Pre-disease state detection device, pre-disease state detection method, pre-disease state detection program, and recording medium

Abstract

A predictor detection device 2 that detects a predictor of an abrupt change in state of a system of interest is provided with: a classification means 11 which, on the basis of correlations of chronological changes in measurement data related to a plurality of nodes of which the system of interest is composed, classifies the plurality of nodes into a plurality of clusters; and a means 12 which selects a cluster corresponding to a pre-set selection condition on the basis of the correlations of chronological changes in the measurement data of the nodes in each of the classified clusters and the chronological changes in the measurement data between all the nodes, and toggles between detection of a node included in the selected cluster as a dynamic network marker serving as a predictor of an abrupt change in state, and verification as to whether a pre-stored specific node can be detected as a dynamic network marker.

Description

Technical Field

[0001] The present invention relates to a system state abrupt change prediction detection device, a system state abrupt change prediction detection method, a system state abrupt change prediction detection program, a traffic jam prediction detection device, a traffic jam prediction detection method, a traffic jam prediction detection program, a pre-disease state detection device, a pre-disease state detection method, a pre-disease state detection program, and a recording medium.

Background Art

[0002] In many complex systems such as biological systems, traffic systems, and economic systems, it is known that when the state reaches a certain critical threshold (hereinafter also referred to as a "bifurcation point"), it rapidly transitions from one stable state (also referred to as a "steady state") to the next stable state (see, for example, Non-Patent Documents 1 to 5). At this time, if it is possible to detect a bifurcation point (hereinafter also referred to as a "sign of an abrupt change in state") at which the state rapidly transitions from one stable state to the next stable state, it is expected that predictions regarding changes in the state of the system (early detection of diseases, early treatment, prediction of traffic jams, etc.) will become possible.

[0003] For example, as research on the dynamic mechanism of complex diseases, the progression process of disease deterioration (for example, asthma attacks, cancer onset, etc.) has been modeled as a time-dependent non-linear dynamics system, and by analyzing this, those who have found the relationship between the state transition at the bifurcation point and the rapid deterioration of the disease are known (see, for example, Non-Patent Document 6).

[0004] Against such a background, a technology for detecting a Dynamical Network Marker (hereinafter also referred to as "DNM"; particularly, when targeting a living body, it is also referred to as a Dynamical Network Biomarker "DNB") that serves as an indicator of the omen (pre-disease state) of a sudden symptom occurrence in the state of a living body has been disclosed (see, for example, Patent Document 1). Such DNM is based on measurement data of a plurality of factor items obtained from measurements related to a living body. Furthermore, in the transition state from a healthy state to a disease state, a change in the state of a factor of interest and a connecting factor that is dynamically directly connected to this factor is captured as local network entropy, and a technology for detecting DNM based on this local network entropy has been disclosed (see, for example, Patent Document 2). Furthermore, a technology has been disclosed in which data obtained by collecting and analyzing biological substances from a plurality of healthy humans is stored, and the analysis data of biological substances collected only once from a subject is added to determine whether DNM can be detected (see, for example, Patent Document 3).

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Patent Document 3

Non-Patent Documents

[0006]

Non-Patent Document 1

Outdoor Tool2

Outdoor Tools3

Outdoor Tools 4

Direct Environment 5

[0007] The technology disclosed in Patent Document 1 detects DNM using an index of Pearson correlation coefficient. This technology has a problem of weak resistance to noise. On the other hand, the technology disclosed in Patent Document 2 detects DNM using an index of network entropy. This technology has improved resistance to noise compared to Patent Document 1, but is still not sufficient. The technology disclosed in Patent Document 3 detects DNM using an index of differential correlation coefficient from a network of measurement data to which one sample has been added. However, both methods have a problem in that they require a huge number of measurements and calculations of indices to measure all factor items every time.

[0008] The present invention has been made in view of the above circumstances, and has an object to make it possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the state of a target system. [Means for solving the problem]

[0009] In order to solve the above problems, a system status sudden change sign detection device of one embodiment of the present invention is a sign detection device that detects signs of a sudden change in the status of a target system, and comprises: a classification means for classifying a plurality of nodes into a plurality of clusters based on the correlation of time series changes in measurement data for a plurality of nodes that constitute the target system; and a means for selecting a cluster that meets predetermined selection conditions based on the correlation of time series changes in the measurement data of the nodes in each classified cluster and the correlation of time series changes in the measurement data among all nodes, and switching between detecting nodes included in the selected cluster as dynamic network markers (DNMs) that are signs of a sudden status change, or verifying whether a specific node that has been stored in advance can be detected as a dynamic network marker.

[0010] Another aspect of the present invention is a method for detecting a sign of a sudden change in a system state, which detects a sign of a sudden change in the state of a target system, and includes a classification step of classifying a plurality of nodes into a plurality of clusters based on correlations of time-series changes in measurement data related to a plurality of nodes constituting the target system, and a step of selecting a cluster that satisfies a preset selection condition based on correlations of time-series changes in measurement data of nodes in each classified cluster and correlations of time-series changes in measurement data between all nodes, and switching between detecting nodes included in the selected cluster as dynamic network markers that are a sign of a sudden change in state, and verifying whether a specific node stored in advance can be detected as a dynamic network marker.

[0011] Yet another aspect of the present invention is a system status sudden change sign detection program, which is a sign detection method for detecting a sign of a sudden change in the status of a target system, and causes a computer to execute a classification step of classifying a plurality of nodes into a plurality of clusters based on correlations of time-series changes in measurement data related to a plurality of nodes constituting the target system, and a step of selecting a cluster that satisfies a preset selection condition based on correlations of time-series changes in measurement data of nodes in each classified cluster and correlations of time-series changes in measurement data between all nodes, and switching between detecting nodes included in the selected cluster as dynamic network markers that are a sign of a sudden status change and verifying whether a specific node stored in advance can be detected as a dynamic network marker.

[0012] Yet another aspect of the present invention is a non-transitory recording medium having a sign detection program recorded thereon for detecting a sign of a sudden change in the state of a target system. This recording medium is a method for detecting a sign of a sudden change in the state of a target system, and records a program for causing a computer to execute the following steps: a classification step of classifying a plurality of nodes into a plurality of clusters based on correlations of time-series changes in measurement data related to a plurality of nodes constituting the target system, and a step of selecting a cluster that satisfies a preset selection condition based on correlations of time-series changes in measurement data of nodes in each classified cluster and time-series changes in measurement data between all nodes, and detecting a node included in the selected cluster as a dynamic network marker that is a sign of a sudden change in state, or a step of switching between verifying whether a specific node stored in advance can be detected as a dynamic network marker.

[0013] Yet another aspect of the present invention is a traffic congestion sign detection device that detects signs of traffic congestion on a road, and includes: a dynamic network marker detection means that detects dynamic network markers based on the distribution of traffic volume for each node and the correlation of traffic volume between the nodes, with multiple points on the road as nodes, and a switching means that switches between determining that there is a sign of traffic congestion at a node detected as a dynamic network marker, and detecting dynamic network markers based on the distribution of traffic volume and the correlation of traffic volume between nodes only for nodes that have been stored in advance as being likely to experience traffic congestion.

[0014] Yet another aspect of the present invention is a traffic congestion sign detection method, which detects a sign of traffic congestion on a road, and includes a dynamic network marker detection step of detecting dynamic network markers based on a distribution of traffic volume for each node and a correlation of traffic volume between the nodes, with multiple points on the road as nodes, and a switching step of switching between determining that there is a sign of traffic congestion at a node detected as a dynamic network marker, and detecting dynamic network markers based on the distribution of traffic volume and the correlation of traffic volume between the nodes only for nodes that have been stored in advance as being likely to experience traffic congestion.

[0015] Yet another aspect of the present invention is a traffic congestion sign detection program, which is a traffic congestion sign detection method for detecting a sign of traffic congestion on a road, and causes a computer to execute a dynamic network marker detection step of detecting dynamic network markers based on the variance of traffic volume for each node and the correlation of traffic volume between the nodes, with multiple points on the road as nodes, and a switching step of switching between determining that there is a sign of traffic congestion at a node detected as a dynamic network marker, and detecting dynamic network markers based on the variance of traffic volume and the correlation of traffic volume between the nodes only for nodes that have been stored in advance as being likely to cause traffic congestion.

[0016] Yet another aspect of the present invention is a non-transitory recording medium having a traffic congestion sign detection program recorded thereon. The recording medium records a program for causing a computer to execute a traffic congestion sign detection method for detecting a sign of traffic congestion on a road, the program including a dynamic network marker detection step of detecting a dynamic network marker based on the variance of traffic volume for each node and the correlation of traffic volume between the nodes, with multiple points on the road as nodes, and a switching step of switching between determining that there is a sign of traffic congestion at a node detected as a dynamic network marker, and detecting a dynamic network marker based on the variance of traffic volume and the correlation of traffic volume between the nodes only for a node previously stored as being likely to experience traffic congestion.

[0017] Yet another embodiment of the present invention is a pre-disease state detection device, which detects a pre-disease state, which is a transition state from a healthy state to a diseased state, and includes a data generation unit that generates reference gene expression data generated from biological samples collected from a subject at multiple past time points or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculation unit that calculates a reference sample network from the reference gene expression data and calculates a perturbed sample network from the test gene expression data, a sample characteristic network calculation unit that calculates a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculation unit that calculates a differential sample characteristic network from the sample characteristic network at a first time point and the sample characteristic network at a second time point after the first time point, and a differential sample characteristic network calculation unit that calculates an expression level from the differential sample characteristic network. The system includes a local network extraction unit for extracting a local network having a structure in which a differential gene, which is a gene whose correlation with the differential gene changes by a predetermined degree or more, is placed around the differential gene, and adjacent genes whose correlation with the differential gene rises sharply by a predetermined degree or more are arranged around the differential gene, a gene expression probability calculation unit for calculating a gene expression probability which is the probability that each adjacent gene is expressed, a network flow entropy calculation unit for calculating a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value for each adjacent gene based on the gene expression probability, a differential network flow entropy calculation unit for calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy, a temporal differential network flow entropy calculation unit for calculating a temporal differential network flow entropy from the differential network flow entropy, and a non-disease state detection unit. The non-disease state detection unit detects a non-disease state if the difference between the temporal differential network flow entropy at a first time and the temporal differential network flow entropy at a second time is equal to or greater than a predetermined value.

[0018] The gene expression probability may be calculated assuming that the expression of each adjacent gene follows a normal distribution.

