Method and device for mining cancer evolution information
Through methods based on rumor dissemination algorithms and gravitational models, gene mutation information is spread and high-gravity networks are built, which solves the problem that existing technology is difficult to explore key information on cancer evolution, and achieves clear description and research support for cancer evolution.
Patent Information
- Application Number
- CN202411102097.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art is difficult to effectively explore and characterize key information on the evolution and pathogenic mechanism of cancer from a macro perspective, and it is impossible to fully describe the driving effect of interactions containing driving effects in the evolution of cancer.
Using a network smoothing method based on rumor propagation algorithm, the mutation information of a single gene node is spread to the functional interaction network, a high-gravity network is built, and an attractive and repelling gravity is introduced, the comprehensive gravity of gene nodes is determined, and the cancer evolution process is demonstrated through a hierarchical visualization model. driver genes and transition factors in.
It has achieved a clear, effective and intuitive description of cancer evolution, provided strong support for cancer research and treatment, and can clearly identify evolutionary factors and transition factors, and deeply reveal key information in the process of cancer evolution.
Smart Images

Figure CN119008031A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of big data analysis and data mining, bioinformatics, and systems biology, and in particular to a method and device for mining cancer evolution information. Background Art
[0002] Cancer, also known as malignant tumor, is a complex disease caused by abnormal proliferation of cells in the body due to loss of normal regulation. Currently, cancer is one of the most difficult complex diseases to overcome in modern humans, and its pathogenic mechanism and disease evolution are attracting more and more attention.
[0003] At present, the main focus of the evolution and pathogenic mechanism of cancer is on cancer-related driver genes. Driver genes are a type of biological molecules that can play a driving role in the process of cancer evolution. They play a very important role in the process of cancer evolution and have become a key biological information for studying disease evolution. At present, most studies on driver genes focus more on single gene mutations as driver genes, that is, genes that can play a role in mutation in the accumulated mutations of cancer cells are called "driver genes". At present, the identification of driver genes can be divided into two major categories. Early studies were mainly based on mutation data processing and identification based on mutation frequency. For example, by performing data statistics and calculations on statistically related indicators to obtain corresponding statistics, and obtaining driver genes through statistical analysis. Later, due to the enrichment of data dimensions, methods based on biological networks gradually became the mainstream method, such as using random walks in biological networks and identifying driver genes at convergence, and constructing gene-related features based on network structure and combining them with machine learning for identification. Among many network-based methods, network smoothing has become a relatively mature technology for transferring the effects of single gene mutations to neighboring nodes, and then combining the network to identify driver genes. For example, the first step in the NBS method is to apply the random walk algorithm for network smoothing. However, these methods currently only focus on transferring single-point mutation information to first-order neighbor nodes, rather than from a global perspective, transferring the impact of this single-point mutation to the global network according to the network topology.
[0004] In addition to the driver genes mentioned above, interactions with driver effects have gradually become a driver gene that has attracted much attention. They contain two interacting gene nodes. This interaction can describe the driving effect of driver genes in the evolutionary process more comprehensively and in detail than the single gene mentioned above. With interactions with driver effects as the main identification target, there are also many means to measure and distinguish the importance of interactions. However, the interactions described by these methods are limited and cannot provide more information about interactions with driver effects.
[0005] Starting with driver genes is to find a fulcrum to describe the evolution of cancer stages. Cancer usually consists of four stages. In the evolution of these four stages, how cancer or tumors gradually change from a precancerous state to a final lethal state is a process quantity. By mining the stage-related driver genes of each stage, we can find the key biological molecules that represent the development of that stage, which will help people understand cancer, intervene in cancer, and treat cancer.
[0006] However, the occurrence and development of cancer is staged. Currently, most methods from the perspective of bioinformatics treat cancer samples as a whole. There is a lack of a macro and intuitive perspective to observe cancer evolution, divide cancer into stages, and explore and characterize key information on the evolution between cancer stages and the pathogenic mechanisms during the evolution process. Summary of the invention
[0007] The purpose of the present invention is to provide a method and device for mining cancer evolution information to clearly, effectively and intuitively describe cancer evolution.
