A test data management and analysis system

Abstract

The present invention relates to the technical field of database management, and specifically to a test data management and analysis system, wherein the system comprises: a data adaptive analysis module receives an RNA sequencing data set, analyzes the connectivity of node features and edges in the data set, adjusts the weights of nodes and edges, and updates the node feature processing mode and edge weight configuration. In the present invention, by updating the gene map in real time and adjusting the weight configuration of nodes and edges, the dynamic parsing ability of data is enhanced, so that data analysis is more in line with the complexity of biological changes, and the construction and comparison of multi-scale topological fingerprints allow in-depth identification of structural features in data, improves the comprehensiveness of analysis and the accuracy of prediction, optimizes the matching and parameter setting of time series data, ensures that the analysis results are close to the real data, improves the quality and efficiency of analysis, extracts the spatial and geometric topological features in biological data, and applies them to training neural network models, accelerates the learning speed, and improves the prediction accuracy and applicability of the model.

Description

Technical Field

[0001] The present invention relates to the technical field of database management, and in particular to a test data management and analysis system. Background Art

[0002] Database management is a core component of the information technology field that involves the use of software tools to organize and manage the storage, retrieval, maintenance, and security of data. This field focuses on the structured storage of data so that it can be efficiently queried and analyzed. Database management systems (DBMS) are key technologies in this field, providing a systematic approach to managing data, including the creation of data models, the insertion, update, deletion of data, and the backup and recovery of data. These systems support a variety of database models, the most common of which are relational and non-relational databases. Database management also includes aspects such as performance optimization, data integrity assurance, and access control to ensure efficient use and security of data.

[0003] Among them, experimental data management and analysis systems refer to software systems designed specifically for processing and analyzing data generated in experiments. Such systems enable researchers to effectively manage experimental data, including data collection, storage, access, and analysis. Experimental data management systems typically provide tools to help users store data in a structured manner and perform complex data analysis to derive support information for scientific or engineering decisions. These systems are particularly important in drug development, biotechnology, materials science, and other fields that require precise data analysis. Their main purpose is to improve the accessibility and reliability of data while ensuring data consistency and compliance.

[0004] Existing technologies rely on static data management methods and lack effective capture and updating of dynamic changes in data, which limits the timeliness and accuracy of analysis. Traditional technologies fail to fully utilize the inherent connections between data when processing complex data structures, resulting in limited depth and breadth of analysis. For example, in the analysis of biological experimental data, static methods are difficult to capture rapidly changing biological processes, which may cause the results to deviate from the actual situation, affecting the reliability of scientific research results and the effectiveness of application. The shortcomings of time series analysis, such as the lack of effective matching and optimization methods, often lead to data analysis results that cannot truly reflect the impact of time changes, thereby affecting the accuracy and timeliness of decision-making. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a test data management and analysis system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A test data management and analysis system comprises:

[0007] The data adaptive analysis module receives the RNA sequencing data set, analyzes the node features and edge connectivity in the data set, adjusts the weights of nodes and edges, updates the node feature processing method and edge weight configuration, updates the map in real time, and generates a dynamic gene map;

[0008] The topological structure comparison module uses the dynamic gene map to analyze the continuous homology group characteristics of the map, identify the ring structure and connected branches from the map, construct a multi-scale topological fingerprint corresponding to the continuous homology group data, perform structural analysis and comparison on the fingerprint, and generate a structure-aware comparison result;

[0009] The biological test time series analysis module analyzes the time series data in the structure-aware comparison result, performs elastic matching on the sequence to determine the degree of fit, adjusts the parameter settings of the sequence matching, optimizes the sequence matching process, and generates a sequence matching optimization result;

[0010] The feature extraction and learning module uses the sequence matching optimization results to extract spatial and geometric topological features in biological data, and applies them to neural networks for model training, optimizes the learning path and parameters of the model, and constructs a topological feature learning model.

