1. No category

動的ネットワークバイオマーカーによる
疾病早期診断技術
「未病」の検出と創薬標的発見
東京大学生産技術研究所
客員教授 陳洛南
1
（１）発明の名称：動的ネットワークバイオマーカー
の検出装置、検出方法及び検出プログラム
• 公開番号: 2014‐064515
• 出願人：科学技術振興機構
• 発明者：陳洛南、合原一幸、劉鋭、劉治平、李美儀
（２）発明の名称：ネットワークエントロピーに基づ
く生体の状態遷移の予兆の検出を支援する検出装置、
検出方法及び検出プログラム
• 公開番号: 2014‐083194
• 出願人：科学技術振興機構
• 発明者: 合原一幸、陳洛南、劉鋭
2
Once disease occurs, difficult to be cured
《黄帝内经》紀元前221年
《Yellow Emperor's Medicine》 221 BC
上医治未病，中医治欲病，下医治已病
the best doctor
treats
unoccurred disease
the better doctor
treats
occurring disease
定性的な概念
→
the inferior doctor
treats
occurred disease.
定量的な指標
Early diagnosis by dynamical network biomarkers
3
特許
動的ネットワークバイオ
マーカーによる定量的に
疾患
「予兆」検出
「未病」診断
創薬 標的
個別化医療
4
臨界状態
正常状态
Normal state
(未病状态:reversible)
前疾病状态
Pre-disease state
(已病状态:irreversible)
疾病状态
Disease state
Early-disease state
疾病早期状态
Disease progression
5
Conceptual illustration from mathematics viewpoint
Charaterizing the limit
1. Bifurcation
• Z(t)=(z1(t),…,zn(t)) are state, P are parameters. • The stability of one equilibrium will change from stable to unstable at bifurcation point Pc .
Normal
Pc
Bifurcation Point
Chen, et al., Scientific Reports, 2, 342, 2012
Critical Transition
(Catastrophy)
Disease
Disease
6
Catastrophic bifurcation
Conceptual illustration from physics viewpoint
Landscape
2. Potential Energy Function
Small
Normal
Disease
Critical
Critical slowing down
Critical
Deviation
Progression
Normal
Large
Disease
Disease
Small
Disease
State
Disorder
7
Conceptual illustration from information-science viewpoint
3. Information Entropy (Disorder)
(Reprogramming)
(Disorder)
Disease progression
disease state
Liu, et al., Scientific Reports, 2, 813, 2012
8
Critical Question?
Can we make early diagnosis?
on pre-disease state
• Constraint: a small number of samples
• Requirement : a large number of observations
e.g. 3 samples for a person
(e.g. microarray data)
( or proteomics data)
9
Observation Data
• High throughput data Z: a few samples but a large number of outputs
e.g. 5 samples for a person
• For example: Genomics and Proteomics Data for control cases (or normal cases), critical cases (pre‐
disease cases), and disease cases. 10
Mathematical foundation
System
Driving Factors (unknown)
Parameters P
Slowly changing factors, e.g. genetic (SNP,CNV), epigenetic (methylation, acetylation) factors
Responses (observable)
State Variables Z
e.g. gene expression, protein expression
The progression of a disease is considered as the evolution of a nonlinear dynamical system.
i.e., identify the leading network which makes the first move into the disease state. 11
Main idea
Mathematical Scheme
Transform
Global network in original space
(observable Z)
Transform
center manifold Subnetwork or DNB in original space
(unobservable Y)
(observable Z)
Generally, one or two dimensions in Surprisingly !
Complicated equations usually in very high dimensional space
Dynamics is constrained to 1 or 2 dimensional abstract space
Complicated equations usually in very high dimensional space
Y(k+1) = g(Y(k); P)
Limit of normal
Before critical period (Normal state)
Critical period (Pre‐disease state)
Disease progression
After critical period (Disease state)
12
Conceptual Illustration
Features of the limit
Center Manifold
• There exists 1‐D center manifold before a generic bifurcation for any systems, regardless of their differences. • We can detect the signal whenever the system approaches the center manifold before the bifurcation. Center
manifold
Before bifurcation
Near bifurcation
After bifurcation
All roads lead to Rome
13
All dynamics converge to the center manifold !
Conceptual Illustration
Analysis on State Space
(from 1‐D center manifold to network)
Y1
Z
Map to the subnetwork
1-D center manifold
(Unobservable Y1)
Network
(Observable Z)
14
Main Theorem
1. If both i and j are in dominant group, Pearson Correlation Standard Deviation 2. If only i is in dominant group, Pearson Correlation Standard Deviation 3. If neither i or j is in dominant group, Pearson Correlation Standard Deviation Three Measurable Conditions
Critical State 15
It always holds provided that a system is becoming increasingly sensitive to external perturbations, regardless of bifurcations.