[0019] When the number of past time points or the number of people of a subject is n, and the number of types of genes is m, the reference gene expression data is an m×n matrix whose (i, j) component represents the expression level of the i-th (i = 1, …, m) gene at the past time point or person of the j-th (j = 1, …, n) subject, and the test gene expression data is an m×(n + 1) matrix whose (i, j) component represents the expression level of the i-th (i = 1, …, m) gene at the past time point or of the j-th (j = 1, …, n) subject, and whose (i, n + 1) component represents the expression level of the i-th (i = 1, …, m) gene of the subject.

[0020] The reference sample network is a weighted gene expression network generated from the reference gene expression data, and the perturbation sample network may be a weighted gene expression network generated from the test gene expression data.

[0021] The sample characteristic network may be calculated by comparing the reference sample network and the perturbation sample network and leaving different edges.

[0022] The differential sample characteristic network may be calculated by comparing the sample characteristic network at the first time and the sample characteristic network at the second time and leaving different edges.

[0023] When the (i, k, j) component of the expression data of the i-th gene gik in the k-th local network is xikj, and μik and σik are the mean and standard deviation of the gene gik, respectively, the gene expression probability p(xikj) is

Equation

[0024] When the expression data of the i-th gene gik in the k-th local network is xik, and the expression data of all genes in the k-th local network is xk, the local network flow entropy NFET(xik) is given by

number

number

[0025] When the number of adjacent genes is Mk, the differential network flow entropy DNFETk is

number

[0026] When the number of local networks is l, the time-difference network flow entropy TNFET is given by

number

[0027] The non-illness state detection unit may detect the non-illness state if the temporal difference network flow entropy at the second time is equal to or greater than twice the temporal difference network flow entropy at the first time.

[0028] Yet another aspect of the present invention is a method for detecting a pre-disease state, which is a state of transition from a healthy state to a diseased state, and includes a data generating step of generating reference gene expression data generated from a biological sample collected from a subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step of calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step of calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step of calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a step of calculating from the differential sample characteristic network a sample characteristic network having an expression amount equal to or greater than a predetermined level. the local network extraction step extracting a local network having a structure in which a differential gene, which is a gene that changes suddenly, is placed at the center and adjacent genes whose correlation with the differential gene rises sharply to a predetermined level or more are arranged around the differential gene; a gene expression probability calculation step calculating a gene expression probability which is the probability that each adjacent gene is expressed; a network flow entropy calculation step calculating, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step calculating a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step.The non-illness state detection step detects a non-illness state if a difference between a temporal difference network flow entropy at a first time and a temporal difference network flow entropy at a second time is equal to or greater than a predetermined value.

[0029] Yet another aspect of the present invention is a non-disease state detection program, which causes a computer to execute a method for detecting a non-disease state, which is a state in which a healthy state transitions to a diseased state, and includes a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step of calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step of calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step of calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a step of calculating a sample characteristic network having a predetermined expression amount from the differential sample characteristic network. the local network extraction step extracts a local network having a structure in which a differential gene, which is a gene that changes to a degree of more than a certain level, is placed at the center and adjacent genes whose correlation with the differential gene rises sharply to a predetermined degree or more are arranged around it; a gene expression probability calculation step calculates a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation step calculates, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step calculates a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step.The non-illness state detection step detects a non-illness state if a difference between a temporal difference network flow entropy at a first time and a temporal difference network flow entropy at a second time is equal to or greater than a predetermined value.

[0030] Yet another aspect of the present invention is a non-transitory recording medium having a program recorded thereon for causing a computer to execute a method for detecting a pre-disease state, which is a state in which a healthy state transitions to a diseased state. This program includes a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step of calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step of calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step of calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a differential gene characteristic network, which is a gene whose expression amount changes to a predetermined degree or more, from the differential sample characteristic network. the local network extraction step extracts a local network having a structure in which adjacent genes whose correlation with the differential genes rises sharply to a predetermined level or more are arranged around the center; a gene expression probability calculation step calculates a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation step calculates, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step calculates a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step.The non-illness state detection step detects a non-illness state if a difference between a temporal difference network flow entropy at a first time and a temporal difference network flow entropy at a second time is equal to or greater than a predetermined value.

[0031] Yet another embodiment of the present invention is a pre-disease state detection device, which detects a pre-disease state, which is a transition state from a healthy state to a diseased state, and includes a data generation unit that generates reference gene expression data generated from biological samples collected from a subject at multiple past time points or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculation unit that calculates a reference sample network from the reference gene expression data and calculates a perturbed sample network from the test gene expression data, a sample characteristic network calculation unit that calculates a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculation unit that calculates a differential sample characteristic network from the sample characteristic network at a first time point and the sample characteristic network at a second time point after the first time point, and a differential sample characteristic network calculation unit that calculates an expression level from the differential sample characteristic network. the local network extraction unit extracts a local network having a structure in which a differential gene, which is a gene whose correlation with the differential gene changes to a predetermined degree or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined degree or more are arranged around the center; a gene expression probability calculation unit calculates a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation unit calculates, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation unit calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation unit calculates a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection unit.The pre-disease state detection unit has a switching means for switching between detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0032] Yet another aspect of the present invention is a method for detecting a pre-disease state, which is a state of transition from a healthy state to a diseased state, and includes a data generating step of generating reference gene expression data generated from a biological sample collected from a subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step of calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step of calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step of calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a step of calculating from the differential sample characteristic network a sample characteristic network having an expression amount equal to or greater than a predetermined level. the local network extraction step extracting a local network having a structure in which a differential gene, which is a gene that changes suddenly, is placed at the center and adjacent genes whose correlation with the differential gene rises sharply to a predetermined level or more are arranged around the differential gene; a gene expression probability calculation step calculating a gene expression probability which is the probability that each adjacent gene is expressed; a network flow entropy calculation step calculating, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step calculating a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step.The pre-disease state detection step includes a switching step of detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0033] Yet another aspect of the present invention is a non-disease state detection program, which causes a computer to execute a method for detecting a non-disease state, which is a state in which a healthy state transitions to a diseased state, and includes a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step of calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step of calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step of calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a step of calculating a sample characteristic network having a predetermined expression amount from the differential sample characteristic network. the local network extraction step extracts a local network having a structure in which a differential gene, which is a gene that changes to a degree of more than a certain level, is placed at the center and adjacent genes whose correlation with the differential gene rises sharply to a predetermined degree or more are arranged around it; a gene expression probability calculation step calculates a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation step calculates, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step calculates a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step.The pre-disease state detection step includes a switching step of detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0034] Yet another aspect of the present invention is a non-transitory recording medium having a program recorded thereon for causing a computer to execute a method for detecting a pre-disease state, which is a state in which a healthy state transitions to a diseased state. This program includes a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step of calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step of calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step of calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a differential gene characteristic network, which is a gene whose expression amount changes to a predetermined degree or more, from the differential sample characteristic network. the local network extraction step extracts a local network having a structure in which adjacent genes whose correlation with the differential genes rises sharply to a predetermined level or more are arranged around the center; a gene expression probability calculation step calculates a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation step calculates, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step calculates a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step.The pre-disease state detection step includes a switching step of detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0035] Any combination of the above components, and any transformation of the present invention into an apparatus, method, system, recording medium, computer program, etc., are also effective as aspects of the present invention. Effect of the Invention

[0036] According to the present invention, it is possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the state of a target system. [Brief description of the drawings]

[0037]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Figure 23

Figure 24

Figure 25

Figure 26

[0038] The present invention will be described below based on preferred embodiments with reference to the drawings. In the embodiments and modified examples, identical or equivalent components and members are given the same reference numerals, and duplicated descriptions are omitted as appropriate. The dimensions of the components in each drawing are enlarged or reduced as appropriate for ease of understanding. Some components that are not important for explaining the embodiments are omitted in each drawing. Terms including ordinal numbers such as first and second are used to describe various components, but these terms are used only for the purpose of distinguishing one component from other components, and the components are not limited by these terms.

[0039] Before describing specific embodiments, the basic knowledge will be described.

[0040] Figure 1 schematically shows the progression process of the system state. For example, Fig. 1a schematically shows the progression process of a disease. Figs. 1b, 1c, and 1d show the stability of the non-linear dynamics system in the process of the progression process using a potential function. In Figs. 1b, 1c, and 1d, the value of the state variable of the system is taken on the horizontal axis, and the value of the potential function is taken on the vertical axis. As shown in Fig. 1a, the progression process of disease deterioration can be represented using the transitions among three states: the healthy state, the pre-disease state, and the disease state. In the healthy state, the system is stable. At this time, as shown in Fig. 1b, the value of the potential function takes a minimum value. In the pre-disease state, as shown in Fig. 1c, the potential function of the system takes a high value. At this time, the system is in a state vulnerable to external disturbances. That is, the state of the system at this time is at a bifurcation point (in other words, the limit of the healthy state) where it will transition to the disease state with just a small external disturbance. However, such a pre-disease state can often be restored to the healthy state by appropriate treatment. On the other hand, in the disease state, the system stabilizes again. That is, as shown in Fig. 1d, the value of the potential function takes a minimum value again. Therefore, once in the disease state, it becomes difficult to recover to the healthy state from here. Usually, when we say "pre-disease state", it means "a state that has not reached the onset but is gradually departing from the healthy state". However, from the perspective of "the bifurcation point of the transition from the healthy state to the disease state", the pre-disease state has extremely remarkable characteristics and is the key to the early detection and early treatment of diseases.

[0041] Therefore, if the pre-disease state can be detected before falling into the disease state, it can be restored to the healthy state by appropriate treatment. Thus, for the early detection and early treatment of diseases, the detection of the pre-disease state, which is the bifurcation point of the transition from the healthy state to the disease state, is extremely important.

[0042] [Underlying Theory] The inventors used genome high-throughput technology that can obtain a large amount (e.g., several thousand pieces) of data (i.e., high-dimensional data) from one sample, and constructed a mathematical model of the time evolution of complex diseases based on bifurcation theory to study the progression mechanism of disease deterioration at the molecular network level. As a result, the existence of DNMs (dynamic network markers) that can detect the immediately preceding bifurcation (sudden deterioration) state before the occurrence of a state transition in a pre-disease state was elucidated. If this DNM can be used as a warning signal for a pre-disease state, disease models will no longer be necessary, and it is expected that early diagnosis of complex diseases will be realized with only a few samples.

[0043] The system related to the disease progression process (hereinafter referred to as "System (1)") is expressed by the following equation (1). Z(k+1)=f(Z(k);P) (1)

[0044] Here, Z(k)=(z1(k),...,zn(k)) is a variable that represents the dynamic state of the system (1) observed at time k (k=0,1,...), and may be specifically considered to be information such as protein expression level, gene expression level, and metabolite expression level (e.g., information on the concentration and number of molecules such as proteins and genes). P is a slowly changing parameter that drives the state transition of the system (1), and can be, for example, information on genetic factors such as SBP and CNV, and non-genetic factors such as methylation and acetylation. f=(f1,...,fn) is a nonlinear function of Z(k).

