Causal relationship prediction method and system based on intervention diffusion model

Abstract

The invention discloses a causal relationship prediction method and system based on an intervention diffusion model, and the method comprises the following steps: (1) calculating the correlation between expression data of a specific representation, and constructing an initial correlation network of the expression data; (2) dividing expression data into a training set and a test set; (3) training a diffusion model by using the training set data; (4) on the basis of the initial correlation network, performing H0 hypothesis and H1 hypothesis on each edge; performing different intervention operations on the trained diffusion model by using the test set data to obtain a predicted value after intervention; (5) judging whether a causal relationship exists between variables with connected edges in the initial correlation network or not according to errors between predicted values and true values of the two hypotheses; and (6) traversing all variable pairs with connecting edges in the initial correlation network according to the steps (4)-(5) to obtain a final causal relationship network. According to the method, the accuracy and generalization of network reasoning are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computational biology and bioinformatics, and in particular to a causal relationship prediction method and system based on an intervention diffusion model. Background Art

[0002] Understanding causal relationships is essential for elucidating the molecular mechanisms of biological processes at the network level. The core of network inference such as gene regulatory networks or protein interaction networks, as well as the inference of disease drivers from genes / molecules to disease phenotypes, fundamentally revolves around causal relationships.

[0003] Traditional network inference methods, including tools such as GENIE3 and PLSNet for gene regulatory networks, often rely on statistical methods or association studies without considering causal relationships, and therefore often lead to inaccurate and spurious correlations. Other methods, such as CellOracle, require specific prior knowledge, such as identification of transcription factors (TFs) and their targets. scMTNI also requires additional information, such as the detection of transposase-accessible chromatin using sequencing (ATAC-seq). Although these methods are effective in their respective fields, the lack of versatility and the struggle with the lack of prior knowledge limit their applicability in reconstructing networks from observational data. On the other hand, the most widely used method for causal inference from genotype / molecule to phenotype is Mendelian randomization, which bridges the genotype-phenotype problem through SNP sites. However, this method is mainly limited to SNP analysis and has multivariate problems such as a high risk of false negative results.

[0004] In the field of traditional causal inference, Granger causality is indeed a commonly used method. However, Granger causality is mainly applicable to time series data and may not be very suitable for non-time series data often encountered in biology. Granger causality is limited to time series data, making it inapplicable to most biological datasets. Another commonly used method is the additive noise model (ANM), which is often used to evaluate causal identifiability. ANM focuses on determining the causal relationship between a pair of variables by incorporating additive noise terms. However, ANM usually only considers one pair of variables and may not fully consider the influence of other variables in the system. CVP is a method for inferring genetic causal relationships, but it can only determine linear relationships.

[0005] Therefore, how to break through the limitations of additional information and properly combine current causal theories to infer causal relationships in real biological or molecular systems with high precision and more effectively reveal the correlation between data to provide a strong basis for practical decision-making is still a technical problem that urgently needs to be improved and broken through. Summary of the invention

[0006] The present invention provides a causal relationship prediction method and system based on an intervention diffusion model, which significantly improves the accuracy and generalization of network reasoning.

[0007] The technical solution of the present invention is as follows:

[0008] A causal relationship prediction method based on an intervention diffusion model comprises the following steps:

[0009] (1) Calculate the correlation between the expression data of specific representations and construct an initial correlation network of the expression data based on the correlation;

[0010] (2) Divide the expression data of a specific representation into a training set and a test set;

[0011] (3) Use the training set data to train the diffusion model;

[0012] (4) Based on the initial correlation network, perform H0 and H1 hypotheses on each edge; H0 assumes that this edge represents the absence of causal relationship, and H1 assumes that this edge represents the existence of causal relationship; use the test set data to perform different intervention operations on the trained diffusion model to obtain the predicted value after intervention;

[0013] (5) Determine whether there is a causal relationship between the variables with edges in the initial correlation network based on the errors between the predicted values ​​and the true values ​​of the two hypotheses;

[0014] (6) According to steps (4)-(5), all variable pairs with edges in the initial correlation network are traversed to obtain the final causal network.

[0015] The expression data of specific phenomena include gene expression data, food chain data, and disease transmission model data. The expression data under normal circumstances is used as reference data for disease research. Such data mainly include gene expression data and whole genome association analysis data. Other data that meet the standards, such as food chain data and disease transmission model data, are also acceptable.

[0016] In step (1), the Pearson correlation coefficient is used to evaluate the correlation between two variables and construct an initial correlation network of the expression data.