Dynamical Network Biomarker (DNB)
DNB  創薬標的
DNB
The leading
Network
16
DNBの条件
• DNBは従来のバイオマーカーと異なり、正常状態と病気状態
を判別するのではなく、正常状態とその極限、即ち、臨界状
態或いは病気前期をはっきりと識別することを目的とする。
• 次の三つの条件を満たす遺伝子或いは分子のグループがDNB
となる。
（１）そのグループの分子の標準偏差が急激的に上昇する；
（２）そのグループの分子間の相関性が急激に上昇する；
（３）そのグループの分子と他の分子との間の相関性が急激に
降下する。
この三つの条件を満たすDNBが存在することが、測定された人
がこの期間において病気早期状態にあることのシグナルとなる
。
17
• Static feature : DNA
From DNA to DNB
• Dynamic feature : DNB
18
Birds
Birds
Social Phenomena
The essential feature of a phase transition is a structural reordering, i.e., long range order – everything gets connected so that things can be reconnected in a different way ‐‐
Collapse of fish stocks, stock market, ice melting to water, diamonds to graphite, revolution
Herding behavior Connectivity Avalanche Critical slowing down
(network science)
(physics science)
(social science)
Fish
Multi-variables, Network
Internet
19
Single-variables, Dynamics
Herding behavior, Connectivity Avalanche, Critical Slowing-Down
Strongly Correlative Fluctuations
(DNB Theory: fluctuated local herding behavior)
Unified Signal
Dynamical Network Marker
Fluctuation
Resonance
Catastrophe
Revolution
Collapse of regime
20
Signal not from statistics but from dynamics
Normal state
Pre-disease
state
Disease state
Disease
progression
It cannot
be detected
by GWAS even with a large
number of samples
Information not from average values but individual variations
21
Composite Indicator based on DNB
PCCd : average PCC of DNB in absolute value
PCCo : average PCC between DNB and others in absolute value
SDd : average SD of DNB
Based on the three conditions
22
Critical Question?
Can we make early diagnosis?
on pre-disease state
• condition: a small number of samples but large data
3 consecutive samples for a person
YES!
(e.g. microarray data)
( or proteomics data)
Not one day in a year but three-consecutive days in a year for physical check
23
Model‐Free Algorithm for DNB
Choose genes with high deviations
Cluster genes at each period
Deviation Test
(Step-1)
Correlation Test
(Step-2)
Determine DNB by three conditions
Functional analysis on DNB
Significance Test
(Step-3)
Functional Test
(Step-4)
24
Three Diseases
• Acute Lung Injury
Mouse
• Liver Cancer (HBV, HCV)
Human
• Lymph Cancer
Human
More examples: Diabetes - Rat ; Influenza - Human
Chen, et al., Scientific Reports, 2, 342, 2012
Liu, et al., Scientific Reports, 2, 813, 2012
Liu, et al., Medicinal Research Reviews, 2013
25
26
Results
in DNB
in DNB
between DNB and others
27
composite
Dynamical changes of whole mouse network (3452 genes and 9238 links) including
DNB during disease progression for lung injury (color: not expression but our criterion)
Even with strong noise on individual
variables, DNB can robustly detect the
signal due to a group of strongly
collective fluctuations.
Lung injury
Normal Lung
Pre-disease
28
The leading network to the disease
Dynamical changes in whole PPI network (2291 genes) including the leading network
during disease progression for HCC (Color: not expression but our criterion)
Pre-disease
metastasis
29
Human HCV-Induce Liver Cancer
Human HBV‐Induce Liver Cancer
(Dynamical Network Biomarker)
30
Clinic symptomatic or asymptomatic subjects in influenza dataset (H3N2/Wisconsin)
17 subjects in live influenza experiment
N
Symptom
(disease state)
S
S1
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S5
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S6
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S7
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S8
N
N
N
N
N
N
N
N
N
N
S
S
S
S
S
S
S10
N
N
N
N
N
N
N
N
N
N
N
S
S
S
S
S
S12
N
N
N
N
N
N
N
N
N
N
N
N
N
S
S
S
S13
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S
S
S15
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S
S
S2
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S3
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S4
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S9
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S11
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S14
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S16
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S17
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
Hour
‐24
0
5
12
21
29
36
45
53
60
69
77
84
93
101
108
Asymptomatic subjects Symptomatic subjects
Non-symptom
(normal state or
pre-disease state)
Time
31
Clinical data
*
*
*
32
Predict the pre‐disease state before the disease
33
Prediction
Clinic symptomatic or asymptomatic subjects in influenza dataset (H3N2/Wisconsin)
Non-symptom
(normal state or
pre-disease state)
N
17 subjects in live influenza experiment
(pre-disease state)
Symptom
(disease state)
S
S1
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S5
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S6
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S7
N
N
N
N
N
N
N
S
S
S
S
S
S
S
S
S
S8
N
N
N
N
N
N
N
N
N
N
S
S
S
S
S
S
S10
N
N
N
N
N
N
N
N
N
N
N
S
S
S
S
S
S12
N
N
N
N
N
N
N
N
N
N
N
N
N
S
S
S
S13
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S
S
S15
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S
S
S2
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S3
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S4
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S9
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S11
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S14
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S16
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
S17
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
N
Hour
‐24
0
5
12
21
29
36
45
53
60
69
77
84
93
101
108
Asymptomatic subjects Symptomatic subjects
N
Early diagnosis
before disease
Time
34
Diabetes Rats (microarray data)
8wk
4wk
12wk
16wk
Beta 细胞损伤
Beta Cell Failure
胰岛素抵抗
Insulin Resistance
5 GK
5 GK
20wk
5 GK
5 GK
5 GK
糖尿病鼠
5 Wistar
5 Wistar
5 Wistar
5 Wistar
5 Wistar
正常鼠
Liver
Tissues Genes
10729
Muscle
10729
Adipose
10729
３tissues，5 stages ， 25 GK，25 Wister
（150 samples）
35
DNB for each tissue of GK rats
36
Concentrations of plasma glucose and insulin of GK and WKY rats
insulin resistance
The first
Critical point
The second
Critical point
β-cell failure
The first
Critical point
The second
Critical point
Consistent with the phenotypes !!!
