Transcript Structure and specificity in protein interactions

Orthology, and its relevance to protein function prediction
Fitch 1970: “Where the homology is the result of gene duplication so that both copies
have descended side by side during the history of an organism, (for example alpha and
beta hemoglobin) the genes should be called paralogous (para= in parallel). Where the
homology is the result of speciation so that the history of the gene reflects the history of
the species (for example, alpha hemoglobin in man and mouse) the genes should be
called orthologous (ortho=exact)”
Comparing genomes for their
genes, orthologs
Species I
A
Gene A
Orthologs
Gene duplication
B
Speciation
Species II
“Which genes do two genomes share, and which don’t they share, and how does that relate
to their phenotypical similarities and differences”
Orthology: Who cares?
Dystrophin related protein 2
Dystrophin
Utrophin
Dystrotelin
Dystrobrevin
The DYS-1 gene from C.elegans is not orthologous to dystrophin, that there is no
effect of the knockout on the muscle cells is not so surprising.
Best-bidirectional hits: a graph based approach
Genome I
35%
23%
25%
30%
Genome II
Orthologs are expected to have relatively high levels of sequence
identity to each other (compared to to other non-orthologous
homologs), because they diverged relatively recently, and …… because
they have similar functions…. (???)
Large scale orthology determination is often done using bidirectional
best hits
An implementation of large-scale detection of pairwise orthology
relations between genomes:
1)
2)
Pairwise comparison of all genomes with each other, using the Smith-Waterman
algorithm to detect “all” homologous relations (E < 0.01) between the predicted
proteins.
Select orthologous relations by selecting best-bidirectional hits between proteins.
-Scoring as level of sequence identity * length of hit.
-Including the possibility of gene fusion/fission (protein A from genome I can be
orthologous to proteins B and C from genome II). By selecting
genome I
genome II
A
B
C
Genome I
35%
35%
25%
23%
25%
Genome II
40%
30% 22%
20%
35%
Genome III
Multiple genomes can be used to check for consistency of bidirectional
best hits.
Solution to the non-transitivity of the concept of orthology sensu
stricto is: “Group orthology”
Conceptually: all proteins that are directly descended from one
protein in the last common ancestor of the species one is interested
in are considered orthologous to each other
Operationally in a “graph-based approach”: Combine all connected
“best triangular hits” into Clusters of Orthologous Groups (COGs,
Tatusov et al, 1997). WWW.NCBI.NLM.GOV
Gene duplications are creative, creating the possibility for
developing new functions (in this case involved in carnitine
synthesis) but ….
They mess up orthology. Furthermore, intricate combinations
of gene duplications and speciations make orthology nontransitive
Inparalogs versus outparalogs:
Inparalogs are due to relatively recent, species-specific gene duplications, e.g.
Q9V6P0 and Q9VY24.
Outparalogs are due to gene duplications that preceded speciations, e.g. Q9V6P0 vs.
Q9VDM7
Disadvantages of graph-based approaches to orthology
Parallel non-orthologous gene-loss can lead to misidentification of orthology relations
when using best bi-directional hits as criterion.
Species I
A
Gene A
Gene loss
Gene duplication
B
Speciation
Non-Orthologs,
although
bidirectional
best hits
Species II
Disadvantages of graph-based approaches to orthology
Variations in the rate of evolution can lead to misidentification of orthology
relations when the latter are based on bi/multi-directional best hits.
Graph based approaches
can recognize
outparalogs as inparalogs
(and vice versa)
Because of independent loss events, and because of variable rates
of evolution, in large gene families, orthology determination using
bi/multi-directional best hits (graph-based approaches) does not
always resolve separate orthologous and/or functional groups.
One solution to this is the creation of phylogenies………
Prediction of orthology using phylogenies (unrooted)
A tree-based approach would also allow a hierarchical, multilevel view on orthology, e.g. by including a numbering
system
Classic usage of phylogeny:
inferring evolutionary history
Dyall et al, Nature 2004
Hrdy et al, Nature 2004
How to make a tree
1)
Distance based methods. Cluster the sequences based on relative levels of sequence
similarity. Fast, but not a direct reconstruction of “what happened in evolution”. Neighbor
Joining is the often used method here
2)
Parsimony methods. Reconstruct the phylogeny that required the least amount of
mutations. Slower (requires in principle the examination of all trees), but branch & bound
makes it faster
N=P I=3…..T(2i-5)
and based on the questionable assumption that the least amount of events possible occurred
in evolution.
3)
Maximum likelihood methods. Find the phylogeny (including branch lengths) that, given a
model of sequence evolution, was most likely to have produced to sequence alignment.
This involves the comparison of all trees, estimating the branch lengths optimal for that
tree, and subsequently estimating the likelihood of the complete tree . The slowest method
of all, that furthermore requires knowledge of a large number of parameters
Likelihood-Based Phylogeny
4 Possible trees for 4 sequences
Sequence W:
Sequence X:
Sequence Y:
Sequence Z:
A
A
A
A
WX Y Z
C
C
C
C
G
G
G
A
C
C
C
C
G
G
A
A
T
T
A
G
WY X Z
T
T
T
G
G
G
G
G
G
G
A
A
G
G
A
A
WZ X Y
All Possible Evolutionary Paths for one
tree, for one collumn in the alignment
T T A G
A
T
G
C
A
T
G
C
A
T
G
C
Likelihood for One Path
TTAG
T
G
G
L(path) = L(root) x P L(branches)
=P(G) P(T|G)P(G|G) P(A|G)P(G|G) P(T|T)P(T|T)
Calculating the likelihood of any path
requires a model of sequence evolution
(and an estimate of the time, and the
mathematics to combine both)
Kimura
Jukes Cantor
A
a
C A
a
a
G
a
a
a
a
2a
2a
T G
General
C A
2a
2a
a
b
C
a
c
T G
d
f
e
T
The Jukes--Cantor Model,
(including time…)
 1  3α 
 αε
S ( )  
αε

