Protein interaction Computational (inferred) Experimental (observed)

Protein interaction

• Computational (inferred)

• Experimental (observed)

Inference of Protein function

Guilt by association

Five methods for inferring functional interactions: Complexes, PathwaysRosetta stone Phylogenetic profiles Expression profiles Gene neighbors Operon method

Rosetta Stone Assumption: Fusion of functionally-linked domains

In organism 1:

A

In organism 2:

Implies proteins A and B may be functionally linked

A

A'

B

B'

Five Known Rosetta Stone Links:

Marcotte et al., Science, 285, pp. 751-753 (1999)

Pathways Detected by the Domain Fusion (Rosetta stone) Method

AroH YDIB AroK PurFAroF AroE AroLAroG Pur2

AroA AroBAroB PurT Pur3

BAroD PurL

AroE Pur7 Pur5 Pur2 PurU

AroK PurE Pur3 PurE CAroL

Pur5 PurKAroA PurK

PurT Pur7AroC GuaA

GuaB PurB PurB A D

GuaA PurH PurA

Marcotte et al., Science, 285, pp. 751-753 (1999)

PHYLOGENETIC PROFILE METHOD

YC083W homology to thioredoxin

MRPL2

MRPL6

MRPL7

MRPL10 known

MRPL16 ribosomal predicted to

YGR021W MRPL23 proteins target

member of MRPS9 mitochondria

highly MRPS28

conserved MRF1 peptide chain release factor protein

protein family YJR113C homology to ribosomal protein S7 synthesis

of unknown MSY1 ttrosyl-tRNA synthetase

function YGL068W probable ribosomal protein L12

MGE1 heat shock protein/chaperone

YDR116C homology to bacterial ribosomal L1 protein

YHR189W homology to peptidyl-tRNA hydrolase

SIS1/XDJ homology to DraJ heat shock protein

PDR13/SSE1/LHS1 homolgoy to Hsp70

RIB2 DRAP deaminase

YDL036C homology to Rib2/pseudouridine synthase

MIS1 C1-THF synthase

ADE3 C1-THF synthase

TPI1 tiose phosphate isomerase

YGL236C homology to conserved gidA family, unknown function

YOL060C homology to hypothetical C. elegans protein M02F4.4

PhylogeneticProfile: e.g.for a yeast protein

Microarray Co-expression Analysis

Conclusion: P2 and P7 are functionally linked P3 and P6 are functionally linked

Expression Profiles

Rich m

ed

Starve

High T

P1 1 0 1P2 1 1 0P3 0 1 1P4 1 0 0P5 1 1 1P6 0 1 1P7 1 1 0

Profile Clusters

P4 1 0 0

P1 1 0 1

P5 1 1 1

P2 1 1 0 P7 1 1 0

P3 0 1 1 P6 0 1 1

Inferring Functional Linkages from the Gene Neighbor Method

genome 1

. . .genome 2 genome 3 genome 4

A

AA

A

B

B

BB

C

C

CC

A

B

C

A statistically significant correlation is observed between the positions of proteins A and B across multiple genomes. A functional relationship is inferred between proteins A and B, but not between the other pairs of proteins:

gene A gene B gene C

OPERON method of inferring functional linkages in the genome of Mycobacterium tuberculosis

distance thresholdnumber of predicted operon groups # of genes with links # of functional linkages

0bp 542 1279 203425bp 792 2071 444250bp 879 2420 589075bp 919 2665 7026100bp 933 2870 8468

Examples of Inferred Protein Networks

Features of Inferred Networks

Linkages are between non-homologs

The function of each proteinis defined in its cellularcontext by its linkages

Each linkage is assigned a probability

Need to separate and visualize each moduleof functional proteins

Prions

• Can spontaneously change shape (10-5)

• Can induce shape change in other proteins

• Contagious, inheritable

• Sup35– Translation termination factor– Prion: reads through stop codons

Sup 35

Interactions

A Network of Proteins Related to Flagella

fliA

fliA

Flagella-Related Transcription Factors

fliA

Flagella-Specific ATP Synthase

fliA

Chemotaxis-Related Proteins

Quantitative Assessment of Inferred Protein Linkages

Assessing Functional Linkages by method of Keyword Recovery

X A Y

Z

Is the linkage inferred between A and Y valid ?