[0008] In order to achieve the above tasks, the present invention adopts the following technical solutions:
[0009] A method for mining cancer evolution information, comprising:
[0010] Based on the stage-specific functional interaction network of cancer stages, a network smoothing method based on a rumor propagation algorithm is used to spread the influence of mutation information of a single gene node to the entire functional interaction network, which is used to smooth and redistribute the mutation frequency of gene nodes;
[0011] Based on the smoothed functional interaction network, gravity modeling is performed on the interactions between gene node pairs in the network; on the basis of gravity modeling, based on the gravity between gene node pairs, a high gravity network is constructed by extracting high gravity interactions, and at the same time, attractive gravity and repulsive gravity are introduced to integrate the expression relationship into the gravity model; the comprehensive gravity of the gene node is determined and used as the weight of the gene node;
[0012] The stage-specific functional interaction networks of each cancer evolution stage are connected together to construct a visualization model of cancer stage evolution displayed in a hierarchical form, which can be used to study cancer evolution, including determining evolutionary factors and transition factors.
[0013] Furthermore, the stage-specific functional interaction network is established for the stage of cancer evolution. The nodes of the network are gene nodes, the weights on the gene nodes are the number of mutations of the gene nodes, and the weights are propagated as rumors in the network. The edges in the network are the interactions between gene nodes.
[0014] Furthermore, the network smoothing method based on the rumor propagation algorithm is expressed as:
[0015]
[0016] Where n represents the total number of gene nodes in the network, u represents the gene node that receives the rumor, vi represents the remaining gene nodes, there are n-1 gene nodes that spread the rumor, and all of them spread the mutation information to node u, t0 represents the initial time point before the rumor spread, and t1 represents the time point when the rumor spread is completed; P vi→u represents the mutation information propagated by each rumor-spreading gene node vi to gene node u, P vi (t0) represents the initial mutation frequency of each rumor-spreading gene node before the rumor spread; k vi Indicates that gene node vi is the kth-order neighbor of gene node u, d vi Represents the degree of gene node vi in the network. The mathematical symbol ! is the factorial symbol; P u (t0) represents the initial mutation frequency of gene node u, and P u (t1) represents the mutation frequency of gene node u after the rumor is spread.
[0017] Furthermore, the method of spreading the influence of the mutation information of a single gene node to the entire functional interaction network to smooth and redistribute the mutation frequency of the gene node includes:
[0018] First, based on the established stage-specific functional interaction network, the initial mutation frequency of each gene node in the network, the degree of each gene node in the network, and the k-order neighbor node set of each gene node are obtained in advance;
[0019] Secondly, based on the information obtained, the rumor spreading coefficient α obtained through experiments is used to balance the mutation information received during the smoothing process and the original mutation information;
[0020] Finally, all the gene nodes in the network are used as rumor receiving gene nodes in turn to receive rumors, and the mutation frequency of each gene node in the network is measured after the rumor is propagated.
[0021] Furthermore, the interaction between gene node pairs in the network is modeled by gravity, and the gravity model is:
[0022]
[0023] Where P(t1) represents the smoothed mutation frequency of the gene node. x (t1) and P y(t1) represents the P(t1) value of gene nodes x and y respectively. The P(t1) value is used here to analogize the quality of gene nodes in the network, and PCC(x,y) represents the Pearson correlation coefficient of gene nodes x and y, which is used to represent the correlation between two gene nodes in gene expression.
[0024] Furthermore, the extraction of high gravity interactions to construct a high gravity network includes:
[0025] In the functional interaction network, after modeling using the gravity model, the gravity corresponding to the interactions between these gene node pairs is ranked from high to low, and at the same time, the interactions containing driving effects are taken as the target, combined with the accuracy index for comprehensive consideration; all interactions with an accuracy of not less than 90% are selected as the final target, and the subnet composed of these screened interacting gene node pairs is regarded as a high-gravity network.
[0026] Furthermore, the introduction of attractive gravity and repulsive gravity, and the integration of the expression relationship into the gravity model, includes:
[0027] In the present invention, the positive and negative signs of the Pearson correlation coefficient PCC of the gene node pair are taken into consideration. For the interaction between two gene node pairs with PCC>0, since they promote each other in the gene expression process, the relationship between their expression product proteins in biological functions is also relatively close, so they can be considered to be mutually attractive, and such interaction is described by attractive gravity; conversely, for the interaction between two gene node pairs with PCC<0, since they inhibit each other in the gene expression process, the relationship between their expression product proteins in biological functions is also relatively distant, so they can be considered to be mutually exclusive, that is, such interaction is described by repulsive gravity.
[0028] Furthermore, the step of determining the comprehensive gravity of the gene node and using it as the weight of the gene node includes:
[0029]
[0030] Where d represents the degree of gene node i, Gravity i Indicates gene section i The gravitational force between its neighboring gene nodes, sumG is the comprehensive gravitational force of gene node i, that is, the sum of the gravitational forces between a gene node and all other adjacent gene nodes;
[0031] Assigning the weight of the comprehensive gravity of the gene nodes in the network can enrich the entire model from the perspective of the gene nodes.
