Global alignment of protein-protein interaction networks by graph - - PowerPoint PPT Presentation
Intro PPI alignment Algorithms Results Global alignment of protein-protein interaction networks by graph matching methods. Mikhail Zaslavskiy 1 , Francis Bach 2 and Jean-Philippe Vert 1 1 Mines ParisTech / Institut Curie / INSERM 2 INRIA /
Intro PPI alignment Algorithms Results
Mikhail Zaslavskiy 1, Francis Bach2 and Jean-Philippe Vert1
1Mines ParisTech / Institut Curie / INSERM 2INRIA / Ecole normale superieure de Paris
SMPGD 2010.
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
1 Biological context 2 Protein-protein interaction network alignment 3 Algorithms 4 Experimental results 5 Conclusions
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
1 Biological context 2 Protein-protein interaction network alignment 3 Algorithms 4 Experimental results 5 Conclusions
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Species 1 Species 2
f1: MKQALAAADDDDAQ... y1: MDDDDALGLLLLA... f2:MGDXLLMMAALLLL... y2: MHHAAKLLDDAS...
... ... Objective: Automatic identification of protein functional orthologs
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Species 1 Species 2
f1: MKQALAAADDDDAQ... y1: MDDDDALGLLLLA... f2:MGDXLLMMAALLLL... y2: MHHAAKLLDDAS...
... ... Objective: Automatic identification of protein functional orthologs
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Species 1 Species 2
f1: MKQDLARIEQFLDALF... y1: MSRLPVLLLLQLLVRGA. . . f2: MSKLKIAVSDSCPDCF... y2: MELAALCRAGLLLALDA. . .
... ...
y1 y2 f1 10 50 f2 27 10 Cij-BLAST similarity scores Optimal assignment : f1 → y2, f2 → y1 Potential difficulties: y is the best hit for f, but f is not the best hit for y (y1, f ) and (y2, f ) produce very close blast scores. Which one to choose ?
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Species 1 Species 2
f1: MKQDLARIEQFLDALF... y1: MSRLPVLLLLQLLVRGA. . . f2: MSKLKIAVSDSCPDCF... y2: MELAALCRAGLLLALDA. . .
... ...
y1 y2 f1 10 50 f2 27 10 Cij-BLAST similarity scores Optimal assignment : f1 → y2, f2 → y1 Potential difficulties: y is the best hit for f, but f is not the best hit for y (y1, f ) and (y2, f ) produce very close blast scores. Which one to choose ?
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Species 1 Species 2
f1: MKQDLARIEQFLDALF... y1: MSRLPVLLLLQLLVRGA. . . f2: MSKLKIAVSDSCPDCF... y2: MELAALCRAGLLLALDA. . .
... ...
y1 y2 f1 10 50 f2 27 10 Cij-BLAST similarity scores Optimal assignment : f1 → y2, f2 → y1 Potential difficulties: y is the best hit for f, but f is not the best hit for y (y1, f ) and (y2, f ) produce very close blast scores. Which one to choose ?
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Increase the quality of similarity score, 3D structures Protein interactions . . .
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Increase the quality of similarity score, 3D structures Protein interactions . . .
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
1 Biological context 2 Protein-protein interaction network alignment 3 Algorithms 4 Experimental results 5 Conclusions
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Fly interaction network
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Maximize:the number of overlapping edges
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Maximize:the number of overlapping edges
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Constrained Alignment: Only proteins having close sequence structures may be matched to each other InParanoid clusters: Maximize:J = the number of overlapping edges Balanced Alignment: Let Sij = BLAST(fi, yj), then the total sequence similarity S =
Sij Maximize: (1 − λ)J + λS λ controls the trade-off between J and S.
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Constrained Alignment: Only proteins having close sequence structures may be matched to each other InParanoid clusters: Maximize:J = the number of overlapping edges Balanced Alignment: Let Sij = BLAST(fi, yj), then the total sequence similarity S =
Sij Maximize: (1 − λ)J + λS λ controls the trade-off between J and S.
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Constrained Alignment: Only proteins having close sequence structures may be matched to each other InParanoid clusters: Maximize:J = the number of overlapping edges Balanced Alignment: Let Sij = BLAST(fi, yj), then the total sequence similarity S =
Sij Maximize: (1 − λ)J + λS λ controls the trade-off between J and S.
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Constrained Alignment:
model, approximate solution Our contribution: exact solution by the Message Passing Algorithm Balanced Alignment:
Our contribution: Graph Matching Algorithms
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Constrained Alignment:
model, approximate solution Our contribution: exact solution by the Message Passing Algorithm Balanced Alignment:
Our contribution: Graph Matching Algorithms
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
1 Biological context 2 Protein-protein interaction network alignment 3 Algorithms 4 Experimental results 5 Conclusions
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
MP is based on a forward-backward recursion similar to the Viterbi recursion for hmm.
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Forward step: compute the number of conserved interactions starting from leaves
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Forward step: compute the number of conserved interactions starting from leaves
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Forward step: compute the number of conserved interactions starting from leaves
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Backward step: find optimal assignments starting from the root cluster
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Backward step: find optimal assignments starting from the root cluster
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Backward step: find optimal assignments starting from the root cluster
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Maximize: (1 − λ)J + λS J - number of overlapping edges S - total similarity score between all matched vertices λ - trade-off
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
G, H — graph adjacency matrices P - matrix encoding the alignment between two graphs Graph matching: max
P
(1 − λ)J(P) + λS(P) subject to P ∈ {0, 1}N×N,
Pij = 1 ∀i,
Pij = 1 ∀j (1)
Approximate algorithms: Grad, Path, Umeyama,. . . ..
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
G, H — graph adjacency matrices P - matrix encoding the alignment between two graphs Graph matching: max
P
(1 − λ)J(P) + λS(P) subject to P ∈ {0, 1}N×N,
Pij = 1 ∀i,
Pij = 1 ∀j (1)
Approximate algorithms: Grad, Path, Umeyama,. . . ..
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
G, H — graph adjacency matrices P - matrix encoding the alignment between two graphs Graph matching: max
P
(1 − λ)J(P) + λS(P) subject to P ∈ {0, 1}N×N,
Pij = 1 ∀i,
Pij = 1 ∀j (1)
Approximate algorithms: Grad, Path, Umeyama,. . . ..
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
1 Biological context 2 Protein-protein interaction network alignment 3 Algorithms 4 Experimental results 5 Conclusions
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Data: PPI networks and InParanoid clusters from (S. Bandyopadhyay, R.Sharan and T.Ideker, 2006) Fly (7k nodes, 20k edges) Yeast (4k nodes,15k edges)
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Table: Performance of the different methods for constrained alignment
Algorithm MP GA PATH MRF IsoRank #cons. interactions 238 238 238 233 228 #HomoloG pairs 41 41 41 36 39 Timing(sec) 1-2 1-2 80 10 1-2
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Maximize: (1 − λ)J + λS
Number of conserved interaction J versus sequence similarity S.
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
1 Biological context 2 Protein-protein interaction network alignment 3 Algorithms 4 Experimental results 5 Conclusions
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Global alignment of PPI networks
Intro PPI alignment Algorithms Results
Jean-Philippe Vert Francis Bach SMPGD Organization committee for the travel grant.
Global alignment of PPI networks