0% found this document useful (0 votes)
29 views

Etwork Motifs Lustering Riteria Imple Xample Lustering: N C C S E C E

This document discusses motif-based clustering of directed networks. [1] Network motifs are small, statistically significant subgraphs that occur more frequently in real networks than in random networks. [2] Motifs can be thought of as simple building blocks of complex networks. The authors propose clustering nodes in a network based on the roles they play in different occurrences of network motifs. They quantify the similarity between nodes by counting how many times each node plays each specific role in a given motif. As an example application, they cluster the transcription network of E. coli bacteria based on the feedforward loop motif.

Uploaded by

suci
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
29 views

Etwork Motifs Lustering Riteria Imple Xample Lustering: N C C S E C E

This document discusses motif-based clustering of directed networks. [1] Network motifs are small, statistically significant subgraphs that occur more frequently in real networks than in random networks. [2] Motifs can be thought of as simple building blocks of complex networks. The authors propose clustering nodes in a network based on the roles they play in different occurrences of network motifs. They quantify the similarity between nodes by counting how many times each node plays each specific role in a given motif. As an example application, they cluster the transcription network of E. coli bacteria based on the feedforward loop motif.

Uploaded by

suci
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

MOTIF-­‐BASED

 CLUSTERING  OF  DIRECTED  NETWORKS  


Thomas  P.  Reith,  Laurie  J.  Heyer  
Mathema3cs  Department,  Davidson  College  

NETWORK  MOTIFS   CLUSTERING  CRITERIA   SIMPLE  EXAMPLE   CLUSTERING  E.COLI    


•  Recurrent,  sta3s3cally  significant   •  Given  a  mo3f,  we  consider  nodes  that   •  Network  contains  423  nodes,  578  edges3  
subgraphs  found  at  higher   play  the  same  role  in  different   7    
•  Each  node  represents  DNA  operon;  each  edge  is  
frequencies  in  real  networks  than   occurrences  of  the  mo3f  across  the   directed  from  operon  encoding  transcrip3on  
randomly  generated  networks1   network  to  be  more  similar   6   5   2   factor  to  operon  regulated  by  said  transcrip3on  
  factor  

•  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.  .

on  different  mo3fs  and  concatenated,  the  


•  Neuronal  networks  (C.  elegans)  
network  can  be  clustered  based  on   1.0  

•  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  

building  blocks  of  complex  networks,  Science  298,  no.  5594  


:  “middleman”   network?   (2002),  824-­‐827.  
  •  M  encodes  an  isomorphism  iff  A  ≤  MBMT  component-­‐  

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),  

1   2   permuta3on  matrix  M,  where  m  =  |V(GA)|  and    


                 n  =  |V(GB)|.  
matrices  M  generated  from  M0  and  check  if  each  
encodes  an  isomorphism  
11980-­‐11985.  
 
 
 

 
•  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  
     
 

You might also like