A quality evaluation system and method for functional factors from paddlefish byproducts

Abstract

This invention relates to the field of quality assessment technology for giant salamander by-products, and discloses a quality evaluation system and method for functional factors of giant salamander by-products. The method includes: acquiring giant salamander by-product samples and preprocessing them to extract functional factors and obtain samples to be tested; identifying and quantifying specific biomarkers based on biochemical analysis and molecular detection methods to generate a biomarker dataset; inputting the biomarker data into a multimodal feature fusion module to extract activity and stability features and perform time-series dynamic modeling; setting independent evaluation branches for activity and stability to generate corresponding score values; dividing functional domains according to functional factor categories, constructing evaluation curves based on historical data, substituting the target sample into each functional domain and adjacent functional domains to generate functional domain evaluation results and weighted fusion; finally, weighted integration of activity scores, stability scores, and comprehensive evaluation values ​​to determine the final quality evaluation result.

Description

Technical Field

[0001] This invention relates to the field of quality assessment technology for giant salamander by-products, and more specifically, to a quality evaluation system and method for source functional factors of giant salamander by-products. Background Technology

[0002] The by-products of giant salamanders contain a variety of functional factors, which play an important role in terms of bioactivity, stability and health benefits.

[0003] However, most existing methods for evaluating the quality of functional factors are limited to the detection of single indicators, such as measuring only the content or activity of a specific component. They lack the ability to comprehensively analyze multiple functional factors simultaneously, making it difficult to meet the needs of systematic and scientific evaluation. Furthermore, the functional factors in salamander byproducts have complex structural characteristics and time dependence; their activity and stability can change with extraction methods, preservation conditions, and processing. Existing technologies typically cannot effectively process multidimensional biomarker data and lack quantitative analysis methods for the performance of functional factors in different functional domains, resulting in evaluation results that fail to accurately reflect the overall quality level of the functional factors.

[0004] Therefore, there is an urgent need for a method that can integrate information from multiple biomarkers, extract activity and stability characteristics, perform dynamic modeling based on time series, and generate a comprehensive quality evaluation through multidimensional scoring and weighted fusion. This method can scientifically and objectively quantify the quality of functional factors in salamander by-products, providing reliable technical support for their research, application, and standardization. Summary of the Invention

[0005] In view of this, the present invention proposes a quality evaluation system and method for functional factors of giant salamander by-products, aiming to solve the problem that existing methods in the current technology cannot comprehensively, multidimensionally, dynamically and quantitatively evaluate the multiple functional factors with complex structural features and time dependence in giant salamander by-products, resulting in a lack of accuracy and systematicity in the evaluation results.

[0006] This invention proposes a quality evaluation method for functional factors derived from salamander byproducts, comprising: Samples of by-products from giant salamanders were obtained and preprocessed to extract functional factors and obtain samples to be tested. Based on biochemical analysis and molecular detection methods, specific biomarkers in the sample to be tested are identified and quantified to generate a biomarker dataset. The biomarker dataset is input into the multimodal feature fusion module to extract the activity and stability features of functional factors, and the features are dynamically modeled based on time series. Two independent evaluation branches are set up to score the activity and stability of the functional factors respectively, and generate activity score and stability score. The biomarker dataset is divided into several functional domains according to the functional factor categories. Evaluation curves for each functional domain are constructed based on the relationship between historical experimental data and functional factors. The biomarker data of the target sample are substituted into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results. A comprehensive evaluation value is generated through a weighted fusion strategy. The final quality evaluation result is obtained by weighting and integrating the activity score, stability score, and comprehensive evaluation value.

[0007] Furthermore, the pretreatment of salamander by-product samples includes: The crude extract was obtained by preliminary separation of the by-products of giant salamanders using a fractional extraction method. The crude extract was subjected to ultrasonic crushing, low-temperature centrifugation and membrane filtration in sequence to remove impurities and enrich functional factors. The functional factors were purified by gradient elution liquid chromatography, and the fraction corresponding to the target peak was collected to obtain the sample to be tested.

[0008] Furthermore, when identifying and quantifying specific biomarkers in samples based on biochemical analysis and molecular detection methods, this includes: Qualitative and quantitative analysis of small molecule biomarkers in samples to be tested was performed using high performance liquid chromatography-mass spectrometry (HPLC-MS). Specific identification and concentration determination of macromolecular biomarkers based on immunoassay technology; The structural characteristics of functional factors are verified using spectral analysis techniques, and a biomarker dataset containing information on multiple biomarkers is generated.

[0009] Furthermore, when inputting the biomarker dataset into the multimodal feature fusion module to extract the activity and stability features of functional factors, the following are included: The biomarker dataset was classified and coded according to functional factor categories, and the data of each category were normalized. Based on feature extraction of normalized data using a multi-channel sensor array, activity and stability features are obtained. The dynamic time warping algorithm is used to model the features in time series and generate dynamic feature sequences.

[0010] Furthermore, when dynamically modeling features based on time series data, this includes: The dynamic feature sequence is input into the recurrent neural network module to capture the time-dependent features of the functional factors; The importance score of the feature at each time step is calculated based on the attention mechanism, and the score is normalized to obtain the time weight. The time-step features are weighted and aggregated according to the time weight to generate a global dynamic feature representation.

[0011] Furthermore, two independent evaluation branches are set up to score the activity and stability of the functional factors, including: An activity evaluation branch and a stability evaluation branch are set up, and each branch is composed of a multilayer perceptron module; Each branch includes at least two fully connected layers, and regularization constraints are introduced between the layers to suppress overfitting; During the training phase, only the parameters of the corresponding branch are updated based on the functional factor category label, while the other branch is frozen. During the prediction phase, the global dynamic feature representation is input into the two branches respectively to generate activity score and stability score.

