Model for identifying senescence critical state and dynamic network marker thereof

Abstract

The invention discloses an aging critical state and dynamic network marker identification model, which can identify a DNB community with a self-adaptive scale based on an SSN (Secure Socket Number) of one sample for a complex biological process such as aging, so that the DNB community scale of each sample has the characteristics of certainty and individuation; for a complex biological process such as aging, the method can greatly improve the calculation efficiency of SSN construction and local network score calculation; for the complex biological process such as aging, the quality of single-sample DNB and group DNB can be remarkably improved; according to the invention, the critical state of male aging and related single-sample DNB and group DNB are successfully identified in aging-related gene expression data of 130 samples.

Description

Technical Field

[0001] The present invention belongs to the technical field of bioinformatics, and in particular relates to an aging critical state and a dynamic network marker recognition model thereof. Background Art

[0002] As the problem of population aging continues to intensify, people's demand for "healthy aging" is increasing day by day. The identification of individual critical states of aging and personalized aging markers is of great significance for the early intervention of aging-related diseases and the study of potential mechanisms of aging. The present invention combines dynamic network marker theory, single sample network construction, community discovery method and parallel optimization strategy to achieve accurate and personalized identification of sample critical states of aging and biomarkers.

[0003] In mathematics, the mutations that occur at bifurcations in complex dynamical systems are described by the transition of system states. Therefore, there are three stages in the evolution of complex dynamical systems: a relatively normal stage, a stage before a critical transition (critical state stage), and a stage after a critical transition. Finding the critical state stage and its corresponding biomarkers in complex biological processes (such as aging and aging-related diseases) makes it possible to conduct targeted interventions before critical transitions occur, which is of great significance for achieving "treating diseases before they occur."

[0004] In order to achieve accurate identification, Professor Luonan Chen proposed the concept of DNB (dynamic network biomarker) based on mathematical theory. According to the theory of DNB, when a complex dynamic system is in a critical state, there is a group of dominant molecules (i.e., DNBs) in the molecular network that exhibit short-term large fluctuations and strong correlations with each other. Specifically, when the system is in a critical state, it has the following three characteristics (necessary conditions):

[0005] (a) The standard deviation of each molecule in DNBs increases dramatically;

[0006] (b) The average absolute value of the Pearson correlation coefficient between DNBs increases sharply;

[0007] (c) The average absolute value of the Pearson correlation coefficient between DNBs and non-DNBs decreased sharply.

[0008] Professor Chen Luonan proposed a comprehensive index CI DNB Used to quantitatively characterize the characteristics of DNB molecules when the critical state occurs:

[0009]

[0010] Among them, SD in 、PCC in and PCC outThey represent the above features (a), (b) and (c) respectively.

[0011] Compared with traditional single-molecule markers, dynamic network markers take into account the view that "complex biological processes are usually determined by the interaction of multiple molecules" and the temporal characteristics of complex biological processes compared with network markers. In addition, due to the use of the second-order statistic of the Pearson correlation coefficient for characterization, many false negative or false positive situations are avoided.

[0012] The single sample network is a molecular regulatory network constructed from the sequencing data of a single sample, which solves the problem that the establishment of a sample network requires multiple sets of samples from the same person in a single sampling period, but such data requirements are often difficult to meet. Figure 1 As shown in Figure 1, sample-specific network is a method for constructing single sample networks that is widely used in the field of critical state identification based on DNB theory. SSN (sample-specific network) characterizes the sample characteristics of a sample by calculating the significant disturbance caused by adding a single sample to a set of reference sample data.

[0013] A community is a subset of nodes in a complex network that are tightly connected. The phenomenon that a complex network contains communities is called a community structure. Community discovery is a process of finding the community structure in a complex network based on an optimization goal. Currently, community discovery methods are widely used in the design of recommendation systems for social networks and in identifying key node molecules in the field of biochemistry.

[0014] Currently, l-DNB (landscape-DNB, a dynamic network marker method based on a single sample "landscape") has been widely and successfully applied to the identification of critical states of biological processes such as breast cancer cell differentiation, coronary atherosclerosis, and skin response to repeated ultraviolet B irradiation. This method combines the SSN method and DNB theory, heuristically providing the possibility of identifying critical states of biological processes and personalized DNB.