[0045] The health state and the disease state can each be expressed by a fixed-point attractor of the state equation Z(k+1)=f(Z(k);P). Because the progression process of a complex disease has very complex dynamic characteristics, the function f is a nonlinear function with several thousand variables. Moreover, it is difficult to identify the elements of the parameter P that drives the system (1). Therefore, it is very difficult to construct and analyze a system model of the health state and the disease state.

[0046] Meanwhile, the system (1) has a fixed point having the following characteristics (A1) to (A3).

[0047] (A1) When Z* is a fixed point of the system (1), Z*=f(Z*;P). (A2) If Pc is the threshold at which the system bifurcates, then if P=Pc, then the absolute value of one real eigenvalue or a pair of complex conjugate eigenvalues ​​of the Jacobian matrix ∂f(Z;Pc) / ∂Z|Z=Z* is 1. (A3) In general, when P≠Pc, the absolute value of the eigenvalue of the system (1) is not 1.

[0048] From the above characteristics, the inventors have theoretically clarified that when the system (1) approaches a state transition point, the following peculiar characteristics appear. That is, when the state of the system (1) approaches a transition point, a dominant group (sub-network) consisting of some nodes appears in the network (1) in which each of the variables z1, ..., zn of the system (1) is configured as a node. The dominant group that appears near the state transition point ideally has the following peculiar characteristics (B1) to (B3).

[0049] (B1) If zi and zj are nodes that belong to the dominant group, PCC(zi,zj) → ±1; SD(zi) →∞; SD(zj) → ∞. (B2) If zi is a node that belongs to the dominant group, but zj is not a node that belongs to the dominant group, PCC(zi,zj) → 0; SD(zi) →∞; SD(zj) → bounded value. (B3) If zi and zj are not nodes in the dominant group PCC(zi,zj)→α, α∈(-1,1); SD(zi) → bounded value; SD(zj) → bounded value.

[0050] where PCC(zi,zj) is the Pearson correlation coefficient between zi and zj, and SD(zi) and SD(zj) are the standard deviations of zi and zj, respectively.

[0051] That is, the appearance of a dominant group having the above-mentioned specific characteristics (B1) to (B3) in the network (1) can be taken as a sign that the system (1) is in a critical transition state (pre-disease state). Therefore, by detecting the dominant group, a sign of a state transition of the system (1) can be detected. That is, the dominant group can be considered as a warning signal indicating a state transition, in other words, a pre-disease state immediately before the disease worsens. Therefore, no matter how complex the system (1) is, or no matter how unknown the driving parameter elements are, by detecting only the dominant group that serves as a warning signal, the pre-disease state can be identified without directly dealing with a mathematical model of the system (1). By identifying a pre-disease state, it becomes possible to realize preventive measures and early treatment against diseases. As disclosed in Patent Document 1, the inventors have named the dominant group that serves as a warning signal indicating the pre-disease state "DNM" (dynamic network marker), and particularly when targeting a biological system, "DNB" (dynamic network biomarker), and as a comprehensive index for detecting it, I = SDd × PCCd / OPCCd However, SDd: The average standard deviation of the nodes in the DNM. PCCd: Mean absolute value of Pearson correlation coefficients between nodes in the DNM; OPCCd: The average absolute value of the Pearson correlation coefficient between a node in the DNM and other nodes. In other words, this DNM is a network that shows logical dynamic connection relationships that produce effective connections only at specific times, and this is used as a marker that indicates the signs of sudden changes in state.

[0052] In this way, the DNM is a dominant group having the specific characteristics (B1) to (B3), and appears in the network (1) as a sub-network consisting of multiple nodes when the system (1) is in a pre-disease state. If each node (z1, ..., zn) in the network (1) is a factor item to be measured for biomolecules such as genes, proteins, and metabolites, the DNM is a group (sub-network) consisting of factor items related to some biomolecules that satisfy the above-mentioned specific characteristics (B1) to (B3).

[0053] Patent Document 1 discloses a method for detecting candidates for DNM by directly utilizing the above-mentioned specific characteristics (B1) to (B3). According to this method, it is possible to detect DNMs that warn of a transition to a disease state from a biological sample. However, when the measurement data contains noise, the detection accuracy deteriorates. In addition, since it is necessary to detect DNMs that satisfy the above conditions (B1) to (B3) from a large amount of measurement data, the amount of calculation is enormous and the detection efficiency is not high.

[0054] [Using memorized DNM] In response to this, the inventors realized that the nodes of the dominant group that constitutes the DNM for each disease are almost the same regardless of the living body, and that if these are stored, the calculations (B1) to (B3) are performed on these, and it is determined that a DNM has been detected only when almost all of the nodes satisfy the conditions (B1) to (B3), then it is possible to omit the enormous amount of calculation required for DNM detection. In other words, if the total number of nodes is N and the number of nodes in the stored dominant group (DNM) is n, then the amount of calculation required for PCC can be reduced to approximately (the number of combinations of two from n) / (the number of combinations of two from N).

[0055] In this case, (B1) to (B3) become as follows. (B1)' For nodes zi and zj that belong to the dominating group, PCC(zi,zj) → ±1; SD(zi) →∞; SD(zj) → ∞; If more than half of the nodes do not satisfy this condition, it is assumed that DNM has not been detected and the detection of DNM is stopped. If (B2)' and (B1)' are satisfied, then for nodes zi that belong to the dominant group and zj that do not belong to the dominant group, PCC(zi,zj) → 0; SD(zi) →∞; SD(zj) → bounded value; However, if more than half of the nodes in the dominant group do not satisfy this condition, SD(zj) is not calculated and the detection of DNM is discontinued, assuming that DNM has not been detected. (B3)' If (B2)' is satisfied, then if zi and zj are not nodes that belong to the dominating group, PCC(zi,zj)→α, α∈(-1,1); SD(zi) → bounded value; SD(zj) → bounded value; Check and I = SDd × PCCd / OPCCd If the difference exceeds a preset threshold, it is determined that DNM has been detected.

[0056] However, there are cases where there are concerns about continuing to use a fixed memorized DNM because of predicted probabilistic fluctuations in DNM and gradual shifts. For this reason, it is effective to provide a switch that can switch the nodes that are the subject of DNM detection so that DNM detection can be performed on all nodes by returning to the original algorithm.

[0057] [Application to traffic congestion prediction] Figure 2 is a schematic diagram of the merging road of the traffic simulator used. The main road has three lanes, and the leftmost lane is connected by a 250m merging ramp. The minimum spacing between vehicles follows a log-normal distribution with a mean of 1.7m, a standard deviation of 0.9m, and a minimum of 1m. The vehicle speed follows a normal distribution with a mean of 100km / h and a standard deviation of 7.5km / h. The length of the vehicle is 4m, and the parameters related to acceleration and deceleration acceleration are 1.5m / s. 2 , 2.0m / s 2It was assumed that the vehicle would not change lanes twice consecutively within 10 seconds. The simulation time step was 0.1 seconds.

[0058] Figures 3 to 6 show two representative traffic flow scenarios. Figures 3 and 4 show the speed and traffic density in Case I, respectively. Figures 5 and 6 show the speed and traffic density in Case II, respectively. In Case I, the inflow traffic volume on the trunk road is set to an average of 4050 vehicles / h, and the traffic volume on the ramp is set to an average of 850 vehicles / h. In Case II, the inflow traffic volume on the trunk road is set to an average of 4800 vehicles / h, and the traffic volume on the ramp is set to an average of 320 vehicles / h. The inflow traffic volume in Case II is slightly higher than in Case I, but the traffic volume on the ramp is low, so the merging zone is relatively smooth. Figures 3 and 4 show the simulation results of the average traffic speed and traffic density in the merging section of the trunk road in Case I, and Figures 5 and 6 show the simulation results of the average traffic speed and traffic density in Case II. In Case I, the traffic flow was modulated after about 10 minutes, and the average traffic speed eventually dropped to about 22 km / h in the merging zone, and the vehicle density exceeded 85 vehicles per km. Stop-and-go congestion was caused at the merging area, leading to congestion.On the other hand, in Case II, although the total traffic volume was slightly higher, a gradual decrease in speed occurred but did not lead to congestion.

[0059] For the purposes of the present invention, we consider the road as a traffic network divided into interconnected cells. Figure 7 shows a schematic diagram of nine cells numbered N1-N9 on a highway. These are the nodes of the DNM. Incidentally, the length of each lane cell is 150m. The first three nodes (N1-N3) are set at the edge of the merging zone, the next three nodes (N4-N6) are set parallel to the merging zone with a distance of 25m from the first set of nodes. The last three nodes (N7-N9) are set at the beginning of the merging zone with the same distance from the second set of nodes.

[0060] Vehicle interval data is acquired at each node every 0.5 seconds. The acquired data is normalized by the mean and variance, and the normalized value z(k) (k is the node number) is used to calculate the standard deviation for each node and the correlation between nodes (Pearson correlation coefficient). Nodes N4 to N6 show a higher standard deviation than the other nodes, especially after 16 minutes, when the state is thought to be transitioning, and also show a very high correlation between N4 to N6. For this reason, these three nodes N4 to N6 were selected as the set of dominant nodes. In addition, the average absolute value of the Pearson correlation coefficient between these nodes and other nodes did not change significantly at any stage and always remained low.

[0061] The effectiveness of the present invention was evaluated in Case I and Case II. Figures 8 and 9 show the average and normalized standard deviation of the vehicle density at all nodes in Case I, respectively. Figures 10 and 11 show the average and normalized standard deviation of the vehicle density at all nodes in Case II, respectively. Figures 9 and 11 also plot the average standard deviation of the measurements of the dominant node set and the non-dominant node set. It can be seen that the dominant node set generally always exceeds the non-dominant node set. In particular, in Case I (Figure 9), the average standard deviation of the dominant node set is clearly higher than the average standard deviation of the non-dominant node set from around 16 minutes to 21 minutes. On the other hand, in Case II (Figure 11), such a clear phenomenon cannot be observed.

