Minimum Dominating Set Approach to Analysis and Control of Biological Networks

Similar documents
Chapter 6 Sampling Distributions

Problem Set 2 Solutions

CS322: Network Analysis. Problem Set 2 - Fall 2009

Information-based Feature Selection

Random assignment with integer costs

Convergence of random variables. (telegram style notes) P.J.C. Spreij

Analysis of Algorithms. Introduction. Contents

CS284A: Representations and Algorithms in Molecular Biology

Rank Modulation with Multiplicity

Large holes in quasi-random graphs

4. Partial Sums and the Central Limit Theorem

Math 2784 (or 2794W) University of Connecticut

Section 11.8: Power Series

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 2 9/9/2013. Large Deviations for i.i.d. Random Variables

INFINITE SEQUENCES AND SERIES

A statistical method to determine sample size to estimate characteristic value of soil parameters

Optimization Methods: Linear Programming Applications Assignment Problem 1. Module 4 Lecture Notes 3. Assignment Problem

Chapter 10: Power Series

CEE 522 Autumn Uncertainty Concepts for Geotechnical Engineering

Chapter 8: STATISTICAL INTERVALS FOR A SINGLE SAMPLE. Part 3: Summary of CI for µ Confidence Interval for a Population Proportion p

A Hadamard-type lower bound for symmetric diagonally dominant positive matrices

Dense H-free graphs are almost (χ(h) 1)-partite

if j is a neighbor of i,

Approximations and more PMFs and PDFs

The coalescent coalescence theory

NYU Center for Data Science: DS-GA 1003 Machine Learning and Computational Statistics (Spring 2018)

ACO Comprehensive Exam 9 October 2007 Student code A. 1. Graph Theory

ENGI 4421 Confidence Intervals (Two Samples) Page 12-01

Lecture 9: Hierarchy Theorems

Advanced Analysis. Min Yan Department of Mathematics Hong Kong University of Science and Technology

ECONOMETRIC THEORY. MODULE XIII Lecture - 34 Asymptotic Theory and Stochastic Regressors

Discrete Mathematics for CS Spring 2008 David Wagner Note 22

CUMULATIVE DAMAGE ESTIMATION USING WAVELET TRANSFORM OF STRUCTURAL RESPONSE

AN INTRODUCTION TO SPECTRAL GRAPH THEORY

The DOA Estimation of Multiple Signals based on Weighting MUSIC Algorithm

18.440, March 9, Stirling s formula

The Growth of Functions. Theoretical Supplement

Lecture 14: Graph Entropy

Singular Continuous Measures by Michael Pejic 5/14/10

Beurling Integers: Part 2

Topic 9: Sampling Distributions of Estimators

Sensitivity Analysis of Daubechies 4 Wavelet Coefficients for Reduction of Reconstructed Image Error

Generalized Semi- Markov Processes (GSMP)

Topic 9: Sampling Distributions of Estimators

1 6 = 1 6 = + Factorials and Euler s Gamma function

Lecture 2. The Lovász Local Lemma

Abstract. Keywords: conjecture; divisor function; divisor summatory function; prime numbers; Dirichlet's divisor problem

EE / EEE SAMPLE STUDY MATERIAL. GATE, IES & PSUs Signal System. Electrical Engineering. Postal Correspondence Course

A Note on Matrix Rigidity

Lecture 20: Multivariate convergence and the Central Limit Theorem

is also known as the general term of the sequence

w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ.

BIOS 4110: Introduction to Biostatistics. Breheny. Lab #9

Lecture 19: Convergence

Chapter 4 : Laplace Transform

CHAPTER 2 NEIGHBORHOOD CONNECTED PERFECT DOMINATION IN GRAPHS

Math 61CM - Solutions to homework 3

Machine Learning Brett Bernstein

Chapter 11 Output Analysis for a Single Model. Banks, Carson, Nelson & Nicol Discrete-Event System Simulation

Fastest mixing Markov chain on a path

Example 2. Find the upper bound for the remainder for the approximation from Example 1.