[0015] like Figure 2 As shown in Figure 1, the l-DNB method consists of three steps: constructing SSN for each sample, calculating the local network score (Local DNB score) for each gene in each sample, calculating the sample global network score and identifying critical states and corresponding DNBs:

[0016] a. Construct SSN for each sample

[0017] First, the gene expression data of the m reference group samples will be used alone to construct the reference network. The reference network is actually a gene co-expression network constructed based on the Pearson correlation coefficient. The perturbation network is also constructed in a similar way: add sample d to the m reference samples and calculate the perturbation network based on m+1 samples. Then, the difference between the reference network and the perturbation network represents the perturbation caused by adding sample d to the reference network. At the same time, a Z score is calculated for each network edge weight and its p-test value is used to calculate whether the edge weights between any two genes x and y in the differential network are statistically significant, and the statistically significant edges are retained. At this point, the SSN network of a single sample has been successfully constructed.

[0018] b. Calculate the local network score for each gene in each sample

[0019] For each sample SSN, a local gene network is formed with each gene in the network as the central node, the gene and its first- and second-order neighbor nodes as members, and the score of the local gene network is calculated. Each local network gene network is heuristically considered that its central gene and first-order neighbor genes are DNB molecules, and all the second-order neighbor genes of the central gene are non-DNB molecules. Thus, the average standard deviation of DNBs in the local gene network, the average value of the absolute value of the Pearson correlation coefficient between DNBs, and the average value of the absolute value of the Pearson correlation coefficient between DNBs and non-DNBs are calculated respectively, and finally the local network score of the local gene network is obtained.

[0020] c. Calculate the sample global network score and identify critical states and corresponding DNBs

[0021] For each sample, the genes with the largest local network scores in the first q genes are taken as the DNBs of the sample, and the average of the local network scores of these q genes is the global network score of the sample. The score of each group is the average of the global network scores of the samples in that group. The critical state of the biological process is identified based on whether there is a statistical difference in the score distribution between samples.

[0022] The above prior art has the following disadvantages:

[0023] 1. High computational resource consumption in dealing with aging issues

[0024] When faced with a complex molecular network of tens of thousands of nodes, relying solely on the computing power of the CPU will face huge time overhead. The challenge faced by the present invention is to build a large and complex aging molecular network, a process involving complex biological mechanisms. In terms of computing speed and resource consumption, finding aging-related DNBs is an extremely challenging task. Taking the syn49637038 project in the Synapse database as an example, the aging-related sequencing data contained in this project is not only computationally intensive but also time-consuming when calculating SSN and LS. On the CPU, it takes about 4.5 days to calculate the SSN and LS of a sample containing more than 20,000 gene data. Considering that the data set contains 113 such samples, if only the CPU is used to perform these tasks, it may take up to 1.3 years, which is obviously unacceptable for practical applications or scientific research.

[0025] In order to solve this problem, the present invention particularly proposes a method to improve the calculation efficiency of SSN and LS, aiming to significantly shorten the calculation time, thereby providing more efficient support for aging-related bioinformatics research.

[0026] 2. The range of candidate DNBs for aging is too large and not personalized enough

[0027] Due to the strong sample specificity and complexity of gene regulatory networks related to aging, the existing l-DNB method has a large range of candidate DNBs, making it difficult to draw more effective and accurate conclusions. This is because the same critical state results can be identified within a fairly wide range of parameter q (number of DNBs), which makes it difficult for researchers to accurately determine the actual scale of single-sample DNBs. At the same time, the l-DNB method subjectively assumes that the scale of DNBs for each sample is fixed, which is inconsistent with the original intention of identifying the DNBs of the sample in an individualized manner.

[0028] The present invention will improve this by adaptively identifying the DNB size for each sample and providing an accurate personalized DNB size number rather than a vague range.

[0029] 3. The quality of single-sample DNBs and group DNBs in dealing with aging is not high

[0030] Whether the DNB identification method can identify high-quality single-sample DNBs and group DNBs in complex biological processes such as aging is an important determinant of the accuracy of downstream analysis results. The original l-DNB method has obvious deficiencies in the performance of the single-sample DNB set average score, group DNB discovery ability, group DNB representativeness and high-quality group DNB number in the complex biological process of aging, which has caused a lot of uncertainty in the subsequent biological significance analysis and reduced the accuracy of aging prediction research. Summary of the invention

[0031] In view of this, the purpose of the present invention is to provide a critical state of aging and its dynamic network marker identification model.