[0011] As a further solution of the present invention, the steps of obtaining the dynamic gene map are specifically as follows:

[0012] Receive RNA sequencing data sets, build a preliminary gene interaction network based on gene expression levels and gene interaction information, and obtain a preliminary network structure representation;

[0013] Based on the preliminary network structure representation, analyzing the connection strength and pattern between nodes, applying the adjacency matrix to represent the connectivity between the polygenic nodes, and generating an adjacency matrix;

[0014] Using the adjacency matrix and the gene expression data of the nodes, the weight of each edge is updated using the formula:

[0015] ;

[0016] Generate updated edge weights;

[0017] in, and Represents the connection nodes and The gene expression values ​​are used to evaluate the interaction strength between nodes. Represents the connection adjustment factor, which is used to adjust the impact of gene expression value differences on network connections. Represents the normalized index of the expression value, which is used to adjust the proportion of the node's expression value in the total expression. represents the nonlinear adjustment index, Represents a single node, Indicates Gene expression value of each node;

[0018] According to the updated edge weights and node features, the network is reconstructed to ensure that the features of each node and the connectivity of the edges are currently reflected, and a dynamic gene map is established.

[0019] As a further solution of the present invention, the step of obtaining the structure-aware comparison result is specifically:

[0020] Using the dynamic gene map, analyzing the gene interaction pattern in the map through topological data, identifying the persistent homology group, capturing the ring structure and connected branches in the map, and generating a homology group feature set;

[0021] The homology group feature set is quantitatively analyzed to calculate the Betti number and Euler feature. The quantitative index describes the number and type of ring structures and connected branches in the graph to obtain quantitative topological data.

[0022] According to the quantized topological data, a multi-scale topological fingerprint corresponding to the persistent homology group data is constructed, and the quantized topological data is converted into a comparable topological fingerprint through the multi-scale persistence theory to generate a multi-scale topological fingerprint;

[0023] The multi-scale topological fingerprint is structurally analyzed and compared using the formula:

[0024] ;

[0025] Calculate differences between structures and generate structure-aware alignment results;

[0026] in, and The quantitative topological measurements are derived from the topological fingerprints at different scales, representing the characteristic quantities of homology groups identified in the gene network at different scales. is an adjustment factor used to adjust the sensitivity of the comparison process to scale differences. Indicates the difference between structures.

[0027] As a further solution of the present invention, the step of obtaining the sequence matching optimization result is specifically:

[0028] Analyzing the time series data in the structure-aware comparison result, determining key time points and change patterns, extracting feature sequences, and obtaining a time series feature set;

[0029] Based on the time series feature set, a flexible matching algorithm is used to calculate the preliminary fit between the sequences, and the time and amplitude errors are adjusted to generate preliminary matching results by adjusting the time window and the matching threshold;

[0030] Using the preliminary matching results, refining the matching parameter settings, including the elastic window size and the matching sensitivity, iteratively optimizing the alignment accuracy between sequences, and obtaining an adjusted matching parameter record;

[0031] Apply the adjusted matching parameter record and re-execute the sequence matching process, using the formula:

[0032] ;

[0033] Calculate the matching fit and generate the sequence matching optimization results;

[0034] in, Represents at a point in time The original time series data, is the time offset used to adjust the time alignment between sequences, Represents the time offset After comparing the time series data, is the amplitude adjustment factor, which is used to adjust the amplitude difference between sequences. Indicates the matching fit.

[0035] As a further solution of the present invention, the steps of acquiring the topological feature learning model are specifically as follows:

[0036] Analyze the biological data extracted from the sequence matching optimization results, apply topological data analysis technology to identify and quantify spatial and geometric topological features in the topological data, and obtain a topological feature data set;

[0037] Importing the topological feature data set into the neural network model, executing standard feedforward and back-propagation algorithms, performing initial training to adjust network weights and biases, and generating unoptimized model training results;

[0038] Extracting performance indicators from the unoptimized model training results, adjusting the learning rate and parameter decay strategy of the neural network according to the indicators, optimizing the learning path of the model, and obtaining optimized network parameters;