[0012] Furthermore, when dividing the biomarker dataset into several functional domains according to functional factor categories, these include: Using the biological activity of functional factors as a variable, functional factors are divided into several functional domains, and a hierarchical structure is constructed according to the intensity of their effects. Within each functional domain, regression fitting is performed using historical experimental data as independent variables and functional factor activity as dependent variables to generate evaluation curves.

[0013] Furthermore, when substituting the biomarker data of the target sample into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results, this includes: Determine the functional domain to which the target sample belongs and its neighboring functional domains, and substitute the biomarker data of the target sample into the corresponding evaluation curves to generate multiple functional domain evaluation results. The weights are calculated based on the distance between the center point of each functional domain and the target sample. The evaluation results of the functional domains are then summed in a weighted manner to generate a comprehensive evaluation value.

[0014] Furthermore, when obtaining the final quality evaluation result by weighted integration of the activity score, stability score, and comprehensive evaluation value, the following are included: Performance indicators were calculated for the activity score, stability score, and comprehensive evaluation value, and subjective and objective weights were determined based on the performance indicators. Weights are generated by using weight normalization and non-negativity constraints, where the sum of all weights in the combined weights is 1. The quality score of the salamander by-products is determined based on the relationship between the combined weights and the activity score, stability score, and comprehensive evaluation value. Based on the relationship between the quality score and the preset quality score, the final quality evaluation result of the salamander by-product is determined: If the quality score is lower than the preset quality score, the final quality evaluation of the giant salamander by-product is determined to be unqualified. When the quality score is higher than or equal to the preset quality score, the final quality evaluation of the giant salamander by-product is determined to be qualified.

[0015] Compared with existing technologies, the advantages of this invention are as follows: By performing biochemical analysis and molecular detection on biomarkers, comprehensive information on the chemical composition and structural characteristics of functional factors can be obtained. Combined with a multimodal feature fusion module, information from biomarkers of different sources and types can be integrated to extract the activity and stability characteristics of functional factors, and the changing patterns of functional factors at different time points can be captured through time-series dynamic modeling. This technique overcomes the limitations of traditional methods that rely solely on a single indicator or a single time point, making the evaluation results more comprehensive and scientific. Secondly, by setting independent evaluation branches, the activity and stability of functional factors are scored separately, and historical experimental data and functional domain evaluation curves are introduced to achieve quantitative analysis of the performance of functional factors in different functional domains. Biomarker data of the target sample can be simultaneously substituted into the evaluation curves of its own functional domain and neighboring functional domains, and a comprehensive evaluation value is generated through a weighted fusion strategy, thereby effectively solving the problem that a single indicator evaluation cannot reflect the overall quality of functional factors, enhancing the accuracy and reliability of the evaluation. Finally, by weighting and integrating the activity score, stability score, and comprehensive evaluation value, a combined weight is generated using a fusion of subjective and objective weights to quantitatively calculate the final quality evaluation result. This not only ensures the reasonable contribution of information from different dimensions to the evaluation results, but also allows for dynamic adjustments based on actual experimental data and historical standards, making the final evaluation results more scientific, objective, and comparable.

[0016] On the other hand, this application also provides a quality evaluation system for functional factors derived from salamander by-products, including: The preprocessing unit is configured to acquire and preprocess samples of giant salamander byproducts, and extract functional factors therein to obtain samples to be tested. The detection unit is configured to identify and quantify specific biomarkers in the sample to be tested through biochemical analysis and molecular detection methods, and generate a biomarker dataset. The feature extraction unit is configured to input the biomarker dataset into the multimodal feature fusion module, extract the activity and stability features of functional factors, and dynamically model the features based on time series. The evaluation unit is configured to set up two independent evaluation branches to score the activity and stability of the functional factors respectively, and generate activity score values ​​and stability score values. The functional domain partitioning unit is configured to divide the biomarker dataset into several functional domains according to the functional factor category, construct evaluation curves for each functional domain based on the relationship between historical experimental data and functional factors, substitute the biomarker data of the target sample into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results, and generate a comprehensive evaluation value through a weighted fusion strategy. The integration unit is configured to weight and integrate the activity score, stability score, and comprehensive evaluation value to obtain the final quality evaluation result.

[0017] It is understood that the quality evaluation system and method for functional factors of salamander by-products in the above embodiments have the same beneficial effects, and will not be described in detail here. Attached Figure Description

[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a method for quality evaluation of functional factors derived from salamander by-products, provided in an embodiment of the present invention. Figure 2 A flowchart illustrating a method for quality evaluation of functional factors derived from by-products of giant salamanders, provided in an embodiment of the present invention. Figure 3 This is a functional block diagram of a quality evaluation system for functional factors of by-products from giant salamanders, provided as an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] like Figures 1-2 As shown in some embodiments of this application, this embodiment provides a method for quality evaluation of functional factors derived from salamander by-products, including: Step S100: Obtain a sample of salamander byproducts and preprocess it to extract the functional factors and obtain the sample to be tested.

[0021] Specifically, the pretreatment of giant salamander by-product samples includes: preliminary separation of giant salamander by-products based on fractional extraction to obtain crude extract; sequential ultrasonic crushing, low-temperature centrifugation and membrane filtration of crude extract to remove impurities and enrich functional factors; purification of functional factors based on gradient elution liquid chromatography, collection of fractions corresponding to target peaks to obtain the sample to be tested.

[0022] Understandably, the preliminary separation of salamander byproducts using a fractional extraction method relies on differences in solubility, polarity, or molecular weight among different components to achieve initial separation of target functional factors from non-target substances through fractionation. This method effectively removes some non-functional impurities, creating conditions for subsequent refined extraction. Secondly, the crude extract undergoes sequential ultrasonic disruption, low-temperature centrifugation, and membrane filtration. Ultrasonic disruption utilizes the cavitation effect and mechanical vibration generated by ultrasound to disrupt cell or tissue structures, thereby releasing functional factors. Low-temperature centrifugation separates impurities from target components by utilizing differences in sedimentation rates under high-speed rotation, while the low temperature maintains the activity and stability of functional factors. Membrane filtration, based on molecular size or molecular retention characteristics, removes large molecular impurities and enriches functional factors. Finally, gradient elution liquid chromatography is used to purify the functional factors. This is achieved by adjusting the composition of the mobile phase to create differences in retention times for components with different polarities or molecular properties on the stationary phase, thus separating the target functional factors. By collecting the fraction corresponding to the target peak, a higher purity sample can be obtained, reducing interference in subsequent analysis.