TR/46 OCTOBER THE ZEROS OF PARTIAL SUMS OF A MACLAURIN EXPANSION A. TALBOT

Information Theory Tutorial Communication over Channels with memory. Chi Zhang Department of Electrical Engineering University of Notre Dame

11 THE GMM ESTIMATION

Sequences A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

THE SYSTEMATIC AND THE RANDOM. ERRORS - DUE TO ELEMENT TOLERANCES OF ELECTRICAL NETWORKS

Statistics Lecture 27. Final review. Administrative Notes. Outline. Experiments. Sampling and Surveys. Administrative Notes

Introduction to Signals and Systems, Part V: Lecture Summary

REGRESSION WITH QUADRATIC LOSS

Discrete-Time Systems, LTI Systems, and Discrete-Time Convolution

ECE 330:541, Stochastic Signals and Systems Lecture Notes on Limit Theorems from Probability Fall 2002

1 Introduction to reducing variance in Monte Carlo simulations

Regression with an Evaporating Logarithmic Trend

Sample Size Determination (Two or More Samples)

Linear Regression Analysis. Analysis of paired data and using a given value of one variable to predict the value of the other

Topic 9: Sampling Distributions of Estimators

ME 539, Fall 2008: Learning-Based Control

Lecture 12: November 13, 2018

University of California, Los Angeles Department of Statistics. Hypothesis testing

The Choquet Integral with Respect to Fuzzy-Valued Set Functions

Infinite Sequences and Series

Week 1, Lecture 2. Neural Network Basics. Announcements: HW 1 Due on 10/8 Data sets for HW 1 are online Project selection 10/11. Suggested reading :

NEW FAST CONVERGENT SEQUENCES OF EULER-MASCHERONI TYPE

A proposed discrete distribution for the statistical modeling of

4.3 Growth Rates of Solutions to Recurrences

5. INEQUALITIES, LIMIT THEOREMS AND GEOMETRIC PROBABILITY

Daniel Lee Muhammad Naeem Chingyu Hsu

Random Variables, Sampling and Estimation

SRC Technical Note June 17, Tight Thresholds for The Pure Literal Rule. Michael Mitzenmacher. d i g i t a l

1 Introduction: within and beyond the normal approximation

ORIE 633 Network Flows September 27, Lecture 8

Filter banks. Separately, the lowpass and highpass filters are not invertible. removes the highest frequency 1/ 2and

Introduction to Extreme Value Theory Laurens de Haan, ISM Japan, Erasmus University Rotterdam, NL University of Lisbon, PT

Continuous Data that can take on any real number (time/length) based on sample data. Categorical data can only be named or categorised

SAMPLING LIPSCHITZ CONTINUOUS DENSITIES. 1. Introduction

Information Theory and Statistics Lecture 4: Lempel-Ziv code

Ma 530 Introduction to Power Series

As stated by Laplace, Probability is common sense reduced to calculation.

Application to Random Graphs

Zeros of Polynomials

Transcription:

Miimum Domiatig Set Approach to Aalysis ad Cotrol of Biological Networks Tatsuya Akutsu Bioiformatics Ceter Istitute for Chemical Research, Kyoto Uiversity Joit work with Jose Nacher i Toho Uiversity

Motivatio: Cotrol Theory for Biological Systems Oe of the mai targets of Systems Biology Though cotrol theory is well established for liear systems, biological systems have o-liear compoets ad are very complex (large-scale) May lead to ew drugs ad treatmet methods Practical cotrol methods exist, but o useful theory Itroductio of 4 gees turs ormal cells ito iduced pluripotet stem cells (ips cells) Abormal がん細胞 Cell Cotrol 制御 Normal 正常細胞 Cell

Cotets Scale-free Networks Cotrollability i Scale-free Networks Miimum Domiatig Set (MDS) Relatio to Structural Cotrollability Theoretical Aalysis of MDS Size Computer Simulatio Database Aalysis Applicatios to Aalysis of Biological Networks Extesios Coclusio

Scale-free Networks

Scale-Free Network [Barabasi & Albert, 999] Degree of a ode The umber of adjacet odes P(k) Degree distributio Frequecy of odes with degree k Scale-free etwork P(k) follows power law Differet from radom etworks degree=2 degree=5 P ( k) degree =3 k

Metabolic Network, Graph ad Degree A B C D E F G H I J Degree Node with degree : J Nodes with degree 2: B, C, D, F, G, H Nodes with degree 3: A,E, I P(k) (degree distributio): P()=., P(2)=.6, P(3)=.3, P(4)=P(5)=P(6)= =