[0032] Furthermore, the construction of a cancer stage evolution visualization model displayed in a hierarchical form for cancer evolution research including determining evolution factors and transition factors includes:
[0033] The network smoothing method and gravity model are used to model the specific functional interaction network of each stage of cancer evolution. After modeling, the interaction networks of each stage are arranged in order from top to bottom, and constructed in a hierarchical form, and the corresponding driver genes are connected to form a top-down hierarchical evolution network;
[0034] By determining the driver genes and interactions containing driver effects in the gene nodes of each interaction network, the driver genes and interactions containing driver effects that appear in adjacent stages are counted, and the ones that appear together are connected separately. The ones that appear together in the functional interaction network of each stage are marked as "evolution factors", that is, the driver genes that always accompany the entire evolutionary process, while those that only appear in adjacent stages but not in all stages are marked as "transition factors", that is, factors that play a stage transition effect in the evolutionary process.
[0035] A device for mining cancer evolution information, comprising:
[0036] The network smoothing module is used for the stage-specific functional interaction network based on the cancer stage. It uses a network smoothing method based on the rumor propagation algorithm to spread the influence of the mutation information of a single gene node to the entire functional interaction network, and is used to smooth and redistribute the mutation frequency of gene nodes;
[0037] The gravity modeling module is used to perform gravity modeling on the interactions between gene node pairs in the network based on the smoothed functional interaction network; on the basis of gravity modeling, based on the gravity between gene node pairs, a high gravity network is constructed by extracting high gravity interactions, while introducing attractive gravity and repulsive gravity to integrate expression relationships into the gravity model; the comprehensive gravity of gene nodes is determined and used as the weight of gene nodes;
[0038] The visualization display module is used to connect the stage-specific functional interaction networks of each cancer evolution stage and construct a cancer stage evolution visualization model displayed in a hierarchical form for cancer evolution research including the identification of evolutionary factors and transition factors.
[0039] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the method for mining cancer evolution information is implemented.
[0040] Compared with the prior art, the present invention has the following technical features:
[0041] The present invention provides a method for mining key information in the evolution process of breast cancer, which can clearly, effectively and intuitively describe cancer evolution and provide strong support for cancer research and treatment. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is a schematic diagram of the basic rumor propagation network smoothing method in the present invention;
[0043] Figure 2 It is a schematic diagram of a rumor propagation network smoothing method based on multiple concurrent starting points in the present invention;
[0044] Figure 3 This is the final visualization result of the evolution information using the breast cancer example in the present invention. DETAILED DESCRIPTION
[0045] Referring to the attached figure, the present invention provides a method for mining cancer evolution information, which is applied to cancer genomics research in cancer evolution information, especially in understanding tumor heterogeneity, revealing key regulatory molecules and functional identification in the process of tumor evolution, tumor diagnosis and prognosis determination, etc. This method needs to integrate disease clinical data, gene mutation data, RNA sequence data, based on the known biological protein interaction network, into multiple stage-specific functional interaction networks, one of which corresponds to a cancer stage, that is, the present invention will mine key information of cancer evolution on the stage-specific functional interaction network.
[0046] The present invention specifically comprises the following steps:
[0047] Step 1: Based on the stage-specific functional interaction network of the cancer stage, a network smoothing method based on the rumor propagation algorithm is used to spread the influence of the mutation information of a single gene node to the entire functional interaction network to smooth and redistribute the mutation frequency of the gene node.
[0048] First, we build a stage-specific functional interaction network for cancer stages, in which multi-omics data are divided into four disease stages (precancerous lesions, early cancer, mid-stage cancer, and late-stage cancer) according to clinical data. In each disease stage, we first build a biological protein interaction network as the underlying network, and also need to combine gene mutation data and RNA sequence data, and further screen the network. Finally, the nodes of the stage-specific functional interaction network are genes, and the gene nodes are assigned weights, which are the mutation frequencies of the genes, and the weights are spread as rumors in the network. The edges in the network are interactions between genes, and each edge is also assigned a weight, which represents the correlation between the two gene nodes connected by the edge in gene expression, expressed by the Pearson correlation coefficient (PCC).