[0039] Using the optimized network parameters, the neural network training process is re-performed using the formula:

[0040] ;

[0041] Construct a topological feature learning model;

[0042] in, Represents the total number of samples, which indicates the total number of data points used to calculate the loss function during training. It is The target output value of the sample represents the actual value predicted by the neural network. It is The input feature data of samples is used for the network input layer. represents the parameters of the network, including the weights and biases of all layers, It is The standard deviation of the error of each sample is used to adjust the impact of the error of each sample. is the regularization strength, a penalty term used to control model complexity, is the regularization index to reduce errors and avoid overfitting, Represents the loss value of the topological feature learning model.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are:

[0044] In the present invention, by updating the gene map in real time and adjusting the weight configuration of nodes and edges, the dynamic parsing ability of data is enhanced, making data analysis more in line with the complexity of biological changes. This method improves the flexibility and accuracy of data processing and provides accurate support information for scientific research and decision-making. Constructing and comparing multi-scale topological fingerprints allows in-depth identification of structural features in data, improving the comprehensiveness of analysis and the accuracy of prediction. Optimizing the matching and parameter settings of time series data ensures that the analysis results are close to real data, improving the quality and efficiency of analysis. In addition, the spatial and geometric topological features in biological data are extracted and applied to training neural network models, which accelerates the learning speed and improves the prediction accuracy and applicability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a system flow chart of the present invention;

[0046] Figure 2 A flowchart of the steps for obtaining the dynamic gene map of the present invention;

[0047] Figure 3 A flowchart of the steps for obtaining the structure-aware comparison results of the present invention;

[0048] Figure 4 A flow chart of the steps for obtaining the sequence matching optimization results of the present invention;

[0049] Figure 5 The figure is a flow chart of the steps for obtaining the topological feature learning model of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0051] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0052] Example:

[0053] See also Figure 1 , a test data management and analysis system comprises:

[0054] The data adaptive analysis module receives the RNA sequencing data set, analyzes the node features and edge connectivity in the data set, adjusts the weights of nodes and edges, updates the node feature processing method and edge weight configuration, updates the map in real time, and generates a dynamic gene map;

[0055] The topological structure comparison module uses the dynamic gene map to analyze the characteristics of the continuous homology group of the map, identify the ring structure and connected branches from the map, construct a multi-scale topological fingerprint corresponding to the continuous homology group data, perform structural analysis and comparison on the fingerprint, and generate a structure-aware comparison result;

[0056] The biological experiment time series analysis module analyzes the time series data in the structure-aware comparison results, performs elastic matching on the sequences to determine the degree of fit, adjusts the parameter settings for sequence matching, optimizes the sequence matching process, and generates sequence matching optimization results;

[0057] The feature extraction and learning module uses sequence matching optimization results to extract spatial and geometric topological features from biological data, and applies them to neural networks for model training, optimizes the model's learning path and parameters, and constructs a topological feature learning model.

[0058] The dynamic gene map includes node features, edge weight configuration, and map update. The structure-aware comparison results include ring structure recognition, connected branch construction, and homology group feature analysis. The sequence matching optimization results are specifically matching elasticity, fit evaluation, and sequence parameter setting. The topological feature learning model specifically refers to spatial features, geometric topological features, and model training optimization.

[0059] See also Figure 2 , the steps for obtaining the dynamic gene map are as follows:

[0060] Receive RNA sequencing data sets, build a preliminary gene interaction network based on gene expression levels and gene interaction information, and obtain a preliminary network structure representation;

[0061] The received RNA sequencing dataset includes batch gene expression information. The data is used to construct a gene interaction network, in which gene expression levels and interaction information between genes are key input parameters. Based on the parameters, a preliminary network structure is constructed using a network analysis method. The structure is obtained by calculating the interaction strength between differentiated genes. The expression level of the gene directly affects the node representation in the network. In this way, the role of the gene expression pattern in biological function can be captured and mapped to the connection of the network. The obtained preliminary network structure representation provides a basic framework for subsequent analysis.

