A method and system for evaluating liver protection of TFF3 based on NF-kappa B pathway inhibition

Abstract

The present application relates to the technical field of biological liver protection, solves the technical problem that the prior art cannot capture the complex nonlinear biological relationship in ALD, lacks biological mechanism support, and researchers have difficulty understanding and trusting the prediction results, and particularly relates to a TFF3 liver protection experimental test method and system based on NF-KB pathway inhibition, the method comprising the following steps: S1, obtaining the original data of the liver protection experiment, preprocessing and data segmentation of the original data to obtain a training set and a test set, the present application effectively captures the basic linear relationship between the TFF3 inhibition effect and the liver injury marker through multiple linear regression, provides interpretable coefficients and clear action direction, the liver protection experiment score plays a key role in quantitative evaluation and clinical transformation bridge, provides intuitive treatment decision support for researchers, improves the interpretability of experimental results, and can also generate targeted protection effect for individual patients.

Description

Technical Field

[0001] This invention relates to the field of biological liver protection technology, and in particular to an experimental testing method and system for evaluating TFF3 liver protection based on NF-κB pathway inhibition. Background Technology

[0002] NF-κB is a family of transcription factor proteins widely distributed in animal cells. These proteins dimerize to form transcription factors, regulate gene expression, and participate in various biological processes. Inhibiting NF-κB can reduce inflammatory responses and is used to treat inflammatory diseases such as rheumatoid arthritis and inflammatory bowel disease. TFF3 is a member of the trefoil factor family, composed of 59 amino acids. Its molecule forms three intrachain disulfide bridges through six conserved cysteine ​​residues. It is mainly distributed in the epithelial region of the small and large intestines and has gastrointestinal mucosal protection and repair functions. It also plays an important role in liver protection. Current technologies cannot capture the complex nonlinear biological relationships in ALD when conducting NF-κB pathway inhibition assessments and TFF3 liver protection experiments, and lack biological mechanism support. Most importantly, researchers find it difficult to understand and trust its predictive results. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and system for evaluating TFF3 liver protection based on NF-κB pathway inhibition. This solves the technical problems of existing technologies failing to capture the complex nonlinear biological relationships in ALD, lacking biological mechanism support, and making it difficult for researchers to understand and trust the predicted results. It achieves the goals of providing interpretable coefficients and clear action directions, increasing support for biological mechanisms, improving the interpretability of experimental results, and facilitating researchers' understanding and trust.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for evaluating the liver protection of TFF3 based on NF-κB pathway inhibition, the method comprising the following steps:

[0005] S1. Obtain the raw data of the liver protection experiment, preprocess the raw data and split the data to obtain the training set and test set;

[0006] S2. Based on the training set, the data is processed through feature decomposition and importance evaluation to obtain the important feature set. Based on the important feature set, the standard dataset is obtained by Z-score standardization.

[0007] S3. Construct a dynamic simulation model based on the standard dataset for dynamic correction and accurate prediction of liver protection experimental data, optimize the dynamic simulation model to obtain an optimized dynamic model, and obtain activity prediction data based on the optimized dynamic model.

[0008] S4. Obtain the TFF3 concentration gradient from the standard dataset, and obtain the quantification index set by sequentially using the least squares method and Bootstrap resampling based on the TFF3 concentration gradient and activity prediction data.

[0009] S5. Based on the quantitative index set, regression prediction and decision prediction are obtained through multiple linear regression combined with optimized decision tree. Based on the regression prediction and decision prediction, the integrated prediction value is obtained through the weighted average formula, and the liver protection experimental score for evaluating the liver protection effect is calculated through the linear transformation formula.

[0010] S6. Based on the integrated predicted values ​​and liver protection experiment scores, a liver protection effect prediction model is constructed to predict the liver protection experiment and form an intuitive experimental effect. The parameters are optimized by grid search method, and the model is calibrated by cross-validation method to obtain the calibrated effect model.

[0011] S7. Input the damage dataset and the set of quantitative indicators into the calibration effect model to obtain experimental results, and generate an experimental report based on the experimental results.

[0012] Furthermore, in S1, the specific implementation steps are as follows:

[0013] S11. Collect raw data from historical data of liver protection experiments, and remove outliers from the raw data using the outlier detection formula to obtain cleaned data.

[0014] S12. Obtain the fill value from the cleaned data using the mean interpolation formula, insert the fill value into the corresponding position in the cleaned data to obtain the fill data, and perform standardization processing on the fill data using the maximum and minimum standardization formula to obtain the standardized data.

[0015] S13. Divide the standardized data into training and test sets by random partitioning.

[0016] Furthermore, in S2, the specific implementation steps are as follows:

[0017] S21. Based on the data in the training set, principal component analysis is used to perform feature decomposition to obtain the principal component feature set.

[0018] S22. Construct a random forest model for obtaining importance data based on the principal component feature set. Input the principal component feature set into the random forest model to obtain the importance evaluation value. Sort the important feature sets in descending order according to the importance evaluation value to obtain the top k important feature sets.

[0019] S23. Obtain the standard dataset based on the important feature set using the Z-score standardization formula.

[0020] Furthermore, in S3, the specific implementation steps are as follows:

[0021] S31. Obtain the initial parameters of the data dynamics model for simulating biological experiments using the maximum likelihood estimation method based on the standard dataset;

[0022] S32. Based on the initial parameters, construct an activity dynamic model for simulating NF-κB activity using the RK4 integral method. Input the standard dataset into the activity dynamic model to obtain the activity time series, and construct the activity dynamic layer.

[0023] S33. Obtain real data from the liver protection experiment. Based on the activity time series obtained from the activity dynamic model, construct a neural network model to correct the liver protection experiment data. Train the neural network model using a loss function and backpropagation algorithm based on the activity time series and real data to obtain the updated dynamic layer.

[0024] S34. Input the active time series and real data into the update dynamic layer to obtain the update time series. Correct the update time series through forward propagation to obtain the corrected data, and obtain the neural network layer.

