Biochemical information database extraction system and method

Abstract

The invention relates to the technical field of biochemical data processing, and discloses a biochemical information database extraction system and method, and the system comprises biochemical data acquisition, preprocessing, feature extraction, distributed storage management and intelligent query optimization modules, and can also comprise a privacy protection and version management module. The acquisition module integrates multi-source heterogeneous data; the preprocessing module detects abnormity and standardizes data; the feature extraction module is used for mining multi-modal features; the storage management module encrypts storage and balances loads; and the query optimization module improves the query efficiency. The privacy protection module guarantees data security, and the version management module realizes data version tracing. According to the method, the biochemical data processing problem is effectively solved, the data processing efficiency, safety and traceability are improved, and powerful support is provided for the fields of biochemical research and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of biochemical data processing, and in particular to a biochemical information database extraction system and method. Background Art

[0002] In the field of biochemistry, with the rapid development of experimental technology and information technology, biochemical data has shown explosive growth. These data come from a wide range of sources, including biochemical experimental equipment, public databases, and real-time sensors. Data from different data sources are significantly heterogeneous, with huge differences in data format, structure, and semantics, which poses a great challenge to the effective integration and utilization of data.

[0003] From the perspective of data acquisition, the data generated by biochemical experimental equipment is usually a binary file or text file in a specific format. The data structure and content specifications generated by equipment from different manufacturers and models are different. Although public databases such as GenBank and PDB provide researchers with a large amount of shared data, the data standards and access interfaces of these databases are also different, making the data acquisition and integration process complicated and cumbersome. Real-time sensors continue to generate a large amount of time-series data. How to efficiently collect and integrate these multi-source, heterogeneous and time-series data has become the primary problem in biochemical information processing. The data preprocessing link also faces many problems. Due to the complexity of the experimental environment, equipment errors, and interference during data transmission, outliers are inevitably present in the original biochemical data. If these abnormal data are not processed in time, they will seriously affect the accuracy and reliability of subsequent data analysis. Traditional anomaly detection methods, such as detection algorithms based on fixed thresholds, cannot adapt to the dynamic changes of biochemical data and are prone to missed detection or false detection. Moreover, data from different data sources differ in terms of dimension and value range. If they are not standardized, the performance of subsequent data analysis models will be reduced.

[0004] Feature extraction is a key step in biochemical data analysis. Biochemical data contains multiple modalities, such as text descriptions, spectral data, and molecular structure diagrams, each of which contains unique biochemical information. However, existing feature extraction methods can often only process single-modal data, making it difficult to fully explore the associations and complementary information between multimodal data. For complex biochemical data, traditional feature extraction algorithms are inefficient and cannot meet the requirements of data processing speed and accuracy in the big data era. In terms of data storage, with the continuous increase in the amount of biochemical data, traditional centralized storage methods face problems such as limited storage capacity, data read and write performance bottlenecks, and single point failures. Although distributed storage technology provides a solution, in a distributed storage environment, secure data storage and load balancing have become new challenges. How to efficiently encrypt and store biochemical data and ensure load balancing between distributed nodes to improve the overall performance and reliability of the storage system is an urgent problem to be solved.

[0005] Query optimization is an important part of biochemical information database applications. When conducting research, researchers need to frequently query specific biochemical information from the database. Existing query methods are often inefficient and difficult to meet the needs of high-frequency queries. Moreover, users' query needs are usually expressed in natural language. How to convert unstructured natural language queries into efficient structured query logic and optimize the query process is the key to improving database query efficiency. In addition, biochemical data involves a large amount of sensitive information, such as personal health data, patent technology data, etc., and data privacy protection is crucial. Traditional access control strategies and encryption technologies cannot meet the strict requirements of biochemical data for privacy protection. In multi-user concurrent access scenarios, permission management and privacy protection face greater challenges. At the same time, with the deepening of research and the updating of data, the management and backtracking of data versions have become increasingly important. Existing version management methods are insufficient in dealing with complex version dependencies and conflicts. Summary of the invention