[0032] An aging critical state identification model, comprising: a sample-specific network, a local gene network, a single-sample DNB community discovery module, and a critical state identification module;

[0033] The sample-specific network is used to construct a sample-specific network SSN of a single studied sample;

[0034] The local gene network is used to:

[0035] For each sample's SSN, a local gene network is formed with each gene in the network as the central node, the gene and its first- and second-order neighbor nodes as members, and the score LS of the local gene network is calculated;

[0036] The single-sample DNB community discovery module is used to:

[0037] For each sample, the local gene network score LS of the sample is used as the initial weight of the gene. On the basis of the initial DNB community, we continuously try to add high-quality genes into the community, and judge whether the addition of this gene brings enough indicator gain to the DNB community, so as to decide whether to add it to the DNB community; until the termination condition is triggered, the DNB community stops iterating, and outputs the DNB community of the sample and the set indicators as the DNB community score;

[0038] The critical state identification module is used for:

[0039] The samples were grouped by age, and a two-sided t-test was performed on the DNB community score distribution of samples in adjacent groups. If the p-test value was less than the set threshold, it was considered that there was a significant change in the scores between the two groups of samples, thereby determining the critical state of aging.

[0040] Furthermore, it also includes a single sample DNB and a group DNB recognition module, which is used to:

[0041] For samples in the research group at the critical stage, the genes in their DNB community are the DNBs of the samples; the high-frequency single-sample DNBs in this group are the group DNBs of the biological process.

[0042] Preferably, the method for the single-sample DNB community discovery module to output the DNB community score includes:

[0043] (a) According to the local gene network score of the sample, at least one gene x0 with the highest score is added to the DNB community as the initial node, and the DNB community discovery is started from this;

[0044] (b) Find genes that meet the following two conditions among the genes newly added to the DNB community and add them to the candidate gene queue: ① genes that have not been added to the candidate gene queue and ② genes that are not neighbor nodes of genes in the DNB community;

[0045] Update the candidate weights of genes in the candidate gene queue using formula (0):

[0046]

[0047] Among them, x is the candidate queue gene whose weight needs to be updated, y i is the DNB neighbor node of gene x, k is the number of DNB neighbors of gene x, sPCC m (x,y i ) represents gene x and gene y i The weights that change between

[0048] (c) Sort the genes in the candidate gene queue according to the candidate weights calculated in step (b), select some genes with larger weights and try to join the DNB community in order from small to large, and calculate the community's composite evaluation index CI according to the following formula set: com :

[0049]

[0050]

[0051] In formula (1), DEV(x) represents the standard deviation of gene x, characterizing the volatility of gene x, and its size is the absolute value of the difference between the expression value of gene x in the research sample d and the average expression value of gene x in the reference sample; H d is the gene set of the DNB community of sample d, h d is the number of genes in the DNB community; The average volatility level of DNBs representing the sample DNB community;

[0052] In formula (2), sPCC m (x,y) represents the edge weight between genes x and y in the SSN of the sample; sPCC i ' n It indicates the average strength of the association between molecules in the DNB community of the sample;

[0053] In formula (3), V d is the set of non-DNB neighbors of genes in the DNB community, v d for its size; It indicates the average strength of association between the molecules in the DNB community of the sample and the non-DNB neighbors in the DNB community;

[0054] If the addition of the new node brings a gain that exceeds the threshold to the composite evaluation index of the community, then the new node is added to the community, and the candidate queue is updated according to the method of step (b) and step (c) is continued until all selected partial genes are traversed and step (d) is executed; otherwise, the node is put back to the original candidate queue, and the gain of the community by nodes with higher candidate weights is continued;

[0055] (d) Stop community discovery and use the composite evaluation index value of the DNB community at this time as the final score of the sample.

[0056] Preferably, the sample-specific network is used to construct a sample-specific network SSN of a single sample under study, specifically:

[0057] First, the gene expression data of m reference group samples will be used separately to construct a reference network; the reference network is a gene co-expression network constructed based on the Pearson correlation coefficient; the perturbation network construction method is: add each single sample d in the study group to the m reference samples, and calculate the perturbation network based on m+1 samples; then, the difference between the reference network and the perturbation network represents the perturbation caused by adding sample d to the reference network; at the same time, a Z score is calculated for each network edge weight and its p-test value is used to calculate whether the edge weights between any two genes x and y in the differential network are statistically significant, and the statistically significant edges are retained. At this point, the SSN of a single sample is successfully constructed.