Scale-Free Distributio P ( k) k Power laws are scale free because if k is rescaled (multiplied by a costat), the P(k) is still proportioal to k May real etworks (e.g., geetic etworks, metabolic etworks, protei-protei iteractio etworks) are reported to have the scale-free property

Poisso Distributio ad Power-Law Distributio Poisso distributio (radom graph) Power-law distributio (scale-free graph) e -λ λ k /k! k - P (k) log P (k) k log(k)

Cotrollability i Scale-free Networks [Liu, Slotie, Barabasi: Nature, 2]

Cotrollability of Liear Systems () Iput: Liear System: Iitial state: x Fial state: x F Output: u(t) (fuctio of t) which drives the system from x to x F i fiite time ) ( ) ( ) ( t B t A dt t d u x x + = + = M N N u u u B x x x A x x x 2 2 2 x(t): N-dim. real vector (iteral odes) u(t): M-dim. real vector (cotrol odes) A: N N real matrix B: N M real matrx

Cotrollability of Liear Systems (2) Fact. System is cotrollable iff N NM matrix C=(B,AB,A 2 B,,A N- B) has full rak (i.e., rak(c)=n). = = 22 4 34 3 2 b b B a a a a A = 4 4 34 3 2 22 b a b a a b a b a b b C rak(c)=4 for most parameters structural cotrollability

Structural Cotrollability G B (V L,V R ;E B ) : bipartite graph costructed from G(V,E) by B R j L i j i i R i R i L i L E x x E x x V x x V V x x V = = ), ( ), ( }, { }, { Thm. [Liu et al. 2] The miimum umber of odes eeded to fully cotrol the system is max {N-M *,}, where M * is the size of the maximum matchig of G B.

Cotrollability of Scale-free Networks The umber of eeded driver odes [Liu et al. 2] Radom etworks: N D Scale-free etworks e < k >/ 2 [ ( ) < > ] exp k N D 2 if <2, may odes must be cotrolled <k>: average degree : umber of odes i a etwork

Miimum Domiatig Set ad Its Relatio to Structural Cotrollability

Miimum Domiatig Set () V D is a domiatig set of udirected graph G(V,E) ( v V-V D )( u V D )({u,v} E) Miimum domiatig set: domiatig set with the smallest umber of odes

Miimum Domiatig Set (2) Well-kow cocept i graph theory ad computer sciece NP-hard, but ca be solved exactly by usig Iteger Liear Programmig (ILP) to some extet Has bee applied to desig/cotrol of mobile ad-hoc etworks (MANET) trasportatio routig computer commuicatio etworks

Relatio betwee MDS ad Cotrollability MDS Thm. Suppose that every edge i a etwork is bi-directioal ad every ode i MDS ca cotrol all of its outgoig liks separately. The, the etwork is structurally cotrollable by selectig the odes i MDS as the driver odes. [Nacher & Akutsu: New J. Phys. 22]

ILP-based Method for MDS Very simple, but works for etworks with a few thousads of odes i may cases {,},,,.. mi } }, { { = = = i E v v i j j j i i x i x s t x j i x i = x i i MDS

Theoretical Aalysis of MDS Size

Estimatio of MDS Size i Scale-free Networks >2 Upper boud: trivially O() Lower boud: Ω() <2 Upper boud: O( -(2-)(-) ) takig the miimum order O(.75 ) whe =.5 Based o a kid of mea-field approximatio [Nacher & Akutsu: J. Phys.: Cof. Ser. 23]

Lower Boud for >2 () Assumig αk -, we have The followig is well kow, where C(S) is the set of edges betwee S ad V-S if S + C(S) <, S is ot a domiatig set If we select all odes with degree > K, we have ) ( = = + α α α dk k 2 2 2 2 2 ) ( ) ( + < = < α K K dk k dk k k S C K K

Lower Boud for >2 (2) Sice we ca assume S </2, we should have The, we estimate a lower boud of S by This meas that the umber icreases as icreases 2 / 2 2 K > K K dk k S K > 2 2 2 α