[0006] The purpose of the present invention is to provide a biochemical information database extraction system and method to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: a biochemical information database extraction system, the system comprising: It includes biochemical data acquisition module, feature extraction module, distributed storage management module and intelligent query optimization module, among which: The biochemical data acquisition module includes a multi-source interface unit and a heterogeneous data integration unit. The multi-source interface unit is used to access biochemical experimental equipment, public databases and real-time sensor data sources. The heterogeneous data integration unit adopts a dynamic priority scheduling algorithm based on a graph structure to dynamically weight and fuse the timing and relevance of different data sources. The feature extraction module includes an adaptive convolutional neural network unit, which dynamically adjusts the convolution kernel parameters according to the dimension of the input data and uses a sparse attention mechanism to extract key biochemical features; The distributed storage management module includes a shard encryption unit and a dynamic load balancing unit. The shard encryption unit adopts a lightweight shard encryption algorithm based on chaotic mapping to encrypt the feature data in blocks and then store them in distributed nodes. The intelligent query optimization module includes a reinforcement learning indexing unit and a semantic parsing unit. The reinforcement learning indexing unit dynamically optimizes the multi-level index structure through a deep deterministic policy gradient algorithm.

[0008] Preferably, the system further comprises a data preprocessing module, wherein the data preprocessing module comprises an anomaly detection unit and a data standardization unit; The anomaly detection unit is based on a dynamic threshold cleaning algorithm, which uses a sliding window to count local data distribution and generate an adaptive cleaning threshold; further comprising: The local outlier factor calculation subunit is used to calculate the local outlier factor based on the KL divergence of the data distribution in the sliding window. The calculation formula is:

[0009] in, is the data distribution in the sliding window, is the historical benchmark distribution, is the discretized interval index in the data distribution; A dynamic threshold generation subunit, used to generate an adaptive cleaning threshold according to the outlier factor and the historical error distribution, and mark abnormal data segments; The data standardization unit is used to perform standardization processing on the data after anomaly detection.

[0010] Preferably, the feature extraction module also includes a multimodal fusion unit, which uses a joint embedding algorithm based on tensor decomposition to map text descriptions, spectral data and molecular structure diagrams to a unified feature space and eliminates modal redundancy through low-rank constraints.

[0011] Preferably, the dynamic load balancing unit of the distributed storage management module includes: Node status monitoring subunit, used to collect storage load and network delay of distributed nodes in real time; The shard migration decision subunit uses a Nash equilibrium strategy based on game theory to dynamically adjust the data shard distribution. Its utility function is:

[0012] in, For Node The utility value under the strategy combination, For Node The current shard storage strategy, A collection of shard storage strategies for other nodes. For Node The load factor, is the network delay coefficient, , is the weight factor.

[0013] Preferably, the semantic parsing unit of the intelligent query optimization module includes: A natural language processing subunit, used to convert unstructured query statements into structured query logic; The syntax tree optimization subunit generates a syntax tree that minimizes the query cost by pruning redundant nodes and merging similar paths.

[0014] Preferably, the system further includes a privacy protection module, and the privacy protection module includes: A differential noise injection unit is used to add Laplace noise to sensitive fields during the data preprocessing stage; The access control unit,adopts a dynamic permission allocation strategy based on attribute-based encryption,,and generates fine-grained access tokens according to user roles and,query contexts.

[0015] Preferably, the access control unit of the privacy protection module further comprises: The policy conflict detection subunit is used to identify and resolve permission policy conflicts when multiple users access the system concurrently. The token dynamic update subunit automatically refreshes the validity period of the access token according to the time decay function and access frequency. The time decay function is:

[0016] in, For access token at time The remaining value of the validity period, is the initial validity period, is the time decay factor, The time the token has been used.

[0017] Preferably, the system further includes a version management module, and the version management module includes: The data snapshot generation unit generates version snapshots using a differential compression algorithm based on incremental hashing; The version backtracking unit records version dependencies through a directed acyclic graph and supports multi-branch backtracking operations.

[0018] Preferably, the version backtracking unit of the version management module further includes: The conflict merge subunit uses a three-way merge algorithm based on operation transformation to resolve version conflicts; The metadata check subunit verifies the integrity and consistency of the version snapshot through the Merkle tree, and its hash calculation is:

[0019] in, is the hash value of the parent node, , They are the hash values ​​of the left and right child nodes respectively. is a cryptographic hash function, Represents a cascade operation on hash values.

