy material, bedding cleanliness, and type of external plants that the pet came into contact with in the 72 hours prior to the time segment are retrieved to determine the recently contacted item type.

9. The pet health monitoring system based on microbiome gene detection according to claim 4, characterized in that, The process of updating the overall parameters of the dynamic baseline model with the parameters of the new time slice, and adjusting the threshold for identifying the normal and abnormal boundaries of the model based on the distribution of the abnormal signal weights in the new time slice, specifically includes: The parameters of the new time slice include the basic predicted values ​​and corrected parameters of each layer distribution of the time slice. The basic predicted values ​​of the new time slice are added to the historical basic feature library of the model. The average value of the historical basic feature library is used to update the basic weights of each layer distribution of the model. The corrected parameters of the new time slice are added to the historical corrected feature library of the model. The median of the historical corrected feature library is used to update the weights of the corrected features of the model. The distribution of abnormal signal weights in the new time slice is statistically analyzed. If the proportion of high-weight abnormal signals increases, the threshold for the model to identify the boundary between normal and abnormal signals is increased, making the model more strict in identifying abnormalities. If the proportion of low-weight abnormal signals increases, the threshold is decreased, making the model more lenient in identifying abnormalities. This achieves the adjustment of the boundary threshold according to the distribution of abnormal signal weights.

10. The pet health monitoring system based on microbiome gene detection according to claim 5, characterized in that, The process involves retrieving sample metadata using the extracted information, reading abnormal background information such as the environmental conditions at the time of sample collection, the specific components of the pet's diet, and the specific types of items the pet recently came into contact with, specifically including: The environmental temperature and humidity records and the cleanliness records of the collection site at the same collection time point are retrieved from the sample metadata using the extracted collection time as the environmental status. The dietary composition records of the corresponding collection time for that location are retrieved from the sample metadata using the extracted collection site. The specific dietary components for that time range are retrieved from the sample metadata using the time range of the extracted recent dietary records. The contact item type records for that time range are retrieved from the sample metadata using the range of the seventy-two hours prior to the extracted collection time.

Images