37
What is the
relation between DNB and DEG
(DEG: Differential Expressed Genes)
动态网络标志物
差异表达基因
There are few overlaps of genes, but large overlaps of pathways
(DNB and DEG are enriched in the same pathways !)
DEG is the result of the disease, but DNB is its cause
38
Relation
between DNB
and DEG
!
Steroid hormone biosynthesis – L1
类固醇激素合成
DNB: Upstream
!
DEG: Downstream !
Members of DNB
stay important
positions to regulate
the downstream
1. Upstream
2. As messengers or receptors
3. As a part of the combined
complex
DNB
over-expressed genes
under-expressed genes
White and Green: background
39
Chen, et al., Scientific Reports, 2, 342, 2012
Liu, et al., Scientific Reports, 2, 813, 2012
Liu, et al., Medicinal Research Reviews, 2013
Dynamic Theory Result in Statistic Theory
Model Free Method
40
Liu, et al., Medicinal Research Reviews, 2013
Biomarkers
Molecular biomarker score
Diseased person
Molecular Biomarker
Distinguish disease from normal state
Static
signal
19XX
Normal person
Time
Network biomarker score
Network Biomarker
Dynamical Network Biomarker
Distinguish disease from normal state
Diseased person
Static
signal
2008
Our work
Normal person
Time
Dynamical network biomarker score
Distinguish pre‐disease
from normal state
Pre-diseased person
2012
Dynamic
signal
Normal person
(Critical Transition, Leading Network, Early Diagnosis)
Our work
Time
41
Chen, et al., Scientific Reports, 2012
;
Liu, et al., Scientific Reports, 2012
Conclusion
• A model free approach to identify early‐signal before critical transitions by small samples of big data
• It can identify the causal network to disease state
• It can be applied to other biological processes
• DNB distinguishes pre‐disease and normal state, although little difference in terms of static features exists between them. • DNB detects pre‐disease state in a dynamical manner, in contrast to static traditional biomarkers
• Handle personal variation (e.g. genetic or epigenetic factors). Each individual may progress to the same disease through different networks (personalized medicine)
42
応用分野
医療、製薬、ヘルスケア分野にて、ビッグデータを
測定し、DNB検出・解析により、
・疾患の超早期診断、病態悪化の予兆検出
・個別疾患、個人対応の個別医療の実現
・創薬的標的の発見
・定期検査、健康診断などでシーケンス、タンパク
質、代謝物のデータより癌、心臓病、糖尿病など
の疾患、医療画像のデータより脳疾患などの超
早期診断
43
実用化に向けた課題
･ 現在、DNBについて理論とその応用枠組みが
可能なところまで開発済み。しかし、個々の病気
の特徴をいかに取り込むかの点が未解決である。
・今後、臨床或いは実験データを取得し、個々の
疾患における早期診断、病態悪化の予兆検出
の実証、実績を積み重ねていく必要がある。
44
企業への期待
･ 個々の疾患について、臨床或いは実験データ
を測定し、そのDNBを開発することについて、企
業との共同研究を希望。
・また、製薬とヘルスケア分野への展開を考えて
いる企業には、薬の標的の同定と健康の定量的
測定について、本技術の導入が有効と思われる。
45
お問い合わせ先
• コーディネータ 科学技術振興機構 知的
財産戦略センター 保護・活用グループ
•
•
•
五月女 勲
TEL ０３－５２１４－８４８６
FAX ０３－５２１４－８４１７
e‐mail: [email protected]
46
Thanks
47