Evolutionary methods and systems of brain region-gene topology in generative adversarial networks

Abstract

This invention relates to the field of artificial intelligence technology, and discloses a method and system for the evolution of brain region-gene topology in generative adversarial networks (GANs) to assist in achieving goals such as disease prediction and intervention feature extraction. The method essentially optimizes the brain region-gene associations of the generator through dynamic separation, edge completion, and cross-structural hole reconnection. Specifically, it includes: finding the edge with the highest topological potential each time, using it as the root edge to separate the tetrahedral structural hole with the highest importance, and completing all compensation edges between nodes adjacent to this hole in the remaining network, repeating until there are no structural holes. Then, at least one edge of each structural hole is deleted according to importance, the correlation of deleted edge pairs of different structural holes is calculated, and pairs exceeding a threshold are retained; then, combinations with node correlation exceeding a second threshold in each pair are selected to generate new edges, and all cross-hole combinations are traversed and fused into a new network. If the termination condition is not met, the iteration continues; otherwise, the fused network is output to the discriminator.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for the evolution of brain region-gene topology in generative adversarial networks. Background Technology

[0002] In existing brain region-gene networks, the applicant's prior patent CN121528300B discloses a "Brain Region-Gene Network Evolution Method and System Based on Generative Adversarial Networks." During the process of updating structural information based on a learning matrix, the patent discloses how the simple complex of a supernode dynamically constructs, merges, and diffuses its structural information in a single iteration. It also discloses the inheritance and renewal relationships between adjacent iterations, perfectly simulating the disease progression of Alzheimer's disease and other intractable diseases, ensuring the reliability of the evolution results. This provides new technical guidance for the research, prediction, and prevention of Alzheimer's disease and other intractable diseases.

[0003] In the existing generative adversarial network, the concept of volume is not introduced, and the topological relationships within a closed star-shaped neighborhood are treated as stable supernodes in a single iteration. However, many other possibilities exist in the evolution of medical records, requiring further cross-sectional research into different evolutionary pathways to provide diverse technical solutions for goals such as disease prediction and intervention feature extraction. Summary of the Invention

[0004] The purpose of this invention is to disclose a method and system for the evolution of brain region-gene topology in generative adversarial networks, so as to assist in the achievement of goals such as disease prediction and intervention feature extraction.

[0005] To achieve the above objectives, the method for the evolution of brain region-gene topology in generative adversarial networks disclosed in this invention includes:

[0006] Step S1: Calculate the edge with the highest topological potential in the brain region-gene topology after the initial or previous iteration, determine the structural hole with the highest importance of the edge, and then separate the structural hole from the current brain region-gene topology; the structural hole is a tetrahedron, and the edge with the highest topological potential is the root edge; Step S2: After completing the compensation edges connecting each pair of nodes adjacent to the hole in the remaining brain region-gene topology, update it to a new brain region-gene topology. Step S3: Determine whether there are still structural holes in the updated brain region-gene topology. If yes, return to step S1; otherwise, proceed to step S4. Step S4: Delete at least one edge of each structural hole according to the preset importance sorting algorithm, and calculate the correlation coefficient between the two deleted edges of different structural hole combinations. Select the pair with the correlation coefficient greater than the third threshold, and then calculate the correlation coefficient between any node of any deleted edge and any node of the other deleted edge in the pair. Select the pair nodes with the correlation coefficient greater than the fourth threshold to generate new edges. After traversing all combinations of different structural holes based on the two deleted edges, merge them into a new brain region-gene topology. Step S5: Determine whether the number of iterations meets the iteration termination condition. If not, return to step S1; otherwise, terminate the iteration and send the brain region-gene topology structure fused by the generator to the discriminator.