[0058] Preferably, the method for calculating the score LS of the local gene network includes:

[0059] Each local gene network is heuristically considered that its central gene and first-order neighbor genes are DNB molecules, and all the second-order neighbor genes of the central gene are non-DNB molecules; thus, the average standard deviation of DNBs in the local gene network, the average value of the absolute value of the Pearson correlation coefficient between DNBs, and the average value of the absolute value of the Pearson correlation coefficient between DNBs and non-DNBs are calculated respectively, and finally the score of the local gene network is obtained.

[0060] The present invention has the following beneficial effects:

[0061] 1. For complex biological processes such as aging, the invention can identify adaptively sized DNB communities based on the SSN of a sample, so that the size of the DNB community of each sample is both deterministic and personalized.

[0062] 2. For complex biological processes such as aging, this invention can greatly improve the computational efficiency of constructing SSN and calculating local network scores.

[0063] 3. For complex biological processes such as aging, this invention can significantly improve the quality of single sample DNB and group DNB.

[0064] 4. This invention has successfully identified the critical state of male aging and the related single-sample DNB and group DNB in ​​the aging-related gene expression data of 130 samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 Constructing a flow chart for the sample-specific network of the present invention;

[0066] Figure 2 It is a framework diagram of the l-DNB method of the present invention;

[0067] Figure 3 This is a framework diagram of the aging critical state identification model based on dynamic network markers and community discovery of the present invention, wherein LS (local score) is the local network score;

[0068] Figure 4 The single-sample DNB community discovery module flow chart of the present invention, wherein w1, ..., w n is the weight of each candidate gene;

[0069] Figure 5 This is a framework diagram of the parallel optimization strategy of the present invention. DETAILED DESCRIPTION

[0070] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0071] In order to identify the critical state and key biomarkers of aging based on the DNB (dynamic network biomarker) theory, all gene expression data samples should be divided into a reference group and a study group, and a SSN (sample-specific network) based on each sample in the study group should be constructed. Then, we calculate the local network score sequence for each local gene network in the SSN, and use it as the initial weight to perform DNB community discovery in each SSN. This will effectively help us identify the critical state and DNB of aging. The framework diagram of the model is shown in the figure below. Figure 3 As shown:

[0072] 1. Sample-specific networks

[0073] First, the gene expression data of the m reference group samples will be used alone to construct the reference network. The reference network is actually a gene co-expression network constructed based on the Pearson correlation coefficient. The perturbation network is also constructed in a similar way: each single sample d in the study group is added to the m reference samples, and the perturbation network based on m+1 samples is calculated. Then, the difference between the reference network and the perturbation network represents the perturbation caused by adding sample d to the reference network. At the same time, a Z score is calculated for each network edge weight and its p-test value is used to calculate whether the edge weights between any two genes x and y in the differential network are statistically significant, and the statistically significant edges are retained. At this point, the SSN of a single sample is successfully constructed.

[0074] 2. Local gene network

[0075] For each sample's SSN, a local gene network is formed with each gene in the network as the central node, the gene and its first- and second-order neighbor nodes as members, and the score of the local gene network is calculated. Each local gene network is heuristically considered that its central gene and first-order neighbor genes are DNB molecules, and all the second-order neighbor genes of the central gene are non-DNB molecules. Thus, the average standard deviation of DNBs in the local gene network, the average value of the absolute value of the Pearson correlation coefficient between DNBs, and the average value of the absolute value of the Pearson correlation coefficient between DNBs and non-DNBs are calculated, and finally the local network score of the local gene network is obtained.

[0076] 3. Single-sample DNB community discovery module

[0077] For each sample, the following single-sample DNB community discovery method can iterate and discover adaptively sized DNB communities in the SSN of the sample. Specifically, this method uses the local network score sequence of the sample as the initial weight of the gene, and continuously tries to add high-quality genes into the community based on the initial DNB community and judges whether the addition of this gene brings sufficient indicator gain to the DNB community, thereby deciding whether to add it to the DNB community. Until the termination condition is triggered, the DNB community discovery stops iterating and outputs the DNB community of the sample. Detailed steps of the single-sample DNB community discovery method (see Figure 4 )as follows:

[0078] (a) According to the local network score sequence of the sample, the gene x0 with the highest score is added to the DNB community as the initial node, and the DNB community discovery is started from this. More genes with higher scores can also be selected to appropriately increase the scale of the community's initial nodes to obtain better community discovery results.