[0023] In a specific embodiment of this application, the above steps are implemented as follows: Using salamander visceral byproducts as raw materials, a fractional extraction method is used for preliminary processing. For example, soluble proteins and small-molecule peptides are first extracted using a water-soluble solvent, and then lipid-soluble components are extracted using ethanol or other organic solvents. This preliminary fractionation process separates functional factors from a large number of non-target substances (such as lipids, impurity proteins, etc.), resulting in a relatively concentrated crude extract. Subsequently, the obtained crude extract is processed using an ultrasonic device. Ultrasonic action breaks down cell membranes and tissue structures, allowing the release of potential functional proteins or peptides. The processed solution is then subjected to low-temperature, high-speed centrifugation, for example, centrifuging at 12,000 rpm for 15 minutes at 4°C. The precipitate and supernatant are separated; the supernatant contains a large number of target factors and retains their activity. Further membrane filtration is performed using an ultrafiltration membrane with a molecular weight cutoff of 3 kDa to retain large-molecule impurities, thereby enriching low-molecular-weight, bioactive peptides. Finally, the supernatant after membrane filtration was injected into a high-performance liquid chromatography (HPLC) system using gradient elution, for example, by gradually increasing the proportion of acetonitrile in a water-acetonitrile system. During gradient elution, functional peptides exhibit different retention times due to their polarity and structural differences. By detecting the appearance times of different peaks on the chromatogram, researchers collected the fractions corresponding to the target peaks. For example, molecules corresponding to characteristic peaks with retention times in the 10-12 minute range were identified by mass spectrometry and collected as functional factor samples. After purification, the purity of these samples was significantly improved, allowing them to be directly used for subsequent activity detection and quality evaluation.

[0024] It can be seen that by adopting the stepwise enrichment and screening approach of "gradual extraction - physical separation - chromatographic purification", impurities can be effectively removed and high-purity samples can be obtained while ensuring the activity and stability of functional factors. This provides a reliable experimental basis for subsequent quantitative detection and quality evaluation.

[0025] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0026] Step S200: Identify and quantify specific biomarkers in the sample to be tested based on biochemical analysis and molecular detection methods, and generate a biomarker dataset.

[0027] Specifically, when identifying and quantifying specific biomarkers in samples based on biochemical analysis and molecular detection methods, this includes: qualitative and quantitative analysis of small molecule biomarkers in samples based on high performance liquid chromatography-mass spectrometry; specific identification and concentration determination of macromolecular biomarkers based on immunoassay techniques; and verification of the structural characteristics of functional factors based on spectral analysis techniques to generate a biomarker dataset containing information on multiple biomarkers.

[0028] Understandably, high-performance liquid chromatography (HPLC) can separate small molecule biomarkers based on their polarity, hydrophobicity, and other physicochemical properties, utilizing the differences in distribution between the stationary and mobile phases. Mass spectrometry (MS), on the other hand, analyzes the mass-to-charge ratio (m / z) of molecules after ionization, enabling structural analysis and quantitative determination. Combining these two methods not only improves separation efficiency and detection sensitivity but also avoids the limitations of single detection methods, thus accurately identifying small molecule functional factors or metabolites in samples. Secondly, for macromolecular biomarkers (such as specific proteins or peptides), immunoassay techniques are used for specific identification and concentration determination. This relies on the highly specific binding between antigens and antibodies. By constructing immunoreaction systems (such as ELISA and immunochromatography), and utilizing signal conversion methods such as colorimetry, fluorescence, or chemiluminescence, sensitive detection and quantification of target macromolecules are achieved. This principle ensures the selectivity and reliability of detection, making it particularly suitable for the analysis of specific biomarkers in complex biological samples. Finally, spectroscopic analysis techniques are used to verify the structural characteristics of functional factors. Spectroscopic analysis (such as ultraviolet-visible spectroscopy, infrared spectroscopy, Raman spectroscopy, etc.) rapidly characterizes molecular structures based on the energy absorption, scattering, or emission characteristics of different chemical bonds or molecular structures at specific wavelengths. By comparing the characteristic absorption peaks or fingerprints of functional factors, the structural integrity and characteristic group information can be verified, providing support for the accurate identification of biomarkers.

[0029] In a specific embodiment of this application, the above steps are implemented as follows: When detecting small molecule biomarkers, high-performance liquid chromatography-mass spectrometry (HPLC-MS) can be used to analyze small molecule amino acids, short peptides, or metabolites in salamander by-products. For example, researchers can inject the sample into a liquid chromatography system and separate amino acids such as glutamic acid and proline using gradient elution conditions; subsequently, the sample enters the mass spectrometry detection module, and its mass-to-charge ratio (m / z) is determined using electrospray ionization (ESI) mode to obtain the corresponding mass spectral peaks and compare them with standards, thereby achieving qualitative and quantitative analysis of these small molecule biomarkers. This not only identifies the main active small molecules in functional factors but also accurately determines their concentration distribution in the sample. Secondly, when detecting macromolecular biomarkers, immunoassay techniques can be used to achieve highly sensitive identification and quantification of specific proteins or peptides. For example, ELISA can be used to detect collagen in salamander byproducts: anti-collagen antibodies are immobilized on a solid support, allowing the target protein in the sample to bind specifically. Then, enzyme-labeled secondary antibodies are added, and after a colorimetric reaction, the absorbance is measured using a spectrometer to calculate the collagen concentration. This method maintains high specificity and accuracy even in complex sample environments. Finally, spectroscopic analysis techniques can be used to verify the structural characteristics of functional factors. For example, Fourier transform infrared spectroscopy (FTIR) can be used to characterize the extracted peptide functional factors, observing the amide I band at 1650 cm⁻¹ and the amide II band at 1540 cm⁻¹ to verify the structural characteristics of the peptide chain; or Raman spectroscopy can be used to detect the vibrational modes of specific functional groups (such as hydroxyl and carboxyl groups) to confirm the presence of specific functional groups in the sample. Comparison of these characteristic spectra can further ensure the structural integrity and characteristicity of the measured markers.