[0062] Based on the preliminary network structure representation, the connection strength and pattern between nodes were analyzed, and the adjacency matrix was applied to represent the connectivity between multi-gene nodes to generate an adjacency matrix;

[0063] Based on the preliminary network structure, the connection strength and pattern analysis between nodes were carried out, and the adjacency matrix was used to represent the connectivity between multi-gene nodes. The adjacency matrix was obtained by calculating the expression level differences and interaction strength between nodes. Each matrix element reflects the interaction strength between two genes. This analysis assists in parsing the relationship in the gene expression regulation network. The generation of the adjacency matrix provides a methodological basis for studying the direct and indirect relationships between genes, and the generated adjacency matrix provides key input data for subsequent weight updates and network reconstruction.

[0064] Using the adjacency matrix and the gene expression data of the nodes, the weight of each edge is updated using the formula:

[0065] ;

[0066] Generate updated edge weights;

[0067] in, and Represents the connection nodes and The gene expression values ​​are used to evaluate the interaction strength between nodes. Represents the connection adjustment factor, which is used to adjust the impact of gene expression value differences on network connections. Represents the normalized index of the expression value, which is used to adjust the proportion of the node's expression value in the total expression. represents the nonlinear adjustment index, Represents a single node, Indicates Gene expression value of each node;

[0068] formula:

[0069] ;

[0070] The formula is useful in that it refers to the gene expression differences between nodes and also uses the parameter and The introduction of expression standardization and nonlinear effects enhances the flexibility of the model and its ability to match differentiated data sets. Detailed explanation of the formula and the formula calculation derivation process are as follows:

[0071] Set in a data set, node and The expression values ​​of are 100 and 300 respectively, and the sum of the expression values ​​of all nodes connected to node i is 1000. .

[0072] First calculate:

[0073] ;

[0074] Then calculate the weight denominator:

[0075] ;

[0076] Then calculate the weight numerator:

[0077] ;

[0078] Get the weight:

[0079] ;

[0080] The results show that at a given expression level and selected parameters, the node and The connection strength between them is weak, indicating that their expression patterns are very different, thus they are in a marginal position in the network.

[0081] According to the updated edge weights and node features, the network is reconstructed to ensure that the characteristics of each node and the connectivity of the edges are currently reflected, and a dynamic gene map is established.

[0082] Based on the updated edge weights and node features, the process of reconstructing the network involves applying the calculated weight values ​​to each connection to ensure that the characteristics of each node and the connectivity of the edges are currently reflected, referring to the interactions between differentiated genes. This step is critical for parsing and predicting changes in gene function and pathological states. The reconstruction of the network not only depends on weight calculation, but also needs to refer to the update of node features, including changes in gene expression and the influence of external conditions. The reconstructed network provides a dynamic view, enabling researchers to monitor the dynamic process of gene expression over time, and establish a dynamic gene map to assist researchers in better parsing the dynamics and complexity of gene regulatory networks.

[0083] See also Figure 3 ,The specific steps for obtaining the structure-aware comparison results are:

[0084] Using dynamic gene maps, we analyze the gene interaction patterns in the maps through topological data, identify persistent homology groups, capture the ring structure and connected branches in the graph, and generate homology group feature sets;

[0085] In the process of analyzing gene interaction patterns using dynamic gene maps, it is first necessary to extract key data from gene expression levels and interaction information. The data includes the expression intensity and expression time of genes. Based on these data, topological data analysis technology is used to analyze the structure in the map. Topological data analysis technology identifies structures that persist in differentiated dimensions by constructing a multidimensional data space, including ring structures and connected branches. This step is critical because it is directly related to whether the subsequent continuous homology groups can be accurately identified. The identified continuous homology groups are manifested as some basic structures in the graph, including closed loops and branches. The identification of the structure depends on calculating the frequency of the structure appearing in the graph and its distribution pattern. In this way, the biological information hidden behind the gene interaction data is captured and a homology group feature set is generated.