[0007] Preferably, step S4 specifically includes: Step S41: Construct the edge information tree structure corresponding to each structural hole. The edge information tree structure takes the adjacent edge of the root edge as the child edge and the opposite edge as the leaf edge, and establishes the propagation connection between each child edge and the root edge, leaf edge and other child edges of the same level. Step S42: Calculate the association weight of each propagation connection based on the adjacency of the two adjacent surfaces of each propagation connection before the separation structure hole, and compare the calculated association weight with the first threshold. Keep the propagation connections that are greater than or equal to the first threshold, and prune the propagation connections that are less than the first threshold. Step S43: For each edge information tree structure after pruning, update the entropy information of the edges according to the association weights of the corresponding propagation connections. Then, perform operations on the updated entropy information matrix and the learnable parameter matrix. Compare the entropy information updated twice after the edge operation with the second threshold. Keep the edges that are greater than or equal to the second threshold and delete the edges that are less than the second threshold. If there are no edges less than the second threshold, delete the edge with the smallest entropy information. The learnable parameter matrix is ​​inconsistent for different iteration numbers, and the learnable parameter matrix is ​​the same for each edge information tree structure in the same iteration number.

[0008] Preferably, the formula for calculating the association weight of each propagation connection is: the adjacency degree of two coplanar edges divided by the sum of the adjacency degrees of the other surfaces adjacent to the two edges.

[0009] Preferably, the specific process of updating the entropy information of the edge according to the association weight of the corresponding propagation connection is as follows: the lower layer edge multiplies the entropy information of the upper layer edge with the association weight of the corresponding propagation connection, then adds it to its original entropy information, and then takes the average value.

[0010] Preferably, the calculation process of topological potential energy includes: Step S11: Based on the initial or previous iteration's brain region-gene topology adjacency matrix Get the adjacency degree of each element. The specific calculation formula is as follows: ; in, It is a matrix of all ones. The identity matrix consists of elements including points, edges, triangle faces, and volumes conforming to Euler's formula. The number of elements, indicated by the superscript. Indicates the current iteration number. Represents the multiplication operator. It represents the Hadamardi (or Hadama) stack; Step S12: Calculate the entropy information matrix The specific calculation formula is as follows: ; ; in, Adjacency matrix The inverse matrix, It represents the Hadamah accumulation. For the intermediate matrix calculated, This indicates the logarithmic operation; Step S13: For the topological potential matrix of the connected edges First, start with the adjacency matrix Separate the edge-edge adjacency matrix ,against Perform a breadth-first search on each row to obtain the edge distance coefficient matrix. , Represented as: ; in, Represents the connection To the border The shortest number of hops between them; This represents the current number of edges; Step S14: Calculate the topological potential matrix of the connected edges. The calculation formula is: ; in, This is the matrix exponentiation function. This is a learnable parameter matrix used to control the influence range of each edge. The edge weight matrix represents the correlation coefficient between the two connected nodes based on the feature sequence.

[0011] Preferably, the importance of the structural hole is the reciprocal of the similarity between the root edge and the neighborhood of the opposite edge.

[0012] Preferably, the method of the present invention further includes: Step S6: Update the entropy information matrix of the fused brain region-gene topology based on convolution operation, then flatten the updated entropy information and input it into the fully connected network. Real-world discrimination and risk prediction of the input brain region-gene topology are achieved through multi-layer linear transformation and nonlinear mapping; and the model parameters of the generative adversarial network are jointly optimized based on the discrimination results.

[0013] Preferably, the method of the present invention further includes: Step S7: For the information entropy matrix of all samples during the evolution process, calculate the change in information entropy of each structural element during the evolution, and sort each element according to the increment to obtain the candidate feature set. Step S8: Using an incremental search strategy, select element subsets of different sizes from high to low according to the change in information entropy, construct multiple sets of candidate feature combinations, and evaluate the prediction performance of each candidate feature set based on the generative adversarial network; select the candidate feature set with the best prediction performance.

[0014] Preferably, the first threshold, the second threshold, the third threshold, and the fourth threshold of the present invention take the same value in different iteration numbers.

[0015] To achieve the above objectives, the present invention also discloses an evolution system of brain region-gene topology in generative adversarial networks, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-described method when executing the computer program.