[0062] Figures 12 and 13 show the average value of the absolute value of the Pearson correlation coefficient (Cdd) in the set of dominant nodes. In case I, it increased very rapidly from 16 to 21 minutes, and then showed a tendency to disappear. However, in case II, such a phenomenon cannot be observed. Therefore, the DNM index, which has the product of the average value of the standard deviation of the set of dominant nodes and the average value of the absolute value of the Pearson correlation coefficient in the set of dominant nodes as the numerator, increases sharply from 16 to 21 minutes in case I, whereas such a phenomenon cannot be seen in case II. Looking at Figures 3 and 4, it can be seen that in case I, the merging zone is completely congested after 22 minutes, and the congestion is expanding further. Therefore, it can be judged that in case I, the DNM index was able to detect the sign of congestion. On the other hand, looking at Figures 5 and 6, it can be seen that in case II, the density of cars increases, but not as much as in case I, and the average car speed is maintained at about 35 km / h, and no obvious congestion occurs.

[0063] In the above, nodes N4 to N6 were selected as the set of dominant nodes. For this reason, it was necessary to calculate the Pearson correlation coefficient between all nodes (36 ways). Here, if it is memorized in advance that N4 to N6 are likely to become the set of dominant nodes, it becomes unnecessary to calculate the Pearson correlation coefficient for the set of dominant nodes and the set of non-dominant nodes. This allows the amount of calculation to be significantly reduced, since it is sufficient to calculate three Pearson correlation coefficients within the set of dominant nodes. However, there are cases where the prerequisite conditions change, such as the position or structure of the merging lamps changing. For this reason, it is dangerous to always fix the set of dominant nodes to the one that was memorized. For this reason, in order to check from time to time whether the prerequisite conditions have changed, it is important to be able to switch the calculation process so that the Pearson correlation coefficient is calculated using all nodes, rather than using the set of dominant nodes that was memorized.

[0064] [Application to pre-disease detection] The present inventors have also proposed a method for detecting a pre-disease state using local network entropy based on a transition state as a method for detecting DNMs more accurately and efficiently (see, for example, Patent Document 2). This method will be specifically described below. Local network entropy is microscopic entropy calculated statistically and mechanically by focusing on one node in a network that shows a logical dynamic connection relationship in which an effective connection occurs only at a specific timing.

[0065] By using the local network entropy, the noise resistance has been improved compared to Patent Document 1, but it is still not sufficient. Furthermore, the problem of the need for a large amount of measurement data remains unsolved.

[0066] In order to solve this problem, the inventors conducted extensive research and discovered a method for detecting pre-disease states using network flow entropy based on transition states, which can detect DNMs with even greater accuracy and efficiency.

[0067] [Network flow entropy based on transition states] The dynamic behavior of the above-mentioned system (1) near the bifurcation point can be approximately expressed by the following equation (2). Z(t+1)=A(P)Z(t)+ε(t) (2)

[0068] Here, ε(t) is the Gaussian noise, and P is the parameter vector that controls the Jacobian A of the nonlinear function f of the system (1). Then, the change in Z is Δz i (t)=z i (t)-z i Denoting (t-1) (i=1,2,…,n), we can prove the following conclusions (C1) and (C2) based on branching theory and central manifold theory.

[0069] (C1) If P is not near a state transition point or a branch point, For any node i, j (including the case where i=j), Δzi (t+T) is Δz i (t) is statistically independent for i,j=1,2,…,n. (C2) If P is near a state transition point or a branch point, When nodes i and j are both in the dominant group or DNM, Δz i (t+T) and Δz j (t) is highly correlated. When neither node i nor j is in the dominant group or DNM, Δz i (t+T) is Δz i (t) is statistically independent.

[0070] Based on the above conclusions (C1) and (C2), the inventors of the present application focus on the transition state and use network flow entropy (hereinafter referred to as "NFE") to detect DNMs with higher accuracy and efficiency. Below, we explain the concept of NFE based on the transition state and the relationship between NFE and DNMs.

[0071] [Transition state] The transition state for any variable zi at time t is represented as xi(t), where xi(t) satisfies the conditions shown in equations (3) and (4). |z i (t)-z i (t-1)|>d i If so, then x i (t)=1 (3) |z i (t)-z i (t-1)|≦d i If so, then x i (t)=0 (4)

[0072] In equations (3) and (4), di is a threshold value that determines whether the change in node i at time t is large or not. In the embodiment, X(t)=(x 1 (t),…,x nLet (t) be the “transition state” of system (1) at time t. From the above-mentioned peculiar properties of DNM and conclusions (C1) and (C2), we can derive the following properties of the transition states (D1) and (D2).

[0073] (D1) If node i and node j belong to the dominant group or DNM, the transition state x i (t+T) and x i (t) increases sharply, and p(x i (t+T)=1|x j (t) = γ) → 1 p(x i (t+T)=0|x j (t)=γ)→0 where γ∈{0,1} and p is the transition probability. (D2) If neither node i nor node j belongs to the dominant group or the DNB, then the transition state xi(t+T) is statistically independent of xj(t), and p(x i (t+T)=γ i |x j (t)=γ j )=p(x i (t+T)=γ i ) →α Here, γ i ,γ j ∈{0,1}, α∈(0,1).

[0074] In addition, if the system is in a healthy state, it can quickly recover from disturbances, but in a non-illness state, it is sensitive to even small disturbances. Therefore, the above threshold d i must be set so that the system can distinguish between a "small change" in health status and a "big change" in the pre-disease state. Here, 0 ), then for each node k, p(|z k (t)|>d k )=α, and each threshold d is set as shown in the following formula (5). i State z at the time of discrimination from (t-1)i It is to be determined whether a significant change or state transition has occurred between (t) and (t). 1 ,i 2 ,…,i m denotes the m adjacent nodes linked with node i.

[0075]

number

[0076] For example, based on samples taken during the normal state, we set each threshold d i Set.

[0077] [Local Network] Node i has links with m nodes, i.e., node i has links with m adjacent nodes (i 1 ,i 2 ,…i m ), the network with node i at the center is defined as a local network. In this case, the transition state of the local network with node i at the center at time t is i (t)=(X i (t),X i1 (t),…,X im (t)). For simplicity, we will omit "i" and use X i (t) is expressed as X(t).

[0078] The connection relationship of each node i is set based on the interaction between the nodes. For example, when a protein is used as a node, information recorded in a database such as PPI (Protein-Protein Interaction) that shows interactions between proteins can be used. Such databases are available from websites such as BioGrid, TRED, KEGG, and HPRD. When a protein is used as a node, adjacent nodes and a local network and an entire network based on the adjacent nodes are set based on a database such as PPI that shows interactions between proteins, but when another factor is used as a node, a database related to the relevant factor is used.

[0079] The transition state at the next time t+1 based on the transition state X(t) at the time t is m+1 Each of these possible transition states is a probability event, {A u} u1,2,… m+1 This can be expressed by the following equation (6).

[0080] Au={x i =γ 0 ,x i1 =γ 1 ,…,x im =γ m} (6) However, γ 1 ∈{0,1}, l∈{0,1,2,…,m}.

[0081] Therefore, the discrete stochastic process in the local network is expressed as follows:

[0082] {X(t+i)} i=0,1,… ={X(t),X(t+1),…,X(t+i),…} (7) where X(t+i)=Au,u∈{1,2,…,2 m+1}.

[0083] That is, when the system (1) is in a normal state or a pre-disease state, the discrete probability process is a Markov process and can be defined by the Markov matrix P = (p u,v ). Thus, the transition rate from state u to state v can be expressed by the following equation (8).

[0084]

Equation

[0085] [Network Flow Entropy] Assume that the state transition matrix of the above-mentioned local network is stationary for a certain period and does not change. p u,v (t) is the element in the u-th row and v-th column of the state transition matrix and is the transition probability between any two possible states Au and Av. Therefore, in a specific period (healthy state or pre-disease state), the probability process shown by the following equation (9) is a stationary Markov process.

[0086]

Equation

[0087] And there exists a stationary distribution π = (π 1 , …, π 2 m+1 ) that satisfies the following equation (10).

[0088]

Equation

[0089] Based on the above considerations, the network flow entropy NFE T (x ik ) is defined by the following equation (11).

[0090]

Equation

[0091] In equation (11), index "i" indicates the central node i of the local network, and X indicates the state transition process X(t), X(t+1), ..., X(t+T) of the local network. The network flow entropy shown in equation (11) is a concept that extends statistical mechanical microscopic entropy.

[0092] [First embodiment] 14 shows a functional block diagram of a system state sudden change sign detection device 2 according to the first embodiment. The system state sudden change sign detection device 2 detects a sign of a sudden change in the state of a target system. The system state sudden change sign detection device 2 includes a classification means 11 and a switching means 12.

[0093] The classification means 11 classifies a plurality of nodes constituting the target system into a plurality of clusters based on the correlation of time-series changes in the measurement data relating to the plurality of nodes.

[0094] The switching means 12 selects a cluster that meets preset selection conditions based on the time series changes in the measurement data of the nodes in each classified cluster and the correlation of the time series changes in the measurement data among all nodes, and switches between detecting the nodes included in the selected cluster as dynamic network markers that are a sign of a sudden change in status, and verifying whether a specific node that has been stored in advance can be detected as a dynamic network marker.

[0095] According to this embodiment, it is possible to provide a device that can avoid the enormous amount of calculation required to detect a sign of a sudden change in the state of a target system.

[0096] [Second embodiment] 15 is a flowchart of a method for detecting a sign of a sudden change in a system state according to the second embodiment. This method detects a sign of a sudden change in the state of a target system.

[0097] This method includes a classification step S11 of classifying multiple nodes into multiple clusters based on the correlation of time-series changes in measurement data for multiple nodes that constitute the target system, and a switching step S12 of selecting a cluster that meets predetermined selection conditions based on the correlation of time-series changes in measurement data for nodes in each classified cluster and the correlation of time-series changes in measurement data among all nodes, and switching between detecting nodes included in the selected cluster as dynamic network markers that are a sign of a sudden change in status, and verifying whether a specific node that has been stored in advance can be detected as a dynamic network marker.

[0098] According to this embodiment, it is possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the state of the target system.

[0099] [Third embodiment] The third embodiment is a symptom detection program for detecting a symptom of a sudden change in the state of a target system. This program causes a computer to execute a classification step S11 of classifying a plurality of nodes into a plurality of clusters based on correlations of time-series changes in measurement data for a plurality of nodes constituting the target system, and a switching step S12 of selecting a cluster that satisfies a preset selection condition based on correlations of time-series changes in measurement data for nodes in each classified cluster and time-series changes in measurement data between all nodes, and switching between detecting nodes included in the selected cluster as dynamic network markers that are a symptom of a sudden change in state, and verifying whether a specific node stored in advance can be detected as a dynamic network marker.

[0100] According to this embodiment, a program that makes it possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the state of the target system can be implemented in software.