[0030] It can be seen that this invention can not only conduct multi-dimensional detection of different types of functional factors in salamander by-products, but also obtain high-precision and high-reliability biomarker datasets through the complementary advantages of chromatography-mass spectrometry, immunology and spectroscopy, providing solid data support for subsequent quality evaluation.

[0031] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0032] Step S300: Input the biomarker dataset into the multimodal feature fusion module, extract the activity and stability features of the functional factors, and dynamically model the features based on the time series.

[0033] Specifically, when inputting the biomarker dataset into the multimodal feature fusion module to extract the activity and stability features of functional factors, the process includes: classifying and encoding the biomarker dataset according to the functional factor categories and normalizing the data of each category; extracting features from the normalized data using a multi-channel sensor array to obtain activity and stability features; and performing time series modeling of the features based on the dynamic time warping algorithm to generate dynamic feature sequences.

[0034] Specifically, when dynamically modeling features based on time series, the process includes: inputting the dynamic feature sequence into a recurrent neural network module to capture the time-dependent features of functional factors; calculating the importance score of each time step feature based on an attention mechanism and normalizing the score to obtain time weights; and weighting and converging the time step features according to the time weights to generate a global dynamic feature representation.

[0035] Understandably, when inputting the biomarker dataset into the multimodal feature fusion module, the focus is on data standardization and multimodal fusion. Classifying and encoding the biomarker data ensures the uniformity of the data structure for different categories of functional factors; normalization eliminates interference from differences in detection methods, numerical ranges, and units, making various data types comparable and fusionable. Subsequently, a multi-channel sensor array is used to extract features from the processed data. The principle is to simulate the sensitive response to the data through different channels, capturing the performance of functional factors in both activity and stability dimensions. This multi-channel approach enables parallel information capture and complementary fusion, generating more representative feature vectors. Secondly, when performing time series modeling based on the Dynamic Time Warping (DTW) algorithm, the technical principle is to capture the changing patterns of functional factors at different time scales through nonlinear sequence alignment. DTW can flexibly handle misalignment issues on the time axis, identifying the similarities and differences in the characteristics of different functional factors over time by calculating the optimal matching path between samples. This allows dynamic modeling to go beyond static feature descriptions, reflecting the dynamic evolution of activity and stability over time. Finally, in the combined application of recurrent neural networks (RNNs) and attention mechanisms, the key lies in deep learning's ability to capture temporal dependencies and model their importance. RNNs can progressively input time-series data, preserving prior information through recursive propagation of hidden states, thus effectively capturing the dependencies between functional factors at different time points. Simultaneously, the attention mechanism calculates the importance scores of features at each time step, reflecting the contribution of different time segments to the overall quality assessment. Normalized time weights are used to weighted convergence of time-step features, ultimately generating a global dynamic feature representation that provides more accurate and interpretable dynamic features of functional factors while considering both overall trends and key time segments.

[0036] In a specific embodiment of this application, the above steps are implemented as follows: In the functional factor classification, coding, and normalization stage, different categories of biomarkers in salamander byproducts can be used as examples, such as small molecule amino acids, polypeptides, and large molecule proteins. Researchers encode these biomarkers from different sources according to their functional categories, for example, classifying amino acids into one category and polypeptides into another. Since different detection methods yield different numerical ranges, for example, the peak intensity of small molecules measured by mass spectrometry is in the range of 10²–10⁻⁶. 4 The protein concentrations obtained from immunoassays may be at the ng / mL level, making the data incomparable without normalization. Normalization maps all data to a unified range (e.g., 0–1), facilitating subsequent feature fusion and comparison. Secondly, in examples of feature extraction using multi-channel sensor arrays, different sensing channels can be set to target activity and stability dimensions respectively. For example, one channel targets the concentration change pattern of the biomarker, another targets degradation rate or half-life changes, and a third channel combines external conditions such as temperature sensitivity or oxidative stress. Through parallel input of multiple channels, the system can simultaneously capture the activity characteristics (e.g., the trend of increased or decreased antioxidant capacity) and stability characteristics (e.g., whether it retains structural integrity after one week of storage). This provides a more comprehensive reflection of the multi-dimensional performance of functional factors. Thirdly, in examples of dynamic time warping (DTW) modeling, the time-varying curves of functional factors under different experimental batches or storage conditions can be considered. For example, the antioxidant activity of sample A decreases rapidly after 3 days, while the decrease in sample B occurs on day 5. Traditional time alignment methods can lead to errors, while DTW (Time-Diminished Wave) allows for non-linear alignment of two time curves, identifying their similarities in the trend of declining activity, even if the occurrences occur at different points in time. This method ensures that dynamic modeling captures the evolution of functional factors more accurately. Finally, in the example combining recurrent neural networks (RNNs) with attention mechanisms, the activity and stability sequences of functional factors can be defined over a 0–10 day storage period. The RNN progressively learns the dependencies between time points, such as the impact of changes on the activity level on day 6. The attention mechanism calculates importance scores at different time steps; for example, days 1 and 7 may contribute more to the final quality assessment result. Through attention-weighted convergence, the globally dynamic feature representation generated by the model not only reflects the overall trend but also highlights key time points, thus providing a more interpretive basis for the final quality assessment.