[0025] S35. Integrate the active dynamic layer, the update dynamic layer, and the neural network layer into a dynamic simulation model. Determine the verification threshold based on the root mean square error formula. Obtain the evaluation value through K-fold cross-validation based on the correction data. Validate the dynamic simulation model based on the evaluation value and the verification threshold to obtain the optimized dynamic model.

[0026] If the evaluation value is greater than or equal to the verification threshold, the dynamic simulation model has a large deviation, and the process returns to S31 to readjust the parameters of the dynamic simulation model.

[0027] If the evaluation value is less than the verification threshold, the current dynamic simulation model will be used as the optimized dynamic model.

[0028] S36. Input the standard dataset into the optimized dynamic model to obtain the activity prediction data.

[0029] Furthermore, in S4, the specific implementation steps are as follows:

[0030] S41. Obtain the TFF3 concentration gradient from the standard dataset, and integrate the activity prediction data and the TFF3 concentration gradient into a dose response dataset using the data aggregation method.

[0031] S42. Construct a two-dimensional rectangular coordinate system with the TFF3 concentration gradient as the independent variable and the activity prediction data as the dependent variable. Obtain the fitting parameters using the nonlinear least squares method based on the dose response dataset.

[0032] S43. Obtain the fitting model based on the fitting parameters and the dose response dataset;

[0033] S44. Based on the fitting parameters of the fitted model, perform multiple resampling operations using Bootstrap resampling to obtain a resampled sample set;

[0034] S45. Confidence intervals are obtained from the resampled sample set using the percentile method. Based on the confidence intervals and fitting parameters, the suppression efficiency score is calculated using the suppression efficiency calculation formula. The suppression efficiency score, confidence intervals, and fitting parameters are then integrated into a set of quantitative indicators.

[0035] Furthermore, in S5, the specific implementation steps are as follows:

[0036] S51. Generate an injury dataset based on liver injury biomarkers in the standard dataset. Obtain regression prediction and regression residuals using a multiple linear regression formula based on the injury dataset and the set of quantitative indicators.

[0037] S52. Based on the damage dataset and the quantization index set, damage standard data and quantization standard data are obtained by feature standardization and corresponding vector standardization, respectively. Based on the damage standard data and quantization standard data, the interaction term is calculated by vector multiplication to obtain the input matrix.

[0038] S53. Construct a decision tree model based on the input matrix and regression residuals. Obtain the initial hyperparameters through the maximum likelihood estimation method. Obtain the information gain and the optimal split point based on the input matrix and regression residuals through the information gain formula and the optimal split point calculation formula, respectively.

[0039] S54. Based on information gain and the optimal split point, construct an initial decision tree using a greedy algorithm. Calculate the cost complexity of each leaf node in the initial decision tree. Sort the multiple cost complexities in descending order and prune the 20% of leaf nodes with the lowest cost complexity to obtain the optimized decision tree.

[0040] S55. Input the input matrix and regression residuals into the optimization decision tree to obtain the decision prediction. Based on the regression prediction and the decision prediction, obtain the integrated prediction value through the weighted average formula.

[0041] S56. Calculate the liver protection experimental score for evaluating the liver protection effect using a linear transformation formula based on the integrated predicted values.

[0042] Furthermore, in S6, the specific implementation steps are as follows:

[0043] S61. Based on S5, construct a liver protection effect prediction model and obtain the initial parameters of the model. Based on the initial parameters, obtain the total effect index through the Sobol formula. Based on the total effect index, sort the initial parameters in descending order and take the larger 20% as the sensitive parameters.

[0044] S62. Adjustment parameters are obtained by using a grid search method based on the sensitive parameters;

[0045] S63. Replace the initial parameters in the liver protection effect prediction model with the adjustment parameters to obtain the optimized effect model. Then, calibrate the optimized effect model using cross-validation to obtain the calibrated effect model.

[0046] Furthermore, in S7, the specific implementation steps are as follows:

[0047] S71. Input the damage dataset and the quantitative index set into the calibration effect model to obtain experimental results, and obtain the experimental data table based on the experimental results through data aggregation.

[0048] S72. Based on the experimental data table, plot the inhibition curve and liver protection score using Matplotlib;

[0049] S73. Generate an experimental report by combining the experimental data table, inhibition curve, and liver protection score chart.

[0050] The technical solution also provides a system for the above-mentioned experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition, the system comprising:

[0051] The preprocessing module is used to acquire the raw data of the liver protection experiment, and to preprocess and segment the raw data to obtain the training set and the test set.

[0052] The standardization module is used to process the data based on the training set through feature decomposition and importance evaluation to obtain an important feature set, and then to obtain a standard dataset by Z-score standardization based on the important feature set;

[0053] The activity prediction module is used to construct a dynamic simulation model for dynamic correction and accurate prediction of liver protection experimental data based on a standard dataset, optimize the dynamic simulation model to obtain an optimized dynamic model, and obtain activity prediction data based on the optimized dynamic model.

[0054] The quantization module is used to obtain the TFF3 concentration gradient from the standard dataset, and obtain the quantization index set by sequentially using the least squares method and Bootstrap resampling based on the TFF3 concentration gradient and activity prediction data.

[0055] The experimental scoring module is used to obtain regression prediction and decision prediction based on the quantitative index set through multiple linear regression combined with optimized decision tree, to obtain integrated prediction value based on the regression prediction and decision prediction through weighted average formula, and to calculate the liver protection experimental score for evaluating the liver protection effect through linear transformation formula.

[0056] The calibration module is used to construct a liver protection effect prediction model based on integrated predicted values ​​and liver protection experiment scores to predict the liver protection experiment and form an intuitive experimental effect. The parameter optimization is performed by grid search method, and the model is calibrated by cross-validation method to obtain the calibration effect model.

[0057] The reporting module is used to input the damage dataset and the set of quantitative indicators into the calibration effect model to obtain experimental results, and generate an experimental report based on the experimental results.