[0101] [Fourth embodiment] The fourth embodiment is a recording medium. This recording medium records a program for making a computer execute a classification step S11 of classifying a plurality of nodes into a plurality of clusters based on correlations of time-series changes in measurement data related to a plurality of nodes constituting a target system, and a switching step S12 of selecting a cluster that satisfies a preset selection condition based on correlations of time-series changes in measurement data of nodes in each classified cluster and correlations of time-series changes in measurement data between all nodes, and detecting a node included in the selected cluster as a dynamic network marker that is a sign of a sudden change in state, or switching between verifying whether a specific node stored in advance can be detected as a dynamic network marker.

[0102] According to this embodiment, a program that makes it possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the state of the target system can be recorded on a recording medium.

[0103] [Fifth embodiment] 16 shows a functional block diagram of a traffic congestion sign detection device 3 according to the fifth embodiment. The traffic congestion sign detection device 3 detects a sign of traffic congestion on a road. The traffic congestion sign detection device 3 includes a dynamic network marker (DNM) detection means 13 and a switching means 14.

[0104] The dynamic network marker detection means 13 detects dynamic network markers based on a plurality of points on a road as nodes, the distribution of traffic volume for each node, and the correlation of traffic volume between the nodes.

[0105] The switching means 14 switches between determining that there is a sign of congestion at a node detected as a dynamic network marker, and detecting a dynamic network marker based on the distribution of traffic volume and the correlation of traffic volume between nodes only for nodes that have been stored in advance as being prone to congestion.

[0106] According to this embodiment, it is possible to provide a device that can avoid the enormous amount of calculation required to detect a sign of a sudden change in the traffic congestion situation on a road.

[0107] [Sixth embodiment] 17 is a flowchart of a traffic congestion sign detection method according to the sixth embodiment. This method detects a traffic congestion sign on a road.

[0108] This method includes a dynamic network marker (DNM) detection step S13 in which multiple points on a road are treated as nodes and a dynamic network marker is detected based on the variance of traffic volume for each node and the correlation of traffic volume between the nodes, and a switching step S14 in which it is determined that there are signs of congestion at a node detected as a dynamic network marker, or to detect a dynamic network marker based on the variance of traffic volume and the correlation of traffic volume between nodes only for nodes that have been stored in advance as being prone to congestion.

[0109] According to this embodiment, it is possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the traffic congestion situation on a road.

[0110] [Seventh embodiment] The seventh embodiment is a sign detection program for detecting a sign of a sudden change in a traffic congestion situation on a road. This program causes a computer to execute a dynamic network marker detection step S13 for detecting a dynamic network marker based on the distribution of traffic volume for each node and the correlation of traffic volume between the nodes, with multiple points on the road as nodes, and a switching step S14 for switching between determining that there is a sign of congestion occurring at a node detected as a dynamic network marker, and detecting a dynamic network marker based on the distribution of traffic volume and the correlation of traffic volume between nodes only for nodes previously stored as likely to cause congestion.

[0111] According to this embodiment, a program that makes it possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the traffic congestion situation on a road can be implemented in software.

[0112] [Eighth embodiment] The eighth embodiment is a recording medium. This recording medium records a program for making a computer execute a dynamic network marker detection step S13 for detecting a dynamic network marker based on the distribution of traffic volume for each node and the correlation of traffic volume between the nodes, with multiple points on a road as nodes, and a switching step S14 for switching between determining that there is a sign of congestion at a node detected as a dynamic network marker, and detecting a dynamic network marker based on the distribution of traffic volume and the correlation of traffic volume between the nodes only for nodes previously stored as likely to cause congestion.

[0113] According to this embodiment, a program that makes it possible to avoid the enormous amount of calculation required to detect a sign of a sudden change in the traffic congestion situation on a road can be recorded on a recording medium.

[0114] [Ninth embodiment] 18 shows a functional block diagram of a non-illness state detection device 1 according to a ninth embodiment. The non-illness state detection device 1 is a device for detecting a non-illness state, which is a state in which a healthy state transitions to a diseased state. The non-illness state detection device 1 includes a data generation unit 10, a sample network calculation unit 20, a sample characteristic network calculation unit 30, a differential sample characteristic network calculation unit 40, a local network extraction unit 50, a gene expression probability calculation unit 60, a network flow entropy calculation unit 70, a differential network flow entropy calculation unit 80, and a temporal differential network flow entropy calculation unit 90.

[0115] The data generation unit 10 generates reference gene expression data generated from a biological sample collected from a group including multiple people, and test gene expression data by adding gene expression data of a biological sample collected from a subject at a specified time to the reference gene expression data.

[0116] FIG. 19 shows a schematic diagram of the operation of the data generating unit 10. FIG. 19(a) shows how reference gene expression data is generated from biological samples collected from a group of n healthy individuals. These biological samples relate to m types of genes. FIG. 19(b) shows how test gene expression data is generated by adding the gene expression data of a biological sample collected from a subject at time t=T to the reference gene expression data. Gene expression data is data that represents information on gene expression levels obtained from a biological sample such as blood. To measure genes from a biological sample, for example, a high-throughput technique may be used.

[0117] In this example, the reference gene expression data is an m-row, n-column matrix whose (i, j) component represents the expression level of the ith (i=1, ..., m) gene of the jth (j=1, ..., n) person. The test gene expression data is an m-row, n+1-column matrix whose (i, j) component represents the expression level of the ith (i=1, ..., m) gene of the jth (j=1, ..., n) person and whose (i, n+1) component represents the expression level of the ith (i=1, ..., m) gene of the subject.

[0118] The sample network calculation unit 20 calculates a reference sample network from the reference gene expression data, and also calculates a perturbation sample network from the test gene expression data.

[0119] Fig. 20 shows a schematic diagram of the operation of the sample network calculation unit 20. Fig. 20a shows how a reference sample network is calculated from reference gene expression data. Fig. 20b shows how a perturbation sample network at time t=T is calculated from test gene expression data.

[0120] In this example, the reference sample network is a weighted gene expression network (Weighted Reference Network: WRN) generated from reference gene expression data, and the perturbed sample network is a weighted gene expression network (Weighted Perturbed Network: WPN) generated from test gene expression data. T ).

[0121] The sample feature network calculation unit 30 calculates a sample feature network, where the sample feature network is a network that represents the relationship between the reference sample network and the perturbed sample network.

[0122] 21 is a schematic diagram showing the operation of the sample characteristic network calculation unit 30. The sample characteristic network calculation unit 30 calculates a reference sample network (WRN) and a perturbation sample network (WPN) at time t=T. T ) and the sample specific network (SSN) at time t = T. T )

[0123] In this example, the sample feature network (SSN T ) is, for example, a reference sample network (WRN) and a perturbed sample network (WPN T ) and keep only the edges that are different. T -WRN| is calculated, and the remaining different edges are called the sample characteristic network (SSN T ) can also be used.

[0124] The differential sample feature network calculation section 40 calculates a differential sample feature network from a sample feature network at a first time and a sample feature network at a second time after the first time.

[0125] 22 is a schematic diagram showing the operation of the differential sample characteristic network calculation unit 40. In this example, the differential sample characteristic network (SSDN) at t=T is calculated. T ) is the sample feature network (SSN) at t=T-1 (first time) T ) and the sample feature network (SSN) at t=T (second time) T-1 ) and keeping only the edges that are different.

[0126] The local network extraction unit 50 extracts a local network. Here, the local network is a network having a structure in which a differential gene, which is a gene whose expression level changes to a predetermined level or more, is placed at the center from the differential sample characteristic network, and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around the center.

[0127] 23 is a schematic diagram showing the operation of the local network extraction unit 50. k Each of (k=1,…,l) is a sample feature network (SSN T-1 ), where l is the number of genes. At the center of the local network, there is a differential gene g k is placed around the adjacent gene g i k (i=1,…,M k ) is placed. The node topological dissimilarity W ik is the central differential gene g k and the surrounding neighboring genes g i k It represents the strength of interaction with

[0128] The gene expression probability calculation unit 60 calculates the gene expression probability, which is the probability that each adjacent gene is expressed.

[0129] The gene expression probability may be calculated assuming that the expression of each neighboring gene follows a normal distribution.

[0130] For example, the mixed sample {s 1,…s n ,s caseT Gene x in} ikj The probability of occurrence of follows a normal distribution, and the probability distribution function f(x ikj ) can be expressed as

number

[0131] In this case, the cumulative distribution function F(x ikj ) is expressed as equation (13).

number

[0132] In this case, the gene expression probability p(x ikj ) is calculated as follows:

number

[0133] The network flow entropy calculation unit 70 calculates, for each neighboring gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value, based on the gene expression probability.

[0134] For example, the network flow entropy calculation unit 70 calculates the entropy of the local network N k Neighboring genes g in (k=1,…,l) i kFor the i-th gene g in the k-th local network, we may calculate the local network flow entropy and the conditional network flow entropy. i k Expression data of x ik Let x be the expression data across all genes in the k-th local network. k Let us assume that. In this case, the local network flow entropy NFE T (x ik )teeth,

number

number

number

number

[0135] A differential network flow entropy calculation unit 80 calculates the differential network flow entropy from the local network flow entropy and the conditional network flow entropy.

[0136] The differential network flow entropy DNFE can be calculated, for example, as follows: The differential network flow entropy at time t=T is the difference between the local network flow entropy and the conditional network flow entropy, i.e., NFE T (x ik )-NFE T (x ik / x k ) is defined as the number of adjacent genes gi k It can be considered as the difference between the local network flow entropy of and the conditional network flow entropy of . By averaging the differential network flow entropy over all neighboring genes, the differential network flow entropy DNFE T k can be calculated as follows:

number

[0137] Differential Network Flow Entropy DNFE T k is the local network N k Differential genes g in (k=1,…,l) k and the adjacent gene g i k This can be interpreted as mutual flow information between

[0138] A temporal differential network flow entropy calculation unit 90 calculates the temporal differential network flow entropy from the differential network flow entropy.

[0139] Temporal Differential Network Flow Entropy TNFE at time t=T T is a differential sample feature network (SSDN T ), it can be calculated as follows by averaging the mutual flow information of all local networks in

number

[0140] Temporal Differential Network Flow Entropy TNFE at time t=T TIf T changes suddenly, it can be considered that T is a branching point of state transition. In other words, the change in the temporal difference network flow entropy can detect a pre-disease state.

[0141] The non-illness state detection unit 100 detects a non-illness state when the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is equal to or greater than a predetermined value.

[0142] For example, the non-illness state detection unit 100 may detect a non-illness state if the temporal difference network flow entropy at the second time is at least twice the temporal difference network flow entropy at the first time.

