A laboratory data analysis method and system

Abstract

This invention provides a laboratory data analysis method and system. The method includes the following steps: acquiring multiple experimental datasets and their associated metadata, which contains dependencies and logical intentions between experiments; constructing an experimental logical dependency graph based on the metadata, with datasets as nodes and dependencies as edges; calculating dynamic cognitive weights for each dataset node by combining the graph's topology with the mutual verification relationships between experiments to quantify its importance; using the topology of the logical dependency graph as a structural prior for a causal analysis model to provide strong constraints, and using the dynamic cognitive weights to perform weighted training on the datasets, ultimately outputting the causal relationships between the data. This invention, by modeling the logic and intention of the experimental process as constraints, enables data analysis to reveal the true causal relationships between data.

Description

Technical Field

[0001] This invention belongs to the field of data analysis technology, specifically relating to a laboratory data analysis method and system. Background Technology

[0002] In modern scientific research, laboratories generate massive amounts of multimodal experimental data every day. To extract valuable knowledge and discover new scientific laws from this complex data, researchers increasingly rely on advanced statistical methods and artificial intelligence algorithms for data analysis. These data-driven methods can efficiently process large-scale datasets, identify hidden patterns and correlations between variables, and greatly accelerate the research process. Automated data analysis platforms can integrate data from different experiments, perform unified modeling and analysis, and attempt to find potential scientific connections.

[0003] However, current mainstream data analysis methods suffer from a fundamental cognitive blind spot when processing a series of experimental data. Existing technologies typically treat all aggregated datasets as a flat, undifferentiated collection, focusing solely on the numerical distribution and statistical characteristics of the data itself. But a complete research project often consists of a series of interconnected, progressively building experiments, with subsequent experiments often designed based on the conclusions or questions raised by previous experiments. Existing analysis tools confuse data from different stages, with different purposes, and even contradictory data, resulting in numerous spurious associations and an inability to effectively distinguish genuine causal relationships. Therefore, understanding and utilizing the inherent logic of the experimental process to accurately extract scientifically significant causal relationships from massive amounts of data is a key technical challenge currently facing the field of laboratory data science. Summary of the Invention

[0004] This invention provides a laboratory data analysis method and system to solve the above-mentioned technical problems.

[0005] In a first aspect, the present invention provides a laboratory data analysis method, the method comprising the following steps: Obtain multiple experimental datasets, as well as experimental metadata associated with each experimental dataset, which includes at least the dependencies on previous experiments and the logical intent of this experiment; Construct an experimental logical dependency graph based on the dependencies and logical intent recorded in the experimental metadata, with the experimental dataset as the dataset nodes and the dependencies as directed edges; Based on the topological structure of the experimental logic dependency graph and the verification relationship formed by the results of subsequent experiments on the conclusions of previous experiments in the experimental logic dependency graph, dynamic cognitive weights are calculated for each dataset node in the experimental logic dependency graph. The topological structure of the experimental logic dependency graph is used as a structural prior to constrain the causal analysis model, and dynamic cognitive weights are used to weight the experimental dataset to train the causal analysis model to output the causal relationships between the data.

[0006] Optionally, obtaining multiple experimental datasets, and experimental metadata associated with each experimental dataset that includes at least the dependencies on previous experiments and the logical intent of the current experiment, includes the following steps: Multiple experimental datasets can be obtained through a pre-defined data interface; The logical intent of the experimental dataset corresponding to this experiment is determined based on the logical intent options pre-selected by the user in the experimental dataset. The dependency relationship between the experimental dataset and the preceding experiments is determined based on one or more predefined preceding experimental data interfaces. Bind the logical intent and the identified dependencies to the experimental dataset as experimental metadata; The bound experimental datasets and experimental metadata are stored uniformly in a pre-defined structured database.

[0007] Optionally, constructing an experimental logical dependency graph based on the dependencies and logical intent recorded in the experimental metadata, with the experimental dataset as the dataset nodes and the dependencies as directed edges, includes the following steps: Initialize the graph data structure, traverse all acquired experimental datasets, and create a corresponding dataset node in the graph data structure for each experimental dataset; Based on the dependencies recorded in the experimental metadata, directed edges are created between dataset nodes pointing to subsequent experimental dataset nodes; The logical intent recorded in the experimental metadata is attached as an attribute to the corresponding directed edge, forming an experimental logical dependency graph that includes all dataset nodes and directed edges.

[0008] Optionally, attaching the logical intent recorded in the experimental metadata as an attribute to the corresponding directed edge includes the following steps: Different numerical influence factors are pre-defined for different logical intentions; When creating directed edges, the numerical influence factor is attached as an initial weight attribute to the corresponding directed edge; The efficiency of weight transfer between dataset nodes is adjusted during the calculation of static authority by utilizing the initial weight attribute.

[0009] Optionally, the topological structure of the comprehensive experimental logic dependency graph and the verification relationships formed by subsequent experimental results on the conclusions of previous experiments in the experimental logic dependency graph, and the calculation of dynamic cognitive weights for each dataset node in the experimental logic dependency graph, include the following steps: The topology based on the experimental logical dependency graph is used to calculate the initial static authority for each dataset node in the graph. The consistency of the results of subsequent experimental dataset nodes with the conclusions of previous experimental dataset nodes is determined by calculating the dynamic confidence of the directed edges connecting the dataset nodes. The dynamic cognitive weight of each dataset node is calculated by weighting and fusing static authority and dynamic confidence.

[0010] Optionally, the topology based on the experimental logical dependency graph is used to calculate the initial static authority of each dataset node in the graph using the following steps: Assign an initial authority score to each dataset node in the experimental logic dependency graph; For each dataset node, an iterative calculation method is used, and the current authority score of the dataset node is updated based on the authority scores of all subsequent dataset nodes. During the iterative computation process, the logical intent attached to the directed edges is used as a regulating factor for the iterative computation. Repeat the iterative calculation until the authority scores of all dataset nodes converge; The converged authority score is used as the static authority score of the dataset node.

[0011] Optionally, the consistency of the results based on the experimental results of subsequent experimental dataset nodes with the conclusions of previous experimental dataset nodes, which constitutes the result of the directed edges connecting the dataset nodes, includes the following steps for calculating the dynamic confidence: For any preceding experimental dataset node and subsequent experimental dataset node that have a dependency relationship, train the benchmark prediction model based on the preceding experimental dataset node. The benchmark prediction model was tested using the subsequent experimental dataset of the subsequent experimental dataset node to obtain the prediction performance index. The predicted performance indicators are compared with the preset benchmark to quantify the consistency score that represents the degree of support or refutation. The dynamic confidence of directed edges connecting nodes in the preceding and subsequent experimental datasets is enhanced or weakened based on the consistency score of the results. The effect of adjusting the dynamic confidence level propagates along the path in the experimental logic dependency graph.