[0017] The calculation formula of Pearson correlation coefficient is as follows:

[0018]

[0019] Where: X and Y are two variables; COV(X,Y) is the covariance between X and Y; σX and σY are the standard deviations of X and Y respectively.

[0020] Furthermore, if the PCC value is less than 0.3, it is considered that there is no correlation edge between the two variables, otherwise it is considered that there is a correlation edge between the two variables, thereby constructing an initial correlation network.

[0021] In step (2), the expression data of the specific representation is divided into a training set and a test set by a cross-validation method; the cross-validation method is K-fold cross-validation or leave-one-out method.

[0022] The training of the diffusion model is divided into the noise addition process and the denoising process. In the noise addition process, the model gradually adds noise to the data until the data is completely transformed into noise. This process can be represented as a series of state transitions, where the observation data D of each state t is t is obtained by observing the data D in the previous state t-1 The noise is added on the basis of

[0023] Noise adding process P(h X |D) is:

[0024]

[0025] Denoising process P(D|h X )for:

[0026]

[0027] In the formula, Indicates that in the process of adding noise, t-1 To D t The state transition probability of Indicates that in the process of denoising, t To D t-1 The state transition probability of ; p represents the probability function of the noise addition process in the diffusion model; q represents the probability function of the noise removal process in the diffusion model; θ ％ represents the parameters of the noise adding process; θ ( represents the parameters of the denoising process; D represents the observed data set, which contains the dependent variable X, the result variable Y and other variables Z; h X Represents the hidden layer corresponding to the dependent variable X; t represents the state during the noise addition and denoising process.

[0028] Step (4) includes: for one edge, in the trained diffusion model, performing a fixed intervention operation and a non-fixed intervention operation on the test set data, respectively, to obtain a predicted value of the fixed intervention operation and a predicted value of the non-fixed intervention operation;

[0029] The fixed intervention operation is to mask the dependent variable data of this edge in the test set data, and the non-fixed intervention operation is to not mask the dependent variable data of this edge in the test set data.

[0030] Step (5) includes: calculating the error ε1 between the predicted value and the true value of the fixed intervention operation, and the error ε2 between the predicted value and the true value of the non-fixed intervention operation; judging whether there is a causal relationship between the dependent variable and the process variable on this edge through the error ε1 and the error ε2.

[0031] In step (5), the error between the predicted value and the true value is the mean square error distance, which is calculated as:

[0032]

[0033] In the formula, Y represents the true value; represents the predicted value; n represents the number of samples of variable Y.

[0034] In step (5), when error ε1 < error ε2, it is considered that there is no causal effect between the two variables on this edge; otherwise, it is considered that there is a causal effect between the two variables on this edge.

[0035] The present invention also provides a causal relationship prediction system based on the intervention diffusion model, comprising:

[0036] An initial correlation network construction module calculates the correlation between the expression data of a specific representation and constructs an initial correlation network of the expression data based on the correlation;

[0037] The data segmentation module divides the expression data of a specific representation into a training set and a test set;

[0038] Diffusion model training module, using training set data to train the diffusion model;

[0039] The intervention module, based on the initial correlation network, performs H0 and H1 assumptions on each edge; the H0 assumption indicates that the causal relationship does not exist, and the H1 assumption indicates that the causal relationship exists; different intervention operations are performed on the trained diffusion model using the test set data to obtain the predicted value after the intervention;

[0040] The causal judgment module determines whether there is a causal relationship between the variables with edges in the initial correlation network based on the errors between the predicted values ​​and the true values ​​of the two hypotheses, and obtains the causal relationship network expressing the data;

[0041] The network generation module traverses all the variable pairs with connected edges in the initial correlation network according to the intervention module and the causal judgment module to obtain the final causal network.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention combines diffusion models and interventions for accurate causal inference, firmly rooted in intervention causal theory rather than just association or observational causal studies, and significantly improves the accuracy and generalization of network inferences even without additional information.

[0044] In addition, the method of the present invention enables manipulation of input variables by intervening in the hidden layer of the diffusion model. The method of the present invention marks a further integration of the field of biological network analysis and the field of causal inference, providing a more reliable and causal perspective for deciphering the complexity of molecular interactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the overall framework principle of the present invention. DETAILED DESCRIPTION

[0046] The embodiments of the present invention are described in detail below, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and cannot be used as a limitation of the present invention.