[0143] By using network flow entropy as an index for detecting DNM, noise resistance can be improved compared to the conventional use of Pearson correlation coefficient or local network entropy. Furthermore, biological samples only need to be collected twice: once from a group including multiple people, and once from a subject at a specified time. Therefore, according to this embodiment, DNM detection that is highly resistant to noise and requires a small number of biological samples from the subject can be realized, thereby making it possible to detect a pre-disease state.

[0144] Below, examples of the present embodiment described above will be summarized.

[0145] [Example 1] The data generating unit 10 generates reference gene expression data in which the (i, j) component represents the expression level of the ith (i=1, ..., m) gene of the jth (j=1, ..., n) person, where n is the number of people, and m is the number of gene types. The data generating unit 10 generates test gene expression data in which the (i, j) component represents the expression level of the ith (i=1, ..., m) gene of the jth (j=1, ..., n) person, where m is the number of people, and m is the number of gene types. The data generating unit 10 generates test gene expression data in which the (i, j) component represents the expression level of the ith (i=1, ..., m) gene of the subject, where (i, n+1) component represents the expression level of the ith (i=1, ..., m) gene of the subject.

[0146] According to this embodiment, reference gene expression data and test gene expression data can be generated in a matrix format.

[0147] [Example 2] The sample network calculation unit 20 calculates a reference sample network as a weighted gene expression network generated from the reference gene expression data, and calculates a perturbed sample network as a weighted gene expression network generated from the test gene expression data.

[0148] According to this embodiment, the reference sample network and the perturbed sample network can be calculated accurately.

[0149] [Example 3] The sample feature network calculation unit 30 calculates the sample feature network by comparing the reference sample network with the perturbed sample network and keeping the different edges.

[0150] According to this embodiment, the sample feature network can be calculated accurately.

[0151] [Example 4] The differential sample feature network calculation unit 40 compares the sample feature network at the first time with the sample feature network at the second time, and calculates the differential sample feature network by retaining the different edges.

[0152] According to this embodiment, the differential sample feature network can be calculated accurately.

[0153] [Example 5] The gene expression probability calculation unit 60 calculates the gene expression probability by assuming that the expression of each adjacent gene follows a normal distribution.

[0154] According to this embodiment, the gene expression probability can be calculated accurately.

[0155] [Example 6] The gene expression probability calculation unit 60 calculates the i-th gene g i k The (ik, j) component of the expression data of x ikj Let μ ik and σ ik respectively for gene g i k When the mean and standard deviation are

number

[0156] According to this embodiment, the gene expression probability can be calculated more accurately.

[0157] [Example 7] The network flow entropy calculation unit 70 Let xik be the expression data of the i-th gene gik in the k-th local network, and xk be the expression data of all genes in the k-th local network.

number

number

[0158] According to this embodiment, the local network flow entropy and the conditional network flow entropy can be calculated accurately.

[0159] [Example 8] The number of adjacent genes is M kThen, the differential network flow entropy calculation unit 80 calculates

number

[0160] According to this embodiment, the differential network flow entropy can be calculated accurately.

[0161] [Example 9] When the number of local networks is l, the time difference network flow entropy calculation unit 90 calculates the following:

number

[0162] According to this embodiment, the temporal difference network flow entropy can be calculated accurately.

[0163] [Example 10] The non-illness state detection unit 100 detects a non-illness state if the temporal difference network flow entropy at the second time is equal to or greater than twice the temporal difference network flow entropy at the first time.

[0164] According to this embodiment, the pre-disease state can be detected more accurately.

[0165] [Tenth embodiment] 24 shows a flowchart of a non-disease state detection method according to the tenth embodiment. This non-disease state detection method detects a non-disease state, which is a transition state from a healthy state to a diseased state. The non-disease state detection method includes a data generation step S10, a sample network calculation step S20, a sample characteristic network calculation step S30, a differential sample characteristic network calculation step S40, a local network extraction step S50, a gene expression probability calculation step S60, a network flow entropy calculation step S70, a differential network flow entropy calculation step S80, and a temporal differential network flow entropy calculation step S90.

[0166] The data generation step S10 generates reference gene expression data generated from a biological sample collected from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from a subject at a specified time to the reference gene expression data.

[0167] In the sample network calculation step S20, a reference sample network is calculated from the reference gene expression data, and a perturbed sample network is calculated from the test gene expression data.

[0168] A sample feature network calculation step S30 calculates a sample feature network, where the sample feature network is a network that represents the relationship between the reference sample network and the perturbed sample network.

[0169] The differential sample feature network calculation step S40 calculates a differential sample feature network from the sample feature network at a first time and the sample feature network at a second time after the first time.

[0170] In the local network extraction step S50, a local network is extracted. Here, the local network is a network having a structure in which a differential gene, which is a gene whose expression level changes by a predetermined degree or more, is placed at the center from the differential sample characteristic network, and adjacent genes whose correlation with the differential gene increases sharply by a predetermined degree or more are arranged around the differential gene.

[0171] The gene expression probability calculation step S60 calculates the gene expression probability, which is the probability that each adjacent gene is expressed.

[0172] The network flow entropy calculation step S70 calculates, for each neighboring gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability.

[0173] A differential network flow entropy calculation step S80 calculates the differential network flow entropy from the local network flow entropy and the conditional network flow entropy.

[0174] A temporal differential network flow entropy calculation step S90 calculates the temporal differential network flow entropy from the differential network flow entropy.

[0175] In the non-illness state detection step S100, if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is equal to or greater than a predetermined value, it is detected that the patient is in a non-illness state.

[0176] According to this embodiment, DNM detection is realized that has high resistance to noise and requires a small number of samplings of biological material from the subject, making it possible to detect a pre-disease state.

[0177] [Eleventh embodiment] The eleventh embodiment is a program that causes a computer to execute a method for detecting a non-disease state, including the above-mentioned data generating step S10, sample network calculating step S20, sample characteristic network calculating step S30, differential sample characteristic network calculating step S40, local network extracting step S50, gene expression probability calculating step S60, network flow entropy calculating step S70, differential network flow entropy calculating step S80, and temporal differential network flow entropy calculating step S90.

[0178] The data generation step S10 generates reference gene expression data generated from a biological sample collected from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from a subject at a specified time to the reference gene expression data.

[0179] In the sample network calculation step S20, a reference sample network is calculated from the reference gene expression data, and a perturbed sample network is calculated from the test gene expression data.

[0180] A sample feature network calculation step S30 calculates a sample feature network, where the sample feature network is a network that represents the relationship between the reference sample network and the perturbed sample network.

[0181] The differential sample feature network calculation step S40 calculates a differential sample feature network from the sample feature network at a first time and the sample feature network at a second time after the first time.

[0182] The local network extraction step S50 extracts a local network. Here, the local network is a network having a structure in which adjacent genes whose correlation with differential genes rapidly increases to a predetermined level or more are arranged around differential genes, which are genes whose expression level changes by a predetermined level or more from the differential sample characteristic network.

[0183] The gene expression probability calculation step S60 calculates the gene expression probability, which is the probability that each adjacent gene is expressed.

[0184] The network flow entropy calculation step S70 calculates, for each adjacent gene, the local network flow entropy and the conditional network flow entropy under the condition that the expression level of the differential gene has a predetermined value, based on the gene expression probability.

[0185] The differential network flow entropy calculation step S80 calculates the differential network flow entropy from the local network flow entropy and the conditional network flow entropy.

[0186] The time-difference network flow entropy calculation step S90 calculates the time-difference network flow entropy from the differential network flow entropy.

[0187] The non-disease state detection step S100 detects a non-disease state if the difference between the time-difference network flow entropy at the first time and the time-difference network flow entropy at the second time is greater than or equal to a predetermined value.

[0188] According to the present embodiment, by realizing DNM detection with strong resistance to noise and a small recovery of biological substances from a subject, a program for detecting a non-disease state can be implemented in software.

[0189] [Embodiment 12] The twelfth embodiment is a recording medium. This recording medium records a program for making a computer execute a non-disease state detection method including the above-mentioned data generating step S10, sample network calculation step S20, sample characteristic network calculation step S30, differential sample characteristic network calculation step S40, local network extraction step S50, gene expression probability calculation step S60, network flow entropy calculation step S70, differential network flow entropy calculation step S80, and temporal differential network flow entropy calculation step S90.

[0190] The data generation step S10 generates reference gene expression data generated from a biological sample collected from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from a subject at a specified time to the reference gene expression data.

[0191] In the sample network calculation step S20, a reference sample network is calculated from the reference gene expression data, and a perturbed sample network is calculated from the test gene expression data.

[0192] A sample feature network calculation step S30 calculates a sample feature network, where the sample feature network is a network that represents the relationship between the reference sample network and the perturbed sample network.

[0193] The differential sample feature network calculation step S40 calculates a differential sample feature network from the sample feature network at a first time and the sample feature network at a second time after the first time.

[0194] In the local network extraction step S50, a local network is extracted. Here, the local network is a network having a structure in which a differential gene, which is a gene whose expression level changes by a predetermined degree or more, is placed at the center from the differential sample characteristic network, and adjacent genes whose correlation with the differential gene increases sharply by a predetermined degree or more are arranged around the differential gene.

[0195] The gene expression probability calculation step S60 calculates the gene expression probability, which is the probability that each adjacent gene is expressed.

[0196] The network flow entropy calculation step S70 calculates, for each neighboring gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability.

[0197] A differential network flow entropy calculation step S80 calculates the differential network flow entropy from the local network flow entropy and the conditional network flow entropy.

[0198] A temporal differential network flow entropy calculation step S90 calculates the temporal differential network flow entropy from the differential network flow entropy.

[0199] In the non-illness state detection step S100, if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is equal to or greater than a predetermined value, it is detected that the patient is in a non-illness state.

[0200] According to this embodiment, by realizing DNM detection that has high resistance to noise and requires a small number of times to collect biological material from a subject, a program for detecting a pre-disease state can be recorded on a recording medium.

[0201] [Thirteenth embodiment] 25 shows a functional block diagram of a pre-disease state detection device 4 according to a thirteenth embodiment. The pre-disease state detection device 4 includes a data generation unit 10 that generates reference gene expression data generated from biological samples collected from a subject at multiple past time points or biological samples collected from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculation unit 20 that calculates a reference sample network from the reference gene expression data and calculates a perturbed sample network from the test gene expression data, a sample characteristic network calculation unit 30 that calculates a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculation unit 40 that calculates a differential sample characteristic network from the sample characteristic network at a first time point and the sample characteristic network at a second time point after the first time point, and a gene whose expression amount changes to a predetermined degree or more from the differential sample characteristic network. the local network extraction unit 50 extracts a local network having a structure in which a differential gene is located at the center and adjacent genes whose correlation with the differential gene rises sharply to a predetermined level or more are arranged around it; a gene expression probability calculation unit 60 calculates a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation unit 70 calculates, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation unit 80 calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation unit 90 calculates a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection unit 100.