[0049] The network smoothing method based on the rumor propagation algorithm is a method that can spread the influence of a single gene node mutation information to the entire network. The rumor propagation algorithm here was first applied to an algorithm mechanism between P2P-based servers in big data processing technology to communicate with each other. The main idea is that the rumor is sent from a gene node in the network and propagated to other gene nodes in the network until other gene nodes in the entire network have received the rumor. Applying this idea to the functional interaction network, the mutation information of the gene node can be regarded as a rumor, and the mutation information of each gene node can be propagated to the remaining gene nodes in the network in a certain way to update the information finally obtained by each gene node. Therefore, the rumor propagation for each gene node in the network can be summarized by the following formula:
[0050] P(t1)=αP mutation +(1-α)P(t0)
[0051] Where P(t0) represents the initial mutation frequency of the gene node receiving the rumor, P mutation represents the mutation frequency of the gene node that spreads the rumor, P(t1) represents the mutation frequency of the gene node after receiving the rumor, where α is the rumor spreading coefficient, and its value range is [0,1]. Its meaning is to serve as a harmonic coefficient to balance the mutation information received and the original mutation information during the smoothing process.
[0052] Based on the above idea, the network smoothing method is further optimized and improved. The first thing to deal with is the problem of multiple rumors concurrently. For a gene node in the network, the mutation information it receives should be the sum of the information from other gene nodes in the network. Therefore, for the situation where multiple rumors are propagated concurrently in the network, multiple rumors need to be accumulated. The second is the problem of rumor propagation efficiency. For the mutation information of a point propagating in the network, the information received by its k-order neighbors should have a certain efficiency loss, and its acceptance efficiency is At the same time, in biological networks, it is necessary to impose a certain degree penalty on points with large degrees to balance the rumors received. Assuming that the degree of the gene node receiving the rumor is d, the final receiving efficiency is
[0053] Based on the above description, we have obtained a rumor propagation algorithm based on multi-starting point concurrency for network smoothing. This smoothing method not only adds multi-starting point concurrency and efficiency reduction on the basis of the original rumor propagation algorithm, but also can satisfy the propagation of single gene node information in the entire network, and can better combine mutation information with network topology. Assuming that the gene node receiving the rumor in the functional interaction network is u, and the other gene nodes are v, the following formula is used to describe it:
[0054]
[0055] Where n represents the total number of gene nodes in the network, u represents the gene node that receives the rumor, vi represents the remaining gene nodes, there are n-1 gene nodes that spread the rumor, and all of them spread the mutation information to node u, t0 represents the initial time point before the rumor spread, and t1 represents the time point when the rumor spread is completed; P vi→u represents the mutation information propagated by each rumor-spreading gene node vi to gene node u, P vi (t0) represents the initial mutation frequency of each rumor-spreading gene node before the rumor spread; k vi Indicates that gene node vi is the kth-order neighbor of gene node u, d vi Represents the degree of gene node vi in the network. The mathematical symbol ! is the factorial symbol; P u (t0) represents the initial mutation frequency of gene node u, and P u (t1) represents the mutation frequency of gene node u after the rumor is spread.
[0056] When applying this network smoothing method, it is first necessary to obtain in advance the network structure and the initial mutation frequency of each gene node in the network, i.e., P(t0), the degree d of each gene node in the network, and the set of k-order neighbor nodes of each gene node based on the established stage-specific functional interaction network; secondly, based on the obtained information, the best rumor propagation coefficient α is obtained through repeated experiments to balance the mutation information received during the smoothing process and the original mutation information; finally, all the nodes in the network need to be regarded as gene nodes that receive rumors in turn, and the mutation frequency of each gene node in the network after the rumor propagation is P(t1).
[0057] Step 2: Based on the smoothed functional interaction network, gravity modeling is performed on the interactions between gene node pairs in the network. On the basis of gravity modeling, based on the gravity between gene node pairs, a high gravity network is constructed by extracting high gravity interactions, and attractive gravity and repulsive gravity are introduced to integrate the expression relationship into the gravity model. The comprehensive gravity of the gene node is determined and used as the weight of the gene node.
[0058] The gravitational model is also used to more fully describe interactions that include driving effects.