[0037] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0038] Step S400: Set up two independent evaluation branches to score the activity and stability of the functional factors respectively, and generate activity score and stability score.

[0039] Specifically, when setting up two independent evaluation branches to score the activity and stability of functional factors, the following steps are taken: setting an activity evaluation branch and a stability evaluation branch, each branch consisting of a multilayer perceptron module; each branch including at least two fully connected layers, and introducing regularization constraints between layers to suppress overfitting; during the training phase, updating only the parameters of the corresponding branch based on the functional factor category label, freezing the other branch; and during the prediction phase, inputting the global dynamic feature representation into the two branches respectively to generate activity and stability scores.

[0040] Understandably, setting up two independent evaluation branches is based on a divide-and-conquer modeling strategy. While both the activity and stability of functional factors are closely related to their quality, they exhibit different mechanisms: activity emphasizes the strength of the functional factor's role in physiological or chemical reactions, while stability emphasizes its resistance to degradation during preservation, processing, or application. By setting separate activity and stability evaluation branches, feature interference caused by mixing these two types of properties in the model can be avoided, making the model more targeted and accurate in its respective subtasks. Secondly, each branch consists of a Multilayer Perceptron (MLP) module, whose technical principle lies in nonlinear feature mapping and pattern learning. A MLP is composed of multiple fully connected layers stacked together, capable of progressively transforming the global dynamic feature representation of the input into a higher-level, more abstract feature space representation. By introducing activation functions, the model can capture the complex nonlinear relationship between input features and target scores. Simultaneously, adding regularization constraints (such as dropout or L2 regularization) between layers can effectively suppress overfitting, ensuring the model has better generalization ability, thereby improving the prediction performance on new samples. Secondly, a "label-driven branch update" mechanism is employed during the training phase. This mechanism improves the model's learning efficiency and robustness through selective parameter updates. Specifically, when an input sample carries an activity category label, only the parameters of the activity evaluation branch are updated, while the parameters of the stability branch are frozen, and vice versa. This mechanism avoids interference from irrelevant labels on other branches, ensuring that each branch focuses on optimizing its own task, thus improving the independence and accuracy of the scores. Finally, during the prediction phase, the global dynamic feature representation is input into two independent branches, generating activity and stability scores respectively. This process relies on parallelized independent inference; the same feature representation passes through two network paths with different parameterizations, outputting evaluation results for the corresponding dimensions. This not only ensures the independence of activity and stability scores but also maintains their consistency and comparability, providing a foundation for subsequent comprehensive quality evaluation.

[0041] In specific embodiments of this application, the above steps are implemented as follows: In the activity assessment branch, certain polypeptide functional factors in salamander byproducts can be considered, whose main function is antioxidation. Researchers input the dynamic features obtained from multimodal fusion and time-series modeling into the activity assessment branch. The model then learns the correspondence between these features and antioxidant capacity (such as free radical scavenging ability) layer by layer. For example, if the input samples during the training phase are labeled "high activity," "medium activity," or "low activity," the model only updates the activity branch parameters, enabling the branch to accurately output the corresponding activity score. During prediction, a new sample passing through this branch will obtain a quantified activity score, such as 0.82 (out of 1), indicating strong antioxidant activity. In the stability assessment branch, the degradation behavior of the same polypeptide functional factors under different storage conditions can be considered. For example, some polypeptides degrade rapidly within 3 days at room temperature, while remaining stable for more than 7 days under low-temperature conditions. During the training phase, input data is labeled with tags such as "high stability" and "low stability." The model only updates the parameters of the stability branch, focusing this branch on capturing features related to degradation rate, half-life, etc. During prediction, a new sample passing through this branch can obtain a stability score, for example, 0.65, reflecting moderate tolerance during storage and application. When used in combination, the global dynamic feature representation of a salamander byproduct sample is simultaneously input into both branches; the activity branch might output 0.82, while the stability branch outputs 0.65. Such independent scoring results not only describe the efficacy and durability of functional factors separately but also provide an intuitive and quantifiable basis for subsequent comprehensive quality evaluation (such as weighted fusion into a final score).

[0042] As can be seen, by setting two independent evaluation branches to score the activity and stability of functional factors respectively, targeted modeling can be achieved, avoiding interference caused by feature coupling. Furthermore, selective updating of branch parameters during the training phase effectively improves the model's convergence efficiency and classification accuracy. Simultaneously, each branch employs a multilayer perceptron structure combined with regularization constraints, enhancing the model's generalization ability and reducing the risk of overfitting. Therefore, this method can generate activity and stability scores more accurately and stably, providing a reliable basis for the comprehensive evaluation of functional factor quality.

[0043] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0044] Step S500: Divide the biomarker dataset into several functional domains according to the functional factor categories. Construct evaluation curves for each functional domain based on the relationship between historical experimental data and functional factors. Substitute the biomarker data of the target sample into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results. Generate a comprehensive evaluation value through a weighted fusion strategy.

[0045] Specifically, when dividing the biomarker dataset into several functional domains according to the functional factor category, the process includes: dividing the functional factors into several functional domains using the biological activity of the functional factors as variables, and constructing a hierarchical structure according to the intensity of action; within each functional domain, performing regression fitting with historical experimental data as independent variables and functional factor activity as dependent variables to generate an evaluation curve.

[0046] Specifically, when substituting the marker data of the target sample into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results, the process includes: determining the functional domain to which the target sample belongs and its neighboring functional domains; substituting the marker data of the target sample into the corresponding evaluation curves to generate multiple functional domain evaluation results; calculating the weights based on the distance between the center point of each functional domain and the target sample; and weighting and summing the functional domain evaluation results to generate a comprehensive evaluation value.