[0079] (b) Find genes that meet the following two conditions among the genes newly added to the DNB community and add them to the candidate gene queue: ① genes that have not been added to the candidate gene queue and ② genes that are not neighbor nodes of genes in the DNB community;

[0080] Update the candidate weights of genes in the candidate gene queue using formula (0).

[0081]

[0082] Among them, x is the candidate queue gene whose weight needs to be updated, y i is the DNB neighbor node of gene x, k is the number of DNB neighbors of gene x, sPCC m (x,y i ) represents gene x and gene y i The weights that vary between them.

[0083] Therefore, the weight of candidate gene x means the strength of the association between gene x and the DNB community.

[0084] (c) Sort the genes in the candidate gene queue according to the candidate weights calculated in step (b). For the sake of computational efficiency, try to add the genes in the top 50% with larger weights to the DNB community in order from small to large, and calculate the community's composite evaluation index CI according to the following set of formulas: com If the addition of a new node brings a gain that exceeds the threshold value to the composite evaluation index of the community, then the new node is added to the community, and the candidate queue is updated according to the method of step (b) and step (c) is continued until all the top 50% of genes are traversed and step (d) is executed; otherwise, the node is put back to the original candidate queue and the gain of the community by nodes with higher candidate weights is continued.

[0085]

[0086] In formula 1, DEV(x) represents the standard deviation of gene x, characterizing the volatility of gene x, and its size is the absolute value of the difference between the expression value of gene x in the research sample d and the average expression value of gene x in the reference sample; H d is the gene set of the DNB community of sample d, h d is the number of genes in the DNB community; therefore, The average volatility level of DNBs representing the sample DNB community;

[0087] In Formula 2, sPCC m (x,y) represents the edge weight between genes x and y in the SSN of the sample; sPCC i ' n It indicates the average strength of the association between molecules in the DNB community of the sample;

[0088] In Formula 3, V d is the set of non-DNB neighbors of genes in the DNB community, v d Because of its size. Therefore, It indicates the average strength of the association between the molecules in the DNB community of the sample and the non-DNB neighbors in the DNB community.

[0089] According to the three characteristics of DNBs in the critical state of complex processes in DNB theory, Formula 4 is composed of the above three formulas and represents the score of the DNB community.

[0090] (d) If the top 50% of genes with larger candidate weights cannot bring sufficient gain to the community, the community discovery is stopped, and the composite evaluation index value of the DNB community at this time is used as the final score of the sample.

[0091] 4. Critical state identification module

[0092] The samples were grouped by age, and a two-sided t-test was performed on the score distribution of samples in adjacent groups. If the p-test value was less than the threshold value of 0.05, it was considered that there was a significant change in the scores between the two groups of samples, thereby determining the critical state of aging.

[0093] 5. Single sample DNB and group DNB identification

[0094] For samples in the critical state research group, the genes in their DNB community are the DNBs of the samples. The high-frequency single-sample DNBs in this group are the group DNBs of the biological process.

[0095] In addition, the sample-specific network and the local gene network of the present invention both adopt a parallel optimization strategy, specifically:

[0096] The parallel computing acceleration process in the present invention is mainly carried out by computing from the differential network to SSN and computing from LS. The implementation methods of each step are as follows: Figure 5 shown.

[0097] In the present invention, the calculation process from the differential network to SSN is a problem constrained by the intensity of mathematical calculations. After calculating the Z score, the algorithm will use this value to further calculate the p-test value to determine the correlation between paired genes in the SSN. This calculation process involves a large number of calculation steps, which places high demands on computing resources. In order to overcome this challenge, the present invention uses parallel processing technology, and each thread independently calculates the Z score and p-test value of a pair of genes, which significantly improves the efficiency of SSN generation.

[0098] In contrast, the calculation of local network scores is a problem subject to storage constraints. The calculation process of local network scores involves a large amount of data summation operations, which requires frequent access to memory for reading and writing. In order to solve this storage bottleneck, the present invention introduces a parallel reduction algorithm to accelerate the summation process, and uses shared memory to optimize the reduction efficiency. Under this framework, each computing block in the GPU is responsible for processing the local network score value of a single gene, while each thread in the block is responsible for calculating the partial summation of the local network score and temporarily storing the intermediate results in the shared memory. When all threads in the block complete the summation, the partial sums in the shared memory are summarized by the parallel reduction algorithm, avoiding frequent access to the global memory. Finally, the algorithm will extract the complete local network score value from the first element of the shared memory.