[0202] The pre-disease state detection unit 100 has a switching means 110 that switches between detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0203] According to this embodiment, it is possible to provide a device for detecting a pre-disease state by realizing DNM detection that has high resistance to noise and requires a small number of samplings of biological material from a subject.

[0204] [Fourteenth embodiment] FIG. 26 shows a flowchart of a method for detecting a pre-disease state according to a fourteenth embodiment. This method for detecting a pre-disease state is for detecting a pre-disease state, which is a state in which a healthy state transitions to a diseased state. The method for detecting a pre-disease state includes a data generating step S10 for generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step S20 for calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step S30 for calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step S40 for calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a step of calculating a differential sample characteristic network centered on differential genes, which are genes whose expression levels change to a predetermined degree or more, from the differential sample characteristic network. the local network extraction step S50 extracts a local network having a structure in which a local network has a structure in which adjacent genes whose correlation with the differential genes rises sharply to a predetermined level or more are arranged around the local network; a gene expression probability calculation step S60 calculates a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation step S70 calculates, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step S80 calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step S90 calculates a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step S100.

[0205] The pre-disease state detection step includes a switching step S110 for switching between detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0206] According to this embodiment, DNM detection is realized that has high resistance to noise and requires a small number of samplings of biological material from the subject, making it possible to detect a pre-disease state.

[0207] [Fifteenth embodiment] The fifteenth embodiment is a program. This program detects a pre-disease state, which is a state of transition from a healthy state to a diseased state. This program includes a data generation step S10 for generating reference gene expression data generated from a biological sample collected from a subject at multiple past times or from a group including multiple people, and test gene expression data by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculation step S20 for calculating a reference sample network from the reference gene expression data and calculating a perturbed sample network from the test gene expression data, a sample characteristic network calculation step S30 for calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculation step S40 for calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a differential sample characteristic network calculation step S50 for calculating a differential sample characteristic network from the differential sample characteristic network, focusing on differential genes, which are genes whose expression levels change by a predetermined degree or more. The computer is caused to execute a local network extraction step S50 for extracting a local network having a structure in which adjacent genes whose correlation with the differential gene rises sharply to a predetermined level or more are arranged around the local network; a gene expression probability calculation step S60 for calculating a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation step S70 for calculating, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step S80 for calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step S90 for calculating a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step S100.

[0208] The pre-disease state detection step includes a switching step S110 for switching between detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0209] According to this embodiment, by realizing DNM detection that has high resistance to noise and requires a small number of times to collect biological material from the subject, a program for detecting a pre-disease state can be implemented in software.

[0210] [Sixteenth embodiment] The sixteenth embodiment is a recording medium. This recording medium includes a data generating step S10 for generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple people, and test gene expression data for adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data, a sample network calculating step S20 for calculating a reference sample network from the reference gene expression data and a perturbed sample network from the test gene expression data, a sample characteristic network calculating step S30 for calculating a sample characteristic network representing the relationship between the reference sample network and the perturbed sample network, a differential sample characteristic network calculating step S40 for calculating a differential sample characteristic network from the sample characteristic network at a first time and the sample characteristic network at a second time after the first time, and a differential gene that is a gene whose expression amount changes by a predetermined degree or more from the differential sample characteristic network. The computer-readable storage device records a program for causing a computer to execute the following steps: a local network extraction step S50 for extracting a local network having a structure in which adjacent genes are arranged in which correlation with subsequent genes rises sharply to a predetermined level or more; a gene expression probability calculation step S60 for calculating a gene expression probability, which is the probability that each adjacent gene will be expressed; a network flow entropy calculation step S70 for calculating, for each adjacent gene, a local network flow entropy and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value based on the gene expression probability; a differential network flow entropy calculation step S80 for calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a temporal differential network flow entropy calculation step S90 for calculating a temporal differential network flow entropy from the differential network flow entropy; and a pre-disease state detection step S100.

[0211] The pre-disease state detection step includes a switching step S110 for switching between detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at a second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

[0212] According to this embodiment, by realizing DNM detection that has high resistance to noise and requires a small number of times to collect biological material from a subject, it is possible to record software that detects a pre-disease state.

[0213] The present invention has been described above based on the embodiments. These embodiments are merely examples, and it will be understood by those skilled in the art that various modifications are possible in the combination of each component and each treatment process, and that such modifications are also within the scope of the present invention.

[0214] [Variation 1] In the embodiment, the reference gene expression data is generated from a biological sample collected from a population including multiple individuals, but the reference gene expression data may also be generated from biological samples collected from the subject at multiple time points in the past.

[0215] [Variation 2] In the embodiment, information on gene expression levels obtained from a biological sample such as blood is used as gene expression data. However, the present invention is not limited to this, and in modified examples, various factors such as the type of disease and the purpose to be detected can be taken into consideration. In particular, various measurement data can be used as factor items, so long as it is information obtained by measuring a living body. For example, measurement data related to the aforementioned genes, proteins, and metabolites, or, without being limited to these, various conditions of each part can be quantified based on internal body images output by a measuring device such as a CT scan, and used as measurement data.

[0216] When understanding the technical ideas abstracted from the embodiments and modifications, the technical ideas should not be interpreted as being limited to the contents of the embodiments and modifications. The above-mentioned embodiments and modifications are merely illustrative examples, and many design modifications, such as changes, additions, and deletions of components, are possible. In the embodiments, the contents in which such design modifications are possible are emphasized by adding the notation "embodiment". However, design modifications are also permitted even in contents without such notation. [Industrial Applicability]

[0217] The technology disclosed herein can be used in fields such as early detection and treatment of diseases in the medical field, new drug development, and medical device development. [Explanation of symbols]

[0218] 1. Pre-disease state detection device, 2. System status sudden change sign detection device, 3. Traffic jam prediction device, 4. Pre-disease state detection device, 10. Data generation unit, 11...classification means, 12. Switching means, 13. Dynamic network marker detection means; 14. Switching means, 20. Sample network calculation section, 30 ··Sample characteristic network calculation part, 40··Differential sample characteristic network calculation unit, 50 Local network extraction unit, 60··Gene expression probability calculation part, 70 Network flow entropy calculation unit, 80 ··Differential network flow entropy calculation part, 90. Temporal difference network flow entropy calculation part, 100: Pre-disease state detection unit, 110...Switching means, S10: Data generation step; S11··Classification step, S12: Switching step, S11: Dynamic network marker detection step; S12: Switching step, S20: Sample network calculation step; S30: Sample characteristic network calculation step; S40: Differential sample characteristic network calculation step; S50: local network extraction step; S60: gene expression probability calculation step; S70: Network flow entropy calculation step; S80: Differential network flow entropy calculation step; S90: Temporal difference network flow entropy calculation step; S100: Pre-disease state detection step; S110··Switching step.

Claims

1. An apparatus for detecting a pre-disease state, which is a transition state from a healthy state to a disease state, comprising: a data generating unit that generates reference gene expression data generated from biological samples collected from a subject at multiple past time points or biological samples collected from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; a sample network calculation unit that calculates a reference sample network from the reference gene expression data and calculates a perturbation sample network from the test gene expression data; a sample feature network calculation unit for calculating a sample feature network representing a relationship between the reference sample network and the perturbed sample network; a differential sample characteristic network calculation unit that calculates a differential sample characteristic network from a sample characteristic network at a first time and a sample characteristic network at a second time that is later than the first time; a local network extraction unit that extracts from the differential sample characteristic network a local network having a structure in which a differential gene, which is a gene whose expression level changes to a predetermined level or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around the differential gene; a gene expression probability calculation unit that calculates a gene expression probability, which is the probability that each of the adjacent genes is expressed; a network flow entropy calculation unit for calculating a local network flow entropy for each of the neighboring genes based on the gene expression probability and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation unit that calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a time differential network flow entropy calculation unit that calculates a time differential network flow entropy from the differential network flow entropy; A non-illness state detection unit; Equipped with The non-disease state detection unit detects a non-disease state if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is greater than or equal to a predetermined value.

2. 2. The device for detecting a non-disease state according to claim 1, wherein the gene expression probability is calculated assuming that the expression of each of the adjacent genes follows a normal distribution.

3. When the number of the subjects at multiple points in the past or the number of the subjects is n and the number of gene types is m, The reference gene expression data is an m-row, n-column matrix whose (i, j) component represents the expression level of the i-th (i=1, ..., m) gene of the j-th (j=1, ..., n) subject or person at a past time point, The pre-disease state detection device according to claim 2, characterized in that the test gene expression data is a matrix with m rows and n+1 columns, in which the (i, j) component represents the expression level of the ith (i=1, ..., m) gene of the jth (j=1, ..., n) human subject at a past time point, and the (i, n+1) component represents the expression level of the ith (i=1, ..., m) gene of the subject.

4. the reference sample network is a weighted gene expression network generated from the reference gene expression data; The pre-disease state detection device according to claim 3 , wherein the perturbation sample network is a weighted gene expression network generated from the test gene expression data.

5. The device for detecting a non-disease state according to claim 4, wherein the sample feature network is calculated by comparing the reference sample network with the perturbed sample network and retaining different edges.

6. The pre-disease state detection device according to claim 5, characterized in that the differential sample feature network is calculated by comparing the sample feature network at the first time with the sample feature network at the second time and retaining the different edges.

7. Let x be the (ik, j) component of the expression data of the i-th gene gik in the k-th local network, and μ and σ be the mean and standard deviation of the gene gik, respectively. The gene expression probability p(xikj) is [0080] 7. The pre-disease state detection device according to claim 6,

8. When the expression data of the i-th gene g in the k-th local network is x, and the expression data of all genes in the k-th local network is x, The local network flow entropy NFET(xik) is [0050] The conditional network flow entropy NFET(x / x) is [0090] 8. The pre-disease state detection device according to claim 7,

9. When the number of adjacent genes is Mk, the differential network flow entropy DNFETk is expressed as follows: ##EQU00012## The pre-disease state detection device according to claim 8,

10. When the number of the local networks is l, the time difference network flow entropy T NFET is expressed as follows: ##EQU00013## The pre-disease state detection device according to claim 9,

11. The non-illness state detection device according to claim 10, characterized in that the non-illness state detection unit detects the non-illness state if the temporal difference network flow entropy at the second time is more than twice the temporal difference network flow entropy at the first time.