[0059] The earliest Newtonian gravitational model can be briefly described as the gravitational force between two molecules is proportional to the product of the masses of the two molecules and inversely proportional to the square of the distance between the two molecules. With this idea, the interaction between nodes is modeled on the functional interaction network in combination with the results of network smoothing. For the gravitational modeling of two adjacent gene nodes x and y in the functional interaction network, the following formula is shown:
[0060]
[0061] Where P(t1) represents the smoothed mutation frequency of the gene node. x (t1) and P y (t1) represents the P(t1) value of gene nodes x and y respectively. Here, the P(t1) value is used to compare the quality of gene nodes in the network, and PCC represents the Pearson correlation coefficient of gene node pairs, which is used to represent the correlation between two gene nodes in gene expression. It is a non-zero value in the range of [-1,1]. Therefore, for |PCC|, the smaller the value, the weaker the correlation between the two genes, and the larger the value, the stronger the correlation between the two genes. Therefore, the reciprocal of |PCC| is used to compare the distance, that is, the more distant the relationship between the two genes, the farther the distance between them, and vice versa, the closer the relationship between the two genes, the closer the distance between them. Using such a gravity model to model the interactions in the network first has a certain theoretical basis. At the same time, it combines the results of the above network smoothing and takes the protein expression relationship into consideration. It is a more comprehensive modeling method.
[0062] The following three extensions can be made to the gravity model to enrich the model and assist in screening interactions with driving effects as key driving genes in each stage:
[0063] (1) Construction of a high-gravity network. The gravity values are normalized and sorted from large to small. Combined with accuracy, as many high-gravity interactions as possible are selected as key information for cancer stage evolution.
[0064] In the functional interaction network, after modeling using the gravity model, there is a specific value to measure the interactions in the entire network. The gravity corresponding to the interactions between these gene node pairs is ranked from high to low, and the interactions with driving effects are taken as the target at the same time, combined with the accuracy index for comprehensive consideration. When the accuracy (an indicator used for classification problems in machine learning, accuracy) is not less than 90%, as many interactions as possible are selected as the final goal, and the subnetwork composed of these screened interacting gene node pairs is called a high-gravity network. Since these high-gravity networks contain more interactions with driving effects, they themselves can also represent stages in the process of cancer evolution and appear as a group with driving effects. Therefore, high-gravity network modeling is a type of driver gene that can reflect the evolution of cancer stages.
[0065] (2) Based on the positive and negative Pearson coefficients (PCC) between gene node pairs, attractive gravity and repulsive gravity are introduced to integrate the expression relationship into the gravity model.
[0066] Since the interaction gravitational forces obtained by the basic gravitational model are all greater than 0, but in the actual network, the force is a vector, so it is necessary to assign a direction to the gravitational force, that is, to distinguish between attractive gravitational forces and repulsive gravitational forces. Therefore, the positive and negative signs of PCC are taken into consideration in the present invention. For the interaction between two gene node pairs with PCC>0, they promote each other in the gene expression process, and the relationship between their expression product proteins in biological functions is relatively close, so they can be considered to be mutually attractive, that is, attractive gravitational forces are used to describe such interactions; on the contrary, for the interaction between two gene node pairs with PCC<0, they inhibit each other in the gene expression process, and the relationship between their expression product proteins in biological functions is relatively distant, so they can be considered to be mutually exclusive, that is, repulsive gravitational forces are used to describe such interactions. Using attractive gravitational forces and repulsive gravitational forces for modeling further enriches the direction of the gravitational model, and at the same time combines the gravitational model with molecular functions.
[0067] (3) Node comprehensive gravity, which is the algebraic sum of all the gravity involved by the gene node.
[0068] Since the gravity model is a model used to measure interactions, and both ends of each interaction are gene nodes, the present invention extends the gravity model from interactions to gene nodes, that is, algebraically sums all the gravitational forces involved in the gene nodes, so that each gene node also has a certain description method. Specifically, it can be described by the following formula:
[0069]
[0070] Where d represents the degree of gene node i, Gravity i Indicates gene section i The gravitational force between a gene node and its neighboring gene nodes, sumG is the comprehensive gravitational force of gene node i, that is, the sum of the gravitational forces between a gene node and all other adjacent gene nodes.
[0071] By introducing node comprehensive gravity, focusing gravity from the interaction to the gene nodes at both ends of the interaction, and giving the gene nodes in the network the weight of the node comprehensive gravity, the entire model can be enriched from the perspective of gene nodes; using node comprehensive gravity, we can see which gene nodes contribute more "gravity" in cancer evolution, as well as the categories of these genes and the roles they play in the cancer evolution process, providing supplementary information on key information in the evolutionary process from the perspective of gene nodes, and enriching the subsequent evolutionary hierarchical network.
[0072] This paper uses the basic gravity model and three extended gravity models in the functional interaction network to provide a more detailed description of the key biomolecules in the evolution of cancer stages. Within each stage, the interactions containing driving effects are mined and identified with the help of high gravity networks, and a systematic description is also given.