[0012] Optionally, the step of using the topological structure of the experimental logic dependency graph as a structural prior to constrain the causal analysis model, and using dynamic cognitive weights to weight the experimental dataset to train the causal analysis model to output the causal relationships between the data includes the following steps: Initialize a Bayesian network as a causal analysis model; All dataset nodes and directed edges in the experimental logical dependency graph are forcibly designated as the initial network topology of the Bayesian network to form a structural prior. When training the causal analysis model, the dynamic cognitive weight of each experimental dataset is applied to weight the contribution of the experimental dataset to the model parameter learning process. Input the experimental dataset into a constrained causal analysis model that requires weighted learning; The parameter learning algorithm is used to complete the training process of the causal analysis model.

[0013] Optionally, the method further includes the following steps: Receive one or more candidate experimental protocols to guide future research; Each candidate experimental scheme is virtualized as a hypothetical dataset node in the experimental logic dependency graph, and the hypothetical dataset node is connected to the relevant existing dataset node. The trained causal analysis model is used to predict the expected experimental results of nodes in the hypothetical dataset. Based on the expected experimental results, the simulation calculation calculates the fluctuation impact of introducing hypothetical dataset nodes on the dynamic cognitive weights of other dataset nodes in the experimental logic dependency graph. A recommendation index is generated for each candidate experimental scheme based on the expected experimental results and the impact of fluctuations.

[0014] In a second aspect, the present invention also provides a laboratory data analysis system, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the laboratory data analysis method as described in the first aspect.

[0015] This invention relates to machine learning and deep learning technologies, and its beneficial effects are: This invention deeply integrates the logical process of scientific research and the researcher's intent into the data analysis model by introducing experimental metadata. By constructing an experimental logical dependency graph, this invention transforms abstract experimental ideas into a computable topological structure and innovatively proposes dynamic cognitive weights. These weights quantify the credibility and importance of different datasets based on the verification relationship between subsequent experiments and prior conclusions. This allows the analysis model to think like a scientist, distinguishing between crucial evidence and preliminary exploration. Its core advantage lies in using this logical graph as a structural prior to constrain the causal analysis model, fundamentally eliminating the possibility of numerous causal inversions or spurious associations. Simultaneously, the dynamic cognitive weights guide the model to focus on high-value data. This mechanism transforms data analysis from a blind-men's-elephant correlation mining process into guided, logical causal reasoning, ultimately producing conclusions that are not only more reliable but also more scientifically interpretable. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a laboratory data analysis method in one embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the process of constructing an experimental logic dependency graph in one embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0019] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0020] Figure 1 This is a flowchart illustrating a laboratory data analysis method in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 1 As shown, the laboratory data analysis method disclosed in this invention specifically includes the following steps: S101. Obtain multiple experimental datasets, as well as experimental metadata associated with each experimental dataset, which includes at least the dependencies on previous experiments and the logical intent of this experiment.

[0021] The goal of this step is to capture the complete contextual information of each independent experiment, not just the raw measurement data. In practice, multiple experimental datasets from different instruments or recording systems are automatically acquired through a standardized data interface. Simultaneously, the user performing the experiment is required to select the logical intent of the experiment from a predefined list via an interactive interface. For example, the experiment may be intended to verify prior conclusions, explore new directions, or attempt to reproduce a known result. Furthermore, the user needs to specify one or more prior experimental datasets that the experiment depends on. This information, including the defined logical intent and dependencies, is integrated into experimental metadata and forcibly bound to the corresponding experimental datasets. Finally, these metadata-bound datasets are uniformly stored in a structured database, such as a relational database or graph database, ensuring data regularity, traceability, and contextual integrity.

[0022] S102. Based on the dependencies and logical intent recorded in the experimental metadata, construct an experimental logical dependency graph with the experimental dataset as the dataset node and the dependencies as directed edges.

[0023] The next step involves constructing an experimental logic dependency graph based on the collected experimental data rich in contextual information. This graph visually reflects the entire experimental process and logical structure. This step transforms abstract experimental dependencies and logical intentions into a concrete, computable network structure. Specifically, a graph data structure is initialized first. All experimental datasets in the database are traversed, creating a unique dataset node for each dataset in the graph. After node creation, the metadata of each dataset is examined again. Based on the recorded dependencies, directed edges are established between the corresponding nodes, with the edges pointing from the preceding node to the subsequent node, clearly indicating the sequence and inheritance of experiments. The experimental logical intentions recorded in the metadata are attached as attributes to the corresponding directed edges. For example, different logical intentions can be assigned different numerical influence factors. This step ultimately generates a complete experimental logic dependency graph, presenting the entire research project in a highly structured form. It not only reveals the direct relationships between various experiments but also quantifies the nature of these relationships through the intention attributes on the edges, providing a structural foundation and initial parameters for subsequent weight calculations.

[0024] S103. Based on the topological structure of the experimental logic dependency graph and the verification relationship formed by the results of subsequent experiments on the conclusions of previous experiments in the experimental logic dependency graph, calculate the dynamic cognitive weight for each dataset node in the experimental logic dependency graph.

[0025] After the experimental logic dependency graph is constructed, the quantitative evaluation phase begins, which involves calculating the dynamic cognitive weight for each experimental dataset node in the graph. This weight is significant in comprehensively evaluating the credibility and influence of each experiment within the overall research framework. The weight is dynamically adjusted based on the validation results of subsequent experiments. The calculation process integrates two dimensions: static authority based on the graph topology and dynamic confidence based on the consistency of experimental results. The calculation of static authority is similar to a webpage ranking algorithm, iteratively evaluating the structural importance of nodes. Its update rule can be expressed as... ,in It is the authority score of node i in the t-th iteration, which is determined by the authority scores of all nodes j pointing to it. and the influence factor of its corresponding edge The weighted summation is performed until the authority scores of all nodes converge. Dynamic confidence is calculated by quantifying the degree to which subsequent experiments support or refute the conclusions of previous experiments.

[0026] For example, for an edge from a preceding node p to a subsequent node s, a baseline prediction model can be trained using the dataset of p, and then tested using the dataset of s. The model's prediction performance is compared with the preset baseline to obtain a consistency score, which will either strengthen or weaken the confidence of this edge. Finally, the static authority and dynamic confidence are weighted and fused to obtain the dynamic cognitive weight of each node. ,in Let be the final weight of node i. It is its static authority after convergence. It is a comprehensive reflection of the dynamic confidence of each edge related to it, and It is a hyperparameter used to balance structural importance and empirical consistency.

[0027] S104. The topological structure of the experimental logical dependency graph is used as a structural prior to constrain the causal analysis model, and dynamic cognitive weights are used to weight the experimental dataset to train the causal analysis model to output the causal relationships between the data.