[0086] Quantitative analysis is performed on the homology group feature set, and the Betti number and Euler feature are calculated. The quantitative indicators describe the number and type of ring structures and connected branches in the graph, and quantitative topological data is obtained.

[0087] In the process of quantitatively analyzing the feature set of homology groups, it is necessary to calculate and compare the relationships between differentiated homology group features. This process includes using algorithms to calculate the Betti numbers and Euler features of multiple homology groups. The Betti numbers provide information about the number of holes in the graph, while the Euler features are a measure of the topological complexity of the graph. By calculation, the number and types of ring structures and connected branches in the graph are described. This information is critical for analyzing how genes interact under differentiated biological processes and conditions. The generation process of quantitative topological data requires collecting sufficient topological information and relationship data, and performing mathematical processing on the data to ensure that the obtained data can reflect real biological phenomena and obtain quantitative topological data of ring structures and connected branches.

[0088] According to the quantitative topological data, a multi-scale topological fingerprint corresponding to the persistent homology group data is constructed. Through the multi-scale persistence theory, the quantitative topological data is converted into a comparable topological fingerprint to generate a multi-scale topological fingerprint.

[0089] When constructing multi-scale topological fingerprints, the multi-scale persistence theory is used based on the quantitative topological data. This theory refers to the changes in structure under differentiated monitoring scales and can assist in analyzing the dynamic performance of genes under differentiated environments and conditions. During the construction process, the quantitative topological data is converted into a comparable topological fingerprint through a series of algorithms. The topological fingerprint can be used for subsequent comparative analysis. In form, it is a collection of multi-dimensional data points, and each dimension represents a topological quantitative indicator. By comparing fingerprints, the fit and difference between gene expression patterns can be identified to generate multi-scale topological fingerprints.

[0090] Structural analysis and comparison of multi-scale topological fingerprints are performed using the formula:

[0091] ;

[0092] Calculate differences between structures and generate structure-aware alignment results;

[0093] in, and The quantitative topological measurements are derived from the topological fingerprints at different scales, representing the characteristic quantities of homology groups identified in the gene network at different scales. is a sensitivity parameter used to adjust the sensitivity of the comparison process to scale differences. Indicates the difference between structures.

[0094] formula:

[0095] ;

[0096] The formula is useful in that by adjusting the comparison sensitivity parameter , which can flexibly cope with data comparisons of differentiated scales, and enhance the adaptability and accuracy of the model in processing differentiated biological data.

[0097] Detailed explanation of the formula and the process of formula calculation and derivation:

[0098] There are two topological fingerprints. and , sensitivity parameter ,

[0099] The calculation process is:

[0100] ;

[0101] The results show that at a given sensitivity setting, the difference between the two topological fingerprints was quantified, and the value of 1.11 can be interpreted as the normalized alignment difference between the fingerprints, which helps to evaluate the degree of difference in the differential gene expression patterns in the subsequent steps.

[0102] See also Figure 4 , the specific steps for obtaining the sequence matching optimization results are:

[0103] Analyze the time series data in the structure-aware comparison results, determine the key time points and change patterns, extract feature sequences, and obtain the time series feature set;

[0104] When analyzing the time series data in the structure-aware comparison results, we first extract key time points and change patterns from the original data. This step involves time series analysis methods, including the use of moving averages and standard deviations to identify abnormal fluctuations and trend changes. Through technology, we can identify periodic fluctuations and anomalies in the data, thereby extracting key feature sequences that affect model analysis. The advantage of this method is that it can match differentiated data patterns and accurately capture the core features of the time series, thereby improving the accuracy and efficiency of data processing. The generated time series feature set will be used in the next step of the elastic matching algorithm.

