A multi-scale causal discovery method and system based on collective behavior modeling
By constructing a multi-scale causal model, using attention mechanisms and multi-scale collaborative learning strategies, the problem of insufficient nonlinear features of collective behavior in complex systems is solved, and more accurate and explainable causal discovery is achieved.
Patent Information
- Application Number
- CN202510486673.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-18
- Publication Date
- 2025-09-02
- Estimated Expiration
- 2045-04-18
AI Technical Summary
Traditional causal discovery methods are difficult to accurately capture the nonlinear characteristics of collective behavior in complex systems, resulting in insufficient identification of causal relationships and unable to reveal the causal mechanism in complex systems.
A multi-scale causal discovery method based on collective behavior modeling is adopted. By constructing a multi-scale causal model, using attention mechanisms and multi-scale collaborative learning strategies, we collaboratively learn the multi-scale causal relationship between individuals and collectives, introduce cross-scale consistency constraints, and optimize the causal structure.
It improves the accuracy, rationality and robustness of causal discovery, can effectively identify key collective behaviors that drive state changes, and supports multi-scale causal analysis.
Smart Images
Figure CN120015115B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of causal discovery, and in particular to a multi-scale causal discovery method and system based on collective behavior modeling. Background Art
[0002] The core of causal discovery lies in uncovering the underlying mechanisms behind observed phenomena. Distinguishing causality from mere correlation in complex systems is crucial for effectively designing interventions and formulating policies. However, traditional causal discovery methods primarily focus on analyzing the interactions between individuals in a system, often overlooking the collective behavior formed by the interaction of multiple individuals.
[0003] In reality, in ecosystems, economic systems, and social networks, the collective effects of individual behaviors can often trigger nonlinear feedback effects, profoundly influencing the dynamics of the system. The complexity of this collective behavior makes it difficult for traditional causal discovery methods to accurately capture the dynamic characteristics within the system. This is especially true in many complex real-world systems, where interactions between individuals are not simply linear superpositions, but rather stem from complex network structures and multi-scale interactions, resulting in complex nonlinear characteristics in collective behavior. This nonlinear relationship suggests that small changes can trigger significant reactions within the system, but traditional methods are inadequate in identifying and modeling this complexity, leading to omissions and deviations in the discovered causal relationships. Furthermore, the abstract nature of collective states and the local optimization strategies of existing score-based causal discovery algorithms further complicate the study of nonlinear influences, making the process of extracting effective causal information from data even more complex.
[0004] Therefore, existing causal discovery methods lack an in-depth understanding of collective behavior and the exploration of its nonlinear effects, and are unable to accurately reveal the causal mechanisms in complex systems. Summary of the Invention
[0005] In order to solve the above problems, this paper proposes a multi-scale causal discovery method and system based on collective behavior modeling, innovatively constructs a collective behavior modeling framework, which can effectively identify the key collective behaviors that drive state changes, and improves the accuracy, rationality and robustness of causal discovery.
[0006] According to some embodiments, the present disclosure adopts the following technical solutions:
[0007] A multi-scale causal discovery method based on collective behavior modeling, including:
[0008] Acquire single-cell RNA sequencing data, wherein the single-cell RNA sequencing data is a gene expression matrix of multiple cells;
[0009] The pre-processed single-cell RNA sequencing data is input into a multi-scale causal model for inference to obtain a multi-scale causal graph, which is used to describe the multi-scale causal relationships between genes and genes, and between genes and gene programs;
[0010] Among them, the multi-scale causal model takes a single gene as an individual variable, aggregates it into a collective variable that represents the genetic program through collective behavior modeling, and adopts a multi-scale collaborative learning strategy to collaboratively learn the multi-scale causal relationship between individuals and individuals, and individuals and collectives.
[0011] According to some embodiments, the present disclosure adopts the following technical solutions:
[0012] A multi-scale causal discovery system based on collective behavior modeling, including:
[0013] The data acquisition module is configured to: acquire single-cell RNA sequencing data, wherein the single-cell RNA sequencing data is a gene expression matrix of multiple cells;
[0014] The causal discovery module is configured to: input the preprocessed single-cell RNA sequencing data into a multi-scale causal model for inference to obtain a multi-scale causal graph, wherein the multi-scale causal graph is used to describe the multi-scale causal relationships between genes and genes, and between genes and gene programs;
[0015] Among them, the multi-scale causal model takes a single gene as an individual variable, aggregates it into a collective variable that represents the genetic program through collective behavior modeling, and adopts a multi-scale collaborative learning strategy to collaboratively learn the multi-scale causal relationship between individuals and individuals, and individuals and collectives.
