Protein-protein interactions

1
Protein-protein interactions

• Ia. A combined algorithm for genome-wide
  prediction of protein function.
• Edward M. Marcotte, Matteo Pellegrini,
  Michael J. Thompson, Todd O. Yeates, David
  Eisenberg(1999) Nature 402,83-86.
• Protein function in the post-genomic era.
• David Eisenberg, Edward M. Marcotte, Ioannis
  Xenarios Todd O. Yeates(2000) Nature 405,
  823-826

2
FUNCTIONAL RELATIONSHIPS AMONG PROTEINS

• GENOME-WIDE PREDICTION (FUNCTIONAL GENOMICS)
• Does not rely on DIRECT SEQUENCE HOMOLOGY
• 3 independent predictions methods available
  experimental data.

3

• STRATEGIES USED TO FUNCTIONALLY LINK
  PROTEINS
• 6217 yeast proteins
• Correlated Evolution Related Phylogenetic
  Profiles (pattern of presence or absence of a
  particular protein across a set of organisms
  whose genomes have been sequenced) proteins,
  which operate together in a common pathway or
  complex, are inherited together.  
• Correlated mRNA Expression Patterns Correlated
  mRNA Expression Patterns under different growth
  conditions
• Correlated Patterns Of Domain Fusion Link 2
  proteins whose homologs are fused into a single
  gene (Rosetta stone sequences) in another
  organism.

4

• STRATEGIES USED TO FUNCTIONALLY LINK
  PROTEINS(continued)
• Gene Neighbour Method if in several genomes, the
  genes that encode 2 proteins are neighbors on the
  chromosome, the proteins tend to be functionally
  linked
• Experimental Evidence Mass spectrometry,
  Coimmunoprecipitaion, Yeast 2-hybrid data (DIP,
  MIPS yeast genome db)
• Metabolic pathway neighbours Proteins, which
  participate in same metabolic pathway, common
  structural complex or biological process or
  closely related physiological function BLAST
  homology searches and pairwise links were defined
  between yeast proteins whose E.Coli homologs
  catalyse sequential reactions in a metabolic
  pathway (EcoCyc db)

5
(No Transcript)
6

• RESULTS
• Phylogenetic profiles 20,749 links
• mRNA expression patterns 26,013 links
• Domain fusion method 45,502 links
•  
• 93,750 pairwise functional links among 76
  (4,701) of yeast proteins
• 4130 HIGHEST CONFIDENCE links (experimental
  proof, valid by 2 of 3 prediction methods)
• 19,251 HIGH CONFIDENCElinks
• (predicted by phylogenetic profiles)
• Remainder predicted by domain fusion or
  correlated mRNA expression patterns

7
VALIDATION

• Excellent reliability if 2 or more prediction
  methods agreed on a link.
• These methods link many proteins that are already
  known to function together on the basis of
  experiments.
• (Ribosomal proteins, proteins from flagellar
  motor apparatus and metabolic pathways)
• Keyword recovery Prediction could be compared
  to the actual annotation compare keyword
  annotation on SwissPDB, for both members of each
  pair of proteins, linked by one of the
  methods-possible when the members have known
  function.
• Keyword recovery if keywords match.
• Average signal to noise ratio for Keyword
  recovery
• Phylogenetic profiles 5
• mRNA expression patterns 2
• When 2 prediction methods gave same linkage 8
• Direct experimental data 8

8

• OUTCOME
• Functional links between proteins of unknown
  function
• General function assigned to more than half of
  2557 previously uncharacterized yeast proteins
  15 from high and highest confidence links, 62
  using all links.
• Functional Links Between Non-Homologous Proteins
  beyond traditional sequence matching Sup35,
  MSH6
• Discovery of potential interactions within and
  across cellular processes and compartments.
• Connections represent a gold mine for
  experimentally testing specific hypotheses about
  gene function.
• Viewing protein-protein interactions globally as
  a network and not as binary data sets, increases
  the confidence levels for individual
  interactions inspection of interaction web at
  different steps identifies unexpected links
  between previously unconnected cellular
  processes.