[0028] This step utilizes the experimental logic dependency graph and dynamic cognitive weights generated in the preceding steps to train a more reliable and logically sound causal analysis model. This process injects the inherent logic of experimental science into a purely data-driven machine learning model to discover more genuine causal relationships between variables, rather than spurious statistical correlations. Specifically, a suitable framework for expressing causal relationships, such as a Bayesian network, is chosen as the foundational causal analysis model. During model initialization, the topology of the experimental logic dependency graph—i.e., all nodes and directed edges—is directly used as the initial network topology of the Bayesian network. This constitutes a strong structural prior, forcing the model to learn within the logical framework defined by the experimental design, preventing the algorithm from learning causal directions that violate the experimental procedure without constraints. In the subsequent model parameter learning phase, dynamic cognitive weights play a crucial role. During model training, the contribution from each experimental dataset is weighted according to its corresponding dynamic cognitive weight.

[0029] In one implementation, acquiring multiple experimental datasets, and experimental metadata associated with each experimental dataset that includes at least dependencies on previous experiments and the logical intent of the current experiment, includes the following steps: Multiple experimental datasets can be obtained through a pre-defined data interface; The logical intent of the experimental dataset corresponding to this experiment is determined based on the logical intent options pre-selected by the user in the experimental dataset. The dependency relationship between the experimental dataset and the preceding experiments is determined based on one or more predefined preceding experimental data interfaces. Bind the logical intent and the identified dependencies to the experimental dataset as experimental metadata; The bound experimental datasets and experimental metadata are stored uniformly in a pre-defined structured database.

[0030] In this implementation, to achieve systematic data collection, the first step is to automatically acquire multiple datasets from different experimental sources through a standardized data interface. The principle behind this process is to establish a unified, automated data inflow channel to eliminate problems such as human error, inconsistent formats, and information omissions that may be introduced by manual import. In practice, an application programming interface (API) or a monitored file location (such as a specific directory on an FTP server) can be configured for various experimental instruments, simulation software, or data recording systems. When an experiment is completed and a data file is generated, the source system automatically calls the API to send the dataset containing the measurement results (e.g., in CSV or HDF5 format) as a data packet to a designated endpoint on the central data server. Alternatively, the system continuously scans a preset directory, triggering the acquisition process upon detecting a new file. Upon receiving the data, the program immediately performs preliminary format validation and assigns a globally unique identifier to ensure that each dataset can be tracked independently. After the raw data is captured by the system, the next step is to accurately capture and determine the logical intent of the experiment. The core principle of this step is to transform the abstract experimental objective in the researcher's mind into structured information that a machine can understand and process. Since the purpose of an experiment (such as verification, exploration, or reproduction) directly affects how its results are interpreted and their value in the knowledge system, it must be made explicit as soon as the data is generated.

[0031] An effective implementation involves embedding a mandatory user interaction step into the data acquisition process. When researchers submit or confirm experimental data, the system displays a graphical user interface (GUI) with a pre-defined list of options (e.g., drop-down menus or radio button groups) encompassing all possible logical intentions. Options may include: preliminary exploration of a new theory, confirmatory replication of previous conclusions, retrospective investigation of unexpected results, or optimization of existing experimental protocols. The user must select one option; the result of this selection, i.e., the logical intention of the experiment, is recorded and associated with the unique dataset identifier generated in the previous step.

[0032] To establish connections between experiments, it is essential to identify which preceding experiments the current experiment inherits from or responds to, i.e., to determine its dependencies. The fundamental principle of this step is to formally record the inherent knowledge inheritance chains in scientific research, ensuring that each new experiment can be accurately placed within the existing research context. In practice, this is usually performed concurrently with the aforementioned selection of logical intent. On the same user interface, in addition to selecting the logical intent, a functional area is provided for specifying one or more preceding experiments that the current experiment depends on. This area can be a searchable list or a visual graph displaying all existing experimental datasets in the database and their summary information. Users can precisely select one or more preceding datasets as the basis or reference for the current experiment by checking boxes, dragging and dropping, or entering IDs. The system records the unique identifiers of these selected preceding datasets, forming a dependency list, denoted as... ,in This is the set of parent node IDs for the current experiment. "predecessors" represents the name of a set that the researchers selected when specifying dependencies for the current experiment. "p" is the iteration variable, which needs to examine each member of the "predecessors" set one by one when processing the set. The unique ID representing the preceding experimental dataset that p points to.

[0033] Once the logical intent and dependencies are successfully determined, this contextual information must be tightly integrated with the original experimental data to form an indivisible, complete unit of information. The principle behind this step is to ensure the atomicity of data and its metadata; that is, the contextual interpretation must be present wherever the data is, thus preventing future misreading or information loss due to the separation of data and metadata. Specifically, this is achieved by creating a composite data structure at the software level, which serves as the smallest unit of experimental record. For example, an experimental object E can be defined, containing at least three core parts: the dataset itself D (or a pointer to its physical storage location), the logical intent of this experiment I, and a list of dependencies P from previous experiments. This binding process can logically be represented as creating a tuple. The metadata M is a collection containing information such as intent and dependencies, i.e. .

[0034] The final step is to store these bound and fully informed experimental units in a pre-defined structured database. The principle behind this step is to build a persistent, queryable, and scalable "single source of truth" for all experimental knowledge, thereby supporting long-term knowledge management and complex analytical queries. In implementation, a database system capable of efficiently handling relationships between entities is typically chosen. For example, a graph database (such as Neo4j) can be used. In such a database, each experimental unit can be directly modeled as a node, and the node's attributes can store its unique identifier, logical intent, data storage path, and other relevant information. Each parent-child relationship recorded in the dependency list can be modeled as a directed edge from the previous node to the current experimental node. This implementation naturally matches the dependency network structure between experiments.

[0035] In one implementation, constructing an experimental logical dependency graph based on the dependencies and logical intent recorded in the experimental metadata, with the experimental dataset as the dataset nodes and the dependencies as directed edges, includes the following steps: Initialize the graph data structure, traverse all acquired experimental datasets, and create a corresponding dataset node in the graph data structure for each experimental dataset; Based on the dependencies recorded in the experimental metadata, directed edges are created between dataset nodes pointing to subsequent experimental dataset nodes; The logical intent recorded in the experimental metadata is attached as an attribute to the corresponding directed edge, forming an experimental logical dependency graph that includes all dataset nodes and directed edges.

[0036] In this embodiment, refer to Figure 2 The first step in constructing the experimental logical dependency graph is to create corresponding entities, i.e., dataset nodes, for all independent experimental data in a virtual graph space. The fundamental principle of this process is to transform the experimental records, stored in a database in a flat list format, into basic building blocks of a network structure. The implementation begins by initializing an empty graph data structure in the program. This structure is conceptually a container, ready to hold the nodes and connections created subsequently. The system then initiates a traversal process, accessing each experimental dataset uniformly stored during the data acquisition phase. For each accessed experimental dataset, the program instantiates a brand new node object in the graph data structure. This node is not merely an empty shell; it is assigned a unique identifier, typically consistent with the ID of the corresponding experimental dataset in the database, to ensure traceability. The process of creating a set of nodes can be formally represented as follows: Where V represents the set of all dataset nodes in the graph, and This is the corresponding node created for each unique experimental dataset D, and ALLDatasets represents the complete set of all experimental datasets.