[0058] By employing the above technical solution, the present invention provides a method and system for evaluating TFF3 liver protection based on NF-κB pathway inhibition, which has at least the following beneficial effects:

[0059] 1. This invention organically integrates mechanistic and data-driven models, preserving the biological interpretability of the dynamic activity model while using dynamic simulation models to correct experimental errors and system biases. This achieves intelligent correction guided by the mechanism, improves processing efficiency, and effectively enhances prediction accuracy. Through innovative optimization of the dynamic model hierarchical structure, it filters out experimental noise and accurately captures nonlinear relationships. While maintaining biological significance, it can also provide more accurate NF-κB activity predictions, laying a reliable foundation for the subsequent quantification of TFF3 inhibition effects.

[0060] 2. This invention integrates TFF3 concentration gradient and activity prediction data into a dose-response dataset through data aggregation, achieving efficient data integration and orderly organization. It fully explores the correlation information between data, providing a solid foundation for subsequent analysis, accurately characterizing the complex relationship between TFF3 concentration gradient and activity prediction data, improving the accuracy of the fitting model, and fully considering the variability and uncertainty of the samples. The confidence interval obtained through the percentile method is more accurate and reliable. The inhibition efficiency score, confidence interval, and fitting parameters are integrated into a set of quantitative indicators, providing rich, accurate, and highly reliable quantitative evidence for in-depth research, and helping to more scientifically evaluate the results of liver protection experiments.

[0061] 3. This invention effectively captures the basic linear relationship between TFF3 inhibition effect and liver injury markers through multiple linear regression, providing interpretable coefficients and clear directions of action. Decision trees specifically handle nonlinear patterns and complex interactions in the data. The liver protection experimental score plays a crucial role in quantitative assessment and clinical translation, providing a standardized measure of TFF3 treatment efficacy, facilitating comparison and evaluation between different experiments and clinical studies, providing intuitive treatment decision support for researchers and medical staff, improving the interpretability of experimental results, and promoting personalized medicine. It can also generate targeted protective effects for individual patients.

[0062] 4. This invention adopts a systematic and multi-level optimization and calibration framework. Through a three-stage process of sensitivity analysis, parameter adjustment and model recalibration, it significantly improves the robustness of the model. The grid search method ensures a comprehensive exploration in the parameter space and avoids missing the optimal solution. At the same time, cross-validation ensures that the found parameter combination has good generalization ability, effectively prevents overfitting, and significantly improves the adaptability to new data, laying a solid foundation for the reliable transformation of research results. Attached Figure Description

[0063] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0064] Figure 1 This is a flowchart of an experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition according to the present invention;

[0065] Figure 2 This is a structural block diagram of an experimental testing system for evaluating TFF3 liver protection based on NF-κB pathway inhibition, according to the present invention.

[0066] In the diagram: 1. Preprocessing module; 2. Standardization module; 3. Activity prediction module; 4. Quantification module; 5. Experimental scoring module; 6. Calibration module; 7. Reporting module. Detailed Implementation

[0067] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0068] Because current technologies cannot capture the complex nonlinear biological relationships in ALD and lack supporting biological mechanisms, researchers struggle to understand and trust the technical issues behind their predictions. Please refer to [link / reference needed]. Figure 1 This embodiment proposes an experimental test method for evaluating the liver protection of TFF3 based on NF-κB pathway inhibition. This method provides interpretable coefficients and a clear direction of action, increases support for the biological mechanism, improves the interpretability of experimental results, and facilitates researchers' understanding and trust. The method includes the following steps:

[0069] S1. Obtain the raw data of the liver protection experiment, and preprocess and segment the raw data to obtain the training set and test set. Since the raw data is easily affected by the instability of the experimental environment or data acquisition equipment, resulting in large data fluctuations and the potential for abnormal data, the specific implementation method is as follows:

[0070] S11. Collect raw data from historical data of liver protection experiments. Remove outliers from the raw data using the outlier detection formula to obtain cleaned data. The raw data includes TFF3 concentration gradient, NF-κB phosphorylation level, TLR4 expression level, and liver injury marker ALT / AST measured in cell culture or mouse models. The outlier detection formula calculates the Z-score using the Z-score calculation formula. For example, if the Z-score of the data is greater than 3, it indicates that the data is an outlier and is removed. The removed data has missing positions, which will be filled in subsequent steps.

[0071] S12. Obtain the imputed value from the cleaned data using the mean interpolation formula, insert the imputed value into the corresponding position in the cleaned data to obtain the imputed data, and then perform standardization processing on the imputed data using the maximum and minimum standardization formula to obtain the standardized data. The mean interpolation formula calculates the mean of two adjacent data points to obtain an imputed value, which is a commonly used method for imputed missing values. The maximum and minimum standardization formula is a calculation method that uses the maximum and minimum values ​​of the data for standardization processing, which will not be elaborated here.

[0072] S13. Divide the standardized data into training and testing sets through random partitioning. Random partitioning is achieved by pre-setting the partition ratio, typically set to 3:7 in practice, with the testing set accounting for 30% and the training set for 70%. In practice, data cleaning and standardization are often used to handle mice in biological experiments, which improves data accuracy and processing efficiency, making subsequent steps more accurate and efficient.

[0073] S2. Based on the training set, the data is processed through feature decomposition and importance evaluation to obtain an important feature set. The standard dataset is then obtained through Z-score standardization based on this important feature set. In practice, it was found that features strongly correlated with the target variable are easily lost during the liver protection experiment, leading to significant bias in the experimental data processing results and a tendency for overfitting. To address this issue, the specific implementation steps are as follows:

[0074] S21. Based on the data in the training set, principal component analysis is used to perform feature decomposition to obtain the principal component feature set. Principal component analysis is an unsupervised dimensionality reduction and feature extraction technique. It projects high-dimensional data into a low-dimensional space through linear transformation, retaining the main features of the data, such as variance, while reducing redundant information. Its core objective is to replace the original variables with a few comprehensive variables, namely principal components, to simplify the data structure and reveal potential patterns. It is a commonly used feature extraction method, which will not be elaborated here.