 αε
 rt

s
 S (t )   t
 st

 st
st
rt
st
st
st
st
rt
st
αε
1  3α 
αε
αε
st 

st

st 
rt 
αε
αε
1  3α 
αε
αε
αε
αε
1  3α 





 4 t
 rt  (1  3 e
)/4

where 
 s t  (1  e  4  t ) / 4
Sum over all paths
TTAG TTAG
A
A
T
T
G
G
C A
C
T
G
C
L(Column Cluster 1) = S L(all possible Evolutionary Paths)
= L(path1) + L(path2) + L(path3) + … + L(path64)
Likelihood of a phylogeny tree for one site
x5
t4
x4
t3
t2
t1
x2
x1
x3
When x4 x5 are unknown,
P ( x x x | T , t )
1

2

5
x ,x
3
5
4
5
3
5
1
4
2
4
P ( x ) P ( x | x , t 4 ) P ( x | x , t 3 ) P ( x | x , t1 ) P ( x | x , t 2 )
4
Whole Sequence Likelihood
WX Y Z
L(Sequence) =  L(each position i)
i
Choose the tree with the Maximum Likelihood.
For All Positions u=1…N
N
P ( x | T , ) 
 P(x
1
u
 x | T , )
n
u
u 1
or
N
log( P ( x | T ,  ) 
 log
u 1
P ( x  x | T , )
1
u
n
u
Maximum Likelihood Phylogeny
•
Pick an Evolutionary Model (Modeltest can help)
•
For each position, Generate all possible tree structures
•
Based on the Evolutionary Model, calculate Likelihood of these Trees
and Sum them to get the Column Likelihood for each OTU cluster.
•
Calculate Tree Likelihood by multiplying the likelihood for each position
•
Choose Tree with Greatest Likelihood
Likelihood-based Phylogeny
• Works well for distantly related sequences
• Works well under different molecular clock
theory
• Can incorporate any desirable evolutionary
model
• Requires a good “model” of evolution
• Requires fast computers
Obtaining confidence in our tree
Bootstrapping:
1) Generate e.g. 1000 alignments by random sampling (with
replacement) from the real alignment.
2) Determine the phylogenetic tree for each alignment.
3) Count how often every subtree appears, and put those values at
the internal branches of the complete tree.
… bootstrap values tend to be on the conservative side….
Unrooted tree topologies
A
B
A
A
B
=
C
=
C
B
D
D
C
D
=
((A,B),(C,D))
Bracket notation
-Unrooted tree topologies only reflect relative evolutionary
relations (In the primates the humans and chimpanzee are closer
related to each other than either is to the the chimpanzee than
they are to the Orang-Otang and the Gibbon)
-Rooted trees reflect relative order of descendance (In the
primates first the Gibbon branched off, then the Orang-Otang
branched off, then the chimpanzee and then the humans)
Orang-Otang Gibbon
Chimp
Human
Orang-Otang
Gibbon
Baboon
Chimp Human
Rooting is important to
separate different
orthologous groups in a
tree  different
functionalities.
Berend’s lecture….
To construct a tree one needs a multiple sequence alignment:
Different optimization criteria:
Global optimization:
The best would be to run multidimensional dynamic programming, which would be
guaranteed to give you the optimal alignment (optimal here means: the best alignment given
the similarity matrix and the gap penalties, not necessarily what is structurally the most likely
course of events).
Reconstructing evolution:
Trying, along an evolutionary tree, to minimize the number of insertion/deletion events.
Alignment
Phylogeny
Automatic sequence alignment use heuristics to obtain
some kind of approximation of the optimal alignment.
ClustalW:
-pairwise sequence alignments of all pairs (N-1)^2 / 2.
-select the most similar pairs of sequences, align those, and
subsequently iteratively align the alignments.
T_Coffee:
-better but slower
Muscle:
Substrate specificities are not necessarily monophyletic (convergent
evolution).
Convergent evolution of Trichomonas vaginalis lactate dehydrogenase from malate
dehydrogenase. Wu et al., PNAS 1999
Lactate/Malate Dehydrogenase
Different small-molecule specificity
H O
CH3 - C - C - O-
LDH
CH3 - C - C - O-
OH
Lactate
O
Pyruvate
H O
O
- C - CH2 - C - C OH
Malate
O-
MDH
-O
- C - CH2 - C - C - O-
O
O
-O
O
O
Oxaloacetate
Lactate/Malate Dehydrogenase
H O
CH3 - C - C - OOH
Lactate
negative
-O
H O
- C - CH2 - C - C - O-
O
positive
Arg 102
OH
Malate
Hannenhalli & Russell, JMB, 303, 61-76, 2000
Another source of information that can be used for orthology
prediction is gene-order conservation.
35%
35%
(be careful for duplicated sets of genes though)
icd or leuB ?
Prediction of orthology/function using a combination of
sequence similarity and gene-order conservation
Further reading
•
The quest for orthologs: finding the corresponding gene across genomes.
Kuzniar A, van Ham RC, Pongor S, Leunissen JA. Trends Genet. 2008
Nov;24(11):539-51