[0047] Understandably, by using the bioactivity of functional factors as the core variable, different categories of functional factors are hierarchically divided according to their intensity of action, thus establishing several functional domains. This domain division method essentially structures the complex feature space of functional factors, enabling factors with similar or identical functions to cluster in the same domain, reducing data heterogeneity and enhancing the relevance and comparability of the evaluation. Secondly, a principle for constructing evaluation curves is proposed. In each functional domain, historical experimental data is used as the independent variable, and the activity of the functional factor is used as the dependent variable. A regression method is used to fit the curve, resulting in an evaluation curve reflecting the "marker-activity" relationship within that functional domain. This means that each functional domain corresponds to a specific mathematical model curve used to characterize the mapping relationship between marker data and activity performance. Then, based on the neighborhood evaluation principle, the system not only substitutes the data of the target sample into the evaluation curve of its own functional domain, but also extends it to neighboring functional domains. This avoids the "jump effect" caused by functional domain boundary division, ensuring that samples at the functional domain boundaries can still be reasonably evaluated, while also introducing neighborhood information to enhance robustness. Finally, a weighted fusion principle is used. After obtaining evaluation results from multiple functional domains, the weight allocation is determined by calculating the distance between the target sample and the center point of each functional domain, thereby performing a weighted summation of the evaluation results for different functional domains. The essence of this strategy is weighted interpolation based on spatial distance, which enables a smooth transition between results from multiple functional domains, improving the continuity and accuracy of the evaluation results.

[0048] In a specific embodiment of this application, the above steps are implemented as follows: In the example of functional domain segmentation, it is assumed that the polypeptide functional factors in salamander byproducts are divided into three functional domains—high, medium, and low—based on their antioxidant activity. Each functional domain contains several polypeptides with similar activity levels. A regression model of independent and dependent variables is established using historical experimental data (such as the activity values ​​of different polypeptides in a standard free radical scavenging experiment) to generate an evaluation curve for each functional domain. For example, the regression curve for the high-activity functional domain can depict the relationship between biomarker concentration and free radical scavenging rate, while the medium-activity and low-activity domains correspond to curves with different slopes or intercepts. Secondly, in the example of target sample evaluation, it is assumed that the activity characteristics of a sample to be tested are measured to be at a medium level. The system first determines that its functional domain is a medium-activity domain, and simultaneously identifies its neighboring high-activity and low-activity domains. Subsequently, the biomarker data of the sample is substituted into the evaluation curves of these three functional domains to obtain three sets of functional domain evaluation results, for example, the medium-activity domain score is 0.65, the high-activity neighbor score is 0.58, and the low-activity neighbor score is 0.62. Finally, in the weighted fusion example, the system calculates the weights based on the distance between the target sample and the center points of each functional domain. Assuming the weight of the active domain is 0.5, the weight of the highly active neighborhood is 0.25, and the weight of the less active neighborhood is 0.25, the final comprehensive evaluation value is obtained by weighted summation: 0.65*0.5 + 0.58*0.25 + 0.62*0.25 = 0.618. This method effectively fuses information from multiple functional domains, ensuring that the final evaluation result considers both the primary functional domain to which the sample belongs and reference data from neighboring functional domains, thus obtaining a more scientific and stable comprehensive quality score.

[0049] It can be seen that by dividing functional domains to achieve data partitioning and modeling, by using regression curves to depict the mapping relationship between markers and activity, and by combining neighborhood evaluation and weighted fusion, a more reasonable and stable comprehensive evaluation value can be obtained.

[0050] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0051] Step S600: The final quality evaluation result is obtained by weighting and integrating the activity score, stability score and comprehensive evaluation value.

[0052] Specifically, the final quality evaluation result is obtained by weighted integration of the activity score, stability score, and comprehensive evaluation value. This includes: calculating performance indicators for each of the activity score, stability score, and comprehensive evaluation value, and determining subjective and objective weights based on these indicators; generating combined weights using weight normalization and non-negativity constraints, where the sum of all weights in the combined weights is 1; determining the quality score of the giant salamander byproduct based on the relationship between the combined weights and the activity score, stability score, and comprehensive evaluation value; and determining the final quality evaluation result of the giant salamander byproduct based on the relationship between the quality score and a pre-set quality score: if the quality score is lower than the pre-set quality score, the final quality evaluation of the giant salamander byproduct is deemed unqualified; if the quality score is higher than or equal to the pre-set quality score, the final quality evaluation of the giant salamander byproduct is deemed qualified.

[0053] Understandably, the process begins with performance index calculations. Performance indices are calculated separately for the activity score, stability score, and overall evaluation score to quantify the reliability and contribution of each score to the overall quality evaluation. Performance indices may include the variance of the score, historical fluctuation range, detection repeatability, and correlation with the actual effect of the functional factor. These indices identify which scores are more critical in reflecting the quality of the functional factor, providing a scientific basis for subsequent weight allocation and preventing fluctuations or instability in certain indices from affecting the accuracy of the overall evaluation. Secondly, weight allocation and combined weight generation are employed. After obtaining the performance indices for each score, a strategy combining subjective and objective weights is introduced. Subjective weights can be determined based on professional experience, industry standards, or laboratory validation results, reflecting the importance of certain scores in actual quality control according to experts. Objective weights are based on data-driven methods, such as the standard deviation of the score, sensitivity analysis, or information entropy calculation, reflecting the reliability and discriminative power of the score itself. Through normalization and non-negativity constraint processing, subjective and objective weights are integrated into a combined weight, ensuring that the sum of all weights is 1. This guarantees a consistent total contribution of the scores and avoids evaluation imbalance caused by excessively large or small individual weights. Next, a combined scoring calculation is performed. The activity score, stability score, and comprehensive evaluation score are multiplied by their respective combined weights and summed to obtain an overall quality score. This comprehensive score is a quantitative representation of multidimensional indicators on a single numerical value, simultaneously reflecting the activity strength, stability changes, and performance of functional factors in various functional domains. This method transforms previously scattered and complex data into intuitive and comparable quantitative results, providing a foundation for quality assessment. Finally, the comprehensive quality score is compared with preset quality standards. A score below the standard is deemed unqualified, while a score meeting or exceeding the standard is deemed qualified. This mechanism not only provides clear quality classification results but also allows for flexible adaptation by adjusting preset standards according to different application scenarios. Through this judgment logic, the entire evaluation method forms a closed loop from multidimensional data collection, feature extraction, dynamic modeling, weighted integration to final judgment, making the quality evaluation process scientific, reliable, and operable.

