le song [email protected] joint work with mladen kolar and eric xing keller: estimating time...

55
Le Song [email protected] Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time KELLER: Estimating Time Evolving Interactions Evolving Interactions Between Genes Between Genes

Upload: weston-foxen

Post on 14-Jan-2016

216 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

Le [email protected]

Joint work with Mladen Kolar and Eric Xing

KELLER: Estimating Time Evolving KELLER: Estimating Time Evolving Interactions Between GenesInteractions Between Genes

Page 2: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

2

Transient Biological Processes

Page 3: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

33

PPI Network

Page 4: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

4

Time-Varying Interactions

Page 5: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

5

The Big-Picture Questions

What are the interactions?

What pathways are activeactive at a particular time point and location?

How will biological networks respond to stimuli (eg. heat shot)?

Page 6: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

6

Regulation of cell response to stimuli is paramount, but we can usually only measure (or compute) steady-state interactions

Transcriptionalinteractions

Protein—proteininteractions

Biochemicalreactions

▲ Chromatin IP ▲Microarrays

▲ Protein coIP▲ Yeast two-hybrid

▲ Metabolic flux measurements

Transcriptionalinteractions

Protein—proteininteractions

Biochemicalreactions

▲ Chromatin IP ▲Microarrays

▲ Protein coIP▲ Yeast two-hybrid

▲ Metabolic flux measurements

Page 7: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

7

t=1 2 3 T

Current PracticeStatic Networks

Microarray Time Series

Dynamic Bayesian Networks

Page 8: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

8

Our GoalReverse engineer temporal/spatial-specific

“rewiring” gene networks

Time

t*

n=1

--- what are the difficulties?--- what are the difficulties?

Page 9: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

9

Two Scenarios

Smoothly evolving networksSmoothly evolving networks Abruptly changing networksAbruptly changing networks

Page 10: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

10

Scenario I (This paper)Kernel reweighted L1-regularized logistic regression

(KELLER)

Key Idea I: reweighting observations

Key Idea II: regularized neighborhood estimation

Page 11: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

11

Key IdeaWeight temporally adjacent observations more than

distal observations

Page 12: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

12

Key IdeaEstimate the neighborhood of each gene separately

via L1-regularized logistic regression

Kernel Reweighting

Log-likelihood

L1-regularization

Page 13: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

13

Consistency

Theorem 1: Under certain verifiable conditions (omitted here for simplicity), KELLER recovers the true topology of the networks:

Page 14: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

14

Synthetic data

DBN and static networks do not benefit from more observations

Number of Samples

Page 15: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

15

Key idea: Temporally Smoothing

Tesla (Amr and Xing, PNAS 2009)

TESLA:

Senario II

Page 16: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

16

Drosophila Life Cycle

Larva

Embryo

Pupa

Adult66 microarrays across

full life cycle

588 genes related to development

Page 17: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

17

molecular function

biological process

cellular component

Page 18: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

18

Page 19: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

19

Page 20: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

20

Page 21: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

21

Page 22: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

22

Page 23: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

23

Page 24: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

24

Page 25: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

25

Page 26: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

26

Page 27: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

27

Page 28: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

28

Page 29: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

29

Page 30: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

30

Page 31: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

31

Page 32: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

32

Page 33: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

33

Page 34: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

34

Page 35: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

35

Page 36: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

36

Page 37: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

37

Page 38: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

38

Page 39: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

39

Page 40: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

40

Network Size vs. Clustering Coefficient

mid-embryonic

mid-pupal

Page 41: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

41

Network Size vs. Clustering Coefficient

mid-embryonic stagetight local clusters

mid-pupal stageloose local clusters

Page 42: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

42

Interactivity of Gene Sets

27 genes based on ontology

Page 43: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

43

Interactivity of Gene Sets

25 genes based on ontology

Page 44: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

44

Transient Gene InteractionsTime

Gene Pairs

Active

Inactive

msn dock

sno Dl

Page 45: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

45

Transcriptional Factor Cascade

Summary networks 36 transcription factorsNode size its total activity

Page 46: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

46

TF Cascade – mid-embryonic stage

Page 47: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

47

TF Cascade – mid-larva stage

Page 48: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

48

TF Cascade – mid-pupal stage

Page 49: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

49

TF Cascade – mid-adult stage

Page 50: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

50

Transient Group Interactions

Page 51: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

51

ConclusionKELLER for reverse engineering “rewiring” networks

Key advantages:Computationally efficient (scalable to 10Computationally efficient (scalable to 1044 genes) genes)

Global optimal solution is attainableGlobal optimal solution is attainable

Theoretical guaranteeTheoretical guarantee

Glimpse to temporal evolution of gene networks

Many interactions are rewiring and transient

Availability: http://www.sailing.cs.cmu.edu/

Page 52: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

52

The EndThanks

Travel fellowship:Office of Science (BER), U.S. Department of Energy, Grant No. DE-FG02-06ED64270

Funding: Lane Fellowship,

Questions?

Page 53: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

53

Interactivity of Gene Sets

30 genes based on ontology

Page 54: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

54

Timing of Regulatory Program

Galactose

Page 55: Le Song lesong@cs.cmu.edu Joint work with Mladen Kolar and Eric Xing KELLER: Estimating Time Evolving Interactions Between Genes

55

ChallengesVery small sample size

Experimental data are scarce and costly

Noisy measurement

More genes than microarraysComplexity regularization needed to avoid over-

fitting

Observations no longer iid since the networks are changing!