[0037] After all experimental datasets have been assigned independent node identities, the next crucial step is to establish directed edges representing the experimental flow and knowledge transfer between these nodes, based on the dependencies explicitly recorded in the metadata. The core principle of this step is to solidify the implicit sequence and dependency chains in the experimental design in a computable way within a graph structure. Specifically, this involves traversing each dataset node in the graph again. For any given node, such as the node representing a subsequent experiment... The system will query its bound metadata and extract a pre-recorded list of prerequisite experiments. This list contains unique identifiers for all prerequisite experiments used as the basis or reference for this experiment. Next, the program will locate the corresponding prerequisite node in the graph for each identifier in the list, such as node [node name missing]. Once these two nodes are located, a directed edge is created, pointing from the previous node to the next node. This process creates a corresponding edge for each dependency recorded in the metadata. The set of all these edges can be defined as... , where E is the set of all directed edges in the graph. This represents a directed edge from node i to node j, while It is the set of preceding experiment identifiers recorded in the metadata of the experiment corresponding to node j.

[0038] Finally, the logical intent information recording the experimental motivation needs to be appended to the corresponding directed edges, thus forming the final experimental logic dependency graph. The principle behind this step is to enrich the semantic information of the graph, providing crucial context for subsequent deeper analysis. Specifically, while creating each directed edge from a preceding node to a subsequent node, or after all edges have been created and a traversal is performed, the system extracts the logical intent of the experiment from the metadata of the subsequent nodes. This logical intent is then appended to the directed edge as an attribute or label; this appending process can be represented as an assignment operation.

[0039] In one implementation, attaching the logical intent recorded in the experimental metadata as an attribute to the corresponding directed edge includes the following steps: Different numerical influence factors are pre-defined for different logical intentions; When creating directed edges, the numerical influence factor is attached as an initial weight attribute to the corresponding directed edge; The efficiency of weight transfer between dataset nodes is adjusted during the calculation of static authority by utilizing the initial weight attribute.

[0040] In this implementation, to transform the subjective intent in experimental design into calculable quantitative indicators, a clear numerical influence factor needs to be pre-defined for each possible experimental logical intent. The fundamental principle behind this step is to acknowledge that different types of subsequent experiments have varying degrees of corroboration strength for prior conclusions. For example, a replication experiment specifically designed to verify prior findings, if successfully reproduced, provides far greater support for the prior conclusions than an experiment merely inspired by it and aimed at exploring entirely new directions. Therefore, by assigning different values ​​to different intents, this difference in logical strength can be quantified. In practice, a static mapping table or a function can be created to map a predefined set of logical intents (e.g., verification, replication, exploration, correction, application) to a set of values. This mapping relationship can be defined as... ,in The text description represents the k-th logical intent, where g is a pre-defined mapping function, and... This is its corresponding numerical impact factor.

[0041] Next, when constructing the directed edges of the graph based on experimental dependencies, the numerical influence factor set in the previous step is directly attached to each newly created edge as an initial weight attribute. The core principle of this operation is to concretely apply the abstract, global intent evaluation value to each connection in the graph, thus making the graph structure itself contain rich semantic information. At the moment of creating a directed edge, the system queries the metadata of the subsequent experiment to obtain its logical intent, and then finds the corresponding numerical influence factor according to the preset mapping relationship. This factor is then stored as an initial weight attribute on this directed edge. Through this step, each connection in the experimental logical dependency graph is no longer a simple pointing relationship, but becomes a channel with weights, whose weight values ​​represent the strength of the logical commitment or expected influence of the subsequent experiment on the preceding experiment.

[0042] Finally, these initial weight attributes attached to the directed edges are used to adjust and optimize the calculation process of static authority, allowing weights to be transferred between dataset nodes in a way that is more in line with scientific research logic. Static authority aims to assess the structural importance of an experiment's conclusion within the entire knowledge network. The basic idea is that an experiment relied upon by numerous high-authority experiments should also have higher authority. This step allows the calculated static authority to reflect not only the quantity of connections but also the quality of those connections. An experimental node pointed to by a high-weight edge (such as a confirmatory edge) will have a higher static authority than a node pointed to by an equal number of low-weight edges (such as an exploratory edge), thus more accurately assessing the core value and influence of each experimental data point within its logical structure.

[0043] In one implementation, the calculation of dynamic cognitive weights for each dataset node in the experimental logic dependency graph, based on the topological structure of the graph and the verification relationships formed by subsequent experimental results on the conclusions of previous experiments, includes the following steps: The topology based on the experimental logical dependency graph is used to calculate the initial static authority for each dataset node in the graph. The consistency of the results of subsequent experimental dataset nodes with the conclusions of previous experimental dataset nodes is determined by calculating the dynamic confidence of the directed edges connecting the dataset nodes. The dynamic cognitive weight of each dataset node is calculated by weighting and fusing static authority and dynamic confidence.

[0044] In this implementation, the prior importance of each experiment is quantitatively evaluated based on the inherent topological structure of the experimental logical dependency graph. A preceding experiment, which forms the basis for numerous subsequent experiments, should be assigned a higher initial authority. Specifically, an iterative algorithm is employed, the core of which is to propagate and accumulate the authority of nodes in the graph along the direction opposite to the experimental dependencies. Initially, all dataset nodes are assigned the same base authority score. Subsequently, in each iteration update, the authority score of a specific node i is recalculated by aggregating the authority scores of all its directly dependent subsequent nodes j, and this aggregation process is influenced by the initial weights of the connecting edges. This iterative update rule can be expressed as follows: ,in It is the static authority score of node i in the t-th iteration. It is the set of all subsequent nodes pointed to by node i, and It is a connecting edge The top represents the initial weight of the logical intent. This process is repeated until the authority scores of all nodes in the graph converge to a stable state.

[0045] Building upon static authority, an empirical verification dimension is introduced, specifically the validation of previous conclusions based on subsequent experimental results. Dynamic confidence is calculated for directed edges connecting dataset nodes. The reliability of a previous experiment's conclusion largely depends on whether subsequent experiments following its path yield consistent results. To quantify this consistency, a specific implementation method is as follows: For any directed edge in the graph from a previous node p to a subsequent node s, a baseline prediction model is first trained using the experimental dataset of the previous node. This model can be seen as the mathematical embodiment of the previous experimental conclusion. Then, the experimental dataset of the subsequent node is input as a test set into this trained model, and a prediction performance metric, such as classification accuracy or regression error, is calculated. This performance metric is compared with a pre-defined benchmark (such as the performance of random guessing) to quantify a consistency score. This score is then used to adjust the dynamic confidence of the directed edge. For example, the initial confidence could be set to 1, and then adjusted based on the consistency score: ,in is the dynamic confidence level of that edge, k is a scaling factor, and B is the expected baseline consistency score.