[0073] Step 3: Based on the key information in the evolutionary process mined above, the stage-specific functional interaction networks of each cancer evolutionary stage are connected together to construct an intuitive cancer stage evolution visualization model to display the cancer evolution information in a hierarchical form.
[0074] Based on the network smoothing and gravity model proposed in this invention, modeling is performed on the specific functional interaction network of each stage of cancer evolution, and the interactions with driving effects and related information about each stage are obtained; the entire perspective is extracted from each stage, and cancer evolution is described from a more macroscopic perspective. We arrange the interaction networks of each stage in order from top to bottom, build them in a hierarchical form, connect the corresponding key driving genes, and form a top-down hierarchical evolution network.
[0075] This evolution visualization idea needs to solve how to model the connection between each stage, which requires the intersection and union strategy of driver genes between stages. By determining the driver genes and interactions containing driver effects in the gene nodes of each interaction network, the driver genes (a category of gene nodes, see the description in the background technology) and interactions containing driver effects that appear in adjacent stages are counted (in this scheme, the gene node pairs that constitute the high gravity network have interactions containing driver effects), and the co-occurring ones are connected separately. The co-occurring ones in the functional interaction network of each stage are marked as "evolution factors", that is, the driver genes that always accompany the entire evolutionary process, and those that only appear in adjacent stages but not in all stages are marked as "transition factors", that is, factors that play a stage transition effect in the evolutionary process.
[0076] In one example of the present invention, Figure 3 As shown, the functional interaction networks in stages 1 to 4 all have interactions between the TP53 gene and the HRAS gene. The interactions between these two genes serve as evolutionary factors throughout the four stages of disease development; while the interactions between the PIK3CA gene and the UBTF gene only appear in individual stages, so they appear as transition factors in the transition between stages.
[0077] At the same time, within each stage, the high gravity network needs to be integrated with attractive gravity, repulsive gravity and node comprehensive gravity for display. Integrating the three different gravity models into a complete network will enrich the key biological molecular information within each stage. At the same time, the types of gene nodes need to be labeled within each stage, and four special types of gene nodes are labeled with different colors: driver genes, DNA repair genes, chromatin regulators and essential genes. These four types of genes perform different biological functions in the process of cancer evolution and play a very important role.
[0078] Therefore, the cancer stage evolution visualization model in the present invention is a top-down stage evolution model, which is a model that intuitively visualizes the mined cancer evolution information, and the first two parts of the present invention are a method for mining key information on cancer evolution in biological networks.
[0079] Another aspect of the present invention provides a device for mining cancer evolution information, comprising:
[0080] The network smoothing module is used for the stage-specific functional interaction network based on the cancer stage. It uses a network smoothing method based on the rumor propagation algorithm to spread the influence of the mutation information of a single gene node to the entire functional interaction network, and is used to smooth and redistribute the mutation frequency of gene nodes;
[0081] The gravity modeling module is used to perform gravity modeling on the interactions between gene node pairs in the network based on the smoothed functional interaction network; on the basis of gravity modeling, based on the gravity between gene node pairs, a high gravity network is constructed by extracting high gravity interactions, while introducing attractive gravity and repulsive gravity to integrate expression relationships into the gravity model; the comprehensive gravity of gene nodes is determined and used as the weight of gene nodes;
[0082] The visualization display module is used to connect the stage-specific functional interaction networks of each cancer evolution stage and construct a cancer stage evolution visualization model displayed in a hierarchical form for cancer evolution research including the identification of evolutionary factors and transition factors.
[0083] The specific designs and related instructions of the above-mentioned network smoothing module, gravity modeling module, and visualization display module correspond to steps 1 to 3 in the previous text and will not be repeated here.
[0084] The aforementioned steps 1 to 3 can also be loaded into a computer-readable storage medium in the form of a computer program, and when the medium is executed by a processor, the method of steps 1 to 3 is implemented.
[0085] Example:
[0086] This embodiment discloses a method for mining cancer information evolution (taking breast cancer as an example), and the specific steps are as follows:
[0087] Data preparation needs to be collected from public data sets first. The collected content includes breast cancer patient information, genes, mutation data, expression data, PPI network, existing cancer driver genes, DNA repair genes, chromatin regulators, essential genes, etc. The collected data needs to be divided according to the stage of the disease, and then the data is processed internally for each stage. The main network is built based on the PPI network and genes retained with a combined score value greater than 400. Then, for each interaction in the network, the PCC at both ends of the interaction is calculated. For the gene nodes in the network, the gene mutation frequency P(t0) is mapped to the corresponding gene category, and finally four stage-specific functional interaction networks are constructed.