[0016] In summary, the essence of this invention is to optimize the brain region-gene association of the generator through dynamic separation, edge completion, and cross-structural hole reconnection. Specifically, it includes: finding the edge with the highest topological potential each time, using it as the root edge to separate the tetrahedral structural hole with the highest importance, and completing all compensation edges between the nodes adjacent to the hole in the remaining network, repeating until there is no structural hole. Then, at least one edge of each structural hole is deleted according to importance, the correlation of the deleted edge pairs of different structural holes is calculated, and the pairs exceeding the threshold are retained; then, combinations with the correlation between nodes in each pair exceeding the second threshold are selected to generate new edges, and all cross-hole combinations are traversed and fused into a new network. If the termination condition is not met, the iteration continues; otherwise, the fused network is output to the discriminator. Its technical implementation path is completely different from the existing approach that treats some star-shaped closed loop domains as supernodes. Structural holes represent the "weak parts" in the system that are most prone to evolution. The iterative process tends to simplify the network topology, effectively solving the complexity problem of information transmission in complex structures, and thus providing new ideas for achieving goals such as assisting in disease prediction and intervention feature extraction.

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0018] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the evolution method of brain region-gene topology in generative adversarial networks disclosed in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the structural hole decomposition process disclosed in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram illustrating the principle of propagation, connection, pruning, and confirmation of edge deletion within the structural hole disclosed in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram illustrating the fusion principle between different structural holes disclosed in an embodiment of the present invention. Detailed Implementation

[0022] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0023] Example 1 This embodiment discloses a method for the evolution of brain region-gene topology in generative adversarial networks, such as... Figure 1 As shown, it includes: Step S1: Calculate the edge with the highest topological potential in the brain region-gene topology after the initial or previous iteration, determine the structural hole with the highest importance of the edge, and then separate the structural hole from the current brain region-gene topology; the structural hole is a tetrahedron, and the edge with the highest topological potential is the root edge.

[0024] In this step, the calculation of the topological potential energy includes: Step S11: Based on the initial or previous iteration's brain region-gene topology adjacency matrix Get the adjacency degree of each element. The specific calculation formula is as follows: ; in, It is a matrix of all ones. The identity matrix consists of elements including points, edges, triangle faces, and volumes conforming to Euler's formula. The number of elements, indicated by the superscript. Indicates the current iteration number. Represents the multiplication operator. It represents the Hadamardi (or Hadama) stack.

[0025] Step S12: Calculate the entropy information matrix The specific calculation formula is as follows: ; ; in, Adjacency matrix The inverse matrix, It represents the Hadamah accumulation. For the intermediate matrix calculated, This represents the logarithmic operation. In this embodiment, The weighted sum of the elements in each row represents the probability of a complex being chosen by its adjacent complexes; this is the entropy information matrix. The element values ​​are used to represent the structural information possessed by the elements in the complex.

[0026] Step S13: For the topological potential matrix of the connected edges First, start with the adjacency matrix Separate the edge-edge adjacency matrix ,against Perform a breadth-first search on each row to obtain the edge distance coefficient matrix. , Represented as: ; in, Represents the connection To the border The shortest number of hops between them; This represents the current number of edges.

[0027] Step S14: Calculate the topological potential matrix of the connected edges. The calculation formula is: ; in, This is the matrix exponentiation function. This is a learnable parameter matrix used to control the influence range of each edge. The edge weight matrix represents the correlation coefficient between the two connected nodes based on the feature sequence.