[0046] Finally, the static authority reflecting structural importance and the dynamic confidence embodying empirical verification are weighted and fused to calculate the final dynamic cognitive weight for each dataset node. In practice, the dynamic confidence scores on edges first need to be aggregated at the node level. For a preceding node p, the dynamic confidence scores of all its outgoing edges (edges pointing to subsequent nodes) can be summarized, for example, by weighted averaging, to obtain a comprehensive confidence score for that node. Then, the static authority of the node is calculated using a weighted summation formula. and its overall confidence score Combined, their final dynamic cognitive weights are calculated. This fusion formula can be expressed as: ,in It is the dynamic cognitive weight of node p, and This is a hyperparameter between 0 and 1, used to adjust the relative weight between structural importance and empirical credibility. The ultimate result of this step is to assign a refined, multi-dimensional weight value to each experimental dataset in the knowledge base. This weight not only reflects the design intent and logical position of the experiment but also incorporates the dynamic feedback from subsequent scientific practice, thus enabling a highly effective assessment of the true contribution of each dataset to building a reliable knowledge system.

[0047] In one implementation, the initial static authority of each dataset node in the graph is calculated based on the topology of the experimental logical dependency graph using the following steps: Assign an initial authority score to each dataset node in the experimental logic dependency graph; For each dataset node, an iterative calculation method is used, and the current authority score of the dataset node is updated based on the authority scores of all subsequent dataset nodes. During the iterative computation process, the logical intent attached to the directed edges is used as a regulating factor for the iterative computation. Repeat the iterative calculation until the authority scores of all dataset nodes converge; The converged authority score is used as the static authority score of the dataset node.

[0048] In this implementation, all experiments are initially considered equally important before any calculations are performed to avoid any preconceived bias. Their final authority is determined entirely by their connections and roles within the network. At the start of the algorithm, all dataset nodes in the graph are traversed, and each node is assigned the same non-zero constant as its authority score at the zeroth iteration. This initial score can be a simple positive number, or, for better numerical stability, can be the reciprocal of the total number of nodes. Then, for each dataset node, its current authority score is updated iteratively based on the authority scores of all its directly dependent dataset nodes. The authority of an experimental dataset is reflected in how many other experiments depend on and reference it. Therefore, a node's authority should be endorsed or contributed by the authority scores of all its directly dependent subsequent nodes. In practice, each iteration performs a comprehensive update of all nodes in the graph. For any given node, its new authority score in the next iteration is calculated by summing the authority scores of all its directly dependent nodes in the current iteration. This process is equivalent to passing the authority of subsequent nodes back along the directed edges to their predecessors. Without considering adjustment factors, the basic logic of this iterative update process can be represented as the accumulation of the authority of all subsequent nodes. The effect of this step is to initiate the dynamic flow and redistribution of authority within the network. After one iteration, the authority scores of nodes depended upon by more subsequent nodes will begin to rise, while the authority scores of nodes at the network edge or at the end of the dependency chain will relatively decrease, thus initially reflecting the structural differences between different nodes.

[0049] To make the calculation of authority more precise and consistent with scientific research logic, the logical intent previously attached to the directed edges needs to be introduced as a moderating factor during the iterative calculation process. The principle behind this step is that subsequent experiments with different logical intents have varying degrees of endorsement strength for preceding experiments. For example, a confirmatory experiment should contribute more to the authority of a preceding conclusion than an exploratory experiment. By introducing a moderating factor, this difference in contribution can be quantified. In practice, this moderating factor is the pre-set numerical initial weight attached to the edges. The transmission of authority is no longer homogeneous but discriminative. A node connected by high-weight edges (such as confirmatory dependencies) will more effectively accumulate authority than a node connected by an equal number of low-weight edges (such as exploratory dependencies), thus allowing the calculation results to more profoundly reflect the intrinsic quality of experimental dependencies, rather than just their quantity.

[0050] The iterative calculation process described above needs to be repeated until the authority scores of all dataset nodes in the graph reach a convergence state. The principle of convergence is that as iterations proceed, the distribution of authority scores in the network gradually shifts from dynamic change to a stable equilibrium state. At this point, further iterations will not significantly affect the scores. This equilibrium state represents the final authority distribution determined by both the graph structure and the logical intent. Specifically, after each iteration, the new authority score vector is compared with the score vector from the previous iteration. The difference between the two (e.g., the sum of the absolute values ​​of the changes in authority scores of all nodes) is used to determine whether convergence has been achieved. When this difference is less than a pre-set, very small positive threshold, the authority is considered to have converged, and the iteration process terminates. Finally, after the iterative calculation process terminates due to reaching the convergence condition, the authority scores obtained by all nodes in the last iteration are formally determined as their static authority scores.

[0051] In one implementation, the process of calculating the dynamic confidence score of the directed edges connecting the dataset nodes based on the consistency of the experimental results of subsequent experimental dataset nodes with the conclusions of previous experimental dataset nodes includes the following steps: For any preceding experimental dataset node and subsequent experimental dataset node that have a dependency relationship, train the benchmark prediction model based on the preceding experimental dataset node. The benchmark prediction model was tested using the subsequent experimental dataset of the subsequent experimental dataset node to obtain the prediction performance index. The predicted performance indicators are compared with the preset benchmark to quantify the consistency score that represents the degree of support or refutation. The dynamic confidence of directed edges connecting nodes in the preceding and subsequent experimental datasets is enhanced or weakened based on the consistency score of the results. The effect of adjusting the dynamic confidence level propagates along the path in the experimental logic dependency graph.

[0052] In this implementation, to quantitatively assess the credibility of the logical relationships between experiments, the conclusions of previous experiments must first be transformed into a mathematical model that can be objectively tested. The conclusion of any scientific experiment can essentially be seen as an assertion about a certain relationship or pattern between variables. By training a baseline prediction model on the dataset of previous experiments, this assertion is essentially fixed in the form of algorithms and parameters, making it a testable and concrete hypothesis. Specifically, for any dependency relationship in the graph from a previous node p to a subsequent node s, the system extracts the complete dataset corresponding to the previous node. Then, based on the focus of the experiment (e.g., which variable is the independent variable and which is the dependent variable), the dataset is divided into features and targets, and a suitable machine learning algorithm (such as linear regression, decision trees, or neural networks) is selected to train a baseline prediction model.

[0053] After successfully constructing a baseline prediction model representing the conclusions of previous experiments, the model is rigorously tested using the datasets from subsequent experiments to obtain a quantifiable prediction performance index. If subsequent experiments are a valid continuation or validation of previous experiments, then the patterns discovered in the data from previous experiments should also apply to the data from subsequent experiments. Specifically, the system acquires the experimental datasets corresponding to subsequent nodes and inputs them as test data into the baseline prediction model trained in the previous step. The model generates a set of prediction results based on the input features in the experimental datasets corresponding to subsequent nodes. Then, these prediction results are compared one by one with the actual observations in the complete datasets corresponding to the previous nodes, and a comprehensive prediction performance index is calculated based on preset evaluation criteria.