[0088] In Python 3.8 and above, it is necessary to prepare data based on the stage-specific network that has been built and processed before smoothing application. The preparation content mainly includes network search. Traverse all gene nodes in the network, and obtain the initial gene mutation frequency P(t0) of the gene node, the degree d of the gene node, and the circle situation of the remaining gene nodes in a gene node network for each gene node traversal, that is, obtain the k value of the remaining gene nodes. Then it is necessary to smooth the network based on rumor propagation in the network, and explore the optimal α value in the process of network smoothing. In the breast cancer example, taking the final mining target (including the interaction of driving effects) as an example, the α rumor propagation parameter used in each stage network was finally determined to be 0.2.
[0089] In the network obtained based on network smoothing, data preparation is required before using the gravity model. The preparation content is to obtain the P(t1) value of each gene node, the first-order neighbor set of each gene node, and the original PCC value of each interaction. The gravity model is used to model each stage-specific network, and the gravity of each interaction is evaluated according to the gravity model formula. The obtained gravity values are normalized within the stage and sorted in order from large to small. Combined with the accuracy, the interactions corresponding to the gravity of the top 100 in the breast cancer example are obtained as high gravity networks. In addition, based on PCC, the gravity of each interaction is divided into attraction and repulsion, and the comprehensive gravity of the node is obtained by combining the gravity values on the interaction with the midpoint of the first-order neighbor set.
[0090] Finally, based on the obtained high-gravity network, the high-gravity networks of the four stages are integrated and sorted in the order of Stage 1, Stage 2, Stage 3, and Stage 4. The intersection and union strategy is used to string the entire four stages together. At the same time, the gene categories of the gene nodes in the four stages are combined to enrich the internal information of the stages. The final effect (breast cancer example) is as follows Figure 3 As shown, the present invention discloses the result of finally visualizing this evolution information in the breast cancer example.
[0091] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.
Claims
1. A method for mining cancer evolution information, characterized in that: include: Based on the stage-specific functional interaction network of cancer stages, a network smoothing method based on a rumor propagation algorithm is used to spread the influence of mutation information of a single gene node to the entire functional interaction network, which is used to smooth and redistribute the mutation frequency of gene nodes; Based on the smoothed functional interaction network, gravity modeling is performed on the interactions between gene node pairs in the network; on the basis of gravity modeling, based on the gravity between gene node pairs, a high gravity network is constructed by extracting high gravity interactions, and at the same time, attractive gravity and repulsive gravity are introduced to integrate the expression relationship into the gravity model; the comprehensive gravity of the gene node is determined and used as the weight of the gene node; The stage-specific functional interaction networks of each cancer evolution stage are connected together to construct a visualization model of cancer stage evolution displayed in a hierarchical form, which can be used to study cancer evolution, including determining evolutionary factors and transition factors.
2. The method for mining cancer evolution information according to claim 1, characterized in that: The stage-specific functional interaction network is established for the stage of cancer evolution. The nodes of the network are gene nodes. The weight on the gene node is the number of mutations of the gene node. This weight is propagated as a rumor in the network. The edges in the network are the interactions between gene nodes.
3. The method for mining cancer evolution information according to claim 1, characterized in that: The network smoothing method based on rumor propagation algorithm is expressed as: Where n represents the total number of gene nodes in the network, u represents the gene node that receives the rumor, vi represents the remaining gene nodes, there are n-1 gene nodes that spread the rumor, and all of them spread the mutation information to node u, t0 represents the initial time point before the rumor spread, and t1 represents the time point when the rumor spread is completed; P vi→u represents the mutation information propagated by each rumor-spreading gene node vi to gene node u, P vi (t0) represents the initial mutation frequency of each rumor-spreading gene node before the rumor spread; k vi Indicates that gene node vi is the kth-order neighbor of gene node u, d vi Represents the degree of gene node vi in the network. The mathematical symbol ! is the factorial symbol; P u (t0) represents the initial mutation frequency of gene node u, and P u (t1) represents the mutation frequency of gene node u after the rumor is spread.
4. The method for mining cancer evolution information according to claim 1, characterized in that: The method of spreading the influence of mutation information of a single gene node to the entire functional interaction network to smooth and redistribute the mutation frequency of the gene node includes: First, based on the established stage-specific functional interaction network, the initial mutation frequency of each gene node in the network, the degree of each gene node in the network, and the k-order neighbor node set of each gene node are obtained in advance; Secondly, based on the information obtained, the rumor spreading coefficient α obtained through experiments is used to balance the mutation information received during the smoothing process and the original mutation information; Finally, all the gene nodes in the network are regarded as rumor-receiving gene nodes in turn to receive the rumor, and the mutation frequency of each gene node in the network is measured after the rumor is propagated.