[0075] S22. A random forest model for obtaining important data is constructed based on the principal component feature set. The principal component feature set is input into the random forest model to obtain importance evaluation values. The top k important features are obtained by sorting them in descending order according to the importance evaluation values. This is a machine learning algorithm based on ensemble learning, which improves model performance by constructing multiple decision trees and combining their prediction results. Random forest feature importance evaluation uses information automatically generated during model training to quantify the contribution of each feature to the model's predictive ability, thereby helping to understand the data, optimize the model, or perform feature selection. Generally, the importance of data is determined by calculating information gain or out-of-bag error, which is a commonly used method for judging data importance and will not be elaborated here.

[0076] S23. Obtain the standard dataset based on the important feature set using the Z-score standardization formula. The Z-score standardization formula is a commonly used data preprocessing method. By converting the data into a distribution with a mean of 0 and a standard deviation of 1, it eliminates dimensional differences and makes features at different scales comparable. This invention uses random forest to evaluate the actual importance of principal components, avoiding the sole reliance on variance to select features and preventing the loss of low-variance features strongly correlated with the target in experimental testing simulations. First, dimensionality reduction is performed using PCA to reduce the number of features, and then random forest is used to evaluate the importance of principal components, thereby improving the efficiency of liver protection experimental data processing and reducing the risk of overfitting.

[0077] S3. Construct a dynamic simulation model based on the standard dataset for dynamic correction and accurate prediction of liver protection experimental data. Optimize the dynamic simulation model to obtain an optimized dynamic model, and obtain activity prediction data based on the optimized dynamic model. Existing NF-κB pathway dynamic simulations have problems in accurately capturing noise, individual differences, and nonlinear responses in experimental systems. They also lack a biological mechanism basis and have weak extrapolation and prediction capabilities. To solve these problems, the specific steps are as follows:

[0078] S31. Based on the standard dataset, the initial parameters of the data dynamic model used to simulate biological experiments are obtained through the maximum likelihood estimation method. The maximum likelihood estimation method is a commonly used parameter estimation method in statistics. Its core idea is to infer the value of unknown parameters in the population distribution by maximizing the probability of the occurrence of sample data. It is a commonly used model parameter estimation method, which will not be elaborated here.

[0079] S32. Based on the initial parameters, construct an activity dynamic model for simulating NF-κB activity using the RK4 integration method. Input the standard dataset into the activity dynamic model to obtain the activity time series, and construct the activity dynamic layer. The RK4 integration method, also known as the fourth-order Runge-Kutta numerical integration method, is a numerical method that approximates the solution of the ordinary differential equation with high accuracy by calculating four intermediate slopes and weighting them in a single time step. The ordinary differential equation is the activity dynamic model used to simulate NF-κB activity.

[0080] S33. Obtain real data from the liver protection experiment. Based on the activity time series obtained from the activity dynamic model, construct a neural network model to correct the liver protection experiment data. Train the neural network model using a loss function and backpropagation algorithm based on the activity time series and real data to obtain the updated dynamic layer. The loss function is set as the mean squared error formula. The loss function and backpropagation algorithm are two core components in neural network training. They work closely together to optimize model parameters. The loss function defines the difference between the model prediction and the real data, while the backpropagation algorithm guides the parameter update direction by calculating the gradient of the loss function with respect to the parameters, thereby obtaining the updated dynamic model.

[0081] S34. Input the active time series and real data into the update dynamic layer to obtain the update time series. Correct the update time series through forward propagation to obtain the corrected data and obtain the neural network layer. Forward propagation refers to the process in which the input data starts from the input layer of the neural network, passes through the hidden layer and the output layer in sequence through linear transformation and nonlinear activation, and finally obtains the model prediction value. The process includes feature extraction, prediction generation and correction combined with the active time series.

[0082] S35. Integrate the active dynamic layer, the update dynamic layer, and the neural network layer into a dynamic simulation model. Determine the verification threshold based on the root mean square error formula. Obtain the evaluation value through K-fold cross-validation based on the correction data. Validate the dynamic simulation model based on the evaluation value and the verification threshold to obtain the optimized dynamic model.

[0083] If the evaluation value is greater than or equal to the verification threshold, the dynamic simulation model has a large deviation, and the process returns to S31 to readjust the parameters of the dynamic simulation model.

[0084] If the evaluated value is less than the validation threshold, the current dynamic simulation model is used as the optimized dynamic model. The root mean square error (RMSE) formula measures the square root of the mean square of the error between the predicted and true values, with units consistent with the target variable. K-fold cross-validation is a robust method for evaluating model performance by repeatedly splitting the dataset, particularly suitable for scenarios with limited data or where reducing evaluation randomness is necessary. Its core idea is to divide the dataset into K subsets (folds), and through K training-validation cycles, each using a different subset as the validation set and the rest as the training set, finally synthesizing all results to obtain a reliable estimate of the model performance. For example, the calibration data can be randomly divided into 5 subsets (folds) of similar size, ensuring that the sample distribution of each fold, such as the target variable range and class proportions, is approximately consistent. Through iterative training and validation, the model performance is estimated using the mean and standard deviation.

[0085] S36. Input the standard dataset into the optimized dynamic model to obtain activity prediction data. This invention organically integrates the mechanistic model and the data-driven model, preserving the biological interpretability of the activity dynamic model while using the dynamic simulation model to correct experimental errors and systematic biases. This achieves mechanism-guided intelligent correction, improves processing efficiency, and effectively enhances prediction accuracy. Through the innovative optimized dynamic model hierarchical structure, it filters out experimental noise and accurately captures nonlinear relationships. While maintaining biological significance, it can also provide more accurate NF-κB activity predictions, laying a reliable foundation for the subsequent quantification of TFF3 inhibition effects.