[0054] After obtaining the original predictive performance metrics, they need to be compared with a reasonable benchmark to quantify a more interpretable consistency score that characterizes the degree of support or refutation. Specifically, a benchmark performance value is pre-defined for the specific task; for example, for a binary classification problem, the benchmark performance value could be 0.5. Then, the actual measured predictive performance metrics are compared with this benchmark value, and the final consistency score is calculated using a normalization function. For example, a formula can be used... To perform the calculation, where This is the consistency score of the results. To predict performance indicators, This serves as the baseline performance value. Through this calculation, the score is normalized to an intuitive range. For example, a score close to 1 indicates that subsequent experimental results are highly consistent with previous conclusions (strong support), a score close to 0 indicates that it is no different from random guessing (no support), and a negative score indicates that the results are even worse than random guessing (potential rebuttal).

[0055] Based on the calculated consistency score, the dynamic confidence of directed edges connecting preceding and subsequent experimental dataset nodes is adjusted accordingly, either strengthening or weakening the confidence level. The principle behind this step is that the edges in the graph not only represent logical dependencies, but their weights should also dynamically reflect the true reliability of this dependency relationship, and this reliability evolves with the emergence of new empirical evidence. Specifically, for each directed edge in the graph... Initialize a dynamic confidence attribute Its initial value can be a neutral value (e.g., 1.0). After calculating the consistency score... The system will then update the score based on the confidence level of the opposite edge. ,in It is a learning rate or adjustment coefficient used to control the magnitude of the impact of a single experiment result on the confidence level. If If the value is positive (indicating support), the confidence of the edge will increase; conversely, if the value is negative (indicating rebuttal), the confidence will decrease.

[0056] When the dynamic confidence of an edge changes due to new experimental evidence, the effect of this adjustment needs to be propagated along the paths in the experimental logic dependency graph to ensure the internal consistency of the entire knowledge network. A reassessment of the confidence of a local connection will inevitably affect the overall confidence of other related knowledge points. For example, if the confidence of a basic experiment drops significantly due to rebuttals from subsequent experiments, the cognitive value of all downstream experiments that depend on this basic experiment should also be re-examined. Specifically, when the dynamic confidence of an edge is updated, the system recalculates the overall confidence of its parent node. This confidence may be an aggregation (such as a weighted average) of the dynamic confidence of all its outgoing edges. The change in the overall confidence of the parent node then serves as input, affecting the calculation of the authority or confidence of upstream nodes. This influence can propagate backward along the dependency path, adjusting step by step until the entire graph reaches a new equilibrium.

[0057] In one implementation, comparing the predicted performance index with a preset benchmark to quantify the consistency score representing the degree of support or refutation specifically includes the following steps: The baseline prediction model is regarded as a cognitive agent that can simulate the inherent laws of the preceding nodes. The cognitive agent is driven to actively predict the data of subsequent nodes, generating a prior belief that includes the expected data pattern and the range of uncertainty. The real experimental dataset of subsequent nodes is used as the external observation input for the cognitive agent; The Bayesian surprise measure is calculated between external observed inputs and prior beliefs. The Bayesian surprise measure quantifies the extent to which the actual data exceeds the model's expectations. The Bayesian surprise is inversely mapped to generate an outcome consistency score, where a lower surprise corresponds to a higher outcome consistency score.

[0058] In this implementation, the trained baseline prediction model is elevated from a simple function fitter to a probabilistic entity capable of expressing its own uncertainty. It must not only learn the main trends in the preceding experimental data but also the confidence level of those trends. An effective implementation method is to use Bayesian methods to train the model, such as Bayesian linear regression. Unlike traditional models that search for a unique optimal parameter, Bayesian methods calculate a posterior probability distribution of the model parameters. This process can be represented as learning a distribution. ,in This represents all the parameters of the model, while This is the dataset of the preceding node p. This distribution itself constitutes a cognitive agent, containing an estimate of the pattern itself and the uncertainty of that estimate. Next, we need to drive this cognitive agent to actively predict based on the conditions of subsequent experiments, thereby generating a prior belief containing the expected data pattern and the range of uncertainty. Based on existing theory, we predict the range of possible outcomes for the new experiment. Here, the cognitive agent plays the role of theory.

[0059] The specific implementation method involves extracting input variables or experimental conditions from the dataset of subsequent nodes s. These conditions are then input into the cognitive agent constructed in the previous step. The agent will generate a hypothesis about the output of subsequent experiments by integrating all possible parameters. The predicted distribution is the prior belief. This process can be represented as calculating the posterior predicted distribution. This distribution It is no longer a single numerical value, but a complete probability distribution, which can usually be described by a mean (the most likely outcome) and a variance (the range of uncertainty of the outcome).

[0060] After the cognitive agent has formed its prior beliefs, the next step is to use the real experimental dataset of subsequent nodes as external observation input and contrast it with the agent's beliefs. This involves using the complete dataset of subsequent nodes s... In this process, the output variable values ​​that are actually measured or observed and correspond to the predicted target are extracted. This value, or set of values, is denoted as... These observations constitute the external observation input received by the cognitive agent. With the model's prior beliefs and the actual external observations, the Bayesian surprise ratio can be calculated. If the observed value falls within the high-probability region of the prior belief distribution, it indicates that the model anticipated it, resulting in low surprise; conversely, if it falls within the low-probability tail region, it indicates that the model did not anticipate it, resulting in high surprise. A standard and effective implementation method is to calculate the negative log-likelihood of the observed data under the model's prior prediction distribution. Bayesian Surprise It can be defined as Here It refers to the probability density of prior beliefs at the point of actual observation. An event with a very low probability density has a large negative logarithm; inverting this logarithm yields a large positive logarithm.

[0061] Finally, the calculated Bayesian surprise score is inversely mapped to generate a more easily understood and usable consistency score. A reasonable mapping relationship should be that the lower the surprise score, the higher the consistency score, and vice versa. A specific implementation method can employ an exponential decay function. For example, the consistency score can be... Defined as ,in The score is the Bayesian surprise level calculated in the previous step, while K is an adjustable normal number used to control the sensitivity of the score. When the surprise level is zero (perfectly meets expectations), the score is 1; as the surprise level increases infinitely, the score will approach zero.

[0062] In one implementation, the topological structure of the experimental logic dependency graph is used as a structural prior to constrain the causal analysis model, and dynamic cognitive weights are used to weight the experimental dataset to train the causal analysis model to output the causal relationships between the data. This includes the following steps: Initialize a Bayesian network as a causal analysis model; All dataset nodes and directed edges in the experimental logical dependency graph are forcibly designated as the initial network topology of the Bayesian network to form a structural prior. When training the causal analysis model, the dynamic cognitive weight of each experimental dataset is applied to weight the contribution of the experimental dataset to the model parameter learning process. Input the experimental dataset into a constrained causal analysis model that requires weighted learning; The parameter learning algorithm is used to complete the training process of the causal analysis model.