Upper Boud for <2 () We select all odes with degree greater tha K= β as DS The, N DS =#odes i DS (domiatig set) is give by O the other had, the total umber of edges E G is E DS (=the umber of edges covered by DS) is The, prob. that a arbitrary edge is NOT covered by DS is ) ( ) ( ) ( ) ( ) ( = = = β β β α O dk k N DS ) ( 2 2 = = α dk k k E G ) ( ) (2 2 2 β β α = = dk k k E DS

Upper Boud for <2 (2) Sice a ode is covered by DS if at least oe edge coectig to the ode is covered by DS, the expected umber (N G-DS ) of odes ot covered by DS is N G β )(2 ) DS O( ) = Here, we balace N G-DS with N DS by lettig β ( ) = + ( β )(2 ) which results i β=2-. Therefore, a upper boud of the size of DS is estimated as which is o() for <<2 O( ( + ( β )(2 ) O( (2 )( ) It is iterestig that it takes the miimum (O(.75 )) whe =.5 ) )

Computer Simulatio

MDS size vs. Scalig Expoet () MDS size decays as decays (especially aroud =2)

Database Aalysis

Data

Why Not Cotradictig [Liu et al.]? Liu et al. assumed oly driver ode values ca be directly cotrolled through exteral sigals. Coversely, MDS approach assumed each driver ode ca cotrol its liks idividually. a ode with degree k is regarded as k driver odes.

Applicatios to Aalysis of Biological Networks

MDS for Aalyzig Biological Networks Applyig cotrol to real cells is far from easy However, MDS may be useful to fid importat proteis, gees, ad other molecules Aalysis of PPI etworks [Milekovic et al. PLoS Oe, 2] (before our work) [Wuchty, PNAS, 24] [Khuri & Wuchty, BMC Bioiformatics, 25] [Wag et al., BIBM 24] Aalysis of metabolic cacer etworks [Asgari et al., PLoS ONE, 23]

Applicatio to Aalysis of PPI Networks Wuchty foud that MDS is useful to fid importat proteis [Wuchty, PNAS 24] Proteis i MDS are eriched with essetial, cacer-related, ad virus-targeted gees. These proteis are highly ivolved i regulatory fuctios, showig high erichmet i trascriptio factors ad protei kiases, ad participate i regulatory liks, phosphorylatio evets, ad geetic iteractios. [Wuchty, PNAS 24]

Extesios

Cotrol of Bipartite Networks May real etworks have bipartite structure (left/right odes) Drug-target, researcher-paper, gee-disease Oly left odes ca be driver odes MDS approach eeds much smaller umber of driver odes [Nacher & Akutsu: Sci. Rep. 23]

Results o Bipartite Networks New feature: Itroductio of degree cutoff (P(k)= for k > H) For <2, the umber of driver odes is ) )( (2 2 H m O

Critical/Redudat Nodes i MDS We applied the cocepts of critical/redudat odes [Jia et al.: Nat. Comm. 23] to MDS because MDS is ot ecessarily uiquely determied Critical ode: appears i every MDS Redudat ode: ever appears i ay MDS Critical odes are expected to be more importat tha MDS [Nacher & Akutsu: J. Comp. Net. 25]

Robust MDS Robust MDS (RMDS): each ode is domiated by at least C odes (C= MDS) Robust agaist deletio of arbitrarily C- edges Upper boud of RMDS size (for <2): (D: miimum degree) RMDS size correspods to MDS size with miimum degree D-C+ ( DC+ )(2 )( ) O ( DC+ )(2 ) + [Nacher & Akutsu: PRE, 25]

Related Work by Molar et al. Aalysis of MDS size with degree cutoff [Sci. Rep. 23] Aalysis of MDS size with degree correlatio [Sci. Rep. 24] Damage-resiliet domiatig sets agaist radom ad targeted attacks [Sci. Rep. 25]

Coclusio

Coclusio Establishmet of a coectio betwee MDS ad structural cotrollability MDS size is small (o()) if <2 Heterogeeous etworks are ot difficult to cotrol This tedecy was verified (to some extet) by computer simulatio ad database aalysis Several extesios Bipartite etworks, Critical/Redudat odes, Robust MDS MDS is useful for idetifyig importat proteis i PPI etworks

Future Work Developmet of a framework/theory which makes cotrol of biological systems easy More rigorous theoretical aalysis o MDS size (our aalyses are based o a kid of mea-field approximatio) More biological applicatios Thak you!