[0086] S4. Obtain the TFF3 concentration gradient from the standard dataset. Based on the TFF3 concentration gradient and activity prediction data, obtain the quantification index set through least squares and Bootstrap resampling. Existing techniques for processing TFF3 concentration gradient and activity prediction data lack a systematic data aggregation strategy, making it difficult to efficiently and accurately integrate the data into a dose-response dataset. This leads to insufficient utilization of data information, affecting the accuracy of subsequent analysis. Furthermore, relying on simple linear fitting methods is problematic because real-world data often exhibits non-linear characteristics, and linear fitting cannot accurately capture the complex relationships between variables, resulting in a significant deviation between the fitted model and the actual data. To address these issues, the specific implementation steps are as follows:

[0087] S41. Obtain the TFF3 concentration gradient from the standard dataset, and integrate the activity prediction data and TFF3 concentration gradient into a dose response dataset using the data aggregation method. The data aggregation method matches the activity prediction data and TFF3 concentration gradient by identifying the data, and creates a data table containing independent and dependent variables. After cleaning and handling missing values, multiple pairs of activity prediction data and TFF3 concentration gradient data are obtained, where the TFF3 concentration gradient is the independent variable and the activity prediction data is the dependent variable.

[0088] S42. Construct a two-dimensional rectangular coordinate system with the TFF3 concentration gradient as the independent variable and the activity prediction data as the dependent variable. Obtain the fitting parameters using the nonlinear least squares method based on the dose response dataset. The fitting parameters include the upper plateau, the lower plateau, the logarithm of the concentration corresponding to the half-maximal effect, the half-inhibition, and the Hill slope. The nonlinear least squares method uses a four-parameter logistic model to calculate and obtain the fitting parameters. The four-parameter logistic model is a commonly used method for obtaining fitting parameters, which will not be elaborated here.

[0089] S43. Based on the fitting parameters and the dose response dataset, a fitting model is obtained. The expression of the fitting model is:

[0090] ;

[0091] Where Y represents the dependent variable, X represents the independent variable, Bottom represents the lower platform, and Top represents the upper platform. IC50 represents the logarithm of the concentration at which the half-maximal effect occurs, IC50 represents the half-maximal inhibition, and Hill represents the Hill slope.

[0092] S44. Based on the fitting parameters of the fitted model, the Bootstrap resampling is used to complete multiple resampling operations and obtain a resampled sample set. The Bootstrap resampling process involves randomly and with replacement drawing the same number of data points as the dose response dataset, re-obtaining the fitting parameters through the Bootstrap statistic, and repeating this process multiple times. For example, in practice, the process is repeated 1000 times to generate the Bootstrap distribution, which is the resampled sample set.

[0093] S45. Confidence intervals are obtained from the resampled sample set using the percentile method. Based on the confidence intervals and fitting parameters, the inhibition efficiency score is calculated using the inhibition efficiency calculation formula. The inhibition efficiency score, confidence intervals, and fitting parameters are integrated into a quantitative index set. This invention integrates TFF3 concentration gradient and activity prediction data into a dose-response dataset through data aggregation, achieving efficient data integration and orderly organization. It fully explores the correlation information between data, providing a solid foundation for subsequent analysis. The fitting parameters are obtained using the nonlinear least squares method, which can better adapt to the nonlinear characteristics of the data, accurately characterize the complex relationship between TFF3 concentration gradient and activity prediction data, and improve the accuracy of the fitting model. Bootstrap resampling is used to complete multiple resampling and obtain a resampled sample set, fully considering the variability and uncertainty of the samples. The confidence intervals obtained through the percentile method are more accurate and reliable. The integration of inhibition efficiency score, confidence intervals, and fitting parameters into a quantitative index set comprehensively and systematically quantifies the impact of TFF3 concentration gradient on activity, providing rich, accurate, and highly reliable quantitative evidence for in-depth research, and helping to more scientifically evaluate the results of liver protection experiments.

[0094] S5. Based on the quantitative index set, regression predictions and decision predictions are obtained through multiple linear regression combined with optimized decision trees. The integrated prediction value is then obtained using a weighted average formula based on the regression and decision predictions, and a liver protection experimental score for evaluating the liver protection effect is calculated using a linear transformation formula. Existing technologies cannot capture the complex nonlinear biological relationships in ALD, such as the dose-response saturation effect of TFF3 and the interactions between liver injury biomarkers. Simply using complex machine learning models becomes a "black box," lacking biological mechanism support, and statistical models oversimplify real-world biological processes, resulting in significant prediction bias. While complex neural networks offer high accuracy, clinicians and researchers find it difficult to understand and trust their predictions. To address these issues, the specific implementation steps are as follows:

[0095] S51. Generate an injury dataset based on liver injury biomarkers in the standard dataset. Obtain regression predictions and regression residuals using the multiple linear regression formula based on the injury dataset and the quantitative index set. The multiple linear regression formula is a method used to analyze the linear relationship between a dependent variable and multiple independent variables. Multiple linear regression can obtain multiple regression coefficients, thereby obtaining the relationship between the dependent variable and the independent variables. The multiple linear regression formula is a commonly used calculation method for finding linear relationships. Its calculation steps include data preprocessing, feature interaction term creation, multiple linear regression fitting to obtain regression predictions, and obtaining regression residuals through residual calculation, which will not be elaborated here.

[0096] S52. Based on the damage dataset and the quantization index set, damage standard data and quantization standard data are obtained through feature standardization and corresponding vector standardization, respectively. The input matrix is ​​obtained by calculating the interaction term through vector multiplication based on the damage standard data and the quantization standard data. Feature standardization and corresponding vector standardization are standardization methods, and the calculation method is the same as the commonly used standardization methods, so they will not be described in detail here.

[0097] S53. Construct a decision tree model based on the input matrix and regression residuals. Obtain the initial hyperparameters using the maximum likelihood estimation method. Calculate the information gain and optimal split point using the information gain formula and optimal split point calculation formula, respectively, based on the input matrix and regression residuals. Maximum likelihood estimation is a statistical method that estimates model parameters by maximizing the probability of sample data occurrences, i.e., the likelihood function. The information gain formula is commonly used to measure the improvement in data purity after splitting a dataset using a certain feature. The optimal split point calculation formula is a method that finds the combination that maximizes the improvement in data purity after splitting among all possible features and values; these details will not be elaborated upon here.