[0016] According to some embodiments, the present disclosure adopts the following technical solutions:
[0017] A computer program product includes a computer program, which, when executed by a processor, implements the multi-scale causal discovery method based on collective behavior modeling.
[0018] According to some embodiments, the present disclosure adopts the following technical solutions:
[0019] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, a multi-scale causal discovery method based on collective behavior modeling is implemented.
[0020] According to some embodiments, the present disclosure adopts the following technical solutions:
[0021] An electronic device includes: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the multi-scale causal discovery method based on collective behavior modeling.
[0022] Compared with the prior art, the present invention has the following beneficial effects:
[0023] (1) This paper proposes a multi-scale causal discovery method based on attention mechanism and multi-scale collaborative learning. This method constructs a collective representation model, aggregates individual variables into collective variables, and adopts a two-stage optimization strategy to collaboratively learn the multi-scale causal relationship between individuals and collectives, thereby effectively revealing the multi-scale causal network structure in complex systems.
[0024] (2) The present invention uses the attention mechanism to dynamically calculate the weight of individual contributions to collective representation, making collective behavior modeling interpretable. It also supports the discovery of causal relationships from the individual to the collective scale, providing a theoretical basis for multi-scale causal analysis.
[0025] (3) The present invention introduces cross-scale consistency constraints in the optimization process to ensure that the causal structures of the three parts, individual-individual, individual-collective, and collective-individual, are coordinated with each other, avoid causal loops, and improve the rationality and robustness of causal discovery.
[0026] (4) The present invention adopts a parameterized causal extraction strategy to directly decode multi-scale causal relationships from the optimized model weights, which is suitable for multi-scale causal analysis in fields such as biomedicine and social sciences. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] The accompanying drawings, which constitute a part of the present disclosure, are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation to the present disclosure.
[0028] Figure 1 This is a flow chart of the method of Example 1.
[0029] Figure 2 This is an example diagram of the multi-scale causal graph of Example 1. DETAILED DESCRIPTION
[0030] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.
[0031] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present disclosure belongs.
[0032] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "comprising" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.
[0033] In response to the research challenges brought by the abstractness of collective states, the present invention proposes a novel multi-scale causal discovery framework, which integrates existing causal learning technologies and uses deep learning methods (such as attention mechanisms) to intelligently aggregate individual observation data, effectively identifying key collective behaviors that drive changes in system states, and simulating their feedback influence mechanisms on individual behaviors through learned collective representations. At the same time, it accurately models the influence of individuals and their collective behaviors, effectively solving the technical difficulties that traditional methods have in quantifying collective states. In terms of causal discovery algorithms, the present invention proposes an innovative two-stage collaborative optimization strategy, which collaboratively optimizes the causal effects at the individual and collective levels, improves the robustness of multi-scale causal discovery, avoids the local optimal trap of traditional score optimization methods, maintains the consistency of cross-scale causal structures, and provides new methodological support for multi-scale causal analysis in complex systems.
[0034] Example 1
[0035] One embodiment of the present disclosure provides a multi-scale causal discovery method based on collective behavior modeling. This method innovatively constructs a collective behavior modeling framework that can effectively identify key collective behaviors that drive state changes, improving the accuracy, rationality, and robustness of causal discovery. The method includes two major steps:
[0036] Step 1: Obtain single-cell RNA sequencing data, which is a gene expression matrix of several cells;
[0037] Step 2: Input the preprocessed single-cell RNA sequencing data into a multi-scale causal model for inference to obtain a multi-scale causal graph, which is used to describe the multi-scale causal relationships between genes and genes, and between genes and gene programs;
[0038] Among them, the multi-scale causal model takes a single gene as an individual variable, aggregates it into a collective variable that represents the genetic program through collective behavior modeling, and adopts a multi-scale collaborative learning strategy to collaboratively learn the multi-scale causal relationship between individuals and individuals, and individuals and collectives.
[0039] As an embodiment, the specific implementation process of the multi-scale causal discovery method disclosed herein is described below by taking the mining of multi-scale causal mechanisms in gene regulatory networks as an example.