5. The method for mining cancer evolution information according to claim 1, characterized in that: The gravitational model is used to model the interaction between gene node pairs in the network. The gravitational model is: Where P(t1) represents the smoothed mutation frequency of the gene node. x (t1) and P y (t1) represents the P(t1) value of gene nodes x and y respectively. The P(t1) value is used here to analogize the quality of gene nodes in the network, and PCC(x,y) represents the Pearson correlation coefficient of gene nodes x and y, which is used to represent the correlation between two gene nodes in gene expression.
6. The method for mining cancer evolution information according to claim 1, characterized in that: The extraction of high gravity interactions to construct a high gravity network includes: In the functional interaction network, after modeling using the gravity model, the gravity corresponding to the interactions between these gene node pairs is ranked from high to low, and at the same time, the interactions containing driving effects are taken as the target, combined with the accuracy index for comprehensive consideration; all interactions with an accuracy of not less than 90% are selected as the final target, and the subnet composed of these screened interacting gene node pairs is regarded as a high-gravity network.
7. The method for mining cancer evolution information according to claim 1, characterized in that: The introduction of attractive gravity and repulsive gravity, and the integration of the expression relationship into the gravity model, includes: In the present invention, the positive and negative signs of the Pearson correlation coefficient PCC of the gene node pair are taken into consideration. For the interaction between two gene node pairs with PCC>0, since they promote each other in the gene expression process, the relationship between their expression product proteins in biological functions is also relatively close, so they can be considered to be mutually attractive, and such interaction is described by attractive gravity; conversely, for the interaction between two gene node pairs with PCC<0, since they inhibit each other in the gene expression process, the relationship between their expression product proteins in biological functions is also relatively distant, so they can be considered to be mutually exclusive, that is, such interaction is described by repulsive gravity.
8. The method for mining cancer evolution information according to claim 1, characterized in that: The step of determining the comprehensive gravity of the gene node and using it as the weight of the gene node includes: Where d represents the degree of gene node i, Gravity i Indicates gene section i The gravitational force between its neighboring gene nodes, sumG is the comprehensive gravitational force of gene node i, that is, the sum of the gravitational forces between a gene node and all other adjacent gene nodes; Assigning the weight of the comprehensive gravity of the gene nodes in the network can enrich the entire model from the perspective of the gene nodes.
9. The method for mining cancer evolution information according to claim 1, characterized in that: The invention constructs a cancer stage evolution visualization model in a hierarchical form for cancer evolution research including determining evolutionary factors and transitional factors, including: The network smoothing method and gravity model are used to model the specific functional interaction network of each stage of cancer evolution. After modeling, the interaction networks of each stage are arranged in order from top to bottom, and constructed in a hierarchical form, and the corresponding driver genes are connected to form a top-down hierarchical evolution network; By determining the driver genes and interactions containing driver effects in the gene nodes of each interaction network, the driver genes and interactions containing driver effects that appear in adjacent stages are counted, and the co-appearing ones are connected separately. The co-appearing ones in the functional interaction network of each stage are marked as "evolution factors", that is, the driver genes that always accompany the entire evolutionary process, while those that only appear in adjacent stages but not in all stages are marked as "transition factors", that is, factors that play a stage transition effect in the evolutionary process.
10. A device for mining cancer evolution information, characterized in that: include: The network smoothing module is used for the stage-specific functional interaction network based on the cancer stage. It uses a network smoothing method based on the rumor propagation algorithm to spread the influence of the mutation information of a single gene node to the entire functional interaction network, and is used to smooth and redistribute the mutation frequency of gene nodes; The gravity modeling module is used to perform gravity modeling on the interactions between gene node pairs in the network based on the smoothed functional interaction network; on the basis of gravity modeling, based on the gravity between gene node pairs, a high gravity network is constructed by extracting high gravity interactions, while introducing attractive gravity and repulsive gravity to integrate expression relationships into the gravity model; the comprehensive gravity of gene nodes is determined and used as the weight of gene nodes; The visualization display module is used to connect the stage-specific functional interaction networks of each cancer evolution stage and construct a cancer stage evolution visualization model displayed in a hierarchical form for cancer evolution research including the identification of evolutionary factors and transition factors.
Citation Information
Cited By
Multilayer regulation and control network key node identification method based on global interactive transitivity and feature fusion
CN119741979A