[0028] In this embodiment, a point-edge adjacency matrix can be used. Edge-face adjacency matrix and the adjacency matrix of surfaces This is used to represent the hierarchical structure in the brain region-gene topology, specifically recording the adjacency relationships between nodes and edges, edges and planes, and planes and polyhedra. A point-to-point adjacency matrix is ​​employed. Edge-edge adjacency matrix Face-to-face adjacency matrix and volume-to-volume adjacency matrix This matrix is ​​used to record the adjacency relationships between structures at the same level. Points, edges, faces, and volumes are all simple complexes (the concept of a simple complex can be found in many existing patent applications published by the applicant in this case, and will not be elaborated here). If a complex is a subset of another complex, or if there is an intersection between complexes of the same dimension, it is considered to have an adjacency relationship and is assigned a value of 1; otherwise, it is assigned a value of 0. Based on the definition of the above matrix, the general formula for constructing the adjacency matrix in different iteration processes is as follows: ; in, , This represents the sum of the number of elements composed of points, edges, faces, and volumes during the corresponding iteration process.

[0029] The process of decomposing the structural hole (which can also be described as the sub-complex in the diagram) in this step can be referred to... Figure 2 In this context, the nodes represented by circles are either brain regions or genes. The conditions for a structural hole to occur are: 1. (i.e., root edge) and opposite side 1. There is no direct connection between them; 2. (i.e., root edge) and opposite side (No direct connection exists, but an indirect connection can be established through plane x); 3. Root edge and opposite side There are bridging edges between them. (Corresponding) Figure 2 In the diagram, the root edge is the connection between nodes 1 and 2, the opposite edge is the connection between nodes 3 and 5, and the adjacent edge is the connection between nodes 2 and 3. During the separation process, all elements related to the structural hole will "collapse" to simulate the functional connection damage after the loss of critical functions in the course of a disease.

[0030] The importance of the structural hole is determined by the reciprocal of the similarity between the root edge and the neighborhood of its opposite edge, thus achieving... Figure 2 The first structural hole to be decomposed is the tetrahedron formed by nodes 1, 2, 3, and 5, not the tetrahedron formed by nodes 1, 2, 7, and 8, nor the tetrahedron formed by nodes 1, 2, 5, and 6. Generally, the less overlap there is between the root edge and the neighboring edges of the opposite edge, the less redundancy there is in the structure on both sides of the structural hole, and the more important the structural hole is.

[0031] Step S2: After completing the compensation edges connecting each pair of nodes adjacent to the hole in the remaining brain region-gene topology, update it to a new brain region-gene topology.

[0032] See Figure 2 The compensation edge in this step corresponds to the connection between nodes 4 and 6 and the connection between nodes 4 and 7.

[0033] Step S3: Determine whether there are still structural holes in the updated brain region-gene topology. If yes, return to step S1; otherwise, proceed to step S4.

[0034] exist Figure 2 In the process, three structural holes were separated through two rounds of iteration.

[0035] Step S4: Delete at least one edge of each structural hole according to the preset importance sorting algorithm, and calculate the correlation coefficient between the two deleted edges of different structural hole combinations. Select the pair with the correlation coefficient greater than the third threshold, and then calculate the correlation coefficient between any node of any deleted edge and any node of the other deleted edge in the pair. Select the pair nodes with the correlation coefficient greater than the fourth threshold to generate new edges. After traversing all combinations of different structural holes based on the two deleted edges, merge them into a new brain region-gene topology.

[0036] The specific process of this step may include: Step S41: Construct the edge information tree structure corresponding to each structural hole. The edge information tree structure takes the adjacent edge of the root edge as the child edge and the opposite edge as the leaf edge, and establishes the propagation connection between each child edge and the root edge, leaf edge and other adjacent child edges of the same level.

[0037] Reference Figure 3 The root edge of this step is 1-2, and the leaf edges include: 1-3, 1-4, 2-3, 2-4, and the leaf edge is 3-4.

[0038] Step S42: Calculate the association weight of each propagation connection based on the adjacency of the two adjacent surfaces of each propagation connection before the separation structure hole, and compare the calculated association weight with the first threshold. Retain propagation connections that are greater than or equal to the first threshold, and prune propagation connections that are less than the first threshold.