[0098] S54. Based on information gain and the optimal split point, an initial decision tree is constructed using a greedy algorithm. The cost complexity of each leaf node in the initial decision tree is calculated, and the multiple cost complexities are sorted in descending order. The leaf nodes with the lowest cost complexity are pruned to obtain the optimized decision tree. The initial decision tree refers to the complete, unpruned decision tree constructed entirely from the training data before any optimization pruning steps. It is a commonly used prediction model, also known as weakest connection pruning, and is a post-pruning technique used to solve the overfitting problem of the initial decision tree. Its goal is to find an optimal balance between tree complexity and prediction accuracy by pruning unnecessary branches. Its core idea is to sacrifice a small amount of fit to the training data, i.e., increase the cost, in exchange for a significant reduction in model complexity.

[0099] S55. Input the input matrix and regression residuals into the optimization decision tree to obtain the decision prediction. Based on the regression prediction and the decision prediction, obtain the integrated prediction value through the weighted average formula. The weighted average formula is a method of summing multiple data by weighting coefficients, which will not be elaborated here.

[0100] S56. Calculate the liver protection experimental score for evaluating the liver protection effect using a linear transformation formula based on the integrated predicted values. The expression is:

[0101] ;

[0102] in, Indicates the liver protection test score. This represents the integrated forecast value. The maximum reference value for the degree of liver injury is generally taken as the maximum value after standardization of the liver injury marker. This invention effectively captures the basic linear relationship between TFF3 inhibition effect and liver injury marker through multiple linear regression, providing interpretable coefficients and clear directions of action. The decision tree is specifically designed to handle nonlinear patterns and complex interactions in the data, such as the synergistic effect between TFF3 and TLR4 receptors, and the threshold effect between liver injury markers. The liver protection experimental score plays a crucial role in quantitative assessment and clinical translation, providing a standardized measure of TFF3 treatment effect, facilitating comparison and evaluation between different experiments and clinical studies. It provides doctors with intuitive treatment decision support; high scores (e.g., >80 points) indicate significant TFF3 intervention effect, while low scores (e.g., <30 points) suggest the need to adjust the treatment plan. It transforms the molecular-level NF-κB inhibition effect into a directly understandable liver protection efficacy indicator, improving the interpretability of experimental results and facilitating the development of personalized medicine. It can also generate targeted protective effects for individual patients.

[0103] S6. A liver protection effect prediction model is constructed based on integrated predicted values ​​and liver protection experiment scores to predict liver protection experiments and form intuitive experimental results. Parameters are optimized using a grid search method, and the model is calibrated using cross-validation to obtain a calibrated effect model. Existing technologies generally use a single algorithm for simple parameter adjustment in liver protection experiment model tuning, lacking a rigorous cross-validation mechanism. To solve the above problems, the specific solution is as follows:

[0104] S61. Based on S5, construct a liver protection effect prediction model and obtain the initial parameters of the model. According to the initial parameters, obtain the total effect index through the Sobol formula. Based on the total effect index, sort the initial parameters in descending order and use the larger 20% as sensitive parameters. The Sobol formula is a method that calculates the contribution of each parameter and its interaction by decomposing the variance of the model output. It will not be elaborated here.

[0105] S62. Adjust the parameters by using the grid search method based on the sensitive parameters. The grid search method is a systematic hyperparameter optimization method that finds the optimal combination of parameters for model performance by exhaustively searching a predefined hyperparameter space. It will not be elaborated on here.

[0106] S63. The initial parameters in the liver protection effect prediction model are replaced with adjusted parameters to obtain the optimized effect model. The optimized effect model is then calibrated using cross-validation to obtain the calibrated effect model. In practice, k-fold cross-validation is used, a statistical method for evaluating the generalization ability of a model. By dividing the dataset into multiple subsets and training and validating the model multiple times, the evaluation bias caused by data partitioning is reduced. Its core idea is to simulate the model's performance on unknown data as much as possible using limited data, which is particularly suitable for scenarios with small samples or requiring high reliability evaluation. Further details are omitted here. This invention adopts a systematic, multi-level optimization and calibration framework. Through a three-stage process of sensitivity analysis, parameter adjustment, and model recalibration, it significantly improves the model's robustness. Sobol sensitivity analysis provides a clear direction for subsequent optimization, and the grid search method ensures comprehensive exploration in the parameter space, avoiding the omission of optimal solutions. Simultaneously, cross-validation ensures that the found parameter combinations have good generalization ability, effectively preventing overfitting and ensuring consistent model performance across different data subsets. This systematic approach enables the model to maintain prediction accuracy while significantly improving its adaptability to new data, laying a solid foundation for the reliable transformation of research results.

[0107] S7. Input the damage dataset and quantitative indicator set into the calibration effect model to obtain experimental results, and generate an experimental report based on the results. Existing experimental reports generally integrate experimental results using various file formats, but lack multi-dimensional and complex data relationship display, easily overlooking the different needs of different audiences. Clinicians, researchers, and decision-makers cannot quickly obtain the core information they are concerned with. To solve the above problems, the specific implementation steps are as follows:

[0108] S71. Input the damage dataset and quantitative index set into the calibration effect model to obtain experimental results. Based on the experimental results, obtain the experimental data table through data aggregation. Data aggregation is a technical process that collects, processes, and summarizes scattered data from multiple data sources. Its core purpose is to transform raw data into a structured and analyzable dataset, which facilitates the extraction of key information and supports decision-making.

[0109] S72. Based on the experimental data table, use Matplotlib to plot the inhibition curve and liver protection score map. Matplotlib is the most well-known 2D plotting library in Python, used to generate static, dynamic and interactive charts. It supports a variety of hard copy formats, such as PNG and PDF, and cross-platform environments, such as Jupyter Notebook and web applications. It can perform basic graph generation, adjustable axis scales, color mapping, multi-plot layout and dynamic interaction.