Compare the keyword annotations common toA and Y, with the number expected at random= “signal-to-noise”

Assessing Inference of Linkages by Recovery of Keywords

YEAST GENOME (Marcotte et al. Nature, 1999)

Inferrence Method # functional Signal links to noise

Individual Methods:Experiment (2-hybrid) 500 8Phylogenetic Profiles 20,749 5Rosetta Stone 50,459 3Correlated mRNA expression 26,013 2

Combined methods:

Links by ≥ 2 methods 1,271 8

Observed Functional Linkages

Database of Interacting Proteins

www.dip.doe-mbi.ucla.edu

Experimentally detected interactions from the scientific literature

The DIP Database

DOE-MBI LSBMM, UCLA

Paul D. Boyer Nature (1999) XRay structure Stock et al. Science (1999)

Two-hybrid assayMoritani C et al. Biochim Biophys (1996)

ATP Synthase ATP Synthase in DIP

HOW PROTEIN INTERACTIONS ARE REPRESENTED IN DIP

0

50

100

150

200

1 2 3 4 5 6 7 8 9 10 11 13 16 687 2138

Number of proteins per network

Num

ber

of c

lust

ers

DIP contains 22,000 interactions from 2,200 articles,

representing hundreds of networks

Assessing DIP reliability with mRNA expression data (ERP)

Coexpression MeasureCoexpression Measure

i ii BAd 2)(

Euclidean distance

Protein A

Pro

tein

B

expression levellog(e/eref)

Extracting the Expression Profile (EPR) IndexExtracting the Expression Profile (EPR) Index

0

4

8

12

16

20

0 10 20 30d2

p x

100

)(p)1()(p)p( 2N

2I

2 ddd

Model:non Y2H all vs all

40% of Y2H is highly reliable40% of Y2H is highly reliable

Overall: 50% of 8000 yeast DIP interactions are reliable = 4000

Interacting

Experimental (Y2H)

Random

fitted =0.4

Many proteins interacting in a Y2H may never meet each other in the cell!

Expression Profile Data Assesses Interaction Data(EPR Index; Deane et al. Molec & Cell Proteomics, 2002)

DOE-MBI LSBMM, UCLA

Uetz, P. et al.; Nature 403: 623-627 (2000)Uetz, P. et al.; Nature 403: 623-627 (2000)

Ito, T. et al.; PNAS 97:1143-47 (2000)Ito, T. et al.; PNAS 98:4569-74 (2001)

Ito, T. et al.; PNAS 97:1143-47 (2000)Ito, T. et al.; PNAS 98:4569-74 (2001)

Ho, Y. et al.; Nature 415:180-183 (2002)Ho, Y. et al.; Nature 415:180-183 (2002)

Gavin, A.C. et al.; Nature 415: 141-147 (2002) Gavin, A.C. et al.; Nature 415: 141-147 (2002)

7.8+/-3.7%

19.5+/-3.7%

48.7+/-6.9%

89.5+/-6.6%

Measured vs Inferred Functional Interactions

EPR Assessment of Inferred Interactions vs. Y2Hybrid & MS

Method Linkages EPR index

Rosetta S +Phylo P 24,200 0.4-0.6

Yeast 2 hybrid ~4,000 0.2-0.5

Mass spec ~4,000 0.1-0.9

DIP yeast overall 6,144 0.3

Some Conclusions

AX Y

Z

B

V

CA protein’s function is defined by the cellular context of its linkages

Many functional linkages are revealed from genomic data

Validity of functional linkages can be assessed computationally by keyword recovery or experimentallyby expression data

Functional complexes can be discovered from genomic data