[0039] Optionally, the formula for calculating the association weight of each propagation connection in this step is: the adjacency degree of two coplanar edges divided by the sum of the adjacency degrees of the other surfaces adjacent to the two edges. For example... Figure 3 As shown, the pruning order is from top to bottom. The pruned propagation connections include the propagation connections between edges 1-2 and 1-4 in the first round, and between edges 1-2 and 2-4; and the propagation connection between edges 1-3 and 3-4 in the second round. The result of the first round of pruning directly causes edges 1-4 and 2-4 to be downgraded to leaf nodes. To simplify the design, this invention does not consider the propagation connections between leaf nodes. Therefore, after edges 1-4 and 2-4 are downgraded to leaf nodes, their original propagation connections no longer exist. In addition, there is no connection between edges 1-3 and 2-4, so there is no propagation connection between them.

[0040] Step S43: For each edge information tree structure after pruning, update the entropy information of the edges according to the association weights of the corresponding propagation connections. Then, perform operations on the updated entropy information matrix and the learnable parameter matrix. Compare the entropy information updated twice after the edge operation with the second threshold. Keep the edges that are greater than or equal to the second threshold and delete the edges that are less than the second threshold. If there are no edges less than the second threshold, delete the edge with the smallest entropy information. The learnable parameter matrix is ​​inconsistent for different iteration numbers, and the learnable parameter matrix is ​​the same for each edge information tree structure in the same iteration number.

[0041] In this step, the learnable parameter matrix is ​​used to learn the magnitude of change in edge weights; the specific calculation can be performed using multiplication between two matrices. Figure 3 The deleted edges include 2-3 connected edges and 3-4 connected edges. Because the deleted edges create gaps in the structural openings, they are also called edge gaps.

[0042] Optionally, the specific process of updating the entropy information of the edge according to the association weight of the corresponding propagation connection is as follows: the lower-level edge multiplies the entropy information of the upper-level edge with the association weight of the corresponding propagation connection, then adds it to its own original entropy information, and finally divides the sum by 2 to take the average value.

[0043] In this step, the entropy information is only used to determine the operation of deleting edges, and the relevant information is not inherited in the next iteration. In the next iteration, the entropy information is determined again in step S12 based on the new brain region-gene topology after subsequent fusion.

[0044] In the above steps, the first threshold, the second threshold, the third threshold, and the fourth threshold all take the same value in different iteration numbers.

[0045] In step S4, the correlation coefficient between the two deleted sides is calculated using the Pearson correlation coefficient between the respective feature sequences of the two sides. The sequence features of the sides can be obtained by taking the average of the feature sequences of the two nodes in a conventional algorithm. The correlation coefficient between subsequent nodes is also obtained using the same algorithm. It is similar to the logic of determining the existence of sides based on correlation in the existing construction of the initial brain region-gene topology network before evolution, and will not be elaborated further.

[0046] For details on the fusion process, please refer to... Figure 4 The edge gaps before merging include: 10-11 connected edge, 1-2 connected edge, 2-5 connected edge, and 4-6 connected edge, while the new edges generated by merging include: 10-1 connected edge, 11-2 connected edge, 2-6 connected edge, and 2-5 connected edge.

[0047] Step S5: Determine whether the number of iterations meets the iteration termination condition. If not, return to step S1; otherwise, terminate the iteration and send the brain region-gene topology structure fused by the generator to the discriminator.

[0048] Furthermore, the method of this embodiment may further include: Step S6: Update the entropy information matrix of the fused brain region-gene topology based on convolution operation, then flatten the updated entropy information and input it into the fully connected network. Real-world discrimination and risk prediction of the input brain region-gene topology are achieved through multi-layer linear transformation and nonlinear mapping; and the model parameters of the generative adversarial network are jointly optimized based on the discrimination results.

[0049] Step S7: For the information entropy matrix of all samples during the evolution process, calculate the change in information entropy of each structural element during the evolution, and sort each element according to the increment to obtain the candidate feature set.