[0047] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the field to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0048] The numbers of the steps mentioned in each embodiment are only for the convenience of description and have no substantial sequence relationship. Different steps in each specific embodiment can be combined in different sequences to achieve the purpose of the invention.

[0049] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0050] Example:

[0051] like Figure 1 As shown, a causal relationship prediction method based on the intervention diffusion model specifically includes the following steps:

[0052] 101) Steps to construct the initial correlation network:

[0053] Using the expression data of specific representations as initial input data, the correlation coefficients between the initial input data are calculated and the initial correlation network is constructed using the correlation coefficients between the initial input data.

[0054] The expression data of specific phenomena include gene expression data, food chain data, and disease transmission model data. For example, expression data under normal conditions is used as reference data for disease research. Such data mainly include gene expression data and whole genome association analysis data. Other data that meet the standards, such as food chain data and disease transmission model data, are also acceptable.

[0055] Specifically, the Pearson correlation coefficient is used to evaluate the correlation between two variables to eliminate the indirect effects of genes, thereby obtaining the direct correlation between genes and constructing a correlation network.

[0056] The specific calculation formula of Pearson correlation coefficient PCC is as follows:

[0057]

[0058] Where: X and Y are two variables; COV(X,Y) is the covariance between X and Y; σX and σY are the standard deviations of X and Y respectively.

[0059] If the calculated PCC value is less than 0.3, it is considered that there is no correlation edge between the two variables. If the PCC value is greater than 0.3, it is considered that there is a correlation edge between the two variables, thereby constructing an initial correlation network.

[0060] 102) Steps to determine causal relationship:

[0061] Based on the initial correlation network, H0 hypothesis and H1 hypothesis are performed on each pair of edges; H0 edge represents the absence of causal relationship, and H1 edge represents the existence of causal relationship; the expression data is divided into training set and test set, and the diffusion model is trained with the training set data, and the intervention operation is performed with the test set data based on the trained diffusion model;

[0062] Among them, in the H0 hypothesis, other variables that do not include the dependent variable are used to regress the result variable; in the H1 hypothesis, the dependent variable and other variables are used to regress the result variable; the specific regression formulas are as follows:

[0063] H0:

[0064] H1:Y=f(X,Z ％ ,Z ( ,…,Z n$( )+ε

[0065] Among them, Y represents the outcome variable; X represents the dependent variable; Z (Z1, Z2, … Z n-2 ) are other variables that do not include X and Y variables; ε and For noise.

[0066] Specifically, the expression data of a specific representation needs to be divided into a training set and a test set. The training set is used as a regression training diffusion model, and the test set is used to judge the error. The division methods include K-fold cross validation and leave-one-out method.

[0067] The specific training of the diffusion model is divided into the noise adding process and the noise removing process. By performing the noise adding process and the noise removing operation on the input data, the generative relationship between the dependent variable and the result variable of the input data is obtained.

[0068] In the process of adding noise, the model gradually adds noise to the data until the data is completely transformed into noise. This process can be represented as a series of state transitions, where the observed data D of each state t is t is obtained by observing the data D in the previous state t-1 The noise is added on the basis of , and the noise adding process can be expressed as:

[0069]

[0070] in, Indicates that in the process of adding noise, t-1 To D t The state transition probability of , θ1 is the model parameter of the noise adding process.

[0071] In the denoising process, the model learns how to recover the original data from the noise. This process is the inverse process of the noise adding process. The denoising process can be expressed as:

[0072]

[0073] in, Indicates that in the process of denoising, t To D t-1 is the state transition probability, and θ2 is the model parameter of the noise adding process.

[0074] The specific generating relationships are as follows:

[0075] Noise adding process:

[0076]

[0077] De-noising process:

[0078]

[0079] Among them, p represents the probability function of the noise addition process in the diffusion model; q represents the probability function of the noise removal process in the diffusion model; θ ％ represents the parameters of the noise adding process; θ (represents the parameters of the denoising process; D represents the observed data set, which contains the dependent variable X, the result variable Y and other variables Z; h X Represents the hidden layer corresponding to the dependent variable X; t represents the state during the noise addition and denoising process.

[0080] Then, intervention operations can be performed on the test set data to evaluate the causal relationship between variables. By intervening in a certain variable, the causal relationship between the remaining two variables can be determined.

[0081] The specific generating relationships are as follows:

[0082]

[0083] Among them, Y represents the outcome variable; X represents the dependent variable; h X Represents the hidden layer corresponding to the dependent variable X.

