Etwork Motifs Lustering Riteria Imple Xample Lustering: N C C S E C E
Etwork Motifs Lustering Riteria Imple Xample Lustering: N C C S E C E
• Can
be
thought
of
as
“simple
building
• Nodes
that
never
play
the
same
role
1
4
1
3
Input
moBf
(FFD)
• Edges
can
be
ac3va3ng,
repressing,
or
both
blocks
of
complex
networks”
are
less
similar
2
3
• Majority
of
feedforward
mo3fs
in
the
network
are
• To
quan3fy
this
similarity,
we
count
Input
network
of
coherent
type,
where
direct
effect
has
same
sign
as
indirect
effect4
2
the
number
of
3mes
each
node
of
the
network
plays
each
specific
role
in
the
! $ ! $ 2
2
2
2
1 0 0 0 2 1 1 2 1 2
given
mo3f
# & # & 1
3
1
3
1
3
1
3
1
3
#
#
0
0
1
0
0
0
&
&
#
#
2
1
0
1
1
0
1
0
0
1
3
2
2
1
&
&
=
acBvaBng
edge
=
repressing
edge
Feedforward
Loop
R
=
# 0 0 0 & D
=
# 1 1 0 0 1 2 1 &
# & # & 1) Clustering
by
single
coherent
feedfoward
loop
# 0 1 0 & # 2 0 1 1 0 3 2 &
1)
r
esults
in
four
clusters
ALGORITHM
STEPS
#
#
1
0
0
0
1
1
&
&
#
#
1
2
3
2
2
1
2
1
3
2
0
1
1
0
&
& 2)
Clustering
by
all
coherent
feedfoward
loops
results
in
nine
clusters.
Three
are
the
same
as
in
1)
and
1
2
1. Create
an
m
x
n
matrix
R,
where
m
is
the
" % " %
the
rest
are
a
refinement
of
the
fourth
cluster
from
1)
number
of
nodes
in
the
network,
n
is
the
number
of
roles
in
the
mo3f,
and
each
3.0
3
4
5
entry
Rij
is
the
number
of
3mes
that
node
i
plays
role
j.
MulB-‐Input
MoBf
2.0
P OTENTIAL
A PPLICATIONS
If
mul3ple
role
matrices
are
created
based
• Gene
networks
(E.
coli)
2. .
• Social
networks
mul3ple
mo3fs.
• Industrial
networks
NODE
ROLES
2.
0
2 5 1 3 4 6 7 • Criminal
networks
Apply
agglomera3ve
hierarchical
clustering
• Computer
networks
• We
define
the
role
of
a
node
as
a
to
the
rows
of
R.
Distance
metrics
and
linkage
criteria
are
user-‐defined.
(ManhaZan
distance;
complete
linkage)
unique
combina3on
of
its
in
and
out
Goal:
Iden3fy
the
“role”
of
each
node
in
the
degrees
network
• Feedforward
loop
has
three
different
roles:
SUBGRAPH
ISOMORPHISM
REFERENCES
2 :
“sender”
• How
do
we
find
all
instances
of
a
given
mo3f
in
a
• Let
A
and
B
be
the
adjacency
matrices
for
GA
and
GB.
R.
Milo,
S.
Shen-‐Orr,
S.
Itzkovitz,
et
al.,
Network
mo3fs:
simple
1
1
3
2
J.
Ullmann,
An
algorithm
for
subgraph
isomorphism,
J.
ACM
23,
:
“receiver”
• Our
method
for
directed
graphs
is
extended
from
wise.
no.
1
(1976),
31-‐42.
Ullmann’s
method
for
undirected
graphs2
3
S.
Shen-‐Orr,
R.
Milo,
S.
Mangan,
et
al.,
Network
mo3fs
in
the
• Create
ini3al
matrix
M0
by
se^ng
mij
=
1
if
it
is
possible
• Suppose
there
is
an
isomorphism
between
a
graph
for
vertex
i
∈
GA
to
map
to
vertex
j
∈ GB;
i.e.
if
indeg(i)
≤
transcrip3onal
regula3on
network
of
Escherichia
coli,
Nature
Gene3cs
31,
no.
1
(2002):
64-‐68.
• Mul3-‐Input
Mo3f
only
has
two
roles:
(mo3f)
GA
and
some
subgraph
of
a
larger
graph
GB.
indeg(j)
and
outdeg(i)
≤
outdeg(j)
4
S.
Mangan
and
U.
Alon,
Structure
and
unc3on
of
the
feed-‐
• We can encode this isomorphism in an m x n • Recursively cycle through all possible permuta3on forward loop network mo3f, PNAS 100, no. 21 (2003),
• If
vertex
i
∈ 2 GA
maps
to
vertex
j
∈2 GB
then
we
set
• Idea
can
be
extended
to
graphs
with
colored
edges
3
4
5
mij
=
1.
• Represent
edge
colors
in
adjacency
matrices
by
1
and
-‐1
2 1
2
3
! 0 1 0 0 0 $ ACKNOWLEDGEMENTS
:
“sender”
:
“receiver”
GA
=
GB
=
#
M
=
#
&
• Then
M
encodes
an
isomorphism
iff
A
=
A
o MBMT,
•
0 0 1 0 0 &
1 3 Davidson
Research
Ini3a3ve
5
4
# 0 0 0 0 1 &
" % where
o
represents
the
Hadamard
product
for
component-‐wise
mul3plica3on
• Dr.
Timothy
Char3er
• Students
and
faculty
in
summer
research
group