Computational functional genomics

1
Computational functional genomics

• Lital Haham
• Sivan Pearl

2
Introduction

• Piles of information but only flakes of knowledge.

• The existing information

Collections of genomic sequences. Expression
profiles Protein-protein interactions And many
more
3
Introduction

• Computational biology strives to extract the
  maximal possible information from known
  sequences, by classifying them according to their
  homologous relationships, predicting their
  biochemical activity, cellular function,
  3-dimensional structures and evolutionary origin.

4
The COG-Clusters of Orthologous Groups of proteins

• Identification of orthologs is critical for
  reliable prediction of gene function in newly
  sequenced genomes.

• The purpose of COG is to serve as a platform for
  functional annotation of newly sequenced genomes
  and for study of genome evolution.

• Reflects one-to-one, one-to-many and many-to-many
  relationships.

5
The COG-statistics

• In 2003, there are 3307 COGs including 74059
  proteins from 43 genomes.

• Genomes from- Bacteria, Archaea and Eukaryota.

• The database includes 17 functional groups.

6
The COG- make on your own

• COG construction procedure is based on the notion
  that any group of at least 3 proteins from
  distant genomes that are more similar to each
  other than to any other protein from the same
  genomes, are most likely to belong to an
  orthologous family.

7
The COG- make on your own
All-against-all protein sequence comparison
8
The COG- make on your own
9
The COG- adding new genomes

• The COGNITOR program adds new proteins to
  pre-existing COGs on the basis of multiple Best
  Hits.

• 60-80 of the proteins of prokaryotes could be
  included.

10
The COG- more applications

• Detecting missed genes.

• Convenient for variety of evolutionary-oriented
  analyses of protein families.

11
Methods

• Experimental method

Biochemical and genetic experiments

• Computational methods

Homology method (BLAST), mRNA expression
Phylogenetic profile
Fusion method (Rosetta stone analysis)
Gene neighbour method
12
Homology method

• Homology method searches proteins whose AA
  sequences are similar.

• 40-70 of new genome can be assigned to some
  function.

• Involve identification of some molecular function.

13
mRNA expression

• Analysis of correlated mRNA expression levels
  enables to establish functional linkages, by
  detecting changes in mRNA expression in different
  cell types, or different environments.

14
Phylogenetic profile

• Describes the pattern of presence or absence of a
  particular protein, across a set of organisms.

• Number of possible profiles

• This number far exceeds the protein families.

15
Phylogenetic profile

• Why would two proteins always both be inherited
  into new species or neither inherited, unless the
  two function together?

• If two proteins have the same phylogenetic
  profile, it is inferred that they have a
  functional link engaged in a common pathway or
  complex.

16
Phylogenetic profile
17
Phylogenetic profile- example

• Analysis of three proteins RL7, FlgL and His5,
  according to their phylogenetic profiles.

• RL7 more than half have function associated with
  the ribosome.

• FlgL more than half include various flagellar
  proteins and cell-wall maintenance proteins.

• His5 more than half involved in amino acid
  metabolism.

18
Phylogenetic profile- example
PgsA phospholipid synthesis YGGH hypothetical
YBEX hypothetical RL34 ribosome L34 RL36 ribosome
L36 RL27 ribosome L27 RL25 ribosome L25 YQCB
hypothetical YABO hypothetical YCEC
hypothetical RFH peptide release factor ClpB geat
shock protein
RL7 ribosome L7 RL15 ribosome L15 RL17 ribosome
L17 PTH peptidyl-tRNA hydrolase RNC ribonuclease
III
YJFH hypothethocal
RS14 ribosome S14
GidB glucose inhib. Division RL24 ribosome
L24 DEF polypeptide deformylase RL20 ribosome
L20 MesJ cell cycle protein RL19 ribosome
L19 RL21 ribosome L21 RL9 ribosome L9 SmpB small
protein B
G3P3 dehydrogenase
RL4 ribosome L4 NONE hypothtical
GrpE co-chaperone
19
Phylogenetic profile

Phylogenetic profiles link protein with
similar keywords
20
Fusion method or the Rosetta stone analysis

• Some pairs of interacting proteins have homologs
  in another organism, fused into a single protein
  chain.