[0110] S73. The experimental report is generated by combining the experimental data table, inhibition curve, and liver protection score chart. While the generation of an experimental report from the combined data table, inhibition curve, and liver protection score chart typically uses a template-based design to automatically integrate all analytical results and charts, this invention achieves fully automated, multi-level intelligent report generation. Through standardized data aggregation, visualization generation, and report compilation processes, it ensures the consistency, accuracy, and reproducibility of the results. Employing multi-dimensional visualization technology, it clearly displays complex information such as the NF-κB inhibition curve, liver protection score distribution, and dose-response relationship. It visually presents the protective effect of TFF3 in a graphical manner, automatically integrating all analytical results and charts, significantly improving work efficiency and eliminating human error. Particularly important is that the report content design in this method caters to the needs of different audiences: providing clinicians with a concise executive summary and therapeutic significance, researchers with detailed methodological and statistical analysis, and decision-makers with cost-effectiveness and translational potential assessments. This systematic report generation system greatly promotes the dissemination of research results, peer review, and clinical translational applications.

[0111] Because existing technologies cannot capture the complex nonlinear biological relationships in ALD and lack support from biological mechanisms, researchers often struggle to understand and trust the predicted results. This embodiment also proposes a system for evaluating TFF3 liver protection experimental testing based on NF-κB pathway inhibition. This system can provide interpretable coefficients and clear directions of action, increasing support from biological mechanisms, improving the interpretability of experimental results, and facilitating researchers' understanding and trust. Figure 2 As shown, the system includes a preprocessing module 1, a standardization module 2, an activity prediction module 3, a quantification module 4, an experimental scoring module 5, a calibration module 6, and a reporting module 7.

[0112] Preprocessing module 1 acquires the raw data from the liver protection experiment, preprocesses and splits the raw data to obtain training and test sets; Standardization module 2 processes the data based on the training set through feature decomposition and importance assessment to obtain an important feature set, and standardizes the data using Z-score to obtain a standard dataset; Activity prediction module 3 constructs a dynamic simulation model based on the standard dataset for dynamic correction and accurate prediction of liver protection experiment data, optimizes the dynamic simulation model to obtain an optimized dynamic model, and obtains activity prediction data based on the optimized dynamic model; Quantization module 4 obtains the TFF3 concentration gradient from the standard dataset, and quantizes the TFF3 concentration gradient and activity prediction data sequentially using least squares and Bootstrap methods. The rap resampling process yields a set of quantitative indicators; the experimental scoring module 5 uses multiple linear regression combined with optimized decision trees to obtain regression and decision predictions based on the quantitative indicator set, then uses a weighted average formula to obtain an integrated predicted value, and calculates the liver protection experimental score for evaluating the liver protection effect using a linear transformation formula; the calibration module 6 uses the integrated predicted value and the liver protection experimental score to construct a liver protection effect prediction model to predict the liver protection experiment and form an intuitive experimental effect, optimizes parameters using a grid search method, and calibrates the model using cross-validation to obtain a calibration effect model; the reporting module 7 inputs the injury dataset and the quantitative indicator set into the calibration effect model to obtain experimental results, and generates an experimental report based on the experimental results.

[0113] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented 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.

[0114] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Since the above embodiments are substantially similar to the method embodiments, they are described relatively simply; relevant parts can be referred to the descriptions of the method embodiments.

[0115] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for assessing TFF3 liver protection based on NF-κB pathway inhibition, characterized in that, The method includes the following steps: S1. Obtain the raw data of the liver protection experiment, preprocess the raw data and split the data to obtain the training set and the test set; S2. The training set data is processed through feature decomposition and importance evaluation to obtain the important feature set. The standard dataset is obtained by standardizing the important feature set through Z-score. S3. Construct a dynamic simulation model based on the standard dataset for dynamic correction and accurate prediction of liver protection experimental data, optimize the dynamic simulation model to obtain an optimized dynamic model, and obtain activity prediction data based on the optimized dynamic model. S4. Obtain the TFF3 concentration gradient from the standard dataset, and obtain the quantification index set by sequentially using the least squares method and Bootstrap resampling based on the TFF3 concentration gradient and activity prediction data. S41. Obtain the TFF3 concentration gradient from the standard dataset, and integrate the activity prediction data and the TFF3 concentration gradient into a dose response dataset using the data aggregation method. S42. Construct a two-dimensional rectangular coordinate system with the TFF3 concentration gradient as the independent variable and the activity prediction data as the dependent variable. Obtain the fitting parameters using the nonlinear least squares method based on the dose response dataset. S43. Obtain the fitting model based on the fitting parameters and the dose response dataset; S44. Based on the fitting parameters of the fitted model, perform multiple resampling operations using Bootstrap resampling to obtain a resampled sample set; S45. Confidence intervals are obtained from the resampled sample set using the percentile method. Based on the confidence intervals and fitting parameters, the suppression efficiency score is calculated using the suppression efficiency calculation formula. The suppression efficiency score, confidence intervals, and fitting parameters are then integrated into a set of quantitative indicators. S5. Based on the quantitative index set, regression prediction and decision prediction are obtained by multiple linear regression combined with optimized decision tree. Based on the regression prediction and decision prediction, the integrated prediction value is obtained by weighted average formula, and the liver protection experiment score is calculated by linear transformation formula. S6. Based on the integrated predicted values ​​and liver protection experimental scores, a liver protection effect prediction model was constructed. The parameters were optimized by grid search method, and the model was calibrated by cross-validation method to obtain the calibrated effect model. S7. Input the damage dataset and the set of quantitative indicators into the calibration effect model to obtain experimental results, and generate an experimental report based on the experimental results.

2. The experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition according to claim 1, characterized in that, In S1, the specific implementation steps are as follows: S11. Collect raw data from historical data of liver protection experiments, and remove outliers from the raw data using the outlier detection formula to obtain cleaned data. S12. Obtain the fill value from the cleaned data using the mean interpolation formula, insert the fill value into the corresponding position in the cleaned data to obtain the fill data, and perform standardization processing on the fill data using the maximum and minimum standardization formula to obtain the standardized data. S13. Divide the standardized data into training and test sets by random partitioning.