[0040] In a gene regulatory network, a gene program refers to a set of co-regulated genes that jointly perform specific biological functions through spatiotemporal expression patterns. Traditional methods (such as co-expression network analysis) can only identify correlations between genes, but have difficulty distinguishing causal regulatory relationships, let alone revealing the potential gene program formed by the synergistic effects of multiple genes. This example systematically explores the multi-scale causal mechanisms hidden in the gene regulatory network by modeling the micro-regulation between individual genes and the macro-causal effects at the gene program level. Figure 1 As shown, the specific steps are:
[0041] S101. Data acquisition and preprocessing
[0042] S1011. Data Acquisition
[0043] Specifically, single-cell RNA sequencing data (scRNA-seq) is obtained, which is a gene expression matrix of several genes in several cells. The gene expression matrix uses cells as rows and genes as columns to represent the expression value of a single gene in a single cell. In this example, the number of cells is 10,000 and the number of genes is 20,000.
[0044] In the following steps, each row of data in the gene expression matrix (i.e., the gene expression data of each cell) is used as a sample, the expression value of a single gene is used as the value of the individual variable, and the gene program is used as the potential collective variable to infer and mine the multi-scale causal relationships between genes and genes, and between genes and gene programs.
[0045] S1012. Data Preprocessing
[0046] Specifically, after obtaining all cell samples, the samples are standardized using statistical quantities such as mean and variance.
[0047] S102. Collective Behavior Modeling
[0048] We use a deep learning model based on the attention mechanism to aggregate individual variables, visualize the abstract collective, and simulate collective behavior based on the impact of the aggregated data on other individuals. The specific implementation method is as follows:
[0049] S1021, Individual Enhancement Representation
[0050] Specifically, let For the Individual variables, build a multi-layer feedforward neural network, map it to the latent semantic space where the collective variables are located, and use it for subsequent prediction of collective representation to support multi-scale causal discovery, multi-layer feedforward neural network It can be expressed as:
[0051] (1)
[0052] in, Individual variables Mapped to the latent representation of the semantic space where the collective variable is located, is the number of individual variables, is the number of potential collective variables, For Parameters of the constructed multi-layer feedforward neural network.
[0053] S1022. Collective Representation Based on Attention Mechanism
[0054] Specifically, let Represent the potential representation matrix of all individual variables, use the attention mechanism to further weight the potential representation, predict reliable collective representation to further study collective behavior, and let the weight matrix of individual variables on collective variables be , the process of calculating the collective representation is expressed as:
[0055]
[0056] (2)
[0057]
[0058] in, is the calculated collective variable, are the key vector and value vector in the attention mechanism, is the corresponding linear transformation matrix in the attention mechanism.
[0059] In the subsequent optimization process, Can represent individual variables The importance of collective representation also serves as the basis for subsequent multi-scale causal structure extraction.
[0060] S103, Multi-scale Collaborative Learning
[0061] Individual variables and collective variables are used together with reconstruction as the training goal, cross-scale consistency constraints are introduced, and the augmented Lagrangian method is used to optimize the learning of the weight matrix of individual variables to individual variables and the weight matrix of collective variables to individual variables.
[0062] S1031, Multi-scale Reconstruction
[0063] Specifically, let is the weight matrix of individual variables to individual variables, is the weight matrix of collective variables to individual variables, through these two weight matrices and collective variables , using Multilayer Perceptron (MLP) to analyze individual variables Reconstruction is expressed as follows:
[0064] (3)
[0065] in, are the multi-layer perceptron parameters to be trained, and ⊕ is the concatenation operation.
[0066] Multi-scale collaborative learning aims to minimize the reconstruction loss, so the loss function is defined as:
[0067] (4)
[0068] S1032, cross-scale consistency constraints
[0069] To collaboratively learn multi-scale causal structures, this embodiment imposes cross-scale consistency constraints in multi-scale collaborative learning:
[0070] (5)
[0071] in, The individual relationship matrix considers the individual and their collective behaviors, and its physical meaning is the edges that different individual variables can directly affect or mediate through the collective.
[0072] On this basis, the optimization objective of multi-scale collaborative learning is expressed as:
[0073]
[0074] (6)
[0075] in, It is a loop-free constraint, which constrains individual variables from affecting themselves directly or indirectly. At the same time, it can be seen from formula (5) that this constraint will simultaneously constrain the influence of individuals on the collective, the influence of the collective on individuals, and the influence of individuals on individuals without loops.