[0020] Preferably, the present invention further comprises a biochemical information database extraction method, which is applied to the above-mentioned biochemical information database extraction system, and the method comprises: Heterogeneous biochemical data are collected through multi-source interface units, and data fusion is performed using a graph-structured dynamic priority scheduling algorithm; Perform anomaly detection and standardization on raw data based on dynamic threshold cleaning algorithm; Adaptive convolutional neural network is used to extract multimodal features, and feature fusion is completed through tensor decomposition algorithm; The chaotic mapping sharding encryption algorithm is used to store feature data in distributed nodes, and the sharding distribution is dynamically adjusted based on the Nash equilibrium strategy; High-frequency queries are optimized through reinforcement learning indexing, and semantic parsing is combined to generate a syntax tree that minimizes the cost.

[0021] Compared with the prior art, the present invention has the following beneficial effects: The biochemical data acquisition module of the present invention is connected to multiple data sources through a multi-source interface unit, and the heterogeneous data integration unit uses a dynamic priority scheduling algorithm based on a graph structure to fully consider the timing and relevance of the data for dynamic weighted fusion. This enables the system to quickly and accurately collect and integrate multi-source heterogeneous biochemical data, greatly improving the efficiency and fusion quality of data acquisition compared to traditional methods, and providing a more comprehensive and accurate data basis for subsequent data analysis. For example, when integrating data from different experimental equipment and public databases, the intrinsic connection between data can be better preserved to avoid information loss.

[0022] The anomaly detection unit in the data preprocessing module is based on a dynamic threshold cleaning algorithm. It generates an adaptive cleaning threshold by sliding a window to statistically analyze the local data distribution. It can accurately detect abnormal data, effectively overcome the defects of the traditional fixed threshold detection method, and reduce missed detection and false detection. The data standardization unit further standardizes the data to ensure the consistency and comparability of the data, providing high-quality data for subsequent feature extraction and analysis, and improving the accuracy and reliability of the analysis results. The adaptive convolutional neural network unit of the feature extraction module can dynamically adjust the convolution kernel parameters according to the input data dimension, and use the sparse attention mechanism to extract key biochemical features, which significantly improves the efficiency and accuracy of feature extraction. The multimodal fusion unit uses a joint embedding algorithm based on tensor decomposition to map multiple modal data to a unified feature space and eliminate modal redundancy, fully mining the complementary information of multimodal data, providing richer and more representative features for biochemical data analysis, and helping to improve the performance of subsequent data analysis models.

[0023] The shard encryption unit of the distributed storage management module adopts a lightweight shard encryption algorithm based on chaotic mapping to encrypt and store feature data in blocks, effectively ensuring data security. The dynamic load balancing unit monitors the node status in real time and uses the Nash equilibrium strategy based on game theory to dynamically adjust the data shard distribution, avoiding the problem of unbalanced node load, improving the overall performance and reliability of the storage system, and ensuring efficient storage and fast access to data. The reinforcement learning index unit of the intelligent query optimization module dynamically optimizes the multi-level index structure through a deep deterministic policy gradient algorithm, which can adapt well to high-frequency query scenarios and significantly improve query speed. The semantic parsing unit converts unstructured query statements into structured query logic, and generates a solution that minimizes the query cost by optimizing the syntax tree, further improving query efficiency and accuracy, making it convenient for researchers to quickly obtain the required biochemical information.

[0024] The differential noise injection unit of the privacy protection module adds Laplace noise to sensitive fields during the data preprocessing stage, effectively protecting the privacy of the data. The access control unit adopts a dynamic permission allocation strategy based on attribute-based encryption, combines user roles and query contexts to generate fine-grained access tokens, and can detect and resolve permission policy conflicts. At the same time, it automatically refreshes the token validity period according to the time decay function and access frequency, fully guaranteeing secure access to data and preventing sensitive information leakage. The data snapshot generation unit of the version management module uses a differential compression algorithm based on incremental hashing to generate version snapshots, which greatly saves storage space. The version backtracking unit records version dependencies through a directed acyclic graph, supports multi-branch backtracking operations, and uses a three-way merge algorithm based on operation transformation to resolve version conflicts. The integrity and consistency of version snapshots are verified through the Merkle tree, which facilitates researchers to trace and manage historical versions of data and ensure the traceability and integrity of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a working principle diagram of the biochemical information database extraction system of the present invention; Figure 2 is the workflow diagram of the anomaly detection unit; Figure 3 This is the workflow diagram of the dynamic load balancing unit; Figure 4 The flowchart for accessing the control unit. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] See also Figure 1-4 The present invention provides a technical solution: a biochemical information database extraction system, the system comprising: Biochemical data acquisition module: access biochemical experimental equipment, public databases and real-time sensor data sources through multi-source interface units to obtain various raw data in the biochemical field. The heterogeneous data integration unit adopts a dynamic priority scheduling algorithm based on graph structure, and performs dynamic weighted fusion according to the time sequence and relevance of data from different data sources, so that the fused data can more accurately reflect the comprehensive characteristics of biochemical information.