[0050] Step S8: Using an incremental search strategy, select element subsets of different sizes from high to low according to the change in information entropy, construct multiple sets of candidate feature combinations, and evaluate the prediction performance of each candidate feature set based on the generative adversarial network; select the candidate feature set with the best prediction performance.

[0051] Steps S6 to S8 have already been implemented in the prior art, and will not be elaborated upon here.

[0052] Example 2 This embodiment discloses an evolution system of brain region-gene topology in generative adversarial networks, 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 a series of steps of the method in Embodiment 1 above.

[0053] In summary, the brain region-gene topology evolution method and system disclosed in the above embodiments of the present invention in generative adversarial networks essentially optimizes the brain region-gene association of the generator through dynamic separation, edge completion, and cross-structural hole reconnection. Specifically, it includes: finding the edge with the highest topological potential each time, separating the tetrahedral structural hole with the highest importance using it as the root edge, and completing all compensation edges between the nodes adjacent to the hole in the remaining network, repeating until there is no structural hole. Then, at least one edge of each structural hole is deleted according to importance, the correlation of the deleted edge pairs of different structural holes is calculated, and the pairs exceeding the threshold are retained; then, combinations with the correlation between nodes in each pair exceeding the second threshold are selected to generate new edges, and after traversing all cross-hole combinations, they are merged into a new network. If the termination condition is not met, the iteration continues; otherwise, the fusion network is output to the discriminator. Its technical implementation path is completely different from the existing approach that treats some star-shaped closed loop domains as supernodes. The structural hole represents the "weak part" in the system that is most likely to evolve. The iterative process tends to simplify the network topology, effectively solving the problem of information transmission complexity in complex structures, and thus providing new ideas for achieving goals such as assisting in disease prediction and intervention feature extraction.

[0054] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for the evolution of brain region-gene topology in generative adversarial networks, characterized in that, include: Step S1: Calculate the edge with the highest topological potential in the brain region-gene topology after the initial or previous iteration, determine the structural hole with the highest importance of the edge, and then separate the structural hole from the current brain region-gene topology; the structural hole is a tetrahedron, and the edge with the highest topological potential is the root edge; Step S2: After completing the compensation edges connecting each pair of nodes adjacent to the hole in the remaining brain region-gene topology, update it to a new brain region-gene topology. Step S3: Determine whether there are still structural holes in the updated brain region-gene topology. If yes, return to step S1; otherwise, proceed to step S4. Step S4: Delete at least one edge of each structural hole according to the preset importance sorting algorithm, and calculate the correlation coefficient between the two deleted edges of different structural hole combinations. Select the pair with the correlation coefficient greater than the third threshold, and then calculate the correlation coefficient between any node of any deleted edge and any node of the other deleted edge in the pair. Select the pair nodes with the correlation coefficient greater than the fourth threshold to generate new edges. After traversing all combinations of different structural holes based on the two deleted edges, merge them into a new brain region-gene topology. Step S5: Determine whether the number of iterations meets the iteration termination condition. If not, return to step S1. Otherwise, terminate the iteration and send the brain region-gene topology structure fused by the generator to the discriminator.

2. The method for the evolution of brain region-gene topology in generative adversarial networks according to claim 1, characterized in that, Step S4 specifically includes: Step S41: Construct the edge information tree structure corresponding to each structural hole. The edge information tree structure takes the adjacent edge of the root edge as the child edge and the opposite edge as the leaf edge, and establishes the propagation connection between each child edge and the root edge, leaf edge and other adjacent child edges of the same level. Step S42: Calculate the association weight of each propagation connection based on the adjacency of the two adjacent surfaces of each propagation connection before the separation structure hole, and compare the calculated association weight with the first threshold. Keep the propagation connections that are greater than or equal to the first threshold, and prune the propagation connections that are less than the first threshold. Step S43: For each edge information tree structure after pruning, update the entropy information of the edges according to the association weights of the corresponding propagation connections. Then, perform operations on the updated entropy information matrix and the learnable parameter matrix. Compare the entropy information updated twice after the edge operation with the second threshold. Keep the edges that are greater than or equal to the second threshold and delete the edges that are less than the second threshold. If there are no edges less than the second threshold, delete the edge with the smallest entropy information. The learnable parameter matrix is ​​inconsistent for different iteration numbers, and the learnable parameter matrix is ​​the same for each edge information tree structure in the same iteration number.