[0054] In a specific embodiment of this application, the above steps are implemented as follows: In the performance index calculation stage, it is assumed that the activity score of a batch of giant salamander by-products is 85, the stability score is 90, and the comprehensive evaluation score is 88. In the laboratory, the standard deviation, coefficient of variation, or information entropy of each score can be calculated using historical experimental data and repeated test results to quantify its reliability. For example, if the activity score fluctuates significantly in multiple experiments, while the stability score is relatively stable, the objective weight of the activity score is relatively reduced, and the weight of the stability score is relatively increased, thereby reflecting the actual reliability of the data. Secondly, in the weight normalization and combined weight generation stage, subjective weights can be set as expert experience allocation values, such as 0.4 for the activity score, 0.35 for the stability score, and 0.25 for the comprehensive evaluation; the objective weights are calculated to be 0.35 for the activity score, 0.4 for the stability score, and 0.25 for the comprehensive evaluation. After the subjective weights and objective weights are merged and normalized, combined weights are generated, for example, the final combined weights are 0.375 for activity, 0.375 for stability, and 0.25 for comprehensive evaluation. This process ensures that the contribution of different scores to the overall score reflects both expert experience and data reliability. Next, in the combined score calculation stage, the quality score is calculated using combined weights: Quality Score = 85 × 0.375 + 90 × 0.375 + 88 × 0.25 = 87.125. Through weighted integration, the multi-dimensional scoring indicators are uniformly converted into a single value, facilitating subsequent judgment and comparison. Finally, in the final quality evaluation stage, assuming the preset quality score standard is 85 points, the calculated quality score of 87.125 is higher than the preset value, therefore the final quality evaluation of this batch of giant salamander by-products is judged as "qualified". If the calculated result is lower than the preset value, such as 82 points, it is judged as "unqualified". This example demonstrates the complete process from raw scoring to combined weights and then to final quality judgment, intuitively reflecting the quantitative and integrative role of multi-dimensional scoring indicators, while achieving a scientific and operable quality evaluation by combining preset standards.

[0055] It can be seen that through performance index analysis, scientific weight allocation, multi-dimensional weighted integration, and standardized judgment, the functional factors of giant salamander by-products have been quantitatively evaluated and their final quality judged.

[0056] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0057] In the above embodiments, comprehensive information on the chemical composition and structural characteristics of functional factors is obtained through biochemical analysis and molecular detection of biomarkers. Combined with a multimodal feature fusion module, information from biomarkers of different sources and types can be integrated to extract the activity and stability characteristics of functional factors. Furthermore, time-series dynamic modeling captures the changing patterns of functional factors at different time points. This technique overcomes the limitations of traditional methods that rely solely on a single indicator or a single time point, making the evaluation results more comprehensive and scientific. Secondly, by setting independent evaluation branches, the activity and stability of functional factors are scored separately, and historical experimental data and functional domain evaluation curves are introduced to achieve quantitative analysis of the performance of functional factors in different functional domains. Biomarker data of the target sample can be simultaneously substituted into the evaluation curves of its functional domain and neighboring functional domains, and a comprehensive evaluation value is generated through a weighted fusion strategy. This effectively solves the problem that a single indicator evaluation cannot reflect the overall quality of functional factors, enhancing the accuracy and reliability of the evaluation. Finally, by weighting and integrating the activity score, stability score, and comprehensive evaluation value, a combined weight is generated using a fusion of subjective and objective weights to quantitatively calculate the final quality evaluation result. This not only ensures the reasonable contribution of information from different dimensions to the evaluation results, but also allows for dynamic adjustments based on actual experimental data and historical standards, making the final evaluation results more scientific, objective, and comparable.

[0058] In another preferred embodiment based on the above embodiments, such as Figure 3 As shown, this embodiment provides a quality evaluation system for functional factors of salamander by-products, including: a preprocessing unit, a detection unit, a feature extraction unit, an evaluation unit, a functional domain division unit, and an integration unit.

[0059] Specifically, the preprocessing unit is configured to acquire and preprocess samples of salamander byproducts, extracting functional factors to obtain samples to be tested; the detection unit is configured to identify and quantify specific biomarkers in the samples to be tested using biochemical analysis and molecular detection methods, generating a biomarker dataset; the feature extraction unit is configured to input the biomarker dataset into a multimodal feature fusion module, extracting the activity and stability features of functional factors, and dynamically modeling the features based on time series; the evaluation unit is configured to set two independent evaluation branches to score the activity and stability of functional factors, generating activity and stability scores; the functional domain segmentation unit is configured to divide the biomarker dataset into several functional domains according to the functional factor categories, construct evaluation curves for each functional domain based on historical experimental data and the relationship between functional factors, substitute the biomarker data of the target sample into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain functional domain evaluation results, and generate a comprehensive evaluation value through a weighted fusion strategy; the integration unit is configured to integrate the activity score, stability score, and comprehensive evaluation value using weighted methods to obtain the final quality evaluation result.

[0060] It is understood that the quality evaluation system and method for functional factors of salamander by-products in the above embodiments have the same beneficial effects, and will not be described in detail here.