[0028] Data preprocessing module: The anomaly detection unit uses a sliding window to count local data distribution and generate an adaptive cleaning threshold based on a dynamic threshold cleaning algorithm to identify and process abnormal data. The data standardization unit standardizes the data after anomaly detection to ensure the consistency and comparability of the data in subsequent analysis and processing.

[0029] Feature extraction module: The adaptive convolutional neural network unit dynamically adjusts the convolution kernel parameters according to the dimension of the input data, and uses a sparse attention mechanism to extract key biochemical features to improve the accuracy and efficiency of feature extraction. The multimodal fusion unit uses a joint embedding algorithm based on tensor decomposition to map multimodal data such as text descriptions, spectral data, and molecular structure diagrams to a unified feature space, and eliminates modal redundancy through low-rank constraints to achieve effective fusion of multimodal data.

[0030] Distributed storage management module: The shard encryption unit uses a lightweight shard encryption algorithm based on chaotic mapping to encrypt feature data in blocks and store them in distributed nodes to ensure the security of data storage. The dynamic load balancing unit collects the storage load and network latency of distributed nodes in real time, and uses the Nash equilibrium strategy based on game theory to dynamically adjust the data shard distribution to ensure the stability and efficiency of the overall system performance.

[0031] Intelligent query optimization module: The reinforcement learning index unit dynamically optimizes the multi-level index structure through a deep deterministic policy gradient algorithm to adapt to high-frequency query scenarios and speed up query speed. The semantic parsing unit converts unstructured query statements into structured query logic, and generates a syntax tree that minimizes query costs by pruning redundant nodes and merging similar paths, further improving query efficiency and accuracy.

[0032] The present invention will be further described below in conjunction with Examples 1 to 5: Embodiment 1: This embodiment mainly describes the working process of the abnormality detection unit in the data preprocessing module in detail. In the process of processing biochemical data, abnormal data will seriously affect the accuracy of subsequent analysis results, so accurate detection and processing of abnormal data is very important.

[0033] The local outlier factor calculation subunit of the anomaly detection unit calculates the local outlier factor based on the KL divergence of the data distribution in the sliding window. Specifically, in the real-time process of data collection and processing, a sliding window of appropriate size is set. As data continues to flow in, the sliding window gradually moves on the data sequence. At each window position, the data distribution in the window is recorded as , the historical benchmark distribution is recorded as The discretized interval index in the data distribution is , through the formula Calculate the KL divergence to get the local outlier factor. This factor can quantify the difference between the data distribution in the current window and the historical benchmark distribution. The greater the difference, the more likely the data in the window is abnormal data.

[0034] The dynamic threshold generation subunit generates an adaptive cleaning threshold based on the calculated outlier factor and historical error distribution. In actual operation, the historical error distribution can be obtained by statistically analyzing a large amount of historical data. Combining the outlier factor with the historical error distribution can more accurately determine a dynamically changing threshold. When the outlier factor of the data in the window exceeds the threshold, the dynamic threshold generation subunit will mark the data segment as abnormal data, and the subsequent data preprocessing module will perform corresponding processing on these marked abnormal data, such as elimination and correction, so as to ensure the quality of data entering the subsequent processing link.

[0035] Embodiment 2: In the feature extraction module, the multimodal fusion unit uses a joint embedding algorithm based on tensor decomposition to process multimodal data. Take the common text descriptions, spectral data and molecular structure diagrams in biochemical experiments as examples. These data contain biochemical information from different angles. The multimodal fusion unit maps these data to a unified feature space. During the mapping process, modal redundancy is eliminated through low-rank constraints. For example, when processing spectral data and molecular structure diagrams, there may be repeated expressions of some information. Low-rank constraints can effectively remove these redundant parts, making the fused features more refined and representative, which is convenient for subsequent data analysis and model training.

[0036] In the dynamic load balancing unit of the distributed storage management module, the node status monitoring subunit collects the storage load and network delay of distributed nodes in real time. Through special monitoring programs and technical means, the storage usage and network transmission delay data of each node are regularly obtained. These data are fed back to the system in real time to provide a basis for subsequent decision-making.