[0076] S1033, two-stage optimization
[0077] In order to find the multi-scale causal structure, multi-scale collaborative optimization learning is performed on all parameters, where all parameters include the weight matrix of individual variables to individual variables , the weight matrix of collective variables to individual variables , the weight matrix of individual variables to collective variables , Multilayer Perceptron parameters and multilayer feedforward neural network parameters If formula (6) is directly used as the optimization target for optimization, the causal discovery results will be too dependent on the initial model values, and an effective causal structure cannot be found. Therefore, this embodiment uses a two-stage optimization algorithm to divide the optimization process into two scales, individual and collective, to optimize the model parameters respectively, effectively improving the accuracy of causal discovery. In the two stages, the augmented Lagrangian method is used to optimize the model parameters of different parts:
[0078] (7)
[0079] in, is the penalty coefficient, is the Lagrange multiplier, the constrained problem in Equation (6) can be transformed into a series of unconstrained subproblems that can be solved by the gradient descent optimization strategy as follows:
[0080] (8)
[0081] (9)
[0082] (10)
[0083] When optimizing individual scales, , when optimizing the collective scale, , , 、 、 It is Optimized in iterations 、 、 ,This embodiment applies the L-BFGS-B algorithm to solve the optimization problem.
[0084] S104. Multi-scale causal relationship extraction
[0085] Specifically, to obtain multi-scale causal relationships, this embodiment intends to extract the causal relationships at the individual to individual level, the causal relationships at the individual to collective level, and the causal relationships at the collective to individual level from the model parameters, and combine the causal relationships of the above three parts to obtain the final multi-scale causal graph.
[0086] S1041. Extraction of individual-scale causal relationships
[0087] According to the set threshold , the weight matrix of individual variables learned previously for individual variables Binarization is used to obtain the causal relationship between individuals:
[0088] (11)
[0089] in, When Individual variables and There is a causal relationship between individual variables, but not vice versa.
[0090] S1042. Collective scale causal relationship extraction
[0091] According to the set threshold , respectively, the weight matrix of individual variables to collective variables obtained in the previous study The weight matrix of collective variables on individual variables Binarization is performed to obtain the causal relationships from individual to collective and from collective to individual.
[0092] (12)
[0093] (13)
[0094] Based on the causal relationship between individuals, individuals to collective parts and collective to individuals, construct Figure 2 The multi-scale causal diagram shown.
[0095] Example 2
[0096] In one embodiment of the present disclosure, a multi-scale causal discovery system based on collective behavior modeling is provided, comprising:
[0097] The data acquisition module is configured to: acquire single-cell RNA sequencing data, wherein the single-cell RNA sequencing data is a gene expression matrix of multiple cells;
[0098] The causal discovery module is configured to: input the preprocessed single-cell RNA sequencing data into a multi-scale causal model for inference to obtain a multi-scale causal graph, wherein the multi-scale causal graph is used to describe the multi-scale causal relationships between genes and genes, and between genes and gene programs;
[0099] Among them, the multi-scale causal model takes a single gene as an individual variable, aggregates it into a collective variable that represents the genetic program through collective behavior modeling, and adopts a multi-scale collaborative learning strategy to collaboratively learn the multi-scale causal relationship between individuals and individuals, and individuals and collectives.
[0100] Example 3
[0101] In one embodiment of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the multi-scale causal discovery method based on collective behavior modeling.
[0102] Example 4
[0103] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, which is used to store computer instructions. When the computer instructions are executed by a processor, the multi-scale causal discovery method based on collective behavior modeling is implemented.
[0104] Example 5
[0105] In one embodiment of the present disclosure, an electronic device is provided, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the multi-scale causal discovery method based on collective behavior modeling.
[0106] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0108] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.