[0105] Based on the time series feature set, the elastic matching algorithm is used to calculate the preliminary fit between the sequences, adjust the time and amplitude errors, and generate preliminary matching results by adjusting the time window and matching threshold;

[0106] When implementing the elastic matching algorithm to calculate the initial fit between sequences, the time and amplitude errors are adjusted through dynamic time warping technology. The technology can dynamically adjust the sequence data at differentiated time points to capture the optimal match, and reduce the deviation between sequences by setting the time window and matching threshold. This method can not only improve the accuracy of sequence matching, but also match differentiated types of data by adjusting parameters. The advantage of this elastic matching lies in its high adaptability and powerful processing capabilities for complex data patterns. The generated preliminary matching results provide basic data and reference basis for subsequent parameter adjustments.

[0107] Using the preliminary matching results, refine the matching parameter settings, including the elastic window size and matching sensitivity, iteratively optimize the alignment accuracy between sequences, and obtain the adjusted matching parameter records;

[0108] When adjusting the matching parameter settings, the effects of the elastic window size and matching sensitivity on the sequence alignment accuracy are captured. The impact of multi-parameter changes on the matching results is analyzed through experimental data, the optimal parameter settings are determined, and the alignment accuracy between sequences is optimized. This process involves multiple iterative experiments, and each iteration adjusts the parameters based on the matching results. By controlling each parameter, not only the matching accuracy is improved, but also the repeatability of the process and the adjustability of the parameters are ensured. The adjusted matching parameter records are obtained, providing parameter support for matching optimization.

[0109] Apply the adjusted matching parameter record and re-execute the sequence matching process, using the formula:

[0110] ;

[0111] Calculate the matching fit and generate the sequence matching optimization results;

[0112] in, Represents at a point in time The original time series data, is the time offset used to adjust the time alignment between sequences, Represents the time offset After comparing the time series data, is the amplitude adjustment factor, which is used to adjust the amplitude difference between sequences. Indicates the matching fit.

[0113] formula:

[0114] ;

[0115] The benefit of the formula is that it improves the matching accuracy by minimizing the square difference of the time series. This method is particularly suitable for time series analysis in a data environment, which enhances the applicability of the model and the reliability of the effect.

[0116] Detailed explanation of the formula and the process of formula calculation and derivation:

[0117] set up The sequence is [1,2,3,4,5], The sequence is [1,1,2,3,4], and the time offset , amplitude adjustment coefficient , amplitude adjustment factor , the calculation steps are:

[0118] ;

[0119] The results show that by adjusting the time offset and amplitude factors, the sequence x(t) is aligned with the sequence y(t) to the minimum difference. The calculation result 13 is the matching fit, which indicates the quality of the match between the sequences, and the results are optimized by adjusting the parameters.

[0120] See also Figure 5 ,The specific steps for obtaining the topological feature learning model are:

[0121] Analyze the biological data extracted from the sequence matching optimization results, apply topological data analysis technology to identify and quantify the spatial and geometric topological features in the topological data, and obtain a topological feature data set;

[0122] Use topological data analysis technology to identify spatial and geometric topological features in biological data. This process first requires calculating the spatial distance between each data point to generate a distance matrix. The matrix reflects the relative position relationship between data points. Based on the distance matrix, the continuous homology group analysis method is used to identify holes and spatial structures in the data. Holes and structures are intuitive representations of the spatial organization of biological tissues or molecular structures. Through this analysis, a high-dimensional spatial mapping of biological data can be obtained. The mapping is an indispensable input feature for subsequent neural network model training. The extracted feature data set will be directly applied to the neural network training process to capture the complex patterns and correlations in biological data.

[0123] Import the topological feature dataset into the neural network model, execute the standard feedforward and backpropagation algorithms, perform initial training to adjust the network weights and biases, and generate unoptimized model training results;

[0124] Importing the topological feature data of the neural network and training the model involves the construction of a multi-layer neural network, in which the number of neurons and the type of activation function in each layer need to be adjusted according to the characteristics of the input data. The training process includes feedforward propagation and backpropagation. In feedforward propagation, the data is passed from the input layer to the output layer after being processed by the hidden layer. The output of each layer is the input of the next layer. The activation function includes ReLU or Sigmoid function to add nonlinear factors and process data relationships. Backpropagation is to adjust the weight of each neuron through the gradient descent method according to the error of the output layer to minimize the prediction error. This process is iterated until the prediction performance of the model reaches the predetermined standard or the number of iterations reaches the upper limit. The optimized model can better parse the topological features of the input and predict unknown data.