[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for quality evaluation of functional factors derived from by-products of giant salamanders, characterized in that, include: Samples of by-products from giant salamanders were obtained and preprocessed to extract functional factors and obtain samples to be tested. Based on biochemical analysis and molecular detection methods, specific biomarkers in the sample to be tested are identified and quantified to generate a biomarker dataset. The biomarker dataset is input into the multimodal feature fusion module to extract the activity and stability features of functional factors, and the features are dynamically modeled based on time series. Two independent evaluation branches are set up to score the activity and stability of the functional factors respectively, and generate activity score and stability score. The biomarker dataset is divided into several functional domains according to the functional factor categories. Evaluation curves for each functional domain are constructed based on the relationship between historical experimental data and functional factors. The biomarker data of the target sample are substituted into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results. A comprehensive evaluation value is generated through a weighted fusion strategy. The final quality evaluation result is obtained by weighting and integrating the activity score, stability score, and comprehensive evaluation value. When inputting the biomarker dataset into the multimodal feature fusion module to extract the activity and stability features of functional factors, the following are included: The biomarker dataset was classified and coded according to functional factor categories, and the data of each category were normalized. Based on feature extraction of normalized data using a multi-channel sensor array, activity and stability features are obtained. The dynamic time warping algorithm is used to model the features in time series and generate dynamic feature sequences. When dynamically modeling features based on time series data, the following are included: The dynamic feature sequence is input into the recurrent neural network module to capture the time-dependent features of the functional factors; The importance score of the feature at each time step is calculated based on the attention mechanism, and the score is normalized to obtain the time weight. The time-step features are weighted and aggregated according to the time weight to generate a global dynamic feature representation; When setting up two independent evaluation branches to score the activity and stability of functional factors, the following are included: An activity evaluation branch and a stability evaluation branch are set up, and each branch is composed of a multilayer perceptron module; Each branch includes at least two fully connected layers, and regularization constraints are introduced between the layers to suppress overfitting; During the training phase, only the parameters of the corresponding branch are updated based on the functional factor category label, while the other branch is frozen. During the prediction phase, the global dynamic feature representation is input into the two branches respectively to generate activity score and stability score.

2. The method for quality evaluation of functional factors derived from salamander by-products as described in claim 1, characterized in that, Pretreatment of salamander byproduct samples includes: The crude extract was obtained by preliminary separation of the by-products of giant salamanders using a fractional extraction method. The crude extract was subjected to ultrasonic crushing, low-temperature centrifugation and membrane filtration in sequence to remove impurities and enrich functional factors. The functional factors were purified by gradient elution liquid chromatography, and the fraction corresponding to the target peak was collected to obtain the sample to be tested.

3. The method for quality evaluation of functional factors derived from salamander by-products as described in claim 2, characterized in that, When identifying and quantifying specific biomarkers in samples based on biochemical analysis and molecular detection methods, this includes: Qualitative and quantitative analysis of small molecule biomarkers in samples to be tested was performed using high performance liquid chromatography-mass spectrometry (HPLC-MS). Specific identification and concentration determination of macromolecular biomarkers based on immunoassay technology; The structural characteristics of functional factors are verified using spectral analysis techniques, and a biomarker dataset containing information on multiple biomarkers is generated.

4. The method for quality evaluation of functional factors derived from salamander by-products as described in claim 1, characterized in that, When dividing a biomarker dataset into several functional domains according to functional factor categories, the following are included: Using the biological activity of functional factors as a variable, functional factors are divided into several functional domains, and a hierarchical structure is constructed according to the intensity of their effects. Within each functional domain, regression fitting is performed using historical experimental data as independent variables and functional factor activity as dependent variables to generate evaluation curves.

5. The method for quality evaluation of functional factors derived from salamander by-products as described in claim 1, characterized in that, When substituting the biomarker data of the target sample into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results, the following steps are included: Determine the functional domain to which the target sample belongs and its neighboring functional domains, and substitute the biomarker data of the target sample into the corresponding evaluation curves to generate multiple functional domain evaluation results. The weights are calculated based on the distance between the center point of each functional domain and the target sample. The evaluation results of the functional domains are then summed in a weighted manner to generate a comprehensive evaluation value.

6. The method for quality evaluation of functional factors derived from salamander by-products as described in claim 5, characterized in that, The final quality evaluation result is obtained by weighting and integrating the activity score, stability score, and comprehensive evaluation score, including: Performance indicators were calculated for the activity score, stability score, and comprehensive evaluation value, and subjective and objective weights were determined based on the performance indicators. Weights are generated by using weight normalization and non-negativity constraints, where the sum of all weights in the combined weights is 1. The quality score of the salamander by-products is determined based on the relationship between the combined weights and the activity score, stability score, and comprehensive evaluation value. Based on the relationship between the quality score and the preset quality score, the final quality evaluation result of the salamander by-product is determined: If the quality score is lower than the preset quality score, the final quality evaluation of the giant salamander by-product is determined to be unqualified. When the quality score is higher than or equal to the preset quality score, the final quality evaluation of the giant salamander by-product is determined to be qualified.

7. A quality evaluation system for functional factors derived from by-products of giant salamanders, used to apply the quality evaluation method for functional factors derived from by-products of giant salamanders as described in any one of claims 1 to 6, characterized in that, include: The preprocessing unit is configured to acquire and preprocess samples of salamander byproducts, extract functional factors from them, and obtain samples to be tested. The detection unit is configured to identify and quantify specific biomarkers in the sample to be tested through biochemical analysis and molecular detection methods, and generate a biomarker dataset. The feature extraction unit is configured to input the biomarker dataset into the multimodal feature fusion module, extract the activity and stability features of functional factors, and dynamically model the features based on time series. The evaluation unit is configured to set up two independent evaluation branches to score the activity and stability of the functional factors respectively, and generate activity score values ​​and stability score values. The functional domain partitioning unit is configured to divide the biomarker dataset into several functional domains according to the functional factor category, construct evaluation curves for each functional domain based on the relationship between historical experimental data and functional factors, substitute the biomarker data of the target sample into the evaluation curves of the corresponding functional domain and its neighboring functional domains to obtain the functional domain evaluation results, and generate a comprehensive evaluation value through a weighted fusion strategy. The integration unit is configured to weight and integrate the activity score, stability score, and comprehensive evaluation value to obtain the final quality evaluation result.

Images