[0099] The present invention provides a critical state of aging and a dynamic network marker identification model, which has the following beneficial effects:

[0100] 1. For complex biological processes such as aging, the invention can identify adaptively sized DNB communities based on the SSN of a sample, so that the size of the DNB community of each sample is both deterministic and personalized.

[0101] Compared with the original l-DNB method, which determines the size of DNBs based on subjective experience (usually 20-50), the method proposed in the present invention can adaptively identify DNB communities of personalized size in SSNs, which provides the possibility for more accurate and personalized aging signal warning and intervention treatment of aging-related diseases.

[0102] This is because the present invention performs community discovery based on a specified objective function, gradually determines whether the addition of new genes to the community can bring sufficient indicator gains, and adds genes to the DNB community one by one until the iteration ends. Therefore, based on the three characteristics of the critical state mentioned in the DNB theory as the optimization target, a DNB community module that meets the requirements can be found in the SSN. The size of each sample DNB community is different for different SSNs.

[0103] 2. For complex biological processes such as aging, this invention can greatly improve the computational efficiency of constructing SSN and calculating local network scores.

[0104] The test results show that in the conversion calculation from differential network to SSN, the RTX 3090 GPU achieved up to 2,700 times acceleration, while the A100 GPU achieved an astonishing 47,000 times acceleration. In LS calculations, the acceleration ratio of the RTX 3090 GPU was 900, while the acceleration ratio of the A100 GPU was 4,300, both significantly improving computing efficiency.

[0105] As mentioned above, the conversion from differential network to SSN and the calculation of LS are the most time-consuming steps in the whole process. They not only involve complex mathematical operations, but are also limited by the frequency of storage access. The present invention effectively solves these bottleneck problems through parallelization strategies, thereby achieving speedups ranging from 2,700 to 47,000. In the LS calculation, the present invention uses shared memory and reduction algorithms to significantly reduce the storage access delay caused by frequent read and write operations, thereby accelerating the summation process.

[0106] 3. For complex biological processes such as aging, this invention can significantly improve the quality of single sample DNB and group DNB.

[0107] This invention can significantly improve the single-sample DNB set average score, group DNB discovery ability, group DNB representativeness and the number of high-quality group DNBs in aging gene expression data, provide more reliable results for subsequent biological significance analysis, and reduce uncertainty in the study of potential mechanisms of aging.

[0108] Because the local network score proposed by the original l-DNB method is a rough heuristic calculation method, it proposes the probability weight of a single gene becoming the DNB of the sample, but does not consider the overall effect between the entire DNB set, and therefore does not consider the overall performance of the DNB set well. In the present invention, the local network score is used as the initial weight for community discovery, and community discovery is performed through the optimization target based on the DNB theory to find more accurate DNBs that meet the requirements.

[0109] 4. This invention has successfully identified the critical state of male aging and the related single-sample DNB and group DNB in ​​the aging-related gene expression data of 130 samples.

[0110] The method provided by the invention identified that 61-65 years old is the critical state of male aging based on two types of cell data. At the same time, the corresponding group DNBs were successfully enriched into four mechanisms closely related to the aging process: lipid metabolism, oxygen metabolism, mitochondrial function, and immune protein synthesis. At the same time, the enrichment analysis results suggest that these DNBs are related to aging-associated diseases such as Parkinson's disease and Alzheimer's disease. In addition, pneumonia caused by new coronavirus infection has been shown to have a strong correlation with these DNBs, which once again verifies the view that the physiological mechanisms of the new coronavirus and aging are similar. These results are highly consistent with current views and research on the aging process.

[0111] The method framework of this invention has made targeted improvements to the characteristics of aging, such as strong sample specificity, huge gene regulatory network, and complex underlying mechanisms. It has introduced a community discovery method and optimized the weight calculation method in the community discovery logic, so that this method performs well in identifying aging critical states and corresponding DNBs.