[0125] Extract performance indicators from the training results of the unoptimized model, adjust the learning rate and parameter decay strategy of the neural network according to the indicators, optimize the learning path of the model, and obtain the optimized network parameters;

[0126] During the neural network training process, the model parameters, including the learning rate and weight decay parameters, are adjusted by analyzing the model's performance indicators, including the value and accuracy of the loss function. This adjustment is based on the model's performance on the validation set. The quality of the performance indicators directly reflects the model's fit to the training data and its ability to generalize to new data. The learning rate determines the step size of the parameter update, and the weight decay is used to control the model's complexity and avoid overfitting. These optimization operations ensure that the model can achieve higher prediction accuracy on the current validation data. The optimized model parameters will be used for model training.

[0127] Using the optimized network parameters, the neural network training process is repeated using the formula:

[0128] ;

[0129] Construct a topological feature learning model;

[0130] in, Represents the total number of samples, which indicates the total number of data points used to calculate the loss function during training. It is The target output value of the sample represents the actual value predicted by the neural network. It is The input feature data of samples is used for the network input layer. represents the parameters of the network, including the weights and biases of all layers, It is The standard deviation of the error of each sample is used to adjust the impact of the error of each sample. is the regularization strength, a penalty term used to control model complexity, is the regularization index to reduce errors and avoid overfitting, Represents the loss value of the topological feature learning model.

[0131] formula:

[0132] ;

[0133] The benefit of this formula is that it refers to the error of each sample and controls the model complexity through the regularization term to avoid overfitting and match the characteristics of differentiated data sets.

[0134] Detailed explanation of the formula and the process of formula calculation and derivation:

[0135] There is a simplified data set, including 3 samples, and the target output of each sample is 1, 0, 1 respectively, model prediction They are 0.8, 0.1, and 0.9 respectively, and the standard deviation of error The regularization strength is 0.1, 0.2, and 0.1 respectively. , the regularization index , model parameters .

[0136] The calculation process is as follows:

[0137] Compute the error term for each sample: ;

[0138] Calculate the square of the error: ;

[0139] Average sum of squared errors: ;

[0140] Compute the regularization term: ;

[0141] loss: ;

[0142] The results show that given the model parameter configuration and data samples, the model loss is 1.7534, reflecting the error of the model on the training data and the impact of complexity adjustment.