[0063] In this implementation, a Bayesian network is an ideal choice. The fundamental principle of this step is that a Bayesian network intuitively represents a set of variables and their conditional probabilistic relationships through a directed acyclic graph (DAG), a structure naturally suited for exploring and expressing causal relationships. Nodes in the network represent variables, while directed edges between nodes represent the direct probabilistic influence of one variable on another. Initially, this network is merely an abstract mathematical structure, defining the model type and basic rules, but its specific internal parameters are not yet determined. First, a Bayesian network object is created, logically containing a set for storing nodes and a set for storing directed edges, while each node is associated with an empty conditional probability table placeholder, to be filled in subsequent steps. Next, all dataset nodes in the experimental dependency graph are traversed, creating a corresponding variable node in the Bayesian network for each node; then, all directed edges in the experimental dependency graph are traversed again, and for each edge from node to node, a directed edge from variable to variable is also created in the Bayesian network. This process can be represented as setting the edge set of the Bayesian network to be exactly the same as the edge set of the experimental logic dependency graph, which ultimately yields a causal analysis model built based on the Bayesian network.

[0064] When training the causal analysis model, the dynamic cognitive weights of each previously calculated experimental dataset are applied to weight the contribution of that dataset to the model parameter learning process. Specifically, this involves modifying the objective function of the parameter learning algorithm. Standard parameter learning processes (such as maximum likelihood estimation) treat all data points equally. However, in weighted learning, data points from high-weight datasets are assigned greater importance. The goal of this process is to find a set of model parameters. This maximizes the weighted log-likelihood function. The objective function can be expressed as: ,in It is the i-th experimental dataset. It refers to its corresponding dynamic cognitive weights, while G is the fixed network structure. The meaning is that under the network structure specified by G, when the parameter is At that time, experimental data were observed. What is the probability? The learning process involves adjusting parameters. This makes the probability maximized after weighting.

[0065] After the model is fully configured (i.e., the structure is constrained and the learning method is weighted), all relevant experimental datasets need to be provided as input to the causal analysis model to be trained. Since different experiments may contain variables with different ranges, these scattered datasets need to be integrated into one or more data tables of a uniform format, corresponding one-to-one with the variable nodes in the Bayesian network. Each row in the data table represents an observation instance, and each column represents a variable. Finally, a parameter learning algorithm is executed to complete the entire training process of the causal analysis model. Using the constrained and weighted data, specific probability values ​​are calculated for the conditional probability table (CPT) of each node in the Bayesian network. Specifically, this is achieved by running a parameter estimation algorithm such as weighted maximum likelihood estimation. The algorithm iterates through every node in the network; for each node, it selects all relevant data rows (weighted according to dynamic cognitive weights) and then fills the node's CPT by statistically analyzing the weighted frequency of different state combinations. For example, calculating... The value is calculated by dividing the weighted total number of times node X is in state x when the parent node is in state pa by the weighted total number of times the parent node is in state pa. The training process is complete once all nodes' CPTs are filled.

[0066] In one embodiment, the laboratory data analysis method further includes the following steps: Receive one or more candidate experimental protocols to guide future research; Each candidate experimental scheme is virtualized as a hypothetical dataset node in the experimental logic dependency graph, and the hypothetical dataset node is connected to the relevant existing dataset node. The trained causal analysis model is used to predict the expected experimental results of nodes in the hypothetical dataset. Based on the expected experimental results, the simulation calculation calculates the fluctuation impact of introducing hypothetical dataset nodes on the dynamic cognitive weights of other dataset nodes in the experimental logic dependency graph. A recommendation index is generated for each candidate experimental scheme based on the expected experimental results and the impact of fluctuations.

[0067] In this implementation, a candidate hypothesis is not merely a vague idea; it encompasses specific experimental variables, anticipated dependencies, and a clear scientific intent. This information can be captured through an interactive interface. Researchers need to define the core variables for each candidate hypothesis, specify the existing experimental foundations upon which it will be built (i.e., select its prerequisite nodes), and choose the logical intent of the hypothesis experiment (such as verification, exploration, etc.) from a pre-defined list. This information is integrated into a hypothesis experiment object, which can be represented as a tuple. ,in This represents the k-th candidate experimental scheme. This is the set of core variables involved in the scheme. It is the set of IDs of the preceding experimental nodes it depends on. This is the logical intent of the scheme. Each candidate experimental scheme is virtualized in the existing experimental logic dependency graph, that is, a temporary hypothetical dataset node is created for it, and it is connected to the graph according to its set dependency relationship.

[0068] For each candidate solution, the system creates a temporary copy of the existing experimental logical dependency graph in memory. Then, within this copy, a new dataset node labeled as the hypothesis is created. Subsequently, based on the dependency set defined in the scheme, the predecessor node of each ID in the set is determined. Create a line from Point to new node Directed edges are created. Simultaneously, the logical intent defined in the scheme is attached as an attribute to these newly created edges. After the hypothesis node is successfully placed in the graph, the previously trained causal analysis model needs to be used to predict the most likely experimental outcome of this hypothesis experiment. Specifically, in the enhanced temporary graph, the hypothesis node is located... This involves the input of the network and all its parent nodes. Then, the known states or data distributions of these parent nodes in real-world experiments are used as evidence and input into the Bayesian network. By executing a probabilistic inference algorithm, the hypothetical nodes are calculated. The variable it represents The posterior probability distribution. This prediction result can be expressed as... ,in It is a complete probability distribution that includes the expected outcome and its uncertainty. Parents represents the found hypothesis node. The function of all parent nodes, where P represents the probability distribution.

[0069] After obtaining the expected experimental results, the next step is simulation calculation. If this expected result actually occurs, how will the introduction of this new hypothetical node affect the dynamic cognitive weights of other nodes in the experimental logic dependency graph? Specifically, the predicted result from the previous step is treated as the actual observed data of the hypothetical node. Then, based on this, the entire calculation process of dynamic cognitive weights is rerun: first, the dynamic confidence of the new edge connecting to the hypothetical node is calculated; then, the change in confidence is backpropagated to update the authority and confidence of its parent node, iterating until the weights of the entire graph reach a new equilibrium. Finally, the fluctuation impact is quantified by calculating the difference between the old and new weights, and its impact index is determined. Can be defined as ,in and These are the dynamic cognitive weights of any node i before and after the introduction of the hypothesis experiment.