9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12

• Ib. A network of protein-protein interactions in
  yeast.
• Schwikowski B, Uetz P, Fields S. (2000). Nat
  Biotechnol. 18, 1257-61

13
DATA SOURCE

• MIPS site
• YPD
• DIPS
• Yeast-2-hybrid studies
• Biochemical experimental data

14
Prediction of function

• Annotated functions of all neighbors of P are
  ordered in a list, from the most frequent to the
  least frequent.
• Functions that occur the same number of times are
  ordered arbitrarily.
• Everything after the third entry in the list is
  discarded, and the remaining three or fewer
  functions are declared as predictions for the
  function of P.
• Evaluation of the quality of the links For
  unknown protein, test predicted function

15
RESULTS

• Analyzed 2,709 published interactions involving
  2,039 yeast proteins
• Single large network containing 2,358 links among
  1,548 individual proteins.Other networks had few
  proteins. 
• 65 of the interactions in the complete set of
  networks occur among proteins with at least one
  common functional assignment.
• 78 of the 1,432 interactions between proteins of
  known localization, the proteins share one or
  more compartments.
• Correctly predicted a functional category for 72
  of 1393 characterised proteins, with at least one
  partner of known function.
• Cross-talk between and within functional
  groups/subcellular compartments.
• Local function vs Contextual/cellular function
  (extended web of interacting molecules)
• Predicted functions of 364 uncharacterised
  proteins.

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
Reliability of the generated networks

• 1,393 of the 2,039 proteins were annotated with
  some function and had at least one neighbor
  annotated with a function.
• In 1,005 of these 1,393 cases (72.1), at least
  one annotated function was predicted correctly by
  the above method.
• Performed the same prediction algorithm 100 times
  on the basis of randomly generated interactions.
• Only 12.2 of the predictions yielded a
  prediction that agreed with the known annotation.

22
PROBLEMS

• Interactions of membrane proteins
  underrepresented Y2H data
• Y2H data lots of false positives.
• Only 15 agreement between this interaction data
  and Marcottes high quality prediction data.
• Uncertainities remain that WILL require
  additional experimentation.

23

• CHALLENGES
• Protein complexes are not static change with
  metabolic state of cell, external stimuli etc.
• Protein chip technology used to study transient
  interactions amenable to variety of assays like
  nucleotide-binding, enzymatic activity etc.

24

• II. Mapping protein family interactions
  intramolecular and intermolecular protein family
  interaction repertoires in the PDB and yeast.
• Park J, Lappe M, Teichmann SA. (2001). J Mol
  Biol. 307, 929-38.

25

• Protein DOMAIN interactions
• interactions between whole structural
  families of evolutionarily related domains as
  opposed to interactions between individual
  proteins.
• Types of domain interactions
• 1)      Domain-domain(intra-chain) interactions
  in multi-domain polypeptide chains
• 2)      Inter-chain protein interactions in
  multi-subunit protein complexes.
• 3) In transient complexes between
  proteins, which can also exist independently

26

• METHODS
• Protein superfamilies from SCOP db
• Interactions between families in the PDB
  (domains of known 3D structure)
• coordinates of each domain were parsed to
  check whether there are 5 or more contacts with
  5A? to another domain
• Interactions between families in the yeast
  genome by homology
• -Protein structures assigned to the yeast
  proteins using the domains from SCOP as queries
  in PSI-BLAST.
• -Yeast sequences also compared to the PDB-ISL
  with FASTA
• Assumption Within polypeptide chains, structural
  domains interact if there are less than 30 amino
  acids separating them.
• If one family F has 2 domains, a and b, and each
  of these interacts with a domain from a different
  family, then the number of interaction families
  for F will be 2.

27

• RESULTS
• 1st attempt at classifying interactions between
  all the known structural protein domains
  according to their families.
• Could classify 8151 interactions between
  individual domains in the PDB and the yeast in
  terms of 664 types of interactions between pairs
  of protein families.
• Scale free network Most protein families only
  interact with 1 or 2 other families.
• A few families are extremely versatile in
  their interactions and are connected to many
  families (Hubs in the graph)-functional reasons.
  Eg -Immunoglobulins, P-loop nucleotide
  triphosphate hydrolases
• In 45 of all families in the PDB, domains
  interact with other domains from the same family
  internal duplication and domain oligomerisation
  is favourable.
• Pairs of families that interact both within and
  between polypeptide chains belong mostly to 2
  types of domains enzyme domains and domains from
  the same family.

28
(No Transcript)
29
(No Transcript)
30

• PROBLEMS
• Multi-domain proteins cannot resolve exactly
  which domains are interacting not used
• Members of 2 families can sometimes interact in
  different ways, using different types of
  interface (different modes of oligomerisation of
  nucleoside diphosphate kinases)
• Does not take account of symmetric homooligomers,
  of which only one monomer is in the PDB entry and
  hence the number of homomultimeric family
  interactions may be underestimated.

31

• FUTURE
• 51 new interactions between superfamilies
  potential targets for structure elucidation and
  experimental investigation of these interacting
  polypeptides that do not have analogs in the PDB.
• For interactions in which one partner does not
  have a structural assignment, possible structures
  can be picked up from the set of known family
  interactions
• Database of domain-domain interfaces