• When two separate proteins in one organism, A and
  B, are expressed as a fused protein in some other
  species, there is a high probability that A and B
  are linked in function.

21
Fusion method
22
The Rosetta Stone model
23
Fusion method what is it good for?

• Predicts protein pairs that have related
  biological functions.
• Predicts potential protein-protein interactions.

• Can turn up complexes of proteins, or protein
  pathways.

24
Fusion method what is it good for?
25
Fusion method

• The group searched the 4290 protein sequences of
  the E.coli genome.

• The proteins could form at most (4290)(4289)/2
  pair interactions. But we expect much less

• There were found 6809 candidate for pair
  interactions.

26
Fusion method validation

• Looking for a similar function in existing
  annotations that would imply at least functional
  interaction.

• Of the E.coli pairs that were found in the
  Rosetta Stone analysis, 68 share at least one
  keyword in their annotations, whereas from E.coli
  proteins that were selected randomly, only 15
  share a keyword.

27
Fusion method validation

• From a database containing protein pairs that
  have been found to interact (experimentally)
  6.4 are linked by Rosetta Stone sequences.

• The phylogenetic profile method was applied to
  the interactions predicted by the fusion method.
  It found more than 8 times as many interactions
  suggested by the phylogenetic profile method, as
  for randomly chosen sets of interactions.

28
Fusion method missing pairs

• False negatives

There was no fusion of the interacting proteins.
The fused protein disappeared during the course
of evolution.
29
Fusion method False alarms

• False positives

False prediction of physical interactions when
the proteins are fused, but are co-regulated and
dont interact.
Cannot distinguish between homologs that bind
and those that do not.
30
Fusion method False alarms

• The false positive rate in E.coli due to the
  inability to distinguish homologs is about 82.

• To reduce these errors the promiscuous domains
  were found and removed during the analysis.

• By filtering of only 5 of all domains, we can
  remove the majority of falsely predicted
  interactions.

31
Fusion method False alarms
32
Neighbour method

• Functional links between genes can be identified
  by examining whether the proximity of the genes
  is conserved across multiple genomes.

• Powerful in uncovering functional linkages in
  prokaryotes where operons are common.

33
Neighbour method
34
Neighbour method- definitions

• close proximate genes are on the same strand
  within 300 bp, and transcribed in the same
  direction.

• Direct link two proximate genes that are also
  proximate in at least two other genomes of
  different phylogenetic groups.

• Inferred link two genes that are not close but
  with orthologs that are close in at least three
  other genomes of different phylogenetic groups.

35
Neighbour method- defenitions
36
Neighbour method

• Proximity between genes is maintained mostly
  because it facilitates their co-transfer to
  another organism.

• Example restriction-modification systems.

37
Neighbour method- validation

• Identification of links that are annotated in
  KEGG or COG and calculate the fraction of those
  in the same functional pathway / category.

• The functional correspondence is correlated to
  the minimal number of phylogenetic groups, in
  which the proximity is detected.

38
Neighbour method- validation
N tradeoff
39
Neighbour method- example
40
Happy end???

• The group analyzed the 6,217 proteins of the
  yeast Saccharomyces combining several methods.
• one can expect each protein to be functionally
  linked to perhaps 550 other proteins, giving
  30,000300,000 biologically meaningful links.

41
Happy end???
42
Networks

• When methods of detecting functional linkages are
  applied to all the proteins of an organism,
  network of interacting, functionally linked
  proteins can be traced.

• As methods improve for detecting protein
  linkages, it seems likely that most of the
  proteins will be included in the network.

43
Networks
44
????? ???