[0037] The shard migration decision subunit uses a Nash equilibrium strategy based on game theory to dynamically adjust the data shard distribution. Its utility function is: ,in For Node The utility value under the strategy combination, For Node The current shard storage strategy, A collection of shard storage strategies for other nodes. For Node The load factor, is the network delay coefficient, , is the weight factor. In practical applications, the weight factor should be reasonably set according to the needs and characteristics of the system. and For example, when the system is sensitive to storage load, you can increase If you are more concerned about network latency, increase By continuously calculating the utility value of each node and dynamically adjusting the distribution of data shards between different nodes according to the Nash equilibrium strategy, the system can achieve a balance between storage load and network latency, thus improving the overall performance of the distributed storage system.

[0038] Embodiment 3: This embodiment describes in detail the specific workflow of the semantic parsing unit in the intelligent query optimization module. The semantic parsing unit plays a key role in improving the accuracy and efficiency of queries.

[0039] The natural language processing subunit is responsible for converting unstructured query statements into structured query logic. In actual application scenarios, users may enter query requirements in various natural languages, such as "find relevant experimental data of a protein with specific functions". The natural language processing subunit uses natural language processing technology to perform lexical analysis, syntactic analysis and semantic understanding on these query statements. Through lexical analysis, sentences are split into individual words or phrases and marked with parts of speech; syntactic analysis constructs the grammatical structure of sentences; semantic understanding further clarifies the semantic relationship between individual words and sentence fragments. After these processing steps, unstructured natural language queries are converted into structured query logic that the system can understand and process, such as converting them into a specific query statement format to clarify the target data type, conditions and other information of the query.

[0040] The syntax tree optimization subunit generates a syntax tree that minimizes the query cost by pruning redundant nodes and merging similar paths. After the initial syntax tree is generated, the syntax tree is analyzed. When querying "activity data of a specific protein under different temperature conditions", there may be some redundant nodes in the syntax tree, which may be caused by repeated statements or unnecessary logical branches in the query statement. The syntax tree optimization subunit identifies and prunes these redundant nodes, and at the same time, merges similar query paths, such as paths that point to the same target data under different combinations of conditions. Through such optimization operations, the resulting syntax tree can execute query tasks at the lowest cost, reduce the consumption of computing resources during the query process, and improve query efficiency.

[0041] Embodiment 4: This embodiment focuses on the working principle and specific implementation of the privacy protection module and the differential noise injection unit and access control unit it contains. This module is of great significance for ensuring the security and privacy of biochemical data.

[0042] In the data preprocessing stage, the differential noise injection unit adds Laplace noise to sensitive fields. Biochemical data may contain some sensitive information, such as experimental sample information involving personal privacy, specific experimental technical details, etc. In order to protect this sensitive information, the differential noise injection unit adds Laplace noise to sensitive fields when the data preprocessing module processes the data. The characteristics of Laplace noise enable it to hide the true value of the original data to a certain extent while maintaining the overall statistical characteristics of the data. For example, for a sensitive value representing the concentration of an experimental sample, after adding Laplace noise to it, it is difficult for an attacker to directly obtain the accurate original concentration value from the processed data, thereby protecting the privacy of the data.

[0043] The access control unit adopts a dynamic permission allocation strategy based on attribute-based encryption to generate fine-grained access tokens according to user roles and query contexts. In actual biochemical information database systems, different users have different roles, such as researchers, managers, students, etc., and their data access requirements and permissions are different. The access control unit first determines the initial access permission range of the user based on the user's role. For example, researchers may have higher data access permissions and be able to view and download detailed experimental data in a specific field; while students may only be able to access partially public data or processed teaching data. At the same time, combined with query contexts such as query time, query frequency, and specific data content of the query, permissions are further adjusted dynamically.

[0044] The access control unit also includes a policy conflict detection subunit and a token dynamic update subunit. The policy conflict detection subunit is used to identify and resolve permission policy conflicts during concurrent access by multiple users. When multiple users access the database at the same time, the permission policies of different users may conflict with each other. For example, a user attempts to modify a certain data while another user has read-only permission to the data at the same time, and a conflict occurs. The policy conflict detection subunit monitors these concurrent access requests in real time, and through analysis and comparison of permission policies, it promptly discovers and resolves conflicts and takes appropriate resolution measures, such as coordinating based on factors such as user priority and the urgency of the access request.