3. The method for the evolution of brain region-gene topology in generative adversarial networks according to claim 2, characterized in that, The formula for calculating the association weight of each propagation connection is: the adjacency degree of two coplanar edges divided by the sum of the adjacency degrees of the other surfaces adjacent to the two edges.

4. The method for the evolution of brain region-gene topology in generative adversarial networks according to claim 3, characterized in that, The specific process of updating the entropy information of an edge according to the association weight of the corresponding propagation connection is as follows: the lower-level edge multiplies the entropy information of the upper-level edge with the association weight of the corresponding propagation connection, then adds it to its original entropy information, and finally takes the average value.

5. The method for the evolution of brain region-gene topology in generative adversarial networks according to any one of claims 1 to 4, characterized in that, The calculation process for topological potential energy includes: Step S11: Based on the initial or previous iteration's brain region-gene topology adjacency matrix Get the adjacency degree of each element. The specific calculation formula is as follows: ； in, It is a matrix of all ones. The identity matrix consists of elements including points, edges, triangle faces, and volumes conforming to Euler's formula. The number of elements, indicated by the superscript. Indicates the current iteration number. Represents the multiplication operator. It represents the Hadamardi (or Hadama) stack; Step S12: Calculate the entropy information matrix The specific calculation formula is as follows: ； ； in, Adjacency matrix The inverse matrix, It represents the Hadamah accumulation. For the intermediate matrix calculated, This indicates the logarithmic operation; Step S13: For the topological potential matrix of the connected edges First, start with the adjacency matrix Separate the edge-edge adjacency matrix ,against Perform a breadth-first search on each row to obtain the edge distance coefficient matrix. , Represented as: ； in, Represents the connection To the border The shortest number of hops between them; This represents the current number of edges; Step S14: Calculate the topological potential matrix of the connected edges. The calculation formula is: ； in, This is the matrix exponentiation function. This is a learnable parameter matrix used to control the influence range of each edge. The edge weight matrix represents the correlation coefficient between the two connected nodes based on the feature sequence.

6. The method for the evolution of brain region-gene topology in generative adversarial networks according to claim 5, characterized in that, The importance of a structural hole is the reciprocal of the similarity between the root edge and the neighborhood of its opposite edge.

7. The method for the evolution of brain region-gene topology in generative adversarial networks according to claim 6, characterized in that, Also includes: Step S6: Update the entropy information matrix of the fused brain region-gene topology based on convolution operation, and then flatten the updated entropy information before inputting it into a fully connected network. Through multi-layer linear transformation and nonlinear mapping, the network can realize the true and false judgment and risk prediction of the input brain region-gene topology. The model parameters of the generative adversarial network are jointly optimized based on the discrimination results.

8. The method for the evolution of brain region-gene topology in generative adversarial networks according to claim 7, characterized in that, Also includes: Step S7: For the information entropy matrix of all samples during the evolution process, calculate the change in information entropy of each structural element during the evolution, and sort each element according to the increment to obtain the candidate feature set. Step S8: Using an incremental search strategy, select element subsets of different sizes from high to low according to the change in information entropy, construct multiple sets of candidate feature combinations, and evaluate the prediction performance of each candidate feature set based on the generative adversarial network; select the candidate feature set with the best prediction performance.

9. The method for the evolution of brain region-gene topology in generative adversarial networks according to claim 2, characterized in that, The first threshold, the second threshold, the third threshold, and the fourth threshold all take the same value in different iteration numbers.

10. An evolutionary system of brain region-gene topology in generative adversarial networks, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method described in any one of claims 1 to 9.

Images