[0084] 103) Steps to construct a causal network: Perform different intervention operations under the assumptions on both sides, and use the data from the test set to perform different intervention operations on the same diffusion model to obtain the results after the intervention; by comparing the size of the fitting error between the intervention result and the true result, it can be determined which hypothesis to accept, and whether the causal relationship exists, thereby constructing the entire causal network.

[0085] The specific judgment criteria are: after the intervention operation in the test set data set, if h is fixed X The error ratio between the estimated Y and the actual Y is not fixed. X If the error between the estimated Y and the actual Y is small, it is considered that there is no causal effect between variables X and Y. X The error ratio between the estimated Y and the actual Y is not fixed. X If the error between the estimated Y and the actual Y is large, it is believed that there is a causal effect between variables X and Y.

[0086] The specific error can be calculated by distance measurement means; the specific distance measurement means for comparison is mean square error distance, and its specific calculation formula is as follows:

[0087]

[0088] Among them, Y represents the true value; represents an estimated value; represents the distance between the true value and the estimated value; n represents the number of samples of variable Y. The specific process is as follows:

[0089] 1) Use cross-validation to divide the data set into training set and test set.

[0090] The cross-validation data division methods in this step include K-fold cross-validation and leave-one-out method.

[0091] 2) Based on the initial network, the training set data is used as input to train the diffusion model. The specific generation equations of the diffusion model are as follows:

[0092] Noise adding process:

[0093]

[0094] De-noising process:

[0095]

[0096] Among them, p represents the probability function of the noise addition process in the diffusion model; q represents the probability function of the noise removal process in the diffusion model; θ ％ represents the parameters of the noise adding process; θ ( represents the parameters of the denoising process; D represents the observed data set, which contains the dependent variable X, the result variable Y and other variables Z; h X represents the hidden layer corresponding to the dependent variable X; t represents the state in the process of adding noise and denoising; P(h X |D) represents the number of layers from D to hidden layer h X The overall noise adding process; P(D|h X ) represents the hidden layer h X The overall denoising process to D.

[0097] 3) Select the target variable pair and make two different hypotheses, H0 and H1.

[0098] In this step, H0 assumes that there is no causal effect between one variable (dependent variable) and the other variable (result variable).

[0099] In this step, H1 assumes that one variable (dependent variable) has a causal effect on another variable (result variable).

[0100] 4) In the trained diffusion model, the test set data is intervened. In the test set data, the fixed intervention operation is to mask the data of the dependent variable, and the non-fixed intervention operation is to not mask the data of the dependent variable. The specific relationship of the intervention operation is as follows:

[0101]

[0102] Among them, Y represents the outcome variable; X represents the dependent variable; h X represents the hidden layer corresponding to the dependent variable X; P represents the probability function; do represents the intervention formula.

[0103] 5) Under different assumptions, the estimated results obtained by intervention are compared with the actual results to obtain the error. By analyzing the error, it can be determined whether there is a causal relationship between one variable and another variable.

[0104] If h is fixed X The error ratio between the estimated Y and the actual Y is not fixed. X If the error between the estimated Y and the actual Y is small, it is considered that there is no causal effect between variables X and Y. X The error ratio between the estimated Y and the actual Y is not fixed. X If the error between the estimated Y and the actual Y is large, it is believed that there is a causal effect between variables X and Y.

[0105] The specific error can be calculated by distance measurement means; the specific distance measurement means for comparison is mean square error distance, and its specific calculation formula is as follows:

[0106]

[0107] Among them, Y represents the true value; represents the estimated value; n represents the number of samples of variable Y.

[0108] By repeating the above steps for the variable pairs that have edges in the initial network, all the causal relationships in the initial network can be obtained, that is, the final causal relationship network (target network) is obtained.

[0109] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0110] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.

[0111] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0112] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0113] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, each function may be stored on or transmitted by a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any medium that facilitates the transfer of a computer program from one place to another. Storage media may be any available medium that can be accessed by a computer. As an example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer. Any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, a server, or other remote source using a coaxial cable, a fiber optic cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, wherein disk often reproduces data magnetically, while disc reproduces data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0114] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A causal relationship prediction method based on an intervention diffusion model, characterized in that: The following steps are involved: (1) Calculate the correlation between the expression data of specific representations and construct an initial correlation network of the expression data based on the correlation; (2) Divide the expression data of a specific representation into a training set and a test set; (3) Use the training set data to train the diffusion model; (4) Based on the initial correlation network, perform H0 and H1 hypotheses on each edge; H0 assumes that this edge represents the absence of causal relationship, and H1 assumes that this edge represents the existence of causal relationship; use the test set data to perform different intervention operations on the trained diffusion model to obtain the predicted value after intervention; (5) Determine whether there is a causal relationship between the variables with edges in the initial correlation network based on the errors between the predicted values ​​and the true values ​​of the two hypotheses; (6) According to steps (4)-(5), all variable pairs with edges in the initial correlation network are traversed to obtain the final causal network.