[0045] The token dynamic update subunit automatically refreshes the validity period of the access token according to the time decay function and access frequency. The time decay function is ,in For access token at time The remaining value of the validity period, is the initial validity period, is the time decay factor, The time the token has been used. As time goes by, the user's access needs and permissions may change. The token dynamic update subunit dynamically adjusts the validity period of the access token based on this time decay function and the user's access frequency. If a user has not accessed the data for a long time or has a low access frequency, the token dynamic update subunit will appropriately shorten the validity period of the access token; conversely, if the user accesses frequently and it is within the scope of his or her permissions, the validity period will be appropriately extended to ensure the convenience of user access and the security of data.

[0046] Embodiment 5: This embodiment describes in detail the working methods of the version management module and the data snapshot generation unit and version backtracking unit included therein. The version management module is crucial for data traceability and historical record management.

[0047] The data snapshot generation unit uses a differential compression algorithm based on incremental hashing to generate version snapshots. In the process of continuous updating and changing of biochemical data, in order to record the different version states of the data, the data snapshot generation unit generates version snapshots regularly or when specific events occur (such as important experimental data updates, system configuration changes, etc.). The differential compression algorithm based on incremental hashing can efficiently capture the changing parts of the data. Specifically, it identifies the state of the data by calculating the hash value of the data. When the data changes, only the hash calculation and storage of the changed part is performed instead of re-storing the entire data. This can greatly reduce the storage space occupied and speed up the generation of version snapshots. For example, in a database containing a large amount of biochemical experimental data, after each experimental result is updated, only the data part that is different from the previous version is stored through this algorithm, which effectively reduces the storage cost.

[0048] The version backtracking unit records version dependencies through a directed acyclic graph and supports multi-branch backtracking operations. In practical applications, the version evolution of biochemical data may have multiple branches, such as data branches generated under different experimental conditions, data branches in different research directions, etc. The directed acyclic graph can clearly record the dependencies between versions, with each version as a node in the graph and the update relationship between versions as a directed edge. In this way, the system can accurately trace back to any historical version.

[0049] The version backtracking unit also includes a conflict merge subunit and a metadata verification subunit. The conflict merge subunit uses a three-way merge algorithm based on operation transformation to resolve version conflicts. Version conflicts may occur during multi-user collaboration or the merging of data from different branches. For example, two users make different modifications to the same data at the same time. The conflict merge subunit uses a three-way merge algorithm based on operation transformation to compare and merge the current version, the target version, and the common ancestor version. By analyzing the operational differences between different versions and performing merge operations according to certain rules, the effective modifications of all parties are retained as much as possible to resolve version conflicts.

[0050] The metadata check subunit verifies the integrity and consistency of the version snapshot through the Merkle tree, and its hash calculation is ,in is the hash value of the parent node, , They are the hash values ​​of the left and right child nodes respectively. is a cryptographic hash function, Represents the cascade operation of hash values. The Merkle tree hashes the data blocks layer by layer to form a tree structure. In the process of version backtracking and data verification, by calculating the hash value of each node and comparing it with the stored hash value, it is possible to quickly verify whether the data of the version snapshot is complete and whether it has been tampered with. If the hash value is inconsistent, it means that there may be a problem with the data, which needs further investigation and repair to ensure the reliability and integrity of the data.

[0051] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0052] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A biochemical information database extraction system, characterized in that: It includes biochemical data acquisition module, feature extraction module, distributed storage management module and intelligent query optimization module, among which: The biochemical data acquisition module includes a multi-source interface unit and a heterogeneous data integration unit. The multi-source interface unit is used to access biochemical experimental equipment, public databases and real-time sensor data sources. The heterogeneous data integration unit adopts a dynamic priority scheduling algorithm based on a graph structure to dynamically weight and fuse the timing and relevance of different data sources. The feature extraction module includes an adaptive convolutional neural network unit, which dynamically adjusts the convolution kernel parameters according to the dimension of the input data and uses a sparse attention mechanism to extract key biochemical features; The distributed storage management module includes a shard encryption unit and a dynamic load balancing unit. The shard encryption unit adopts a lightweight shard encryption algorithm based on chaotic mapping to encrypt the feature data in blocks and then store them in distributed nodes. The intelligent query optimization module includes a reinforcement learning indexing unit and a semantic parsing unit. The reinforcement learning indexing unit dynamically optimizes the multi-level index structure through a deep deterministic policy gradient algorithm.