[0112] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A critical state identification model for aging, characterized in that: include: Sample-specific network, local gene network, single-sample DNB community discovery module, and critical state identification module; The sample-specific network is used to construct a sample-specific network SSN of a single studied sample; The local gene network is used to: For each sample's SSN, a local gene network is formed with each gene in the network as the central node, the gene and its first- and second-order neighbor nodes as members, and the score LS of the local gene network is calculated; The single-sample DNB community discovery module is used to: For each sample, the local gene network score LS of the sample is used as the initial weight of the gene. On the basis of the initial DNB community, we continuously try to add high-quality genes into the community and judge whether the addition of this gene brings enough indicator gain to the DNB community, so as to decide whether to add it to the DNB community; Until the termination condition is triggered, DNB community discovery stops iterating and outputs the DNB community of the sample and the set indicators as the DNB community score; The critical state identification module is used for: The samples were grouped by age, and a two-sided t-test was performed on the DNB community score distribution of samples in adjacent groups. If the p-test value was less than the set threshold, it was considered that there was a significant change in the scores between the two groups of samples, thereby determining the critical state of aging.

2. The aging critical state identification model according to claim 1, characterized in that: It also includes single sample DNB and group DNB identification modules for: For samples in the research group at the critical stage, the genes in their DNB community are the DNBs of the samples; the high-frequency single-sample DNBs in this group are the group DNBs of the biological process.

3. A critical state identification model for aging as claimed in claim 1 or 2, characterized in that: The method for the single-sample DNB community discovery module to output the DNB community score includes: (a) According to the local gene network score of the sample, at least one gene x0 with the highest score is added to the DNB community as the initial node, and the DNB community discovery is started from this; (b) Find genes that meet the following two conditions among the genes newly added to the DNB community and add them to the candidate gene queue: ① genes that have not been added to the candidate gene queue and ② genes that are not neighbor nodes of genes in the DNB community; Update the candidate weights of genes in the candidate gene queue using formula (0): Among them, x is the candidate queue gene whose weight needs to be updated, y i is the DNB neighbor node of gene x, k is the number of DNB neighbors of gene x, sPCC m (x,y i ) represents gene x and gene y i The weights that change between (c) Sort the genes in the candidate gene queue according to the candidate weights calculated in step (b), select some genes with larger weights and try to join the DNB community in order from small to large, and calculate the community's composite evaluation index CI according to the following formula set: com : In formula (1), DEV(x) represents the standard deviation of gene x, characterizing the volatility of gene x, and its size is the absolute value of the difference between the expression value of gene x in the research sample d and the average expression value of gene x in the reference sample; H d is the gene set of the DNB community of sample d, h d is the number of genes in the DNB community; The average volatility level of DNBs representing the sample DNB community; In formula (2), sPCC m (x,y) represents the edge weight between genes x and y in the SSN of the sample; sPCC i ' n It indicates the average strength of the association between molecules in the DNB community of the sample; In formula (3), V d is the set of non-DNB neighbors of genes in the DNB community, v d for its size; It indicates the average strength of association between the molecules in the DNB community of the sample and the non-DNB neighbors in the DNB community; If the addition of the new node brings a gain that exceeds the threshold to the composite evaluation index of the community, then the new node is added to the community, and the candidate queue is updated according to the method of step (b) and step (c) is continued until all selected partial genes are traversed and step (d) is executed; otherwise, the node is put back to the original candidate queue, and the gain of the community by nodes with higher candidate weights is continued; (d) Stop community discovery and use the composite evaluation index value of the DNB community at this time as the final score of the sample.

4. The critical state recognition model for aging as claimed in claim 3, characterized in that: The sample-specific network is used to construct a sample-specific network SSN of a single sample under study, specifically: First, the gene expression data of m reference group samples will be used separately to construct a reference network; the reference network is a gene co-expression network constructed based on the Pearson correlation coefficient; the perturbation network construction method is: add each single sample d in the study group to the m reference samples, and calculate the perturbation network based on m+1 samples; then, the difference between the reference network and the perturbation network represents the perturbation caused by adding sample d to the reference network; at the same time, a Z score is calculated for each network edge weight and its p-test value is used to calculate whether the edge weights between any two genes x and y in the differential network are statistically significant, and the statistically significant edges are retained. At this point, the SSN of a single sample is successfully constructed.

5. The critical state recognition model for aging as claimed in claim 3, characterized in that: The method for calculating the score LS of the local gene network by the local gene network includes: Each local gene network is heuristically considered that its central gene and first-order neighbor genes are DNB molecules, and all the second-order neighbor genes of the central gene are non-DNB molecules; thus, the average standard deviation of DNBs in the local gene network, the average value of the absolute value of the Pearson correlation coefficient between DNBs, and the average value of the absolute value of the Pearson correlation coefficient between DNBs and non-DNBs are calculated respectively, and finally the score of the local gene network is obtained.

Images