2. The causal relationship prediction method based on the intervention diffusion model according to claim 1 is characterized in that: In step (1), the Pearson correlation coefficient is used to evaluate the correlation between two variables and construct an initial correlation network of the expression data.

3. The causal relationship prediction method based on the intervention diffusion model according to claim 2 is characterized in that: The calculation formula of Pearson correlation coefficient is as follows: Where: X and Y are two variables; COV(X,Y) is the covariance between X and Y; σX and σY are the standard deviations of X and Y respectively.

4. The causal relationship prediction method based on the intervention diffusion model according to claim 3 is characterized in that: If the PCC value is less than 0.3, it is considered that there is no correlation edge between the two variables, otherwise it is considered that there is a correlation edge between the two variables, thereby constructing an initial correlation network.

5. The causal relationship prediction method based on the intervention diffusion model according to claim 1 is characterized in that: In step (2), the expression data of the specific representation is divided into a training set and a test set by a cross-validation method; the cross-validation method is K-fold cross-validation or leave-one-out method.

6. The causal relationship prediction method based on the intervention diffusion model according to claim 1 is characterized in that: The training of the diffusion model is divided into the noise addition process and the denoising process; the noise addition process P(h X |D) is: Denoising process P(D|h X )for: In the formula, Indicates that in the process of adding noise, t-1 To D t The state transition probability of Indicates that in the process of denoising, t To D t-1 The state transition probability of ; p represents the probability function of the noise addition process in the diffusion model; q represents the probability function of the noise removal process in the diffusion model; θ ％ represents the parameters of the noise adding process; θ ( represents the parameters of the denoising process; D represents the observed data set, which contains the dependent variable X, the result variable Y and other variables Z; h X Represents the hidden layer corresponding to the dependent variable X; t represents the state during the noise addition and denoising process.

7. The causal relationship prediction method based on the intervention diffusion model according to claim 1 is characterized in that: Step (4) includes: for one edge, in the trained diffusion model, performing a fixed intervention operation and a non-fixed intervention operation on the test set data, respectively, to obtain a predicted value of the fixed intervention operation and a predicted value of the non-fixed intervention operation; The fixed intervention operation is to mask the dependent variable data of this edge in the test set data, and the non-fixed intervention operation is to not mask the dependent variable data of this edge in the test set data.

8. The causal relationship prediction method based on the intervention diffusion model according to claim 1 is characterized in that: Step (5) includes: calculating the error ε1 between the predicted value and the true value of the fixed intervention operation, and the error ε2 between the predicted value and the true value of the non-fixed intervention operation; judging whether there is a causal relationship between the dependent variable and the process variable on this edge through the error ε1 and the error ε2.

9. The causal relationship prediction method based on the intervention diffusion model according to claim 8 is characterized in that: When the error ε1 is less than the error ε2, it is considered that there is no causal effect between the two variables on this edge; otherwise, it is considered that there is a causal effect between the two variables on this edge.

10. A causal relationship prediction system based on an intervention diffusion model, characterized in that: include: An initial correlation network construction module calculates the correlation between the expression data of a specific representation and constructs an initial correlation network of the expression data based on the correlation; The data segmentation module divides the expression data of a specific representation into a training set and a test set; Diffusion model training module, using training set data to train the diffusion model; The intervention module, based on the initial correlation network, performs H0 and H1 assumptions on each edge; the H0 assumption indicates that the causal relationship does not exist, and the H1 assumption indicates that the causal relationship exists; different intervention operations are performed on the trained diffusion model using the test set data to obtain the predicted value after the intervention; The causal judgment module determines whether there is a causal relationship between the variables with edges in the initial correlation network based on the errors between the predicted values ​​and the true values ​​of the two hypotheses, and obtains the causal relationship network expressing the data; The network generation module traverses all the variable pairs with connected edges in the initial correlation network according to the intervention module and the causal judgment module to obtain the final causal network.

Images