3. The experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition according to claim 1, characterized in that, In S2, the specific implementation steps are as follows: S21. Based on the data in the training set, principal component analysis is used to perform feature decomposition to obtain the principal component feature set. S22. Construct a random forest model for obtaining importance data based on the principal component feature set. Input the principal component feature set into the random forest model to obtain the importance evaluation value. Sort the important feature sets in descending order according to the importance evaluation value to obtain the top k important feature sets. S23. Obtain the standard dataset based on the important feature set using the Z-score standardization formula.

4. The experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition according to claim 1, characterized in that, In S3, the specific implementation steps are as follows: S31. Obtain the initial parameters of the data dynamics model for simulating biological experiments using the maximum likelihood estimation method based on the standard dataset; S32. Based on the initial parameters, construct an activity dynamic model for simulating NF-κB activity using the RK4 integral method. Input the standard dataset into the activity dynamic model to obtain the activity time series, and construct the activity dynamic layer. S33. Obtain real data from the liver protection experiment. Based on the activity time series obtained from the activity dynamic model, construct a neural network model to correct the liver protection experiment data. Train the neural network model using a loss function and backpropagation algorithm based on the activity time series and real data to obtain the updated dynamic layer. S34. Input the active time series and real data into the update dynamic layer to obtain the update time series. Correct the update time series through forward propagation to obtain the corrected data, and obtain the neural network layer. S35. Integrate the active dynamic layer, the update dynamic layer, and the neural network layer into a dynamic simulation model. Determine the verification threshold based on the root mean square error formula. Obtain the evaluation value through K-fold cross-validation based on the correction data. Validate the dynamic simulation model based on the evaluation value and the verification threshold to obtain the optimized dynamic model. If the evaluation value is greater than or equal to the verification threshold, the dynamic simulation model has a large deviation, and the process returns to S31 to readjust the parameters of the dynamic simulation model. If the evaluation value is less than the verification threshold, the current dynamic simulation model will be used as the optimized dynamic model. S36. Input the standard dataset into the optimized dynamic model to obtain the activity prediction data.

5. The experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition according to claim 1, characterized in that, In S5, the specific implementation steps are as follows: S51. Generate an injury dataset based on liver injury biomarkers in the standard dataset. Obtain regression prediction and regression residuals using a multiple linear regression formula based on the injury dataset and the set of quantitative indicators. S52. Based on the damage dataset and the quantization index set, damage standard data and quantization standard data are obtained by feature standardization and corresponding vector standardization, respectively. Based on the damage standard data and quantization standard data, the interaction term is calculated by vector multiplication to obtain the input matrix. S53. Construct a decision tree model based on the input matrix and regression residuals. Obtain the initial hyperparameters through the maximum likelihood estimation method. Obtain the information gain and the optimal split point based on the input matrix and regression residuals through the information gain formula and the optimal split point calculation formula, respectively. S54. Based on information gain and the optimal split point, construct an initial decision tree using a greedy algorithm. Calculate the cost complexity of each leaf node in the initial decision tree. Sort the multiple cost complexities in descending order and prune the 20% of leaf nodes with the lowest cost complexity to obtain the optimized decision tree. S55. Input the input matrix and regression residuals into the optimization decision tree to obtain the decision prediction. Based on the regression prediction and the decision prediction, obtain the integrated prediction value through the weighted average formula. S56. Calculate the liver protection experimental score for evaluating the liver protection effect using a linear transformation formula based on the integrated predicted values.

6. The experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition according to claim 1, characterized in that, In S6, the specific implementation steps are as follows: S61. Based on S5, construct a liver protection effect prediction model and obtain the initial parameters of the model. Based on the initial parameters, obtain the total effect index through the Sobol formula. Based on the total effect index, sort the initial parameters in descending order and take the larger 20% as the sensitive parameters. S62. Adjustment parameters are obtained by using a grid search method based on the sensitive parameters; S63. Replace the initial parameters in the liver protection effect prediction model with the adjustment parameters to obtain the optimized effect model. Then, calibrate the optimized effect model using cross-validation to obtain the calibrated effect model.

7. The experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition according to claim 1, characterized in that, In S7, the specific implementation steps are as follows: S71. Input the damage dataset and the quantitative index set into the calibration effect model to obtain experimental results, and obtain the experimental data table based on the experimental results through data aggregation. S72. Based on the experimental data table, plot the inhibition curve and liver protection score using Matplotlib; S73. Generate an experimental report by combining the experimental data table, inhibition curve, and liver protection score chart.

8. A system for applying the experimental test method for evaluating TFF3 liver protection based on NF-κB pathway inhibition as described in any one of claims 1-7, characterized in that, The system includes: The preprocessing module (1) is used to obtain the raw data of the liver protection experiment, and to preprocess and segment the raw data to obtain the training set and the test set. The standardization module (2) is used to process the data based on the training set through feature decomposition and importance evaluation to obtain an important feature set, and to obtain a standard dataset by Z-score standardization based on the important feature set; The activity prediction module (3) is used to construct a dynamic simulation model for dynamic correction and accurate prediction of liver protection experimental data based on a standard dataset, and to optimize the dynamic simulation model to obtain an optimized dynamic model, and to obtain activity prediction data based on the optimized dynamic model. Quantization module (4), the quantization module (4) is used to obtain the TFF3 concentration gradient from the standard dataset, and obtain the quantization index set by sequentially passing the least squares method and Bootstrap resampling according to the TFF3 concentration gradient and activity prediction data; Experimental scoring module (5), the experimental scoring module (5) is used to obtain regression prediction and decision prediction by multiple linear regression combined with optimization decision tree according to the quantitative index set, obtain integrated prediction value by weighted average formula according to regression prediction and decision prediction, and calculate liver protection experimental score for evaluating liver protection effect by linear transformation formula; The calibration module (6) is used to construct a liver protection effect prediction model based on the integrated predicted value and liver protection experiment score to predict the liver protection experiment and form an intuitive experimental effect. The parameter is optimized by grid search method and the model is calibrated by cross-validation method to obtain the calibration effect model. The report module (7) is used to input the damage dataset and the quantitative index set into the calibration effect model to obtain experimental results, and generate an experimental report based on the experimental results.

Images