2. A biochemical information database extraction system according to claim 1, characterized in that: The system also includes a data preprocessing module, which includes an anomaly detection unit and a data standardization unit; The anomaly detection unit is based on a dynamic threshold cleaning algorithm, which uses a sliding window to count local data distribution and generate an adaptive cleaning threshold; further comprising: The local outlier factor calculation subunit is used to calculate the local outlier factor based on the KL divergence of the data distribution in the sliding window. The calculation formula is: ； in, is the data distribution in the sliding window, is the historical benchmark distribution, is the discretized interval index in the data distribution; A dynamic threshold generation subunit, used to generate an adaptive cleaning threshold according to the outlier factor and the historical error distribution, and mark abnormal data segments; The data standardization unit is used to perform standardization processing on the data after anomaly detection.

3. The biochemical information database extraction system according to any one of claims 1 to 2, characterized in that: The feature extraction module also includes a multimodal fusion unit, which uses a joint embedding algorithm based on tensor decomposition to map text descriptions, spectral data and molecular structure diagrams to a unified feature space and eliminates modal redundancy through low-rank constraints.

4. The biochemical information database extraction system according to claim 1, characterized in that: The dynamic load balancing unit of the distributed storage management module includes: Node status monitoring subunit, used to collect storage load and network delay of distributed nodes in real time; The shard migration decision subunit uses a Nash equilibrium strategy based on game theory to dynamically adjust the data shard distribution. Its utility function is: ； in, For Node The utility value under the strategy combination, For Node The current shard storage strategy, A collection of shard storage strategies for other nodes. For Node The load factor, is the network delay coefficient, , is the weight factor.

5. The biochemical information database extraction system according to claim 1, characterized in that: The semantic parsing unit of the intelligent query optimization module includes: A natural language processing subunit, used to convert unstructured query statements into structured query logic; The syntax tree optimization subunit generates a syntax tree that minimizes the query cost by pruning redundant nodes and merging similar paths.

6. The biochemical information database extraction system according to claim 1, characterized in that: It also includes a privacy protection module, which includes: A differential noise injection unit is used to add Laplace noise to sensitive fields during the data preprocessing stage; The access control unit,adopts a dynamic permission allocation strategy based on attribute-based encryption,,and generates fine-grained access tokens according to user roles and,query contexts.

7. The biochemical information database extraction system according to claim 6, characterized in that: The access control unit of the privacy protection module further includes: The policy conflict detection subunit is used to identify and resolve permission policy conflicts when multiple users access the system concurrently. The token dynamic update subunit automatically refreshes the validity period of the access token according to the time decay function and access frequency. The time decay function is: ； in, For access token at time The remaining value of the validity period, is the initial validity period, is the time decay factor, The time the token has been used.

8. The biochemical information database extraction system according to claim 1, characterized in that: It also includes a version management module, which includes: The data snapshot generation unit generates version snapshots using a differential compression algorithm based on incremental hashing; The version backtracking unit records version dependencies through a directed acyclic graph and supports multi-branch backtracking operations.

9. The biochemical information database extraction system according to claim 8, characterized in that: The version backtracking unit of the version management module further includes: The conflict merge subunit uses a three-way merge algorithm based on operation transformation to resolve version conflicts; The metadata check subunit verifies the integrity and consistency of the version snapshot through the Merkle tree, and its hash calculation is: ； in, is the hash value of the parent node, , are the hash values ​​of the left and right child nodes respectively, is a cryptographic hash function, Represents a cascade operation on hash values.

10. A method for extracting a biochemical information database, characterized in that: A system as claimed in any one of claims 1 to 9, comprising: Heterogeneous biochemical data are collected through multi-source interface units, and data fusion is performed using a graph-structured dynamic priority scheduling algorithm; Perform anomaly detection and standardization on raw data based on dynamic threshold cleaning algorithm; Adaptive convolutional neural network is used to extract multimodal features, and feature fusion is completed through tensor decomposition algorithm; The chaotic mapping sharding encryption algorithm is used to store feature data in distributed nodes, and the sharding distribution is dynamically adjusted based on the Nash equilibrium strategy; High-frequency queries are optimized through reinforcement learning indexing, and semantic parsing is combined to generate a syntax tree that minimizes the cost.

Images