Modeling Com binatorial I ntervention Effects in Transcription Netw - - PowerPoint PPT Presentation

Achim Tresch Com putational Biology Gene Center Munich

( The Sound of One-Hand Clapping) Modeling Com binatorial I ntervention Effects in Transcription Netw orks

The Question

(Japanese Kōan)

If two hands clap and there is a sound; what is the sound of one hand?

Kōan A paradoxical anecdote or riddle, used in Zen Buddhism to demonstrate the inadequacy of logical reasoning and to provoke enlightenment.

Synthetic Genetic I nteractions

modified after Collins, Krogan et al., Nature 2007

How to define “Interaction“ mathematically?

Synthetic Genetic Array

ΔA

GrowthYA of single manipulation of A

ΔB

GrowthYB of single manipulation of B

ΔA ΔB

Growth YAB

• f double

manipulation

• f A and B

Synthetic Genetic I nteractions ΔB ΔA ΔA ΔB

Phenotype Measurement YA

• f single perturbation

Phenotype Measurement YB

• f single perturbation

Phenotype Measurement YAB

• f double perturbation

How to define “Interaction“ mathematically?

The interaction score SAB is a function

• f the two single perturbations and the

combined perturbation,

SAB = SAB (YA ,YB ,YAB )

Synthetic Genetic I nteractions

Common Interaction Scores Common choices for f : f = min(YA ,YB ) (v. Liebig s minimum rule for plant growth) f = YA ·YB (chemical equilibrium a + b ↔ ab , [a][b] = [ab]) f = YA + YB (log version of YA ·YB ) f = log2[(2YA - 1)(2YB - 1) + 1] (essentially the same as YA + YB ) Define an expected phenotype of the double perturbation as a function f(YA ,YB ) of the single perturbation phenotypes YA and Yb. The interaction score SAB is then the deviation from the expected phenotype SAB = YAB - f(YA ,YB )

Interaction Scores are not very reliable Results crucially depend on f

Mani, Roth et al., PNAS 2007

Synthetic Genetic I nteractions

Pan, Boeke et al., Cell 2006 Cartoon by Van de Peppel et al, Mol. Cell 2005 Collins, Krogan et al., Nature 2007

Breakthrough: Combine a set of weak predictors to create a strong predictor (guilt by association = correlation of interaction scores)

Synthetic Genetic I nteractions

Costanzo M, Myers CL, Andrews BJ, Boone C, et al.: Science 2010

Take home message: Two components are likely to interact (physically) whenever they have the same interaction partners

Screening for TF interactions ΔA

One manipulation High dimensional readout

If two hands clap and there is a sound; what is the sound of one hand?

Genetic interactions from one perturbation

Harbison, Fraenkel, Young et al. Nature 2004 MacIsaac, Fraenkel et al. BMC Bioinformatics 2006 Ansari et al., Nature Methods 2010 Berger, Bulyk et al., Nature Biotech 2006

a) From ChIP binding experiments b) From protein binding arrays, followed by PWM-based predictions

Step 1: Construct a transcription factor - target graph

Genetic interactions from one perturbation

Step 1: Construct a transcription factor - target graph

Intersection size of target sets of TF1 and TF2 can be used alone to assess TF cooperativity.

(Beyer, Ideker et al., PlOS Comp. Biol 2006)

Genetic interactions from one perturbation

~2.000 target genes 118 transcription factors

Graph obtained from MacIsaac et al. (BMC Bioinformatics 2006)

Established Methods for the detection of univariate TF activity :

GSEA (Subramanian, Tamayo PNAS 2005) Globaltest (Goemann, Bioinformatics 2004) MGSEA (Bauer, Gagneur, Nucl. Acids Res. 2010) and many more …

Step 2: Combine TF-target information and expression data

Common Idea: A TF is active if its set of target genes shows significantly altered expression. To quantify this, various tests are constructed.

gene 4

time

Antagonistic interaction of TF 1+2 TF 1+2 active

Genetic interactions from one perturbation

gene 1 TF 1 TF 2 Synthesis rates during salt stress gene 2 TF 1 is active gene 3 TF 2 is active Binding sites

TF1 TF1 TF2 TF2

Step 3: Given TF1 and TF2, group genes into 4 interaction classes

Genetic interactions from one perturbation

gene 1 TF 2 is inactive gene 3 TF 1 is inactive gene 2 TF 1 TF 2 Synthesis rates during salt stress Binding sites

time

Synergistic interaction of TF1+2 gene 4 TF 1+2 active

Step 3: Given TF1 and TF2, group genes into 4 interaction classes

Genetic interactions from one perturbation

Our interaction score for the pair (T1,T2) is then β12.

TF2) and TF1

• f

target a is ( TF2)

• f

target a is ( ) TF1

• f

target a is ( ~ ) induced not is ( ) induced is ( log

12 2 1

g Ind g Ind g Ind g P g P ⋅ + ⋅ + ⋅ +         β β β β

Step 4: Use these 4 groups to define an interaction score (for all genes g) For any pair of transcription factors T1 and T2, we perform a logistic regression.

Genetic interactions from one perturbation ) , ( ) ( ) ( ~ . . .

2 1 12 2 2 1 1

g TF TF Ind g TF Ind g TF Ind → ⋅ + → ⋅ + → ⋅ + β β β β

gene 1 gene 3 gene 2 gene 4

time

TF 1 is active TF 2 is active Antagonistic interaction

1

> + β β

2

> + β β

12 2 1

< + + + β β β β

Example:

~ β

0 <

β

1

~ β β +

2

~ β β +

12 2 1

~ β β β β + + +

TF 1+2 active

Step 4: Use these 4 groups to define an interaction score

TF 1 TF 2 Binding sites

) ( ) (

1 1 12

< + − + − < β β β β β β

Application: Osm otic stress in yeast

Use the guilt by association trick to construct an interaction matrix for all transcription factors using only a two group microarray comparison!

Inclusion criterion:

• nly TFs

with >70 targets

Miller, Tresch, Cramer et al.,

• Mol. Syst.
• Biol. 2010,

in revision

„One hand clapping“

Validation with BioGRID database:

Application: Osm otic stress in yeast

Among 84 TFs under consideration (with enough targets), 3486 potential interactions

• Exist. Only 97

interactions are recorded.

Application: Osm otic stress in yeast

Single interactions scores don‘t work well Profile correlations do work

Validation with BioGRID database:

Genetic interactions from one intervention

One hand clapping can be applied to: Microarray data, Pol II ChIP data, nascent RNA data 3 stress responses:

• smotic stress NaCl,
• smotic stress KCl,

heat shock

(Mitchell, Pilpel at al. Nature 2009):

Application to a similar dataset leads to similar results:

Acknow ledgem ents

20

Gene Center Munich: Patrick Cramer Dietmar Martin Björn Schwalb Sebastian Dümcke

My Answ er

Two hands clap and there is a sound; what is the sound of one hand? It is similar for transcription factors that interact.

Zen Biology Systems Buddhism