Claims
1. A multi-scale causal discovery method based on collective behavior modeling, characterized by: include: Acquire single-cell RNA sequencing data, wherein the single-cell RNA sequencing data is a gene expression matrix of multiple cells; The pre-processed single-cell RNA sequencing data is input into a multi-scale causal model for inference to obtain a multi-scale causal graph, which is used to describe the multi-scale causal relationships between genes and genes, and between genes and gene programs; The multi-scale causal model takes individual genes as individual variables, aggregates them into collective variables that represent genetic programs through collective behavior modeling, and adopts a multi-scale collaborative learning strategy to collaboratively learn the multi-scale causal relationships between individuals and individuals, and between individuals and collectives; The multi-scale causal model includes a collective behavior modeling module, which uses a deep learning model based on the attention mechanism to aggregate individual variables to obtain collective variables and dynamically calculate the weight matrix of individual variables to collective variables, specifically: Individual enhancement means: For the Individual variables, build a multi-layer feedforward neural network, map it to the latent semantic space where the collective variables are located, and use it for subsequent prediction of collective representation to support multi-scale causal discovery, multi-layer feedforward neural network It can be expressed as: in, Individual variables Mapped to the latent representation of the semantic space where the collective variable is located, is the number of individual variables, is the number of potential collective variables, For Parameters of the constructed multi-layer feedforward neural network; Collective representation based on attention mechanism: Let Represent the potential representation matrix of all individual variables, use the attention mechanism to further weight the potential representation, predict reliable collective representation to further study collective behavior, and let the weight matrix of individual variables on collective variables be , the process of calculating the collective representation is expressed as: in, is the calculated collective variable, are the key vector and value vector in the attention mechanism, is the corresponding linear transformation matrix in the attention mechanism.
2. The multi-scale causal discovery method based on collective behavior modeling according to claim 1, characterized in that: The gene expression matrix, with cells as rows and genes as columns, is used to represent the expression value of a single gene in a single cell; The preprocessing is to take each row of data in the gene expression matrix as a sample and perform standardization processing on all samples.
3. The multi-scale causal discovery method based on collective behavior modeling according to claim 1, characterized in that: The multi-scale causal model also includes a multi-scale collaborative learning module, which uses individual variables and collective variables to jointly reconstruct as a training goal, introduces cross-scale consistency constraints, and adopts an augmented Lagrangian method to optimize the learning of the weight matrix of individual variables to individual variables and the weight matrix of collective variables to individual variables.
4. The multi-scale causal discovery method based on collective behavior modeling according to claim 3, characterized in that: The reconstruction is used as the training goal, and the cross-scale consistency constraint is introduced, which can be expressed as follows: in, Individual variables and reconstructed individual variables, To consider the individual relationship matrix of individuals and their collective behaviors, They are the weight matrix of individual variables to individual variables, the weight matrix of collective variables to individual variables, and the weight matrix of individual variables to collective variables, It is an acyclic constraint, which means that individual variables cannot directly or indirectly affect themselves. is the number of individual variables.
5. The multi-scale causal discovery method based on collective behavior modeling according to claim 1, characterized in that: The multi-scale causal model also includes a multi-scale causal relationship extraction module, which extracts the causal relationship from individual to individual level, the causal relationship from individual to collective level, and the causal relationship from collective to individual level based on the weight matrix of individual variables to individual variables, the weight matrix of collective variables to individual variables, and the weight matrix of individual variables to collective variables, to form the final multi-scale causal graph.
6. A multi-scale causal discovery system based on collective behavior modeling, characterized by: A multi-scale causal discovery method based on collective behavior modeling according to any one of claims 1 to 5 is adopted, comprising: The data acquisition module is configured to: acquire single-cell RNA sequencing data, wherein the single-cell RNA sequencing data is a gene expression matrix of multiple cells; The causal discovery module is configured to: input the preprocessed single-cell RNA sequencing data into a multi-scale causal model for inference to obtain a multi-scale causal graph, wherein the multi-scale causal graph is used to describe the multi-scale causal relationships between genes and genes, and between genes and gene programs; Among them, the multi-scale causal model takes a single gene as an individual variable, aggregates it into a collective variable that represents the genetic program through collective behavior modeling, and adopts a multi-scale collaborative learning strategy to collaboratively learn the multi-scale causal relationship between individuals and individuals, and individuals and collectives.
7. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the multi-scale causal discovery method based on collective behavior modeling according to any one of claims 1 to 5 is implemented.
8. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by the processor, the multi-scale causal discovery method based on collective behavior modeling as described in any one of claims 1 to 5 is implemented.
9. An electronic device, characterized in that: include: A processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement a multi-scale causal discovery method based on collective behavior modeling as described in any one of claims 1 to 5.