[0143] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A test data management and analysis system, characterized in that: The system comprises: The data adaptive analysis module receives the RNA sequencing data set, analyzes the node features and edge connectivity in the data set, adjusts the weights of nodes and edges, updates the node feature processing method and edge weight configuration, updates the map in real time, and generates a dynamic gene map; The topological structure comparison module uses the dynamic gene map to analyze the continuous homology group characteristics of the map, identify the ring structure and connected branches from the map, construct a multi-scale topological fingerprint corresponding to the continuous homology group data, perform structural analysis and comparison on the fingerprint, and generate a structure-aware comparison result; The biological test time series analysis module analyzes the time series data in the structure-aware comparison result, performs elastic matching on the sequence to determine the degree of fit, adjusts the parameter settings of the sequence matching, optimizes the sequence matching process, and generates a sequence matching optimization result; The feature extraction and learning module uses the sequence matching optimization results to extract spatial and geometric topological features in biological data, and applies them to neural networks for model training, optimizes the learning path and parameters of the model, and constructs a topological feature learning model; The steps for obtaining the dynamic gene map are specifically as follows: Receive RNA sequencing data sets, build a preliminary gene interaction network based on gene expression levels and gene interaction information, and obtain a preliminary network structure representation; Based on the preliminary network structure representation, analyzing the connection strength and pattern between nodes, applying the adjacency matrix to represent the connectivity between the multi-gene nodes, and generating an adjacency matrix; Using the adjacency matrix and the gene expression data of the nodes, the weight of each edge is updated using the formula: ； Generate updated edge weights; in, and Represent the gene expression values ​​of connecting nodes i and j, respectively, and are used to evaluate the interaction strength between nodes. Represents the connection adjustment factor, which is used to adjust the impact of gene expression value differences on network connections. A standardized index representing the expression value, which is used to adjust the proportion of the node's expression value in the total expression. represents the nonlinear adjustment index, Represents a single node, Indicates Gene expression value of each node; Reconstructing the network according to the updated edge weights and node features to ensure that the features of each node and the connectivity of the edges are currently reflected, and establishing a dynamic gene map; The steps for obtaining the structure-aware comparison result are specifically as follows: Using the dynamic gene map, analyzing the gene interaction pattern in the map through topological data, identifying persistent homology groups, capturing the ring structure and connected branches in the map, and generating a homology group feature set; The homology group feature set is quantitatively analyzed to calculate the Betti number and Euler feature. The quantitative index describes the number and type of ring structures and connected branches in the graph to obtain quantitative topological data. According to the quantized topological data, a multi-scale topological fingerprint corresponding to the persistent homology group data is constructed, and the quantized topological data is converted into a comparable topological fingerprint through the multi-scale persistence theory to generate a multi-scale topological fingerprint; The multi-scale topological fingerprint is structurally analyzed and compared using the formula: ； Calculate differences between structures and generate structure-aware alignment results; in, and The quantitative topological measurements are derived from the topological fingerprints at different scales, representing the characteristic quantities of homology groups identified in the gene network at different scales. is a sensitivity parameter used to adjust the sensitivity of the comparison process to scale differences. Indicates the difference value between structures; The steps for obtaining the sequence matching optimization result are specifically as follows: Analyzing the time series data in the structure-aware comparison result, determining key time points and change patterns, extracting feature sequences, and obtaining a time series feature set; Based on the time series feature set, a flexible matching algorithm is used to calculate the preliminary fit between the sequences, and the time and amplitude errors are adjusted to generate preliminary matching results by adjusting the time window and the matching threshold. Using the preliminary matching results, refining the matching parameter settings, including the elastic window size and the matching sensitivity, iteratively optimizing the alignment accuracy between sequences, and obtaining an adjusted matching parameter record; Apply the adjusted matching parameter record and re-execute the sequence matching process, using the formula: ； Calculate the matching fit and generate the sequence matching optimization results; in, Represents at a point in time The original time series data, is the time offset used to adjust the time alignment between sequences, Represents the time offset After comparing the time series data, is the amplitude adjustment factor, which is used to adjust the amplitude difference between sequences. Indicates the matching fit; The steps for acquiring the topological feature learning model are specifically as follows: Analyze the biological data extracted from the sequence matching optimization results, apply topological data analysis technology to identify and quantify spatial and geometric topological features in the topological data, and obtain a topological feature data set; Importing the topological feature data set into the neural network model, executing standard feedforward and back-propagation algorithms, performing initial training to adjust network weights and biases, and generating unoptimized model training results; Extracting performance indicators from the unoptimized model training results, adjusting the learning rate and parameter decay strategy of the neural network according to the indicators, optimizing the learning path of the model, and obtaining optimized network parameters; Using the optimized network parameters, the neural network training process is re-performed using the formula: ； Construct a topological feature learning model; in, Represents the total number of samples, which indicates the total number of data points used to calculate the loss function during training. It is The target output value of the sample represents the actual value predicted by the neural network. It is The input feature data of samples is used for the network input layer. represents the parameters of the network, including the weights and biases of all layers, It is The error standard deviation of each sample is used to adjust the impact of each sample error. is the regularization strength, a penalty term used to control model complexity, is the regularization index to reduce errors and avoid overfitting, Represents the loss value of the topological feature learning model.

Images