12. A method for detecting a pre-disease state, which is a state of transition from a healthy state to a diseased state, using a pre-disease state detection device including a data generation unit, a sample network calculation unit, a sample characteristic network calculation unit, a differential sample characteristic network calculation unit, a local network extraction unit, a gene expression probability calculation unit, a network flow entropy calculation unit, a differential network flow entropy calculation unit, a temporal differential network flow entropy calculation unit, and a pre-disease state detection unit, a data generating step of generating, by the data generating unit, reference gene expression data generated from biological samples collected from the subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; a sample network calculation step of calculating a reference sample network from the reference gene expression data and calculating a perturbation sample network from the test gene expression data by the sample network calculation unit; a sample characteristic network calculation step of calculating a sample characteristic network representing a relationship between the reference sample network and the perturbed sample network by the sample characteristic network calculation unit; a differential sample characteristic network calculation step of calculating a differential sample characteristic network from a sample characteristic network at a first time and a sample characteristic network at a second time after the first time by the differential sample characteristic network calculation unit; a local network extraction step of extracting, from the differential sample characteristic network, a local network having a structure in which a differential gene, which is a gene whose expression amount changes to a predetermined degree or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined degree or more are arranged around the differential gene by the local network extraction unit; a gene expression probability calculation step of calculating a gene expression probability, which is the probability that each of the adjacent genes is expressed, by the gene expression probability calculation unit; a network flow entropy calculation step of calculating, by the network flow entropy calculation unit, a local network flow entropy for each of the adjacent genes based on the gene expression probability, and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation step of calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy by the differential network flow entropy calculation unit; a time differential network flow entropy calculation step of calculating a time differential network flow entropy from the differential network flow entropy by the time differential network flow entropy calculation unit; A non-disease state detection step; Including, The non-disease state detection step is characterized in that the non-disease state detection unit detects that the patient is in a non-disease state if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is greater than or equal to a predetermined value.

13. A program for causing a computer to execute a method for detecting a pre-disease state, which is a transition state from a healthy state to a disease state, a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; A sample network calculation step of calculating a reference sample network from the reference gene expression data and calculating a perturbed sample network from the test gene expression data; a sample feature network calculation step of calculating a sample feature network representing a relationship between the reference sample network and the perturbed sample network; a differential sample feature network calculation step of calculating a differential sample feature network from a sample feature network at a first time and a sample feature network at a second time after the first time; a local network extraction step of extracting a local network from the differential sample characteristic network, the local network being structured such that a differential gene, which is a gene whose expression level changes to a predetermined degree or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined degree or more are arranged around the differential gene; a gene expression probability calculation step of calculating a gene expression probability, which is the probability that each of the adjacent genes is expressed; a network flow entropy calculation step of calculating a local network flow entropy for each of the neighboring genes based on the gene expression probability and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation step of calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a time differential network flow entropy calculation step of calculating a time differential network flow entropy from the differential network flow entropy; A non-disease state detection step; Including, The non-disease state detection step detects a non-disease state if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is greater than or equal to a predetermined value.

14. A non-transitory recording medium having a program recorded thereon for causing a computer to execute a method for detecting a non-illness state, which is a state in which a healthy state transitions to a diseased state, The program is a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; A sample network calculation step of calculating a reference sample network from the reference gene expression data and calculating a perturbed sample network from the test gene expression data; a sample feature network calculation step of calculating a sample feature network representing a relationship between the reference sample network and the perturbed sample network; a differential sample feature network calculation step of calculating a differential sample feature network from a sample feature network at a first time and a sample feature network at a second time after the first time; a local network extraction step of extracting a local network from the differential sample characteristic network, the local network being structured such that a differential gene, which is a gene whose expression level changes to a predetermined degree or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined degree or more are arranged around the differential gene; a gene expression probability calculation step of calculating a gene expression probability, which is the probability that each of the adjacent genes is expressed; a network flow entropy calculation step of calculating a local network flow entropy for each of the neighboring genes based on the gene expression probability and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation step of calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a time differential network flow entropy calculation step of calculating a time differential network flow entropy from the differential network flow entropy; A non-disease state detection step; Including, The recording medium is characterized in that the non-illness state detection step detects a non-illness state if a difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at the second time is greater than or equal to a predetermined value.

15. An apparatus for detecting a pre-disease state, which is a transition state from a healthy state to a disease state, comprising: a data generating unit that generates reference gene expression data generated from biological samples collected from a subject at multiple past time points or biological samples collected from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; a sample network calculation unit that calculates a reference sample network from the reference gene expression data and calculates a perturbation sample network from the test gene expression data; a sample feature network calculation unit for calculating a sample feature network representing a relationship between the reference sample network and the perturbed sample network; a differential sample characteristic network calculation unit that calculates a differential sample characteristic network from a sample characteristic network at a first time and a sample characteristic network at a second time after the first time; a local network extraction unit that extracts from the differential sample characteristic network a local network having a structure in which a differential gene, which is a gene whose expression level changes to a predetermined level or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around the differential gene; a gene expression probability calculation unit that calculates a gene expression probability, which is the probability that each of the adjacent genes is expressed; a network flow entropy calculation unit for calculating a local network flow entropy for each of the neighboring genes based on the gene expression probability and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation unit that calculates a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a time differential network flow entropy calculation unit that calculates a time differential network flow entropy from the differential network flow entropy; A non-illness state detection unit; Equipped with The pre-disease state detection unit is characterized in having a switching means for switching between detecting a pre-disease state if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

16. A method for detecting a pre-disease state, which is a state of transition from a healthy state to a diseased state, using a pre-disease state detection device including a data generation unit, a sample network calculation unit, a sample characteristic network calculation unit, a differential sample characteristic network calculation unit, a local network extraction unit, a gene expression probability calculation unit, a network flow entropy calculation unit, a differential network flow entropy calculation unit, a temporal differential network flow entropy calculation unit, and a pre-disease state detection unit, a data generating step of generating, by the data generating unit, reference gene expression data generated from biological samples collected from the subject at multiple past times or from a group including multiple people, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; a sample network calculation step of calculating a reference sample network from the reference gene expression data and calculating a perturbation sample network from the test gene expression data by the sample network calculation unit; a sample characteristic network calculation step of calculating a sample characteristic network representing a relationship between the reference sample network and the perturbed sample network by the sample characteristic network calculation unit; a differential sample characteristic network calculation step of calculating a differential sample characteristic network from a sample characteristic network at a first time and a sample characteristic network at a second time after the first time by the differential sample characteristic network calculation unit; a local network extraction step of extracting, from the differential sample characteristic network, a local network having a structure in which a differential gene, which is a gene whose expression amount changes to a predetermined degree or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined degree or more are arranged around the differential gene by the local network extraction unit; a gene expression probability calculation step of calculating a gene expression probability, which is the probability that each of the adjacent genes is expressed, by the gene expression probability calculation unit; a network flow entropy calculation step of calculating, by the network flow entropy calculation unit, a local network flow entropy for each of the adjacent genes based on the gene expression probability, and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation step of calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy by the differential network flow entropy calculation unit; a time differential network flow entropy calculation step of calculating a time differential network flow entropy from the differential network flow entropy by the time differential network flow entropy calculation unit; A non-disease state detection step; Including, The pre-disease state detection step is characterized in that it includes a switching step in which the pre-disease state detection unit switches between detecting a pre-disease state if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is a predetermined value or more, or detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

17. A program for causing a computer to execute a method for detecting a pre-disease state, which is a transition state from a healthy state to a disease state, a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; A sample network calculation step of calculating a reference sample network from the reference gene expression data and calculating a perturbed sample network from the test gene expression data; a sample feature network calculation step of calculating a sample feature network representing a relationship between the reference sample network and the perturbed sample network; a differential sample feature network calculation step of calculating a differential sample feature network from a sample feature network at a first time and a sample feature network at a second time after the first time; a local network extraction step of extracting a local network from the differential sample characteristic network, the local network being structured such that a differential gene, which is a gene whose expression level changes to a predetermined degree or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined degree or more are arranged around the differential gene; a gene expression probability calculation step of calculating a gene expression probability, which is the probability that each of the adjacent genes is expressed; a network flow entropy calculation step of calculating a local network flow entropy for each of the neighboring genes based on the gene expression probability and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation step of calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a time differential network flow entropy calculation step of calculating a time differential network flow entropy from the differential network flow entropy; A non-disease state detection step; Including, The non-disease state detection step is characterized in that it includes a switching step of detecting a non-disease state if the difference between the temporal difference network flow entropy at the first time and the temporal difference network flow entropy at the second time is a predetermined value or more, or detecting a non-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

18. A non-transitory recording medium having a program recorded thereon for causing a computer to execute a method for detecting a non-illness state, which is a state in which a healthy state transitions to a diseased state, The program is a data generating step of generating reference gene expression data generated from biological samples collected from a subject at multiple past times or from a group including multiple individuals, and test gene expression data obtained by adding gene expression data of a biological sample collected from the subject at a predetermined time to the reference gene expression data; A sample network calculation step of calculating a reference sample network from the reference gene expression data and calculating a perturbed sample network from the test gene expression data; a sample feature network calculation step of calculating a sample feature network representing a relationship between the reference sample network and the perturbed sample network; a differential sample feature network calculation step of calculating a differential sample feature network from a sample feature network at a first time and a sample feature network at a second time after the first time; a local network extraction step of extracting a local network from the differential sample characteristic network, the local network being structured such that a differential gene, which is a gene whose expression level changes to a predetermined degree or more, is placed at the center and adjacent genes whose correlation with the differential gene increases sharply to a predetermined degree or more are arranged around the differential gene; a gene expression probability calculation step of calculating a gene expression probability, which is the probability that each of the adjacent genes is expressed; a network flow entropy calculation step of calculating a local network flow entropy for each of the neighboring genes based on the gene expression probability and a conditional network flow entropy under the condition that the expression amount of the differential gene has a predetermined value; a differential network flow entropy calculation step of calculating a differential network flow entropy from the local network flow entropy and the conditional network flow entropy; a time differential network flow entropy calculation step of calculating a time differential network flow entropy from the differential network flow entropy; A non-disease state detection step; Including, The recording medium is characterized in that the pre-disease state detection step includes a switching step of switching between detecting a pre-disease state if the difference between the temporal difference network flow entropy at a first time and the temporal difference network flow entropy at the second time is a predetermined value or more, and detecting a pre-disease state using a local network having a structure in which a pre-stored differential gene is centered around it and adjacent genes whose correlation with the differential gene increases sharply to a predetermined level or more are arranged around it.

Images