[0070] Finally, by combining the expected experimental results obtained from the preceding steps with the simulated fluctuation effects, a final recommendation index is generated for each candidate experimental scheme. Specifically, this is achieved by designing a weighted scoring function that considers multiple factors. For example, the recommendation index of an experiment should not only be determined by its impact on cognitive weight fluctuations. The correlation is direct and may also be related to other factors, such as the scientific significance or importance of the expected result itself. (Can be defined by experts or determined based on its position in the causal network). Recommendation Index It can be a weighted sum of these factors: , ,in and These are weighting coefficients used to balance different consideration dimensions. After calculating the recommendation index for all candidate options, they are ranked.

[0071] The present invention also discloses a laboratory data analysis system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the laboratory data analysis method as described in any of the above embodiments.

[0072] The processor can be a central processing unit (CPU). Of course, depending on the actual use, it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it.

[0073] The memory can be an internal storage unit of a computer device, such as a hard disk or RAM, or an external storage device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) provided on the computer device. Furthermore, the memory can be a combination of internal storage units and external storage devices of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0074] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0075] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A laboratory data analysis method, characterized by, The method includes the following steps: Obtain multiple experimental datasets, as well as experimental metadata associated with each experimental dataset, which includes at least the dependencies on previous experiments and the logical intent of this experiment; Construct an experimental logical dependency graph based on the dependencies and logical intent recorded in the experimental metadata, with the experimental dataset as the dataset nodes and the dependencies as directed edges; Based on the topological structure of the experimental logic dependency graph and the verification relationship formed by the results of subsequent experiments on the conclusions of previous experiments in the experimental logic dependency graph, dynamic cognitive weights are calculated for each dataset node in the experimental logic dependency graph. The topological structure of the experimental logical dependency graph is used as a structural prior to constrain the causal analysis model, and dynamic cognitive weights are used to weight the experimental dataset to train the causal analysis model to output the causal relationships between the data. The topological structure of the comprehensive experimental logical dependency graph and the verification relationships formed by subsequent experimental results on the conclusions of previous experiments in the experimental logical dependency graph, and the calculation of dynamic cognitive weights for each dataset node in the experimental logical dependency graph include the following steps: The topology based on the experimental logical dependency graph is used to calculate the initial static authority for each dataset node in the graph. The consistency of the results of subsequent experimental dataset nodes with the conclusions of previous experimental dataset nodes is determined by calculating the dynamic confidence of the directed edges connecting the dataset nodes. The dynamic cognitive weight of each dataset node is calculated by weighting and fusing static authority and dynamic confidence. The topology based on the experimental logical dependency graph calculates the initial static authority of each dataset node in the graph using the following steps: Assign an initial authority score to each dataset node in the experimental logic dependency graph; For each dataset node, an iterative calculation method is used, and the current authority score of the dataset node is updated based on the authority scores of all subsequent dataset nodes. During the iterative computation process, the logical intent attached to the directed edges is used as a regulating factor for the iterative computation. Repeat the iterative calculation until the authority scores of all dataset nodes converge; The converged authority score is used as the static authority score of the dataset node; The consistency of the experimental results based on the quantized subsequent experimental dataset nodes with the conclusions of the preceding experimental dataset nodes is calculated by using the directed edges connecting the dataset nodes to determine the dynamic confidence score, which includes the following steps: For any preceding experimental dataset node and subsequent experimental dataset node that have a dependency relationship, train the benchmark prediction model based on the preceding experimental dataset node. The benchmark prediction model was tested using the subsequent experimental dataset of the subsequent experimental dataset node to obtain the prediction performance index. The predicted performance indicators are compared with the preset benchmark to quantify the consistency score that represents the degree of support or refutation. The dynamic confidence of directed edges connecting nodes in the preceding and subsequent experimental datasets is enhanced or weakened based on the consistency score of the results. The effect of adjusting the dynamic confidence level is propagated along the path in the experimental logic dependency graph; The step of using the topological structure of the experimental logical dependency graph as a structural prior to constrain the causal analysis model, and using dynamic cognitive weights to weight the experimental dataset to train the causal analysis model to output the causal relationships between the data includes the following steps: Initialize a Bayesian network as a causal analysis model; All dataset nodes and directed edges in the experimental logical dependency graph are forcibly designated as the initial network topology of the Bayesian network to form a structural prior. When training the causal analysis model, the dynamic cognitive weight of each experimental dataset is applied to weight the contribution of the experimental dataset to the model parameter learning process. Input the experimental dataset into a constrained causal analysis model that requires weighted learning; The parameter learning algorithm is used to complete the training process of the causal analysis model.

2. The laboratory data analysis method of claim 1, wherein, The process of acquiring multiple experimental datasets, and experimental metadata associated with each experimental dataset that includes at least the dependencies on previous experiments and the logical intent of the current experiment, includes the following steps: Multiple experimental datasets can be obtained through a pre-defined data interface; The logical intent of the experimental dataset corresponding to this experiment is determined based on the logical intent options pre-selected by the user in the experimental dataset. The dependency relationship between the experimental dataset and the preceding experiments is determined based on one or more predefined preceding experimental data interfaces. Bind the logical intent and the identified dependencies to the experimental dataset as experimental metadata; The bound experimental datasets and experimental metadata are stored uniformly in a pre-defined structured database.

3. The laboratory data analysis method of claim 1, wherein, The process of constructing an experimental logical dependency graph based on the dependencies and logical intent recorded in the experimental metadata, with the experimental dataset as the dataset nodes and the dependencies as directed edges, includes the following steps: Initialize the graph data structure, traverse all acquired experimental datasets, and create a corresponding dataset node in the graph data structure for each experimental dataset; Based on the dependencies recorded in the experimental metadata, directed edges are created between dataset nodes pointing to subsequent experimental dataset nodes; The logical intent recorded in the experimental metadata is attached as an attribute to the corresponding directed edge, forming an experimental logical dependency graph that includes all dataset nodes and directed edges.

4. The laboratory data analysis method of claim 3, wherein, The step of attaching the logical intent recorded in the experimental metadata as an attribute to the corresponding directed edge includes the following steps: Different numerical influence factors are pre-defined for different logical intentions; When creating directed edges, the numerical influence factor is attached as an initial weight attribute to the corresponding directed edge; The efficiency of weight transfer between dataset nodes is adjusted during the calculation of static authority by utilizing the initial weight attribute.

5. The laboratory data analysis method according to claim 4, characterized in that, The method further includes the following steps: Receive one or more candidate experimental protocols to guide future research; Each candidate experimental scheme is virtualized as a hypothetical dataset node in the experimental logic dependency graph, and the hypothetical dataset node is connected to the relevant existing dataset node. The trained causal analysis model is used to predict the expected experimental results of nodes in the hypothetical dataset. Based on the expected experimental results, the simulation calculation calculates the fluctuation impact of introducing hypothetical dataset nodes on the dynamic cognitive weights of other dataset nodes in the experimental logic dependency graph. A recommendation index is generated for each of the candidate experimental protocols based on expected experimental results and fluctuation effects.

6. A laboratory data analysis system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor implements the laboratory data analysis method as claimed in any one of claims 1 to 5 when executing the computer program.

Images